PHYS3010

PHYS 3010:
Classical Mechanics
Lecture Notes
Tom Kirchner1
Department of Physics and Astronomy

York University
April 2015
1 tomk@yorku.ca
Contents
1 Introduction 3
1.1 Prelude and Overview . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Recap of Newtonian Mechanics . . . . . . . . . . . . . . . . . 5
1.2.1 Newton’s Laws (1687) . . . . . . . . . . . . . . . . . . 5
1.2.2 (Linear) momentum, angular momentum, work, and
energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Hamilton’s Principle — Lagrangian and Hamiltonian dynam-

ics 17
2.1 Preliminary formulation of Hamilton’s principle (1834/35) . . 17
2.1.1 Calculus of variations . . . . . . . . . . . . . . . . . . . 18
2.1.2 HP for a simple case . . . . . . . . . . . . . . . . . . . 22
2.2 Constrained systems and generalized coordinates . . . . . . . . 23
2.2.1 Preliminary for one mass point . . . . . . . . . . . . . 23
2.2.2 N-particle systems . . . . . . . . . . . . . . . . . . . . 31
2.3 General formulation of Hamilton’s principle and Langrange’s
equations for N-particle systems . . . . . . . . . . . . . . . . . 35
2.3.1 Hamilton’s principle . . . . . . . . . . . . . . . . . . . 35
2.3.2 Equivalence of Lagrange’s and Newton’s equations of
motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4 Conservation theorems revisited . . . . . . . . . . . . . . . . 41
2.4.1 Generalized momenta . . . . . . . . . . . . . . . . . . . 41
2.4.2 Energy and the Hamiltonian . . . . . . . . . . . . . . . 43
2.5 Hamiltonian dynamics . . . . . . . . . . . . . . . . . . . . . . 49
2.6 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.6.1 Generalized forces and potentials . . . . . . . . . . . . 53
2.6.2 Friction . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.6.3 Lagrange’s equations with undetermined multipliers . . 56
1
2
2.6.4 d’Alembert’s principle . . . . . . . . . . . . . . . . . . 57
3 Applications 58
3.1 Central-force problem . . . . . . . . . . . . . . . . . . . . . . . 58
3.1.1 Preliminary . . . . . . . . . . . . . . . . . . . . . . . . 58
3.1.2 Reduction of the two-body problem to an effective one-
body problem . . . . . . . . . . . . . . . . . . . . . . . 59
3.1.3 Relative motion . . . . . . . . . . . . . . . . . . . . . . 61
3.2 Dynamics of rigid bodies . . . . . . . . . . . . . . . . . . . . . 69
3.2.1 Preparations . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.2 Kinetic energy and inertia tensor . . . . . . . . . . . . 70
3.2.3 Structure and properties of the inertia tensor . . . . . 71
3.2.4 Generalized coordinates and the Lagrangian . . . . . . 75
3.2.5 Equations of motion . . . . . . . . . . . . . . . . . . . 76
3.2.6 Angular momentum . . . . . . . . . . . . . . . . . . . 77
3.2.7 Applications: symmetric tops . . . . . . . . . . . . . . 79
3.3 Coupled oscillations . . . . . . . . . . . . . . . . . . . . . . . . 85
3.3.1 An illustrative example: two coupled oscillators . . . . 85
3.3.2 Lagrangian and equations of motion for coupled oscil-
lations: general case . . . . . . . . . . . . . . . . . . . 90
3.3.3 Solution of the EoMs . . . . . . . . . . . . . . . . . . . 91
A Supplementary material 100

A.1 Energy conservation of a conservative N-particle system . . . 100
A.2 Does S assume a minimum for the actual path (i.e., is the
stationary point always a minimum)? . . . . . . . . . . . . . . 101
A.3 Differential constraints . . . . . . . . . . . . . . . . . . . . . . 103
A.4 Details regaring the proof of equivalence of Newton’s and La-
grange’s equations of motion . . . . . . . . . . . . . . . . . . 105
A.5 Some details regarding rigid body dynamics . . . . . . . . . . 108
Chapter 1
Introduction
1.1 Prelude and Overview

What is Classical Mechanics (CM)?
CM deals with the motion of material objects through space and time and
with the laws that govern that motion.
Analysis:
Let’s go through this definition step by step:
(i) Material objects: The central property of the objects of CM is (inertial)

mass. We will see soon that mass is to be interpreted as resistance to
acceleration.
An important concept is that of a mass point: ⇔ a pointlike particle
with mass as its only property. Certainly, the idea of a (classical) mass
point is an idealization, but it is a useful one for two reasons: it greatly
simplifies actual calculations and it often is a good approximation even
if the objects are not small. Think of the Solar System. What makes
the idea of viewing the planets and the Sun as mass points work is that
the distances between them are large compared to their sizes.
(ii) Motion through space and time: There are a few (operational) things
to be said about space and time (pertaining to this course).
Space: is three-dimensional and Euclidian.
3
4
→ One can define Cartesian coordinate systems, and the mathematical

description of what happens in space is facilitated by vector algebra in
R3 .
Time: is just a homogeneous parameter.
→Motion: an object passes (continuously) through different positions
as time goes by, i.e., motion can be characterized by a trajectory r(t)
and its derivatives:

• trajectory r(t) 


 ⇒ ’kinematics’:

d mathematical description
• velocity v(t) = dt
r(t) = ṙ(t)

 of motion without

d  consideration of causes (forces)

• acceleration a(t) = dt
v(t) = v̇(t) = r̈(t)
Extensions of the notions of space and time (beyond this course)
– special theory of relativity → 4-dimensional ”Minkowski” space

(space-time)
– general theory of relativity → (locally) curved space
– quantum mechanics → ∞-dimensional Hilbert space
(iii) The governing laws: So, what causes the motion of objects? The
answer—according to Newton—is: forces. Analyzing the motion of
objects subject to forces is called dynamics (in contrast to kinematics,
where forces are not considered). This analysis is built on Newton’s
Laws, which we will recap shortly.
But first, let’s ask: What are we going to do in PHYS 3010 given that
Newton’s Laws are studied in PHYS 1010 and 2010?
The answer is: We will develop and apply alternate formulations of CM, the
so-called Lagrangian and Hamiltonan formulations (which are equivalent to
Newton’s). Why?
• To obtain deeper insights into the foundations of CM and physics

in general. For instance, we will discuss Hamilton’s principle which
is an example of a variational principle. Variational principles are
widespread in other branches of physics, e.g. in quantum mechanics.
5
• To learn a powerful problem-solving strategy (i.e., the application of

the Lagrangian Equations of Motion).
1.2 Recap of Newtonian Mechanics

1.2.1 Newton’s Laws (1687)
Lex prima: ”Every body continues in its state of rest, or uniform motion in
a straight line, unless it is compelled to change that state by forces impressed
upon it.”
Lex secunda: ”The change of motion is proportional to the motive force

impressed and is made in the direction of the line in which that force is
impressed.”
Lex tertia: ”To every action there is always imposed an equal reaction;
or, the mutual actions of two bodies upon each other are always equal and
directed to contrary parts.”
In addition, Newton formulated a corollary which is sometimes called his
fourth law. Its content is the principle of superposition of forces, i.e., it
states that forces add like vectors and that it is the net force that causes
the change of motion of an object according to his second law. Moreover, he
gave a definition of the ’motion’ he refers to in that law. It is the (linear)
momentum
p = mv
with the mass m and the velocity v.
Analysis
• The first law is Galileo’s principle of inertia. It contains the insight that
the states of rest and uniform motion are equivalent. As a consequence,
the descriptions of physical processes from the perspective of two ref-
erence frames which move uniformly with respect to each other are
also equivalent. In other words, this postulate introduces, somewhat
implicitly, inertial reference frames and Galilean transformations.
6
Definition of an inertial reference frame: a reference frame in which

a forcefree body moves uniformly or is at rest (i.e., a system in which
Newton’s first law holds).
S 2
S 1
m r 1(t) = R (t) + r 2(t)
r 2
= r re l + v re l t + r 2(t)
r v 1(t) = v re l + v 2(t)
1
R a 1(t) = a 2(t)
Figure 1.1: Mass point m seen from two reference frames which move uni-
formly (R(t) = rrel + vrel t) with respect to each other.
According to the first law we have:
if F = 0 −→ a1 = a2 = 0
if F 6= 0 −→ a1 = a2 6= 0
• The second law tells us (quantitatively) what happens if one or more

than one forces act on a body. It is the fundamental equation of motion
(EoM) of CM:
X
ṗ = Fnet ≡ Fi
i
d
⇐⇒ (mv) = Fnet
dt
7
if ṁ = 0: mv̇ = ma = mr̈ = Fnet
ṁ = 0 is certainly fulfilled for mass points, but ṁ 6= 0 is possible,

too: Think of a rocket whose fuel is burnt. In more general terms (and
beyond the scope of this course) Einstein showed us that ṁ 6= 0 is
actually not the exception but the rule, because the mass of a moving
object depends on its speed, and therefore (in general) on time.
Consequences:
(i) If 0 = Fnet = ṗ =⇒ p = const −→ recover first law!

(ii) Invariance of EoM wrt. Galilean transformations (see above):
S1 : F1 = ma1 = ma2 = F2 : S2
This is why inertial reference frames are so important: the forces

are the same in all of them, hence they are all equivalent for the
description of a physical process.
• The third law expresses a fundamental property of physical forces1 :

action equals reaction. Let’s consider two interacting mass points:

F12 : force on particle 1 due to particle 2 
F12 = −F21

F21 : force on particle 2 due to particle 1
2nd law
−→ m1 a1 = F12 = −F21 = −m2 a2
m1 |a2 | a2
=⇒ = ≡
m2 |a1 | a1
If one fixes the absolute scale by introducing a standard mass (the

kilogram) the last equation expresses a dynamical definition of mass in
1
which, according to our current understanding of the fundamental interactions, has
to be formulated somewhat differently in modern physics.
8
terms of accelerations: the smaller mass speeds up faster. Hence, we

can interpret mass as the resistance of an object to acceleration.
a ) b )
z z
F 1 2
m 1 F m 1
1 2
r 1 r 1
F 2 1
m 2
m 2
F 2 1
r 2 r 2
y y
Figure 1.2: Illustrations of the third law: the force vectors can be, but are
not necessarily directed along the line that joins the particles.
Note that Newton’s third law is fulfilled for the gravitational and the
electric forces, but not (at least not directly) for magnetic forces.
Further comments
(i) The physical origin of forces is (normally) not discussed in CM.
(ii) Instead, the basic problem of CM is to solve Newton’s EoM for given
forces. The EoM (normally) is a second-order ordinary differential
equation (ODE) of the form
r̈(t) = f(r, ṙ, t) (+ initial conditions)
Unfortunately, analytical solutions are only known for a few cases, but
numerical procedures are readily available (see, e.g., the various com-
puter problems in [FC]).
(iii) Conservation laws—briefly reviewed in the next section—emerge as

consequences of Newton’s Laws.
9
1.2.2 (Linear) momentum, angular momentum, work,

and energy
a) Momentum
simplest situation: one forcefree mass point (MP):
F = 0 =⇒ p = mv0 = const momentum conservation
֒→ r(t) = r0 + v0 t uniform motion

∢ system of N MPs and some useful notions:
• ’internal force’ fki : (force exerted on MP k by MP i)
• ’external force’ Fk : external force on kth MP
• ’isolated system’: a system without external forces
(Fk = 0 for k = 1, ..., N)
P
• total mass: M = N k=1 mk
PN
• centre of mass R = M1 k=1 mk rk
PN
• centre-of-mass velocity: V = Ṙ = M1 k=1 mk vk
P PN
• total = centre-of-mass momentum: P = MV = N k=1 mk vk = k=1 pk
• position of a MP wrt. centre of mass: r′ k = rk − R
B s p .: N = 2 z
m 1
r '1
r 1 r '2
R m 2
r 2
Figure 1.3: Definition of the centre of mass.

10
∢ EoM of kth MP:

N
X
ṗk = Fk + fki
i=1
note that fki = −fik −→ fkk = 0
N
X XN N
X
−→ ṗk = Fk + fki
k=1 k=1 i,k=1
q q q (3rd law)
Ṗ = Fext + 0
The centre of mass of a particle system moves as if it were a single particle

of mass M acted on by the total external force.
Momentum conservation (holds in isolated systems):
if Fext = 0 =⇒ Ṗ = 0 =⇒ P = const
This makes the centre-of-mass a convenient choice for the origin of an

inertial reference frame (for an isolated system).
b) Angular momentum
definition for one particle:
l = r × p = m(r × v)
|l| = l = rp sin γ
11
z z
p
p
r
g r
y y
l l g
Figure 1.4: On the left panel the angular momentum vector points out of the
yz-plane, on the right panel it points into it.
d
∢ l̇ = dt
(r × p) = m(v × v) + r × ṗ = r × F
definition: torque N = r × F
−→ l̇ = N
if N=0 =⇒ l̇ = 0 =⇒ l = const
The former equation is the fundamental EoM for rotational motion. The
latter expresses angular momentum conservation.
N = 0 if (i) F=0
(ii) Fkr central force
Let’s consider a system of N MPs. The total angular momentum is defined

12
as
N
X
L(t) = lk (t)
k=1
X
= rk (t) × pk (t)
k
X
−→ L̇ = (rk × ṗk )
k
N
X N
X
= (rk × Fk ) + (rk × fki )
k=1 i,k=1
X X 1 X
∢: (rk × fki ) = (ri × fik ) = (rk × fki ) + (ri × fik )
i,k i,k
2 i,k
1 X
= (rk × fki ) − (ri × fki )
2 i,k
1 X
= (rk − ri ) × fki
2 i,k
= 0 if fki k (rk − ri )
total torque is defined as

X X
N= (rk × Fk ) = Nk
k k
Accordingly, we have derived
L̇ = N
if N=0 =⇒ L̇ = 0 =⇒ L = const
In particular, total angular momentum is conserved in isolated systems.

c) Work and energy
13
Let’s start again with one MP. The most general definition of work is as
follows: if r(t) is the path on which the MP travels in the time interval [t0 , t]
the associated work W is
Z t

W = F r(t′ ), v(t′ ), t′ · v(t′ ) dt′ .
t0
If the force is a vector field, i.e., if F = F(r) W is given by a line integral
Z
W = F(r) dr with r (t)
K
dr = v(t′ ) dt′
K
r (t0)
Discussion:
(i) For uniform F and rectilinear motion: W = F · r
(ii) W = 0 if F ⊥ dr
Example 1: If you simply hold (but do not move) a mass m in Earth’s

gravity field −→ W = 0 (since v = 0)
Example 2: uniform circular motion
=r (R cos ωt, R sin ωt)

=v (−Rω sin ωt, Rω cos ωt)
=a (−Rω 2 cos ωt, −Rω 2 sin ωt)
= −ω 2 r
−→ ma = F(r)
= −mω 2 r
−→ F · v = 0 ⇐⇒ F ⊥ dr = v dt =⇒ W = 0
14
Let’s connect work with Newton’s EoM:

Z t

∢ W = F r(t′ ), v(t′ ), t′ · v(t′ ) dt′
t0
Z t
EoM
= m v̇(t′ ) · v(t′ ) dt′
t0
Z
m t d 2 ′ ′
= v (t ) dt
2 t0 dt′
m 2
= v (t) − v2 (t0 )
2
1
= p2 (t) − p2 (t0 )
2m
definition: kinetic energy

m 2 p2
T = v = ≥0
2 2m
−→ T (t) = T (t0 ) + W (t0 → t) ’work-energy theorem’
The work-energy theorem holds for all forces. A stronger relation is obtained
if the force field is conservative. For a conservative force a scalar potential
energy function U exists such that
F(r) = −∇U(r)
Note that U is defined only up to a constant by this equation. Normally one

r→∞
fixes this constant by requiring U(r) −→ 0. The conservativity of a force
can be formulated in different (but equivalent) ways:
F = −∇U ⇐⇒ ∇×F= 0
m տց m
Z 2 I
F(r) · dr ⇐⇒ F · dr = 0
1 C
independent of path
15
work − energy theorem −→ W = U(1) − U(2) = T (2) − T (1)

⇐⇒ T (1) + U(1) = T (2) + U(2)
E = T + U = const conservation of energy
In a more general situation the total force might consist of conservative and
dissipative parts
F = Fconservative + Fdissipative
∇ × F = ∇ × Fdiss 6= 0
∢ Newton’s EoM:
mv̇ = −∇U(r) + Fdiss |·v

Fdiss · v = mv̇v + ∇Uv
d m 2
⇐⇒ Fdiss · v = v + U(r(t))
dt 2
d dE
⇐⇒ T +U = = Fdiss · v
dt dt
energy is not conserved in this case, but changes according to the power
associated with the dissipative force.
Let’s consider a P
conservative N-particle system. Starting from Newton’s
EoM ṗk = Fk + N i=1 fki one obtains (see Appendix A.1)
d
T +U = 0 energy conservation
dt
T + U = E = const
where
P Pmk 2
• T = k Tk = k 2 vk : total kinetic energy
16
P P P P
• U = k Uk + k<i Ūki = k Uk + 21 k6=i Ūki : total potental energy
(note that Ūki = Ūik )
if the following applies:
(i) Fk = Fk (rk ) = −∇k Uk (rk ) (conservative external forces)
(ii) ∇k × fki = ∇i × fik = 0 (conservative internal forces)
(iii) fki = fki (rk − ri ) = −fik (rk − ri ) (internal forces fulfil Newton’s third
law and depend only on the relative coordinate rk − ri ).
Chapter 2
Hamilton’s Principle —
Lagrangian and Hamiltonian
dynamics
2.1 Preliminary formulation of Hamilton’s prin-

ciple (1834/35)
Of all the possible paths along which a particle may move from one point to
another in a given time interval [t1 , t2 ] the actual path followed is that which
minimizes the integral Z t2
S= (T − U) dt
t1
Comments:
(i) L := T − U: Lagrangian (function)
obviously, the Lagrangian has the dimension of an energy, but it is not
the energy of the system. Note that the definition implies that the
system under study is conservative.
Rt
(ii) S = t12 L dt: ’action integral’
(iii) Hamilton’s principle (HP) is an integral principle (in contrast to New-

ton’s law of motion)
17
18
(iv) HP is a fundamental principle of modern physics.
r 2 = r (t2)
s2 > s m in
s m in
s1 > s m in
r 1 = r (t1)
Figure 2.1: Illustration of Hamilton’s principle
(v) EoMs can be derived from HP

in order to do this we need a few basics of the calculus of variations.
2.1.1 Calculus of variations

given: reasonably well-behaved function f (x, ẋ, t)
sought-after: ’path’ x(t) with x(t1 ) = x1 and x(t2 ) = x2 such that

Z t2
I= f (x, ẋ, t) dt assumes extremum value
t1
Theorem: a necessary condition for I to assume an extremum value for x(t)

is Euler’s equation:
19
∂f d ∂f
− =0
∂x dt ∂ ẋ
Proof: assume that x(t) is the sought-after path

define neighbouring paths by
x(α, t) = x(t) + αη(t) with η(t1 ) = η(t2 ) = 0
֒→ ẋ(α, t) = ẋ(t) + αη̇(t)

the arbitrary function η(t) describes the deformation of the path and the
(real) parameter α is a scale factor that determines its magnitude
Z t2
∢ f (x(α, t), ẋ(α, t), t) dt = I(α)
t1
the integral can be viewed as an ordinary function of α. If this function

assumes an extremum value at α = 0 (i.e., for x = x(0, t)) it follows that
dI
=0
dα α=0
20
let’s work out the derivative:

Z t2
dI d
= f (x(α, t), ẋ(α, t), t) dt
dα dα t1
Z t2
∂
= f (x(α, t), ẋ(α, t), t) dt
t1 ∂α
Z t2
∂f ∂x ∂f ∂ ẋ
= + dt
t1 ∂x ∂α ∂ ẋ ∂α
Z t2
∂f ∂f
= η(t) + η̇(t) dt
t1 ∂x ∂ ẋ
Z t2 t2 Z t2 d ∂f
∂f ∂f
integration by parts : = η(t) dt + η(t) − η(t) dt
t1 ∂x ∂ ẋ t1 t1 dt ∂ ẋ
Z t2
∂f d ∂f
= − η(t) dt
t1 ∂x dt ∂ ẋ
(since η(t1 ) = η(t2 ) = 0)
Z t2
dI ∂f d ∂f
֒→ = 0 = − η(t) dt for x = x(0, t)
dα α=0 t1 ∂x dt ∂ ẋ
η(t) arbitrary ∂f d ∂f
=⇒ − =0
∂x dt ∂ ẋ
Elementary example:
x (t)
x 2
x v(t) shortest distance between two points in a plane
x (t) R t2 √
x 1
length of path: I = t1
1 + ẋ2 dt
t1 t2 t
√
֒→ f= 1 + ẋ2 = f (ẋ)
∂f ∂f ẋ Euler d ẋ
=0; =√ −→ √ =0
∂x ∂ ẋ 1 + ẋ2 dt 1 + ẋ2
ẋ
=⇒ √ = const = C1
1 + ẋ2
21
s
C12
֒→ ẋ = ± = C2 =⇒ x(t) = C2 t + C3 (rectilinear path)
1 − C12
Remarks:
(i) A more interesting (and quite famous) application is the so-called

brachistochrone problem. It is example 6.2 in both [Tay] and [TM].
(ii) If the first derivative of a function vanishs the function is said to be

stationary at that point. Stationarity is necessary, but not sufficient for
a a minimum (not even for an extremum). Since the Euler equation is
equivalent to the stationarity of I at α = 0 we don’t know if its solution
really minimizes the integral. In fact, there are counterexamples (see,
e.g., Chap 6.3 of [Tay]) and, accordingly, a more cautious statement of
HP (which will be discussed later) refers only to the stationarity of the
action integral. For practical applications, this is of no concern.
(iii) δ-notation
It is quite common to use a short-hand notation in variational calculus—
the so-called δ-notation. To introduce it let’s go back to the calculation
dI
of dα on the previous page. We had
Z t2 ∂f
dI d ∂f
= − η(t) dt
dα t1 ∂x dt ∂ ẋ
∂x
this can be rewritten by noting that η(t) = ∂α and multiplying both
sides by a ’small’ scale factor dα
Z t2
dI ∂f d ∂f ∂x
dα = − dα dt
dα t1 ∂x dt ∂ ẋ ∂α
Z t2
∂f d ∂f
⇐⇒ δI = − δx dt
t1 ∂x dt ∂ ẋ
dI ∂x
with the ’variations’ δI = dα
dα and δx = ∂α
dα.
22
The stationarity of I is then expressed as
δI = 0Z Z
t2 t2
⇐⇒ δ f (x, ẋ, t) dt = δf dt
t1 t1
Z t2
∂f d ∂f
= − δx dt = 0
t1 ∂x dt ∂ ẋ
More about the δ-notation: [TM], Chap. 6.7
(iv) A more rigorous definition of variations δI and δx is based on the

mathematical notion of a functional. A functional I[x] is a generalized
function: it maps a function to a number: x(t) 7→ I. Our integral is a
typical example Z t2
I[x] = f (x.ẋ, t) dt
t1
One can construct a calculus for functionals, i.e., set up precise defi-
nitions and rules of how to differentiate them. Functional derivatives
then turn out to be closely related to our variations, i.e., the calculus for
functionals provides a foundation of the symbolic notation introduced
above.
(v) More on calculus of variations: [TM], Chap. 6; [Tay], Chap. 6
2.1.2 HP for a simple case

∢ 1 single mass point in one-dimensional world
m 2 dU
L=T −U = ẋ − U(x) = L(x, ẋ) ; F (x) = −
2 dx
Z t2
HP : δS = δ L(x, ẋ) dt = 0
t1
d ∂L ∂L
⇐⇒ − =0 Lagrangian equation
dt ∂ ẋ ∂x
23
let’s work out the derivatives:

∂L d ∂L
= mẋ ; = mẍ ;
∂ ẋ dt ∂ ẋ
∂L dU
= − = F (x)
∂x dx
֒→ mẍ = F (x) ⇐⇒ Lagrange ⇐⇒ HP
Appendix A.2 deals with the question whether S always assumes a minimum
on the actual path (in the 1D world).
2.2 Constrained systems and generalized co-

ordinates
2.2.1 Preliminary for one mass point
a) Examples for constraints
(i) Motion on an inclined plane (without friction)
z constraint: z = (− tan α)x + h

h
−→ number of degrees of freedom (dofs) is
reduced to 2
a
x
24
(ii) Motion on the surface of a sphere (no friction)
constraint: x2 + y 2 + z 2 = R2
R
x
−→ two dofs
(iii) Motion on the rim of a circle of radius r 2 = R2 − z02
z constraints: x2 + y 2 + z 2 = R2
z = z0 < R
y −→ one dof
z 0
x special case: z0 = R:
zero dof (particle cannot move)
if z0 >R the two constraints are not compatible

(iv) Planar pendulum
y constraints:
p z=0
x l = x2 + y 2 = const
l
j same as example (iii)
m
all constraints in examples (i) - (iv) are characterized by equations

of the form f (x, y, z) = 0. They are called scleronomic-holonomic
constraints.
25
But there are other types of constraints:

(v) Bead on a uniformly rotating wire
y constraints: y = tan(ωt)x
z=0
j = w t −→ one dof
x
the first constraint is characterized by an equation of the form

f (x, y, z, t) = 0, i.e., the constraint involves the time. It is called
rheonomic-holonomic.
(vi) Mass point within a spherical container
constraint: x2 + y 2 + z 2 < R
x
variant:
y
x2 + y 2 + z 2 ≥ R2
−→ these are nonholonomic constraints (characterized by inequal-

ities)
−→ number of dofs is not reduced.
26
(vii) Upright disk rolling on a horizontal plane (without slipping)
centre of mass velocity is locked to angular velocity

ω = ψ̇. Hence:
y vCM = ±vCM cos ϕî ± sin ϕĵ
vCM = Rψ̇
j
x → differential (nonholonomic) constraints:
bird eye’s view on the dxCM = ±Rdψ cos ϕ , dyCM = ±Rdψ sin ϕ
upright disk (for a spe-
cial situation) → number of dofs is not reduced
Summary

scleronomic : f (x, y, z) = 0
holonomic constraints
rheonomic : f (x, y, z, t) = 0

inequalities
nonholonomic constraints
differential
holonomic constraints reduce the number of dofs, but nonholonomic
ones do not.
b) Generalized coordinates and Lagrangian equations

∢ one-particle system with holonomic constraint f (x, y, z) = 0. Let’s
assume that the equation of constraint can be solved for z = z(x, y).
Then, the z-coordinate can be eliminated from the Lagrangian:
−→ L(x, y, z, ẋ, ẏ, ż, t) = L(x, y, z(x, y), ẋ, ẏ, ż(x, y, ẋ, ẏ), t)
= L̃(x, y, ẋ, ẏ, t)
see example (i): inclined plane

constraint: f (x, y, z) = h − x tan α − z = 0 ⇐⇒ z(x) = h − x tan α
in addition let’s assume y = 0 and that gravity acts on the particle:
27
m 2 m 2 m
v = (ẋ + ẏ 2 + ż 2 ) = (ẋ2 + ẋ2 tan2 α)
T =
2 2 2
U = mgz = mg(h − x tan α)
m 2
֒→ L=T −U = ẋ (1 + tan2 α) − mg(h − x tan α)
2
= L(x, ẋ)
d ∂L ∂L
HP ⇐⇒ − =0
dt ∂ ẋ ∂x
work out the derivatives:
∂L d ∂L mẍ ∂L
= mẋ(1 + tan2 α), = , = mg tan α
∂ ẋ dt ∂ ẋ cos2 α ∂x
Lag.eq.
=⇒ mẍ − mg tan α cos2 α = 0
⇐⇒ ẍ = g sin α cos α
g
solution : x(t) = (sin α cos α)t2 + C1 t + C2
2
g
z(t) = h − (sin2 α)t2 − (C1 t + C2 ) tan α
2
One obtains—of course— the same result in the Newtonian framework.
The treatment is, however, rather different. One has to recognize that
the plane exerts a normal force on the particle such that the resulting
(net) force points in the direction of the inclination. The acceleration is
then determined by the net force, i.e., the projection of the gravitational
force in this direction:
ma = Fnet = mg sin α
integration yields
g
s(t) = t2 sin α + C3 t + C4
2
28
this result can be written in terms of the x- and z-coordinates by using

sin α = (h − z)/s and x = s cos α. Apparently, the advantage of the
Lagrangian treatment is that it is not necessary to think about the
normal force (i.e., the force of constraint).
see example (iv): planar pendulum
let’s assume that gravity is acting: U = mgy
constraint : z = 0
p
x = ± l2 − y 2
m 2
L=T −U = (ẋ + ẏ 2 ) − mgy
2
m y 2 ẏ 2 + ẏ 2 (l2 − y 2 )
= − mgy
2 l2 − y 2
m ẏ 2l2
= − mgy = L(y, ẏ)
2 l2 − y 2
−→ this looks complicated!
It is a much better idea to work in polar coordinates:

p
r = x2 + y 2 = l = const
x
tan ϕ = −
y
⇐⇒ x = r sin ϕ y = −r cos ϕ
= l sin ϕ = −l cos ϕ
U = mgy = −mgl cos ϕ

ẋ = lϕ̇ cos ϕ , ẏ = lϕ̇ sin ϕ
m 2 2
֒→ L = l ϕ̇ + mgl cos ϕ = L(ϕ, ϕ̇)
2
Lagrangian equation of motion:
∂L d ∂L ∂L
= ml2 ϕ̇ , = ml2 ϕ̈ , = −mgl sin ϕ
∂ ϕ̇ dt ∂ ϕ̇ ∂ϕ
=⇒ ml2 ϕ̈ + mgl sin ϕ = 0

29
r
2 g
ϕ̈ + ω sin ϕ = 0 , ω=
l
compare this to the Newtonian treatment (where another force of con-

straint, the tension of the rod, must be introduced).
For small angles one can assume sin ϕ ≈ ϕ
=⇒ ϕ̈ + ω 2 ϕ = 0 =⇒ ϕ(t) = a sin(ωt − β)
for larger angles the oscillation is nonlinear and the solution much more
involved (see [TM], Chap. 4.4).
another example: (a variant of) the cycloidal pendulum
a cycloid is the path of a point on the edge of a circular wheel that
roles on a straight line. If the wheel roles in x-direction the equations
of the cycloid are
x = R(α − sin α)
y = R(1 − cos α)
where α is the angle through which the rolling circle rotates.

Let’s consider a bead that slides on a wire that has the form of an
inverted cycloid
x = R(α − sin α)
y = R(1 + cos α)
The suitable (generalized) coordinate to express the Lagrangian is the

angle α. The potential energy is
U = mgy = mgR(1 + cos α)
we need
ẋ = Rα̇(1 − cos α)
ẏ = −Rα̇ sin α
30
to rewrite the kinetic energy

m 2
T = (ẋ + ẏ 2) = mR2 α̇2 (1 − cos α)
2
֒→ L = T − U = mR2 α̇2 (1 − cos α) − mgR(1 + cos α) = L(α, α̇)
derivatives:
∂L
= 2mR2 α̇(1 − cos α)
∂ α̇
d ∂L
= 2mR2 (α̈(1 − cos α) + α̇2 sin α)
dt ∂ α̇
∂L
= mR2 α̇2 sin α + mgR sin α
∂α
Lagrangian equation of motion
g
=⇒ 2α̈(1 − cos α) + α̇2 sin α − sin α = 0
R
this seemingly complicated equation of motion can be simplified by an
ingenious coordinate transformation:
step (i): use 1 − cos α = 2 sin2 α2 , sin α = 2 sin α2 cos α2
α α g α
֒→ 2α̈ sin + α̇2 cos − cos = 0
2 2 R 2
step (ii): substitute u = − cos α2 :

α̇ α
֒→ u̇ = sin
2 2
α̈ α α̇2 α
ü = sin + cos
2 2 4 2
1 α α
= (2α̈ sin + α̇2 cos )
4 2 2
EoM g
−→ ü + u=0
4R
this is nothing butpthe equation of motion of an harmonic oscillator
g
of frequency ω = 4R . The motion is said to be isochronous, i.e., it
31
has a frequency that is independent of the amplitude. To complete the

solution one has to find α(t) by using the solution u(t) in the coordinate
transformation, and then insert α(t) in the cycloid equations to find
x(t) and y(t) (try it!).
Conclusions:
– Holonomic constraints can be used to eliminate coordinates.

– Suitable coordinates yield simple Lagrangian equations. Unfortu-
nately, there is no general recipe of how to find such coordinates.
2.2.2 N-particle systems

Let’s generalize things to be able to deal with systems with many particles
and many degrees of freedom. We start by introducing a convenient nomen-
clature:
r1 r2 .... rN
x1 y1 z1 x2 y2 z2 xN yN zN
↓ ↓ ↓ ↓ ↓ ↓ ւ ↓ ց
x1 x2 x3 x4 x5 x6 x3N −2 x3N −1 x3N
m1 m2 mN
ւ ↓ ց ւ ↓ ց ւ ↓ ց
m1 = m2 = m3 m4 = m5 = m6 m3N −2 = m3N −1 = m3N
N 3N
X mi 1X
yields, e.g. : T = vi2 = mi ẋ2i
i=1
2 2 i=1
a) Classification of constraints
(i) k1 scleronomic-holonomic constraints
fi (x1 , ..., , x3N ) = 0 ; i = 1, ..., k1

(ii) k2 rheonomic-holonomic constraints
fj (x1 , ..., , x3N , t) = 0 ; j = 1, ..., k2

32
→ k1 + k2 = k holonomic constraints reduce the number of

degrees of freedom of an N-particle system from 3N to 3N − k (if
they are independent and compatible with each other).
(iii) m nonholonomic differential constraints
... can also be characterized by equations (see Appendix A.3), but
they do not reduce the number of dofs.
b) Generalized coordinates and configuration space

goal: description of N-particle system in terms of suitable (generalized)
coordinates
typical starting point: Cartesian coordinates
consider coordinate transformation (x1 , ..., x3N ) ←→ (q1 , ..., q3N )
xi = xi (q1 , ..., q3N , t) i = 1, ..., 3N
generalized coordinates qµ = qµ (x1 , ..., x3N , t) µ = 1, ..., 3N
assume k independent holonomic constraints
fj (x1 , ..., x3N , t) = 0 [or fj (q1 , ..., q3N , t) = 0], j = 1, ..., k
choose

q3N −k+1 = f1 (x1 , ..., x3N , t) = 0 


q3N −k+2 = f2 (x1 , ..., x3N , t) = 0 
.. (ignorable coordinates)
. 


q3N = fk (x1 , ..., x3N , t) = 0 
=⇒ 3N −k independent generalized coordinates remain. They describe

the system completely.
q = (q1 , ..., q3N −k ) : ′′ configuration (vector)′′

= point in (3N − k)−dim. configuration space
֒→ xi = xi (q1 , ..., q3N −k , t)
d
֒→ ẋi = xi (q1 , ..., q3N −k , t)
dt
3N −k
X ∂xi ∂xi
= q̇µ +
µ=1
∂qµ ∂t
= ẋi (q, q̇, t), i = 1, ..., 3N
33
q̇µ : generalized velocity
֒→ L = L(q, q̇, t)
example: double pendulum (N = 2)
z y Cartesian coordinates (x1 , ..., x6 )

x 
l constraints
p : x 3 = x6 = 0 
A
ϑ 2
lA = px1 + x2 2
=⇒ 4 constraints
A mA

lB lB = (x4 − x1 )2 + (x5 − x2 )2
ϑ mB
B
=⇒ 2 dofs
generalized coordinates:

q1 = ϑA 
characterize motion in 2 − dim. configuration space

q2 = ϑB

q3 = x3 = 0 , q6 = x6 = 0 



p 
q4 = lA − x21 + x22 = 0 ignorable


p 


q5 = lB − (x4 − x1 )2 + (x5 − x2 )2 = 0
6
1X
L=T −U = mi ẋ2i − g(m2 x2 + m5 x5 )
2 i=1
coordinate transformation:
x1 = lA sin ϑA = lA sin q1
x2 = −lA cos ϑA = −lA cos q1
x3 = 0
x4 = lA sin q1 + lB sin q2
x5 = −lA cos q1 − lB cos q2
x6 = 0
general form xi = xi (q1 , q2 ) , i = 1, ..., 6
34
ẋ1 = lA q̇1 cos q1

ẋ2 = lA q̇1 sin q1
ẋ3 = 0
ẋ4 = lA q̇1 cos q1 + lB q̇2 cos q2
ẋ5 = lA q̇1 sin q1 + lB q̇2 sin q2
ẋ6 = 0
general form ẋi = ẋi (q1 , q2 , q̇1 , q̇2 ) , i = 1, ..., 6
use conventional nomenclature for the masses
(m1 , m2 , m3 ) −→ mA (m4 , m5 , m6 ) −→ mB
to obtain the Lagrangian

mA 2 2 2 m h
B 2
֒→ L = lA q̇1 cos q1 + sin2 q1 + lA q̇1 cos q1 + lB q̇2 cos q2
2 2
2 i
+ lA q̇1 sin q1 + lB q̇2 sin q2 + mA glA cos q1 + mB g lA cos q1 + lB cos q2
mA 2 2 mB 2 2 mB 2 2
= lA q̇1 + lA q̇1 + lB q̇2 + mB lA lB q̇1 q̇2 cos q1 cos q2 + sin q1 sin q2
2 2 2
+ (mA + mB )glA cos q1 + mB glB cos q2
mA + mB 2 2 mB 2 2
= lA q̇1 + l q̇ + mB lA lB q̇1 q̇2 cos(q1 − q2 )
2 2 B 2
+ (mA + mB )glA cos q1 + mB glB cos q2
= L(q1 , q2 , q̇1 , q̇2 )
35
2.3 General formulation of Hamilton’s prin-

ciple and Langrange’s equations for N-
particle systems
2.3.1 Hamilton’s principle
Of all the possible paths along which a particle system with 3N-k degrees of
freedom may move from one point in configuration space to another one in a
given time interval [t1 , t2 ] the actual path is such that the action integral
Z t2
S= L(q, q̇, t) dt
t1
is stationary, i.e., δS = 0.
Comments:
(i) ”Path”: motion in 3N − k dimensional configuration space
q(t) = {q1 (t), ..., q3N −k (t), t1 ≤ t ≤ t2 }
(ii) In most cases the stationary point is a minimum.
(iii) Derivation of Lagrangian equations of motion from HP

it is straightforward to generalize the arguments of Sec. 2.1.1
– assume that q(t) is the actual path for which δS = 0

– neighbouring paths:
qµ (α, t) = qµ (t) + αηµ (t) (µ = 1, ..., 3N − k)

q̇µ (α, t) = q̇µ (t) + αη̇µ (t)
with ηµ (t1 ) = ηµ (t2 ) = 0
R t2
– ∢ S[q] = L(q(α, t), q̇(α, t), t) dt = S(α)
t1

dS
– q(t) is the sought-after path ⇔ dα =0
α=0
36
Z t2
dS ∂
• = L(q(α, t), q̇(α, t), t) dt
dα t1 ∂α
3N −k Z t2
X ∂L ∂qµ ∂L ∂ q̇µ
= + dt
µ=1 t 1
∂q µ ∂α ∂ q̇µ ∂α
3N −k Z t2
X ∂L ∂L
= ηµ (t) + η̇µ (t) dt
µ=1 t1 ∂qµ ∂ q̇µ
3N −k t2 3N −k Z t2
X ∂L X ∂L d ∂L
= ηµ (t) + − ηµ (t) dt
µ=1
∂ q̇µ t1
µ=1 t1
∂q µ dt ∂ q̇µ
| {z }
=0
d ∂L ∂L
dS
dα
⇐⇒ − =0, µ = 1, ..., 3N − k
α=0 dt ∂ q̇µ ∂qµ
(for qµ = qµ (α = 0, t))
2.3.2 Equivalence of Lagrange’s and Newton’s equa-

tions of motion
a) System without constraints in Cartesian coordinates
show:
d ∂L ∂L
− =0, i = 1, ..., 3N
dt ∂ ẋi ∂xi
XN
⇐⇒ ṗk = Fk + fkj , k = 1, ..., N
j=1
37
proof:
3N
1X
L = T −U = mj ẋ2j − U(x1 , ..., x3N )
2 j=1
d ∂L d ∂T
֒→ = = mi ẍi = ṗi (i = 1, ..., 3N)
dt ∂ ẋi dt ∂ ẋi
3N
∂L ∂U X
= − = Fi + fij
∂xi ∂xi j=1
a detailed analysis of the last equation can be found in Appendix A.4.
b) Invariance of the Lagrangian equations to coordinate transformations
goal: show
d ∂L ∂L d ∂L ∂L
− =0 ⇐⇒ − =0
dt ∂ ẋi ∂xi dt ∂ q̇µ ∂qµ
with xi = xi (q1 , ..., q3N , t) , i = 1, ..., 3N

=⇒ Lagrangian equations of motion in 3N generalized coordinates ⇐⇒
Newtonian equations of motion in Cartesian coordinates
let’s be a bit more general and show invariance of the Lagrangian equa-
tions with respect to general coordinate transformations
qµ → Qα = Qα (q, t) , α = 1, ..., n
qµ = qµ (Q, t) , µ = 1, ..., n
n
d X ∂qµ ∂qµ
ingredient : q̇µ = qµ (Q1 , ..., Qn , t) = Q̇β +
dt β=1
∂Qβ ∂t
= q̇µ (Q1 ...Qn , Q̇1 ...Q̇n , t)
∂ q̇µ ∂ X ∂qµ ∂qµ ∂qµ

֒→ = Q̇β + =
∂ Q̇α ∂ Q̇α β ∂Qβ ∂t ∂Qα
38
d ∂L ∂L
assume − =0, µ = 1, ..., n
dt ∂ q̇µ ∂qµ

L(q1 ...qn , q̇1 ...q̇n , t) = L q1 (Q1 ...Qn , t), q2 (Q1 ...Qn , t), ..., q̇1 (Q1 ...Qn , Q̇1 ...Q̇n , t), ..., t
= L̃(Q1 ...Qn , Q̇1 ...Q̇n , t)
d ∂ L̃ ∂ L̃
show − = 0, α = 1, ..., n
dt ∂ Q̇α ∂Qα
∂ L̃ X ∂L ∂qµ ∂L ∂ q̇µ
• = +
∂Qα µ
∂qµ ∂Qα ∂ q̇µ ∂Qα
∂ L̃ X ∂L ∂ q̇µ X ∂L ∂qµ
• = =
∂ Q̇α µ
∂ q̇µ ∂ Q̇α µ
∂ q̇µ ∂Qα
" #
d ∂ L̃ d X ∂L ∂qµ X d ∂L ∂qµ ∂L d ∂qµ
• = = +
dt ∂ Q̇α dt µ ∂ q̇µ ∂Qα µ
dt ∂ q̇µ ∂Qα ∂ q̇µ dt ∂Qα
| {z }
∂ q̇µ
=
∂Qα
" #
d ∂ L̃ ∂ L̃ X d ∂L ∂L ∂qµ
−→ − = − =0 (q.e.d.)
dt ∂ Q̇α ∂Qα µ
dt ∂ q̇µ ∂qµ ∂Qα
| {z }
=0
Comments:
(i) Special case: n = 3N und Qα ≡ xi

֒→ Lagrangian equations in 3N generalized coordinates ⇐⇒ Newtonian
equations in Cartesian coordinates
(ii) Newtonian equations of motion are invariant to Galilean transforma-

tions, but not to general coordinate transformations!
i.e. mi ẍi = Fi does not imply mµ q̈µ = Fµ

39
(iii) Holonomic constraints
assume q = {q1 ...q3N −k , q3N −k−1 ...q3N }

| {z }
ignorable
HP : δS = 0 for L = L(q1 ...q3N , q̇1 ...q̇3N , t)

3N Z t2
dS X ∂L d ∂L
⇐⇒ 0 = = − ηµ (t) dt
dα α=0 µ=1 t1 ∂qµ dt ∂ q̇µ
3N −k Z t2
X ∂L d ∂L
= − ηµ (t) dt
µ=1 t1
∂qµ dt ∂ q̇µ
3N Z t2
X ∂L d ∂L
+ − ηµ (t) dt
t1 ∂qµ dt ∂ q̇µ | {z }
µ=3N −k+1
= 0 for µ = 3N − k + 1, ..., 3N
(ignorable coordinates are not varied)
∂L d ∂L
⇐⇒ − = 0 for µ = 1, ..., 3N − k
∂qµ dt ∂ q̇µ
−→ HP ⇐⇒ Lagrangian eqs. for 3N − k generalized coordinates ⇐⇒

Newtonian eqs. for 3N Cartesian coordinates + forces of constraint
(iv) Lagrangian recipe
– formulate k (holonomic) constraints

– choose 3N generalized coordinates, identify k of them with con-
straints
– set up L = T − U in 3N Cartesian (or other) coordinates
– work out coordinate transformation to 3N − k generalized coor-
dinates
– find L = L(q1 ...q3N −k , q̇1 ...q̇3N −k , t)
40
– work out
d ∂L ∂L
− =0 (µ = 1, ..., 3N − k)
dt ∂ q̇µ ∂qµ
– solve equations of motion and analyze solutions

41
2.4 Conservation theorems revisited

2.4.1 Generalized momenta
definition of a generalized (aka canonical, aka conjugate) momentum
∂L
pµ =
∂ q̇µ
example 1: Cartesian coordinates
∂L ∂T 1 ∂ X 2

pi = = = mj ẋj = mi ẋi
∂ ẋi ∂ ẋi 2 ∂xi j
−→ usual linear momentum component

example 2: pendulum in xy-plane
m 2 2
L = l q̇ + mgl cos q , (q = ϕ)
2
∂L
֒→ p = = ml2 q̇ = ml2 ϕ̇ = lz = (r × p)z
∂ q̇
−→ z-component of the angular momentum
question: when is ṗµ = 0?
d ∂L ∂L
֒→ Lagrangian eqs. : = ṗµ =
dt ∂ q̇µ ∂qµ

∂L  ṗµ = 0
֒→ if =0 ⇒
∂qµ  p = ∂L
= const
µ ∂ q̇µ
example 1: free particle
1X
L = T = mi ẋ2i
2 i
∂L
֒→ =0 ⇐⇒ pi = mi ẋi = const
∂xi
42
example 2: particle on the inside surface of a cone
let’s assume that the cone (angle α) is standing upright with its symmetry
axis parallel to the z-axis.
√ 2 2
x +y
• constraint tan α = z
• generalized (cylindrical) coordinates

p
q1 = r = x2 + y 2
q2 = ϕ
r
q3 = tan α − = 0 (ignorable)
z
• Lagrangian in Cartesian coordinates
m 2
L= T −U = (ẋ + ẏ 2 + ż 2 ) − mgz
2
• coordinate transformation
x = r cos ϕ = q1 cos q2
y = r sin ϕ = q1 sin q2
z = r cot α = q1 cot α
ẋ = q̇1 cos q2 − q1 q̇2 sin q2
ẏ = q̇1 sin q2 + q1 q̇2 cos q2
ż = q̇1 cot α
• Lagrangian in generalized coordinates

m
L = ((q̇1 cos q2 − q1 q̇2 sin q2 )2 + (q̇1 sin q2 + q1 q̇2 cos q2 )2 + q̇12 cot2 α) − mgq1 cot α
2
m q̇12 2 2
= + q1 q̇2 − mgq1 cot α
2 sin2 α
= L(q1 , q̇1 , q̇2 )
• conserved momentum
∂L ∂L
=0 ֒→ p2 = = mq12 q̇2 = mr 2 ϕ = lz = const
∂q2 ∂ q̇2
43
• remaining equation of motion
d ∂L mq̈12
=
dt ∂ q̇1 sin2 α
∂L
= mq1 q̇22 − mg cot α
∂q1
=⇒ q̈1 − q1 q̇22 sin2 α + g sin α cos α = 0

p2
eliminate q̇2 by using q̇2 = mq12
p22 sin2 α
֒→ q̈1 − + g sin α cos α = 0
m2 q13
Comments:
(i) equation of motion in conventional notation
lz2 sin2 α
r̈ − + g sin α cos α = 0
m2 r 3
(ii) alternate treatment with (z, ϕ) as generalized coordinates is also pos-

sible (and is very similar due to the linear relation between r and z)
(iii) a coordinate which is conjugate to a conserved momentum (such as

q2 = ϕ in the previous example) is called a cyclic coordinate1
2.4.2 Energy and the Hamiltonian

a) Preparation 1: Euler’s theorem for homogeneous functions
definition: f (x1 ...xm ) is a homogeneous function of degree n:
⇐⇒ f (λx1 , λx2 , ..., λxm ) = λn f (x1 ...xm )
Euler: if f is a homogeneous function of degree n

m
X ∂f
=⇒ xi = nf (x1 , ..., xm )
i=1
∂xi
1
some authors call these coordinates ignorable
44
proof: yi = λxi
∂f X ∂f ∂yi
∢ (y1, ..., ym ) =
∂λ i
∂yi ∂λ
X ∂f
= xi
i
∂y i
∂
= (λn f )
∂λ
= nλn−1 f (x1 , ..., xm )
for λ = 1 ֒→ yi = xi and
X ∂f
xi = nf (x1 , ..., xm ) .
i
∂xi
b) Preparation 2: a theorem concerning kinetic energy

theorem: For time-independent transformations xi ←→ qµ
(xi = xi (q)) T is a homogeneous function of degree 2 in the generalized
velocities.
1X
proof : T = mi ẋ2i
2 i
1 X X ∂xi ∂xi
= mi q̇µ q̇ν
2 i µ,ν
∂qµ ∂qν
T (λq̇) = λ2 T (q̇)
Euler
X ∂T
=⇒ q̇µ = 2T
µ
∂ q̇µ
c) The Hamiltonian
The Lagrangian is not the total energy of a conservative system (unless
U = 0) and normally not a conserved quantity. Still, it is useful to look
at its total time derivative:
45
d X ∂L ∂L ∂L
L(q, q̇, t) = q̇µ + q̈µ +
dt µ
∂qµ ∂ q̇µ ∂t
Lag.eqs.
X h d ∂L ∂L d i ∂L
= q̇µ + q̇µ +
µ
dt ∂ q̇µ ∂ q̇µ dt ∂t

d X ∂L ∂L
= q̇µ +
dt µ ∂ q̇µ ∂t
d nX o ∂L
⇐⇒ pµ q̇µ − L = −
dt µ
∂t
definition: Hamiltonian function
X
H= pµ q̇µ − L
µ
dH ∂L
֒→ =−
dt ∂t

∂L  Ḣ = 0
=⇒ if =0 =⇒
∂t 
H = const
Discussion:

(i) H = L = E = Joule
(ii) H ≡ E = T + U if
• system conservative
46
• at most holonomic constraints

• time-independent transformation xi → qµ
the latter two conditions are normally realized through sclero-
nomic constraints and resting reference frames
proof:
∂L ∂T ∂U
pµ = = ( = 0)
∂ q̇µ ∂ q̇µ ∂ q̇µ
X X ∂T
֒→ H= pµ q̇µ − L = q̇µ − L = 2T − T + U
µ µ
∂ q̇µ
= T +U
(iii) conditions for H = E = T + U are sufficient, but not necessary

(iv) H = E and Ḣ = 0 are independent
֒→ it is possible that Ḣ = 0 and H 6= E
֒→ it is possible that Ḣ 6= 0 and H = E
∂L
(iv) if conditions listed in (ii) are fulfilled and if ∂t
=0
=⇒ H = E = T + U = const
d) Examples
(i) one-dimensional harmonic oscillator

m 2 m m
ẋ , U = ω 2 x2 , L = T − U = (ẋ2 − ω 2x2 )
T =
2 2 2
m 2 m 2 2 m 2
֒→ H = pẋ − L = mẋ − ẋ + ω x = (ẋ + ω 2 x2 ) = E = const
2
2 2 2
(ii) planar pendulum
m 2 2
L = l ϕ̇ + mgl cos ϕ
2
with pϕ = ml2 ϕ̇
m
H = pϕ ϕ̇ − L = ml2 ϕ̇2 − l2 ϕ̇2 − mgl cos ϕ
2
m 2 2
= l ϕ̇ − mgl cos ϕ = T + U = E = const
2
47
(iii) planar double pendulum

mA + mB 2 2 mB 2 2
T = lA ϑ̇A + l ϑ̇ + mB lA lB ϑ̇A ϑ̇B cos(ϑA − ϑB )
2 2 B B
U = −(mA + mB )glA cos ϑA − mB glB cos ϑB
∂L 2
p1 = = (mA + mB )lA ϑ̇A + mB lA lB ϑ̇B cos(ϑA − ϑB )
∂ ϑ̇A
∂L 2
p2 = = mB lB ϑ̇B + mB lA lB ϑ̇A cos(ϑA − ϑB )
∂ ϑ̇B
H = p1 ϑ̇A + p2 ϑ̇B − T + U
2 2
= (mA + mB )lA ϑ̇A + mB lA lB ϑ̇A ϑ̇B cos(ϑA − ϑB )
2 2
+ mB lB ϑ̇B + mB lA lB ϑ̇A ϑ̇B cos(ϑA − ϑB )
− T + U = T + U = E = const
(iv) bead on a rotating wire
rheonomic constraint : y = x tan ωt ⇐⇒ ϕ−ωt = 0

y

q1 = r
generalized coordinates
q2 = ϕ − ωt = 0 (ignorable)
j = w t
x
• assume U = 0
m 2 m 2
L = T = (ṙ + r 2 ϕ̇2 ) = (ṙ + r 2 ω 2 ) = E
2 2
∂L m m
H = pṙ − L = ṙ − L = mṙ 2 − ṙ 2 − r 2 ω 2
∂ ṙ 2 2
m 2 2 2
= (ṙ − r ω ) 6= E
2
∂L dH
−
= =0 −→ H = const
∂t dt
Lagrangian equation of motion:
d ∂L ∂L
− = 0
dt ∂ ṙ ∂r
k k
mr̈ − mω 2 r = 0 ⇐⇒ r̈ − ω 2 r = 0
48
general solution
r(t) = C1 eωt + C2 e−ωt

ṙ(t) = C1 ωeωt − C2 ωe−ωt
m n ωt −ωt
2
2

ωt −ωt
2 o
֒→ H = C1 ωe − C2 ωe − ω C1 e + C2 e
2
m 2 2 2ωt
= C1 ω e + C22 ω 2 e−2ωt − 2C1 C2 ω 2 − C12 ω 2e2ωt
2
− C22 ω 2 e−2ωt − 2ω 2C1 C2
= −2mω 2 C1 C2 = const

֒→ L = mω 2 C12 e2ωt + C22 e−2ωt = E(t)
• assume U = mgy = mgr sin ωt

m 2
L= T −U = (ṙ + r 2 ω 2) − mgr sin ωt
2
∂L
֒→ ∂t
= −mgrω cos ωt 6= 0 =⇒ Ḣ 6= 0
m 2
H(t) = (ṙ − r 2 ω 2) + mgr sin ωt
2
m
6 = E(t) = (ṙ 2 + r 2 ω 2 ) + mgr sin ωt
2
L, H, E are all different and none is conserved!
(v) particle in a time-dependent homogeneous force field (1D)
F (t) = F0 t
֒→ U(x) = −F0 xt
m 2
L = T −U = ẋ + F0 xt
2
m
H = pẋ − L = ẋ2 − F0 xt = T + U = E = E(t)
2
∂L
= F0 x = −Ḣ
∂t
49
2.5 Hamiltonian dynamics

We know that
L = L(q, q̇, t)
and X
H= pµ q̇µ − L
µ
Question: H = H(?)
To clarify on what variables H depends consider the total differential
X
dH = d pµ q̇µ − L(q, q̇, t)
µ
( )
X ∂L ∂L ∂L
= pµ dq̇µ + q̇µ dpµ − dqµ − dq̇µ − dt
µ
( )
X ∂L
use Lg.EoM : = q̇µ dpµ − ṗµ dqµ − dt
µ
∂t
= dH(q, p, t)
on the other hand:

( )
X ∂H ∂H ∂H
dH(q, p, t) = dqµ + dpµ + dt
µ
∂qµ ∂pµ ∂t
∂H ∂H
compare =⇒ ṗµ = − , q̇µ =
∂qµ ∂pµ
−→ Hamilton’s (canonical) equations of motion
Remarks:
(i) Hamilton’s EoMs ⇐⇒ Lagrange’s EoMs ⇐⇒ HP ⇐⇒ Newton II

50
show equivalence with Lagrange’s EoMs:

X
L = pµ q̇µ − H
µ
X
֒→ dL = pµ dq̇µ + q̇µ dpµ − dH(q, p, t)
µ
X ∂H ∂H ∂H
= dqµ −
pµ dq̇µ + q̇µ dpµ − dpµ − dt
µ
∂q µ ∂p µ ∂t
use X ∂H
= pµ dq̇µ + q̇µ dpµ + ṗµ dqµ − q̇µ dpµ − dt
Hamilton′ s EoMs ∂t
µ
= dL(q, q̇, t)
X ∂L ∂L ∂L
= dqµ + dq̇µ + dt
µ
∂L 
pµ = ∂ q̇µ  d ∂L ∂L
֒→ =⇒ − =0
∂L  dt ∂ q̇µ ∂qµ
ṗµ = ∂qµ
(ii) example 1: 1D oscillator

m 2
L = T −U = ẋ − ω 2 x2
2
m 2 m 2 2 m 2
H = pẋ − L = mẋẋ − ẋ + ω x = ẋ + ω 2 x2 = T + U = E
2 2 2
∂L p
p= = mẋ ⇐⇒ ẋ =
∂ ẋ m
p2 m
֒→ H(x, p) = + ω 2x2
2m 2
Note that ẋ has been eliminated to make H a function of x and p. This
is important to obtain the correct EoMs.

EoM : ṗ = − ∂H
∂x
= −mω 2 x  ẍ + ω 2x = 0
−→ ↑
p
ẋ = ∂H ẍ = mṗ

∂p
= m
∂H
=0 −→ H = E = const
∂t
51
∂H ∂L dH ∂H
(iii) =− = ⇐⇒ H = const if =0
∂t ∂t dt ∂t
(iv) system with n := 3N − k dofs:
ˆ n − ODEs of second order

Lagrange =
(2n initial conditions qµ (0), q̇µ (0), µ = 1, ..., n)
ˆ 2n − ODEs of first order
Hamilton =
(2n initial conditions qµ (0), pµ (0), µ = 1, ..., n)
(v) the transformation from L to H is called a Legendre transformation
(vi) example 2: central-force problem in polar coordinates (to be discussed

later)
m 2
L=T −U = ṙ + r 2 ϕ̇2 − U(r) = L r, ṙ, ϕ̇
2
∂L ∂L ∂U
EoM : = mṙ , = mr ϕ̇2 −
∂ ṙ ∂r ∂r
= pr
∂L ∂L
= mr 2 ϕ̇ , =0 (ϕ cyclic)
∂ ϕ̇ ∂ϕ
= pϕ
∂U
mr̈ − mr ϕ̇2 + = 0
=⇒ ∂r
r 2 ϕ̈ + 2r ṙϕ̇ = 0
pr pϕ
H = pr ṙ + pϕ ϕ̇ − L(r, ṙ, ϕ̇) with ṙ = and ϕ̇ =
m ! mr 2
pr pϕ m pr 2 2
p 2
ϕ
= pr + pϕ − + r + U(r)
m mr 2 2 m mr 2
p2r p2ϕ
= + + U(r)
2m 2mr 2
= H(r, pr , pϕ )
52
EoM:
∂H p2ϕ ∂U ∂H
ṗr = − = 3
− , ṗϕ = − = 0
∂r mr ∂r ∂ϕ
∂H pr ∂H pϕ
ṙ = = , ϕ̇ = =
∂pr m ∂pϕ mr 2
⇐⇒ Lagrange-EoM
Note:

m 2 p2ϕ
H=E= 2
ṙ + 2mr 2
+ U(r) 
 correct but problematic :
only H = H(q, p, t) and L = L(q, q̇, t)
p2ϕ 
L= T −U = m 2
ṙ + − U(r)
 yield correct EoMs!
2 2mr 2
(vii) ”phase space”: populated by
Π = (q1 ...q3N −k , p1 ...p3N −k )
−→ phase trajectorie Π(t) characterizes dynamics of the system
Hamilton−EoMs
Π(t0 ) = Π0 −→ Π(t)
state of a mechanical system ⇐⇒ phase Π
−→ all observables are functions of Π : f = f (q, p, t) = f (Π, t)
(viii) further reading: [Tay], Chap. 13

53
2.6 Extensions
2.6.1 Generalized forces and potentials
a) Formulation
3N
∂U X
so far : − = Fi + fij ≡ Fi i = 1, ..., 3N
∂xi j=1
let’s translate this to generalized coordinates:

∂U X ∂U ∂xi X ∂xi
∢ − =− = Fi
∂qµ i
∂xi ∂qµ i
∂qµ
′′
≡ Qµ generalized force components′′

= Qµ (q1 ...q3N −k , t)
standard Lagrangian: L(q, q̇, t) = T (q, q̇, t) − U(q, t)
d ∂L d ∂T ∂L ∂T ∂U
EoMs : = = = −
dt ∂ q̇µ dt ∂ q̇µ ∂qµ ∂qµ ∂qµ
d ∂T ∂T
⇐⇒ − = Qµ , µ = 1, ..., 3N − k
dt ∂ q̇µ ∂qµ
−→ alternate form of the Lagrangian equations of motion

extension: consider a generalized (velocity-dependent) potential U ∗ = U ∗ (q, q̇, t)
!
∂U ∗ d ∂U ∗
with Qµ = − −
∂qµ dt ∂ q̇µ
d ∂T ∂T d ∂U ∗ ∂U ∗
−→ − = −
dt ∂ q̇µ ∂qµ dt ∂ q̇µ ∂qµ
d ∂L ∂L
− = 0
⇐⇒ dt ∂ q̇µ ∂qµ
for L = T − U∗
54
b) Application: charged particle in an electromagnetic (EM) field
• the Lorentzian force

FL = q E + (v × B)
is obviously velocity dependent.

∗ One can construct a generalized po-
∗ i ∂U d ∂U ∗
tential U such that FL = − ∂xi − dt ∂ ẋi . To do so we need to move
from the electric and magnetic fields E and B to the corresponding
potentials:
• EM potentials
∂A
E = −∇φ −
∂t
B = ∇×A
φ, A: scalar and vector potentials
∂A
֒→ FL = q(−∇φ − + (v × (∇ × A)))
∂t
• generalized potential

U∗ = q φ − v · A
∂U ∗ d ∂U ∗
֒→ FLi = − − (i = 1, 2, 3)
∂xi dt ∂ ẋi
• Lagrangian
L = T − U∗
m 2
= v − qφ + qv · A
2
one can show (try it!) that the Lagrangian equations of motion are
nothing else but Newton’s equations for the Lorentzian force. This
shows the consistency of the argument.
• Hamiltonian
H =p·v−L
55
with the generalized momentum components

∂L
pi = = mẋi + qAi
∂ ẋi
note that the generalized momentum vector is in general not parallel
to the velocity vector (due to the occurrence of the vector potential)
m
֒→ H = (mv + qA) · v − v2 + qφ − qv · A
2
m 2
= v + qφ
2
1
= (p − qA)2 + qφ
2m
= E
obviously, H is the total energy of the system (kinetic energy plus

’electric’ potential energy; the magnetic field (vector potential) does
not exert work on the charged particle. However, in general (time-
dependent fields) the energy is not conserved.
further reading: [GPS], Chaps. 1.5, 8.1
2.6.2 Friction
Fricitional forces are also velocity dependent, but they cannot be associated
with a generalized potential. However, one can take them into account by
introducing another function into the Lagrangian equations.
assume : Fifric = −βi ẋi
X ∂xi X ∂xi
֒→ Qfric
µ = Fifric = − βi ẋi
i
∂qµ i
∂qµ
X ∂ ẋi
∂ X βi 2
= − βi ẋi = − ẋ
i
∂ q̇µ ∂ q̇µ i
2 i
definition: Rayleigh’s dissipation function

X βi
R := ẋ2i = R(q, q̇, t)
i
2
56
d ∂T ∂T
Lag. eqs. : − = Qµ = Qcon
µ + Qµ
fric
dt ∂ q̇µ ∂qµ
∂U ∂R
= − −
∂qµ ∂ q̇µ
d ∂L ∂L ∂R
⇐⇒ − + =0
dt ∂ q̇µ ∂qµ ∂ q̇µ
note the difference to the previous case of generalized potentials U ∗ : in the

present case we have a standard Lagrangian L = T − U and the frictional
force enters the equations of motion thorugh an additional term. In the
former case the velocity-dependent force is accounted for by a non-standard
potential, but the form of the Lagrangian equations of motion is unchanged.
example: free fall with air resistance (1D)
m
(1 -d im ) m 2
L = ż − mgz
2
β 2
R = ż
2
d ∂L ∂L ∂R
=⇒ = mz̈ , = −mg , = β ż
dt ∂ ż ∂z ∂ ż
Lag. eq.
=⇒ mz̈ + mg + β ż = 0
for the solution of this EoM see, e.g. [FC], Chap. 4.3
for (a bit) more on the dissipation function see [GPS], Chap. 1.5
Let’s end this chapter with mentioning two further topics, which are of
some interest, but which we do not cover (for time reasons).
2.6.3 Lagrange’s equations with undetermined multi-

pliers
This variant can be used if the constraints are given in differential form. It
enables the calculation of the forces of constraint, which is of interest from
57
an applied (engineering) point of view.

further reading: [TM], Chap. 7.5; [Tay], Chap. 7.10
2.6.4 d’Alembert’s principle

This is an independent postulate (involving strange things like virtual dis-
placements and virtual work), which can be used to derive the Lagrangian
equations of motion without invoking Hamilton’s principle.
further reading: [GPS], Chap. 1.4
Chapter 3
Applications
3.1 Central-force problem

3.1.1 Preliminary
Consider the following situation:
1. isolated system (Fext = 0)
2. f12 = −f21
= −∇1 Ū(|r1 − r2 |) = ∇2 Ū(|r1 − r2 |)
3. no constraints → 6 dofs
m1 2 m2 2
֒→ Lagrangian : L = T − Ū = v + v − Ū (|r1 − r2 |)
2 1 2 2
m1 m2
gravity : Ū (|r1 − r2 |) = −G
|r1 − r2 |
G = 6.67 × 10−11 Nm2 kg−2
solar system: Sun + eight planets (plus smaller objects)
let’s estimate the mutual forces. First, a few masses (in units of the mass of
the Earth):
Sun M⊙ = 330.000 mE
1
mercury (the smallest planet) MM e = 20 mE
jupiter (the biggest planet) MJu = 320 mE
58
59
2
force on Earth due to Sun fE⊙ M⊙ RXE
∢ = = 2
force on Earth due to X fEX mX R⊙E
Venus Mars Jupiter Moon

mX [mE ] 0.81 0,11 320 0.012
min
RXE [R⊙E ] 0.27 0.52 4.2 0.0026
fE⊙ /fEX 30000 810000 18300 180
=⇒ it is a good (first-order) approximation to view the Earth-Sun problem

as an isolated two-body problem
3.1.2 Reduction of the two-body problem to an effec-

tive one-body problem
momentum conservation : Ṗ = Fext = 0

with P = MV = (m1 + m2 )Ṙ = m1 v1 + m2 v2
m1 r1 + m2 r2
R = centre of mass (CM)
m1 + m2
The CM moves uniformly. Three coordinates are necessary to describe this
motion. Three coordinates are then left to describe the internal dynamics of
the system — the relative motion of the two bodies.
Positions wrt. CM: r′k = rk − R (k = 1, 2)
1X 1X
֒→ T = mk vk2 = mk (vk′ + V)2
2 k 2 k
1X 1X X
= mk V2 + mk vk′2 + mk vk′ · V
2 k 2 k
| k {z }
=0
1 1X
֒→ T = TCM +T ′ ; TCM = MV2 ; T ′ = mk vk′2 (holds for N ≥ 2)
2 2 k
60
For N = 2 one can further rewrite the expressions:
m1 r1 + m2 r2 (m1 + m2 )r1 − m1 r1 − m2 r2
∢ r′1 = r1 − R = r1 − =
m1 + m2 m1 + m2
m2 m2
= (r1 − r2 ) = r
m1 + m2 m1 + m2
m1 r1 + m2 r2 (m1 + m2 )r2 − m1 r1 − m2 r2
r′2 = r2 − R = r2 − =
m1 + m2 m1 + m2
m1
= − r
m1 + m2
with r = r1 − r2 = r′1 − r′2 ′′

relative vector′′
m2 m1
֒→ v1′ = v ; v2′ = − v
m1 + m2 m1 + m2
m1 ′ 2 m2 ′ 2 1 m1 m2 2 1 2
֒→ T′ = v + v = v = µv
2 1 2 2 2 m1 + m2 2
m1 m2 ′′ ′′
µ= reduced mass
m1 + m2
1 1 µM
→ T = MV2 + µv2 ; Ū = −G
2 2 r
M 2 µ µM
−→ L = Ẋ + Ẏ 2 + Ż 2 + ṙ2 + G
2 2 r
= L Ṙ, r, ṙ
= LCM (Ṙ) + Lrel (r, ṙ)
The first three Lagrangian equations are simple (and contain no news):

∂L ∂L ∂L ∂L
∂X
= ∂Y
= ∂Z
= 0 =⇒ P X = ∂ Ẋ
= M Ẋ = const  total

∂L
(X, Y, Z cyclic) PY = ∂ Ẏ = M Ẏ = const momentum
∂L 
PZ = = M Ż = const  conservation
∂ Ż
61
3.1.3 Relative motion

a) Lagrangian
choose spherical coordinates (r, θ, ϕ)
µh 2 i µM
Lrel (r, θ, ṙ, θ̇, ϕ̇) = ṙ + (r ϕ̇ sin θ)2 + (r θ̇)2 + G
2 r
generalized momenta:
∂L
pr = = µṙ
∂ ṙ
∂L
pθ = = µr 2θ̇
∂ θ̇
∂L
pϕ = = µr 2(sin2 θ)ϕ̇ = const (ϕ cyclic)
∂ ϕ̇
choose a reference system such that x(0) = y(0) = 0 (→ θ(0) = 0)
pϕ (0) = 0 = pϕ (t) =⇒ ϕ̇ = 0
z
t = 0 =⇒ 2D motion in plane ϕ = const
µ 2 µM
y ֒→ Lrel = (ṙ + r 2 θ̇2 ) + G
2 r
j
x
now θ is also cyclic and pθ = µr 2 θ̇ = const

Newtonian view on angular momentum conservation:
l = µ(r × v) = const for central forces
the conservation of the angular momentum vector implies
(i) conservation of direction → motion in a plane (⊥ l)

(ii) conservation of magnitude
l = µ|r × v| = µr 2θ̇ = pθ
62
this also has a geometrical interpretation: the area per unit time
swept by r(t) is given by
1 l
Ȧ = |r × v| = = const
2 2µ
this is nothing else but Kepler’s second law (see below).
Let’s work out the equation of motion:
∂Lrel ∂Lrel µM
= µṙ , = µr θ̇2 − G
∂ ṙ ∂r r
EoM µM
=⇒ µr̈ = µr θ̇2 − G
r2
2
l µM
⇐⇒ µr̈ = − G (3.1)
µr 3 r2
b) Hamiltonian, energy, and a qualitative discussion of the Kepler orbits
Hrel = pr ṙ + lθ̇ − Lrel

µ µM
= .... = (ṙ 2 + ṙ 2 θ̇2 ) − G
2 r
p2r l2 µM
= + −G = Erel = const (energy conservation)
2µ 2µr 2 r
∂L
(since − = Ḣ = 0)
∂t
= Trad + Trot + Ugrav
= Trad + Uef f (r)
l2 µM
with Uef f (r) = − G
2µr 2 r
Trad = Erel − Uef f (r) ≥ 0
⇐⇒ Erel ≥ Uef f (r)

63
E
p Q = 0
2
1 /r
U e ff (r)
E 3
riR r a
r
E 2
E 1
-1 /r
Figure 3.1: Effective potential of the Kepler problem
From now on use E ≡ Erel (and l ≡ pθ )
(i) E = E1 = Uef f (R) −→ Trad = 0

l
−→ circular orbit with angular velocity θ̇ = µR2
= const
(ii) E1 < E = E2 < 0
−→ finite (bounded) orbit in [ri , ra ]
(ri , ra : turning points of radial motion, Trad (ri ) = Trad (ra ) = 0)
with changing angular velocity θ̇ = µrl 2
(iii) E = E3 ≥ 0
t→∞
infinite (unbounded) orbit in [r3 , ∞) , (in general r −→ ∞)
64
∢ l=0:
E
E 2
r 1
r
E 1
- 1 /r
(i) E = E1 < 0 finite (bounded) orbit in [0, r1 ] (cf. free fall)

t→∞
(ii) E = E2 > 0 infinite (unbounded) orbit (in general r −→ ∞)
Remarks:
(i) Analysis refers to ’quasiparticle’ of mass µ and relative motion.

For the Earth-Sun system we have
M⊙ mE
µ= ≈ mE ; r ≈ rE ; M ≈ M⊙ ; R ≈ r⊙
M⊙ + mE
and an orbit of type (ii) (for the case l 6= 0)

(ii) Similar analysis is possible for other (central) potentials and is
also useful for quantum-mechanical problems
(iii) Quantitative analysis requires solution of EoM
(iv) Further reading: e.g. [TM], Chap. 8.5, 8.6
65
c) Quantitative analysis
µ 2 l2
starting point : E= ṙ + + U(r)
2 2µr 2
s
dr 2 l2
⇐⇒ ṙ = =± E − U(r) −
dt µ 2µr 2
Z t Z r
′ dr ′
−→ t − t0 = dt = ± r
t0 r0 2 l2
µ
E − U(r ′ ) − 2µr ′2
−→ inversion yields r(t)

Z t
l l dt′
obtain θ(t) from θ̇ = 2 −→ θ(t) − θ0 =
µr µ t0 r 2 (t′ )
the problem with this procedure is that the integral cannot be calcu-
lated in closed analytical form (for the gravitational potential).
alternative consideration: analyze orbits r(θ)
s
dr dr dθ l dr 2 l2
ṙ = = = 2 =± E − U(r) −
dt dθ dt µr dθ µ 2µr 2
Z θ Z r
′ l dr ′ 1
֒→ θ(r) − θ0 = dθ = ± r
θ0 µ r0 r ′2 2

l2

µ
E − U(r ′ ) − 2µr ′2
inversion −→ r(θ)
for U(r) = − kr (k = GµM) this integral can be solved by using the
substitution u = 1/r and the indefinite integral
Z
du 1 b + 2cu
√ = √ arccos − √
a + bu + cu2 −c b2 − 4ac
One obtains the Kepler orbits: conic sections (in polar coordinates)
with one focus at the origin (look up ’conic sections’ on wikipedia!)
66
1 1
= 1 + ε cos(θ − θ′ )
r α
l2
α = >0
µk
s
2El2 ′′
ε = 1+ 2
≥0 eccentricity′′
µk
Classification of Kepler orbits

2

E = − µk
2l2
≡ Ecirc 


ε=0 circle 



l2 
r=α= µk
= const 







planets
Ecirc < E < 0 
 (+ comets)


0<ε<1 ellipse 


α α 
rmax = 1−ε
, rmin = 1+ε







′′ 
aphelion′′ ′′
perihelion′′

ε=1 E=0 parabola 
comets

ε>1 E>0 hyperbola
d) Kepler’s laws (1609, 1619)
I. ”Planets move in elliptical orbits about the Sun with the Sun at
one focus.”
II. ”The area per unit time swept out by the radius vector from the
Sun to a planet is constant.”
III. ”The square of a planet’s period is proportional to the cube of the
major axis of the planet’s orbit.”
67
e) Further remarks and references
(i) Instead of using energy conservation (the first integral of the mo-
tion) one can obtain the Kepler orbits directly from the EoM:
Let’s first rewrite r̈:
dr dθ l dr
ṙ = = 2
dθ dt µr dθ

d l dr d l dr l2 d 1 dr
֒→ r̈ = = θ̇ = 2 2
dt µr 2 dθ dθ µr 2 dθ µ r dθ r 2 dθ

r=1/u l2 2 d 2 dr du l 2 2 d2 u
= u u = − u .
µ2 dθ du dθ µ2 dθ2
Plug this result into the EoM (3.1) to obtain
d2 u(θ) µk
2
+ u(θ) = 2 = α−1 .
dθ l
This is a relatively simple inhomogeneous differential equation
that can be solved without great difficulty (try it!). It yields —
of course — the same results for the Kepler orbits as spelled out
on the previous page ([Tay], Chap. 8.6).
(ii) An analysis of the motion as a function of time can be found in
[GPS], Chap. 3.8.
(iii) It turns out that the Kepler problem has another constant of mo-
tion: the Laplace-Runge-Lenz (LRL) vector A:
r
A = p × l − µk
r
where p and l are the linear and the angular momentum vectors
and k = GµM.
dA l̇=0 d r
proof : = ṗ × l − µk
dt dt r
µk r ṙ − ṙr
= − 3 (r × (r × ṙ)) − µk
r r2
r(r · ṙ) − r 2 ṙ + r 2 ṙ − ṙrr
= −µk =0
r3
68
Note that in the second step Newton’s EoM in the form ṗ = − rk2 rr
was used. It looks now as if we had too many conserved quanti-
ties: l, A, E involve seven components, which cannot all be inde-
pendent. It turns out that only five of them are; see [GPS], Chap.
3.9 (this chapter also includes a discussion of further properties of
the LRL vector).
(iv) Relative motion → two-body motions
Keep in mind that the Kepler orbits are associated with the rel-
ative motion of the two-body problem. If m1 ≫ m2 the relative
motion is a good approximation of the motion of m2 , while m1 is
always close to the centre of mass (CM). If the two masses are not
that different one has to consider the coordinate transformations
discussed in Sec. 3.1.2. If the orbits are bounded, one obtains
similar ellipses of both bodies about the CM.
(v) Deviations from the ideal elliptical orbits in the solar system are
observable. They are (with decreasing importance) due to
∗ gravitational forces among the planets
∗ effects associated with Einstein’s theory of general relativity
∗ solar oblateness (deviations of the mass distribution of the
sun from a sphere)
(vi) Further reading: [TM], Chap. 8; [Tay], Chap. 8 (both book
chapters contain proofs of Kepler’s third law).
69
3.2 Dynamics of rigid bodies

3.2.1 Preparations
definition: ”rigid body”
A rigid body is an aggregate of particles (mass points), whose relative dis-
tances are constrained to remain (absolutely) fixed.
∢ rigid body consisting of N mass points at positions (r1 ...rN ):
constraints:

N N (N −1)
|ri − rj | = cij = const ∀ ij −→ = 2
constraints
2

N N
N 3N −
2 2
2 1 5
3 3 6
4 6 6
5 10 5
6 15 3
7 21 0
8 28 -4

N
−→ constraints cannot be independent!
2
In fact, one can show that for N > 2 one has (in general)
3N − 6 independent constraints =⇒ 6 dofs
Lagrangian:
N N
1X X
L= T −U = mi vi2 − U(ri )
2 i i=1
Note that internal forces—if present—are neutralized by the constraints and

do not contribute.
70
Chasles’ theorem: the general motion of a rigid body is composed of a trans-

lation of a point and a rotation about that point.
The natural choice for the 3 coordinates of P

the translational motion are the
(Cartesian) coordinates of the CM: R = M1 i mi ri
Let’s introduce three reference frames:
So : fixed (inertial) system
Sf : non-rotating system with fixed axes and

S ' with CM as origin
R
S
Sb : rotating body system also with CM as
origin
coordinates and velocity of i-th mass point:
coordinates and velocity of i-th mass point:

ri = R + ri
S Sf,b
o

vi = Ṙ + ω × ri (for details see [TM], Chap. 10.2)
So Sf,b
3.2.2 Kinetic energy and inertia tensor
1X 1 X 2
2
T = mi vi = Ṙ + (ω × ri )
2 i So 2 i Sf,b
1 X n o
mi Ṙ2 + 2Ṙ · (ω × ri ) + (ω × ri )2

=
2 i Sf,b Sf,b
1 X 1X
M Ṙ2 + Ṙ · ω × mi (ω × ri )2

= mi ri +
2 i
Sf,b 2 i Sf,b
| {z }
k = 0 in k
= Ttrans CM system + Trot
71
N
1X 2 2
Trot = mi ω ri − (ω · ri )2 write it out in body system:
2 i=1 Sb
N 3 3 3
1 X nX 2 2 X (j)
X
(k)
o
= mi ωj ri − ωj xi ωk xi
2 i=1 j=1 j=1 k=1
N 3 n o
1X X (j) (k)
= mi r2i δjk − xi xi ωj ωk
2 i=1 j,k=1
3
1 X
= Ijk ωj ωk
2 j,k=1
N
X n o
2 (j) (k)
with Ijk = mi δjk ri − xi xi inertia tensor (inertia matrix)
i=1
1 T
summary: Trot = ω I ω
2
3.2.3 Structure and properties of the inertia tensor

a) Continuous mass distributions
mi = ∆mi = ρ(ri )∆Vi −→ ρ(r) d3 r = dm

N
X Z Z
M= mi −→ dm = ρ(r) d3 r
i=1 V V
Z n o
Ijk = ρ(r) δjk r2 − xj xk d3 r
V
R 
explicitly : I11 = V
(x22 + x23 )ρ(r) d3 r 



R 
′′
I22 = V
(x21 + x23 )ρ(r) d3 r moments of inertia′′


R 


I33 = V
(x21 + x22 )ρ(r) d3 r
72
R 
I12 = I21 = − V
x1 x2 ρ(r) d3 r 



R 
′′
I13 = I31 = − V
x1 x3 ρ(r) d3 r products of inertia′′


R 


I23 = I32 = − V
x2 x3 ρ(r) d3 r
note that I is symmetric (Ijk = Ikj )
b) Examples
(i) Homogeneous cube (I)
x 3
 M
 ρ0 = a3
rǫV
ρ(r) =

x 1
0 else
a
Z Z a Z a Z a
M M 2 2 2
I11 = 3 (x22 + x23 ) 3
dr = (x22 + x23 ) dx1 dx2 dx3
a V a3 − a2 − a2 − a2
Z a Z a
M 2 2
= dx2 dx3 (x22 + x23 )
a2 − a2 − a2
Z a Z a
M 2
2
2
2

= a x2 dx2 + a x3 dx3
a2 − a2 − a2
M x32 a2 x33 a2
= +
a 3 − a2 2 − a2
1
= Ma2 = I22 = I33
6
Z
M
I12 = − 3 x1 x2 dx1 dx2 dx3
a V
Z a Z a Z a
M 2 2 2
= − 3 x1 dx1 x2 dx2 dx3
a − a2 − a2 − a2
M 2 a2 2 a2
= − 2 x1 a x2 a
4a −2 −2
= 0 = I13 = I23
73
(ii) Homogeneous cube (II)

Z a Z a Z a
M
I11 = x22 + x23 dx1 dx2 dx2
a30 0 0
!
3 a
M x2 x33 a
= +
a 3 0 2 0
a
2
= Ma2 = I22 = I33
3
Z a Z a Z a
M
I12 = − 3 x1 dx1 x2 dx2 dx3
a 0 0 0
! !
M 1 2 a 1 2 a
= − 2 x x
a 2 10 2 20
Ma2
= − = I13 = I23
4
c) Principal axes of inertia

Theorem: For any rigid body and any choice for the origin of the body
system there exists a set of perpendicular (”principal”) axes such that
Ijk = Ik δjk
”proof”: I is a (real) symmetric matrix ⇒ can be diagonalized
diagonalization of I ⇐⇒ ∃ orthogonal matrix
D−1 = D T ⇐⇒ (D−1 )kj = Dkj

−1
= Djk ,
 
I1 0 0 X
T T
such that D I D =  0 I2 0  ⇐⇒ Dnj Ijk Dkm = δmn In
0 0 I3 jk
74
 
I1 0 0 X X
⇐⇒ I D = D  0 I2 0  ⇐⇒ Ijk Dkm = Djk Ik δkm
0 0 I3 k k
= Djm Im
X
= δjk Dkm Im
k
X
⇐⇒ Ijk − Im δjk Dkm = 0
k
    
I11 − Im I12 I13 D1m 0
⇐⇒  I21 I22 − Im I22   D2m  =  0 , m = 1, 2, 3
I31 I32 I33 − Im D3m 0
condition for a nontrivial solution:
det(Ijk − Im δjk ) = 0
this (cubic) equation is called secular or characteristic equation.
Diagonalization procedure:
(i) Solve secular equation → obtain ”eigenvalues” I1 , I2 , I3
(ii) Find D by inserting eigenvalues into the system of homogeneous
equations above.
Remarks:
(i) Ik :”principal moments of inertia”; the axes of the corresponding
body system are the principal axes
(ii) Transformation matrix D characterizes a rotation in R3
(iii) The principal
R moments of inertia are consistent with the equation
Ik = V ρ(r)(r − x2k ) d3 r
2
(iv) Change of I when moving body system from CM to other origin:

Steiner’s parallel-axis theorem ([TM], Chap. 11.6)
(v) Kinetic energy
1X
if Ijk = Ik δjk −→ Trot = Ik ωk2
2 k
75
3.2.4 Generalized coordinates and the Lagrangian

Lagrangian:
L = T − U = Ttrans + Trot − U
M 2 1X
= Ẋ + Ẏ 2 + Ż 2 + Ik ωk2 − U
2 2 k
This form of the Lagrangian reinforces what is stated by Chasles’ theorem:

the translational motion of the rigid body can be described by the three CM
coordinates, while we need three additional coordinates to characterize the
rotational motion. Recall that we introduced a fixed inertial reference frame
So , a fixed (i.e. non-rotating) reference system Sf with the CM as origin and a
rotating body system Sb , the origin of which is also the CM. The translational
motion of the body (the CM) is described by the transformation from So to
Sf , while rotations correspond to the transformation from Sf to Sb :
translation rotation
So −→ Sf −→ Sb
Theorem: any rotation can be described by a series of three rotations through

the Eulerian angles (α, β, γ) about designated coordinate axes.
Definition of the Euler angles: (note that other conventions are also in use)
1 . D re h u n g (a ) 2 . D re h u n g (b ) 3 . D re h u n g (g )
3 '= 3 3 '
3 '' 3 '''= 3 '' 2 '''
2 '' 2 ''
2 ' b 2 '
2
g
a
1 1 '= 1 '' 1 ''
1 ' (K n o te n lin ie ) 1 '''
Figure 3.2: Definition of the Eulerian angles.
Remarks:
• rotations are characterized by orthogonal 3 × 3 matrices (det D = 1)

76
• rotation matrices form a nonabelian group, i.e., D1 D2 6= D 2 D 1
• If we consider a vector r in 3D space, its components in the systems

Sf and Sb are related by
r Sb = D γ D β
D α
r Sf ≡ D r Sf ,
where D α , D β , D γ are the rotation matrices that correspond to the

rotations through the Euler angles. They are spelled out in Appendix
A.5.
Using the rotation matrices Dα , D β , D γ one can determine the components

of ω in the body system (see Appendix A.5):
ω1 = α̇ sin β sin γ + β̇ cos γ

ω2 = α̇ sin β cos γ − β̇ sin γ
ω3 = α̇ cos β + γ̇
This has to be inserted in the rotational kinetic energy expression in the

Lagrangian. One obtains:
M
L = Ẋ 2 + Ẏ 2 + Ż 2
2
1 n 2 2
+ I1 α̇ sin β sin γ + β cos γ + I2 α̇ sin β cos γ − β̇ sin γ
2
2 o
+ I3 α̇ cos β + γ̇ − U X, Y, Z, αβγ

= L XY Z, αβγ; Ẋ Ẏ Ż, α̇β̇ γ̇
3.2.5 Equations of motion
d ∂L ∂L ∂U
translational motion : = ⇐⇒ M Ẍ = −
dt ∂ Ẋ ∂X ∂X
d ∂L ∂L ∂U
= ⇐⇒ M Ÿ = −
dt ∂ Ẏ ∂Y ∂Y
d ∂L ∂L ∂U
= ⇐⇒ M Z̈ = −
dt ∂ Ż ∂Z ∂Z
77
example: rigid body in uniform gravitational field

X X

U= mi gzi = g mi zi = MgZ = U(Z)
So So
i i
The EoMs for the translational motion are simple:
Ẍ = Ÿ = 0, Z̈ = −g
If the body rotates (freely) about the CM (see Sec. 3.2.7) one has
∂U ∂U ∂U
= = =0
∂α ∂β ∂γ
֒→ rotational motion is not influenced by U, but is far from being trivial
(this situation is called motion of a ’free top’).
In the general case, one obtains a set of second-order nonlinear, coupled
differential equations for the Euler angles α, β, γ for the description of the
rotational motion (see Appendix A.5). These EoMs can be summarized as:

I1 ω̇1 − I2 − I3 ω2 ω3 = N1

I2 ω̇2 − I3 − I1 ω1 ω3 = N2

I3 ω̇3 − I1 − I2 ω1 ω2 = N3
Euler’s equations
Ni : components of torque in the body system. For the example with U =

MgZ all torque components Ni vanish.
3.2.6 Angular momentum

For details see [TM], Chap. 11.4 and 11.9
recap: angular momentum of a system of particles (cf. Sec. 1.2.2)
X
L = (ri × pi )
i
X
L̇ = (ri × Fi ) = N
i
78
One can show:

L = LCM + L
So Sf
X
= M R×V + mi ri × vi
Sf
i
X
= M R×V + mi ri × (ω × ri )
Sf
i
The second term can be rewritten by decomposing the vectors in the body
system and using the definition of the inertia tensor. One obtains

L = I ω
Sf
Note that ths is the angular momentum measured by an observer in Sf who

does not rotate with the rigid body (otherwise he would find L = 0). But
the components of this representation of L refer to the body system Sb , i.e.
X

Lj = Ijk ωk
Sb Sb
k
Remarks:
(i) Lj = Ij ωj in principal axes body system
(ii) in general L ∦ ω → motion of rigid bodies is complicated!
P P
(iii) Trot = 21 jk Ijk ωk ωj = 12 j Lj ωj = 21 L · ω
1X 1 X L2j
in principal axes system : Trot = Lj ωj =
2 j 2 j Ij
Note that this is the rotational kinetic energy measured by a non-

rotating observer (otherwise she would measure Trot = 0).
If one rewrites L̇ = N in body coordinates one reobtains Euler’s equations

L̇1 = I1 ω̇1 − I2 − I3 ω2 ω3 = N1
Sb

L̇2 = I2 ω̇2 − I3 − I1 ω3 ω1 = N2
Sb

L̇3 = I3 ω̇3 − I1 − I2 ω1 ω2 = N3
Sb
79
3.2.7 Applications: symmetric tops

top: a rigid body which is free to turn about a fixed point. The top is
torque-free if the fixed point is the CM and U = MgZ (see above).
a) Free spherical top: I1 = I2 = I3 ≡ I
Euler:
֒→ ω̇k = 0 −→ ωk (t) = ωk (0) = const
−→ body rotates uniformly about fixed axis
b) Free symmetric top: I1 = I2 = I, I3 6= I

Two cases can be distinguished:
’prolate’ top: I > I3

3
3 ’oblate’ top: I < I3

Euler : I ω̇1 + I − I3 ω2 ω3 = 0

I ω̇2 − I3 − I ω3 ω1 = 0
I3 ω̇3 = 0 =⇒ ω3 (t) = ω3 = const
I3 −I
define: Ω := I
ω3 (= const)

ω̇1 + Ωω2 = 0 (1)

֒→

ω̇2 − Ωω1 = 0 (2)
80
decouple:
1
(1′ ) : ω̈1 + Ωω̇2 = 0 ⇐⇒ ω̇2 = − ω̈1
Ω
1
(2′ ) : − ω̈1 − Ωω̇1 = 0 ⇐⇒ ω̈1 + Ω2 ω1 = 0
Ω
֒→ general solution : ω1 (t) = C1 cos Ωt + C2 sin Ωt
′′d
(1 ) : C1 cos Ωt + C2 sin Ωt = −Ωω2
dt
⇐⇒ ω2 (t) = C1 sin Ωt − C2 cos Ωt
∢ initial conditions: ω1 (0) = A , ω2 (0) = 0
ω1 (t) = A cos Ωt
ω2 (t) = A sin Ωt (components of ω wrt Sb )
ω3 (t) = ω3 = const
q
|ω| = A2 + ω32 = const
Visualization:
3
W < 0 W > 0
”free precession”
   
I1 ω 1 IA cos Ωt
w (t) L =  I2 ω2  =  IA sin Ωt 
I3 ω 3 I3 ω 3
L3 I3 ω 3 ω3 ω3
prolate : p < =p
L21 + L22 IA A ω12 + ω22
I3 ω 3 ω3
oblate : >
IA A
81
p ro la t o b la t
3 3
w L
L w
ω, L, e3 are in a plane at all times (show e3 · (ω × L) = 0)

1 1
Trot = L · ω = IA2 + I3 ω32 = const
2 2
To better understand the motion of the torque-free top one has to look
at it from the viewpoint of the inertial reference frame Sf . This can be
done in two ways:
(i) A qualitative analysis starts from noting that the angular mo-
mentum in Sf is constant (since N = 0). Our solution seems to
contradict this, but the time dependence of the components of L
is a consequence of decomposing it in Sb . When viewed from Sf ,
L is fixed, while ω and the 3-axis of the rigid body are precess-
ing. The motion of the rigid body can be visualized by considering
two cones, the so-called body and space-fixed cones, see, e.g. [TM],
Chap. 11.10 for details and illustrations.
(ii) The quantitative analysis involves the determination of the Eule-
rian angles. This can be done by comparing the explicit solution
obtained for ω in Sb with the general expression derived in Ap-
pendix A.5:
   
A cos Ωt α̇ sin β sin γ + β̇ cos γ
ω =  A sin Ωt  =  α̇ sin β cos γ − β̇ sin γ 
ω3 α̇ cos β + γ̇
82
We know from the previous analysis that the angle between L and
the 3-axis of Sb is constant (steady precession). If we choose the
3-axis of Sf to point along L this angle can be identified with the
Euler angle β = β0 =const.
 
A cos Ωt = α̇ sin β0 sin γ
֒→  A sin Ωt = α̇ sin β0 cos γ 
ω3 = α̇ cos β0 + γ̇
A
֒→ A2 = α̇2 sin2 β0 ⇒ α̇ = ± = const
sin β0
We obtain:  
α(t) = ± sinAβ0 t + α0

 
 β0 = tan−1 IIA 3 ω3

γ(t) = −Ωt ± π2
Interpretation: α(t) characterizes the steady precession of the
symmetry axis of the rigid body about the 3-axis of Sf , β0 the
constant inclination of the symmetry axis, and γ(t) the rotation
of the rigid body about its symmetry axis.
c) The heavy symmetric top: N 6= 0
(i) Potential energy

U = MgZ = MgS cos β = U(β)
Sf
(ii) Kinetic energy

1 1
Trot = [I(ω12+ω22 )+I3 ω32 ] = . . . = [I(α̇2 sin2 β+β̇ 2 )+I3 (α̇ cos β+γ̇)2 ]
2 2
83
(iii) Lagrangian
L = Trot − U = L(β, α̇, β̇, γ̇)
֒→ α, γ cyclic and
∂L
pα = = I α̇ sin2 β + I3 (α̇ cos2 β + γ̇ cos β) = const
∂ α̇
∂L
pγ = = I3 (α̇ cos β + γ̇) = I3 ω3 = const
∂ γ̇
֒→ pα = I α̇ sin2 β + pγ cos β
(iv) Analysis
Instead of working out the EoMs let’s look at the (conserved)
energy of the top and do something similar to what we have done
for the central-force problem in Sec. 3.1:
1
E = H = Trot +U = [I(α̇2 sin2 β+β̇ 2)+I3 (α̇ cos β+γ̇)2 ]+MgS cos β
2
We can eliminate α̇ and γ̇ by using the constant momentum com-
ponents and obtain
1 2 1 (pα − pγ cos β)2 1 p2γ
E = I β̇ + + + MgS cos β
2 2 I sin2 β 2 I3
1 p2γ 1
⇐⇒ E − ≡ E ′ = I β̇ 2 + Ueff (β)
2 I3 2
1 (pα − pγ cos β)2
with Ueff (β) = + MgS cos β
2 I sin2 β
A formal solution is obtained from rewriting the equation for E ′ as a
differential equation for β:
r Z Z
2 ′ dβ
β̇ = ± (E − Ueff (β)) ⇒ dt = ± q
I 2
(E ′ − U (β))
I eff
Inversion yields β(t), which in turn can be used to obtain α(t) and γ(t)
from integrating
pα − pγ cos β(t)
α̇ =
I sin2 β(t)
pγ
γ̇ = − α̇ cos β(t)
I3
84
Unfortunately, these steps cannot be carried out analytically. A more

qualitative discussion can be based on inspecting Ueff (β), which is a
u-shaped function:
Observations
(i) Shallow minimum at β0 : If E ′ = E0′ = Ueff (β0 ) we have steady

precession at a fixed inclination angle. This can be analyzed in
dUeff
more detail by inspecting dβ = 0.
β0
′
(ii) For E = E1′> Ueff (β0 ) the inclination angle varies between β1 , β2
with 0 < β1 ≤ β ≤ β2 < π. This is called nutation. Depending
on whether or not α̇ = pαI−psinγ 2cos
β
β
changes sign in the allowed
β-interval three different types of motion can be distinguished
(monotonic precession, looping motion, cusplike motion).
For details and illustrations see [TM], Chap. 11.11.

85
3.3 Coupled oscillations

3.3.1 An illustrative example: two coupled oscillators
k 1 k 1 2 k 2
m 1 m 2
q 1 q 2
m1 2 m2 2
• T = q̇ + q̇
2 1 2 2
k1 2 k2 2
• Uext = q + q
2 1 2 2
k12 2
• Ū = q1 − q2
2
֒→ L = T − Uext − Ū
m1 2 m2 2 1 2
= q̇1 + q̇2 − k1 q1 + k2 q22 + k12 (q12 − 2q1 q2 + q22 )
2 2 2
Obtain equations of motion:
d ∂L ∂L
= m1 q̈1 , = −k1 q1 − k12 (q1 − q2 )
dt ∂ q̇1 ∂ q̇1
d ∂L ∂L
= m2 q̈2 , = −k2 q2 + k12 (q1 − q2 )
dt ∂ q̇2 ∂ q̇2
k1 k12
֒→ q̈1 + q1 + (q1 − q2 ) = 0
m1 m1
k2 k12
q̈2 + q2 + (q2 − q1 ) = 0
m2 m2
86
• special case #1: k12 = 0 (no coupling)

r
kλ
qλ (t) = Aλ cos(ωλ t + δλ ), λ = 1, 2; ωλ =
mλ
• special case #2: m1 = m2 ≡ m, k1 = k2 ≡ k

k + k12 k12
֒→ q̈1 + q1 − q2 = 0
m m
k + k12 k12
q̈2 + q2 − q1 = 0
m m
The form of the (internal) potential energy suggests the following trans-
formation to another set of coordinates Qλ :
1
Q1 = (q1 + q2 )
2
1
Q2 = (q1 − q2 )
2
Inserting the inverse transformation into the equations of motion and
adding and subtracting them yields the uncoupled equations and solu-
tions
r
k k
Q̈1 + Q1 = 0, Q1 (t) = A1 cos(Ω1 t + δ1 ), Ω1 =
m rm
k + 2k12 k + 2k12
Q̈2 + Q2 = 0, Q2 (t) = A2 cos(Ω2 t + δ2 ), Ω2 =
m m
so that we obtain
q1 (t) = A1 cos(Ω1 t + δ1 ) + A2 cos(Ω2 t + δ2 )

q2 (t) = A1 cos(Ω1 t + δ1 ) − A2 cos(Ω2 t + δ2 )
as general solutions.
Discussion
(i) Q1 , Q2 : ”normal coordinates”, Ω1 , Ω2 : ”normal frequencies”

87
(ii) pick initial conditions: q1 (0) = q2 (0) = A, q̇1 (0) = q̇2 (0) = 0
A A
q1 (0) = A = A1 cos δ1 + A2 cos δ2

q2 (0) = A = A1 cos δ1 − A2 cos δ2
q̇1 (0) = 0 = −A1 Ω1 sin δ1 − A2 Ω2 sin δ2
q̇2 (0) = 0 = −A1 Ω1 sin δ1 + A2 Ω2 sin δ2
֒→ A1 = A , A2 = 0 , δ1 = 0
symmetrical
֒→ q1 (t) = A cos Ω1 t = q2 (t) →
normal mode
(iii) pick initial conditions: q1 (0) = −q2 (0) = A, q̇1 (0) = q̇2 (0) = 0
antisymmetrical
֒→ q1 (t) = A cos Ω2 t = −q2 (t) →
normal mode
(iv) Note that Ω2 > Ω1 reflects the fact that more energy is stored in
the antisymmetrical than in the symmetrical mode.
88
(v) Consider a variant: hold m2 fixed, i.e., introduce the constraint

q2 = 0.
EoM k + k12
֒→ q̈1 + q1 = 0
m
general solution:
r
k + k12
q1 (t) = A1 cos(ω0 t + δ1 ), ω0 =
m
Similarly, one finds for q1 = 0:
q2 (t) = A2 cos(ω0 t + δ2 )
Observation: the ’common frequency’ ω0 splits into two frequen-

cies according to Ω1 < ω0 < Ω2 if coupling is turned on.
(vi) pick initial conditions: q1 (0) = A, q2 (0) = q̇1 (0) = q̇2 (0) = 0
solution:
A (Ω1 + Ω2 )t (Ω1 − Ω2 )t
q1 (t) = cos Ω1 t + cos Ω2 t = A cos cos
2 2 2
A (Ω1 + Ω2 )t (Ω1 − Ω2 )t
q2 (t) = cos Ω1 t − cos Ω2 t = −A sin sin
2 2 2
→ beats
89
0,8
0,4
-0,4
-0,8
0 5 10 15 20 25
t
Figure 3.3: Beats for Ω1 = 1, Ω2 = 1.5; black curve: q1 (t), red curve: q2 (t)
special case: k12 ≪ k (weak coupling)


q1 (t) ≈ A cos ∆t cos ω0 t 
Ω1 k12
∆= ≪ Ω1 , ω0
 2 k
q2 (t) ≈ A sin ∆t sin ω0 t
0,5
-0,5
-1
0 50 100 150 200 250
t
Figure 3.4: Beats for Ω1 = 1, Ω2 = 1.05; black curve: q1 (t), red curce: q2 (t)
90
3.3.2 Lagrangian and equations of motion for coupled

oscillations: general case
Let us characterize the systems of interest by the following properties:
(i) At most holonomic constraints
(ii) Time-independent transformations xi ←→ qµ
(iii) Conservative forces
(iv) System is in vicinity of stable equilibrium
֒→ L = T − Uext − Ū ≡ T − U
H = T + U = E = const
• Taylor-expand U(q) about equilibrium configuration q0 = 0
X ∂U 1 X ∂ 2 U
U(q) ≈ U(0) + qµ + qµ qν + ...
| {z } ∂q µ
µ | {z }
0 2 µν
∂q µ ∂q ν 0
=0 =0
1X ∂ 2 U
≈ Uµν qµ qν , Uµν := = Uνµ
2 µν ∂qµ ∂qν 0
X mi X ∂xi
• T = ẋ2i with ẋi = q̇µ
i
2 µ
∂qµ
1X
֒→ T = Tµν (q)q̇µ q̇ν
2 µν
X ∂xi ∂xi
with Tµν (q) = mi = Tνµ (q)
i
∂qµ ∂qν
X ∂Tµσ
Taylor-expand: Tµν (q) ≈ Tµν (0)+ qλ ≈ Tµν (0) ≡ Tµν = Tνµ
λ
∂qλ 0
(0th order for Tµν (q) is consistent with second order for U(q))
1 X 1
֒→ L = Tµν q̇µ q̇ν − Uµν qµ qν = q̇T T q̇ − qT U q
2 µν 2
note that the matrices T and U are symmetrical

91
Lagrangian EoMs:
∂L ∂T 1X
= = Tµν δµλ q̇ν + δνλ q̇µ
∂ q̇λ ∂ q̇λ 2 µν
1X 1X
= Tλν q̇ν + Tµλ q̇µ
2 ν 2 µ
X
= Tλµ q̇µ
µ
∂L ∂U X
− = = ... = Uλµ qµ
∂qλ ∂qλ µ
X
=⇒ Tλµ q̈µ + Uλµ qµ = 0 , λ = 1, ..., n for n dofs
µ
T q̈ + U q = 0
explicitly:
       
T11 .... T1n q̈1 U11 .... U1n q1 0
 ..   ..   ..   ..   .. 
 .  .  +  .  .  =  . 
Tn1 .... Tnn q̈n Un1 .... Unn qn 0
3.3.3 Solution of the EoMs

a) Simple situation: uncoupled oscillators
characterized by Tλµ = δλµ Tλ , Uλµ = δλµ Uλ
EoMs Uλ
֒→ q̈λ + qλ = 0 , λ = 1, ..., n
Tλ
r
Uλ
solution : qλ (t) = Aλ cos(ωλ t + δλ ) , ωλ =
Tλ
92
b) Intermediate situation: diagonal T -matrix
characterized by Tλµ = Tλ δλµ

EoMs X
֒→ Tλ q̈λ + Uλµ qµ = 0
µ
√
transformation (i): q̃λ = Tλ qλ
EoMs X
֒→ q̃¨λ + Sλµ q̃µ = 0
µ
Uλµ ′′
with Sλµ = p dynamical matrix′′
Tλ Tµ
S is symmetric, positive semi-definite matrix, i.e.,
Sλµ = Sµλ , ξ T S ξ ≥ 0 ∀ ξ = {ξ1 ...ξn } =

6 {0...0}
=⇒ S can be diagonalized; eigenvalues are real (since S is symmetric)

and non-negative (since S is positive semi-definite)
That is, there is an orthogonal transformation matrix V (V −1 = V T )

such that:
 
Ω21 0
V TS V = 
 .. 
=Ω ,
2
(Ω2i ≥ 0)
.
0 Ω2n
This motivates transformation (ii):
Q = V T q̃ ⇐⇒ q̃ = V Q
EoMs
֒→ V Q̈ + S V Q=0
V T V Q̈ + V TS V Q = 0
⇐⇒ Q̈ + Ω2 Q = 0
93
⇐⇒ Q̈λ + Ω2λ Qλ = 0 λ = 1, ..., n
solutions : Qλ (t) = Aλ cos(Ωλ t + δλ )

↑ ↑
normal coordinates normal frequencies
Recipe:
(i) Set up and diagonalize S: solve secular equation
det(Sµν − Ω2Λ δµν ) = 0
−→ obtain
0 ≤ Ω21 < Ω22 < ... < Ω2n
(ii) Insert Ω2λ into matrix equations to obtain V :
X
(Sµν − Ω2λ δµν )Vνλ = 0
ν
(iii) Obtain
1 X Vµλ
qµ (t) = p q̃µ (t) = p Qλ (t)
Tµ λ
Tµ
X Vµλ
= p Aλ cos(Ωλ t + δλ ) , µ = 1, ..., n
λ
Tµ
Applications
example (i): two coupled oscillators again (cf. Sec. 3.3.1)

The Lagrangian can be written as:
( )
1 m1 0 q̇1 k1 + k12 −k12 q1
֒→ L = (q̇1 , q̇2 ) −(q1 , q2 )
2 0 m2 q̇2 −k12 k2 + k12 q2
94
diagonalize Sµν = √Uµν

Tµ Tν

k1 +k12

m1
− Ω2 − √mk12
1 m2

−→ =0

− √mk12
1 m2
k2 +k12
m2
− Ω2
k + k
1 12
k + k
2 12
k2
⇐⇒ − Ω2 − Ω2 − 12 = 0
m1 m2 m1 m2
" # 21
2
1 k1 + k12 k2 + k12 1 k1 + k12 k2 + k12 2 4k12
=⇒ Ω21,2 =+ + ± − +
2 m1 m2 2 m1 m2 m1 m2
for the special case m1 = m2 ≡ m , k1 = k2 ≡ k one reobtains

r r
k k + 2k12
Ω1 = , Ω2 =
m m
example (2): longitudinal vibrations of a linear triatomic molecule
k k
m M m
q 1 q 2
q 3
m 2 2 M 2 1 X
• T = q̇ +q̇ + q̇2 = Tµν q̇µ q̇ν
2 1 3 2 2 µν


m 0 0
֒→ T =  0 M 0 
0 0 m
95
k k
• U = (q1 − q2 )2 + (q3 − q2 )2
2 2
k 2
= q1 + 2q2 + q32 − 2q1 q2 − 2q2 q3
2
1X
= Uµν qµ qν
2 µν
 
k −k 0
֒→ U =  −k 2k −k 
0 −k k
Uµν
• Sµν = p
Tµ Tν
 k k

m
− √mM 0
 
 
֒→ S =  √ k 2k k 
 − mM m
− √mM 
k k
0 − √mM m
• solve characteristic equation

k − Ω2 − √ k 0
m mM

−√ k 2k k
− Ω2 − √mM =0
mM m
k k
0 − √mM m
− Ω2
k n 2k k2 o
k
⇐⇒ − Ω2 − Ω2
− Ω2 −
m M m mM
k k k
−√ √ − Ω2 = 0
mM mM m
k
2
n 2k
2
k
2
2k 2 o
⇐⇒ −Ω −Ω −Ω − = 0
m M m mM
k
⇐⇒ Ω22 =
m
96
2k 2 k 2k 2k 2
− Ω2 + + Ω4 − = 0
mM m M mM
k 2m
⇐⇒ Ω2 Ω2 − 1+ = 0
m M
=⇒ Ω21 = 0
2 k 2m
Ω3 = 1+
m M
r r
֒→ k k
Ω1 = 0 , Ω2 = , Ω3 = 2m + M
m mM
• transformation matrix (= eigenvectors)

 k    
− Ω2
− √k 0
m λ mM V1λ 0
 −√ k 2k
− Ω2
− k 
∢  mM M λ
√
mM 
V 2λ
 =  0 
k k 2 V3λ 0
0 − √mM m − Ωλ
Ω1 = 0 :
 pm 
k k

V
m 11
− √ V
mM 21
= 0
M
− √ k V + 2k V − √ k V = 0
mM 11 M 21 mM 31 =⇒ V 1 = α  p1 
k k m
− √mM V21 + m V31 = 0 M
k
Ω22 = m :


k

− √mM V22 = 0 1
− √ k V + k( 2 − 1 k
)V22 − √mM V32 = 0 =⇒
mM 12 M m V 2 =β 0 
k
− √mM V22 = 0 −1
k
Ω23 = mM
(2m + M) :
 q 

− 2k
V
M 13
− √ k
mM 23
V = 0
− 21 M
m
−√ k V − k V − √ k V = 0  
mM 13 m 23 mM 33 =⇒ V 3 = γ
 1
q


k 2k
− √mM V23 − m V33 = 0 − 12 M
m
97
normalization:
r
2m M
=⇒ α2 (1 + ) = 1 ⇐⇒ α =
M M + 2m
1
2β 2 = 1 ⇐⇒ β = √
2
r
1M 2m
γ 2 (1 + ) = 1 ⇐⇒ γ =
2m M + 2m
 p q 
m √1 M
−
 q M +2m 2
q 2(M +2m) 
 M 2m 
=⇒ V =  M +2m
0 M +2m 
 p q 
m √1 M
M +2m
− 2
− 2(M +2m)
• general solution
s
1 1 M
q1 (t) = √ Q1 (t) + √ Q2 (t) − Q3 (t)
M + 2m 2m 2m(M + 2m)
s
1 2m
q2 (t) = √ Q1 (t) + Q3 (t)
M + 2m M(M + 2m)
s
1 1 M
q3 (t) = √ Q1 (t) − √ Q2 (t) − Q3 (t)
M + 2m 2m 2m(M + 2m)
with Q1 (t) = A1 t + B1 (Q̈1 = 0)

Q2 (t) = A2 cos(Ω2 t + δ2 )
Q3 (t) = A3 cos(Ω3 t + δ3 )
• initial conditions to bring out normal oscillations
case (i) : q1 (0) = q2 (0) = q3 (0) = 0

q̇1 (0) = q̇2 (0) = q̇3 (0) = v0
=⇒ q1 (t) = q2 (t) = q3 (t) = v0 t

֒→ translation of entire molecule
98
case (ii) : q1 (0) = A = −q3 (0) , q2 (0) = 0

q̇1 (0) = q̇2 (0) = q̇3 (0) = 0
=⇒ q1 (t) = A cos Ω2 t = −q3 (t)

q2 (t) = 0
2m
case (iii) : q1 (0) = q3 (0) = −A , q2 (0) = A
M
q̇1 (0) = q̇2 (0) = q̇3 (0) = 0
=⇒ q1 (t) = −A cos Ω3 t = q3 (t)

2m
q2 (t) = A cos Ω3 t
M
• visualization of normal models

Ω1 : • −→ • −→ • −→ translation
Ω2 : • −→ • ←− •
Ω3 : ←− • • −→ ←− •
c) General case (just briefly)

X
• EoMs : Tµν q̈ν +Uµν qν = 0 , µ = 1, ..., n
ν
• ansatz (motivated by case b)):

X
qν (t) = Vνλ Qλ (t) with Qλ (t) = Aλ cos(Ωλ t + δλ )
λ
X
֒→ q̈ν = Vνλ Q̈λ
λ
X
= − Vνλ Ω2λ Qλ
λ
insert X
֒→ Uµν − Ω2λ Tµν Vνλ Qλ = 0
νλ
99
since Qλ are linearly independent, one has:

X
Uµν − Ω2λ Tµν Vνλ = 0 (∗)
ν
• This set of equations has a nontrivial solution if:

det Uµν − Ω2λ Tµν = 0
=⇒ obtain 0 ≤ Ω21 < ... < Ω2n
• Insert Ω2λ into (∗) to obtain the transformation matrix V . One

imposes the orthonormality relations
V TT V = 1
and verifies
V T U V = Ω2
i.e. U and T are diagonalized simultaneously.
Details: [TM], Chap. 12.4, [GPS], Chap. 6
Appendix A
Supplementary material
A.1 Energy conservation of a conservative N -

particle system
N
X
∢ ṗk = Fk + fki
i=1
X X X
−→ fki · vk = mk · v̇k · vk − Fk · vk
i,k k k
1d X X drk
rhs : = mk vk2 + ∇k Uk ·
2 dt k k
dt
d X mk 2
= v + Uk (rk (t))
dt k 2 k
d X
= Tk + Uk
dt k
1 X
lhs : = fki · vk + fik · vi
2
i6=k
1 X
= fki · vk − vi
2 i6=k
100
101
define : rki := rk − ri
vki := vk − vi
֒→ ∇k Ūki = ∇ki Ūki = −∇i Ūik

֒→ fki = −∇k Ūki = −∇ki Ūki
1X
−→ lhs = fki · vki
2 i6=k
1X drki
= − ∇ki Ūki ·
2 i6=k dt
d 1X
= − Ūki rki (t)
dt 2 i6=k
−→ lhs − rhs = 0
d X 1X
(Tk + Uk ) + Ūki = 0
dt k 2 i6=k
d
T +U = 0.
dt
A.2 Does S assume a minimum for the actual

path (i.e., is the stationary point always
a minimum)?
(i) Prelude: U(x) = 0
m 2
L=ẋ = T
2
The (Lagrangian) equation of motion ẍ = 0 implies ẋ = v0 = const.
Recall the definition of neighbouring paths
x(α, t) = x(t) + αη(t)

ẋ(α, t) = ẋ(t) + αη̇(t)
with η(t1 ) = η(t2 ) = 0

102
Z
m t2 2
∢ S(α) = ẋ (α, t) dt
2 t1
Z Z t2 Z
m t2 2 m 2 t2 2
= ẋ (t) dt + mα ẋ(t)η̇(t) dt + α η̇ (t) dt
2 t1 t1 2 t1
Z t2 Z
m 2 t2 2
= Sactual + mαv0 η̇(t) dt + α η̇ (t) dt
t1 2 t1
Z
t2 m 2 t2 2
= Sactual + mαv0 η(t)t1 + α η̇ (t) dt
2 t1
= Sactual + β
m 2
R t2
with β = 2
α t1
η̇ 2 dt > 0. We do have a minimum, since S(α) >
Sactual .
(ii) Let’s add a potential
m 2
L= ẋ − U(x)
2
and Taylor-expand U around the stationary point
dU 1 d2 U
U(x(α, t)) = U(x)α=0 + αη + α2 η 2 + . . .
dx α=0 2 dx2 α=0
1
= U(x) − F (x)αη − F ′ (x)α2 η 2 + . . .
2
insert this into S:
Z t2
S(α) = L(x(α, t), ẋ(α, t)) dt
t1
Z Z t2 Z
m t2 2 m 2 t2 2
= ẋ (t) dt + mα ẋ(t)η̇(t) dt + α η̇ (t) dt
2 t1 t1 2 t1
Z t2 Z t2 Z
α2 t2 ′
− U(x(t))dt + α F (x(t))η(t)dt + F (x(t))η 2 (t)dt + O(α3)
t1 t1 2 t1
Z t2 Z t2
m 2
= ẋ − U(x) dt + α (mẋη̇ + F (x)η) dt
t1 2 t1
Z
α2 t2
+ mη̇ 2 + F ′ (x)η 2 dt + O(α3)
2 t1
Z t2 Z
α2 t2
= Sactual + αm (ẋη̇ + ẍη) dt + mη̇ 2 + F ′ (x)η 2 dt + O(α3 )
t1 2 t1
103
Z t2 Z t2 Z t2
t2
I1 = (ẋη̇ + ẍη)dt = ẋη̇dt + ẋη t1 − ẋη̇dt = 0
t1 t1 t1
Z t2 Z t2 Z t2
2 2
2
I2 = ′
mη̇ + F (x)η dt = m η̇ dt + F ′ (x)η 2 dt
t1 t1 t1
the term linear in α always vanishs, but the second-order term can be
anything, since the second integral in I2 can be positive or negative (or
zero). Let’s check two simple examples:
Example 1: homogeneous gravitational field
F = −mg → F ′ (x) = 0
Z t2
֒→ I2 = m η̇ 2 dt > 0
t1
in this case we have a minimum, since S(α) > Sactual .
Example 2: linear force
F = kx → F ′ (x) = k > 0
Z t2
֒→ I2 = (mη̇ 2 + kη 2 )dt > 0
t1
this also yields a minimum, but for the harmonic oscillator i.e., for
F = −kx we can’t draw any conclusion from our expression for I2 .
Indeed, one can show that—for long enough time intervals—the sta-
tionary point of the action is a saddle point in this case.
A.3 Differential constraints

m differential constraints can be characterized by
3N
X
aij dxj + ait dt = 0 ; i = 1, ..., m
j=1
e.g. rolling disk (previous example (vii) in Chap. 2.2.1.a):
dxCM ± (R cos ϕ)dψ = 0 , dyCM ± (R sin ϕ)dψ = 0

104
֒→a11 = 1 , a21 = 1
a14 = ±R cos ϕ , a24 = ±R sin ϕ
all other coefficients including a1t and a2t are zero

 =0 scleronomic
if ait

6= 0 rheonomic
∢ total differential of a holonomic constraint f (x1 , ..., x3N ) = 0
3N
X ∂f
df (x1 , ..., x3N ) = dxj
j=1
∂xj
= ∇f · dr (with ∇ = (∂x1 , ..., ∂x3N ))
֒→ reobtain f by integration
Z Z
f (x1 , ..., x3N ) = df = ∇f · dr (path − independent)
∂2f ∂2f
⇐⇒ =
∂xj ∂xl ∂xl ∂xj
(fulfilled, if f sufficiently well-behaved)
3N
X
=⇒ differential constraints aij dxj + ait dt are holonomic
j=1
∂aij ∂ail ∂aij ∂ait

if = ; =
∂xl ∂xj ∂xt ∂xj
proof: for holonomic constraint fi (x1 , ..., x3N , t) = 0
3N
X ∂fi ∂fi
dfi = dxj + dt = 0
j=1
∂xj ∂t
3N
X ∂fi ∂fi
= aij dxj + ait dt with aij = , ait =
j=1
∂xj ∂t
105
f fulfils integrability conditions
∂ 2 fi ∂ 2 fi ∂ 2 fi ∂ 2 fi
= , =
∂xj ∂xl ∂xl ∂xj ∂xj ∂t ∂t∂xj
k k k k
∂ail ∂aij ∂ait ∂aij
= , = q.e.d.
∂xj ∂xl ∂xj ∂t
One can use the integrability conditions to check that the constraints of the
rolling disk are not holonomic.
A.4 Details regaring the proof of equivalence

of Newton’s and Lagrange’s equations of
motion
Let us look more closely at the force components in
3N
∂L ∂U X
=− = Fi + fij
∂xi ∂xi j=1
Basically this is a (tedious) counting exercise:
(i) external forces:

N
X
Uk (rk ) = U1 (x1 x2 x3 ) + U2 (x4 x5 x6 )
k=1
+ ... + UN (x3N −2 x3N −1 x3N )
≡ Uext (x1 , . . . , x3N )
→ external force on k-th mass point: Fk = −∇k Uk ֒→ m-th component

of the force on k-th mass point:
∂ ∂Uext
Fkm = − U =− m ,
m k
(k = 1, ..., N; m = 1, 2, 3)
∂xk ∂xk
106
explicitly:
 1 
Fk ≡ Fkx = − ∂x∂ k Uext
 
 2 
 Fk ≡ Fky = − ∂ Uext
 ⇐⇒ Fi = − ∂Uext ,

 ∂yk i = 1, ..., 3N



 ∂xi
 F 3 ≡ F z = − ∂ Uext 
k k ∂zk
F1x F1y F1z F2x F2y F2z ... FNx FNy FNz
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
F1 F2 F3 F4 F5 F6 F3N −2 F3N −1 F3N
(ii) internal forces (note that Ūik = Ūki ):

fij = ∇i Ūij = −∇j Ūij
N
X
Ūij (ri − rj ) = Ū21 (x1 x2 x3 , x4 x5 x6 )
j<i
+ Ū31 (x1 x2 x3 , x7 x8 x9 ) + ... + Ū32 (x4 x5 x6 , x7 x8 x9 ) + ...

+... + ŪN,N −1 (x3N −5 x3N −4 x3N −3 , x3N −2 x3N −1 x3N )
some examples:
∂ Ū ∂
− = − (Ū21 + Ū31 + ... + ŪN 1 )
∂x1 ∂x1
N
∂ x
X
x
= − (Ū12 + Ū13 + ... + Ū1N ) = f1 = f1j
∂x1 j=1
3N
X
= f1j
j=1
∂ Ū ∂
− = − (Ū21 + Ū23 + Ū24 + ... + Ū2N )
∂x5 ∂x5
X N 3N
X
y y
= f2 = f2j = f5j
j=1 j=1
..
.
107
∂ Ū ∂
− = − (ŪN 1 + ŪN 2 + ... + ŪN,N −1 )
∂x3N ∂x3N
XN X3N
= fNz = fNz j = f3N j
j=1 j=1
3N
∂ Ū X
֒→ in general : − = ................ = fij
∂xi j=1
with matrix fji (3N × 3N) of the following structure:
i\j 1 2 3 4 5 6 7 .. .. 3N-2 3N-1 3N

1 0 0 0 x 0 0 x .. .. x 0 0
2 0 0 0 0 x 0 0 .. .. 0 x 0
3 0 0 0 0 0 x 0 .. .. 0 0 x
4 x 0 0 0 0 0 x .. .. x 0 0
5 0 x 0 0 0 0 0 .. .. 0 x 0
6 0 0 x 0 0 0 0 .. .. 0 0 x
7 x 0 0 x 0 0 0 0 0 .
.. .. .. ..
. . . 0 .. .. .
3N
summary: fij 6= 0 if 1 ≤ i = j ± 3n ≤ 3N ; (n = 1, ..., N − 1)

3N
∂ Ū X
it follows : − = fij , i = 1, ..., 3N
∂xi j=1
⇐⇒ −∇i Ūij = fij , i, j = 1, ..., N

3N
d ∂L ∂L X
=⇒ − = ṗi − Fi − fij = 0
dt ∂ ẋi ∂xi j=1
N
X
⇐⇒ ṗk = Fk + fkj , k = 1, ..., N
j=1
108
A.5 Some details regarding rigid body dy-

namics
Any rotation about a fixed point can be represented by a series of three
rotations about designated axes through the Eulerian angles (α, β, γ).
coordinates coordinates
wrt. Sf wrt. Sb
α β γ
(x′1 , x′2 , x′3 ) −→ (x′′1 , x′′2 , x′′3 ) −→ (x′′′ ′′′ ′′′
1 , x2 , x3 ) −→ (x1 , x2 , x3 )
r Sb = D γ
D β
D α
r Sf = D r Sf
Each rotation is characterized by a rotation matrix. The product of these
matrices (note their order) characterizes the resulting rotation.
rotation 1: rotate reference system Sf counterclockwise through α about x′3 -

axis:
    ′ 
x′′1 cos α sin α 0 x1
 x′′2  =  − sin α cos α 0   x′2  (r′′ = D α
r Sf )
x′′3 0 0 1 x′3
rotation 2: rotate the new system (S ′′ ) counterclockwise through β about

x′′1 -axis:
 ′′′     ′′ 
x1 1 0 0 x1
 x′′′
2
= 0 cos β sin β   x′′2  (r′′′ = D β
r′′ )
′′′
x3 0 − sin β cos β x′′3
rotation 3: rotate the new system (S ′′′ ) counterclockwise through γ about

x′′′
3 -axis:
     ′′′ 
x1 cos γ sin γ 0 x1
 x2  =  − sin γ cos γ 0   x′′′ 2
 (rSb = D r′′′ )
γ
x3 0 0 1 x′′′
3
109
multiply the matrices:
D = Dγ DβDα
 
cos α cos γ − sin α cos β sin γ sin α cos γ + cos α cos β sin γ sin β sin γ
=  − cos α sin γ − sin α cos β cos γ − sin α sin γ + cos α cos β cos γ sin β cos γ 
sin α sin β − cos α sin β cos β
Since the total rotation consists of three rotations, the angular velocity
vector ω (which describes the orientation of the body system) is the sum of
the three corresponding angular velocity vectors:
ω = ωα + ωβ + ωγ
We need to find the components of these vectors in the body system.
(i) ω α points in x′3 -direction in the fixed system (cf. rotation 1). The com-
ponents of this vector in the body system are obtained from applying
D = D γ D β D α:
     
α̇1 0 α̇ sin β sin γ

֒→ ωα =  α̇2  = D  0  =  α̇ sin β cos γ 
Sb
α̇3 α̇ α̇ cos β
| {z }
k
 
0 (note that the componentes
D γ D β 0  of ω α do not change
α̇ during the first rotation)
(ii) ω β points in x′′1 = x′′′ ′′ ′′′

1 direction in the systems S and S . The com-
ponents with respect to the body system are obtained from applying
D γ:
     
β̇1 β̇ β̇ cos γ

ωβ =  β̇2  = D γ  0  =  −β̇ sin γ 
Sb
β̇3 0 0
110
(iii) ω γ points in x′′′

3 = x3 direction in the S
′′′
and the body system:
     
γ̇1 0 0

ωγ =  γ̇2  = Dγ  0  =  0 
Sb
γ̇3 γ̇ γ̇
Collecting terms yields ωk = α̇k + β̇k + γ̇k
ω1 = α̇ sin β sin γ + β̇ cos γ

ω2 = α̇ sin β cos γ − β̇ sin γ
ω3 = α̇ cos β + γ̇
In the principal axes body system we have

1X
Trot = Ik ωk2
2 k
1 2
= I1 ω1 + I2 ω22 + I3 ω32
2
1n
= I1 α̇2 sin2 β sin2 γ + β̇ 2 cos2 γ + 2α̇β̇ sin β sin γ cos γ
2
+ I2 α̇2 sin2 β cos2 γ + β̇ 2 sin2 γ − 2α̇β̇ sin β sin γ cos γ
o
+ I3 α̇2 cos2 β + γ̇ 2 + 2α̇γ̇ cos β

= Trot βγ, α̇β̇ γ̇
We are after the Lagrangian equations of motion for

Lrot = Trot − U = Trot βγ, α̇β̇ γ̇ − U αβγ
!  
q1 = α
d ∂Trot ∂Trot ∂U
− =−  q2 = β 
dt ∂ q̇k ∂qk ∂qk
q3 = γ
Let’s look at one coordinate explicitly:
q3 = γ :
∂Trot d ∂Trot
• = I3 γ̇ + α̇ cos β ֒→ = I3 α̈ cos β − α̇β̇ sin β + γ̈
∂ γ̇ dt ∂ γ̇
111
∂Trot
• = I1 α̇2 sin2 β sin γ cos γ − β̇ 2 sin γ cos γ + α̇β̇ sin β cos2 γ − α̇β̇ sin β sin2 γ
∂γ

+ I2 − α̇2 sin2 β sin γ cos γ + β̇ 2 sin γ cos γ − α̇β̇ sin β cos2 γ + α̇β̇ sin β sin2 γ

I3 α̈ cos β − α̇β̇ sin β + γ̈
n 2 2
֒→ EoM: − I1 − I2 α̇ sin β − β̇ 2 sin γ cos γ
2 2
o ∂U
+ α̇β̇ sin β cos γ − sin γ = −
∂γ
One can show that this equation can be written in the form
∂U
I3 ω̇3 − I1 − I2 ω1 ω2 = −
∂γ
let’s check:

ω1 ω2 = α̇ sin β sin γ + β̇ cos γ α̇ sin β cos γ − β̇ sin γ
= α̇2 sin2 β sin γ cos γ + α̇β̇ sin β cos2 γ
− β̇ 2 sin γ cos γ − α̇β̇ sin β sin2 γ

= α̇2 sin β − β̇ 2 sin γ cos γ + α̇β̇ sin β cos2 γ − sin2 γ
Similar (quite lengthy) calculations yield EoMs for q1 = α and q2 = β:
q1 = α :
n o
α̈ I1 sin2 γ + I2 cos2 γ sin2 β + I3 cos2 β

+ 2α̇β̇ I1 sin2 γ + I2 cos2 γ − I3 sin β cos β

+ 2α̇β̇ I1 − I2 sin2 β sin γ cos γ + β̈ I1 − I2 sin β sin γ cos γ

+ β̇ 2 I1 − I2 cos β sin γ cos γ
n o ∂U
+ β̇ γ̇ I1 − I2 cos2 γ − sin2 γ − I3 sin β + γ̈I3 cos β = −
∂α
112
q2 = β :

α̈ I1 − I2 sin β sin γ cos γ

+ α̇2 I3 − I1 sin2 γ − I2 cos2 γ sin β cos β
n o
+ α̇γ̇ I3 + I1 − I2 cos2 γ − sin2 γ sin β
∂U
+β̈ I1 cos2 γ + I2 sin2 γ − 2β̇ γ̇ I1 − I2 sin γ cos γ = −
∂β
Let’s consider linear combinations of the three EoMs:

1 ∂U ∂U ∂U
I1 ω̇1 − I2 − I3 ω2 ω3 = − sin γ − sin β cos γ + cos β sin γ
sin β ∂α ∂β ∂γ
≡ N1
1 ∂U ∂U ∂U
I2 ω̇2 − I3 − I1 ω3 ω1 = − cos γ + sin β sin γ + cos β cos γ
sin β ∂α ∂β ∂γ
≡ N2
Summary (using N3 = − ∂U∂γ

):

I1 ω̇1 − I2 − I3 ω2 ω3 = N1 




I2 ω̇2 − I3 − I1 ω3 ω1 = N2 Euler equations





I3 ω̇3 − I1 − I2 ω1 ω2 = N3
X
i.e. εijk Ik ω̇k − Nk − Ii − Ij ωi ωj = 0
k


 1 (ijk) even permutation of (1, 2, 3)



with ǫijk = −1 (ijk) odd permutation of (1, 2, 3)





0 else
Bibliography
[FC] Fowles, G. R. and G. L. Cassiday: Analytical Mechanics. Thom-

son.
[GPS] Goldstein, H., C. Poole and J. Safko: Classical Mechanics.

Addison Wesley.
[Tay] Taylor, J. R.: Classical Mechanics. University Science Books.
[TM] Thornton, S. T. and J. B. Marion: Classical Dynamics. Thom-

son.
113

PHYS3010

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PHYS3010

Uploaded by

Copyright:

Available Formats

PHYS 3010:

Department of Physics and Astronomy

2 Hamilton’s Principle — Lagrangian and Hamiltonian dynam-

2.6.4 d’Alembert’s principle . . . . . . . . . . . . . . . . . . 57

A Supplementary material 100

1.1 Prelude and Overview

(i) Material objects: The central property of the objects of CM is (inertial)

→ One can define Cartesian coordinate systems, and the mathematical

Extensions of the notions of space and time (beyond this course)

– special theory of relativity → 4-dimensional ”Minkowski” space

• To obtain deeper insights into the foundations of CM and physics

• To learn a powerful problem-solving strategy (i.e., the application of

1.2 Recap of Newtonian Mechanics

Lex secunda: ”The change of motion is proportional to the motive force

Definition of an inertial reference frame: a reference frame in which

According to the first law we have:

• The second law tells us (quantitatively) what happens if one or more

if ṁ = 0: mv̇ = ma = mr̈ = Fnet

ṁ = 0 is certainly fulfilled for mass points, but ṁ 6= 0 is possible,

(i) If 0 = Fnet = ṗ =⇒ p = const −→ recover first law!

This is why inertial reference frames are so important: the forces

• The third law expresses a fundamental property of physical forces1 :

If one fixes the absolute scale by introducing a standard mass (the

terms of accelerations: the smaller mass speeds up faster. Hence, we

(i) The physical origin of forces is (normally) not discussed in CM.

r̈(t) = f(r, ṙ, t) (+ initial conditions)

(iii) Conservation laws—briefly reviewed in the next section—emerge as

1.2.2 (Linear) momentum, angular momentum, work,

֒→ r(t) = r0 + v0 t uniform motion

• position of a MP wrt. centre of mass: r′ k = rk − R

Figure 1.3: Definition of the centre of mass.

∢ EoM of kth MP:

The centre of mass of a particle system moves as if it were a single particle

Momentum conservation (holds in isolated systems):

This makes the centre-of-mass a convenient choice for the origin of an

Let’s consider a system of N MPs. The total angular momentum is defined

total torque is defined as

Accordingly, we have derived

In particular, total angular momentum is conserved in isolated systems.

If the force is a vector field, i.e., if F = F(r) W is given by a line integral

(i) For uniform F and rectilinear motion: W = F · r

Example 1: If you simply hold (but do not move) a mass m in Earth’s

Example 2: uniform circular motion

=r (R cos ωt, R sin ωt)

Let’s connect work with Newton’s EoM:

definition: kinetic energy

−→ T (t) = T (t0 ) + W (t0 → t) ’work-energy theorem’

Note that U is defined only up to a constant by this equation. Normally one

work − energy theorem −→ W = U(1) − U(2) = T (2) − T (1)

E = T + U = const conservation of energy

mv̇ = −∇U(r) + Fdiss |·v

if the following applies:

(i) Fk = Fk (rk ) = −∇k Uk (rk ) (conservative external forces)

(ii) ∇k × fki = ∇i × fik = 0 (conservative internal forces)

2.1 Preliminary formulation of Hamilton’s prin-

(iii) Hamilton’s principle (HP) is an integral principle (in contrast to New-

(iv) HP is a fundamental principle of modern physics.

Figure 2.1: Illustration of Hamilton’s principle

(v) EoMs can be derived from HP

2.1.1 Calculus of variations

sought-after: ’path’ x(t) with x(t1 ) = x1 and x(t2 ) = x2 such that

Theorem: a necessary condition for I to assume an extremum value for x(t)

Proof: assume that x(t) is the sought-after path

x(α, t) = x(t) + αη(t) with η(t1 ) = η(t2 ) = 0

∂ q̇µ ∂ X ∂qµ ∂qµ ∂qµ