Beruflich Dokumente
Kultur Dokumente
Drop formation from a capillary tube: Comparison of one-dimensional and two-dimensional analyses and
occurrence of satellite drops
Physics of Fluids 14, 2606 (2002); 10.1063/1.1485077
I. INTRODUCTION AND MAIN RESULTS Fig. 1, which we call the ‘‘primary filament,’’ as it falls
under gravity. Our objective is to describe the motion of this
Anyone who has had a leaky water faucet is familiar falling liquid filament, including: 共i兲 its time-dependent
with the phenomenon of falling liquid drops. A slow flow shape and dynamics; 共ii兲 its instabilities; and 共iii兲 whether the
rate causes a drop to grow to a critical size, after which it filament pinches off near its ends or at interior points.
falls under the force of gravity; then the process repeats. A nearby but different problem, on the stability of a
High-speed imaging now permits observation of this process uniform liquid jet, was studied long ago by Plateau,12
with a level of detail that was formerly unavailable. The Rayleigh,13,14 and Chandrasekhar.15 People often identify the
sequence of images in Fig. 1 exhibits the detailed process for primary filament of a falling drop with a uniform jet, and an
a particular liquid. As Fig. 1 shows, the drop starts to fall instability on the primary filament is sometimes described in
from the fluid source. As the drop falls, it remains connected the literature as a ‘‘Rayleigh instability.’’ The starting point
to the source by a long, straight liquid filament. This filament of our analysis is to deny this identification, because a uni-
grows thinner and longer, until eventually it pinches off, first form liquid jet and a falling liquid filament are different. This
at the bottom and then at the top of the filament. After pinch- can be seen in at least two ways.
ing off, the elongated filament contracts vertically. Depend- 共a兲 The axial velocity of a uniform liquid jet is uniform
ing on parameter values, the filament may exhibit spatial in both time and space. The axial velocity of a falling liquid
oscillations before it pinches off, and it may contract into one filament cannot be uniform in time, because it accelerates in
satellite droplet 共as in Fig. 1兲 or into more than one after a uniform gravitational field. It is not uniform in space either.
pinching off. As we discuss in detail in the following, this 共b兲 As Fig. 1 shows, the radius of the falling filament
sequence of events is typical for many 共but not all兲 fluids. decreases as the filament grows in length, before pinching
The process of drop formation, with or without gravity, off; the radius of a uniform jet is uniform in time. The two
has been the subject of a great deal of recent research.1–11 motions are different, and neither is a limit of the other.
Inspired in part by the results in Ref. 1, most of these papers In Sec. II, we derive an exact solution of the Navier–
have focused on the detailed structure of the pinch-off. This Stokes equations, in which the radius of the filament de-
paper focuses instead on the motion of the filament visible in creases in time according to
H
a兲
Author to whom all correspondence should be addressed; electronic mail: h共 t 兲⫽ , 共1兲
dmh@math.psu.edu 冑t⫹t *
Before pinch-off, we define the primary filament to be the motion of actual filaments. To test 共1兲, we conducted several
portion of the falling liquid with these properties. It is easy to experiments like that shown in Fig. 1, using a variety of
find solutions to 共2兲–共8兲 with these properties. Let ⫽t liquids. The experimental apparatus was described in detail
⫹t * . Then in Ref. 5; additional information can be found in Appendix
A. In each experiment we measured the thickness, 2h, of the
z⫹z 0 g
w 0 共 r,z,t 兲 ⫽ ⫺ , liquid filament at a fixed z location as a function of time. The
2 constants 兵 H,t * 其 for that experiment were determined by
r applying 共1兲 to two points in the resulting time series. 共See
u 0 共 r,z,t 兲 ⫽⫺ , Appendix A for more detail.兲 For each experiment, we used
2
data for the time interval over which 共1兲 applies 共i.e., for
共9兲
H which z h⫽0 in that experiment兲. We show experimentally
h 共 z,t 兲 ⫽h 0 共 兲 ⫽ , in Sec. V that this time interval increases with increasing
冑 viscosity of the fluid.
1 3 h 20 共 兲 ⫺r 2 1 Figure 2 shows the results for eight experiments, in
p 0 共 r,z,t 兲 ⫽ ⫺ ⫹ ⫹ p amb , which we varied not only the fluid properties 共viscosity and
8 2
h 0共 兲
surface tension兲, but also the external parameters of the ex-
where (t * ,H,z 0 ) are arbitrary constants. The reader can periment 共flow rate and orifice size兲. The parameters for
verify directly that this is an exact solution of the Navier– these experiments are listed in Table I in Appendix A. The
Stokes equations for a fluid with a free surface at r data are graphed with h 冑t⫹t * as a function of t, to deter-
⫽h 0 ( ). For a physical solution, we require H⬎0, t * ⬎0. mine whether the data for each experiment lie on a horizon-
The solution works equally well for a viscous 共⬎0兲 or in- tal line, as predicted by 共1兲. They do, with noticeable oscil-
viscid 共⫽0兲 fluid, with or without surface tension 共兲, with lations. The horizontal lines shown in Fig. 2 correspond to
or without gravity 共g兲. the best value of H for the particular experiment. Thus, Fig.
With g⫽0, 共9兲 describes the motion of a liquid filament 2 shows that for a wide range of fluid and experimental pa-
that is uniformly extending in the z direction. This special rameters, there are values of 兵 H,t * 其 such that 共1兲 models
case was previously discovered by Frankel and Weihs,16,17 well the radial contraction of falling liquid filaments. In the
who used it to describe the extensional flow of a shaped experiments shown in Fig. 2, the radius of the filament typi-
charge. cally contracted by a factor of more than 2 during the time
The solution in 共9兲 with g⬎0 has a simple interpreta- interval shown, so Rayleigh’s model of a uniform jet cannot
tion: it describes the free fall of a liquid filament in a uniform describe these filaments. On the other hand, these data do not
gravitational field. To see this, denote the Lagrangian coor- exclude the possibility of some other model in which the
dinates of a particular fluid particle by 兵 r( ),z( ) 其 . One radius of the filament also contracts, like that in Ref. 19.
finds the motion of this fluid particle by solving two ordinary Note that the data in each experiment in Fig. 2 exhibit
differential equations: small oscillations about the value of H for that experiment.
dz z⫹z 0 g dr r This might be evidence of fluid instability, which we discuss
⫽w 0 ⫽ ⫺ , ⫽u 0 ⫽⫺ . in Sec. III.
d 2 d 2
Before doing so, we mention a variation on 共1兲 and 共9兲.
The result is For t⬍T, the Navier–Stokes equations also admit a solution
g in which
z 共 , 兲 ⫽⫺ 2 ⫹ ⫺z 0 , 共10a兲
2 K
h共 t 兲⫽ ;
冑T⫺t
r共 , 兲⫽ , 共10b兲
冑 the other variables in 共9兲 are unchanged if we reinterpret:
⫽t⫺T in 共9兲. This solution is as robust as that in 共9兲: it
where , are constants. Equation 共10a兲 is the familiar for-
applies for either a viscous or inviscid fluid, with or without
mula for the position of a particle that is falling in a uniform
surface tension, with or without gravity. In 共9兲, the filament
gravitational field. The constant of integration, , can be in-
stretches vertically as it falls; here the filament contracts ver-
terpreted as the vertical velocity of the particle at ⫽0.
tically and grows radially. The solution in 共9兲 describes the
Equation 共10b兲 also has a simple interpretation. All of
motion of the filament before it pinches off. It is tempting to
the fluid particles in Fig. 1 began their motion at the fluid
speculate that this second solution might describe the motion
source. The particles lower in the filament are further from
of the filament after it pinches off, as it contracts into one or
the source, so they have been falling longer and are falling
more droplets 关i.e., in frames 共e兲 and 共f兲 of Fig. 1兴. Unfortu-
faster than those above. Therefore z w⬎0, as one confirms
nately, we have no experimental evidence to support this
directly from 共9兲. Conservation of mass requires that the fila-
speculation.
ment balance this vertical stretching by contracting radially.
This radial contraction accounts for the time dependence
III. STABILITY OF AN INVISCID LIQUID FILAMENT
seen in both 共10b兲 and 共1兲.
Thus, the solution in 共9兲 and 共1兲 has a simple explana- The viscosity of the fluid does not affect the velocity
tion: free fall. We show next that it describes the observed fields in 共9兲, but it affects the stability of the solution. In this
Phys. Fluids, Vol. 12, No. 3, March 2000 The motion of a falling liquid filament 553
冉 冊
terms in the perturbed variables. The result is a set of linear-
ized equations for the perturbed variables, 兵 u 1 (r,z, ), r z⫹z 0 g 1
1⫺ r 1⫹ ⫺ z 1⫹ ⫽0. 共13兲
w 1 (r,z, ),p 1 (r,z, ),h 1 (z, ) 其 . For example, the linearized 2 2 2
forms of 共3兲 and 共4兲 are 共for 0⬍r⬍h 0 ( )兲
冉 冊
The governing equations for this linearized flow are
r u1 z⫹z 0 g 1 more complicated than those for many stability analyses, in-
tu 1⫺ ru 1⫺ ⫹ ⫺ z u 1 ⫹ r p 1 ⫽0,
2 2 2 cluding those of Rayleigh’s jet, because of the variable co-
共11a兲 efficients in 共11兲 and 共13兲. Some of this difficulty can be
冉 冊
overcome by changing variables. Let 兵 r,z, 其 → 兵 , , 其 ac-
r w1 z⫹z 0 g 1
tw 1⫺ rw 1⫹ ⫹ ⫺ z w 1 ⫹ z p 1 ⫽0. cording to
2 2
共11b兲 z⫹z 0 g
⫽r 冑 , ⫽ ⫹ , 共14兲
Define the vorticity of the linearized flow: 2
554 Phys. Fluids, Vol. 12, No. 3, March 2000 Henderson et al.
Viscosity Density Sur ten Flow rate Flow rate Orifice radius
Fluid 共cm2/s兲 共g/cm3兲 共dyn/cm兲 共s/drop兲 共ml/s兲 共cm兲
冉 冊
C(m, ) turns out to be
r z⫹z 0 g
→ ⫹ ⫺ ⫹ ⫹ .
2 冑
冉 冊冋 册
2 2
d I 0 共 兲 dC 3/2 3
冑 ⫺ 3 共 1⫺ 兲 ⫺
2
冑 C⫽0,
Comparing 共14兲 with 共10兲 shows that the new variables can d I 1共 兲 d H 42
be interpreted as Lagrangian-like coordinates of the unper- 共18兲
turbed flow. This identification was first made in Ref. 16.
Moreover, the effect of gravity disappears when the linear- where
ized equations are written in these variables, and the linear-
ized equations here become identical with those in Ref. 16. mH
⫽ 共19兲
Hence we may simply state the main points of the resulting 3/2
analysis.
In terms of the variables in 共14兲, the solution of 共13兲 is is a dimensionless, time-dependent, axial wave number.
1共 , , 兲 ⫽ 冑 0
1共 , , 0 兲 . 共15兲
D(m, ) also satisfies 共18兲. Therefore, for each fixed m, 共18兲
determines the stability/instability of the corresponding wave
mode, for an invisicid fluid.
If the fluid starts from rest, as it does in the experiment Comment: An equation equivalent to 共18兲 was first de-
shown in Fig. 1, then its vorticity vanishes initially 共i.e., at rived in Ref. 16, without the restriction to axisymmetric
⫽ 0 兲, and 共15兲 guarantees that the motion remains irrota- modes. Those authors showed from their more general equa-
tional for ⭓ 0 . We now assume that the linearized motion tion that only the axisymmetric modes can be unstable, so
of the inviscid fluid is irrotational, so 1 ⬅0. Then the lin- 共18兲 contains all of the unstable modes if g⫽0. It is not
earized velocities can be represented in terms of a velocity difficult to show that their conclusion also holds if g⬎0, and
potential, which satisfies a form of Laplace’s equation. It that the analysis presented here captures all the linearized
follows that for an infinitely long filament, the velocity field instabilities of 共9兲.
has the form Comment: The analysis leading to 共18兲 assumed that the
冕 冉 冊
⬁ m filament is infinitely long. For a filament of finite length, the
u 1共 , , 兲 ⫽ I1 3/2 关 A 共 m, 兲 cos共 m 兲 integrals in 共16兲 and 共17兲 should be replaced by discrete
0 sums, and m should be restricted to a countable set of non-
⫹B 共 m, 兲 sin共 m 兲兴 dm, 共16a兲 negative values. These changes do not affect 共18兲, which
applies to each relevant mode 共m兲.
w 1 共 , , 兲 ⫽⫺ 冕 冉 冊
⬁
0
I0
m
3/2 关 A 共 m, 兲 sin共 m 兲
Note that if ⬎0 and m⬎0, then ⬎0 and
( 冑 I 0 ( )/I 1 ( ))⬎0. Therefore 共18兲 has the standard form
of a Sturm–Liouville equation.20 In this form, it is easy to
⫺B 共 m, 兲 cos共 m 兲兴 dm. 共16b兲 identify the stable and unstable modes: 共18兲 necessarily has a
The free surface is given by growing 共i.e., unstable兲 solution if
h 1共 , 兲 ⫽
1
冑
冕 ⬁
0
关 C 共 m, 兲 sin共 m 兲 冋 3/2
H 3 共 1⫺ 兲 ⫺
2
3
42
⬎0. 册 共20兲
These inequalities provide the criteria for the linear sta- more complicated than it is for a liquid jet. This point was
bility of an inviscid liquid filament as it falls. Equations emphasized by Frankel and Weihs,16 who integrated an
共18兲–共20兲 are the main results of this section, so we now equation equivalent to 共18兲 numerically.
examine them in detail.
共1兲 The gravitational constant, g, does not appear in IV. STABILITY OF A VISCOUS LIQUID FILAMENT
共18兲. Thus, the flow in 共9兲 is a result of gravity, but the
inviscid stability of that flow does not depend on gravity. The linearized stability analysis for a viscous liquid is
共2兲 We may write 共20兲 in terms of a conventional wave more complicated than that for an inviscid liquid. We carry
number, as follows. Let k⫽m/ ; then k is an ordinary, di- this analysis to completion only for a ‘‘very viscous’’ liquid,
mensional wave number of a spatial oscillation in the z di- which is defined precisely in 共35兲. In this limit, the main
rection. From 共19兲 and 共9兲, result is that the set of unstable wave numbers is given by
共43兲. The growth of these unstable modes can be found by
mH integrating 共40兲 numerically. In Sec. V, we use these results
⫽ ⫽kh 0 共 兲 . 共21a兲
3/2 to predict whether a given filament pinches off first near its
ends or in the interior of the filament.
Therefore, the set of linearly unstable wave numbers for a A similar analysis was given in Ref. 17. Those authors
falling filament of an inviscid fluid is given by 共20兲, or did not restrict their attention to ‘‘very viscous’’ liquids, so
equivalently by their results are less explicit than those obtained here.
3 共 h0兲3
1⫺ 共 kh 0 兲 ⫺
2
⬎0. 共21b兲 A. The linearized equations
4 2
In terms of the variables defined in 共14兲, the linearized
Note that 共21b兲 can only hold for (kh 0 ) small enough, so this equations for the evolution of the perturbed quantities with
is a long-wave instability, like Rayleigh’s instability for a ⬎0 are as follows.
uniform jet.
共3兲 We may compare the results for a falling inviscid For 0⬍ ⬍H,
filament to Rayleigh’s13 results for an inviscid jet of 共con- 共 u 1 兲 ⫹ ⫺3/2 w 1 ⫽0, 共24a兲
冉 冊
stant兲 radius a. Rayleigh found that a jet is always unstable,
with a set of unstable wave numbers given by u 1 冑 1 u1
u 1⫺ ⫹ p 1 ⫽ 2 u 1 ⫹ u 1 ⫺ 2
2
1⫺ 共 ka 兲 2 ⬎0. 共22兲
2
Thus, the results for an inviscid falling filament differ from ⫹ u , 共24b兲
those for an inviscid jet in three ways. 2 1
共i兲 The constant radius of the jet 共a兲 must be replaced by
the time-dependent radius of the filament 关 h 0 ( ) 兴 .
共ii兲 The constant growth rate of the instability of the jet
w 1⫹
w1
⫹
1
冉 1
冊
p 1 ⫽ 2 w 1 ⫹ w 1 ⫹ 2 2 w 1 ,
共24c兲
must be replaced by the variable growth rate of the instabil-
ity of the filament, obtained by integrating 共18兲. at ⫽0,
共iii兲 An extra term in 共21b兲, 43 关 (h 0 ) 3 / 2 兴 , has no u 1 ⫽0, w 1 ⫽0, 共24d兲
counterpart in 共22兲. This term arises from the radial pressure
gradient of the falling filament, which makes the filament at ⫽H,
more stable than a jet of the same liquid and the same 共in- h1
stantaneous兲 radius. h 1⫹ ⫽u 1 兩 ⫽H , 共24e兲
2
共4兲 A consequence of the additional term in 共21b兲 is that
a filament of sufficiently large radius is linearly stable: this 1 3 Hh 1 h 1 2
p 1 兩 ⫽H ⫽ 5/2 ⫺ 2⫺ h
occurs if 4 H 2 1
3 共 h0兲3 ⫹2 冑 u 1 兩 ⫽H , 共24f兲
⬎1. 共23兲
4 2
As increases, h 0 ( ) decreases, so eventually 共23兲 fails;
every filament becomes unstable eventually. Even so, some
冋
⫺
3
1
册
2 h 1 ⫹ 冑 w 1 ⫹ u 1 ⫽0. 共24g兲
filaments enjoy a time interval of stability, whereas uniform Again, note that the gravitational constant 兵g其 does not ap-
liquid jets are always unstable. pear in 共24兲, so 兵g其 does not affect the linear stability of the
共5兲 In Ref. 13, Rayleigh introduced the concept of the system.
‘‘most unstable wave number.’’ This is the wave number of The vorticity equation follows from 共24b兲 and 共24c兲:
the fastest growing wave mode, so it is the wave number that
one is liable to observe in an experiment. For a falling liquid
filament, 共18兲 is time dependent, so the ‘‘most unstable wave
1⫹
1
2 冉 1 1
冊
⫽ 2 1 ⫹ 1 ⫺ 2 ⫹ 2 2 1 . 共25兲
number’’ changes in time. Thus, the process by which a fall- For a viscous liquid 共⬎0兲, this is a partial differential equa-
ing liquid filament ‘‘selects’’ a particular wave number is tion, so it needs boundary conditions. They are as follows.
556 Phys. Fluids, Vol. 12, No. 3, March 2000 Henderson et al.
共i兲 The vorticity should be bounded for all . We now assume that initially, at ⫽ 0 , P(m, 0 )⫽0
共ii兲 From 共24d兲, 1 vanishes at ⫽0. ⫽Q(m, 0 ), and we use this fact in Appendix B. Then make
共iii兲 From 共24e兲 and 共24g兲, one shows that use of the identity 共p. 111 of Ref. 22兲,
冉 冊
⬁
1 2
1 兩 ⫽H ⫽⫺2 h .
1
共26兲 ⫽ 兺 J 共 n 兲,
H k⫽1 n k HJ 2 共 n k H 兲 1 k
0⭐ ⬍H, 共29a兲
The vorticity equation, 共25兲, is essentially a diffusion where 兵 n k H 其 denote the zeroes of J 1 (x),
equation; its solution decays in time. Even so, 共26兲 shows
that motion of the free surface can create new vorticity at this J 1 共 n k H 兲 ⫽0, k⫽1,2,3,..., 共29b兲
surface. The competition between these two processes 共i.e., so n 1 H⫽3.832..., n 2 H⫽7.016..., etc. The final result is that
creation and decay of vorticity兲 determines the strength of for 0⭐ ⬍H, the vorticity generated at the free surface has
the rotational velocity field. Thus, the velocity field in the the form
viscous problem consists of three parts: ⬁
1 2
共1兲 an irrotational velocity field, analogous to that in the g 共 , , 兲 ⫽⫺ 兺 n k HJ 2共 n k H 兲 J 1共 n k 兲
冑 k⫽1
inviscid problem;
冕
共2兲 a rotational velocity field, generated by the original vor- ⬁
ticity distribution of the fluid 共at ⫽ 0 兲, which decays as ⫻ 关 R 共 m,n k , 兲 cos共 m 兲
→⬁; 0
共3兲 another rotational velocity field, generated by new vor- ⫹S 共 m,n k , 兲 sin共 m 兲兴 dm, 共30a兲
ticity created at the free surface.
where
冉 冊冕 冉 冊
B. Three velocity fields
m2 n 2k 共 s 2 ⫺ 2 兲
R⫽exp exp
The irrotational velocity field is identical in form to that 0 2
in 共16兲. It is one part of the velocity field in the viscous
problem.
Because 共25兲 is a diffusion equation, the original vortic- 冉 再 冎冊
⫻ s P 共 m,s 兲 exp ⫺
m2
s
ds⫺ P 共 m, 兲 . 共30b兲
冉 冊冕 冉 冊
ity decays in time. The corresponding velocity fields also
decay in time. Moreover, this part of the velocity field cre- m2 n 2k 共 s 2 ⫺ 2 兲
S⫽exp exp
ates no new instabilities, beyond those already seen in Sec. 0 2
冉 冉 冊冊
III. Therefore, to simplify the analysis as much as possible,
we now assume that the original velocity field is irrotational, m2
⫻ s Q 共 m,s 兲 exp ⫺ ds⫺Q 共 m, 兲 . 共30c兲
so the original vorticity vanishes 共at ⫽ 0 兲. Then it follows s
that this 共second兲 part of the velocity field also vanishes, for
⭓ 0. Verifying that g ( , , ) satisfies 共26兲 is delicate. The
Finally, consider the vorticity generated by motion of representation in 共29兲 is discontinuous at ⫽H, and this dif-
the free surface, according to 共26兲. This vorticity satisfies a ficulty also appears in 共30兲. Therefore, one must show that
nonhomogeneous equation, with the forcing occurring at the g ( , , ) satisfies 共26兲 in the limit as →H. To do so, use
free boundary. One way to find this vorticity is to find solu- 共30b兲 and 共30c兲 to replace R(m,n k , ) and S(m,n k , ) in
tions of the homogeneous problem, then to use variation of 共30a兲 each by two terms, one integral and one nonintegral. In
parameters to build a Green’s function of the nonhomoge- each case, the exponential factor in the integral terms makes
neous problem 共e.g., see Chap. 8 of Ref. 21兲. Here we simply that sum 共over n k 兲 well behaved, so that these sums are con-
list the main ingredients in this construction, and the final tinuous at ⫽H. Therefore, one can simply evaluate them at
result. First, represent the right-hand side of 共26兲 by ⫽H, where they vanish because J 1 (n k H)⫽0 by 共29b兲. The
two sums that remain are discontinuous at ⫽H, but these
⫺2 冉 冊 冑冕
1
h ⫽
1
1
⬁
0
关 P 共 m, 兲 cos共 m 兲
two sums 共over n k 兲 each decouple from the integral 共over m兲.
For these sums, using 共29a兲 allows one to evaluate the limit
( →H) easily, and to show that g ( , , ) satisfies 共26兲 as
⫹Q 共 m, 兲 sin共 m 兲兴 dm. 共27兲 →H.
Verifying that g ( , , ) satisfies 共25兲 for 0⬍ ⬍H is
With h 1 ( , ) represented as in 共17兲, the relations between even more delicate, because formal differentiation inside the
the coefficients are sum in 共30a兲 leads to divergent series. One way to resolve
this difficulty is to show that g ( , , ) is a weak solution of
P 共 m, 兲 共25兲. Equivalently, one can establish conventions to interpret
⫽⫺2m 共 ⫺3/2C 共 m, 兲兲 , particular divergent sums. The following conventions arise
冑 by formally differentiating 共29a兲 twice:
共28兲
⬁
Q 共 m, 兲 1
冑
⫽⫹2m 共 ⫺3/2
D 共 m, 兲兲 . 1⫽ 兺
k⫽1 J 2 共 n k H 兲
J 0共 n k 兲 , 0⭐ ⬍H, 共31a兲
Phys. Fluids, Vol. 12, No. 3, March 2000 The motion of a falling liquid filament 557
⬁
n tions 兵 A(m, ),B(m, ),C(m, ),D(m, ), P(m, ),Q(m, ) 其 .
0⫽ 兺 k
k⫽1 2 k H 兲
J 共 n
J 1共 n k 兲 , 0⭐ ⬍H. 共31b兲 These equations, along with 共28兲, determine the motion of a
viscous liquid filament.
We reiterate that these are conventions; the sums in 共31兲 may We omit the equations obtained in this way, which are
diverge. linear but complicated integro-differential equations for the
By direct calculation from 共30兲, unknown functions listed above. Equations corresponding to
冉 冊
these can be found in Ref. 17. In order to simplify the analy-
m2
R 共 m,n k , 兲 ⫹ n 2k ⫹ R⫽⫺ n 2k P 共 m, 兲 , sis, we now make an additional assumption.
2
冉 冊
共32兲 D. Assumption: The liquid is very viscous
m2
S 共 m,n k , 兲 ⫹ n 2k ⫹ S⫽⫺ n 2k Q 共 m, 兲 . Recall that R(m,n k , ) and S(m,n k , ) are defined in
2
共30兲. We show in Appendix B that for large viscosity 共→⬁兲
Using 共32兲 and 共31b兲, one can verify by differentiating and ⬎ 0 ,
within the sum and within the integral that g ( , , ) satis-
fies 共25兲, at least formally. It is evident that g ( , , ) van- n 2k 3
ishes at ⫽0, and that it is bounded for all , for suitable R 共 m,n k , 兲 →⫺ P 共 m, 兲 ,
m 2 ⫹n 2k 3
兵 P,Q 其 . This completes the verification that g ( , , ) is a 共35兲
suitable representation of the forced vorticity field. n 2k 3
The corresponding velocity fields are S 共 m,n k , 兲 →⫺ Q 共 m, 兲 .
m 2 ⫹n 2k 3
⬁
2 In terms of the differential equations in 共32兲, 共35兲 asserts that
u g1 共 , , 兲 ⫽ 冑 兺 J 共n 兲
k⫽1 n k HJ 2 共 n k H 兲 1 k the homogeneous solutions of 共32兲 decay much faster than
冕冋 册
the time scale on which the forcing terms, P(m, ) and
⬁ m
⫻ 关 R 共 m,n k , 兲 sin共 m 兲 Q(m, ), evolve. This rapid decay occurs in a very viscous
0 m ⫹n 2k 3
2
fluid. We now define a ‘‘very viscous fluid’’ to be one for
which 共35兲 holds.
⫺S 共 m,n k , 兲 cos共 m 兲兴 dm,
This assumption simplifies the representation of the ve-
⬁ 共33兲 locity field generated by the motion of the free surface,
2
w g1 共 , , 兲 ⫽ 2 兺
k⫽1
J 共n 兲
n k HJ 2 共 n k H 兲 0 k
u g1 ( , , ) and w g1 ( , , ). It also reduces the system of lin-
earized equations from third order in time to second order. In
⫻ 冕冋 0
⬁ nk
m ⫹n 2k 3
2 册 关 R 共 m,n k , 兲 cos共 m 兲
terms of from 共19兲 and (n k H) from 共29b兲, define
⬁
2
⫹S 共 m,n k , 兲 sin共 m 兲兴 dm.
共 兲 ª2 兺 2
k⫽1 ⫹ 共 n k H 兲
2. 共36兲
For a viscous filament with no initial vorticity, the complete In addition, recall one more identity 共Ref. 23, p. 361兲:
velocity field is the sum of those in 共16兲 and 共33兲.
J 0 共 n k H 兲 ⫹J 2 共 n k H 兲 ⫽0.
C. Conditions at the free surface Using all of these plus 共35兲 in 共33兲 leads to limiting values
One condition at the free surface results from eliminat- 共as →H兲 of the quantities that appear in the boundary con-
ing the pressure between 共24c兲 and 共24f兲, both evaluated at ditions at the free surface:
⫽H: u g1 共 , , 兲 →0,
1
冉 1
1
冊
p 1 ⫽ 2 w 1 ⫹ w 1 ⫹ 2 2 w 1 ⫺ 共 w 1 兲 ,
w g1 共 , , 兲 →
H
冕 ⬁ ⬘共 兲
2
关 P 共 m, 兲 cos共 m 兲
冉 冊
共34兲 0
1 3H 3
p 1 兩 ⫽H ⫽ ⫺ h ⫺ h ⫹Q 共 m, 兲 sin共 m 兲兴 dm,
4 7/2 H 2 1 3 1
共37兲
⫹
2
u 1 兩 ⫽H .
u g1 共 , , 兲 →
冑
2H
冕 ⬘
⬁
共 兲关 P 共 m, 兲 cos共 m 兲
冑 0
⫹Q 共 m, 兲 sin共 m 兲兴 dm,
Two other conditions are shown in 共24e兲 and 共24g兲. These
conditions are not independent, because 共24e兲 and 共24g兲
were already used to derive 共26兲. In what follows, we use
共34兲 and 共24e兲, but not 共24g兲.
冉 1
冊
2 w g1 ⫹ w g1 ⫹ 2 2 w g1
The next step is to substitute the representations for the
velocity fields from 共16兲, and 共33兲, plus the representation for →
H
冕 0
⬁
关 2⫹ 共 兲兴关 P 共 m, 兲 cos共 m 兲
the free surface from 共17兲, into the conditions at ⫽H, in
order to obtain evolution equations for the unknown func- ⫹Q 共 m, 兲 sin共 m 兲兴 dm.
558 Phys. Fluids, Vol. 12, No. 3, March 2000 Henderson et al.
the other for 兵 (A(m, ),D(m, ),⫺Q(m, ) 其 . The first set is Observe that ⌬ is exponential with a real-valued exponent,
冉 冊
so ⌬⬎0, and 共42兲 is indeed in Sturm–Liouville form. In
H ⬘共 兲 1 Appendix D, we show that ␣共兲⬎0 for ⭓0. Therefore 共42兲,
共 2⫹ 共 兲兲 P⫺ P ⫺ 共 I 0共 兲 B 兲
H 2 and 共40兲, have a nonoscillatory growing solution, and the
冉 冊
filament is unstable, if
3
⫽ 5/2⫺ 3 共 1⫺ 兲 C⫹ ⬘ 共 兲 P
冉 冊
2
4 H H m 2 冑 m 2
⌫ , , ⬎0, 共43兲
H
2
⫹ 2 2 I ⬘1 共 兲 B, where ⌫ is given in 共41c兲. This is the criterion for unstable
H
共38兲 wave numbers for a very viscous filament; it is one of the
C⫽ 冑 I 1 共 兲 B. two main results in this section.
This set of equations, along with 共28兲, determine the For which dimensionless wave numbers 共兲 is ⌫⬎0? In
linearized stability of a very viscous filament. To analyze 共41c兲, the first term on the right-hand side represents the
this system, it is convenient to rewrite it as a second-order effect of surface tension; it is positive for 0⬍⬍1. The sec-
differential equation for C(m, ). In addition, we eliminate ond term represents the effect of viscosity; because ␣共兲⬎0
共兲 by using an identity derived in Appendix C: for all ⬎0, this term is always positive. We show in Ap-
pendix D that the last term in 共41c兲 is positive for 0⬍⬍X
I 0共 兲 共X⬇2兲. Thus, every filament is unstable in a very viscous
共 兲⫽ ⫺2. 共39兲
I 1共 兲 liquid, because the set of unstable wave numbers always in-
The final result, after some algebra, is cludes 0⬍⬍1, and it may include more.
冉 冊 冊 冉
show in Appendix D that
I 0共 兲 2
I 0共 兲
␣共 兲⫽ ⫺ ⫺ 2, 共41a兲 ␣ 共 兲 ⬎0 for ⬎0, 共44a兲
I 1共 兲 I 1共 兲
 ,冉 m2
⫽3 冊 冉 冊 冉 冊 冉 冊
I 0共 兲 3
⫺7
I 0共 兲 2
⫹4
I 0共 兲 ␣ 共 兲 →2 as →0, 共44b兲
I 1共 兲 I 1共 兲 I 1共 兲
冉 m2
冊 m2
冉 冊 冉 冊
 , ⫺ ␣ 共 兲 ⬎0 for ⬎0, ⭓0, 共44c兲
I 0共 兲 2m2
⫺3 2
⫹4 2 ⫹
I 1共 兲
冉 m2
冊
冋冉 冊 册
 , ⫺ ␣ 共 兲 →2 as →0, m→0. 共44d兲
I 0共 兲 2
⫻ ⫺ ⫺1 ,
2
共41b兲
I 1共 兲 In addition, recall that 共40兲 or 共42兲 applies only for a
⌫ ,冉 m m
H
,
2
冑
冊 冑 2
⫽
m
H
2
共 1⫺ 2 兲 ⫹
3m
␣共 兲
2 very viscous liquid. For wave number m⬎0, the viscosity 共兲
of the liquid appears in both  and ⌫, so each of these coef-
ficients is asymptotically large. We show next that in the
冋冉 冊 ⫹
3
4
6
I 0共 兲
I 1共 兲
3 ‘‘very viscous’’ limit, 共40兲 approximately splits into two
first-order equations, which we solve approximately. Let
Ⰷ1 denote a large dimensionless parameter representing
冉 冊 冉 冊 ⫺16
I 0共 兲
I 1共 兲
2
⫹8
I 0共 兲
I 1共 兲
the viscosity of the filament, e.g., ⫽ m 2 / . For a very
viscous liquid and m⬎0, it follows from 共41兲 that
冉 冊 册 ⫺6 2
I 0共 兲
I 1共 兲
⫹9 2 . 共41c兲
␣ ⫽O 共 1 兲 ,  ⫽O 共 兲 , ⌫⫽O 共 兲 .
Then 共40兲 has a fast-changing solution, C f (m, ), with
共45兲
冉冕 冊
of the unstable mode is approximately given by 共47兲. For a
 given filament, a graph like that in Fig. 3 shows which wave
C f 共 m, 兲 ⬃C f 共 m, 0 兲 exp ⫺ d . 共46兲
0 ␣ numbers grow most rapidly, and how their instantaneous
From 共44兲, 兵/␣其⬎1, so this solution decays, relatively fast. growth rates change with time. A particular liquid filament
The other solution of 共40兲, C s (m, ), changes slowly: qualifies as ‘‘very viscous’’ if the time scale of growth in
共47兲 is slow in comparison with the time scale of decay in
Cs 2C s 共32兲.
C s ⫽O 共 1 兲 , ⫽O 共 1 兲 , ⫽O 共 1 兲 .
2 In Sec. V, we use these results to predict whether a given
liquid filament pinches off first near its ends, like the fila-
For this solution, 共40兲 becomes approximately
ment in Fig. 1, or if it pinches off in its interior.
 Cs ⌫
⫺ C ⬃0,
2 s V. WHERE DOES A FILAMENT PINCH OFF?
so that The filament in Fig. 1 pinches off near its ends 共first at
C s 共 m, 兲 ⬃C s 共 m, 0 兲 exp 冉冕
0
⌫

d . 冊 共47兲
the bottom of the filament, then at the top兲. This behavior is
commonly observed in experiments.2–5,11 It is called ‘‘end-
pinching’’ in Ref. 11, and it occurs in filaments that form
If 共43兲 holds for a particular m⬎0, then C s (m, ) grows with either because of gravity, as in our Fig. 1 and in Refs. 2–5 or
, relatively slowly. This is the unstable mode for this m. without gravity.11 However, not all filaments pinch off first
Both  and ⌫ depend on , so the unstable modes do not near their ends, as noted in Ref. 11 for filaments and drops
exhibit simple exponential growth. The instantaneous growth created without gravity in a four-roll mill. Figure 4 shows the
rate of an unstable mode is given by results of an experiment like that in Fig. 1, but for a more
⌫ viscous fluid. 共This experiment uses fluid E, described in
G⫽ , 共48兲 Table I.兲 Several differences appear in the motion of these

two filaments.
and it changes with time. Moreover, the ‘‘most unstable 共1兲 More viscous filaments survive longer. 共This can al-
wave number’’ 关i.e., the wave number m that maximizes the ready be inferred from Fig. 2.兲 As a result, the experiment in
growth rate in 共48兲 at fixed 兴, changes with time; this is Fig. 4 lasts longer than that in Fig. 1.
another difference between a falling liquid filament and a 共2兲 Because it lasts longer 共in time兲, the filament in Fig.
uniform jet. Figure 3 shows graphs of instantaneous growth 4 grows to a much longer length than that in Fig. 1. As a
560 Phys. Fluids, Vol. 12, No. 3, March 2000 Henderson et al.
The contracting filament of fluid C exhibited instabilities that was initiated, and of the Mathematics Department, Hong
grew substantially, causing the filament to break up into sev- Kong University of Science and Technology, where it was
eral satellite drops. completed. His work in Japan was supported by the Japan
For all of our experimental data, therefore, this criterion Society for the Promotion of Science. The work was also
successfully predicts whether a filament pinches off first near supported in part by NSF Grant Nos. DMS-9304390 and
its ends or at internal points. To our knowledge, this is the DMS-9731097. The experiments were supported by fellow-
first theoretical prediction of this behavior. ships from the David & Lucile Packard Foundation and the
Alfred P. Sloan Foundation, and from NSF Grant No. DMS-
ACKNOWLEDGMENTS 9257456.
We thank Bernard Deconinck for discovering 共39兲,
James Meiss for catching a logical error in an early version
APPENDIX A: EXPERIMENTAL APPARATUS,
of this paper, Joe Hammack for many helpful discussions, MATERIALS, AND PROCEDURES
and an anonymous referee for bringing Refs. 16 and 17 to
our attention. H. S. acknowledges the kind hospitality of the The experimental apparatus comprised silicone and com-
Physics Department, University of Tokyo, where this work mercial vegetable oils, Lucite reservoirs, and an imaging sys-
562 Phys. Fluids, Vol. 12, No. 3, March 2000 Henderson et al.
TABLE II. Spatial and temporal resolution for each experiment shown in filament widths and filament lengths were obtained directly
Fig. 2.
from the monitor, from the hard copies and/or using PhotoA-
Spatial resolution Temporal resolution dobe on the computer for image enhancement and edge de-
Fluid 共m pixel兲 共frame/s兲 tection. Comparisons among the different techniques did not
show significant variability.
A1 110 3000
A2 111 3000 Measurements for filament width and length were ob-
B1 33.6 3000 tained as follows. First, for all fluids, the experiment was
B2 25.6 3000 imaged with a coarse resolution so that the entire drop was
B3 33.6 3000 visible until pinch-off. From this global view, reference
C 39.6 6000
D 63.7 3000
points were obtained. In particular, we noted the location of
E 42.1 3000 the drop when we first judged a filament with h/ z⫽0 to
have formed. Then we noted the interval of time that passed
while the drop fell from this location until we first observed
either necking at the ends of the filament 共for fluids A, B,
tem. Table I lists the fluid properties 共viscosity, surface ten- and C or wavelike instabilities on the filament 共for fluids D
sion, density兲 as well as the experimental parameters 共flow and E兲. Measurements of filament width for fluid A were
rate and orifice size兲. obtained from this global view during this time-interval.
Fluids B, C, D, and E were silicone oils 共Dow Corning兲 Second, measurements of length were obtained from this
chosen for their ranges of viscosities from about 100 to 1000 time series by placing the reticule from the imager onto the
times more viscous than water. The surface tension of these filament boundary and measuring the length under it for
oils did not vary significantly. Fluid A was a commercial which h/ z⫽0. These measurements are accurate to within
vegetable oil 共Weis brand兲, with a surface tension that varied ⫾0.01 cm.
significantly from that of the silicone oils. Viscosities were Third, we zoomed-in the imager to obtain higher-
measured in a temperature-controlled bath; see Ref. 5 for resolution for measurements of filament width, for all fluids
details of fluids A and B. Surface tension was measured with except C. The resolutions are listed in Table II. In these
a DuNouy tensiometer at room temperature. Experiments A1 images a portion of the filament was in the image, but neither
and A2 used the same fluid with different flow rates. Experi- the orifice nor the drop was in the image during measure-
ments B1–B3 used the same fluid with different orifice sizes. ments. The position of the image was known relative to the
The flow rates were measured as an average over several orifice and to the reference location obtained from the global
drops. view. We waited a known 共small兲 amount of time after the
For a given experiment, the oil was poured into a reser- drop passed this location. Then this time was defined to be
voir and left open to the atmosphere. It dripped from an t⫽0 in the corresponding measurements of filament width.
orifice under gravity at a rate controlled by a needle valve. 共Filament lengths are also referenced to this time.兲 Measure-
The orifice was machined to have a flat edge so that the oil ments were taken during the time interval over which
coated the surface area between the inner and outer edges. dh/dz⫽0, determined from the global view.
The radius of the outer edge is listed in Table I. The dimen- To obtain values of H and t * from the measurements of
sions of the overall apparatus are given in Ref. 5. The appa- filament width, we calculated (H,t * ) from every possible
ratus was enclosed in a plastic box that was not temperature combination of two data points. Then we considered the er-
controlled but stopped air-currents and external contamina- ror 共in a least-squares sense兲 between the calculated values
tion. of H and the corresponding values of h 冑 . We considered
A Kodak EktaPro 1012 EM Motion Analyzer captured the values of (H,t * ) that provided the smallest error, the
images of the falling drop with spatial and temporal resolu- values obtained as an average over some small error, and in
tions listed in Table II. the end, we chose the values that, by eye, showed the best
Illumination was obtained using silhouette photography, agreement with the entire set of data points. These values of
following Ref. 24, with a 600 W lamp, an experimental H correspond to the horizontal lines in Fig. 2 and are listed
grade one-way transparent mirror 共Edmund Scientific, along with t * in Table III. They could have changed by
A40,047兲 and reflective screen material 共Scotchlite 3M about ⫾5% each. This variation did not effect the results of
7615兲. The mirror, placed between the camera and drop, was the stability calculations. The results of these calculations are
oriented at a 45° angle to the camera’s face. Approximately also listed in Table III. In particular, we list the values of the
50% of the incident light was reflected toward the drop. The wave number, m, that had the maximum growth rate 关from
reflective material, placed directly behind the drop, reflected Eq. 共48兲兴 at time t⫽0. For fluids 共A, B兲, this occurred at the
the incoming light rays back toward the drop. The effect of longest wavelength that fit into the filament at that time; for
this setup was to render the drop as a shadow. The reflective fluids 共C, D, E兲, it occurred at the peak of the curve corre-
material, rated as 98% efficient, sends back the light within sponding to that shown in Fig. 3.
an angle of 0.5° of its initial path,24 resulting in less scatter- Finally, here are two comments about our observations.
ing of light around the edges of the imaged drops and a First, because of inadequate spatial and temporal resolution,
higher-contrast image. we were unable to measure modal amplitudes as a function
Images were visible on a monitor and down-loaded to of time. Instead, we measured wavelengths that were observ-
video, to hard copy, and/or to a computer. Measurements of able at various times and at various spatial resolutions. We
Phys. Fluids, Vol. 12, No. 3, March 2000 The motion of a falling liquid filament 563
0
e 共 s 兲 s P 共 m,s 兲 ds
B1 0.0018 0.0045 0.130
B2
B3
C
0.0021
0.0025
0.0020
0.0043
0.0070
0.0048
0.094
0.180
0.099
⫹ 冕
0
e 共 s 兲
m2
s2
P 共 m,s 兲 ds⫺ P 共 m, 兲 ,
D 0.0019 0.0083 0.148
E 0.0028 0.0211 0.253 ⫽R1⫹R2⫺ P.
But
0
e 共 s 兲 s 冉 m 2 /s 2
n 2k s⫹m 2 /s 2 冊 P 共 m,s 兲 ds
冉 冊
resolution, we could not resolve the short wavelengths; when
we observed with a fine resolution, we could not resolve the m 2 /s 2
⫽e 共 s 兲 P 共 m,s 兲 兩
long wavelengths. For example, when viewing experiment E n 2k s⫹m 2 /s 2 0
at ⫽0.17 s, with a fine resolution that allowed only a por-
tion of the filament into the image, we observed wavelengths
of about 0.6 cm. The image size was such that we could not
⫺ 冕
0
e 共 s 兲 s 冋 m 2 /s 2
n 2k s⫹m 2 /s 2
P 共 m,s 兲 ds,册
冉 冊
also observe wavelengths larger than about 1 cm. The am-
plitudes were essentially 1 pixel. Thus, when viewing the m2
⫽ P 共 m, 兲
filament at a coarser resolution, we could not resolve these n 2k 3 ⫹m 2
冕 冋 册
amplitudes and would not observe those modes. Instead,
m 2 /s 2
when viewing the filament at a coarse resolution that allowed ⫺ e 共 s 兲 s P 共 m,s 兲 ds,
the whole filament-drop combination into the image, we ob- 0 n 2k s⫹m 2 /s 2
served waves with lengths of about 3 cm. They also had where we have used P(m, 0 )⫽0 and 共兲⫽0. Therefore,
amplitudes of about 1 pixel. 共We note that the most unstable
wavelength at t⫽t * gives a wavelength at ⫽0.17 of 4.2
cm.兲 Presumably, fluids such as D and E, for which pinch-off R 共 m,n k , 兲 ⫽⫺ 冉 n 2k 3
n 2k 3 ⫹m 2
冊 P 共 m, 兲
冋 册
occurs within the filament, admit many unstable modes and
at present we cannot observe their evolutions.
Second, we also note that previous work2,5 showed that
⫹ 冕
0
e 共 s 兲 s 2
n 2k s
n k s⫹m 2 /s 2
P 共 m,s 兲 ds.
pinch-off occurred in a secondary filament localized at the
end of the filament near the drop. Reference 2 observed this In the last integral, both the quantity in square brackets and
same structure at the orifice, while Ref. 5 did not observe it its derivative have bounds that are independent of , so the
for fluid B2. It was conjectured in Ref. 5 that this difference integral vanishes as →⬁ by Watson’s Lemma.25 This leads
was due to the difference in orifice sizes, since that of Ref. 2 to 共35兲.
was significantly larger than that in Ref. 5. Our observations To evaluate R(m,n k , ) as →0, expand (s) in a Tay-
support this conjecture: when we used the same fluid with a lor series in powers of , and integrate term by term. The first
larger orifice 共in experiment B3兲, the secondary filament did integral cancels (⫺ P(m,t)), and the other integrals vanish
form. as →0. Because R(m,n k , ) and S(m,n k , ) both vanish as
→0, one can show that the theory for a viscous fluid repro-
duces that of an inviscid fluid as →0.
APPENDIX B: LIMITING BEHAVIOR OF R „ m , n k , …
冋 册
⬁
m2 共 n H 兲2
s ⫽ n 2k s⫹
s2
⬎0 for 0⬍s⬍⬁, ⬘ 共 兲 ⫽4 兺
k⫽1 关 2
⫹
k
共 n kH 兲2兴 2
. 共C1兲
564 Phys. Fluids, Vol. 12, No. 3, March 2000 Henderson et al.
⬎0. To establish an upper bound on 共兲 for ⬎0, observe Proof: Both I 0 ( ) and I 1 ( ) have convergent Taylor
that series expansions near ⫽0.22,23 The representation in 共i兲 fol-
lows from these. Similarly, both functions have asymptotic
⬁ ⬁
1 1 expansions as →⬁, and 共ii兲 follows from these. The differ-
共 兲 ⫽2 2 兺
k⫽1 ⫹共 n kH 兲
2 2 ⬍2
2
兺
k⫽1 ⫹ 共 k 兲
2 2, ential equation in 共iii兲 follows from those in 共C4兲.
Lemma 4:
so
共 兲 ⬍2 2
冕 ⬁ dk
2⫽
2
冕 ⬁ dy
⫽.
I 0共 兲 1
⬎ ⫹
I 1共 兲 2
冑 1
共 2 兲2
⫹1 for 0⬍ ⬍⬁.
0 ⫹共 k 兲
2
0 1⫹y 2
Proof: Define
冑
Remark: Lemma 2 is due to Bernard Deconinck.
Lemma 2: 1 1
y 共 兲ª ⫹ ⫹1,
2 共 2 兲2
I 0共 兲
共 兲⫽ ⫺2. so
I 1共 兲
y
Proof: The Bessel function J 1 (Y ) can be written in terms of 1⫺y 2 ⫹ ⫽0. 共C5兲
an infinite product:26
冉 冊
⬁ Lemma 4 asserts that I 0 ( )/I 1 ( )⬎y( ) for 0⬍⬍⬁. The
Y Y2
J 1共 Y 兲 ⫽
2 兿
k⫽1
1⫺
共 n kH 兲2
, assertion is valid as →0 by 共i兲 of Lemma 3, so we must
prove that it remains valid for ⬎0. The proof is by contra-
diction. Assume that there exists X * , 0⬍X * ⬍⬁, such that
where J 1 (n k H)⫽0. Taking logarithms and differentiating
at ⫽X * ,
yields
I 0 共 X * 兲 /I 1 共 X * 兲 ⫽y 共 X * 兲 . 共C6兲
冉 冊
⬁
Y2
ln J 1 共 Y 兲 ⫽ln Y ⫺ln 2⫹ 兺
k⫽1
ln 1⫺
共 n kH 兲2
, If 共C6兲 occurs more than once, then let X * be the smallest
positive value at which 共C6兲 is valid. Comparing 共iii兲 of
so Lemma 3 with 共C5兲 shows that
J 1⬘ 共 Y 兲1
⫽ ⫺
⬁
2Y / 共 n k H 兲 2
兺
J 1 共 Y 兲 Y k⫽1 1⫺Y 2 / 共 n k H 兲 2
d I0
d I1冉冊
⫽0 at ⫽X * . 共C7兲
I ⬘1 共 兲1 2 2
⬁
1 共 兲
d I0
d I1冉冊
⬍0 at ⫽X * .
⫽ ⫹ 兺
I 1 共 兲 k⫽1 共 n k H 兲 ⫹
2 2⫽ ⫹
. 共C3兲 This contradicts 共C7兲, so there can be no such point X * , and
Lemma 4 holds for all ⬎0.
From Ref. 22,
d d
关 I 1 共 兲兴 ⫽ I 0 共 兲 , I 共 兲 ⫽I 1 共 兲 , 共C4兲 APPENDIX D: THE COEFFICIENTS IN „41…
d d 0
Therefore I 1⬘ ( )/I 1 ( )⫽I 0 ( )/I 1 ( )⫺1/ . Substituting this Lemma 5: ␣共兲⬎0 for 0⬍⬍⬁.
into 共C3兲 yields Lemma 2. Proof: From 共41a兲,
Lemma 3: I 0 ( )/I 1 ( ) has the following properties.
冉 冊冋冉 冊 册
I0 I0
冋 冉冊 冉冊 册
␣共 兲⫽ ⫺1 ⫺ 2 . 共D1兲
I 0共 兲 2 1 2
1 4
I1 I1
共 i兲 ⫽ 1⫹ ⫺ ⫹O 共 6 兲
I 1共 兲 2 2 12 2 From Lemma 4,
as →0,
冉 冊
I0 1
⫺1⬎ ⫹ 冑 1
⫹ 2 ⫺1⬎0 for ⬎0.
共 ii兲
I 0共 兲
I 1共 兲
⫽1⫹
1
⫹
3 1
2 2 2 冉 冊 2
⫹O 共 ⫺3 兲 ]
I1 2 4
Using Lemma 4 again shows that for ⬎0,
as →⬁, ␣ 共 兲 ⬎ 关 21 ⫹ 冑 14 ⫹ 2 兴 • 关 ⫺ 21 ⫹ 冑 14 ⫹ 2 兴 ⫺ 2 . 共D2兲
Phys. Fluids, Vol. 12, No. 3, March 2000 The motion of a falling liquid filament 565
The right-hand side of 共D2兲 is identically zero, so this estab- This is a contradiction, so there can be no such point X * .
lishes Lemma 5. Therefore T 2 ( )⬎0 for all ⭓0. Therefore both factors in
Lemma 6: As →0, ␣共兲→2. 共D4兲 are positive for ⭓0, so T 1 ( )⬎0. Therefore both
Proof: Use 共41a兲 and 共i兲 of Lemma 3. terms in 共D3兲 are positive for ⬎0, m 2 / ⭓0. This com-
Lemma 7: With  ( , m 2 / ) and ␣共兲 defined in 共41兲, pletes the proof.
冉 冊
Lemma 8: As →0 and m→0, ⫺␣→2.
m2 m2
 , ⫺ ␣ 共 兲 ⬎0 for ⬎0, ⭓0. Proof: This follows from 共D3兲, 共D4兲, and 共i兲 of
Lemma 3.
Proof: From 共41兲, we may write
 ⫺ ␣ ⫽T 1 共 兲 ⫹ 冉 冊冋冉 冊
2m2
I0
I1
⫺ 2 ⫺1 , 册 共D3兲
1
2
J. Eggers, ‘‘Universal pinching of 3D axisymmetric free-surface flow,’’
Phys. Rev. Lett. 71, 3458 共1993兲.
X. D. Shi, M. P. Brenner, and S. R. Nagel, ‘‘A cascade of structure in a
where drop falling from a faucet,’’ Science 265, 219 共1994兲.
冉 冊 冉 冊 冉 冊 冉 冊
3
M. P. Brenner, X. D. Shi, and S. R. Nagel, ‘‘Iterated instabilities during
I0 3
I0 2
I0 I0 droplet fission,’’ Phys. Rev. Lett. 73, 3391 共1994兲.
T 1 共 兲 ⫽3 ⫺8 ⫹5 ⫺3 2 4
X. Zhang and O. A. Basaran, ‘‘An experimental study of dynamics of drop
I1 I1 I1 I1 formation,’’ Phys. Fluids 7, 1184 共1995兲.
5
D. M. Henderson, W. G. Pritchard, and L. B. Smolka, ‘‘On the pinch-off
⫹5 2 . of a pendant drop of viscous fluid,’’ Phys. Fluids A 9, 3188 共1997兲.
6
From Lemma 4, for ⬎0, D. H. Peregrine, G. Shoker, and A. Symon, ‘‘The bifurcation of liquid
bridges,’’ J. Fluid Mech. 212, 25 共1990兲.
冋 冉 冊 册冋 冉 冊 冉 冊 册
共Soc. Ind. Appl. Math.兲 J. Appl. Math. 43, 268 共1983兲.
I0 I0 2
I0 11
H. A. Stone, B. J. Bentley, and L. G. Leal, ‘‘An experimental study of
T 1共 兲 ⫽ 3 ⫺5 ⫺ ⫺2 . 共D4兲 transient effects in the breakup of viscous drops,’’ J. Fluid Mech. 173, 131
I1 I1 I1 共1986兲.
12
Comparing with 共D1兲 shows that the second factor in 共D4兲 is J. Plateau, Statique Expérimentale et Théoretique des Liquides Soumis aux
Seules Forces Moléculaire 共Gauthier-Villars, Paris, 1873兲.
␣共兲, so by Lemmas 5 and 6 it is positive for ⭓0. 13
Lord Rayleigh, ‘‘On the instability of jets,’’ Proc. London Math. Soc. 10,
To show that the first factor in 共D4兲 is positive, define 4 共1879兲.
冉 冊
14
Lord Rayleigh, ‘‘On the instability of a cylinder of viscous liquid under
I0 5 capillary force,’’ Philos. Mag. 34, 145 共1892兲.
T 2共 兲 ⫽ ⫺ . 15
S. Chandrasekhar, Hydrodynamic and Hydromagnetic Stability 共Dover,
I1 3
New York, 1961兲, especially Chap. 12,
It follows from 共i兲 of Lemma 3 that as →0, T 2 ( ) 16
I. Frankel and D. Weihs, ‘‘Stability of a capillary jet with linearly increas-
→1/3⬎0. We need to show that T 3 ( )⬎0 for all ⬎0. Use ing axial velocity,’’ J. Fluid Mech. 155, 289 共1985兲.
17
I. Frankel and D. Weihs, ‘‘Influence of viscosity on the capillary instabil-
共iii兲 of Lemma 3, plus algebra, to show that ity of a stretching jet,’’ J. Fluid Mech. 185, 361 共1987兲.
18
G. K. Batchelor, An Introduction to Fluid Dynamics 共Cambridge Univer-
dT 2 5 共 T2兲2 4共 T2兲
⫽⫹ ⫺ ⫺ . 共D5兲 sity Press, Cambridge, 1967兲.
d 9 3 19
S. Tomotika, Proc. R. Soc. London, Ser. A 153, 302 共1936兲.
20
E. L. Ince, Ordinary Differential Equations 共Dover, New York, 1944兲,
Now mimic the proof of Lemma 4. Assume that there exists especially Chap. 10.
X * , with 0⬍X * ⬍⬁, such that at ⫽X * , 21
G. Hellwig, Differential Operators of Mathematical Physics 共Addison–
Wesley, Palo Alto, CA, 1964兲.
T 2 共 X * 兲 ⫽0. 共D6兲 22
F. Bowman, Introduction to Bessel Functions 共Dover, New York, 1958兲.
23
M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions
If 共D6兲 occurs more than once, then let X * be the smallest 共USGPO, Washington, DC, 1964兲.
positive value at which 共D6兲 is valid. At ⫽X * , dT 2 /d 24
J. Scoria, W. K. Chiu, and M. P. Norton, ‘‘A study of unsteady laminar
⬎0 from 共D6兲. But T 2 ( )⬎0 for ⬍X * , so this plus 共D6兲 boundary layer flow on a flat plate using a smoke-wire/silhouette flow
visualization technique,’’ Exp. Therm. Fluid Sci. 3, 291 共1990兲.
assures that 25
C. M Bender and S. A. Orszag, Advanced Mathematical Methods for
Scientists and Engineers 共McGraw–Hill, New York, 1978兲.
dT 2
⭐0 at ⫽X * .
26
E. T. Whittaker and G. N. Watson, A Course in Modern Analysis 共Cam-
d bridge University Press, London, 1980兲.