You are on page 1of 27

The Rayleigh-Plateau instability of incompressible axisymmetric

bodies of fluid
Bruno Figeys & Sam Roelants
Supervisor: Marco Caldarelli
April 30, 2009
Abstract
It is a long known and easily verified fact that a steady jet of fluid, like the one coming out of
an ordinary kitchen faucet, eventually breaks up into spherical drops. We can thus conclude the
cylindrical body of fluid is somehow unstable and decays to the more stable spherical droplets. It is
exactly this instability, first researched by Plateau and by Rayleigh after him, that we will investigate
in this paper. We study the stability of static and rotating cylinders of incompressible fluid under
small perturbations in a classical framework. We also look for possible axisymmetric configurations
which extremize the system energy, again for static and rotating bodies, and relate them in phase
diagrams, showing the branches of possible stable and unstable fluid configurations.

Contents
1 Introduction

2 Boundary Conditions
2.1 The Young-Laplace equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 The complete set of linearized boundary conditions . . . . . . . . . . . . . . . . . . . . . .

4
4
6

3 Static axisymmetric bodies of fluid


3.1 Stationary configurations . . . . . . . .
3.1.1 Extremizing the action . . . . . .
3.1.2 The solutions in a phase diagram
3.2 Stability of a static cylinder in 3D . . .
3.3 The dispersion relation . . . . . . . . . .
3.4 Stability of a cylinder in d dimensions .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

7
7
7
11
12
14
16

4 Stationary rotating configurations


4.1 Minimizing the potential energy . . . .
4.2 Through the Young-Laplace equation .
4.3 Solutions . . . . . . . . . . . . . . . .
4.4 The phase diagram . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

17
17
18
18
19

.
.
.
.

5 Stability of a rotating cylinder of fluid


19
5.1 Warming up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.2 General study of the stability of the uniform rotating cylinder of fluid . . . . . . . . . . . 21
5.3 The dispersion relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6 An application to black hole physics
25
6.1 Nuclear physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7 Conclusion

25

A Appendix
A.1 cylindrical coordinate system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.2 Generalized cylindrical coordinate system . . . . . . . . . . . . . . . . . . . . . . . . . . .

I
I
I

Introduction

This article is the result of an exploration of the instability of cylinders. The cylinders and all other
axisymmetric objects discussed in this paper all have a periodicity or length L. Later on we shall encounter
some examples showing how this can be achieved.
In the first section of this paper we will discuss the lesser known surface tension energy and the related
Young-Laplace equation. In this first part the Young-Laplace equation will also be situated in the context
it will used in in this article, i.e. as one of the frequently used boundary conditions. In the second part
the discussion we will discuss static axisymmetric bodies. We show how we obtain the equation that
describes these bodies and put them in a phase diagram. Next we take perhaps the most familiar of
these shapes, the cylinder, and study its stability. The third and the fourth part are the extension of the
second part, but this time the axisymmetric bodies are undergoing a rigid body rotation. This makes it
all a lot harder to calculate, and still some open questions remain unanswered.
Despite the fact that these capillarity effects are a less widely known subject in physics it already
has some history. The breakup of a steady jet of fluid into smaller droplets is a well known and quite
well understood phenomenon. It was Plateau who first ascribed this effect to the capillarity of the fluid.
After analysis, he concluded that a cylindrical body of fluid of radius R is unstable to all perturbational
modes of wavelength > 2R. He thus (wrongly) assumed that after the breakup, the drops ought to
be equally spaced with an intermediate distance of 2R. It was Lord Rayleigh who correctly argued
that the intermediate distance is in fact determined not by the marginally unstable mode, but instead
in the maximally unstable mode. Indeed, it is the perturbational mode which grows the fastest which
will determine the eventual spacing between the droplets, which Rayleigh found to be a fraction 0.697 of
Plateaus initial guess. We will come across these results and many more in the course of this paper.
The Rayleigh-Plateau instability of fluids is a capillarity effect, and thus caused by the surface tension
of the fluid. The tensile properties of fluids are characterized by a parameter , called the surface energy
or surface tension and which has dimensions of energy per unit surface or force per unit length. We
can thus write down the potential energy of a fluid configuration due to the surface tension as U = A,
where A is of course the surface of the body of fluid.
In a simple calculation, we can verify that when considering solely surface tension as a means of
cohesion, the cylinder is an unstable configuration, in the sense that certain modes will decrease the
area of the body and thus result in a configuration of lesser energy. Suppose we start from a cylinder of
incompressible fluid of radius R0 and length L, aligned along the z-axis. Its volume V and area A are
naturally given by
V = R02 L, A = 2R0 L.
We can then perturb the radius of the cylinder to
R(z) = R0 + R1 cos(kz) + 2 R2
We will calculate the difference is surface A up to O(2 ). As we are dealing with a fluid of constant
density, here, as in the rest of this paper, we need to keep the volume of fluid constant. This is why the

Figure 1: A typical example of the Rayleigh-Plateau instability on a jet of water. This image was borrowed
from [7]

second order change 2 R2 is important


R to compensate for the first order change. The volume of a solid
of revolution is well known to be r2 dz, by which we easily find
1
Vpert = R02 L + R12 L2 + 2R0 R2 L2 + O(3 )
2

1
V = 2 L R12 + 4R0 R2 .
2

From this, since V = 0 we immediately find R2 to be


R2 =

R12
4R0

Next, we calculate the Rperturbed


area Apert and the net change in area A. The area of a solid of

revolution is simply 2 r 1 + r 2 dz, where r denotes differentiation with respect to z. A rather quick
calculation immediately yields

1 2 2
k R1 R0 + R2 + O(3 ).
Apert = 2R0 L 22 L
4
Upon plugging in our value for R2 , we immediately find A to be

A = 2R2 L2 k 2 R02 1 .

Indeed, we find that A < 0 when kR0 < 1, or equivalently, when the mode wavelength is greater than
the circumference 2R0 , exactly as Plateau predicted. Intuitively, one can say that longer cylinders are
more unstable than shorter ones, in the sense that they are unstable under a wider range of perturbation
modes. Later in this paper we will linearize the hydrodynamic equations. It should then be clear that
we do not need to study any other perturbations to the fluid than purely harmonic modes, since we can
always expand a more general perturbation into its Fourier series.
Thus, we have established that the cylinder is indeed an unstable configuration, which will eventually
decay into a body of lesser surface energy, like spheres. It is then possible to study the instability of the
cylinder, which is exactly what we will do in this paper. The relevance this research extends far beyond
classical hydrodynamics and has applications in an almost inconceivable range of physical domains. A
few of these domains include nuclear physics, magnetohydrodynamics and plasma physics, astrophysics
and string theory. We will come back to a few of these applications later on in the text.

Boundary Conditions

As is most common in fluid dynamics, throughout this paper, we will turn to the Navier-Stokes equations
in order to study the behaviour of the fluid in question. Throughout this text we will resort to a model
using incompressible, inviscid fluids. More succinctly, we will resort to the continuity equation that
expresses local conservation of mass and the Euler equation for ideal inviscid fluids which paraphrases
conservation of momentum. These equations are given by

= (~v )
t
1
d~v
= p,
dt

(mass)

(1)

(momentum)

(2)

where the d/dt operater denotes the convective derivative. As with any set of differential equations one
wishes to solve, the appropriate boundary conditions are needed to arrive at an actual result. In this
section we look precisely for the boundary conditions to supplement our equations.

2.1

The Young-Laplace equation

Of all the hydrodynamical equations and identities that will be used throughout this work, the next
one proves to be paramount to the entire discussion. We will now discuss the Young-Laplace equation,
4

which relates the discontinuity of the pressure p at a fluid interface due to surface tension to the local
shape of the surface. The Young-Laplace equation is perhaps not as famous as many of the other notable
equations in fluid dynamics, but it is none the less indispensible when considering fluid capillarity.
The exact statement of the Young-Laplace equation is
p = n
,

(3)

with again being the surface energy, and n


denoting the unit vector normal to the surface pointing from
region 1 to region 2. p is then the pressure in region 1 minus the pressure in region 2 on the boundary.
We will not attempt to derive this equation here, but gladly defer the reader to more specialised literature
such as [11].
The nature of this equation is a purely geometrical one and will, more specifically, require some
differential geometry in order to be evaluated correctly. For the readers convenience, we added an
appendix to the end of this paper with some useful relations obtained from differential geometry.
It is also generally known from differential geometry that, if we denote by the mean curvature of
the surface, n
= 2, so that the Young-Laplace equation may equivalently be written as
p = 2,
or alternatively, using the pricipal radii of curvatures R1 and R2 ,

1
1
.
+
p =
R1
R2
This is all just side information, as all of the relevant physics is contained in (3), which is the form we
will almost exclusively use. Note however, that the equation in terms of the mean curvature allows for a
more transparent interpretation than the original formulation. A pressure gradient proportional to the
local curvature of the fluid arises due to the surface tension. This pressure gradient then exerts a force
that attempts to restore the fluid to a smoother configuration. As a consequence, we are unlikely to find
fluids with sharp edges or corners, but will almost always observe configurations with smooth surfaces,
such as drops, planes and cylinders.
The profile of a fluid surface may be parametrized by a function z = h(r, ) in cylindrical coordinates
(see A.1). It proves to be useful, however, to define the function f (r, , z) = z h(r, ), so that the
surface of the fluid is nothing but the level surface f (r, , z) = 0. In this context, we easily find the unit
normal n
to the surface to be f /||f ||, evaluated at f = 0. It is then straightforward, but perhaps
a bit tedious, to obtain an expression for n
in terms of the parametrization h(r, ). We will simply
provide our result (use (A.1) and (A.2)), and to lighten the notation, we will denote differentiation with
respect to a variable by a subscript.

h2
1
2
1
+ 2 (h + rhr ) 1 + h2r 2 h hr hr ,
(4)
n
= 3 hrr 1 + 2

r
r
r
q
where = ||f || = 1 + h2r + h2 /r2
In this text, we will keep only to axisymmetric bodies, so the parametrization h does not depend on .
We can then simply drop all terms in this rather frightful expression containing -derivatives, i.e. h ,
hr and h .
We can, however, choose to define our surface in a different way. For example, through a function
g = r R(z, ). In this case, we can again compute n
. We find

!
!
2
2
2R Rz Rz
R
R
1
R
2
2
n
= 3
,
(5)
+ 3 Rzz 1 + 2 2 1 + Rz +

r
r
r
r
r2

with still defined as ||g||. Again, the equation simplifies significantly if we stick to axisymmetric
profiles. We will, in fact, use both parametrizations and both expressions in the future.
It is interesting to note that the Young-Laplace equation results from imposing the condition that
the surface is in equilibrium, which is realised by balancing the normal stresses to the surface. This is
completely equivalent to imposing that the energy of the fluid configuration be at a minimum. Hence,
we will later show that minimizing the surface area (and thus the potential energy due to the surface
tension) by use of the Euler-Lagrange equation is completely equivalent to the Young-Laplace equation.
5

2.2

The complete set of linearized boundary conditions

The Young-Laplace equation already gives us one condition that must hold at the fluid boundary. In the
next paragraph we will find another.
A second boundary condition
In this section, we consider a slightly different situation, in view of the stability analysis of the cylinder
we will conduct later. This time we start from a fluid surface to which we have added a small perturbation.
We can then again parametrize the fluid surface as a level surface f (r, , z, t) = 0. The boundary condition
then follows from observing that fluid particles at the boundary remain at the boundary. Expressed in
terms of the Lagrangian derivative, this gives us
df
=0
dt
We can rewrite this as

if f = 0.

f
v f
f
f
+ vr
+
+ vz
= 0 if f = 0.
t
r
r
z

(6)

This is the second boundary condition we were looking for. Next, we will attempt to exploit the fact
that the perturbations we will study are very small to significantly simplify the boundary conditions, as
well as the rest of the calculations.
Linearizing the equations
Suppose our perturbed surface is defined by a function r = R(z, , t), or equivalently by a level surface
f = r R(z, , t) = 0. If the perturbation is small enough ( O()), we can neglect any contributions
to our equations of quadratic order in or higher. In the context of a perturbed cylinder, we can for
example set R = R0 +R(z, , t), where R0 is the radius of the cylinder, and R is a perturbation of order
. This will induce a change in the velocity ~v = ~v0 + ~v , where again ~v0 is the initial velocity field of the
fluid, and ~v is the induced perturbation, again of order . We can then for example linearize our second
boundary condition (6) by neglecting the products of any two or more pertubations. Remembering that
f = r R(z, , t), we obtain

v R
R
R
+ vr
vz
= 0 at r = R(z, , t).
t
r
z

Wich results in the linearised equation


R
R v0, R
+
+ v0,z
= vr
t
r
z

at r = R(z, , t)

(7)

Another important simplification is found when we expand vr around R0 in a Taylor expansion. We


find that
vr
vr (R, , z, t) = vr (R0 , , z, t) + R
+ ,
r
meaning, as long as v0,r does not depend on r, after linearization only the leading term survives. Thus,
evaluating the boundary condition at r = R in the linear approximation becomes nothing but evaluating
at r = R0 .
We can also linearize the found expressions for n
. Since we are discussing perturbations on a
cylinder parametrized by f = r R(z, , t) = 0, the more natural expression to use is (5). Since R0
is constant, Ri = Ri . The derivatives of R have no zeroth order contribution. We can therefore
neglect any products of derivatives of R, and keep only terms linear in Ri . One easily shows, for example
through a binomial expansion, that = 1 + O(2 ). What remains is
n
=

R
1
Rzz
.
r
r2

This expression is to be evaluated at the surface r = R:


n
|R =

1
R
Rzz
.
R
R2
6

A binomial expansion tells us that R1 = R01 RR02 + O(2 ) and R2 = R02 + O(2 ). The YoungLaplace boundary condition is thus linearized to the following easier result:

R
R
1
.
2 Rzz
p =
R0
R0
R02
To sum up the work done in this section, we formulate here the two linearized boundary conditions we
have derived, which we will use several times throughout the text. We find that for small perturbations
of order , the fluid must satisfy to linear approximation the following two equations:

R
R
1
2 Rzz
p =
R0
R0
R02
R v0, R
R
vr =
+
+ v0,z
t
R0
z

(8)
(9)

Static axisymmetric bodies of fluid

With the above machinery we will first study static bodies (~v0 = 0). First we derive configurations
of fluid with an extremal potential energy. After that we will study the stability of the cylinder, in 3
dimesions. As for more advanced physical topics in high-energy physics a more dimensional variant is
required, we will perform a less complete, but still worthy, analysis without confining ourselves to the
three dimensional space. The necessary expressions derived up until now have very simple generalizations
to d dimensions by using appendix A.2.

3.1

Stationary configurations

In this section we search for the configurations of fluid with a minimal energy in the form of surface
tension. There would be an easy way to drop this surface tension and thats by decreasing the number
of particles or, equivalently, decreasing the volume. This is of course a rather trivial way of decreasing
the surface area, and is not quite what were looking for. Given a constant volume, we want to know
which configurations have an extremal energy. When we find the shapes that extremize the energy, we
still have no way of telling if it actually minimizes the energy, or if the shape is simply a saddle point.
In the latter case, the configuration will be a stationary solution, but not at all a stable one.
3.1.1

Extremizing the action

The extremization problem at hand can easily be solved by the use of a Lagrange multiplier. The action
to be minimized is
I = A V.
with the Lagrange multiplier, A the area of the body and V the volume. We wont search for every
kind of configuration, but only for axisymmetric configurations. Using the coordinates defined in (A.5),
the fluid body is defined by the set of points satisfying
0 < r < R(z),

0 < z < L,

(10)

0 < i < , for i = 1 . . . d 3


0 < d2 < 2.
We must now extremize the action for such a parametrization of a body P :
Z
Z
p
rd2 drdzdd2 ,
I[P] =
rd2 1 + r2 dd2 dz
P

where r denotes differentiation with respect to z. These integrals are nothing but the generalisation of
the surface and volume of a solid of revolution to our new coordinate system. Since the integrands do
not depend on the angular coordinates, the integration over d2 simply yields a constant factor, the
7

r 2

r+

Figure 2: The behaviour of r2 as a function of r. The zeros correspond to maximal and minimal values of r
for the surface.

surface of a unit (d 2)-sphere Sd2 . Since this overall factor does not matter in our optimization, we
discard it. What remains is a functional which can be written more compactly as
I[P] =
with

Ldz

d1
r
.
d1
One way to extremize this action is by using the well known Euler-Lagrange equation

dL
d dL
= 0,

dr
dz dr
L(r, r , z) rd2

p
1 + r2

(11)

which yields a second order differential equation. A more clever trick however, is to use our knowledge
of Lagrangian and Hamiltionian mechanics (or, more specifically, knowledge of the Legendre transform).
We compare L with the Lagrangian in classical mechanics and observe that L
z = 0, there is no zdependence. If z were seen as the time in classical mechanics this would mean that the Hamiltonian is
a conserved quantity. Thus, we have that the Legendre transform of our Lagrangian with respect to r
must also be conserved.
L
H = r L = c
r
Plugging in our Lagrangian, we obtain the following first order differential equation:
r2 =

(d 1)2 r2(d2)
1.
2 (rd1 + c)2

(12)

This equation carries several parameter freedoms (c and ). It describes all the shapes that extremize
the surface, among which are the shapes of minimal surface area, which are the ones we are especially
interested in. One may argue that c < 0 results in physically unrelevant solutions, so we will not discuss
them here. We will, however, briefly discuss the several other possible outcomes. Note that there is
another way to look at the things than proposed here. You could rewrite formula (12) as
1 2
r + V (r) = 0.
2
This way you could use your knowledge in classical mechanics to better understand this problem. Figures
2 and 4 can then nearly be seen as the graphs of V (r), they just need to be flipped over.

L
2

0
L2

L
2

0
z

Figure 3: A plot of several unduloid shapes. The limiting profiles of a sphere and a cylinder are also plotted for
reference.

Spheres Should we choose c = 0, an analytical solution is available to us. The differential equation
describing the shape of the fluid is
r
dr
(d 1)2 1
=
1.
dz
2
r2
Rewriting this equation, we obtain

which is readily integrated to

dz = q

rdr
(d1)2
2

,
r2

(d 1)2
r2 .
2
We have discarded the integration constant that determines the z-offset, since it is of little importance
to us. Squaring this expression, we obtain the equation of a d-dimensional sphere:
z=

r2 + z 2 =

d1

This is of course something we had at least hoped for: spherical shapes minimize (hence extremize) the
surface area for a given volume. It is therefore not very surprising that we see the jet coming out of the
kitchen faucet contract into spherical drops. We can check the mean curvature of this shape R11 + R21
with R1 and R2 the principal radii of curvature of the shape. In this case, both R1 and R2 are simply
2/, so that the mean curvature is precisely .
Cylinders There is another analytical solution. This is the one with constant radius, r = 0. In this
case , c and R must be taken so that r = 0.

d1
d2
R
c=R

This is what one can see as the generalization of a cylinder to more dimensions. We will call them the
cylinders in d dimensions.
Unduloids Before giving the actual numerical solutions to the differential equation, we can take a quick
glance at the graph of r to learn quite a bit about the solutions. This will also allow for a first check if
the found numerical solutions are at all plausible. In Figure 2 we have graphed r2 . We immediately find
9

r 2

Figure 4: Demonstrating the behaviour of r2 as we vary c for fixed . When c = 1/, the curve has only one
zero, corresponding to a solution with constant radius: the cylinder.

that r2 , so r as well, has 2 zeros, marked by r . Indeed, since these are radii for which r = 0, these
must be extremal values for r(z). It is also obvious there can be no values of r outside the zone marked
by r , since r2 becomes negative in this area. These solutions are already known to mathematicians,
and are called unduloids. It is obvious that the cylinder and the sphere are just limiting cases of the
unduloids. To understand this it is useful to understand Figure 4, in which we have plotted r2 for a few
combinations of c and to show the general behaviour. More precisely, we have chosen to be constant
and allow c to vary. It should be clear that depending on c, there can be two, one, or zero intersections
with the r-axis. For c very large there are no such points, consequently there cannot be any physical
stationary configuration. For the limiting case of c = ccrit there is just one such point. There is just one
possible value of r and that is r+ = r . This solution represents the cylinder because we see that for
this value r = 0. For 0 < c < ccrit there are two such points, these values of c represent the unduloids
we did not know yet. For the critical value c = 0 we see that the curve is not like the ones before. This
is because this limiting case is the sphere for which r should be zero and r = at r = 0. The
cylinder is the unduloid corresponding with r+ = r , the lowest curve in graph 4. Negative values for c
do not give physical realistic configurations, we will explain why in a later paragraph. We plotted a few
examples of unduloids in Figure 3, along with the limiting cases of the sphere (c 0) and the cylinder
(c 2 , in 3D).
Constant mean curvature and the Lagrange multiplier Another way to derive the differential
equation found by Euler-Lagrange is by using Young-Laplaces. With the NSE we can calculate, we will
do this later in more detail, that the pressure in the fluid must be constant everywhere in the non-rotating
case. With Young-Laplaces equation we can then formulate a very important statement. Looking back
to the equation (3), proves that as the pressure is constant in the fluid and outside;
The mean curvature of the surface ( n
) must be constant.
There exists a cute way to interpret the Lagrange multiplier we have introduced. We begin by
plugging the Lagrangian (11) in the Euler-Lagrange equation. Rewriting, one quickly obtains a second
order differential equation:

d2

(1 + r2 ) r =
r
( 1 + r2 )3
We can then start from an entirely different, seemingly unrelated angle: We now know that n
is
constant. Call this constant quantity . We can evaluate n
using the generalised formula (A.9). Since
we are considering axisymmetric shapes, all -derivatives drop out, and we are left with a differential
equation

d2
2

n
==
(1 + r ) r
r
( 1 + r2 )3
We find the exact same differential equation! By direct comparison we find that the Lagrange multiplier
we introduced is up to a proportionality nothing but the mean curvature n
of the shape!
10

L
2

0
L2

L
2

0
z

Figure 5: A plot of a solution corresponding to c < 0. The shape has a flex point on either side. There tangent
lines are drawn for reference.

Spheres
Cylinders
Unduloids

5.4

A/V 2/3

5.2
5
4.8
4.6
0.1

0.2

0.3

0.4

0.5

V /L

Figure 6: The phase diagram of the normalized area versus the normalized volume for axisymmetric stationary
solutions. All bodies are periodic with periodicity L. The values for the unduloid branch were obtained
numerically using MatlabTM .

Another class of solutions With the previous result we can give an argument why shapes found
with c < 0 are no realistic configurations. In Figure 5 we can see that there is a flex point. We wish
to calculate the mean curvature in this point. To do this, it is easier to parametrize the surface as
z h(r) = 0. In the flex point h = h = 0, thus n
= 0 with equation (4). The previous paragraph
proves then that the mean curvature is zero everywhere and the pressure inside the fluid is the same as
outside, zero. Which means that there can be no fluid inside the defined surface. These kind of shapes
can therefore not be good configurations for fluid bodies. Unlike the soap bubbles, since these adopt the
form of minimal surfaces. Which are exactly those with mean curvature zero.
3.1.2

The solutions in a phase diagram

It is interesting to plot the several possible solutions in a phase diagram showing the surface of energy
versus the volume, where we will stick to regular three dimensional bodies. Such a plot can be found
in Figure 6. The area A is normalized by a factor V 2/3 in order to improve the visibility of the details
of the plot. The different branches represent different solutions, all of periodicity L. The sphere-branch
and cylinder-branch are easily obtained analytically, but for the unduloid branch, one has to resort to
numerical computation of the volume and surface of several solutions. At the far left, the cylinder branch
11

Cylinders
Unduloids

A/V 2/3

A/V 2/3
(4)2/3

Spheres
Unduloids
(36)1/3

6
3

1
4

V /L3

V /L

Figure 7: A closer look of the branch point of the


cylinder and unduloid branch. The branch
point is situated exactly at the marginally
unstable cylinders with 2R0 = L.

Figure 8: A closer look at the merger point. One can


easily show this point is where the spheres
no longer fit in our periodicity of L

consists of the stable cylinders. This may seem strange, since we have already shown that all cylinders
are stable to certain wavelengths, depending on their radius. The only reason these are stable is because
of the periodicity we imposed on them. This makes it impossible for harmonic modes of > L to exist on
the cylinder. Since we know that cylinders are unstable to modes with < 2R0 , we find precisely the
cylinders for which 2R0 < L are indeed stable. A simple calculation allows us to situate the marginally
unstable cylinder for which 2R0 = L on the phase diagram. We find that its volume is L3 /4, with a
normalized area of (4)2/3 . The marginally unstable cylinder is situated in Figure 7. It is indeed precisely
at the bifurcation point between the cylinders and the unduloids. One may also examine the point where
the unduloid branch and the sphere branch merge. This point is named the merger point. Intuitively
we may anticipate there is something special about the sphere located at the merger point. Indeed, a
simple calculation reveals that the merger point is precisely the sphere that has a diameter equal to the
periodicity L, which means the spheres of volume past the merger point are forbidden, since they do not
fit in our periodicity. We find that the volume of a sphere of diameter L is L3 /6. The merger point is
looked closer upon in Figure 8. After the next section, when we have obtained the dispersion relation for
perturbations on a cylinder, we will look closer upon these branch points. It will allow us for example
to predict where we may expect such branch points, and also give an indication of how the new solution
will look.

3.2

Stability of a static cylinder in 3D

In this part we discuss the stability of the non-rotating cylinder in 3 dimensions. To study its stability we
will simply put a perturbation on the cylinder and study how this perturbation will evolve. We will use
linearized differential equations, this will allow us to decompose the perturbation into a Fourier series.
Which offers us the possibility to calculate the time evolution of each term. The problem is thus reduced
into studying the evolution of a perturbation of the form ei(kz+m) .
The unperturbed cylinder
In the unperturbed case, a cylinder of incompressible fluid with length L and radius R0 is at rest. It
is shown in the previous section that this is a stationary solution. The fluids behaviour is determined
completely by the continuity equation and the Euler equation, while our boundary condition is supplied
by the Young-Laplace equation as discussed in section 2. Since ~v0 = 0 and is constant the continuity
equation (1) holds. The NSE reduces to p0 = 0, which means that the pressure p0 must be constant
in the fluid. The boundary condition tells us what the value of p0 is. If we assume that the pressure
outside is zero, then the pressure jump of the Young-Laplace equation is equal to the pressure inside the

12

cylinder.
p0 = n

The boundary is defined by r = R0 , or again by f = 0, where f (r, , z) = r R0 . Formula (5) tells us


that n
= R10 . The equations for the unperturbed cylinder are

p0 = R10
.
~v0 = 0

The perturbed cylinder in 3D


Now we want to explore what happens when we perturb the static cylinder through

r = R0 + R
p = p0 + p .

~v = ~v0 + ~v

(13)

These perturbations are all of order . The continuity equation and NSE reduce to

~v = 0
.
v
~
v ~v = p
t + ~

Linearising the second equation kills the second term. After taking the divergence of both sides, the left
side vanishes completely. We are left with the Laplace equation for the perturbation on the pressure:
2 p = 0.

(14)

As we mentioned before, because of the linearity of the equations, we can construct an arbitrary
perturbation out of simple harmonic functions. This means we only have to study the behaviour of
perturbations ei(kz+m) . We do so, and assume our perturbations are of the following form:

r = R0 + Aei(kz+m)+t
p(r, , z, t) = p0 + P (r)ei(kz+m)+t .

~ (r)ei(kz+m)+t
~v (r, , z, t) = ~v0 + V

We can plug this form for p in our Laplace equation (A.3) and find:

m2
2
2
P (r) k + 2 P (r) = 0
r

m2
1
2

P + P k + 2 P = 0.
r
r

The resulting equation is a modified Bessel equation. This equation has long been extensively studied,
and its solutions are generally known. We find that the solution is of the form [1]
P BIm (kr) + CKm (kr).
We have plotted a few of the modified Bessel functions in Figure 9 for reference. As we know that
the pressure must stay finite in the center, we discard those solutions that are singular at r = 0 (being
the Modified Bessel functions of the second kind, Kn ). Thus we find the solution for the pressure to be
P (r) = BIm (kr).
Now lets see what we can do about our boundary conditions. The boundary conditions are given by
p(R0 + R) = n
|
R0
dR
vr (R0 + R) =
dt
R0

13

4
I0 (x)
I1 (x)
I2 (x)
I3 (x)

K0 (x)
K1 (x)
K2 (x)
K3 (x)

1
0

0
0

(a) modified Bessel functions of the first kind

(b) modified Bessel functions of the second kind

Figure 9: Plots for the first few modified Bessel Functions

For the Young-Laplace equation we refer back to equation (5). Linearizing both boundary conditions,
we get

m2 1
1
2
+ k R +
R
p0 + p(R0 ) =
R0
R02
vr (R0 ) = R.

After plugging in our exponential forms of the perturbations, we get:

m2 1
A
P (R0 ) = k 2 +
R02
Vr (R0 ) = A

The Navier-Stokes equation gives us that Vr (r) = P (r), so that we get a set of equations

2 2
k R0 + m2 1 A Im (kR0 )B = 0
2
R0
1

A +
(kIm
(kR0 )) B = 0

(15)
(16)

If this system of linear homogeneous equations is to have non-trivial solutions for A and B, its determinant
must be zero. This results in the dispersion relation
2 =

3.3

kR0 Im
2 2
(kR0 )
2
k
R
+
m

1
0
3
R0
Im (kR0 )

(17)

The dispersion relation

Finally, we have come to the point where we can study the actual stability of the cylinder. The dispersion
relation gives the rate at which any perturbational mode evolves. It should be clear that when the right
hand sign turns negative, will be purely imaginary and the perturbation we put on the cylinder
will oscillate stably. The result of the perturbations are simply waves that will run away along the
cylinder surface. We are thus especially interested in the part where 2 (k) 0 to study the rates at
which different modes grow exponentially. In Figure 10 we have plotted the real part of the dispersion
relation with respect to the dimensionless frequency and wavenumber for an axisymmetric perturbation
(m = 0). Upon inspection, we find that the modified Bessel functions Im and their derivatives are
always positive (as one can anticipate from Figure 9). Thus, the sign change is completely determined
by the factor k 2 R02 + m2 1, which we have dubbed the discriminant factor, for obvious reasons. It is
immediately clear that for m 1, the discriminant remains positive for all k. This result has a very
14

R03

0.4

0.3

0.2

0.1

0
0

0.697

R0 k

Figure 10: A plot of the dispersion relation in 3 dimensions. We have chosen to plot dimensionless quantities.
The maximum of the curve lies at precise kR0 = 0.697, as Rayleigh predicted. The curve changes
sign in kR0 = 1, as Plateau predicted.

natural interpretation. Should we expand our perturbation in angular components eim , we find that
the only components that contribute to the break up are the m = 0 modes, since the m 6= 0 are stable
modes. Thus the angular variations of the perturbation do not matter and will not contribute to the
breakup: instability is completely determined by the z-variations of the perturbation. In the m = 0 case,
then, the discriminant changes sign in exactly kR0 = 1. This is indeed the stability condition Plateau
determined, as well as the result of our little toy calculation in the introduction!
The mode for which kR0 = 1, the marginally stable mode, is one of special importance. It is
immediately clear from Figure 10 that this mode does not depend on time. In the linear approximation,
this mode, also called the threshold mode, brings the cylinder into a new stationary configuration. It is
of course no coincidence that the linear theory predicts a new kind of stationary solutions precisely at
the point in the phase diagram where the unduloid branch of solutions forks off. Indeed, this new kind of
stationary solutions predicted by the linear theory is precisely the zeroth and first order approximation of
the unduloid branch in its Fourier series! This may sound like an obvious statement, but its importance
should not be underestimated. Without any knowledge of possible stationary solutions, we were are able
to predict branch points on the phase diagram. More generally, no matter what the situation is, when a
shape flips over from a stable to unstable region, the dispersion relation flips sign. And thus, there exists
a time-independant threshold mode where (k) = 0, which indicates a new branch of solutions. We will
use this later on to predict branch points on another phase plot.
Deriving the disperion relation allows us to locate the maximum of the curve in Figure 10. We find
it to be at kR0 0.697. Again we retrieve a known result, as this is precisely Rayleighs correction
to Plateaus hypothesis we have mentioned at the beginning of this text. Upon perturbing the system,
we find that the mode for which kR0 grows exponentially faster than all other modes. It is therefore
natural to assume that the intermediate distance between two drops of a contracted jet corresponds to
the wavelength of this mode. We find that this wavelength is simply 2R0 /0.697 = 9.01R0 . Please note
that we cannot make exact predictions using the disperion relation, or any other part of our model for
that matter, since we are only studying the phenomena up to a level where the linearization is plausible.
Since the perturbations on an unstable body will grow exponentially, we quickly leave the linear region
and are forced to include non-linear terms, which complicates the calculations tremendously. We have,
however, gotten quite a few results from the dispersion relation, most of which are very relevant and
experimentally measurable. Just to show how our predictions are pretty good, we have included a figure
taken from [5]. The experimental data included in Figure 11 was taken from the the work of Donnelly and
Glaberson, who did a great deal of measurements concerning the capillary break up of a jet of water [6].
We see that the experimental results stick to the theoretical result quite nicely. Besides what we have
discussed here, there resides a great deal more of information in the dispersion relation, which makes it
a key result of this work.

15

Figure 11: The dispersion relation with the experimental data of Donelly & Glaberson (1966) plotted onto it.
Taken from [5].

3.4

Stability of a cylinder in d dimensions

Unfortunately, doing the entire calculation up until now in d dimensions proved to be quite difficult.
This is partly due to our lack of experience in calculations in higher dimensions, using exotic coordinate
systems with their own distinct exotic geometries. We have, however, performed the calculation in d
dimensions when supposing perturbations which do not depend on any angular coordinate. This means
we only consider m = 0 modes. We will not repeat the entire calculation, as there is hardly any
difference with the calculation done in the previous section. One needs only to look up the d dimensional
generalizations of the used formulas, which are all present in the appendix. We will simply state the
results here.
After performing an identical calculation, one obtains a slightly different modified Bessel equation for
the pressure:
d2
P k2 P = 0
P +
r
To which the solutions are simply
P = Br I (kR0 ),
with defined als = (d 3)/2. Next we find the linearized boundary conditions:

d2
A
P (R0 ) = k 2
R02
P (R0 ) = 2 A.

In these equations, A represents the perturbation strength. Plugging our solution for P in these boundary
conditions, and equation the determinant of the system to 0 as usual, we obtain the dispersion relation:
2 =

I (kR0 ) + kR0 I (kR0 )


2 2
k R0 (d 2)
.
3
R0
I (kR0 )

(18)

We immediately see that in the d = 3 case, = 0 and we retrieve the dispersion relation (17). We note
that the threshold wavenumber is kthres = d 2/R0 .

16

Stationary rotating configurations

In this section we search for axisymmetric stationary configurations of a rotating fluid in 3D. There are
two ways to find such a configuration. One is by minimizing the (potential) energy, the other is by using
the Young-Laplace equation on the boundary. We suppose the fluid is flowing like ~v = r.
The reason
for this is to be found in viscosity. If the fluid were a little viscous, any configuration with shear flows
could never be a steady flow. The only rotating flow that doesnt have such shear flows is the flow of a
rigid body rotation.

4.1

Minimizing the potential energy

We are again facing an optimization problem with a constraint. We want to extremize the energy, keeping
the number of particles constant. Because we are studying incompressible fluids, the density must be
constant and so must be the volume. We can introduce a Lagrange multiplier and define the action to
be extremized as
I := U + Urot V.
We know that U = A, but we still need to find an expression for Urot .
Therefore we first derive an expression for a point mass. If we choose the potential to be zero on the
rotation axis then we get
Z r
F~ d~r.
U (r) =
0

Here F~ represents the force that moves a point mass from the center to a radius r. From the point of
view of the particles in motion, they are being pushed by a fictional centrifugal force F~ = mr2 r. This
results in the following effective potential for a point mass
1
U (r) = mr2 2
2
To get the potential energy, we just need to integrate last equation over the entire fluid
Z
1
Urot = 2
r2 dm
2
M
For symmetry reasons we can study only one half of the configuration, so that we can define a function
h(r) which defines the boundary as the values where f = 0 with f (r, z, ) = z h(r). The action becomes
Z
Z
Z
1
I[P] =
r2 dV
dA 2
dV
2
P
P
P

Z
Z
p
1 2 2
2

I[P] =
r + 2rh(r)dr
2r 1 + h dr
2
r
r
Just as in the static case, we can write this action in terms of a Lagrangian:
Z
I = L(h, h , r)dr,

where L is given by

L(h, h , r) := 2r

1+

h 2

1 2 2
r + 2rh.
2

To extremize this action we again resort to the Euler-Lagrange equation.

d dL
dL
=0

dh
dr dh
Filling in this form of L and rewriting, we obtain
1

+ 2 r2 = 3
2

17

h
1 + h2 + h
r

L
2
L
2

L
2

L
2

(a) Shapes of a few rotating drops.

4.2

(b) Shapes of a few rotating non-uniform


cylinders.

Through the Young-Laplace equation

The same equation can be derived by using the Young-Laplace equation on the boundary. For starters
we find the pressure jump simply as the pressure in the fluid, which we can find through the NSE.
r2 r = p
Which means that the pressure must vary as
1
p(r) = pc + r2 2 .
2

(19)

The Young-Laplace equation tells us to equate this pressure to the curvature of the surface, given
by the divergence of the normal. If the boundary is defined as the points for which f = 0, with
f (r, z, ) = z h(r), we find that the divergence is given by (4):
n
=

h
h
3
r

The Young-Laplace equation becomes;


1
pc + r2 2 =
2

h
h
+ 3
r

This way we obtain the same differential equation as before. Through direct comparison we find that
= pc . So it doesnt really matter whether we extremize the energy to find the stationary surfaces, or
if we use the Young-Laplace equation, since we are basically doing the same thing. This is no different
than the choice one has when solving a mechanical problem by balancing the forces or by minimizing
the potential energy.

4.3

Solutions

Now that we have the differential equation that describes our surfaces of stationary energy, we can try
to solve it. There is a little trick to solving equation (4.2) the divergence of the normal vector can be
written in the following form:

1 2 2
rh

.
pc + r =
2
r
1 + h2
Integrating and rewriting this equation gives;
h2 =

f2
2 r2 f 2
18

6
5
4

3
2
1
0
0

0.05

0.1

Figure 12: The phase plot of the dimensionless angular momentum J versus the dimensionless rotational
velocity .

where f c + 21 pc r2 + 18 2 r4 and c is the integration constant. There are several kinds of solutions
to this equation. In the non-rotating case we had as main solutions the spheres, the cylinders and the
unduloids. In the rotating case, however, things get a bit more interesting. Depending on the boundary
conditions we impose on h and h , we get different types of solutions. We will again discuss the different
solutions briefly.
Drops It is desirable that in the 0 limit, we retrieve the solutions for the static surfaces. This
gives us reason to believe that at least one kind of solution exists where h (0) = 0 and h (R) = ,
since the we know the spheres comply to these conditions. We therefore impose these conditions and
look for numerical solutions satisfying them. These numerical solutions are plotted in Figure 12(a). We
see that as increases, a pinching effect becomes clear. There exists a critical for which the pinching
is complete, and what remains is a toroidal shape.
Non-uniform cylinders In analogy to our previous reasoning, we would also like to find solutions
that resemble the non-uniform cylinders, by imposing the boundary conditions h (Rmin ) = and
h (Rmax ) = , for some Rmin and Rmax . The uniform cylinders are then simply the solutions where
Rmin = Rmax . These solutions are plotted in Figure 12(b). Again, there is a definite pinching effect as
the angular velocity increases. Just as before, there exists a critical for which the pinching is complete.
What remains after this pinch off is a cylindrical shape surrounded by a toroidal shape.
There are, in fact, a whole lot more solutions possible. We will not begin to discuss them all. We
will discuss the drops and non-uniform cylinders in the phase diagram in the following section.

4.4

The phase diagram

This time we choose to plot the angular momentum versus the angular velocity of configurations. We do
this for configurations of fixed volume. The result can be found in Figure 12. Note that at first glance,
this plot seems plausible. For slow rotation, the angular velocity the angular momentum J = I are
indeed proportional, where the moment of inertia I is the proportionality. As increases, however, the
body starts getting deformed, and fluid mass is displaced to a greater distance from the rotation axis, as
shown in Figure 12(b). We find that I can no longer be approximated as constant, and the curve bends
over.

Stability of a rotating cylinder of fluid

The last major thing we will discuss is the stability of a uniform rotating cylinder in 3 dimensions. We
will begin by repeating the calculation in the introduction of this text. This is a quick and easy way to
19

get a first glimpse of the stability behaviour of the cylinder.

5.1

Warming up

We attempt to study the instability of a perturbed cylinder by examining how the total potential energy
changes as a consequence of the perturbation. The first time we did this computation, the energy
depended solely on the surface area. This time, however, we must also take into consideration the
effective potential energy associated with the rotational motion. The total potential energy of the motion
is therefore given by
U = U + Urot .
We start from a regular cylinder which we perturb to a new shape given by
R(, z) = R0 + R1 cos(kz + m) + 2 R2 .
Again, the incompressibility of the fluid forces us to take along a second order change to compensate for
the change in volume due to the first order perturbation. By imposing this incompressibility, we easily
obtain that
R2
R2 = 1 ,
4R0
which is indeed very similar to relation we found the first time. We find that the surface of our non
axisymmetric perturbed cylinder is
Z L Z 2 q
2 /R2 + R2 ddz
R 1 + R
A=
z
0

Dropping all the higher order terms and doing the straightforward integration just like before, we find
an expression for U :
LR12 2
U = (k 2 R02 + m2 1)
.
2R0
This is indeed the result we would expect after our discussion of the stability of static cylinders. The
most notable thing is that we have in a fairly simple manner determined the discriminant factor that
determines the stability.
Next we compute the contribution by the rotational potential energy. We have already shown in 4.1
that the potential energy of the flow associated to the rotation is precisely
Z
Z Z
1 2
1 2 L 2 4
2
r ddz.
Urot =
r dV =
2
4
0
0
There is not much to learn from the straightforward calculations, so we immediately go to the result:
Urot =

LR02 R12 2 2

Adding both results an arranging them in a nicer way, we get the total change of the energy:

2 R03 LR12 2
2 2
2
U = k R0 + m 1

2R0
This result is quite nice, since now we know exactly how the discriminant factor ought to change in the
rotating cylinder dispersion relation.
The new term that appears has a special significance and is called the rotational Bond number, but
we will often refer to it simply as the Bond number. The Bond number is a dimensionless quantity that
indicates the importance of a certain force in comparison to capillary effects. It is generally defined as
=

a
,
/L2

where a is the acceleration of the force we are comparing, and L is a general length scale of the motion.
Conventionally, the length scale is chosen to be R0 /2. [3, 10] For the moment, its more convenient for us
20

to define the rotational Bond number with the length scale L = R0 . Since the force we compare to the
capillary force is the centrifugal repulsion, we substitute R0 2 for a to obtain precisely the new term
=

2 R03
.

So the Bond number is a dimensionless quantity that directly indicates the rotational velocity of the flow
as well as its importance compared to the capillary effects.

5.2

General study of the stability of the uniform rotating cylinder of fluid

We follow the same strategy as for the non-rotating cylinder. First we solve the NSE and the continuity
equation for the rotating cylinder without a perturbation to find the zeroth order contributions. After
that we put a perturbation on it.
The unperturbed rotating cylinder
As already said before, we suppose the cylinder is rotating as a rigid body with ~v = r.
We have
already solved the NSE for this unperturbed case and found equation (19) that describes the pressure.
We also find an expression for the central pressure pc through the Young-Laplace equation. The pressure
outside is chosen to be zero again, so that we get
pc =

2 R02 .
R0
2

We have used the fact that the mean curvature n


is constant and equal to 1/R for a cylinder. This
way we find that the unperturbed cylinder is determined by the following equations

p0 (r, z, ) = pc + 12 2 r2
~v0 (r, z, ) = r.

The perturbed rotating cylinder


Just as in the static case, we put a perturbation of the form (13) on the cylinder. The linearized
continuity equation and NSE become:

~v = 0
v
~
v0 ~v + ~v ~v0 = p.
t + ~
We can again expand the perturbation in a Fourier series with terms ei(kz+m)+t . This is the same
perturbation as in the static case and we will use the same notation as we did in (3.2). Using (A.4) to
evaluate the terms in the NSE, we quickly get a set of equations that relate the velocity components to
the pressure:
V
r
+ Vr + im

r V + ikVz = 0
r
(Vr 2V ) = P
(20)
(2Vr + V ) = im

r P

Vz = ikP,

where + im. Upon solving the last three equations for the velocity components, and filling these
into the first equation, we again obtain a modified Bessel equation for the pressure P (r).

2
+ 42 2 m2
1

(21)
k + 2 P =0
P + P
r
2
r
It is interesting to note that in the non-rotating limit 0, and we recover exactly the Bessel
equation we got in Section 3.2, when d is taken to be 3. The solutions are again easily found to be
P = BIm (r),
where 2

(2 +42 ) 2
k .
2

21

Next, we impose the boundary conditions (8) and (9). The kinematic boundary condition (9) leads
us to
A = Vr .
From equations (20), one immediately finds that
Vr =

P
P
2 im

.
2
+ 42
r 2 + 4

The kinematic boundary equation becomes

(R0 ) 2 im Im (R0 )
Im
B=0
A +
+
2 + 42
R0 2 + 42

(22)

The Young-Laplace equation (8) must be handled with more care. If the pressure outside is again taken
to be zero, we get the linearized equation

m2 1
1
2
+ k +
R .
p0 (R0 + R) + p(R0 + R) =
R0
R02
The first term on the left hand side is easily calculated, since we know p0 . For the second term, a quick
Taylor expansion yields
p(R0 + R) = p(R0 ) + p (R0 )R +
So, we find that the first order contribution to p(R0 + R) is simply p(R0 ). The final form of our
linearized Young-Laplace equation is

m2 1
1
2
2
+ k +
R .
p(R0 ) + R0 R + p(R0 ) =
R0
R02
Plugging in the perturbations and dividing out the exponentials, this can be rewritten as

m2 1 2 R0
2
A Im (R0 )B = 0.

k +
R02

(23)

So, the two boundary conditions (22) and (23) are

Im
(R0 ) 2 im Im (R0 )
A +
B=0
+
2 + 42
R0 2 + 42

m2 1 2 R0
A Im (R0 )B = 0.

k2 +
R02

If these equations are to have a non-trivial solution, the determinant must again vanish. Just like before,
this provides us with the dispersion relation:
=

5.3

R03

k 2 R02 + m2 1

2 R03

R0 Im
(R0 ) + 2imIm (R0 )
2
( + 42 )Im (R0 )

(24)

The dispersion relation

This equation looks very heavy, but in the end, it isnt as bad as it looks. At first glance, we are pleased
by the fact that the discriminant factor is exactly what we anticipated from our quick computation at
the start of this section. The new term appearing is indeed precisely the Bond number describing the
rotational effects. A major feature that needs to be pointed out is the sign of this new term. The
new term introduces new modes of instability. A m = 1 mode, for example, could become unstable for
sufficiently large . A second feature of this equation is that we can no longer express as an explicit
function of k, since it appears in the argument of the Bessel functions through . We can still, however,
plot the dispersion relation for m = 0. We do this by defining, through equation (24), an implicit function
F (, k). The dispersion relation graph will simply be the level contour where F (, k) = 0.
22

R03

0.8
4
0.6

0.4

2
1

0.2
=0
0
0

R0 k

Figure 13: The dispersion relation, plotted for a m = 0 mode for different values of . The graphs are labeled
by their Bond number. The = 0 graph corresponds to the non rotating dispersion in Figure 10

The dispersion for m = 0 modes


We have plotted the dispersion function in Figure 13 for a m = 0 mode for several values of . The
different curves are labeled with their corresponding Bond number. Several features of this plot are quite
interesting to point out. First of all, the maximum of the curve shifts upward as the rotation increases.
We conclude from this that faster spinning cylinders will tend to break up faster, as their exponential
growth rate is greater. Next, we notice that the marginally stable mode shifts upward as well. The
cylinder becomes unstable to a greater range of modes. Less formally, we may summarize these results
by stating that rotating cylinders are more unstable than static cylinders, and the instability increases
with . As the maximum of the dispersion plot shifts to the right in Figure 13, max decreases. This
max determines, as already explained, the distance between two drops when a jet of fluid collapses.
Thus this distance between two distinct drops after the breakup will decrease in rotating jets.
The dispersion for m 6= 0 modes Thanks to the Bond number appearing in the discriminant factor,
there is a term that can negate the m2 term in the discriminant that was keeping all m 6= 0 modes
stable. We can then expect to find unstable modes for m 6= 0. Plotting the dispersion relation for these
modes proves to be more difficult. We can, however, obtain quite a bit of information from some basic
analytical properties of the dispersion relation. For example, we can check whether or not the = 0
and k = 0 mode is is unstable, meaning it lies on the dispersion curve. Filling in these two values into
the dispersion relation (24), we obtain an equation in and m, wich gives us a condition for which the
= 0, k = 0, m 6= 0 mode indeed exists. A quick calculation teaches us this relation is precisely when
= m + 1.
One may wonder what significance this result may have, as we have obtained the value of the dispersion
curve for only a single value. The point doesnt even seem that special, because the dispersion curves
for m = 0 also all passed through the origin of the , k-plane. These is, however, a major difference.
In the m = 0 case, the = 0, k = 0 mode was hardly interesting, because it simply corresponded to

Figure 14: An example of a new branch of solutions with broken angular symmetry. This cylinder is approximated to first order by the m = 3 threshold mode with k = 0. Image is taken from [10]

23

Figure 15: A plot of the dimensionless angular momentum versus the dimensionless angular velocity for drops
undergoing rigid body rotation. Image borrowed from [12].

zero perturbation. In this case however, a = 0, k = 0, m 6= 0 mode is a time independent perturbation


which breaks only axial symmetry, but leaves the cylinder z-invariant. Indeed, since = 0 for this mode,
the time dependence exp(t) drops out. So does the z dependence in the factor exp(ikz). This means
we have new threshold modes. Therefore, we can expect to find new branch points on the phase diagram
corresponding to families of cylinders with broken axisymmetry. An example of such a cylinder is given
in Figure 14. Since this text is mainly a study of the rotationally invariant shapes of fluid, we will not
study these new solutions in full detail. Now that we have the dispersion relation at our disposal, it
becomes so easy to make a few notes about these solutions, that it would be a waste not to.
We know that for every m there exists a threshold mode, determined by the condition = m+1. This
easily allows us to determine the bifurcation points on the fase diagram. We find that the bifurcation
frequencies are
s
(m + 1)
.
thr =
R03
It should be noted, though, that by no means do we claim that these are the only bifurcation points,
others may exist. We have simple stated here the ones we could find, that at the same time break the
rotational symmetry.
It is possible to perform a linear investigation of these branch points to determine how the new
branch will start of closely to the bifurcation point. Such a study, however, is far beyond the scope of
this text. What we can do, is take a look at a similar phase plot for rotating drops, which has long been
explored. [12] Such a plot, given in Figure 15 resembles the one we have obtained to a high degree, and
so it is a nice way to see how one might expect such bifurcations to evolve.
We indeed see a similar behaviour of the curve. Starting from the origin, we are on an axisymmetric
branch. Then we encounter a 2-lobed threshold mode, from which a new branch of 2-lobed solutions
forks. At the bifurcation point, the stability flips from stable to unstable. At every next branch point, we
encounter a new branch of solutions, each time with an extra lobe, which is of course precisely what we
would expect. These branches, however, have very peculiar features. For example, there are very specific
regions on the phase diagram where the drops are unstable only towards certain m modes, while they are
stable with respect to all other angular modes. Going into a deeper discussion about these drops would
bring us too far, so we simply defer the reader to more complete and technical articles such as [10, 12].

24

An application to black hole physics

Cylinders of fluid held together by surface tension can be used as a model for black strings. Black strings,
black holes with the shape of cylinders, exist in more than four dimensions. These objects have a long
wavelength instability under gravitation perturbations, the Gregory-Laflamme instability, that shares
many features with the Rayleigh-Plateau instability. This analogy, pointed out by Vitor Cardoso, Oscar
J. C. Dias and Leonardo Gualtieri [4, 14], is a useful tool to get some intuition of the properties of black
objects, and extends the old ideas of the membrane paradigm. This is why the higher dimensional cylinder
which we explored above is interesting. Just as the cylinders of fluid have a threshold wavelength above
which they become unstable, the black strings have one. Thus the Rayleigh-Plateau instability becomes
the hydrodynamical analog of the Gregory-Laflamme instability, and a new branch of non-uniform
black

strings emerges from the threshold mode. For large dimensions d the equation R0 kthres = d 2 for
fluids agrees with the Gregory-Laflamme instability. For comparison, see [4]. Just as in hydrodynamics,
the angular modes of the perturbation (m 6= 0 in the discussion 3.2) are stable modes for the black
strings, as was calculated by Kudoh [8]. There are a lot of other similarities, as for example a remarkable
qualitative agreement in the dispersion relations. In 3.1.2 we plotted the solutions in a phase diagram
for 3 dimensions. As summarized by Cardoso and Dias [4] this phase diagram has some fundamental
changes at about dimension 11. Where in lower dimensional spaces the spheres were the preferred shapes
of a fixed volume fluid, as can be deducted from the phase diagram in 3.1.2, this will change for more
than 11 dimensions. Above this critical dimension the cylinders will not any more be unstable in favor
of the spheres. This phenomenon also occurs in the Gregory-Laflamme instability, as discussed in the
paper of Evgeny Sorkin who first observed it [13]. A more extensive study should be done in the rotating
case. A rotating drop could at high rotation form a rotating torus, which should be the analog of a black
ring. Also these drops begin to get lobed as in Figure 15. This little summary is just a beginning of what
can be said about this analogy. Recently, this analogy between black objects and fluid lumps has been
elevated to a precise duality [9] for particular models of gravity, and the Rayleigh-Plateau instability
has been proven to be the fluid dynamical dual of the Gregory-Laflamme instability [2]. This opens the
possibility to study, using fluid dynamics techniques, the complex phase structure of higher dimensional
black holes.

Conclusion

Throughout this text we have learned that the Rayleigh-Plateau mechanism, while quite simple in its
original scheme, has analogies and similarities with many other theories, which makes it a very relevant
branch of research, not just from a hydrodynamical point of view. We have obtained the dispersion
relations for perturbations on a cylinder, which are a great source of all kinds of information and allow for
many kinds of experimental verification. Upon comparing our results with such experimental verification,
we found that the linear approximation we have utilised here describes correctly the dispersion relation
for the perturbation, although it is not enough to follow the full time evolution of the instability. We
have been able to use the linear theory to predict new branches of stationary configurations and their
properties, which is another key result obtained through the dispersion relation. We have interpreted
the phase diagrams, and plotted the stationary solutions.
However, there still remain many open questions and many possible outsets for new research. Firstly,
while this has already been done for the most part, it will prove useful to generalize the entire framework
of the Rayleigh-Plateau instability, and all its extensions, to higher dimensional spaces, as black hole
applications are situated in such spaces. One may also look further into the m 6= 0 modes for rotating
cylinders. We did not explore in depth the dispersion relation for non-axisymmetrical perturbations.
While these modes could perhaps be simply stable (excluding a discrete set of threshold modes that we
found), there is no fundamental reason to believe that this is so.
Another extension, moving a bit further from the actual Rayleigh-Plateau instability, is the research
of non-axisymmetric configurations that extremize the potential energy. The dispersion relation has
already narrowed down several of the branch points where such configurations may start, and the linear
theory can perhaps predict their behaviour very near to the branch point, but the general behaviour of
these solutions on the phase diagram and their stability must be researched further in order to complete
the full phase diagram and obtain the cylinder analog of Figure 15.

25

Appendix

In this appendix we will introduce some useful coordinate systems.

A.1

cylindrical coordinate system

The 3 dimensional cylindrical system is defined by following equations;


x = rcos
y = rsin
z=z
There are some useful formulae.
f
f
1 f
r +
z +

r
z
r
vz
1 v
1
(rvr ) +
+
~v =
r r
z
r

1
f
2 vz
1 2 v
f =
r
+
+ 2
2
r r
r
z
r 2

~ B
~ = Ar Br + A Br B + Az Br r
A
r
r

B
B
B
A
+ Ar
Br +
+ Az

+
r
r

Bz
A Bz
Bz
+ Ar
z
+
+ Az
r
r
z
f =

A.2

(A.1)
(A.2)
(A.3)

(A.4)

Generalized cylindrical coordinate system

This is a coordinate system which is the more dimensional variant of the cylindrical coordinate system
in 3D. Each point is described by a vector of d components: (r, z, 1 , . . . , d2 ).
x1 = r sin cos 1
x2 = r sin sin 1
x3 = r cos cos 2
x4 = r cos sin 2 cos 3
..
.
xd2 = r cos sin 2 cos d3

(A.5)

xd1 = r cos sin 2 sin d3


xd = z

It can be verified that ds2 = dz 2 + dr2 + r2 d2 + sin2 d21 + cos2 d2d4 . Where dd4 is a compact
notation to write the rest of the angles, which arent of any interest in this context. From this we can
derive the following expressions. (We dont give the full expressions, the rest is in this context equal to
zero. )
f
f
1 f
r +
z +
1 + . . .
r
z
r 1
1 d2 vz
r
vr +
+ ...
~v = d2
r
r
z 2
vz
f
1
rd2
+
+ ...
f = d2
r
r
r
z 2
f =

(A.6)
(A.7)
(A.8)

When writing a fluid profile as z = h(r, ), we find the divergence of the normal to be generalized to
"
!
#
2
2
R
R
d2 2
2Rz R Rz
1
R
+ 3 2 +
n
= 3 1+ 2 2
Rzz 2 2 (1 + Rzz ) +
(A.9)

r
r sin
r sin
r sin
r2 sin2
2
where = (1 + Rz2 + R
/r2 sin2 )1/2 .

References
[1] F. Bowman. Introduction to Bessel Functions. Dover Publications, New York, 1958.
[2] Marco M. Caldarelli, Oscar J. C. Dias, Roberto Emparan, and Dietmar Klemm. Black holes as
lumps of fluid. JHEP, 0904:024, 2009.
[3] S. Chandrasekhar. The stability of a rotating liquid drop. Proc. R. Soc. Lond., 286(1404):126,
1965.
[4] V. Cardoso & O. J. C Dias. Title: Rayleigh-plateau and gregory-laflamme instabilities of black
strings. Phys. Rev. Lett., 96:181601, 2006.
[5] P.G. Drazin. Introduction to Hydrodynamic Stability. Cambridge University Press, Cambridge, 2002.
[6] R.J. Donnelly & W. Glaberson. Experiments on the capillary instability of a liquid jet. Proc. R.
Soc. Lond. A, 290(1423):547556, 1966.
[7] D. F. Rutland & G. J. Jameson. A non-linear effect in the capillary instability of liquid jets. Journal
of Fluid Mechanics, 46:241265, 1971.
[8] Hideaki Kudoh. Origin of black string instability. Phys. Rev. D, 73:104034, 2006.
[9] Subhaneil Lahiri and Shiraz Minwalla. Plasmarings as dual black rings, 2007.
[10] O.A. Basaran R.E Benner Jr. and L.E. Scriven. Equilibria, stability and bifurcations of the rotating
columns of fluid subjected to planar disturbances. Proc. R. Soc. Lond. A, 433(1887):8199, 1991.
[11] P. Roura. Thermodynamic derivations of the mechanical equilibrium conditions for fluid surfaces:
Youngs and laplaces equations. Am. J. Phys, 73(12):11391146, 2005.
[12] R.A. Brown & L.E. Scriven. The shape and stability of rotating liquid drops. Proc. R. Soc. Lond.
A, 371(1746):331357, 1980.
[13] E. Sorkin. A critical dimension in the black-string phase transition. Phys. Rev. Lett., 93:031601,
2004.
[14] O. J. C Dias & L. Gualtieri V. Cardoso. The return of the membrane paradigm? black holes and
strings in the water tap. Int. J. Mod. Phys. D, 17:505, 2008.

II

You might also like