SPINNING DISKS UNDER GRAVITY

Shayak Bhattacharjee
Department of Physics, Indian Institute of Technology Kanpur, NH-91, Kalyanpur, Kanpur – 208016, Uttar Pradesh, India.
email : [email protected], [email protected]

PACS : 45.20.dc 45.40.Cc

Keywords : Euler’s disk precession contact friction finite-time singularity

Abstract : In this work I examine the counter-intuitive accelerating rotation of a disk-shaped object such as a coin when it is
released on a smooth surface with a spin about a vertical axis. The only existing theoretical model examines the terminal stages of
the motion and proposes air drag as the primary agency behind the sudden stop. Subsequent experimental work has tended to
favour rolling friction or sliding friction as the principal motivator. Here I show that a combination of rolling friction and sliding
friction is necessary and sufficient to understand the complete motion of the disk. Rolling friction is dominant in the initial stages
of the fall but it changes to sliding friction after the disk dips below a certain angle with the horizontal. The change occurs as a
result of the inability of the normal reaction to provide the torque for the motion. Qualitatively, this model can account for all the
observations associated with the disk. Quantitative analysis has been carried out with the friction treated as perturbations on the
corresponding steady state frictionless solutions and a finite time singularity in the precession rate has been obtained with a
critical exponent of 1/2.

§1. The classical motion of disks under the influence of friction is often interesting and complicated. One such case
has been treated by Farkas et. al. [1] and subsequently by Weidman and Malhotra [2]. They proved the surprising
result that if a disk is released flat on a rough surface with both a translation and a rotation about the centre of
mass, then the two motions stop at the same instant and there is a definite relationship between their magnitudes.
Another interesting behaviour arises when, instead of being released flat on its face, the disk is released on its
edge with a spin about a more or less vertical axis. In this case, its centre of mass (CM) comes down slowly while it
precesses and rotates with ever-increasing speed. Suddenly however the motion stops and the object comes to
rest flat on its face. The toy called Euler’s disk [3] is tailored to make this phenomenon appear at its most
spectacular but virtually any disk and any smooth surface such as a table top suffice for an effective
demonstration of this bizarre motion.

Interestingly, explanations of this motion are as elusive as the phenomenon itself is commonplace. Moffatt [4]
attributed the motion to the viscous drag of air on the basis of a theoretical analysis. Subsequently, van den Engh
et. al. [5], Easwar et. al. [6] and Petrie et. al. [7] have done careful experiments which indicate that rolling friction
rather than air drag is the principal agency behind the observed motion. Caps et. al. [8] however claim that it is not
rolling but sliding friction which causes the disk to behave the way it does.

In this Letter I propose a theory of the disk which explains its complete motions and achieves good agreement
with experimental measurements of the critical exponent of precession. In addition, I also address the following
hitherto unsolved question regarding the disk : (a) how exactly does friction, which acts in a horizontal direction,
make the disk come down ? (b) in the terminal stages of the motion, why does the spin rate of the disk about its
axis decrease instead of increasing along with the precession ? (c) given the release parameters how long will the
disk take to come to a stop and (d) why does the CM of the disk often execute circular motions during the falling
process ?

The equation of motion for the disk can be written in terms of balances of forces and torques. The forces involved
are the weight mg of the disk, the normal reaction N from the ground and the friction F with the ground. The first
of these acts on the CM while the other two act at the point of contact. For the torque balance, we need to apply
Euler’s equation to transform from the body frame to the lab frame. The complete equations of motion are

1
 mɺɺrCM = ( −mg + N ) zˆ + F , (1a)
dL
+ ω × L = rcontact × N + rcontact × F . (1b)
dt

In the above equation, rCM denotes the position of the centre of mass while rcontact denotes the vector from the CM
to the point of the disk instantaneously in contact with the ground. A direct analysis of this equation is very
difficult hence I will use physical arguments to penetrate
penetrate the system and make a few pertinent assumptions.
Under the aegis of these assumptions, I will obtain a few ‘intermediate’ equations of motion, which describe only
part of the dynamics but are analytically tractable.
able. Combining these intermediate solutions will lead to a
(approximate) solution for Eqs. (1a) and (1b).

I will work with a combination of the stationary laboratory frame and the body-aligned
body aligned frame i.e. a frame which is
instantly aligned with the disk but is still inertial. Note that this frame is not genuinely attached to the disk – it is
aligned with the disk at only one instant of time. The conversion between the true disk frame and this frame is
achieved using Euler’s equation which has been mentioned in Eq. (1b). A variety of reference referenc frames in this
context has been discussed in Reference [9]. The origin is the CM of the disk. I will convert between the two
[10 Let x, y and z be a set of mutually perpendicular right-handed
frames using the Euler angles [10]. right axes in the
laboratory frame with gravity along the –z direction. Orthographic views of the transformations are shown in Fig.
1.

Figure 1 : Orthographic views of orthogonal transformations. Adapted from Reference [9]. The middle panel shows the disk in profile view. The
right panel has it in top view. The complete three-dimensional
three dimensional view of the transformation is cumbersome and is a pretty standard figure in any
pedagogical rotational dynamics
namics material, for instance [11],
], hence I omit it here.

The transformation rule from the x-y-z to the d-q-o body frame is given below:
below

dˆ   cosϕ cosψ − cosθ sinϕ sinψ sinϕ cosψ + cosθ cosϕ sinψ sinθ sinψ   xˆ 
  
qˆ  =  − cosϕ sinψ − cosθ sinϕ cosψ − sinϕ sinψ + cosθ cosϕ cosψ sinθ cosψ   yˆ  (2)
 oˆ   sinθ sinϕ − sinθ cosϕ cosθ   zˆ 
  

The inverse transformation is achieved by transposing the matrix in the above equation.

Clearly, d-q-o constitutes a principal basis for the disk, which I assume to be symmetric with moments of inertia
Id=Iq=I and Io about its own axis. Hence, the angular momentum vector satisfies Ldqo = ( I ωd , I ωq ,I o ωo ) where ω is
the angular velocity vector of the disk as a whole. As for the Euler angles, φ describes the disk’s precession, ψ its
spin and θ its nutation. I mention below the relations connecting these motions with ωdqo :

2
 ωd = θɺ cosψ + ϕɺ sinθ sinψ , (3a)
ωq = −θɺ sinψ + ϕɺ sinθ cosψ , (3b)
ωo =ψɺ + ϕɺ cosθ . (3c)

Since the rate of nutation is at least one order of magnitude smaller than the precession and spin, I will leave out
the nutation in the subsequent discussion.

§2. First I consider the disk without friction when the only forces on it are gravity and the normal reaction N. Since
there is no torque about either the z or the o axis, Lz and Lo are constants of the motion, determined by the initial
conditions i.e. the angular velocity of the disk when it was released. The expression for Lo is self-evident while that
for Lz follows by breaking it up into d-q-o components using Eq. (2) and again converting back to x-y-z
coordinates:

Lo = I o ωo = const. (4a)
Lz = I Ω sin θ + I o cosθ (ψɺ + Ω cosθ ) = I Ω sin θ + I o ωo = const. ,
2 2
(4b)

where ϕɺ = Ω is the precession frequency. Since the term in parentheses on the right hand side of Eq. (4b) is
constant (ωo), as θ decreases, the precession rate tends to blow up. This is just what is seen in the disk – as the
angle becomes shallower, the precession speed increases dramatically.

In the frictionless case the disk will enter a steady state featuring precession and spin at a constant rate and a
constant nonzero angle to the horizontal which are determined by initial conditions and torque balance condition.
The equation for this state can be obtained from Eq. (1b) by setting the normal reaction equal to mg and the
friction equal to zero. The state is stable and small perturbations will induce oscillatory motions about this state.
The disk is generally released with little or no rotation about the o-axis and in particular if Lo=0 the vector L lies
in the plane of the disk. I now introduce friction.

The contact existing between the disk and the ground is a point contact. Friction opposes relative motion between
the contact point and the ground. At any instant, the point of the disk in contact with ground is r ( cosα ,sinα )
where r is the disk radius and α=-90o-ψ and the velocity of this point, using the relation v = ω × r in the d-q-o
( )
frame, is rω cosψ dˆ − sinψ qˆ . Hence if ωo is zero, there will be no slip between the disk and the ground. Contrary
o

to the ideal nonslip situation, a little friction, smaller than when there is relative motion, is always there due to
deformation of the contact surfaces and this is of paramount importance in governing the motion of the disk. I will
refer to the friction during slipping and during nonslip as ‘strong’ and ‘weak’ friction respectively. The first
question is, in which direction will the weak friction act ? This is not immediately clear from Eq. (1b) so let me
turn to energy considerations, since friction must necessarily drain away the disk’s energy. The relation
E = (1 / 2)L ⋅ ω enables me to write

1 1 1
E = Ldqo ⋅ ωdqo = ( Ω sinθ sinψ , Ω sinθ cosψ , 0 ) ⋅ ( I Ω sinθ sinψ ,I Ω sinθ cosψ , 0 ) = I Ω 2 sin 2 θ , (5)
2 2 2

from which it follows that the friction tries to drag down the rate of precession. As soon as precession slows down
infinitesimally, the disk will start to slip. The strong friction will immediately correct for this by reducing the spin
speed too. At the end of this process, the magnitude and precession rate of L will have become slightly smaller
hence the torque on the disk will have reduced. And this can only happen if the strength of the reaction N
decreases. In the absence of friction, N was balancing the disk’s weight but now the reduced value will no longer
be able to do this and the CM of the disk will fall down by a slight amount. This process will keep going on and thus
rolling friction i.e. the weak friction will drag the disk down slowly but surely. From the torque balance, the disk’s
precession speed will increase. This argument gives a mechanism by which friction, which ostensibly acts in the
tangential direction, drags the disk down in the vertical direction, and hence answers question (a) I posed at the
end of §1. It should be noted that the frictional force will cause a displacement of the CM of the disk; since this
effect will typically be quite small, I neglect it in the analysis.

3
§3. At sufficiently small θ however, it is impossible
mpossible to find a frictionless steady state with ωo=0 which requires L
to lie flat on the disk. The finite torque of the reaction cannot make an ever-increasing
ever increasing L precess at an ever-
accelerating rate in an ever-widening
ever widening cone. Since the rolling friction states
states are very close to the corresponding
steady states, it follows that such states too are not possible at shallow angles. To satisfy torque balance at small
angle, the large L must project out from the disk plane, as shown in the final configuration of Fig. 2. This implies
that the rate of spin should be smaller than that of precession whereby ωo=0 is no longer possible and the contact
becomes a sliding one. The mechanism of descent of the disk however remains unchanged.

Figure 2 : Schematic representation of the emergence of L from the plane of the disk. Profile views of the disk can be seen. The initial position
shown is the limit of validity of rolling without slipping and L remains attached to the plane of the disk. Now, friction
frictio causes the CM of the disk
to come down by an infinitesimal amount, causing a change in the angle between the disk and the ground (for the sake of clarity clari I have
exaggerated this angle greatly and also ignored the z-displacement
displacement of the CM). This change in angle is determined entirely by the force balance
and torque does not enter the equation. If L still remains attached to the disk, its tip gets displaced by a very large amount (note the distance
between the initial L and the red hypothetical L). The frictional
tional torque however is not that large – the blue vector shows the change in L
accruing during this time on its account. The final L must be the sum of the initial L and the change in L hence it comes out of the plane of the
disk.

Now it is time to express the above logic in a mathematical form. For the frictionless steady state at small angle, if
θ and Ω are specified, then the spin rate ψɺ and hence the entire vector L get determined uniquely from the torque
balance equation (1b). Since frictionless steady states are being considered I set N=mg z and F=0. Further since L
precesses about the z axis with frequency Ω, dL/dt must equal Ωzˆ × L :

dL  I −I 2 
= Ωzˆ × L = nˆ  o Ω sin 2θ + I o Ωψɺ sinθ  = nˆ ( mgr cosθ )
dt  2 
Io − I . (6)
mgr cosθ − Ω 2 sin 2θ
⇒ ψɺ = 2
I o Ω sinθ

At the risk of repetition I once again clarify the distinction between the steady
steady state solutions given in Eq. (4) and
Eq. (6).
). The first of these assumes that ωo is constant (equal to zero for rolling without slipping), determined from
initial conditions.
ons. If the disk is released on a frictionless surface to begin with, then it will always enter a state
given by Eq. (4). ). The torque balance condition puts a cap on the angle which it will make with the ground –
arbitrarily shallow angles will not be permitted.
permitted. If the friction is there however, then at some time this limiting
angle will be breached and then L will start coming out of the plane of the disk. At this point the nature of the

4
friction will change from rolling friction to sliding friction. Equation (6) examines the steady state which will be
attained if the (now sliding) friction is again turned off after the shallow-angle motion with L out of the plane is
well developed. Of course, friction is essential to drive the disk into the range of angle where Eq. (6) supersedes
Eq. (4). In other words, a steady state Eq. (4) corresponds to the disk being started off without friction while a
steady state (6) corresponds to friction acting on the disk for some time and then being turned off. In reality of
course, friction will neither be nonexistent nor can it be turned off, but if its magnitude (both rolling and sliding) is
small, I can assume that the actual motion will be very close to the corresponding steady state in the inviscid
condition. Clearly, when the terminal stages of the motion have to be examined, it is the form Eq. (6) and not Eq.
(4) which will be relevant.

§4. The interesting feature of Eq. (6) is that the expression for the spin rate features a sinθ term in the
denominator. This implies that the spin blows up at shallow angles, contrary to what is observed in practice
(energy balance does not allow for an infinite spin rate). Hence I now impose the condition that ψɺ be finite i.e. the
numerator in the last expression of Eq. (6) go to zero when θ 0; using the standard reductions of trigonometric
functions this implies that
1/2
 mgr  1
Ω=  . (7)
 Io − I  θ 1/2

This is quite a different behaviour from what is predicted by a naive application of the inappropriate Eq. (4).
Substituting Eq. (7) in Eq. (6) shows that the spin rate in fact goes to zero as the angle becomes shallower. This
deceleration of spin is clearly borne out by simple observation of a spinning coin or medallion (a videographic
example is given in Reference [12]) and has been documented by Caps et. al [8]. This answers question (b) I posed
at the end of §1. It should be noted that the accelerating precession and retarding spin rate are inimical to the
condition of rolling without slipping which stipulates a proportionality relation between these two rates.

After relaxing the rolling without slipping assumption, the dynamics near the point θ=0 is of considerable interest
and I now try to evaluate it. Clearly, the phenomenon to be modelled is the decrease of the normal reaction on
account of the reduction in precession rate. In the transient mode, the term mgr in the first line of Eq. (6) will get
replaced by Nr. This should give me the dynamics of N, which I should then be able to couple to the dynamics of
the CM of the disk. The latter [Eq. (1a)] is governed by

d 2z d 2θ
m = mr = −mg + N , (8)
dt 2 dt 2

where the small angle approximation has been used. The equation for N however is more tricky and I now start
making bolder approximations than just small angle [if the end result is plausible then the approximation scheme
is justified – I will show the a posteriori justification after Eq. (14)]. First, since Ω >>ψɺ I keep only the Ω2 term in
the modified Eq. (6) to get

1
N = (I o − I ) Ω 2θ  . (9)
r

A coupling with the CM equation will require dN/dt and this derivative must be evaluated by the product rule.
What is interesting and important is that there are two ways by which Ω will change : one on account of the
friction which gives say a retardation rate of β (which may or may not depend on Ω, let me keep this fluid for now)
and the other on account of change in θ. Then I get

r dN   dΩ dθ   dθ
= θ  2Ω  − β +   + Ω2 . (10)
I o − I dt   dθ dt   dt

The boldest approximation comes now : since the rate of descent of the disk is generally very slow, the normal
reaction remains very close to mg throughout. While evaluating the derivatives in Eq. (10) I assume that N is not

5
just close but exactly equal to mg i.e. the product Ω2θ is a constant (since dependences and not coefficients are key,
I will not bother about exact values any longer). Under this assumption, the coefficient of dθ/dt cancels out and
only the term involving θβΩ is left. Even so, the coupling of Eq. (8) and the simplified Eq. (10) is not easy to solve
in general. There is one exception however, which occurs if β is directly proportional to Ω (fortunately a realistic
scenario). In this case, the combination Ω2θ appears in this remaining troublesome term and makes it equal to a
constant. The coupled system becomes

d3θ
= −6 A , (11)
dt 3

where A is positive. The reasonable choice of initial conditions is that the disk was in frictionless steady state
corresponding to angle θ0 when the friction was turned on at t=0. Under such conditions, the first two derivatives
of θ at t=0 become zero, and the permissible solution is θ = − At 3 + θ0 . At time t = t 0 = (θ0 / A )
1/3
, θ clearly becomes
zero, demonstrating the finite time singularity in the system (indeed, this is the feature which has attracted the
most attention of the researchers). To get the critical exponent, I can write the solution in terms of t0 as
( ) ( )
θ = A t 03 − t 3 which leads to θ = A (t 0 − t ) t 02 + t 0t + t 2 . Finally, using the relation Eq. (7) between θ and Ω, I have

1
Ω∝ . (12)
(t 0 − t )1/2

Typically the singularity is expressed in the form Ω ∝ (t 0 − t )

−1/n
; the corresponding n from the present model is 2.
This is in good agreement with the observations of Easwar et. al. [6] and Caps et. al. who have found it to be in the
range of 2.5 to 3.3 (Moffatt [4] has found it to be 6 and Bildsten [13] has 4.5). Incidentally, using the present
approach, air drag should also produce an exponent similar to the one I have got. At the simplest level air drag
exerts a torque proportional to angular velocity, which is just what was used to obtain Eq. (10) and its solution Eq.
(11). This exponent differs from Moffatt’s result because I have relaxed his assumption of rolling without slipping
while doing the analysis. This shows that while air drag is definitely not a mandatory requirement for achieving
the motion of the disk, its presence (which is realistic) will not cause a significant deviation from the behaviour
that would have been observed in its absence, which is what is seen in the entire experimental literature.

§5. I now answer question (c) of §1 by presenting a calculation for the total time of fall of the disk, a parameter
which has not been considered by any of the earlier researchers. The fall can be divided into three stages : (i)
initial phase when disk starts from the release conditions and reaches an approximate steady-state corresponding
to rolling friction, (ii) rolling phase when rolling friction acts on the disk and it falls quasi-statically under gravity
and (iii) sliding phase in the terminal stage of the motion. Of these three phases, (ii) is the longest and as a first
approximation the total time can be set equal to the time spent in (ii). To estimate this time, I write Eq. (6) and
introduce the rolling without slipping condition ψɺ + Ω cosθ = 0 . This leads to

I I
N = − Ω 2 sinθ = − 2 Ω 2z , (13)
r r

which is similar to Eq. (8). Once again taking the derivative by the chain rule and substituting the relation between
Ω and θ I get zɺɺɺ = −2 β ΩzI / r 2 . Since Ω2z must be constant, for the case β=ξΩ, this becomes zɺɺɺ = −2ξ mg and the
time taken for z to go to zero from an initial value z0 evaluates to ( 3z 0 / ξ mg )
1/3
. And z0, or equivalently the angle
θ0, can be obtained from the release angular momentum L=L00z by solving the system

I Ω 00 sin 2 θ0 = L00 , (14a)

2
I Ω00 sin 2 θ0 = mgr , (14b)

6
where I have used the double nought subscripts to avoid confusion with o-axis components. Using typical
dimensions and weights of an Euler’s disk (r=3 cm, m=200g) and choosing ξ to match the rolling frictional
coefficient of 0.001 to 0.01 at a precession rate of 100 rpm, a time of approximately 101 to 102 seconds is obtained,
in agreement with experiments.

It should be noted that the results of this and the preceding sections validate the approximations we made when
going from Eq. (8) to Eq. (9) and Eq. (10) to Eq. (11). The assumption of Ω >>ψɺ is justified since the first of these
diverges as θ 0 while the second goes to zero in the same limit. The large time taken by the disk to fall through a
short distance shows that N must indeed remain close to mg throughout – a significant difference between N and
mg would have caused the disk to hit the ground quickly. Hence the assumption of constant N is not too bad
either.

Before concluding this Letter I would like to answer question (d) of §1, which deals with the circular motions
often executed by the disk CM during the falling process. This too has been documented by Caps et. al. [8] and can
be explained from my model. As the CM falls, the disk nutates so as to keep the lowest point still in contact with
the ground. This nutation imparts to the contact point an ‘outward’ velocity which friction will oppose with an
inward force. This force, coupled with the tangential frictional force provides the centripetal acceleration which
causes the CM to travel in a circle.

How does the motion stop ? When the angle becomes too shallow, more than one point of the disk comes into
contact with the ground and the increased surface area causes an increase in the frictional torque which causes
the precession to stop quickly. The visual feeling of a sudden stop is further enhanced because the spin has
already become very slow and the precession cannot be made out when the disk becomes horizontal and the z and
o axes coincide.

* * * * *
Acknowledgement

I am grateful to KVPY, Government of India, for a generous Fellowship. I would also like to thank the two anonymous referees, whose
comments have resulted in substantial improvement in the manuscript quality. Finally I would like to thank Keith Moffatt for discussion on this
topic.

References

[1] Z Farkas, G Bartels, T Unger and D E Wolf, “Frictional coupling between sliding and spinning motion,” Physical Review Letters 90, 248302
(2003)
[2] P D Weidman and C P Malhotra, “Regimes of terminal motion of sliding spinning disks,” Physical Review Letters 95, 264303 (2005)
[3] Euler’s Disk : The official website. http://www.eulersdisk.com/
[4] H K Moffatt, “Euler’s disk and its finite time singularity,” Nature 404 (6780), 833-834 (2000)
[5] G van den Engh, P Nelson and J Roach, “Analytical dynamics : Numismatic gyrations,” Nature 408 (6812), 540 (2000)
[6] K Easwar, F Rouyer and N Menon, “Speeding to a stop : the finite time singularity of a spinning disk,” Physical Review E 66, 045102 (2002)
[7] D Petrie, J L Hunt and C G Gray, “Does the Euler disk slip during its motion ?” American Journal of Physics 70 (10), 1025-1028 (2002)
[8] H Caps, S Dorbolo, S Ponte, H Croisier and N Vandewalle, “Rolling and slipping motion of Euler’s disk,” Physical Review E 69, 056610 (2004)
[9] S Bhattacharjee, “Rotating frame analysis of rigid body dynamics in space phasor variables,” American Journal of Physics 81 (7), 518-526
(2013)
rd
[10] H Goldstein, C P Poole and J L Safko, “Classical Mechanics,” 3 Ed. Pearson, Upper Saddle River, New Jersey, USA (2002) : Chapter 4
[11] J Peraire and S Windall, “3D Rigid body dynamics : tops and gyroscopes,” MIT Open Course Ware 16.07 Dynamics (2009)
[12] The video demonstrates the discussion graphically
[13] L Bildsten, “Viscous dissipation for Euler’s disk,” Physical Review E 66, 056309 (2002)