Physics of the granite sphere fountain

Jacco H. Snoeijer1 and Ko van der Weele2

1

Physics of Fluids Group and J.M. Burgers Centre for Fluid Dynamics,
University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands
2
Department of Mathematics, University of Patras, 26500 Patras, Greece
(Dated: April 3, 2014)
A striking example of levitation is encountered in the kugel fountain where a granite sphere,
sometimes weighing over a tonne, is kept aloft by a thin film of flowing water. In this paper we
explain the working principle behind this levitation. We show that the fountain can be viewed as
a giant roller bearing and thus forms a prime example of lubrication theory. It is demonstrated
how the viscosity and flow rate of the fluid determine (i) the remarkably small thickness of the film
supporting the sphere and (ii) the surprisingly long time it takes for rotations to damp out. The
theoretical results compare well with measurements on a fountain holding a granite sphere of one
meter in diameter. We close by discussing several related cases of levitation by lubrication.

I.

INTRODUCTION

Granite sphere or kugel fountains (see Fig. 1) are

a familiar sight in town squares and science museums,
and smaller ones often with a marble sphere decorate
many private homes and gardens. These fountains consist of a perfectly polished ball floating in a socket that
fits precisely around it. The fluid that wells up around
the rim of the socket is pumped into the fountain via a
hole at the base. In spite of its considerable weight, the
sphere is easily brought into a spinning motion, which
is an attractive sight especially when the surface of the
sphere is engraved with the Earths map, a soccer ball,
the night sky, or something of the kind. The fluid layer
between the socket and sphere is very thin (thinner than
a credit card1 ), which is important for any kugel on display in a public place, since it means there is no risk of
childrens fingers being caught under the spinning sphere.
Despite its popularity, the granite sphere fountain is
poorly understood by most people. When we asked visitors of the House of Science in Patras, Greece, which
physical mechanism they thought was responsible for the
floating of the sphere in front of the main entrance (a
granite ball with a diameter of 1 m), the most common answer was Archimedes law of buoyancy, as if
the sphere were an iceberg or a ship. Perhaps the visitors who gave this answer were under the impression that
the sphere was hollow. In reality, however, it is solid and
the buoyant force is by no means capable of keeping the
sphere afloat, since granite has a density 2.75 times as
large as that of water.
The second most common answer was the incompressibility of water. This is not too convincing either, since
it fails to explain why the sphere does not squeeze the
water out of the space between itself and the socket, and
simply sit on top of the inlet nozzle like a giant granite
cork.
A third answer was Pascals principle, which states
that a pressure applied to an enclosed incompressible
fluid at rest is transmitted undiminished and isotropically to every part of the fluid, as well as to the walls
of the container. This comes much closer to the truth,

FIG. 1: One of the largest granite sphere fountains in the

world, the Grand Kugel at the Science Museum of Virginia,
Richmond. The sphere has a diameter of 2.65 m and a mass
of about 27 tonnes.

as we will see, even though the water in the fountain is

neither fully enclosed (it is open at the rim of the socket)
nor at rest.
A search on the internet did not yield much in the
way of a conclusive answer. On the site of one of the
leading manufacturers of these fountains it is stated that
basic physical principles and very accurate working of
the stone allow granite objects weighing tonnes to float
on air or water,2 without however giving any hint as
to which these basic principles are. Another site, describing the Millennium Globe in Kenilworth, UK, says
that complex physics and precision engineering are
involved.3 The lemma on the electronic encyclopedia
wikipedia about the kugel ball, as the fountain is widely
known (from the German Kugel, meaning bullet or
ball), states that the sphere is supported by a very thin
film of water and because the thin film of water lubricates it, the ball spins.4 Finally, we came across several
physics forums where students asked about the working
of the kugel fountain, without however getting any answer that went much deeper than the above statements.

2
In our view, therefore, there is some reason for a paper
that explains the physics of the granite sphere fountain.
It turns out that the levitation hinges on the principle of
lubrication. The key observation is that the pressure that
builds up in the thin fluid layer, squeezed as it is between
the kugel and the socket, supplies the force required to
balance the colossal weight of the granite sphere. The
pressure integrated over the submerged area gives an upward force that equals the weight of the sphere plus the
force exerted by the atmospheric pressure on the equivalent area around the top. For the sake of clarity we neglect buoyancy and also the minor contribution to the upward force from the shear stresses at the submerged surface of the kugel. Hence, if Fg is the weight of the sphere
and Fup the resultant upward force due to the pressure
inside the fluid layer minus the atmospheric counterforce,
the balance gives:

Fg = Fup =

Z Z

[P () Patm ] cos dA,

FIG. 2: A granite wheel fountain, in which the levitated

object is a cylinder instead of a sphere. The disk in this
particular fountain has an estimated radius of R = 0.50 m
and approximate width 0.30 m, and is immersed in the fluid
to an angle max of about 35 degrees or 0.60 rad (cf. Fig. 3).

(1)

Asub

where Asub denotes the submerged area of the sphere.

The angle appearing in this relation runs from 0 at the
center of the fountains basin to max at the rim. In the
next sections we will work out Eq. (1) in detail, but we
can already make a back-of-the-envelope estimate right
here.
For a granite sphere with a diameter of 1 m, the mass
is m = 43 R3 gr = 1440 kg (with R = 0.5 m and
gr = 2750 kg/m3 ), so its weight is Fg = mg 1.4 104 N.
Given that the submerged area of such a sphere will be
approximately Asub 1.5 m2 , the average fluid overpressure (above atmospheric pressure) on this surface must
be 1.4 104 N/1.5 m2 = 0.93 104 N/m2 0.1 atm. Thus,
with the pressure at the rim of the socket (where the
water meets the surrounding air) being 1 atm, the pressure under the sphere must exceed this on average by
0.1 atm. This is the surprisingly low pressure required to make the fountain work. The excess pressure
above atmospheric pressure (0.1 atm) is usually called
gauge pressure, and the total pressure (1.1 atm) is called
absolute pressure. In the present work we must take care
to distinguish between these two. The absolute pressure
will be denoted by P , the gauge pressure by P Patm as
in Eq. (1).
In Section 2 we first turn our attention to the cylindrical version of the Kugel fountain and, performing the
above calculation in more detail, we will see that it is in
fact a beautiful exercise in lubrication theory. It may well
find its way to the classroom as part of an introductory
course in fluid dynamics. In Section 3 we analyze the
spherical fountain, which is slightly more advanced both
from a mathematical and a physical point of view. In
Section 4 we address the spinning motion of the sphere
and especially the rate with which the rotations damp
out due to the viscosity in the fluid layer. Finally, in
Section 5 we discuss the analogies between the granite
sphere fountain and other instances in which objects are

levitated by a thin fluid layer, such as water drops floating on their own vapor layer above an overheated surface
(the so-called Leidenfrost phenomenon) and also the airborne variety of the kugel fountain.
II.

CYLINDRICAL FOUNTAIN

We first consider a two-dimensional version of the fountain where the levitated object is a cylinder instead of a
sphere. This is known as the granite wheel, an example
of which is shown in Fig. 2. The analysis for the cylinder
is easier than for the sphere and therefore provides a more
direct illustration of the physical mechanism. In this section we will not include rotation yet, so the floating cylinder is supposed to be at rest. We compute the pressure
at the inlet nozzle (or equivalently, the inflow rate Qin )
required to give the fluid layer the desired thickness h of
a few tenths of a millimeter, i.e., sufficiently large for two
well-polished surfaces to not grind each other and at the
same time sufficiently small to guarantee that no fingers
(not even those of the smallest children) can get caught
between them.
A.

Physical mechanism: balance of forces

The mechanism of levitation requires an upward force

that balances the weight of the levitated object. In the
case of a cylinder of radius R and length L, this weight
is
Fg = R2 Lggr ,

(2)

where gr = 2.75 103 kg/m3 is the density of granite

and g = 9.81 m/s2 is the gravitational acceleration. The
net levitation force is provided by the gauge pressure

3
shear stress is much smaller than the pressure (by a factor
h/R) and therefore the contribution of the shear stress
may be omitted in Eq. (3).
For simplicity we also omit the contribution from the
buoyant force. Given that the depth of the basin is R(1
cos max ) = 0.175R for max = 0.60 rad, only a small part
of the cylinders volume is under water: the submerged
fraction is given by (max cos max sin max )/ = 0.043.
The associated buoyant force is g times the submerged
volume (0.043R2Lg, with the density of water), which
amounts to a fraction 0.043/gr = 0.016 (or 1.6 per
cent) of the cylinders weight. A similar calculation for
the spherical fountain yields that the submerged volume
fraction in that case is (2 + cos max )(1 cos max )2 /4 =
0.0069, meaning that the buoyant force compensates no
more than 0.0025 (0.25 per cent) of the kugels weight.
Clearly, the levitation owes little to Archimedes law of
buoyancy.
B.

FIG. 3: (a) Sketch of the cylindrical fountain. The gap coordinates (x, y, z) parallel and perpendicular to the curved
surface are related to the cylindrical coordinates (r, , z) as
x = R, y = R + h r and z = z, where R denotes the radius
of the cylinder and h the thickness of the fluid layer. The
cylinder is submerged up to the angle max . The thickness of
the layer has been greatly exaggerated for the sake of clarity.
(b) The flow inside the water layer is essentially straight since
h R. After leaving the inlet nozzle, the velocity profile u(y)
quickly takes a parabolic shape.

P () Patm inside the fluid layer. We anticipate that

this pressure is not uniform, but rather a function of the
angle defined in Fig. 3. The associated force then follows from an integral of [P () Patm ] cos (the excess
pressure on the cylinder, taken in the vertically upward
direction) over the submerged surface:
Z Z
[P () Patm ] cos dA
Fup =
Asub
max

= LR

[P () Patm ] cos d.

(3)

max

The desired balance between gauge

pressure and weight
R
is achieved when Fup = Fg , or [P () Patm ] cos()d =
Rggr . In order to proceed we thus need to know the
gauge pressure P () Patm inside the fluid layer.
One may note that, next to the pressure, also the shear
stress in the liquid contributes to the force on the cylinder. As the stress induces a force parallel to the solid
surface, rather than perpendicular as is the case for the
pressure, the contribution to the upward force is of the
form dF = sin dA. As we will show, however, the

The fluid mechanical equations

The pressure field (and the related velocity field) can

be found by solving the set of partial differential equations that express the mass and momentum balance
within the fluid. The mass balance is represented by
the continuity equation:

+ (u) = 0,
t

(4)

where is the density of the fluid and u the velocity field.

For an incompressible fluid like water (with = constant)
this simplifies to u = 0. The natural coordinates for
the cylindrical fountain are r, and z (as in Fig. 3a) or,
in view of the fact that the liquid film is extremely thin
(h/R 1) we may also treat the flow as being essentially
along a straight line and use the coordinates (x, y, z)
parallel and perpendicular to the surface of the cylinder
(Fig. 3b). In principle the velocity field could have three
components u = (uex , vey , wez ) but due to symmetry in
the z direction and ignoring the edges of the cylinder
the velocity in the z direction may be assumed to be
identically zero. Also, in the steady state, the velocity
in the y direction will be zero: After leaving the nozzle
the flow field rapidly orients itself in the x-direction,16
u = (u(x, y)ex , 0, 0), independent of z or t. The continuity equation u = u/x + v/y + w/z = 0 then
reduces to
u
= 0,
x

(5)

from which we infer that the velocity is also independent

of x, so u = (u(y)ex , 0, 0).
The momentum balance is expressed by the NavierStokes equation, which for fluids with constant density
and viscosity is given by:


u
+ (u )u = g P + 2 u.
(6)

4
It is the presence of the nonlinear term (u)u on the left
hand side that makes this equation so notoriously difficult
to solve in general. Fortunately, in the present case, all
terms on the left hand side (proportional to the fluid
density ) are negligibly small in comparison with the
viscous term 2 u on the right hand side. This means
that inertia of the fluid plays a negligible role, making
the cylindrical fountain an example of creeping flow, or
Stokes flow, as it is known since Stokes seminal paper
(from 1851) on the subject.5 Usually this type of flow is
associated with low Reynolds number (Re < 1) but in the
present case it also holds for larger values of Re. Indeed,
the first term vanishes because we consider steady flow
(u/t = 0) and also the second term (u )u =
uu/x ex is identically zero on account of Eq. (5).
In a more general setting, when the derivative u/x
would be unequal to zero, inertial effects could come into
play. This is for instance the case for the spherical fountain. In that case one estimates the relative importance
of the fluid inertia by inserting the order-of-magnitude
estimates u/x U/R and 2 u/y 2 U/h2 (where U
denotes the characteristic value of the velocity). Then,
the condition uu/x 2 u/y 2 can be written as
U 2 /R U/h2 , or equivalently:6,7
h2 U R
R2

h2
Re 1,
R2

(7)

where Re = U R/ is the Reynolds number based on

the Kugel radius R. It is at once apparent (even when
u/x is not identically zero) that our analysis need not
be restricted to low Reynolds number. Provided the film
thickness h is small with respect to the radius of the levitated object (h R), the Reynolds number Re based on
R may in fact be quite large. This puts our problem into
a special class of Stokes flows called lubrication flows,8
first analyzed by Reynolds in 1886.9 According to (7),
the appropriate dimensionless parameter can be thought
of as a Reynolds number based on the length scale h2 /R,
which is a subtle combination of the Kugel radius and
the thickness of the water layer. We note that the cylindrical fountain, with u/x = 0, is an exceptionally pure
example of lubrication flow.
Also the gravitational term g in the Navier-Stokes
equation is negligibly small in the problem at hand. This
is readily seen if we recognize that the effect of gravity is simply to add a hydrostatic component to the
pressure. With = 103 kg/m3 and an estimated film
thickness of h = 0.3 mm, the magnitude of this gravitational contribution to the fluid pressure is at most
gh 3 Pa = 3 105 atm. Comparing this to the
required gauge pressure at the inlet nozzle (which is in
the order of several tenths of an atmosphere) we see that
the contribution from gravity is indeed marginal and may
safely be neglected. The Navier-Stokes equation then reduces to:
0 = P + 2 u,

(8)

which is known as the Stokes flow or creeping flow

approximation.6,7,10,11 The components of this equation
represent a balance of pressure and viscous stress inside
the fluid layer. Inertia of the fluid (represented by the
fluid density ) plays no role for the cylindrical fountain.
C.

Pressure field and velocity inside the fluid layer

Solving the creeping flow equation The problem is

now properly laid out and we are ready to solve for the
flow and pressure inside the fluid layer. With the fluid
speed u(y) depending only on the perpendicular position
within the layer, the components of Eq. (8) take the form
d2 u
P
(9)
= 2,
x
dy
P
= 0,
(10)
y
P
= 0.
(11)
z
The latter two equations imply that the pressure is a
function of x only. The first equation (9) can then be
solved by separation of variables. Recognizing that the
expression on the left-hand side does not depend on y,
and the right-hand side not on x, the equation implies
that both sides depend neither on x nor y but are simply
constant. This constant has to be negative, say K,
and thus Eq. (9) yields:
dP
1 dP
=
= K,
dx
R d

d2 u
= K.
dy 2

(12)

The first of these two equations immediately reveals the

reason why the constant (K) had to be chosen negative: The pressure gradient dP/d from the inlet nozzle
to the rim of the socket must necessarily be negative. Integrating this first equation of Eq. (12) yields the form
of the pressure profile:
P () = P (0) RK,

(13)

where P (0) is the pressure at the inlet nozzle. The pressure at max where the flow meets the surrounding air
must be Patm (= 1atm), so Patm = P (0) RKmax .
This gives the gauge pressure P (0) Patm = RKmax .
The only unknown in this relation, K, will follow when
we solve the second equation of Eq. (12).
Integrating the second equation, we find that the velocity profile has the form u(y) = A+ By (1/2)Ky 2, where
A and B are integration constants to be determined from
the boundary conditions. We employ no-slip boundary
conditions at the socket (y = 0) and at the surface of
the cylinder (y = h). If the cylinder is not rotating this
means that the speed vanishes at both boundaries, giving
respectively A = 0 and B = (1/2)Kh, hence we arrive at
the following parabolic velocity profile (see also Fig. 3b):
u0 (y) =

1
Ky(h y),
2

(14)

5
and P () for every angle between 0 and max is then
readily obtained from Eq. (13):
P () Patm =

FIG. 4: The linear pressure distribution P ()/Patm in the

water layer under the levitated cylinder, given by Eq. (18).
The radius and width of the cylinder are taken to be R =
0.50 m and L = 0.30 m, mimicking the granite wheel of Fig.
2, and the inflow rate is Qin = 0.30 liter per second. The
associated thickness of the fluid layer is 0.29 mm, cf. Eq. (20).
Above and close to the inlet nozzle (from = 0 to about
0.05 rad) the pressure field may be expected to deviate from
the straight line; that is why this part of the plot has been
hatched. The outer rim of the fountain, where the water meets
the air and hence the fluid pressure equals the atmospheric
pressure, lies at = max = 0.60 rad ( 35 degrees).

Qin
=L
2

u0 (y)dy =

1
LKh3 ,
12

(15)

where we have simply integrated the velocity profile (14).

So the constant K is
K=

6Qin
,
Lh3

(16)

which is, as expected, directly proportional to the inflow

rate.
Pressure field With the above value of K, the value
of the gauge pressure at the inlet nozzle becomes:
P (0) Patm = RKmax =

6RQin
max
Lh3

(17)

(18)

The pressure field P () is shown in Fig. 4 for typical

parameter values (see next subsection). It is symmetric
around the flow inlet, = 0, and decreases with a constant negative gradient (RK) until it becomes equal
to Patm at the edge of the socket at max .
The strong h-dependence of the pressure, P
Qin R/Lh3 , is one of the foremost features of lubrication flow. If we compare P to the typical shear stress at
the kugel surface, = du/dy Qin /Lh2 , we see that
/P h/R. The smallness of this ratio is the origin of
the renowned low friction in lubrication layers. Without the lubricant, Coulombs friction law would dictate
a frictional force (shear stress times area) comparable in
magnitude to the normal force (pressure times area).

D.

Thickness of the fluid layer

Now that we know P () Patm in terms of and Qin ,

we are in a position to compute the integral in Eq. (3),
i.e., the net levitation force. The result is:
Fup =

where we use the subscript 0 to indicate no rotation.

This is the well-known planar Poiseuille velocity profile
for flow between parallel plates under the influence of
a constant pressure gradient. The constant K sets the
strength of the velocity field and is directly related, as
we shall see in a moment, to the fluid influx Qin at the
nozzle.
Mass balance The inflow rate Qin has dimensions of
volume per unit time (e.g. liters per minute) and is, in
the absence of rotation, equally distributed over the left
and right sides of the cylinder. By mass conservation
this must be equal to the flow integrated across the fluid
layer:

6Qin R
(max ||) .
Lh3

12Qin R2
(1 cos max ) .
h3

(19)

In equilibrium this upward force is equal to the weight

of the cylinder Fg , Eq. (2), and thus we arrive at the
expression for the film thickness:
h = hcyl =

1/3
1/3 
12
Qin
(1 cos max )
. (20)

Lggr

The subscript cyl indicates that this result concerns the

cylindrical fountain, not yet the spherical one. The inlet
of the cylindrical fountain is not point-like, but rather
distributed along a line of length L, and Qin /L represents
the inflow rate per unit length along this line. For clarity
the expression (20) has been split into a dimensionless
prefactor (depending on the geometry of the fountain via
max ) and a factor that has the dimension of a length.
Interestingly, in the case of the cylindrical fountain the
thickness of the fluid layer is independent of the radius R
of the cylinder. The physical explanation for this is that
both the levitation force and the weight of the cylinder
scale as R2 , and thus cancel in the end result for h.
To get a feel for Eq. (20), let us insert typical values
of the fountain parameters, taking the granite wheel in
Fig. 2 as an example. From the figure we estimate that
max is a little under 35 degrees (or 0.60 rad), giving
a prefactor of 0.87. With L 0.30 m, the density of
3
the wheel gr = 2.75 103 kg/m , the viscosity of water
3
= 1.00 10 Pa s, and a typical inflow rate Qin of
0.30 liter per second (0.30 103 m3 /s) we arrive at a film

6
thickness of 0.87 (Qin /gL)1/3 0.3 mm. This is satisfactorily small: there is no danger for fingers being caught
between the wheel and the socket. The small value of h
also justifies our earlier assumption that h/R 1. In
the present example this ratio is only h/R = 0.6 103 ,
which is very good news in the context of the lubrication
condition Eq. (7).
In practice it is wise to choose the inflow rate such that
the film thickness is several tenths of a millimeter. We
note, however, that there is no specific threshold value of
Qin below which the fountain could not work in principle. As long as Qin is positive, the pressure P (0) at the
inlet will always exceed the atmospheric pressure and a
thin lubrication layer establishes itself between the basin
of the fountain and the cylinder. If the surfaces were
perfectly smooth, any supramolecular thickness h (corresponding to tiny inflow rates Qin ) would be sufficient to
make the fountain work. The only problem with choosing
a very small value of Qin is that it will render the system
rather vulnerable: small irregularities in the masonry, a
slight unbalance, or even sand grains caught in the fluid
layer may be enough to cause scratches on the polished
surfaces.

III.

SPHERICAL FOUNTAIN

We now turn to the spherical fountain. In Fig. 5

we show the kugel fountain that adorns the main entrance of the House of Science in Patras, Greece, where
we were allowed to do some elementary measurements.
Its granite sphere has a diameter of precisely one meter
(R = 0.50 m) and, by trying to fit plastic sheets of different thickness inside the gap between the sphere and the
socket, we found that the thickness of the water layer is
h = 0.30 0.05 mm. In this section we will show that
such a thickness is indeed consistent with hydrodynamic
theory. For sufficiently small inflow rates the calculation
can again be based on Stokes flow alone, i.e., neglecting
the liquid inertia. At higher flow rates, however, inertial
effects become increasingly important for the spherical
fountain and this turns out to have a negative effect on
the levitation force. Consequently, Qin should neither
be too small (as before, to avoid scratches) nor too large
(to prevent the inertial effects from becoming dominant),
meaning that there exists some intermediate value of Qin
for which the fountain works optimally.

A.

FIG. 5: The kugel fountain at the House of Science in Patras,

Greece. The granite sphere has a diameter of precisely 1 m
and is immersed in the water basin up to an angle max
35 degrees. We measured the thickness of the water layer to
be 0.30 0.05 mm.

Creeping flow approximation

The spherical fountain calls for a three-dimensional

analysis (see Fig. 6). The flow is now radially outward
from the inlet nozzle, u = (ur , u , u ) = (0, u0 (r, )e , 0),
and as the fluid is being spread over a region of increasing
area its speed must decrease with to ensure conservation of mass. (This deceleration means that the inertial
term (u )u in the Navier-Stokes equation is not iden-

FIG. 6: The spherical coordinates (r, , ) for the kugel fountain. Within the fluid layer one may also conveniently use the
coordinates (x, y, ) with x = R and y = R + h r as in
Fig. 3. The thickness of the fluid layer has been exaggerated
for clarity.

tically zero anymore, but for the time being we shall assume it is still small in comparison with the terms P
and 2 u, which is a valid approximation as long as the
inflow rate Qin is sufficiently small.) When reaching an
angle , the circumference of the cross-section has become
2R sin , and the total flux through this circumference
per unit time is simply:
Qin = 2R sin

u0 (y, )dy.

(21)

7
Once more, the velocity profile across the fluid layer is
parabolic in the creeping flow approximation. So we set
u0 (y, ) = 12 K()y(h y) and the integral in the above
Eq. (21) is then readily evaluated:
Qin =

1
R sin K()h3 ,
6

(22)

showing that the factor K() in the expression for the

velocity decreases as 1/ sin :
K() =

6Qin
,
Rh3 sin

(23)

and hence u0 (y, ) = 3(Qin /Rh3 )y(h y)/ sin .

We proceed along the same lines as for the cylindrical
fountain, picking up the analysis at Eq. (12). The relation Eq. (12)b corresponds to the above assumption of a
parabolic velocity field. With the factor K being given
by Eq. (23), the gauge pressure P () Patm can then
be computed from the equation for the pressure gradient
(12)a:

P () Patm =

(1 cos max ) sin

6Qin
ln
.
3
h
(1 cos ) sin max

(24)

This pressure profile is depicted in Fig. 7 (solid curve).

Interestingly, the pressure exhibits a singularity at = 0,
where the logarithmic factor diverges (this can be traced
back to the diverging velocity 1/ sin ). This poses no
problem, however: In the first place, we should exclude
the immediate neighborhood of the nozzle, i.e. a small
area around = 0, since our analysis does not cover this
region (the velocity of course does not diverge in reality).
Secondly, even if we choose to use the above expression
for the pressure down to = 0, the contribution to the
levitation force remains finite:
Z Z
[P () Patm ] cos dA
Fup =
=

A
2 Z max

[P () Patm ] cos R2 sin dd

6Qin R2
(1 cos max ).
h3

(25)

So we see that the logarithmic singularity in the pressure

is tamed by the smallness of the area around the origin
over which it is integrated.
Now, by equating the above levitation force Eq. (25) to
the weight of the sphere, Fg = (4/3)R3 ggr , one readily
obtains the thickness of the fluid layer for the spherical
fountain (in the creeping flow approximation):
hsph =

FIG. 7: Pressure distribution P ()/Patm under the spherical

fountain of Fig. 5. The solid curve is the pressure due to
viscosity alone, given by Eq. (24), while the dashed curve
[Eq. (29)] includes also the contribution from the inertial
effects. We take a typical water influx of Qin = 1.5 liter
per second, and the associated thickness of the fluid layer
is 0.31 mm. The hatched area indicates the nozzle region
(from = 0 to 0.05 rad) where the water flows into
the system and the actual pressure will deviate from our theory; thus both the logarithmic singularity in the solid curve
for 0, as well as the strong Bernoulli suction when the
dashed curves dives to negative values, are shrouded and made
harmless by the presence of the nozzle. The outer rim of the
fountain, where the water meets the air (and hence the fluid
pressure becomes equal to the atmospheric pressure), lies at
= max = 0.60 rad.

1/3
1/3 
9
Qin
(1 cos max )
,
2
Rggr

(26)

This result for the spherical fountain has essentially the

same structure as that for the cylindrical fountain, see
Eq. (20). It illustrates that a perfectly polished spherical

fountain (just like the cylindrical one) does not require a

specific threshold value of Qin in order to function. As
long as Qin is positive, small as it may be, a lubrication
layer with a finite thickness h will establish itself.
Taking the fountain depicted in Fig. 5 as an example (with R = 0.50 m, max = 0.6 rad, = 103 Pa s,
3
gr = 2750 kg/m ) and setting the inflow rate to Qin =
0.3 liter/s = 0.3 103 m3 /s, Eq. (26) yields a value for
the gap width of hsph 0.18 mm. A smoothly polished
kugel may just be able to operate with such a thin water layer. To be on the safe side, however, it is good
to have a somewhat thicker layer and this can readily
be accomplished by choosing a larger inflow rate. At
Qin = 1.5 liter/s the above equation (26) predicts a layer
thickness hsph = 0.31 mm, in excellent agreement with
the 0.300.05 mm we measured on the fountain of Fig. 5.

8
B.

Inertial effects: Bernoulli suction

Contrarily to the cylindrical fountain, the velocity in

the film below the sphere is not uniform, but decreases
as 1/ sin . This means that the inertia of the liquid can
become important: the deceleration of the fluid mass
induces an extra contribution to the pressure field inside the water layer, originating from the advection term
(u )u in Eq. (6). To estimate the importance of
this advection term with respect to the viscous term
2 u, we check whether the condition (7) is still satisfied. For the parameter values cited above (using the
flow rate Qin = 1.5 liter/s) the Reynolds number is
Re = U R/ 106 . Given that (h/R)2 106 we
must conclude that (h/R)2 Re is of order unity, so inertial effects cannot really be neglected at these high values
of Qin . As an aside, we note that the Reynolds number
based on the film thickness (Reh = U h/) is, for the
same value of Qin , still small enough for the flow to remain laminar.
1.

Inviscid flow: the yarn spool effect

To illustrate the effect of inertia in its purest form, we

first consider the idealized case of inviscid flow. This is
the opposite limit of creeping Stokes flow: one now assumes that the inertia (or kinetic energy) of the liquid is
so large that it completely dominates over viscous friction. If in addition the flow is steady and irrotational,
one can integrate the Navier-Stokes equations (6) to a
very simple form: 21 |u|2 + gy + P = constant, or if we
neglect gravity (just as before):
1
|u|2 + P = C,
2

FIG. 8: The classical demonstration of Bernoulli suction, using a yarn spool and a paper card with a thin needle pierced
through it, to keep it centered with respect to the hole of the
spool. If one blows air through the spool, the increased air
velocity in the narrow layer between the spool and the card
induces (by Bernoullis law) a region of low pressure there. As
a result, the atmospheric pressure of the ambient air pushes
the card against the spool. The demonstration is usually done
upside down, as in the above picture, to show that (apart from
the surprising fact that the card is not simply blown away)
the Bernoulli suction can even beat gravity.

If u(r) denotes the radially outward velocity in the

layer at a distance r from the inlet, and P (r) the local
pressure, Bernoullis law tells us that 12 u2 (r) + P (r) =
1
2
2 u (rmax )+Patm (with rmax the radius at the rim where
the layer meets the ambient atmosphere). Now, by mass
conservation Qin = 2ru(r)h, or u(r) = Qin /(2hr), and
thus we get:


1
P (r) = Patm u2 (r) u2 (rmax )
2


1
1
Q2
.

= Patm 2in2
2
8 h
r2
rmax

(27)

where C is a constant. This is the celebrated Bernoullis

law that expresses the conservation of energy in an inviscid flow: regions of high kinetic energy correspond to
low pressure, and vice versa. This has a remarkable consequence for the spherical fountain, where mass conservation dictates that the velocity in the water below the
sphere decreases from the nozzle to the outlet. According
to Bernoullis law (27), this means that the P is lowest at
the nozzle; if it were for the inertial contribution alone,
the pressure would in fact everywhere be below the atmospheric pressure that is reached at the outer rim. Rather
than providing a levitating force, the inertial pressure induces a downward force that attracts the sphere towards
the socket. This effect is known as Bernoulli suction.
A classic demonstration of Bernoulli suction is the experiment with a paper card and a yarn spool illustrated
in Fig. 8. When air is blown through the spool, the increased air velocity in the layer between the spool and
the card means (according to Bernoullis law) that a region of low pressure is created here. As a result, the
atmospheric pressure of the ambient air pushes the card
against the spool.

(28)

Clearly, the pressure in the layer is everywhere smaller

than Patm ; it only attains this value at r = rmax .
2.

Inertial effects and the granite sphere fountain

From the above analysis it is evident that the granite

sphere can only be levitated when viscosity dominates
over inertia. If not, the pressure reduction due to the decelerating liquid would disable the fountain. This means
that one should not make the inflow rate too large (since
the relative influence of inertia grows with Qin ). On the
other hand, as we have noted before, one should not make
Qin too small in order to avoid damaging contact between
the kugel and the basin. Hence, there is an intermediate range of Qin for which the spherical fountain works
optimally.
A consistent description that includes both viscous
and inertial effects is notoriously difficult for the Navier-

9
Stokes equations. This is why we have until now concentrated on the limiting cases of purely viscous and purely
inertial flow. If one wants to combine the two, one generally has to reside to approximation schemes. For a radial
flow between two parallel discs in the horizontal plane
(such as the yarn spool of Fig. 8, and arguably also the
spherical fountain, since the curvature of the kugel plays
only a minor role) this problem has recently been addressed by Armengol et al. in this journal.12 They derive the following approximate expression for the pressure
field in the fluid layer:
P (r) = Patm +

6Qin
rmax
27Q2in
ln

3
h
r
140 2 h2

1
1
2
r2
rmax

(29)
In this equation one recognizes the pressure contributions
from viscosity and inertia, respectively; their structure is
very similar to the previously derived exact expressions
(24) and (28) for the idealized limiting cases. As expected, the pressure due to viscosity generates a positive
levitating force whereas the inertial pressure works in the
negative direction.
We note that the magnitudes of both contributions increase with the inflow rate, but while the viscous pressure
grows linearly with Qin , the inertial contribution scales as
Q2in . This confirms our earlier observation that the dominance shifts from the viscous to the inertial regime as Qin
is gradually increased. Using the estimates Qin hRU
and r R one verifies from (29) that the cross-over takes
place when U h2 /(R) is of order unity, consistent with
(7). Eventually, at large flow rates, the generated levitating force will no longer be able to lift the Kugel. In
Fig. 7 we have sketched how inertial effects change the
pressure profile in the fluid layer below the granite sphere
(dashed curve).

IV.

DAMPING OF ROTATIONS

Anybody who has ever put his or her hands on the

kugel fountain will have experienced that the sphere is
easily set into a rotating motion and that it takes a surprisingly long time before the sphere comes to a halt.
This is because there is no direct contact between the
sphere and the socket, making friction very low. In fact,
the only source of friction lies in the viscous drag the
fluid layer exerts on the sphere and this is exactly the
same principle on which roller bearings work. In this section we compute this small viscous drag and show that
it causes the angular speed to slow down exponentially,
as (t) = (0) exp(t/trel ), with a relaxation time trel of
the order of 10 minutes. We first address the cylindrical
fountain, which we treat analytically, and then present
experimental results for the case of a sphere.

FIG. 9: (a) Rotation of the levitated object breaks the symmetry of the flow inside the fluid layer. (b) The velocity
field is now a superposition of the parabolic Poiseuille profile
(caused by the pressure gradient from the inlet nozzle to the
surrounding air) and the linear Couette profile caused by the
velocity difference between the socket, which is at rest, and
the surface of the kugel moving at speed R.

A.

The cylindrical fountain: analysis

When the cylinder is set into rotation with angular

frequency , the velocity at the cylinder surface becomes
u = R. This has an effect on the flow inside the water layer. The no-slip boundary condition now becomes
u(y = h) = R. As illustrated in Fig. 9, the resulting

10
profile can be seen as a superposition of the Poiseuille
parabolic profile and a simple linear Couette profile. Such
a superposition is allowed since the Stokes equation (8)
is linear with respect to the velocity. Mathematically the
profile can thus be written as u(y) = u0 (y)+u (y), where
u0 (y) is the profile without rotation given by Eq. (14) and
y
u (y) = R .
h

(30)

Interestingly, the rotational profile has zero second

derivative (d2 u /dy 2 = 0) and therefore does not contribute to the pressure balance P/x = d2 u/dy 2 =
d2 u0 /dy 2 , cf. Eq. (9). This means that the pressure
distribution P () is unaffected by rotation.
The main effect of the rotation is to break the left-right
symmetry of the system and hence it produces a nonzero
frictional torque (i.e. force moment) on the cylinder. The
cylinder applies a force on the water and, by Newtons
third law, the water applies an equally strong reaction
force on the cylinder. This is accomplished via the shear
stress in the water layer, which can be computed as


du0
du
= 0 + =
.
(31)
+
dy
dy
Shear stress is also present when there is no rotation,
but by symmetry the Poiseuille contribution 0 does not
yield a net torque on the cylinder. The entire net torque
comes from the Couette contribution = R/h. We
can compute the torque on the cylinder by integrating
the force moment dT = R dA over the submerged
surface,
Z Z
R dA
T =
A

max

max

R
LR d
h

2max LR3
=
.
h

(32)

The frictional torque is thus proportional to and opposite to the direction of rotation. It will slow down the
rotation according to the equation of motion
T = Icyl

d
,
dt

(33)

where Icyl = 12 M R2 = 21 gr LR4 is the moment of inertia of the granite wheel. Given the expression for T
derived above, this equation of motion takes the form
4max
d
=

dt
gr Rh

(34)

which is easily solved to give an exponential decay:

(t) = 0 et/trel ,

(35)

with the relaxation time trel given by:

trel =

gr Rh
.
4max

(36)

This relaxation time consists of a characteristic time scale

for the damping, gr Rh/, multiplied by a dimensionless
geometric prefactor. Using the same fountain parame3
ters as in Section 2 (gr = 2750 kg/m , R = 0.5 m,
h = 0.3 mm, max = 0.6 rad and = 0.001Pa s) the
characteristic time scale gr Rh/ is 412 seconds. Putting
in the geometric prefactor, the relaxation time is found to
be 540 s, i.e., no less than 9 minutes! Indeed, the damping of the rotations turns out to be a slow process, with
the weak viscous drag only very gradually wearing down
the angular momentum of the massive granite wheel.
B.

The spherical fountain: scaling argument and

experiment

The calculation of the torque on the sphere is much

more involved than for the cylinder. To begin with, the
axis of rotation is no longer fixed. There are two basic
modes of rotation: around the horizontal axis, as for the
cylinder, but also around the vertical axis. The effectiveness of the viscous drag is different for the two modes.
This is due to the fact that the fluid velocity (and thus
the strength of the drag) as well as the effective moment
arm are not uniform over the surface. Rather than pursuing a detailed analysis of the rotating sphere it is more
insightful to give a scaling argument that focuses on the
essential physics.
In analogy to the torque on the cylindrical wheel, given
by Eq. (32), we find for the sphere:
T

R4
.
h

(37)

Here the width of the wheel L appearing in (32) has been

replaced by R. Different modes of rotation will have different prefactors, but these are not captured by a scaling
analysis. This torque must be equated to Id/dt, where
the moment of inertia I M R2 gr R5 . Therefore,
d
R4
gr R5 .
h
dt

(38)

Hence we recover a similar exponential decay of the angular velocity as in Eq. (34), with a relaxation time
that once more reads
trel

gr Rh
.

(39)

To verify this scaling argument, we have performed a

series of experiments on the granite sphere fountain of
Fig. 5. Bringing the sphere in a rotation around the
horizontal axis, we monitored the decay of the angular
velocity during approximately 40 revolutions. By tracking a distinct spot on the surface of the sphere, we were

11

FIG. 11: The Leidenfrost phenomenon: a water drop hovering

above a hot plate, levitated by its own thin vapor layer. Image courtesy of Raphaele Thevenin and Dan Soto, reproduced
from D. Quere, 2013.13

FIG. 10: Measured decrease of the angular velocity (t)

for the spherical fountain of Fig. 5 rotating around a horizontal axis. At the start of the experiment, the kugel was
given a spinning motion with (0) = 1.57 rad/s, i.e., one
complete revolution in precisely 4 s. The plot shows that
ln[(t)/(0)] = Ct with C = 4.55 103 s1 , or equivalently, that the angular velocity decays exponentially as
(t) = (0) exp(Ct). The corresponding relaxation time
is trel = 1/C = 220 s.

able to determine the time for each complete revolution

and thus the angular velocity (t). The results are presented in Fig. 10, showing ln versus time. The data
are seen to lie on a straight line, in excellent agreement
with the predicted exponential decay of (t). The slope
of the curve can be identified with 1/trel and yields
trel 220 s; the same value was found (within about
15 seconds) for different runs of the experiment. This
is consistent with the order-of-magnitude estimate (39),
3
which for the sphere in question (with gr = 2750 kg/m ,
R = 0.50 m, h = 0.3 mm and = 0.001 Pa s) predicts
that the relaxation time should be of the order of 400
seconds.
In the above discussion, we have left out several effects
that might influence the relaxation time. For one thing,
we expect the inlet nozzle not to be positioned precisely
in the center of the socket but several centimeters besides
it. This induces a spontaneous rotation even in the absence of any human intervention, adding a realistic effect
to kugels decorated with a map of the Earth, keeping the
whole surface properly wetted, and in all likelihood
affecting the relaxation time. Another feature that has
not been taken into account is that the rotation generates
an asymmetry in the outflow at the edge of the socket, as
sketched in Fig. 9: the water comes out with much more
vigor at one side (in the direction of the rotation) than at
the other. The possible extra source of dissipation this
introduces is beyond the scope of the present analysis.

V.

DISCUSSION

In conclusion, we are now in a position to give a definitive answer to our original question: What makes the
fountain work?. It is not Archimedes law of buoyancy,
the favorite of the Science Museum visitors. Instead, it is
another basic principle, less familiar to the general public
but of key importance in almost every type of machinery:
lubrication.
As a matter of fact, the kugel fountain can be thought
of as a giant roller bearing. The pressure inside the
thin fluid layer (thickness h) scales as 1/h3 , and is perfectly able to carry the heavy granite sphere. For a given
flow rate Qin , usually around 1 liter per second, the water layer automatically adjusts itself to the thickness required to lift the weight. The water acts as a lubricant
and is responsible for the surprisingly low friction experienced by the sphere.
Another phenomenon that relies on the same lubrication principle is the Leidenfrost drop, shown in Fig. 11.
This is a water drop hovering above a hot plate (typically around 250 degrees Celsius) without touching it,
carried by its own vapor layer.13,14 The fact that air is
a poor conductor of heat ensures that the drop, instead
of instantly boiling away, only slowly evaporates and can
survive for more than a minute. The same effect is observed when pouring liquid nitrogen on a table: nitrogen
drops are skating freely over the surface, with negligible
friction, thanks to the lubricating layer of nitrogen vapor.
Also the popular game of air hockey works on the same
principle, only in this case the puck does not evaporate of
course but the lubrication layer is provided by air flowing
out of tiny pores in the table.11
In all these examples, the thickness of the layer adjusts
itself such that the integrated pressure exactly balances
the weight of the levitated object. Owing to the smallness
of the gap, the viscous forces inside the flow dominate
over the inertial ones. This is, as we have shown, a necessary condition for achieving an upward levitation force.
In addition, one may wonder whether the compressibility

12
of the air does not fundamentally change the physics of
levitation. The effects of compressibility, however, only
begin to play a role when the Mach number becomes of
order unity. This number is defined as Ma = u/c, where
u is the velocity of the flow and c the speed of sound in
the air. Since the latter is roughly 330 m/s, the Mach
number in all examples remains much smaller than 1 and
hence the flow may safely be treated as incompressible.
In fact, also the granite sphere fountain itself can operate on air. As compared to water, both viscosity and
inertial effects become smaller in air but not to the same
degree. While the viscosity is reduced by a factor 50,
the density is reduced by no less than a factor 1000.
This means that the troublesome inertial effects are relatively smaller for the air-borne kugel than for the water
version, at least as long as we may keep the air inflow
rate Qin at a reasonably low level. (Recall that the viscous effects grow linearly with Qin , whereas the inertial
effects grow quadratically.) The condition that Qin be
kept small poses just one practical challenge: it implies,
by Eq. (26), that the gap width h will be very small,
which calls for a sphere and basin that are both perfectly
spherical and exceptionally well polished. The successful lubrication of the kugel with air thus relies, not on a
strong airflow, but on the craftsmanship of the stonema-

See the information on various websites, for instance that

of Clutes Kugel in the State Botanical Garden of Georgia
in Athens, Georgia at http://clutebarrow.org/kugel.html.
From the brochure by Kusser Aicha Granitwerke
on Floating sphere and floating object fountains, see
http://www.kusser.com/.
See http://midwarks.info/kenglobe/ for a highly readable
report on the granite sphere fountain (also known as the
groovy ball project) in Kenilworth, UK.
See the lemma Kugel ball at http://en.wikipedia.org/, including the frequently-asked-questions section.
G.G. Stokes, On the effect of the internal friction of fluids on the motion of pendulums, Cambridge Phil. Trans. 9
(1851), pp. 8106.
G.K. Batchelor, An Introduction to Fluid Dynamics (Cambridge University Press, Cambridge UK, 1967). Flow fields
in which inertia forces are negligible (including lubrication
flows) are discussed in section 4.8.
F.M. White, Viscous Fluid Flow, second edition (McGrawHill, New York, 1991). Creeping flows (including lubrication flows) are discussed in section 3-9.
A. Cameron, Basic Lubrication Theory, Ellis Horwood Series in Engineering Science (Ellis Horwood, New York,
1976).
O. Reynolds, On the theory of lubrication and its application to Mr. Beauchamp Towers experiments including
an experimental determination of the viscosity of olive oil,
Phil. Trans. Roy. Soc. London Ser. A 186 (1886), pp.157
234.

son. Not without reason an airborne kugel (a perfectly

polished black granite sphere with a diameter of 0.40 m)
was awarded a special prize at the International Granite
and Stone Fair Stona 2004 in Bangalore, India.

Acknowledgments

We wish to thank Devaraj van der Meer, Dimitrios

Razis, and Leen van Wijngaarden for many insightful
comments and stimulating discussions. We are grateful to Dimitrios Razis and Vasilis Verganelakis for their
kind help with the measurements on the kugel fountain at
the House of Science in Patras. KvdW cordially thanks
the Physics of Fluids group at the University of Twente
for its hospitality during the final write-up of the paper. He acknowledges partial financial support from the
research project COVISCO, grant code MIS 380238,
which is co-financed by the European Union (European
Social Fund) and Greek national funds through the Operational Program Education and Lifelong Learning of
the National Strategic Reference Framework - Research
Funding Program THALES: Investing in knowledge society through the European Social Fund.

10

11

12

13

14

15

16

Y.A. C
engel and J.M. Cimbala, Fluid Mechanics: Fundamentals and Applications (McGraw-Hill, New York, 2006).
The creeping flow approximation is treated in section 10-3.
L.G. Leal, Advanced Transport Phenomena: Fluid Mechanics and Convective Transport Processes (Cambridge
Univ. Press, Cambridge, 2007). The thin-gap approximation and lubrication problems are discussed in Chapter 5.
J. Armengol, J. Calb
o, T. Pujol and P. Roura, Bernoulli
correction to viscous losses: Radial flow between two parallel discs, Am. J. Phys. 76, 730737 (2008).
D. Quere, Leidenfrost Dynamics, Annu. Rev. Fluid Mech.
45, 197215 (2013).
J.H. Snoeijer, P. Brunet, and J. Eggers, Maximum size of
drops levitated by an air cushion, Phys. Rev. E 79, 036307
(2009).
A photograph of the prizewinning airborne kugel can
be found on the website of Brahma Granitech at
http://www.brahmagranitech.com/.
Due to the extremely small thickness of the fluid layer, the
water has hardly any freedom to explore the normal direction once it has left the nozzle region, and is immediately
forced into the parallel direction. As noted in Figs. 4 and 7,
the size of the nozzle region covers only about 10% of the
total range of . For the spherical fountain this constitutes
an area of (0.1)2 0.01 of the total immersed area,
making the contribution of the region that is not treated
by our analysis of the order of 1%.