Chloe Elliott

The Spinning Top

April 30th 2009

Abstract
In this project I will look into the phenomenon of precession, in particular into
the motion of the spinning top, and further onto how precession allows the
Levitron top achieve stable levitation over a magnetic base.

Contents

1 Introduction............................................................................................4
1.1 Context and Motivation.....................................................................4

2 Rigid Bodies and the Kinematics of their Motion.........................6

2.1 Independent Coordinates of a Rigid Body.........................................6
2.2 Orthogonal Transformations..............................................................8
2.21 Transformation Matrix Properties..............................................9
2.22 Further Properties of Transformation Matrices.........................13
2.3 The Euler Angles..............................................................................14
2.4 Euler's Theorem on the Motion of a Rigid Body..............................15
2.41 Proof of Euler's Theorem..........................................................15
2.5 Innitesimal Rotations.....................................................................19
2.6 Rate of Change of a Vector..............................................................21

3 Equations of Motion of Rigid Bodies..............................................23

3.1 Angular Momentum of a Rigid Body About a Fixed Point.............23
3.2 Tensors.............................................................................................24
3.3 Moment of Inertia Tensor.................................................................25
3.4 The Principal Axis...........................................................................26
3.5 Euler's Equations of Motion............................................................28
3.6 Force-Free Motion of a Rigid Body..................................................29
3.61 Poinsot's Construction..............................................................30
3.62 Symmetric Rigid Bodies............................................................31
3.7 The Heavy Symmetrical Top With One Point Fixed.......................33
3.71 The Lagrangian and the Generalised Momenta........................33
3.72 The Energy Equation................................................................34
3.73 Motion of the Spinning Top......................................................36
3.74 Uniform Precession...................................................................37
3.75 The Sleeping Top......................................................................38

4 The Levitron........................................................................................40
4.1 Introduction to the Levitron............................................................40
4.2 An Overview of How the Levitron Works........................................40
4.3 Conditions for Stable Equilibrium...................................................42
4.4 The Magnetic Field on the Axis.......................................................44
4.5 Adiabatic and Geometric Magnetism...............................................47
4.6 The Eect of Geometric Magnetism on Stability.............................49
4.7 Adiabatic Conditions........................................................................50
4.8 Analogy with Microscopic Particle Traps.........................................52

5 Conclusion.............................................................................................53
Bibliography.............................................................................................55
3

Chapter 1

Introduction
1.1 Context and Motivation
In this project, I am going to be considering the spinning motions of rigid
bodies. I will begin by discussing the kinematics of the motion of rigid bodies,
and then go on to look at their equations of motion. I am most interested in
the phenomenon of precession.
The Spinning Top is a toy that can be spun on an axis, balancing on a
point. It is one of the oldest recognisable toys found on archaeological sites, and
it seems to have originated independently in cultures all over the world. The
action of the spinning top is reliant on a gyroscopic eect for its motion. This
eect can sometimes suggest counter-intuitive ideas, and it certainly amazed
physicists Niels Bohr and Wolfgang Pauli!

Figure 1.1: Pauli & Bohr marvel at the spinning top1

All rotating objects can undergo precession. The gyroscope is a device which
is used for measuring or maintaining orientation. The classic image of a gyroscope is a relatively large rotor suspended in light supporting rings, known as

gimbals, which have frictionless bearings and ensure that the central rotor is

1 image

obtained from www.damtp.cam.ac.uk/user/tong/dynamics/three.pdf

4

isolated from external torques. The gyroscope is capable of extreme balanced

stability when at high speeds, and can maintain the direction of the high speed
rotation axis of the central rotor.

Figure 1.2: A gyroscope2

The angular momentum of the rotor maintains its magnitude in absence of
external torques, and the orientation also remains almost xed, regardless of
the motion of the platform on which it is mounted.
The classical gyroscope has several applications, with the gyrocompass being
arguably the most important. Gyrocompasses were rst developed around 1906.
They make use of a fast-spinning wheel in addition to friction forces in order
to exploit the Earth's rotation and enable true north to be found.

They are

widely used as navigation devices on ships and aircrafts, and were particularly
important in the early 1900's when they had signicant military uses.
There are many more examples of gyroscopic motion: The wheels of bicycles,
the spin of the Earth in space and even the behaviour of a boomerang all exhibit
this type of motion.
The Levitron is an amazing anti-gravity top which is exceptional in that
Earnshaw's theorem seems to indicate that its operation is impossible, and the
majority of scientists thought that it would never be able to work. Precession
is in fact responsible for the Levitron being able to apparently oat in mid-air
over a magnetic base. I will be looking closely into the Levitron and attempting
to explain the principles behind its operation in this project.

2 image obtained from http://commons.wikimedia.org/wiki/File:3D_Gyroscope.png

Chapter 2

Rigid Bodies and the Kinematics of their

Motion
A rigid body can be described as a system of mass points subject to the
constraint that the distance between any 2 points of the system must remain
permanently xed.

In this chapter we will be considering the nature of rigid

body motion.

2.1 Independent Coordinates of a Rigid Body

It is convenient to consider a coordinate system whose axes point along xed
directions in the body. We shall call this the body-xed system. The origin is
chosen to be some point held xed in the body.
A rigid body of N particles has up to 3N degrees of freedom, however, there
is the constraint that the distance between the particles must remain xed:

| ri rj |= rij
where

rij

is the distance between the

ith

and

cij

j th

(2.1)
particles and

cij

is a

constant.

Figure 2.1: A rigid body has xed distances between its points3
1
2 N (N - 1) constraint equations exist, but these constraint
equations are not all independent. If we x a point in a rigid body, we only
For N particles,

need to specify distances to any three non-collinear points. Once these rst three
points are xed, any additional particle in the body gives 3 new coordinates, but
also three new constraint equations, so the net increase in degrees of freedom is
zero. There is therefore a maximum of 9 degrees of freedom for a rigid body.
The rst three particles we chose have xed distances between them:

r12

c12

r23

c23

r13

c13 .

3 image obtained from http://www.damtp.cam.ac.uk/user/tong/dynamics/three.pdf

So now the number of degrees of freedom is reduced to six.

We now know that six coordinates are required to completely specify the
position of an unconstrained rigid body in 3-dimensional space: three of these
are cartesian coordinates which we need to specify position of the centre of mass
of the body axes and three are angles needed to specify the orientation of the
body axes relative to a co-incident external set of axes.
Additional constraints on the body could reduce the number of independent
coordinates further.
Many dierent ways exist of specifying the orientation of one cartesian system of axes with respect to another. A popular and useful method is to specify
the direction cosines of a set of primed axes relative to unprimed axes. If we set

x, y, z

the external axes to be labelled as

axes, the

x0

axis in the body frame

can then be dened by its three direction cosines.

Convention is to label the three unit vectors along
the unit vectors in the body system as i', j' and
cosines for

x0

k'.

x, y, z

as i,

and

k,

and

We then dene the direction

as:

cos11

cosi0 i

i0 i

cos12

cosi0 j

i0 j

cos13

cosi0 k

(2.2)

i0 k.

The relation between unit vectors in the two systems can be expressed as:

i0 = cos11 i

cos12 j

cos13 k

j0 = cos21 i

cos22 j

cos23 k

k
An arbitrary vector,

v,
v

cos31 i

cos32 j

(2.3)

cos33 k.

can be expressed in the two systems as:

xi

yj

zk

x0 i 0

y 0 j0

z 0 k0

(2.4)

and the relationship between primed and unprimed coordinates is:

x0 = cos11 x

cos12 y

cos13 z

y 0 = cos21 x

cos22 y

cos23 z

z0

cos32 y

cos33 z

cos31 x

with similar inverse relations.

Because the primed axis are xed in the body, the nine direction cosines
will change in time as the body rotates. The direction cosines can be thought
of as generalised coordinates describing the orientation of the body, however,
they cannot all be independent since we know that only three coordinates are
required to specify an orientation. The relations between the direction cosines
are due to the basis vectors being orthogonal to each other in each of the two
coordinate systems, and all having unit magnitude:

(2.5)

with similar relations for

ij

jk

ii

jj

i0 , j0

k0 .

and

ik

= 0

kk

= 1

=
=

(2.6)

Combining the set of equations (4.3)

with the dot-product relations, we can then dene six orthogonality relations
between the direction cosines:

3
P

cospq cosp0 q

p,q

= 0

= 1, 2, 3,

p 6= p0

q=1
3
P

cos2 pq

= 1.

(2.7)

q=1
We can use the Kronecker delta function,

p0 p

= 1
= 0

pq
p

, dened by

p0

p 6= p0

to reduce the orthogonality relations to:

3
P

cospq cosp0 q

p0 p .

(2.8)

p=1
Due to there being six orthogonality parameters between the nine direction
cosines, the number of independent coordinates has now been reduced to just
three.
The direction cosines are therefore convenient and useful to use as a method
of specifying the relative orientation of a cartesian coordinate system to another
coincident system, but they cannot be used as generalized coordinates to set up
a Lagrangian. We must instead choose some set of three independent functions
of the direction cosines. Although not unique, the Euler angles are a popular
choice for this set, and we will look at these in more detail later.

2.2 Orthogonal transformations

A general linear transformation equation can be written as

x0
where

Ax

(2.9)

is the matrix of transformation which can be considered as an

operator which transforms the unprimed system into the primed system. The
vector remains unchanged;

acts on the coordinate system only, yielding the

coordinates of the vector components in the new, primed frame.

elements

has matrix

aij.

Denoting the

x, y , z

axes instead as the

x1, x2, x3

axes will prove to be

advantageous here. In matrix form, the linear transformation can be written as:


a11
x01
x02 = a21
x03
a31

a13
x1
a23 x2
a33
x3

a12
a22
a32

(2.10)

to give three transformation equations. The set of equations (2.5) is a special

case of a linear transformation, as there is not independence of all the direction
cosines. We can re-derive the connection between the cosine coecients in terms
of our new

aij

notation.

We are able to express each of the primed coordinates as a summation:

x0i
The actual vector

3
P

aij xj

= 1, 2, 3.

(2.11)

i=1

remains unchanged in both the old and new coordinate

x (given in terms of the sum of squares

systems, and so magnitude of the vector

of the components) therefore also remains unchanged in both systems:

3
P

x0i 2

i=1

3
P

x2i

(2.12)

i=1

Using the transformation equation (2.11), this now becomes

 3
3
P
P
i=1


aij xj

i=1

3
P


aik xk

3 P
3
P

i=1

aij aik xj xk .

(2.13)

i=1j,k=1

We can rearrange the summations and write this expression as

3
P
j,k=1

3
P


aij aik xj xk

(2.14)

i=1

which is equal to the right hand side of equation (2.12) if and only if

3
P

aij aik

jk

for

j, k

= 1, 2, 3

(2.15)

i=1
where

jk

is the Kronecker delta. If we express the

aij

components in terms of

the cosine coecients, we now obtain a set of equations identical to equations

(2.8).
Equation (2.15) is known as the orthogonality condition.

The transition

between coordinates xed in space and coordinates xed in the rigid body is
achieved by means of an orthogonal transformation.

2.21 Transformation Matrix Properties

We will now think about what will happen when we perform two successive orthogonal transformations. Physically, this corresponds to the rigid body
undergoing two successive displacements.

We can denote the rst transformation as

x0

B:

Bx.

(2.16)

In matrix form,

x01
x02
x03

b11
b21
b31

b12
b22
b32

b13
x1
b23 x2 .
b33
x3

We will denote the second transformation from

00

by

x0

(2.17)

to the next coordinate set

A:
x00

Ax0 = ABx

(2.18)

Or in matrix form:

x001
x002
x003

0
a11 a12 a13
x1
a22 a23 x02
= a21
a31 a32 a33
x03

a11 a12 a13

b11 b12
a22 a23 b21 b22
= a21
a31 a32 a33
b31 b32

b13
x1
b23 x2 .
b33
x3

We can also use component form to show that the application of

(2.19)

on system

is:

xk

P
bkj xj

(2.20)

j
and the application of

on

x00

is:

x00i

P
aik x0k
k

so combining these two equations above gives:

x00i

aik

P
bkj xj

(2.21)

k
which is the same as

x00i =



P P
aik bkj xj .
j

(2.22)

If we set

cij

P
aik bkj
k

then we are able to rewrite equation (2.21) as:

10

(2.23)

x00i

P
cij xj .

(2.24)

We can therefore dene an orthogonal transformation

elements of the square matrix

AB

where the

are given by the equation (2.23). This tells us

that the successive transformation of orthogonal matrices

equivalent to a third linear transformation

A and B is completely

C.

It is worth noting that this matrix multiplication is generally not commutative:

AB 6= BA.

(2.25)

Matrix multiplication is, however, associative:

(AB)C =

A(BC).

(2.26)

In the addition of two transformation matrices:

and the elements of the matrix are

cij

aij

(2.27)

bij .

The inverse transformation is one by which the new, primed system is transformed into the old, unprimed system, as shown by the following equation:

xi

P
j

where

a0ij

a0ij x0j

are the elements of the inverse transformation matrix

(2.28)

A1

This must be consistent with the set of equations:

x0k
We can then substitute the

xi

P
aki xi .

from (2.28) into (2.29) to obtain:

x0k

P P 0 0
aki aij xj
i

P P
aki a0ij xj
j

which, since the

(2.29)

x0 components

(2.30)

are independent, can only possibly be correct if:

P
aki a0ij

kj .

(2.31)

If we let

and

A1

be the matrices with elements denoted by

respectively, we see that equation (2.31) can instead be written as:

11

aki

and

a0ij

AA1
where

(2.32)

is known as the identity transformation:

Because

1x,

1
0
0

0
0 .
1

0
1
0

(2.33)

no change is produced in the coordinate system.

It can be shown that

and

A1

are commutative:

1A= (AA1 )A

A(A1 A)

A1

(2.34)

and therefore

AA1
The transpose,

AT

of matrix

A1 A

1.

(2.35)

can be obtained by interchanging the rows

and columns:

a11
a21
a31

a12
a22
a32

a13
a23
a33

AT

a11
a12
a13

a21
a22
a23

For an orthogonal transformation, it can be shown that

a31
a32 .
a33

A1

AT

(2.36)

. We

can prove this by rst considering the sum:

P
akl aki a0ij .
k,i
If we rst sum over

k,

we obtain the expression:

P
i

If instead we sum over

P
akl aki a0ij .
k

rst, we get:

P
aki a0ij akl .
i

We can then use the earlier obtained orthogonality condition to reduce the
rst sum to

P
il a0ij
i

a0lj .

The second sum can also being reduced, using equation (2.31):

12

P
kj akl

ajl .

k
This indicates that the elements

aij

of

a0ji

and

of the reciprocal

A1

are

related by the equation

a0ij

AT

A1

and we have thus proved that

aji

(2.37)

for orthogonal matrices. We can

then substitute this result into the equation (2.35) to get the relation

AT A= 1

AAT .

(2.38)

2.22 Further Properties of Transformation Matrices

We already know that if

x0 = Ax.
0
vector x

B,

the

in this new coordinate system would now be:

Bx0 = BAx
where

is an orthogonal transformation matrix, then

If we then transform the coordinate system by another matrix,

Bx0

is the vector

r in the new
matrix A under the
vector

x0

Bx(B-1 B)x

-1 )Bx

= (BAB

(2.39)

Bx is the
A0 = (BAB-1 ) is the transformation
system B. Any transformation of a matrix

in the new coordinate system, and

system. The matrix

new coordinate

having this form is known as a similarity transformation.

The determinant of a matrix is notated as:
det(A) =
For two square matrices,

A and B,

| A |.

(2.40)

the following rule applies for their deter-

minants:

| AB | = | A || B |
If

(2.41)

is an orthogonal transformation matrix, then since the determinant of

a unit matrix is 1 and the orthogonality condition indicates that

=

1,

AAT

AT A

it becomes apparent that

| A || AT |

| AT || A |

= 1,

(2.42)

implying that interchanging columns or rows of an orthogonal matrix does

not have an eect on the determinant, i.e.

| A |=| AT | .

This then tells us that

we can write:

| A |2

= 1

which then allows us to conclude that

|A|

(2.43)
=

: there are only 2 possi-

bilities for the determinant of an orthogonal matrix: +1 or -1.

Finally, we can show that the value of a determinant is invariant under a
similarity transformation, by nding the determinant of both sides of a similarity transformation equation. For orthogonal matrices

-1
(BAB ), this gives:

13

and

B,

where

A0 =

| A0 |= | B || A || B-1 |= | A | (| B || B1 |)

| A |.

(2.44)

2.3 The Euler Angles

As discussed earlier, we only require three independent parameters to describe the orientation of a rigid body.

There are many choices of generalised

coordinates that could be used to describe an arbitrary rotation of a coordinate system from one orientation to another, but Eularian angles are the most
common and useful choice.
The transformation from an initial cartesian coordinate system to another
can be achieved by means of three successive rotations performed in a specic
sequence, with the three successive angles of rotation being dened as the Eularian angles.

Figure 4.2: The rotations which dene the Eularian angles4

xyz ,

The initial system of axes,

is initially rotated anticlockwise by an angle

about the z-axis, giving a transformation matrix

cos sin
sin cos
0
0

D:

0
0 .
1

(2.45)

This transformation results in a new system of axes, which will be labelled

the

axes. The new set of axes are then rotated anticlockwise about the

axis by an angle

This transformation is given by the matrix

1
C = 0
0

0
cos
sin

0
sin
cos

(2.46)

This produces the intermediate set of axes, to be labelled the

The intersection of the

xy

and

0 0

C:

planes is the

0 0 0 axes.

axis, and it is called the line

of nodes.
The

0 0 0

axes are then rotated anticlockwise by an angle

axis, this transformation is given by the matrix

cos
B = sin
0

sin
cos
0

about the

B:

0
0 .
1

4 image obtained from http://tabitha.phas.ubc.ca/wiki/index.php/Rigid_Bodies

14

(2.47)

x 0 y 0 z 0 system.

This gives the nal set of axes, the

The Eularian angles
the

x 0y 0z 0

system to the

mation matrix

, and therefore specify the relative orientation of

xyz system completely, and the the complete transfor-

A,where
x0 =Ax

(2.48)

is simply the product of the three successive matrices:

A = BCD

(2.49)

where

coscos cossinsin
= sincos cossincos
sinsin

cossin + coscossin
sinsin + coscoscos
sincos

sinsin
cossin .
cos

(2.50)

2.4 Euler's Theorem on the Motion of a Rigid Body

At any instant, the orientation of a body can be specied by an orthogonal
transformation matrix,

R.

Orientation will change as time progresses, and the

R(t).

matrix is therefore a function of time,

We can choose the body-xed axis to be coincident with the space axis at

= 0, so that

R(0)

where

is the identity matrix as before. Physically,

R(t)

the motion must be continuous, so

is a continuous function of time and it

evolves continuously from the identity transformation.

Euler's theorem states that any transformation in the 3-dimensional real

space which has at least one xed point can be described as a simple rotation
about a single axis. This axis, through the xed point, means that the result of
several consecutive rotations can be replaced by a single rotation. The axis of
rotation is unaected by the operation and so any vector lying along the axis of
rotation must have identical components in both the initial and nal axis.

2.41 Proof of Euler's Theorem

To prove Euler's theorem, we must show that there exists a vector
the same components in both initial and nal systems i.e.
vector

unchanged by transformation

v0

having

there must exist

R.

Rv

v.

(2.51)

This is a special case of the more general equation,

v0

Rv

15

(2.52)

where
of

is an unspecied constant, which is allowed to be complex. The values

which allow equation (2.52) to be solved are called the eigenvectors of the

matrix. The vector solutions are known as the eigenvectors of

R.

We are now able to state an alternative form of Euler's theorem:

The real orthogonal matrix specifying the physical motion of a rigid body with

1.

one point xed always has eigenvalue

We can therefore rewrite equation (2.52) as

(R

1)v

= 0

(2.53)

v = (X1 X2 X3 )T (where we have chosen to denote the

coordinates x, y, z as X1 , X2 , X3 ) to give 3 homogeneous simultaneous equations
and restate it with

with 4 unknowns:
(r11

r21 X1
r31 X1

)X1
+
+

r12 X2

r13 X3

= 0

(r22 )X2

r23 X3

= 0

r32 X2

(r33 )X3

(2.54)

= 0.

From this set of equations (2.54), we are unable to exactly specify

and

X3 , but they do allow us to nd ratios of the components.

X1 , X2

This corresponds

to the situation where only the direction of the eigenvector can be xed and the
magnitude remains unknown, i.e. if

is an eigenvector then so is

tv

where

is

any scalar constant.

We only have non-trivial solutions to the set of equations (2.54) when the
determinant of the coecients is zero:


 r11

|R - 1|= r21
 r31

r12
r22
r32

r13
r23
r33

= 0

(2.55)

Equation (2.55) is known as the characteristic equation of the matrix. The

values of

which allow the characteristic equation to be satised are the required

eigenvalues.
We are now once again able to restate Euler's theorem: the characteristic

equation must have the root

1 .

Generally, we will have 3 roots with 3

eigenvectors. The components of the eigenvectors may be labelled

Xik

, where

the rst subscript indicates which particular component is being considered, and
the second subscript indicates which of the three eigenvectors is involved.
The set of equations (2.54) would then have typical members written as

P
rij Xjk

j
We can then expand this for each

k:
16

k Xik .

(2.56)

r11 X1k + r12 X2k

r13 X3k

k X1k

r21 X1k + r22 X2k

r23 X3k

k X2k

r31 X1k + r32 X2k

r33 X3k

k X3k .

(2.57)

We are now able to put this into matrix form:

X11
X21
X31

X12
X22
X32

0
2
0

X13
X23
X33

(2.58)

and

1
0
0

0
0
3

(2.59)

to give a new expression for the matrix equation:

RX = X.

(2.60)

This allows us to write

X1 RX
by simply multiplying equation (2.60) by
are trying to diagonalise

X1

(2.61)
on the left hand side. Here, we

by a similarity transformation equation. The

elements of the diagonal matrix obtained by transforming

will then be the

eigenvectors we are looking for. Euler's theorem for rigid body motion with
one point xed can thus be proved by using the orthogonality properties of

R.

We are trying to solve the characteristic equation, so we will begin by considering

(R

1)RT

RT R

1 RT

(2.62)

RRT

(2.63)

and then take the determinant of both sides of equation (2.62):

| R 1 || RT |

| 1 RT |.

In order to describe rigid body motion, the matrix

(2.64)

must correspond to a

proper rotation, i.e.

|R|

| RT |

= +1

(2.65)

and we can therefore rewrite equation (2.64) as

|R-1|

| 1 R |.

17

(2.66)

We can set

R - 1,

and then rewrite the above equation again as:

|D|
If

| D |.

(2.67)

is an n x n matrix, then it has the well-known property that

| D |

= (1)

|D|

(2.68)

and so for our 3 x 3 matrix,

| D |

|D|

(2.69)

and therefore equation (2.67) can now be written as

|D|
Obviously this can only be true for

| D |.

| D |=

(2.70)

0 and therefore,

can then solve the characteristic equation (2.55) with

|R - 1|= |R
This implies that

|R - 1|=

0 . We

= +1:

- (+1)1|= 0.

+1 is an eigenvalue of

R.

(2.71)
We shall now seek to nd

is the product
1 2 3 . The determinant of a matrix is unaected by
1
a similarity transformation. X
RX = is a similarity transformation and
therefore the determinant of R is equal to the determinant of :
the remaining two eigenvalues. The determinant of the matrix

of the three eigenvalues

| R |= 1 2 3
We already know that

|R|

(2.72)

= +1 and that one of the eigenvectors,

say, is

equal to +1, and we are left with the equation:

1 2
We know that

is real, and so if

then its complex conjugate

= +1.

(2.73)

is a solution to the characteristic equation,

is also a solution.

It is a property of complex numbers that

k c k=k c* kfor c  C.

From this we

know that all the eigenvalues have unit magnitude.

All three eigenvalues of the real orthogonal matrix with determinant +1 may
be real. There is the trivial solution that

1 = 2 = 3

= 1. There is also the

possibility for there to be one real root and two complex roots. The real root
is

3 =

+1 and the two complex roots must be complex conjugates and have

product +1, to give the correct determinant. There is thus one and only one
+1 eigenvalue in any non-trivial physical transformation. This is the statement
of Euler's theorem.
We can obtain the direction cosines of the axes of rotation by nding the
eigenvector associated with

= +1 in equation (2.53) and scaling it to unit

length. We can nd the angle of rotation,

without too much diculty. We

are able to use a similarity transformation to transform

R to a coordinate system
R0 represents the

where the z-axis lies along the axis of rotation. The matrix
rotation through the angle

about the z-axis. It has the form

18


R0

The trace of

R0

sin 0
cos 0 .
0
1

cos
sin
0

(2.74)

is then
1 + 2cos.

(2.75)

Under a similarity transformation, the trace is always invariant. This must

therefore also be the value of the trace of

P
rii

R:

= 1 + 2cos.

(2.76)

i
We can then solve for

Chalses' theorem immediately follows from Euler's theorem, and it states

that the most general displacement of a rigid body is a translation plus a rotation.

2.5 Innitesimal Rotations

A nite rotation about the z-axis has a rotation matrix of the form

0
0 .
1

cos sin
A = sin cos
0
0

(2.77)

Two rotations performed one after the other, i.e. the addition of two rota-

AB. Non-commutativity
A,B are not commutative in

tions, corresponds to the product of the two matrices,

of matrix multiplication allows us to conclude that

addition and thus cannot be considered as vectors. This leads to the conclusion
that the sum of nite rotations is dependent on the order in which the rotations
are carried out.

The same is not true if we consider innitesimal rotations.

Goldstein denes an innitesimal rotation as: an orthogonal transformation of

coordinate axes in which the components of a vector are almost the same in both
sets of axes - the change is innitesimal .
An innitesimal transformation of a vector

x0

is given by

x0 = (1 + )x
where the vector

is the identity matrix and

(2.78)
is innitesimal.

It can be shown that two innitesimal transformations

1 and 2

are com-

mutative, by the equivalence of

(1 + 1 )(1 + 2 )

12 +1 1

12

1 2 1

(1 + 2 )(1 + 1 )

12 +2 1

11

2 1 1

19

(2.79)

We can dene

A 1+

as the transformation matrix, and the inverse of

this is

A-1 = 1

(2.80)

since

AA-1
For small angles, sin

(1 + )(1 )

and cos

12 1.

(2.81)

1. From this we can obtain the matrix

of innitesimal rotation:

1+
And the innitesimal matrix

d 0
1 0 .
0 1

1
d
0

(2.82)

is therefore

0
d 0
= d 0 0
0
0 0

0 1 0
0 0 .
= d 1
0 0 0

(2.83)

Since the innitesimal transformation is dened to be a rotation, orthogonality

of rotation matrices requires

AT
but

A-1

1 ,

A-1

(2.84)

so

T =

(1+)

1T

1+T

(2.85)

and so

= -

(2.86)

which is the denition of an antisymmetric matrix.

is an antisymmetric matrix, the diagonal elements are therefore

.
matrix must therefore be of the form

0
d3 d2
0
d1
= d3
d2 d1
0

Given that

equal to zero and there can only possibly be three distinct elements in
The

where

d1 , d2

and

d3

are quantities associated with the three independent

parameters responsible for specifying the rotation of the body.

20

(2.87)

When the innitesimal rotation matrix is applied to the vector

ential change in

x,

the dier-

is given by

dx =x0 x

= (1+)xx =

x.

(2.88)

In matrix form, this is

dx

0
d3
d2

d2
x
d1 y
0
z

d3
0
d1

yd3 zd2
zd1 xd3
xd2 yd1

and so

dx

= (yd3 xd2 )i + (zd1 xd3 )j + (xd2 yd1 )k

=

x d

(2.89)

This implies that

dx
dt rotation =

=
The vector

d
dt

(2.90)

here is the angular velocity of the body. It is dened as the

instantaneous angular rate of rotation of the body,

d
dt .

(2.91)

2.6 Rate of Change of a Vector

If we set

as an arbitrary vector involved in describing motion of a rigid

body with time, how the vector will vary in time as the body is in motion will be
dependent on the coordinate system of observation. Only the eects of rotation
of the body axes will result in the components of
axes diering from the components of

with respect to the body

with respect to the space axes.

This allows us to write:

(dR)body

= (dR)space

+ (dR)rotation .

(2.92)

But as we saw before,

(dR)rotation =

21

R d.

(2.93)

The dierential

dR

observed in the space set therefore is related to the same

dierential when observed in the body set by the following equation:

(dR)space

= (dR)body

d R.

We can then obtain the rate of change of the vector

dR
dt

space =

dR
dt

body +

(2.94)

with respect to time:

R.

(2.95)

This can be written as an operator equation due to the arbitrary nature of

d
dt

space =

d

dt body +

R:

(2.96)

Caution should be taken when taking the time derivative of a vector with
respect to one coordinate system, as components can only be taken along a
dierent set of coordinate axes

after

dierentiation has been carried out.

It will prove useful to be able to use the Euler angles and their time derivatives to express the angular velocity vector,

.We

can consider the general

innitesimal rotation associated with the angular velocity vector to be made up

= ,

= and = . The angular velocity vectors do not have symmetrically

placed directions: is along the space z-axis, points along the line of nodes
0
and is along the body z -axis. In order to obtain as the sum of the three
of three sequential innitesimal rotations with the angular velocities

angular velocities we need to use the orthogonal transformations described in

section (2.3) to get the components of the vectors

and

along any set

of axes we desire.
We shall obtain the components of

for the set of body axes, as this will

be the most useful coordinate system to consider.

Being parallel to the space z-axis,
formation

BCD

requires the complete orthogonal trans-

to be applied to obtain its components along the body

axes:
( )x0 =

sinsin

Since the direction of

0 -axis,

( )y 0 =

sincos

(2.97)

along the body axes by applying the

last orthogonal transformation in the sequence,

The vector

cos
.

lies along the line of nodes, which coincides with the

we can nd the components of

( )x0 =

( )z 0 =

cos

lies along the

( )y 0 =

z 0 -axis

(2.47):

sin

( )z 0 = 0.

(2.98)

and so does not require an orthogonal

transformation. We now are able to add the separate angular velocity

components with respect to the body axes and we obtain:

x0

sinsin

y0

sincos
-sin

z0

cos

22

cos
(2.99)

Chapter 3

Equations of Motion of Rigid Bodies

In this chapter we will look into the nature of rigid body motion, using the
ideas and techniques explained in the previous chapter. We will be using the
Euler angles as generalised coordinates and will apply other tools previously
described to obtain the equations of motion for rigid bodies.

3.1 Angular Momentum of a Rigid Body About a Fixed Point

In general rigid body rotation, every particle lies a xed distance from the
origin and a xed angle to the rotation axis. The velocity

vi

of a rigid body

about a xed point is therefore

vi = ri
where

ri

is the radius vector of the

ith

(3.1)

particle, relative to the xed point.

Thus the total angular momentum of the rigid body about the xed point, given
by

J=

P
mi (ri vi )

(3.2)

i
can now be written as

P
J = mi (ri ( ri

)).

(3.3)

i
We can expand the triple cross product to obtain

J=

P
mi (ri 2

- (ri

i
Since

lies in the same plane as

r,

it rotates as

ri

rotates.

J are given by
P
P
x mi (ri2 x2i ) - y mi xi yi

).

(3.4)

The components of

Jx

P
P
y mi (ri2 yi2 ) - x mi yi xi

P
z mi yi zi

P
P
mi (ri2 zi2 ) - x mi zi xi

Jy

Jz

P
mi xi zi

(3.5)

P
mi zi y

Angular momentum of the body is therefore linearly related to its angular

velocity.

23

The linear transformation can be written in matrix form:

Jx
Jy
Jz
so that

Jx

Ixx x

Ixx
Iyx
Izx

Ixy y

Ixz z

Ixz
x
Iyz y
Izz
z

Ixy
Iyy
Izy

and similarly for

Jy

and

(3.6)

Jz .

The elements Ixx , Iyy and Izz are known as the moment of inertia coecients. For a rigid body made up of discrete particles,

Ixx

P
mi (ri2

with similar relations for

Iyy

x2i ).

Izz . For continuous

(r)(r2 x2 )dV
=

and

Ixx

(3.7)

bodies,
(3.8)

dV
where

(r)

is the mass density of the body.

The rest of the elements of the elements are known as products of inertia
and they have the form given by

Ixy

P
m i xi yi .

(3.9)

and

are

vectors that are independent of our choice of axes. The relation

between them can therefore be expressed as

J
where

(3.10)

is dened as the moment of inertia tensor.

3.2 Tensors
A tensor acts on a vector to produce a new vector which is linearly related
to the old one but will, in general, have a dierent direction. The action of a
tensor on a vector

can be denoted by

T a

(3.11)

which can be written as

b1
T11
b2 = T21
b3
T31

T12
T22
T32

or

24

T13
a1
T23 a2
T33
a3

(3.12)

Ti

P
Tij aj .

(3.13)

j
Familiar methods of multiplying two vectors are by means of a dot product
or cross product, but there is a third type of multiplication, a dyadic product,
which will be useful here.

, produces a tensor:

a1 b1 a1 b2 a1 b3
a b = a2 b1 a2 b2 a2 b3 .
a3 b1 a3 b2 a3 b3

A dyadic product, denoted

(3.14)

A tensor of this type is known as a tensor of the 2nd rank. A tensor of zero
rank has only one component, which is invariant under orthogonal transformation.

A scalar is a tensor of zero rank.

A tensor of rst rank is completely

equivalent to a vector.
The most important property of a vector is the way its components transform
under a rotation of the coordinate axes; keeping its geometrical or physical
meaning invariant.

3.3 Moment of Inertia Tensor

The kinetic energy of energy about a point is given by

T
where

vi

1
2

P
mi vi2

(3.15)

is the velocity relative to a xed point and

vi = ri .

This allows us to write the kinetic energy as

1
2

mi vi

ri )

(3.16)

which is equivalent to

Pm r v
i

( i

i)

(3.17)

and so the rotation kinetic energy corresponding to equation (3.10) has the
form:

T
If we set

1
2

(3.18)

to be a unit vector in the direction of

we are then able to dene

, and can rewrite the kinetic energy in the form

1 2
2

nIn

25

1
2
2 I

(3.19)

where we dene

I,

the moment of inertia about the axis of rotation, as

nIn

P
mi (ri2

- ( ri

n)2 ).

(3.20)

The direction of the axis of rotation has an eect on the value of the moment
of inertia. The direction of

changes in time with respect to the body, and so

the moment of inertia must also be a function of time. It is possible to keep the
moment of inertia constant by constraining the body to allow it to rotate only
about a xed axis. If this constraint is applied then the form of kinetic energy
in equation (3.19) is very close to the form we need to set up the Lagrangian
for the system. The last step required is to express

as the time derivative of

some angle. This is usually possible without too much diculty.

3.4 The Principal Axis

We seek to show that any rigid body has three principal axes that can be
chosen so that they are always mutually orthogonal. The equation (3.9) dening
the products of inertia shows that the components of the inertia tensor are in
fact symmetrical, that is

Iij

Iji .

(3.21)

This indicates that the only six coordinates of the tensor are independent.
We have already seen the relationship between a rigid body's angular momentum and angular velocity in equation (3.10), where

is now known to be a

tensor of the second rank, known as the moment of inertia tensor. If the angular
momentum of a body is parallel to its angular velocity, then the rigid body is
described as being dynamically balanced :

J
where

is some scalar number. The angular velocity

(3.22)

must point along a

principal axis of the moment of inertia tensor for this to be true. The value of

is then called the principal moment of inertia.

The eigenvectors of the tensor

provide the principal axes, while the eigen-

values give the principal moments. We know that the eigenvectors satisfy the
linear set of equations:

Ixx
Ixy
Izx
Here, the symmetry of

Ixy
Iyy
Iyz

Izx
x
Iyz y
Izz
z

has explicitly been displayed.

26

= 0.

(3.23)

Non-trivial solutions for

will exist only for the case where the determinant

vanishes:


 Ixx

 Ixy

 Izx

Ixy
Iyy
Iyz

Izx
Iyz
Izz

The characteristic equation is cubic in

the desired principal moments

I1 , I2

and

= 0.

(3.24)

, and there will be three solutions:

I3 . The eigenvectors can then be

obtained by substituting each eigenvalue back into the set of linear equations
in turn.

Only two of the equations we get will be independent and we can

then use these to work out the direction of the corresponding eigenvector. Unit
eigenvectors,

ei , ej

and

ek

will be used to specify the directions of the principal

axes, as the magnitudes of the eigenvectors are not determined.

Symmetry of the moment of inertia tensor means that its eigenvalues and
eigenvectors have similar properties to those of a real symmetric matrix. The
eigenvalues of any symmetric matrix are real. The principal moments are therefore also real.
If

ei

and

ej

are the eigenvectors corresponding to dierent principal mo-

ments, then these eigenvectors (principal axes) are orthogonal:

ei ej

= 0.

(3.25)

If there is a repeated root of the characteristic equation then the rigid body
has degenerate principal moments and any vector that is in the plane of

ej

ei

and

can be a principal axis and any pair of suitable vectors in this plane can be

chosen as principal axes. If a body has an axis of symmetry then that axis is a
principal axis and rotations around that axis will be dynamically balanced. We
can thus use the symmetries of a rigid body in order to recognise the principal
axes: If there is an axis of symmetry through the origin of the body then this
axis is a principal axis for rotations about the origin. Another principal axis is
the normal to a plane of reection symmetry through the origin.
We can use these properties along with the property that any rigid body
has three principal axes that can always be chosen so that they are mutually
orthogonal, to determine the three principal axes for many rigid bodies. If the
coordinate axes are chosen to lie along the directions of the principal axes, then
the principal moments can be found more easily, they are just the moments of
inertia about our three principal axes.
There is another method that can be used to approach the idea of principal
axes. Earlier, we dened the moment of inertia about a given axis as

In

. If we let the direction cosines of the axis be

vectors be denoted by

i, j

and

k,

n
We can then write

and

and the unit

then

as

27

k.

(3.26)

Ixx 2

Iyy 2

Izz 2 + 2Ixy

once again using the symmetry of

2Iyz

2Izx ,

(3.27)

explicitly.

It is useful to dene a new vector

as

n
=

(3.28)

and its magnitude is related to the moment of inertia about the axis whose

n. The equation of some surface in space is a function

1 , 2 and 3 :

direction is provided by
of the three variables
1 =

Ixx 21

Iyy 22

Izz 233 + 2Ixy 1 2

2Iyz 2 3

2Izx 3 1 .

(3.29)

This is the equation of the inertia ellipsoid. It is possible to transform to

a set of cartesian axes so that the ellipsoid equation takes on the normal form
with the principal axes of the ellipsoid along the new coordinate axes:
1 =

I1 01 2

I2 02 2

I3 03 2

(3.30)

Equation (3.30) is the same as equation (3.29) when the inertia tensor is
diagonal.

The coordinate transformation which turns the ellipsoid equation

into normal form is therefore the same as the previously discussed principal
axis transformation. The lengths of the inertia ellipsoid axes are determined by
the principal moments of inertia.

The inertia ellipsoid will be an ellipsoid of

revolution if two of the roots of the characteristic equation are equal because
the inertia ellipsoid will therefore have two equal axes.
In the case of all principal moments being equal, the inertia ellipsoid is in
fact a sphere.

3.5 Euler's Equations of Motion

Six generalised coordinates are required to describe the motion of an unconstrained rigid body: 3 cartesian coordinates to describe the translational motion
and three Euler angles to describe the rotational motion.
Kinetic energy can be written as the combination of the translational energy
of the centre of mass and the rotational energy about the centre of mass:

1
1
2
2
2 M v + 2 I .

(3.31)

If we work in the principal axes, rotational energy has the form

1
1
1
2
2
2
2 I1 x + 2 I2 y + 2 I3 z

(3.32)

and if one point in the body is xed then the kinetic energy only contains
rotational terms. If the forces are conservative, we can write the Lagrangian as

1
2
2 (I1 x +

I2 y2

28

I3 z2 ) - V (,,)

(3.33)

with Euler angles being used to express the components of

The components of the tensor

I
and

I1, I2, I3

form a diagonal matrix:

I1
0
0

0
I2
0

0
0
I3

(3.34)

are the principal moments of inertia.

When we are considering the motion of a rigid body with one point xed, it
is useful to use Euler's equations of motion rather than the Lagrange equations
of motion.
denition of

We can derive the Euler equations of motion by considering the

torque .

In an inertial frame, torque is equal to rate of change of

angular momentum:

dJ
dt =

(3.35)

In this case, the time derivative is referring to the space axes. In a body-xed
frame, this equation of motion becomes

dJ
dt body +

The component of the equation of motion along the

dJx
dt

(3.36)

principal axis is then

+ y Jz z Jy = x

In this body-xed frame, the tensor

(3.37)

only depends on the positions of atoms

in the body, and due to the rigidity condition on the body, these particles do
not move and

is therefore constant.

The equation

can then be used to re-write the equation of motion

as

(I )

This allows us to reduce the equation (3.37) to

I1 x y z (I2 I3 ) = x .

3.6 Force Free Motion of a Rigid Body

If we consider a rigid body subject to no net forces or torques, the centre
of mass of the body must be either at rest or moving uniformly. Without loss
of generality, we are able to consider the frame where the centre of mass is
stationary. In this case, rotation about the centre of mass is the sole motion
from which the angular momentum arises.

29

(3.38)

Euler's equations are thus the equations of motion of the system, and without
the presence of net torques they become:

I1 x

y z (I2 I3 )

I2 y

z x (I3 I1 )

I3 z

x y (I1 I2 ).

(3.39)

These equations will also apply to the motion of a rigid body with no net
torques and with one point xed.

3.61 Poinsot's Construction

Poinsot's construction is a geometric description of the motion which we are
able to derive without needing a complete solution to the problem above.

It

is a perfectly adequate method of completely describing force-free rigid body

motion.
Earlier, we dened a vector

equation (3.28), whose components are mea-

sured by the principal axes of a body, with the coordinate system also oriented
along the principal axes of the body. We can dene a function in this space:

F ()
and the surfaces of constant
dened by the surface with

(3.40)

are ellipsoids, with the inertia ellipsoid being

= 1. The vector

is parallel to the axis of rotation

and it moves accordingly as the rotation axis changes in time. There is always a
point on the inertia ellipsoid that is dened by the tip of
of

and if the gradient

is evaluated at this point then it gives the direction of the normal vector

to the inertia ellipsoid.

Earlier we saw that

has diagonal form in the principal axes. This makes it

simple for us to evaluate the partial derivative of

with respect to

1 :

F
1 = 2I1 1 .
We can also dene

(3.41)

as

(3.42)

which then allows us to write

F
2
2

1 = I I1 x = I Jx
and similar results follow for the other components of the gradient of

(3.43)

F :

F
2

2 = I Jy
F
2

3 = I Jz .

30

(3.44)

The angular velocity vector

always moves so that the normal correspond-

ing to the inertia ellipsoid is always in the same direction as the angular momentum,

J.

In this case, it is the inertia ellipsoid which moves in space, while

the direction of

is xed to preserve the relationship between the angular mo-

mentum and angular velocity.

The distance between the origin of the ellipsoid and its tangent plane at

must always be xed. This xed distance is given by

J
J
2T

J = J I = J I 2
and is the projection of

on

(3.45)

J.

This can also be written as

J
J =
where the angular momentum
of motion.

2T
J

and the kinetic energy

(3.46)

are just constants

The normal to the plane has a xed direction along the angular

momentum vector

J, and so

the tangent plane is known as the invariable plane.

The motion of the force-free rigid body can be pictured as being such that
the inertia ellipsoid rolls, on the invariable plane, without slipping, with the
ellipsoid's centre always remaining a constant height above the plane.

The

point of contact of the ellipsoid with the plane is dened by the position of
Because

is along the instantaneous axis of rotation, the body is momentarily

at rest in this direction.

The non-plane curve traced on the surface of the inertia ellipsoid with xed
centre by its point of contact with the xed plane on which it rolls is known
as the polhode.

The curve traced out on the invariable plane by the point of

contact between the plane and the inertia is known as the herpolhode.
The values of the kinetic energy

and the angular momentum

determine

the direction of the invariable plane and the height of the inertia ellipsoid above
the plane. The direction of

furnishes the direction of the angular velocity in

space and the orientation of the inertia ellipsoid, which is body-xed, provides
the instantaneous direction of the body.

3.62 Symmetric Rigid Bodies

If the body we are considering is symmetrical, the inertia ellipsoid is an
ellipsoid of revolution and the polhode is a circle on the axis of symmetry. The
vector

then moves on the surface of a cone correspondingly, and its direction

is said to precess in time about the symmetry axis of the body.

If we allow the axis of symmetry of a rigid body to be the z principal axes of
the moment of inertia tensor, then the other two principal axes can be chosen
to be any two orthogonal vectors in the plane orthogonal to the z principal axis.

31

The principal moments about these two axis are equal:

I1

I2

(3.47)

and the equations of motion for a symmetric body can then be simplied to

I1 x
I2 y

(I1 I3 )z x

I3 z

(I1 I3 )z y

(3.48)

= 0.

The third equation shows us that the rate of rotation around a symmetry
axis is constant and we are therefore able to treat it as one of the initial known
conditions. The two remaining equations are a pair of coupled linear dierential
equations for

and

y .

We are able to dene a constant

I1 I3
I1 z ,

(3.49)

known as the angular frequency. It is then possible to rewrite the equations for

and

as

y =

y x =

0.

(3.50)

There are various methods we could employ to solve these equations. One
technique is to nd the time derivative of the rst equation:

(3.51)

and to then use the second equation to substitute for

x
A typical solution for

= -

2 x .

(3.52)

can be written as

x
where

y :

Asin t

(3.53)

is some constant and if we substitute this solution for

of the set of equations (3.48) we can solve for

y
The magnitude of the vector

x i

Acos t.

y j

(3.54)

is constant and it rotates uniformly

about the z-axis of the symmetrical body with angular frequency

nitude of the total angular velocity
body-xed frame

precesses

into the rst

to obtain:

The mag-

of the body is also constant and in the

about the symmetry axis with the same frequency,

It is worth noting that the body axes are themselves rotating in space, at

the greater frequency

Equation (3.49) tells us that smaller the dierence between

slower the precession frequency
rotation,

I1

and

I3

is, the

will be in comparison with the frequency of

The kinetic energy and the magnitude of the angular momentum

can be used to evaluate the constants

A
32

and

z :

J2
Here,

1
1
2
2
2 I1 A + 2 I3 z

I12 A2

I32 z2 .

(3.55)

is actually the amplitude of precession. It is possible to solve these

equations for both

and

in terms of

and

L.

3.7 The Heavy Symmetrical Top with One Point Fixed

A heavy symmetrical top is a symmetrical body, pivoted at a point on its
axis of symmetry, moving in a gravitational eld. In the previous section, we
chose the z-axis of the body-xed system to the the symmetry axis of the body.
The symmetry axis is also one of the principal axes of the body.
The constraint that one point of the top must be xed reduces the number
of degrees of freedom of the system to 3, and the Euler angles are able to
completely specify the motion of the body.

Figure 3.1: The orientation of a spinning top is completely specied by the

Euler Angles5 .

5.71 The Lagrangian and the Generalised Momenta

The centre of gravity of the body lies along its symmetry axis, and if we
denote the distance from here to the xed point by

R,

we can then write the

potential energy of the body as

M gRcos.

(3.56)

5 image adapted from http://teacher.pas.rochester.edu/PHY235/LectureNotes/Chapter11.htm

33

The kinetic energy of the symmetrical body can be written as

1
2
2 I1 (x +

y2 )

1
2
2 I3 z .

(3.57)

We can write this in a dierent form, using Euler angles, by using the set of
equations (2.99), obtaining

2 sin2 )

1
2
2 I1 ( +

because the cross terms in

x2

and

y2

2 I3 ( +

cos)

(3.58)

cancel each other out.

We can then write the Lagrangian for the system:

2 sin2 )

1
2
2 I1 ( +

2 I3 ( +

cos)

M gRcos.

(3.59)

The corresponding generalized momenta are then:

L
2
= (I1 sin +

L
=

I1

L
=

I3 (

I3 cos2 )

I3 cos

(3.60)

(3.61)

cos)

I3 z .

(3.62)

It can be seen that the Lagrangian does not involve the Euler angles
and

explicitly.

This indicates that these coordinates are ignorable and the

generalised momenta corresponding to these angles are constant in time and the
system has only one degree of freedom,

5.72 The Energy Equation

As well as the generalised momenta for

and

the total energy

of the

conservative system will also be conserved:

Since the momenta

constants

and

V
p

I1
2 (

and

+ 2 sin2 ) +

I3 2
2 z

+ M gRcos.

(3.63)

are constants of motion, we can dene two

b,
a

I3 3
I1 ,

p
I1

(3.64)

which we can use to write

bacos
sin2 .

We can also obtain an equivalent expression for

I1 a
I3

(3.65)

cos bacos
sin2 .

34

(3.66)

 as a function of time then it is possible to integrate the equations

and in time.
and from the
We can use equation (3.65) and equation (3.66) to eliminate
If we know

to give the dependence of

equation for the total energy, leaving a dierential equation that is dependent
only on

z is constant in time, as shown by

I1
a
. We can then dene a constant of motion, the
I3

First we need to be aware that

equation (3.62), and equal to

reduced energy,

E0

as:

E 0 E 21 I3 z2 .

(3.67)

It is then possible for us the rewrite the total energy equation as

E0

I1 2
2 (

+ 2 sin2 )

M gRcos.

(3.68)

If we dene two constants as

2E 0
I1 ,

2M gR
I1

(3.69)

then we can substitute in equation (3.65) into equation (3.68) and rearrange
the terms to give

sin2 2

sin2 ( cos) (b acos)2 .

To simplify the analysis, we will set

cos

(3.70)

and then it is possible to rewrite

the above equation as

u 2 = (1 u2 )( u) (a bu)2 = f (u).

(3.71)

We are also able to reduce the equations (3.65) and (3.66) to

bau
1u2

(3.72)

and

I1 a
I3

u(bau)
1u2

(3.73)

We can immediately take the square root of equation (3.71) and integrate it to
get a quadrature:

u(t)

u(0)

du
(1u2 )(u)(bau)2

We could nd similar integrals for

and

(3.74)

but we do not need to perform

these integrations to get a general idea of the motion. The equation for

f (u) is
, at

cubic and the roots of the cubic polynomial provide the turning angles of

changes. .When u is large, u3 becomes the dominant

f (u)
as u
, and f (u)
as u
.

which the sign of

of

f (u),

and so

term

u is equal to 1, f (u) = (ba)2 0, except in the instance when u =

1 is a root of f (u), which physically corresponds to a vertical top. Physically,
the top can only be in motion when u
2 is positive and u must lie in the interval
u = -1 to u = +1, so, since u = cos, we see that must lie between - and +.
When

35

A plot of the function

f (u)

looks like Figure (3.2):

Figure 3.2: A plot of the function f (u)6

For any real top,

f (u)

therefore has two roots

-1 and +1. The top must move so that

cos

u1

and

u2 ,

that lie between

always stays within this region.

5.73 Motion of the Spinning Top

The motion of the top can be depicted by tracing the curve of the intersection
of the gure axis (the

locus of the gure axis) onto a unit sphere about the xed
1 = arc cosu1

point. The locus lies between the co-latitude bounding circles

and

2 = arc cosu2 .

Any point on the locus has polar coordinates which are

identical to the Euler angles for the body system. The value of the root

b au

determines the locus curve shape.

Figure 3.3: The dierent types of motion depend on the direction of precession
at the extrema 7
6 image obtained from http://www.damtp.cam.ac.uk/user/tong/dynamics/three.pdf
7 image obtained from http://teacher.pas.rochester.edu/PHY235/LectureNotes/Chapter11.htm
36

There are three possibilities for the motion, and they are dependent on the
sign of

as determined by equation (3.72) at the roots

Motion as depicted in Figure 3.3(a) occurs if

If

> 0

at

u = u1 ,

but

< 0

u = u2 ,

at

> 0

u = u1 and u = u2 .
u = u1 and u = u2 .

at both

then the type of motion as shown in

Figure 3.3(b) will occur. Figure 3.3(c) shows the path of motion that will come

> 0 at u = u1 , but = 0 at u = u2 .
Motion in is known as precession, whilst

about if

motion in

is called

nutation,

and can be visualised as the gure axis nodding up and down between the
bounding angles

and

as it goes around.

The motion as shown in Figure 3.3(c) is not as unlikely as it may seem. If

we spin the top and let it go at some angle

t = 0, = 0

and

= = 0.
p

0 ,

we have the initial conditions

Also, we should remember that the quantity

2
I1 sin

I3 z cos

I3 z cos0

(3.75)

is constant. We now have sucient information to describe the qualitative

motion of the top. It will initially begin to fall under gravity, and so
increases. As the top is falling,

must turn and increase so that

remains constant. The direction of precession,

as

z ,

always

must be in the same direction

and we get the type of motion as shown by Figure 3.3(c).

When

and

are initially equal to zero, and if we assume that the ini-

tial kinetic energy of rotation is large compared with the maximum change in
potential energy:

1
2
2 I3 z >>

2M gR,

(3.76)

then we are able to quantitatively predict how the top will move. The
precession and the nutation will only be small disturbances to the rotation of
the top about its gure axis. In this case, we would call the top a fast top.
For a fast top, the angular momentum is along the axis of spin of the top.

5.74 Uniform Precession

If

is equal to zero, and if

is constant, then the top will precess without

bobbing up and down. This situation requires

f (u) to have a single real root u0

lying between -1 and +1. This root must satisfy the following equations:

f (u0 )

(1 u20 ) ( u0 )

- (b

au0 )2

= 0

(3.77)

and

f 0 (u0 )

= -2u0 (

u0 ) - (1 u20 )

+ 2a(b

au0 )

= 0.

(3.78)

Combining these two equations, we nd that

2 =

a 2 u0 .
37

(3.79)

Substituting in the relation

I1 a

I3 z

and the denition for

as seen in

equations (3.69), this gives us the equation

3 z I1 cos

M gR =(I
0 ).
For a xed value of
shove

and xed

0 ,

(3.80)

giving the top exactly the right initial

will allow the top to spin without nutation. The quadratic nature of

equation (3.80) in

indicates that there are two frequencies with which it is

possible for the top to precess without nutation, known as slow precession and
fast precession. We can also see that equation (3.80) can never be satised by

= 0, so an initial shove must always be given to the top for it to be able to

precess uniformly. The fast and slow precessions will only exist if equation
(3.80) can actually be solved. This requires

z > I23
For a given

0 ,

M gRI1 cos0 .

(3.81)

the top therefore has to be spinning fast enough to have

uniform solutions, otherwise it will topple over.

5.75 The Sleeping Top

We can start a top spinning with its axis vertical, with

= = 0.

(3.82)

When it is spinning quietly about the vertical, we call it a sleeping top. In

this situation,

f (u)

has a root at

= 0

(or

u = +1), so that f (1) = 0.

a = b and = here.

can use equations (3.64) and (3.69) to check that

function has two roots of

u=

f (u)

+1,

(1 u)2 ((1 + u) a2 )

2
1 . If u2 is greater than 1, then
0
critical angular velocity , where
and a root of

u2

0
and the graph of

f (u)

We
The

(3.83)
will be greater than a

4I1 M gR
I32

looks like Figure 3.4(a). This motion is stable and the

only possible motion is for

u = 1,

so the top just continues to spin about the

f (u)

vertical.
If

u2

< 1 , then

<

and

looks like Figure 3.4(b) and the top is

unstable.

38

(3.84)

Figure 3.5: A plot of f (u) for the stable and unstable sleeping top8
Practically, the top will spin about the vertical until friction gradually reduces the frequency of rotation to below the critical angular velocity and the
top will then start to wobble as it slows down and will eventually fall.

8 image obtained from http://www.damtp.cam.ac.uk/user/tong/dynamics/three.pdf

39

Chapter 4

The Levitron

4.1 Introduction to the Levitron

The Levitron is a popular mechanical toy invented by Mr. R Harrigan and

developed by Mr. W Hones. It consists of a small, magnetic spinning top and a
permanently magnetized ceramic base plate. The magnetic forces and torques
couple uniquely with the gyroscopic motion of the spinning top, allowing stable
levitation to occur. The top is seen to oat above the base in mid-air, precessing
and nutating about an equilibrium point until air resistance slows the top down
suciently and the top then becomes unstable and falls.

Figure 4.1: The Levitron9

Knowledgeable physicists were adamant that Harrigan was wasting his time
with the Levitron, because stable magnetic levitation for permanent static magnetic dipoles is forbidden by Earnshaw's theorem, but in this chapter I will
discuss the mechanics of the device and attempt to explain the mechanical principles which allow it to work, providing a strict set of conditions are satised.

4.2 An Overview of How the Levitron Works

The heavy, symmetrical top is a rigid body, with mass
mentum

J.

and angular mo-

The mass of the top can be altered by adding small washers to it.

The centre of mass of the top is located at

r = (x, y, z). The top can be thought

which is located at the centre of

of as a magnetic dipole, with vector moment

mass and points along the symmetry axis.
A magnetic eld

B(r) is provided by the base plate, and its gradients provide

mg , on the top. This

a repulsive force which opposes the gravitational force,

9 image

obtained from http://www.vnix.nl

40

repulsive force is responsible allowing the top to achieve stability whilst spinning
above the base; it acts on the vector moment in the presence of the spin

J.

It is not sucient to just stabilize the top against ipping. If we assume that
the magnetic dipole moment of the top,
vertically, in the downwards

approximately vertical, but pointing in the

B(r)

energy

is always oriented approximately

B(r) is also
+z direction, then the magnetic
Bz . The total potential energy

direction and the repulsive eld

is approximately equal to

is given by:

V = B(r)
V

mgz

(4.1)

Bz + mgz .

(4.2)

The rst of two conditions that must be satised for stable levitation to occur is that the upward repulsive force, which is approximately equal to

z Bz ,

must balance the force of gravity on the top to allow it to oat in equilibrium.
The potential energy must have a critical point at the equilibrium point, and the
second condition for stable levitation is that this critical point must be a mini-

10 since both gravity

mum. This is actually impossible, by Earnshaw's theorem

and

Bz

are both harmonic potential elds. The energy minimum condition can-

not be satised in both the vertical and horizontal directions simultaneously

and instead there is a saddle point.
It is, however, still possible for the top to oat stably in a time-independent
potential eld. If we take the potential to be the sum of the gravity and the
magnetic energy of

then the average

rather than the

z -component. A
m only. The

averaged over its precession around

energy makes use of the magnitude

of

minimum of the potential then exists for a narrow range of mass

range is narrow because it is the small deviations of

from vertical near the

axis which are responsible for allowing the potential to posses a minimum. The
form of
of

m.

B(r) varies with temperature,

and is responsible for dictating the range

The mass required to allow the top to stably levitate may need to be

frequently adjusted.
We need to make a correction to the adiabatic averaging underlying the static
stability, as it is not exact. The rst adjustment is to introduce an additional
force called geometric magnetism which has the form

r is the
Bef f (r)

v Bef f (r)

where

velocity at which the top is moving through the magnetic eld, and
is the eective eld constructed from component derivatives of

B(r).

Three speeds must be very dierent to allow the top to oat stably in the
static potential eld. The spin angular velocity of the top must be the fastest,

10 Earnshaw's Theorem states that it is not possible to achieve stable suspension of an object
against gravity, using any combination of electric charges and xed magnets. The proof is
fairly straight forward: The static force as a function of position F(x) which acts on a body
due to the combination of gravitational, electrostatic and magnetic elds will always have
divergence equal to zero, 5 F(x) = 0. The force at a point of equilibrium is zero. In the
case of stable equilibrium, the force must point inwards towards the equilibrium
point
on

some small sphere around the point. Gauss's theorem implies, however, F(x) dS = 5 F
S

dV , and since the divergence of the force over the volume inside is equal to zero, the radial

component of the force over the surface must also be equal to zero.
41

the precession angular velocity of the top, and the rate at which

changes in

the frame of the top must be the slowest.

If the top and base were both made from metal, the top would fall faster,
because eddy currents would be induced, which would be an additional source
of dissipation (energy is also dissipated by resistance from the air) and the top
would fall more quickly.

4.3 Conditions for Stable Equilibrium

We are able to write the potential energy of top is given in equation (4.1).
The magnetic torque on the top is equal to

and the spin of the top

changes as the result of this torque,

J(t)

(t) B(r(t)).

(4.3)

We can make two assumptions for the top which will allow us to simplify the
problem. Assuming that the top is small allows us to approximate its magnetism
as a point dipole., located at coordinates

x.

The second assumption is that the

top is fast, so that the angular momentum is along the axis of spin of the top,
and this also coincides with the magnetic moment axis. The condition for a fast
top is that the spin must be much faster than the precession and we shall see
later that this condition is indeed satised.
We can write an equation for the magnetic eld so that it is expressed in
terms of its magnitude

and direction

b as seen from the spinning top's frame:

B(r(t)) B(t)b(t).

(4.4)

We are now able to write a new equation for the spin as:

J(t)
where

(t)b(t) J(t)

(4.5)

is the angular precession frequency with which the top rotates about

the instantaneous direction of the magnetic eld as seen in the body-xed

frame:

B
J .

(4.6)

As seen in the previous chapter, saying that the precession is fast is dierent
to saying that the top is fast, and here the condition that the precession is fast
is equivalent to the expression

| || b |, and here the adiabatic

invariant

Jb

must be approximately conserved, and this relation is known as an adiabatic

slaving of

to

b.

The component

B (t) b(t)

(4.7)

must then also be an adiabatic invariant, and this allows us to rewrite the
potential energy as:

V = V (r) = B B(r) + mgz .

42

(4.8)

Later, we will examine this approximation more closely.

With the assumptions made above, the top can how achieve stable levitation

V (r)

above the base if

has a possesses a minimum there.

This requires the

following three conditions to be satised:

(i) equilibrium:

V (r)

= 0

(ii) vertical stability:

z2 V (r)

> 0

(iii) horizontal stability :

x2 V (r)

> 0 and

(4.9)

y2 V (r)

> 0.

I am only going to be considering the situation where the minimum lies on

the vertical symmetry axis of the base and neglecting the possibility of any
o-axis minima.
Since the centre of mass of the top is not located inside the base, the base
does not contribute any currents to

B(r). B

is curl and divergence free, so we

are able to write the eld as

B(r)

(r)

(4.10)

with

2 (r)
The potential

is stationary at

= 0 .

(4.11)

x = y = 0 in horizontal planes, and has

x and y , which is true for the Levitron

circular symmetry to second order in

with a square base.
If we dene

R (x, y)

(4.12)

| R |.

(4.13)

and

then we can write the potential close to the axis as

(r) = (0, 0, z) + 21 x2 (0, 0, z)R2

+... .

(4.14)

Conveniently, we can use the notation

n (z) = zn (0, 0, z).

(4.15)

Laplace's equation (4.11) must be satised by equation (4.14), and this gives

(r)

0 (z) 41 2 (z)R2 + ...

(4.16)

The adiabatic energy equation (4.8) makes use of the magnitude of the vector

B,

and we can use equations (4.10) and (4.11) with (4.16) to write

second order in

R:
43

B(r)

to the

B(r)

h
1 sgn1 1 +

without explicitly writing the

R2
8

22
21

2 31

dependence of the

i

...,

(4.17)

n .

In the rst of the set of equations (4.9), horizontal equilibrium on the axis
is guaranteed by symmetry. For the system to also be in vertical equilibrium,
gravity must be balanced by the repulsive magnetic force determined by the
gradients of the magnitude of the eld provided by the base plate, which is

mg
The gravitational force

B z B = B 2 sgn1 .

mg

(4.18)

is always positive, and so we can combine this

fact with the second and third stability equations of set (4.9) to give three new
conditions:
(i) equilibrium:

B 2 sgn1

> 0

B 3 sgn1

(ii) vertical stability:

(iii) horizontal stability:

< 0

(4.19)

B sgn1 (23 21 )

> 0.

We cannot possibly satisfy these conditions when

will imply that

and

is then disobeyed. It is then a necessity that

that the projection of

B > 0

because then (ii)

have opposite signs, and clearly equation (4.19(iii))

along

B < 0, which
B, and

is anti-parallel to

in practice means
we get three new

conditions for stable equilibrium:

(i) equilibrium:

and

(ii) vertical stability:

(iii) horizontal stability:

The dierence between

have opposite signs

and

have the same signs

22 23 1 > 0.

B and B gives rise to the term 22 in (iii), and it allows

both equations (ii) and (iii) to be simultaneously satised, conrming that the
adiabatic potential in equation (4.8) does actually permit the top to achieve
stable levitation over the base in spite of Earnshaw's seemingly contradictory
theorem.

4.4 The Magnetic Field on the Axis

The Levitron has a square base which, with the exception of an unmagnetized
central hole, is uniformly magnetized.

The purpose of the hole is to allow a

region which is almost eld-free, so that the top can be spun by hand before it
is lifted on a plastic plate to a position where it is able to oat in equilibrium.
We can consider the base to be a planar distribution of dipole sources which
are vertically oriented, with density

(R) and the following formula gives us the

potential of a dipole:

44

(4.20)

0 (z) = z

d2 R

base
where

(R)

(4.21)

(R2 +z 2 ) 2

= 0 corresponds to the centre of the hole.

the dipoles in the base point up, then

top must then point down, so

We can assume that

is positive. The dipoles in the spinning

is negative, and we can authenticate this

assumption by seeing that the an upright top will repel the top of the base
when held close; unlike dipoles will repel.
The analysis of stability for the top will almost be the same regardless of
whether the base is a square with a hole in the centre or just a circular disc.
It is simpler to carry out analysis for the case when the base is a uniformly
magnetized disc, so this is what we will do. If the disc has a radius

and

is

constant, then equation (4.21) gives:


0 (z) = 2 1

z
a2 +z 2

The condition given in equation (4.20(i)) requires

(4.22)
and

to have opposite

signs. Equation (4.18) makes it seem as if magnetic repulsion can balance gravity

3 > 0,
z > 12 a and so vertical stability is guaranteed by by
1
2
condition (4.20(ii)) for z > a. The function 2 23 1 from equation (4.20(iii))
2
2
2
is directly proportional to 2a 5z and so for the top to be horizontally stable,
q
at any height above the base if an appropriate mass is chosen. When

z < 21 a

and when

we need

< 0,

to be less than

2
5 . Stable equilibrium can therefore occur in the

range

1
2

<

z
a

<

2
5.

(4.23)

It is also possible to nd the range of masses which are allowed by the above
condition (4.23). We can dene three new coordinates,

x
a

which are all dimensionless.

y
a

z
a

(4.24)

We can also dene energy and mass by the

following relations:

Va
2|B | ,

mga2
2|B | .

(4.25)

We can combine equations (4.8), (4.17) and (4.22) to nd that the potential
energy close to the axis is equal to

V (, , ) = M +

1
3
(1+ 2 ) 2

3(2 + 2 )(25 2 )
7

8(1+ 2 ) 2

The condition for equilibrium (4.18) determines the range of masses,

(4.26)

M,

which

are allowed by condition (4.23), giving

M=

3
3

(1+ 2 ) 2

45

(4.27)

1
2 , which is the lower stability limit for
. If the mass is greater than this maximum, stable equilibrium is simply not

has a maximum

M+

when

possible, and the top will not be able to oat. The maximum is equal to

M+ =
The upper stability limit for

48
5

52

= 0.85865010...

is equal to

(4.28)

2
5 and the mass has a minimum

here:

M = 75 5 2
72

= 0.818146658... .

(4.29)

If the mass is less than the minimum, vertical stability can actually be
achieved, but the condition for horizontal equilibrium is not satised and so
stable equilibrium can only occur when the mass is in the interval between
and

M+ .

This interval is small - it is approximately only 5% of the mean stable

mass.
It is possible to nd a mass which yields the most stable motion of the top,
and this is

MS = 0.847837.

If we no longer consider the base to be a magnetized disc but instead the

more accurate model of a uniformly magnetized square slab with sides of length

2a

and with an unmagnetized hole of radius

in the centre, then we get a new

equation:

0 (z) = 2 z2z+w2 8arcsin

z
2(z 2 +a2 )


.

(4.30)

If we carry out stability analysis on this function for the Levitron, which has

w
1
a 4 , we obtain the result that the top is able to achieve
stable levitation in the region
the relation that

3.976 <

z
w < 4.360

(4.31)

which we can see is very narrow. We must make sure that we choose the
mass of the top carefully so that equation (4.18) has a solution which lies in the
stable range.

A change

height of equilibrium

dm

in the mass of the top results in a change in the

and we can use equation (4.18) to nd that this gives

d = dm
m |

2
3

2
3 | is known as an amplif ication f actor which decreases in the
stable interval from innity to a minimum value. When we use the base model
1
w
of a square slab with a hole in as before, and
a 4 , this minimum value is
7.05. Even the lightest of the washers which we can add to the top to increase
where

its mass will result in the height of levitation changing by around one tenth of
the interval of stability.
One feature of the Levitron which may seem puzzling is the need to constantly adjust the weight of the top over short periods of only a few minutes.

46

(4.32)

This has been shown experimentally to be due to handling of the top and ambient temperature changes resulting in temperature variation which has an eect
on the mass of the top required to allow stable levitation to occur.

4.5 Adiabatic and Geometric Magnetism

In the Levitron, we know that in addition to gravity, magnetism also contributes to the motion of the top. From the equation for the potential energy
of the top (4.1), we nd the sum of these forces on the top to be

F = mgk + FM
where

is the unit vector in the upward

(4.33)

zdirection

and

FM

is the magnetic

force, which we can rewrite to get:

F= mgk + (t) B(r).

It is possible to split
pendicular to the

into its components which are parallel to and per-

eld at any instant, to get

(t)
where

(4.34)

B b(t) + (t)

(4.35)

is the perpendicular component. We can now write the magnetic

force as:

FM FA +FG
with

FA

(4.36)

being the adiabatic component of the magnetic force, and

FG

being

the geometric component. We can write this as

FM
where

and

B B(r) + (t) B(r)

(4.37)

are connected by the dot product.

Since we have assumed that both the top and the precession are fast, the
equation of motion (4.5) for

will also be satised by

to parallel and perpendicular components of

Separating this into

we obtain

= B b + B b +
= b

(4.38)

We need to make some adiabatic approximations. It is necessary to allow the

precession of the top to be about a direction which is slightly dierent from the
instantaneous eld direction

b(t). We also need to make the approximation that

to zero, but not the component

we can set the precession averaged velocity

We now nd the parallel and perpendicular components of equation (4.38)

give:

B 0
47

(4.39)

and

B b b .

(4.40)

Equation (4.39) is the conservation of the adiabatic invariant, and the component

FA

in equation (4.36) becomes the adiabatic force of the lowest order

that would be obtained from equation (4.8). This leads to the conditions obtained earlier for static stability.
Using equation (4.40) as well as the assumption that the top is fast and
therefore

is parallel to

magnetic force,

FG

we are able to write the geometric component of

(which is a post-adiabatic force) as

B(r)
FG = B (b b)
JB

=
B (b b) B(r)
where

JB J b.
b :

(4.41)

As the top moves through the inhomogeneous eld

B(r),

it

causes a change

b = (v )b(r).
The force

FG

(4.42)

is dependent on both the position of the top and its velocity:

FG = JBB3 [B (v )B] B(r).

(4.43)

It is possible to use vector algebra to rewrite equation (4.43) as

FG = v Bef f (r)
with the vector

v = r ,

and where

Bef f

(4.44)

is dependent on the components of

B,

as dictated by

Bef f

JBB3 (Bx By BZ + By BZ Bx + Bz Bx By

The geometric magnetism force,

FM

(4.45)

has the same dependence on velocity

as the Lorentz force, it seems as if the top carries a unit charge in response to
the eective eld

Bef f .

The geometric post-adiabatic magnetism force acts as

a reaction to the fast spin

J(t)

of the top on the slow motion

r(t)

of its centre

of mass.
In all the calculations, we have regarded the precession to be fast, so that
it is slaved to the slow variable

and

then reacts magnetically on

r.

We can

instead show that the precession can be considered to be geometric because

J is

actually slaved to the spin of the top - the motion of the axis of the top is slow
when compared to the spin. When we average the precession over the nutation
of the body axes, we can consider it to be a geometric reaction which results
from a monopole source of magnetism xed at the point of precession. For the
Levitron, this xed point is simply the centre of mass of the top, but in the
regular spinning top it would be the point where the top is in contact with the
surface while spinning.

48

4.6 The Eect of Geometric Magnetism on Stability

We can use the previously seen equations (4.33), (4.37) and (4.44) to write
an equation of motion for the top when it is subjected to the force of gravity as
well as the two magnetic forces. We obtain

mr = mgk + B B(r) + v Bef f (r).

FA

Using the adiabatic force

(4.46)

from equation (4.63) with earlier obtained

equations (4.12), (4.13) and (4.17), we get

B = sgn1

h 
1
4

22
1


i
23 R + 2 k .

(4.47)

The eective magnetic eld on the axis in equation (4.45) can be used with
equations (4.10) and (4.16) to give

Bef f = JB sgn1 422 k.

(4.48)

This tells us that geometric magnetism does not have an eect on vertical
motion; only the static gravitational force and the adiabatic magnetism force
will aect the motion vertically. The geometric magnetism is instead responsible
for aecting the horizontal motion of the top, and we can write a linear equation
for the horizontal acceleration:

=
R

g
4

2
1


22
2 32 R + JB sgn1 4m
2R k

(4.49)

where there are coecients which are dependent on the height of the top, and

B .

we have made use of equation (4.18) to eliminate

Dening

u(t)

as

u(t) x(t) + iy(t)

(4.50)

will allow us to write equation (4.49) as

u
= A(z)u + iB(z)u ,

(4.51)

in scalar form. This has the general solution:

u(t) = u+ ei+ (z)t + u ei (z)t

(4.52)

where

(z) = 21 (B

B 2 4A).

to be real for the top to be stable horizontally, which

B 2 > 4A. With equation (4.49), this gives the condition


2
JB
1 3 1 3
1
>
G(z)

32
|
|

2
2
m g
3
2 .

We need
requires

49

(4.53)
in turn

(4.54)

If the condition (4.20(iii)) is satised, then the condition (4.54) is also satised, since

G(z) will be negative.

If

G(z) is positive then condition (4.20(iii)) will

not be satised and horizontal stability cannot be achieved. Condition (4.54)

shows that if the top is spun fast enough, geometric magnetism can indeed provide post-adiabatic stabilization. If we set the top to have a vertical axis with
radius of gyration

and spin frequency

we are able to evaluate the last part

a from equation (4.22).

z
satises the condition:
a

of equation (4.54) with the potential of a disc of radius

The horizontal motion will then be stable if
2
JB
m2 ga3

G(z)
4 2 2 d4
>
ga3
a3

16(5 2 2)( 2 +1)3

.
5

(4.55)

This is equivalent to the condition:

2
5

<

81 2 2 d4
686ga3 ..

(4.56)

a 5cm, d = 1.13cm,
q and
2
hand spinning gives a spin frequency 20Hz. Stability requires
5 <
0.0062. The total range of stable spin rates is approximately 18Hz 40Hz,
Physical measurements for the Levitron are

and is achievable by hand-spinning.

The lower spin limit corresponds to the

sleeping top condition. Geometric magnetism only contributes a small (about

5%) increase to the stable interval of

when spun at

contribute up to a 20% increase if the top were spun at

= 20Hz,
= 40Hz.

but would

4.7 Adiabatic Conditions

Practically, the ratio of the moments of inertia is somewhere between

1
2 and

1; we can assume that all of the principal moments of inertia of the top are
approximately the same size, and we then have the following condition for the
top to be fast:

2 | |
where

is the spin angular velocity and

||

is the precession angular

velocity. We can use equations (4.6) and (4.18) to express the above inequality
in terms of the magnetic potential. We get

 min

(4.57)

where

min =

1
2d

r  
 
g  12 .

We can then write the precession frequency in terms of

2
min
.

50

(4.58)

min :
(4.59)

By the adiabaticity, the rate at which the eld

is changing (equation (4.42))

must be much slower than precession. This condition is written as

| || (v )b |.

(4.60)

We can use equations (4.10), (4.11) and (4.16) to give the result

n
o
2
2
b = x 2|
,
y
,
sgn
+ ...
1
2|1 |
1|

(4.61)

and this means that equation (4.59) is dependent only on the transverse speed
of the top,

v .

We can apply some algebra to the adiabatic condition (4.60) to

give the new condition

v 

2 g
d

  23
 1 
 2 

min
.

(4.62)

The nutation frequency is related to the transverse speed of the top. If we

consider the frequency of this vertical motion, equation (4.46) gives

z =

1
2

r  
 
g  32  =

1
2

g (4 2 1)
a ( 2 +1)

(4.63)

where we are referring to the circular disc base.

Using the previously stated measurements of

and

for the Levitron, and

using an approximate spin frequency of 20Hz, the potential from equation (4.22)
implies that 0.73a

<|

1
2

|< 0.83a

and

min
over the interval of stability,

1
2

0.88
2

<<

p ga

d2

8.7Hz

(4.64)

2
5 . This frequency is much less than

the lowest speed achievable by hand-spinning. Equation (4.59) then tells us

that the corresponding precession frequency here would be

= 3.8Hz

, and

the Levitron is indeed a fast top.

The inequality (4.62) gives an upper limit for the transverse speed of the
top. When the top is spun by hand, its horizontal speed is much slower than
this limit and so we have satised the adiabatic condition for motion.
Earlier we discussed

MS ,

the mass for which the most stable motion can

occur. Putting this together with equation (4.63), we obtain

z =

0.61106
2

pg
a

0.69min ad 1.4Hz.

This nutation frequency is much less than the precession frequency.

If we spin the top too fast, the precession frequency will be too slow to
prevent the top from tipping over so much that it can no longer be supported
by the magnetic eld, and the top will fall over.

51

(4.65)

4.8 Analogy with Microscopic Particle Traps

While I will not look into it here, it is worth noting that the trap mechanism
in the Levitron is analogous to magnetic gradient traps for neutral particles
with a quantum magnetic dipole moment. These traps were rst used to trap
cold neutrons and currently are being used to trap atoms.

52

Chapter 5

Conclusion
Rigid body dynamics is a broad and interesting topic, with many useful
applications. In this project, we began by dening a rigid body and went on
to look at the basics of rigid bodies and their motion, describing Euler angles
and considering innitesimal rotations.

We then went on to look at angular

momentum and the moment of inertia tensor, as well as the principal axis,
combining these ideas to nd the equations of motion for a rigid body to be
Euler's equations.

We looked into the motion of symmetric rigid bodies and

nally were able to consider the motion of a heavy symmetrical top.

We have seen that precession is an exciting phenomenon and although it may
imply some counter-intuitive ideas, fully understood it can be used to describe
the motion of the spinning top, and further the motion of the Levitron, an
exceptional toy which had scientists baed for a long time, since it seems to
violate Earnshaw's theorem.
In examining the principles behind the Levitron, we have seen that it is the
unique coupling of the magnetic forces and torques with the gyroscopic motion
of the top that is responsible for permitting stable levitation to occur.
There are many areas where I could have extended the study of precession.
An obvious one is to study the precession of the Earth.

This would involve

considering local torque-induced precession due to the gravitational eects of

the sun and moon acting upon the axis of the Earth.

There is also a minor

amount of non-local torque-free precession due to the motion of the solar system.
The Earth has a central bulge at the equator, so it is in fact not a sphere, but
a symmetrical top. We could also look into the eects of the Earth's precession
on astronomical observations and the precession of orbital objects. Clearly, this
topic is very rich and there is wide scope for further study here.
Very recently, I discovered a special type of spinning top called the Rattleback, or Celt, which is a semi-ellipsoidal top which can be initially spun in
any direction, but if not spun in its preferred direction, it will become unstable,
rattle, stop and reverse its spin. This spin-reversal seems to violate angular
momentum conservation laws, and I think it would be very interesting to study
the mathematical principles behind the Rattleback more closely.
There are many other areas of the project which I haven't touched on at all.
We could have considered Thomas precession, which is a special correction to
gyroscopic precession in a rotating non-inertial frame. It has many applications,
such as in quantum mechanics where it is a correction to the spin-orbit interaction, and takes into account the relativistic time dilation between the electron
and the nucleus in Hydrogen atoms.
We could further extend the study of the Levitron to consider more closely
the analogy between it and microscopic particle traps.

53

It is astounding to think that the simple spinning top, one of the oldest
discovered toys has such stimulating mathematics behind it, and similar mathematics can help explain such a wide range of dierent phenomena in the areas
of astrology, electromagnetism and quantum mechanics.

54

Bibliography

[1] A. Modi, Dynamics of a Spinning Top:

http://www.anirudh.net/courses/emch520/html/node3.html
[2] D. Kleppner and R. J. Kolenkow: An Introduction to Mechanics
(McGraw-Hill)
[3] D. Tong, Cambridge University: Dynamics Lectures:
http://www.damtp.cam.ac.uk/user/tong/dynamics/three.pdf
[4] G. R. Fowles and G. L. Cassiday: Analytical Mechanics (Harcourt Brace)
[5] H. Goldstein: Classical Mechanics (Addison Wesley)
[6] Hyperphyics, Gyroscopes:
http://hyperphysics.phy-astr.gsu.edu/hbase/gyr.html
[7] J. B. Marion and S. T. Thornton: Classical Dynamics of Particles and

Systems (Saunders College)

[8] J. Huang, Texas Tech University:
http://www.phys.ttu.edu/~huang24/Teaching/Phys5306/
[9] M. Burske, University of Colarado, Advanced Dynamics:
http://theory.ph.man.ac.uk/~mikeb/lecture/pc167/rigidbody/contents.html
[10] Morehouse, Orthogonal Transformations:
http://www.morehouse.edu/facsta/cmoore/Orthogonal%20Transformations.htm
[11] M. Simon, L. Heinger & S. Ridgway: Spin Stabilized Magnetic

Levitation : Am. J. Phys., Vol. 65, No. 4 (1997)

[12] M. V. Berry: The Levitron: An Adiabatic Trap for Spins: The Royal
Society (1996)
[13] Queen's University Mechanics Lectures:
http://me.queensu.ca/courses/MECH494/documents/06-EulersTheorem.pdf
[14] Tabitha Rigid Bodies:
http://tabitha.phas.ubc.ca/wiki/index.php/Rigid_Bodies
[15] The Levitron: http://www.levitron.com/physics.html
[16] T. W. B. Kibble and F. H. Berkshire: Classical Mechanics (Longman)

55

[17] V. D. Barger and M. G. Olsson: Classical Mechanics A Modern

Perspective (McGraw-Hill)
[18] Wikipedia, Gyroscopes: http://en.wikipedia.org/wiki/Gyroscope
[19] Wolfram MathWorld, Innitesimal Rotations:
http://mathworld.wolfram.com/InnitesimalRotation.html

56