H.

Hergert
Facility for Rare Isotope Beams
and Department of Physics & Astronomy

PHY422/820: Classical Mechanics

FS 2021
Worksheet #12 (Nov 15 – Nov 19)

November 15, 2021

1 Preparation
• Lemos, Sections 4.8-4.9, 5.1, 5.3-5.7
• Goldstein, Sections 6.1-6.5

2 Spinning Tops
2.1 The Symmetric Heavy Top
2.1.1 Defining the Lagrangian
Let us now consider a symmetric top with A = B 6= C with a fixed point that is moving under the
influence of gravity. Its dynamics is best described using a Lagrangian whose degrees of freedom
are the Euler angles (φ, θ, ψ). Since we have no translation, the kinetic term is purely rotational,
1
T = ω~ ·I ·ω
~. (1)
2
Since T is expressed in terms of tensors it is form-invariant under rotations, and we can evaluate
it in the principal axis frame, where we have
1 1
T = A(ωx20 + ωy20 ) + Cωz20 . (2)
2 2
Using the expression for the body-fixed components of ω in terms of the Euler angles
~ = (φ̇ sin θ sin ψ + θ̇ cos ψ, φ̇ sin θ cos ψ − θ̇ sin ψ, φ̇ cos θ + ψ̇)T ,
ω (3)
we have
1   1
T = A θ̇2 + φ̇2 sin2 θ + C(ψ̇ + φ̇ cos θ)2 . (4)
2 2
Denoting the distance of the top’s center-of-mass from the fixed point by l, the potential term can
be written as
V = M gl cos θ , (5)
where we have used that θ is the angle between the space-fixed axis z and the body-fixed axis z 0
(see worksheet #10). Thus, the Lagrangian for the top reads
1   1
L = A θ̇2 + φ̇2 sin2 θ + C(ψ̇ + φ̇ cos θ)2 − M gl cos θ. (6)
2 2

1
2.1.2 Conservation Laws and General Solution
Inspecting the Lagrangian (6), we immediately notice that φ and ψ are cyclic variables, so we
obtain
∂L
= A sin2 θ + C cos2 θ φ̇ + C ψ̇ cos θ ≡ pφ = const. ,


(7)
∂ φ̇
∂L  
= C ψ̇ + φ̇ cos θ = Cωz 0 ≡ pψ = const. . (8)
∂ ψ̇

We see that pψ is the z 0 -component of the angular momentum in the body-fixed frame, while pφ
is the z-component of the angular momentum in the space-fixed frame. The latter is conserved
because gravity does not exert any torque in the space-fixed x or y direction. Moreover, the
Lagrangian does not explicitly depend on the time, hence the total energy of the top is conserved:
1   1
E = T + V = A θ̇2 + φ̇2 sin2 θ + C(ψ̇ + φ̇ cos θ)2 − M gl cos θ . (9)
2 2
We can rewrite it in terms of the conserved momenta as

1 (pφ − pψ cos θ)2 p2ψ

E = Aθ̇2 + + + M gl cos θ . (10)
2 2A sin2 θ 2C
Since the third term on the right-hand side is constant, it is convenient to introduce

p2ψ 1 (pφ − pψ cos θ)2

E0 = E − = Aθ̇2 + + M gl cos θ (11)
2C 2 2A sin2 θ

which now only depends on θ and θ̇. This allows us to follow the same strategy we used for one-
dimensional potentials to determine the general solution for θ(t): We separate the variables and
integrate to obtain
dθ0
Z θ
q = t − t0 , (12)
θ0 2 0 − V (θ 0 ))
A (E eff

where we have introduced the effective potential

(pφ − pψ cos θ)2

Veff (θ) ≡ + M gl cos θ . (13)
2A sin2 θ
Once we have θ(t), we can plug it into Eqs. (7) and (8) and integrate them to obtain φ(t) and ψ(t).
The integral in Eq. (12) cannot be solved analytically in general, but we can discuss the dynamics
of the top qualitatively using the effective potential, as in our treatment of central-force problems.

2.1.3 Discussion of the Effective Potential

First, we will determine the general shape of the effective potential:

• A physical solution for θ(t) must be confined to the interval [0, π] at all times. At the borders
of the interval, the denominator of the first term in the effective potential (13) will vanish,
and Veff (θ) → +∞.

2
 g1 (u) f (u)

g2 (u)

−1 u1,0 0 u2,0 1
u

Figure 1: Graphical determination of the extrema of Veff (θ) (cf. Eq. (18)).

• The derivative of Veff (θ) is

0 2 (pφ − pψ cos θ) pψ sin θ · 2A sin2 θ − (pφ − pψ cos θ)2 · 4A sin θ cos θ

Veff (θ) = − M gl sin θ
4A2 sin4 θ
(pφ − pψ cos θ) pψ · sin2 θ − (pφ − pψ cos θ)2 · cos θ
= − M gl sin θ
A sin3 θ
(pφ − pψ cos θ) (pψ sin2 θ − pφ cos θ + pψ cos2 θ)
= − M gl sin θ
A sin3 θ
(pφ − pψ cos θ) (pψ − pφ cos θ)
= − M gl sin θ , (14)
A sin3 θ
which allows us to find the extrema as solutions of
0 (pφ − pψ cos θ0 ) (pψ − pφ cos θ0 )
Veff (θ0 ) = − M gl sin θ0 = 0 . (15)
A sin3 θ0
Substituting u = cos θ, we have
dVeff dVeff dθ
=
du dθ du  
(pφ − pψ cos θ) (pψ − pφ cos θ) 1
= − M gl sin θ −
A sin3 θ sin θ
(pφ − pψ cos θ) (pψ − pφ cos θ)
=− + M gl
A sin4 θ
(pφ − pψ u) (pψ − pφ u)
=− + M gl
A(1 − u2 )2
!
= 0. (16)

We can find the solutions by introducing functions

f (u) ≡ M glA(1 − u2 )2 , g(u) = (pφ − pψ u) (pψ − pφ u) , (17)

3
 50Veff (θmin )
Veff (θ)

20Veff (θmin )

Veff (θmin )

0 (2) (1)
θ1 θ1 θmin (1)
θ2
(2)
θ2 π
θ

Figure 2: The effective potential for the heavy top.

and determining their intersection

!
f (u) = g(u). (18)
Typical examples are shown in Fig. 1: For physical parameters, the functions intersect in a
single point u0 . Since Veff (θ) → ∞ at the borders of the integral, this extremum must be a
global minimum.
From the preceding discussion, we can infer that the effective potential has the shape shown in
Fig. 2, and this allows us to classify the main types of motion for the symmetric heavy top.

2.1.4 General Features of the Motion

The motion of the symmetric heavy top will be a superposition of the following effects:
• a precession of the symmetry axis around the space-fixed z axis with instantaneous angular
velocity φ̇
• a nutation1 , that is, a periodic change of the tilt angle θ(t) between the top’s symmetry axis
and the space-fixed z axis,
• and a precession of body-fixed angular momentum around the symmetry axis (chosen to be
the body-fixed z 0 axis) with angular velocity ψ̇, just as in the case of the free symmetric top.
Combining Eqs. (7) and (8), we can determine the angular velocity φ̇:

pφ − pψ cos θ = (A sin2 θ + C cos2 θ)φ̇ + C ψ̇ cos θ − C ψ̇ cos θ − C φ̇ cos2 θ , (19)

hence p
φ
pφ − pψ cos θ pψ − cos θ
φ̇ = = p ψ . (20)
A sin2 θ A sin2 θ
We distinguish the following cases:
1
From Latin nutare, “to nod, to sway.”

4
 0 q1 q0 q q2 p
Fig. 4.15 The effective potential Veff (θ ).

q´ . .
q1 f f q1 q ´= q1
q´
q2 q2 q2
. . . .
f f f f

(a) (b) (c)

Fig. 4.16 The possible combinations of nutation and precession.
Figure 3: Types of motion for the symmetric heavy top.
General Features of the Motion
• E 0 > Veff (θ0 ), |pφ | > |pψ |: The angular velocity φ̇ has the same sign at all times, and the
topInwill
general, for in
precess any given
the same the angle θThe
E′ direction. periodically
symmetry oscillates
axis willbetween
trace theoutvalues
a curve andthe unit
θ1 on
θ2 shown in Fig. 4.15. The symmetry axis of the top is said to perform a nutation about turning
sphere, as shown in Fig. 3(a). While precessing, the axis will nutate between the
the vertical.
points θ1 and θDuring
2 that arethe determined from E 0 = Veffangular
nutation the instantaneous (θ1 ) = velocity
Veff (θ2 ), associated
as usual. with the

• E 0 precession
> V (θ ), of |p
eff
the |symmetry
0 < |p |: axis
φ Thisofcase
the top
ψ
around
leads to athe vertical
richer is
variety of phenomena. Introducing
p pφ
the auxiliary angle θ0 = arccos ppψφ ,− we p ψ can
cos θhave thep ψthree− cosbehaviors
θ shown in Figs. 3(a-c). If
φ
φ̇ = = p ψ . 0 (4.157)
the sign of φ̇ stays the same duringI1the sin 2entire
θ motion,I1 the sin 2 angle
θ θ lies outside of the physically
allowed range θ(t) ∈ [θ1 , θ2 ] for a top with a fixed energy. If θ1 < θ0 ≤ θ2 , φ̇ will change sign
There are several possibilities, to wit:
during a nutation, which causes the corkscrew-shaped trajectory shown in Fig. 3(b).
(a) |p φ | > |p ψ |. The precession rate φ̇ has the same sign all the time.
Last but not least, θ0 = θ1 yields the trajectory shown in Fig. 3(c). This case would occur if
′ ∈ (0, π ) defined by cos θ ′ = p φ , we
(b) |p φ | < |p ψ |. Introducing the ancillary angle θ
a top with a fixed energy that is spinning around its symmetry axis is released: p ψ The initial
have φ̇ =are
conditions p ψ (cos θ ′ − cos θ )/I1 sin 2 θ . In order to determine the position of θ ′ , note that
Eq. (4.155) gives Veff ′ (θ ′ ) = − mgl sin θ ′ < 0, and it follows that
θ = θ1 , θ̇ = 0 , φ̇ = 0 , ψ̇ = ωz 0 , θ ′ < θ0 . If θ ′ < θ1 (21)
thenimplies
which φ̇ has the same sign during the entire motion. In this case, as well as in the previous
one, the symmetry axis of the top traces out on the unit sphere with centre at O a curve
similar to the one shown in Fig. 4.16(a). ′ >θ θ, , the precession rate φ̇ changes sign
pφ = CωIfz 0θcos 11 (22)
periodically during the nutation and the motion
pψ = Cωz 0 , resembles the one depicted in Fig. 4.16(b). (23)
The situation portrayed in Fig. 4.16(c) corresponds
1 to θ ′ = θ . The last case is not so rare
1
2
as it seems. Suppose, for instance,Ethat = theCω z0 + M
2 initial
gl cos θ1are θ̇ = 0, φ̇ = 0, ψ̇ = ω3 ,
conditions (24)

and
pφ
cos θ0 = = cos θ1 . (25)
pψ

• E 0 = Veff (θ0 ): The tilt angle stays constant θ(t) = θ0 , and the axis precesses without
nutating, with a constant precession frequency:
pφ − pψ cos θ0
φ̇ = = const. (26)
A sin2 θ0

Let us now determine the initial conditions under which this can be achieved. Introducing

σ = pφ − pψ cos θ0 (27)

we also have
pψ − pφ cos θ0 = pψ sin2 θ0 − σ cos θ0 (28)

5
 and this allows us to rewrite Eq. (15) as a quadratic equation in σ:

σ 2 cos θ0 − σpψ sin2 θ0 + M glA sin4 θ0 = 0 . (29)

The general solutions are

q
pψ sin2 θ0 ± p2ψ sin4 θ0 − 4M glA sin4 θ0 cos θ0
σ± =
2 cos θ0
s !
Cωz 0 sin2 θ0 4M glA cos θ0
= 1± 1− , (30)
2 cos θ0 C 2 ωz20

where we have used pψ = Cωz 0 in the final step. We see that real solutions can only exist if
2p
ωz 0 ≥ ω0 = M glA cos θ0 , (31)
C
which requires θ0 ≤ π2 . Clearly, the existence of two solutions implies that there will be two
separate precession frequencies associated with σ+ and σ− .
In the limit ωz 0  ω0 , we have

Cωz 0 sin2 θ0 Cωz 0 sin2 θ0

  2   2
ω0 ω0
σ+ = 2+O = + O (32)
2 cos θ0 ωz20 cos θ0 ωz20

and
Cωz 0 sin2 θ0 1 ω02 Cωz 0 sin2 θ0 ω02
  
σ− = 1− 1− = , (33)
2 cos θ0 2 ωz20 4 cos θ0 ωz20
where we have expanded the square-root terms for small (ω0 /ωz 0 )2 . Note that σ− is suppressed
by one order in the expansion compared to σ+ . Plugging these results back into the expression
for the precession frequency, we find a fast precession frequency,
 2
σ+ Cωz 0 ω0
φ̇+ = 2 = +O , (34)
A sin θ0 A cos θ0 ωz20

and a slow precession frequency

σ− Cωz 0 ω02 Cωz 0 4 M gl

φ̇− = 2 = 2 = 2 2 M glA cos θ0 = . (35)
A sin θ0 4A cos θ0 ωz 0 4A cos θ0 C ωz 0 Cωz 0

2.1.5 The Vertical Top

One of the most frequently encountered examples is that of a symmetric top that is set to spin in
an upright position, with initial conditions θ(0) = 0 and ψ̇ = ωz 0 . From Eqs. (7) and (8), we see
that
pφ = pψ = Cωz 0 , (36)
and the effective potential becomes

C 2 ωz20 (1 − cos θ)2

Veff (θ) = + M gl cos θ
2A sin2 θ
C 2 ωz20 4 sin4 2θ
= + M gl cos θ
2A 4 sin2 2θ cos2 2θ

6
 C 2 ωz20 θ
= tan2 + M gl cos θ . (37)
2A 2
Now we consider what happens for small nutations with respect to the vertical axis. Expanding
the effective potential for small angles, we have

C 2 ωz20 θ 2 1 C 2 ωz20
     
1 2
Veff (θ) = + M gl 1 − θ = − M gl θ2 + M gl . (38)
2A 2 2 2 4A
Thus, the effective potential is a parabola that is offset by the constant M gl in energy. For
C 2 ωz20
− M gl > 0 , (39)
4A
the tops’ motion with. This implies that its initial angular velocity must exceeds a critical value
(cf. Eq. (31))
2p
ωz 0 > ωc = M glA . (40)
C
This explains the typical behavior of the upright top: As it is set in motion with a sufficiently
large ωz 0 , it will be spinning with barely a wobble. Eventually, the rotation will slow down due to
frictional energy losses, and the top will start to wobble, before eventually falling over when ωz 0
drops below the critical value.

3 Small Oscillations
Harmonic oscillators are ubiquitous in classical mechanics (and beyond). In fact, their importance
extends beyond free wave phenomena or systems that explicitly contain springs: As we have seen
throughout this course, oscillatory motion is also so prevalent in physics because it occurs when a
system is displaced out of a stable equilibrium. Consider the potential (or effective potential) as
a function of the generalized coordinates and perform a Taylor expansion around an equilibrium
point:
n
1 X ∂ 2 V 
 
X ∂V 
V (~q0 + ∆~q) = V (~q0 ) + ∆qi + ∆qi ∆qj + O(∆q 3 ) . (41)
∂qi  q~0 2 ∂qi ∂qj  q~0
i=1 ij

In an equilibrium configuration, the (generalized) forces on the system vanish, i.e.,

∂V
Qi = − = 0, (42)
∂qi
and the potential is approximately quadratic in the vicinity. If the matrix of second derivatives
(i.e., the Hessian) is positive definite at ~q0 , the potential will be that of a harmonic oscillator — and
that is precisely the condition under which an equilibrium is stable, as we saw in our discussions
of orbits or the motion of rigid bodies!
In this next segment of the course, we will develop the tools to describe small oscillations around
equilibria for general mechanical systems with multiple degrees of freedom.

3.1 Coupled Oscillators

3.1.1 Example: Two Coupled Pendula
We start the discussion of coupled oscillators by recalling the procedure for solving the equations
of motion of such systems. As an example, we consider two equal pendula of length l and mass m

7
that perform small oscillations while being coupled by an ideal spring with spring constant k and
natural length d (see Fig. 4). Letting the y axis point upward, the coordinates of the masses are
given by

x1 = l sin θ1 , ẋ1 = lθ̇1 cos θ1 , (43)

y1 = l(1 − cos θ1 ) , ẏ1 = lθ̇1 sin θ1 , (44)
x2 = d + l sin θ2 ẋ2 = lθ̇2 cos θ2 , (45)
y2 = l(1 − cos θ2 ) ẏ2 = lθ̇2 sin θ2 , (46)

so the kinetic energy reads

1
1  
T = m ẋ21 + ẏ12 + ẋ22 + ẏ22 = ml2 θ̇12 + θ̇22 . (47)
2 2
For the potential term, we first consider the spring potential:
k p 2
Vsp = (x2 − x1 )2 + (y2 − y1 )2 − d
2 
s  2
k y2 − y1
= (x2 − x1 ) 1 + − d , (48)
2 x2 − x1

where we have used that x2 − x1 > 0 for small-angle oscillations. For small angles, we also have

x2 − x1 = d + lθ2 − lθ1 + O(θ3 ) , (49)

θ22 θ2 l 2
− l 1 + O(θ4 ) = θ2 − θ12 + O(θ4 ) .


y2 − y1 = l (50)
2 2 2
We can expand the ratio in the potential,
l 2 2 4


y2 − y1 2 θ2 − θ1 + O(θ ) l 1
θ22 − θ12 + O(θ4 )


= 3
= l
x2 − x1 d + l(θ2 − θ1 ) + O(θ ) 2d 1 + d (θ2 − θ1 ) + O(θ3 ))
 
l l
θ22 − θ12 + O(θ4 ) 1 − (θ2 − θ1 ) + O(θ2 )


=
2d d
l
θ22 − θ12 + O(θ3 ) ,


= (51)
2d
so s
y2 − y1 2 1 y2 − y1 2
   
1+ =1+ + . . . = 1 + O(θ4 ) (52)
x2 − x1 2 x2 − x1
Now the spring potential reads
k
2 k
x2 − x1 − d + O(θ4 ) = lθ2 + d − lθ1 − d + O(θ3 )


Vsp = (53)
2 2
and the total potential becomes
 2
θ22

k 2 3 2 θ1
+ O(θ3 )


V = l θ2 − θ1 + O(θ ) + mgl +
2 2 2
 2
θ22

k 2 2 θ1
= l (θ2 − θ1 ) + mgl + + O(θ3 ) . (54)
2 2 2

8
 while the potential energy is written
k
V= [(d0 + η2 − η1 ) − d0 ]2 + mg[l(1 − cos θ1 ) + l(1 − cos θ2 ) ] . (5.107)
2

l l
q1 q2

h1 h2
Fig. 5.7 Two equal pendulums coupled by a spring (Example 5.7).
Figure 4: Coordinates for two pendula that are coupled by a spring.

Putting everything together, we obtain the Lagrangian

θ12 θ22
 
1   k
L = ml2 θ̇12 + θ̇22 − l2 (θ2 − θ1 )2 − mgl + . (55)
2 2 2 2
Next, we derive the Lagrange equations. The partial derivatives are
∂L d ∂L
= ml2 θ̇j , = ml2 θ̈j , (56)
∂ θ̇j dt ∂ θ̇j
∂L
= kl2 (θ2 − θ1 ) − mglθ1 , (57)
∂θ1
∂L
= −kl2 (θ2 − θ1 ) − mglθ2 , (58)
∂θ2
and the equations of motion read

ml2 θ̈1 − kl2 (θ2 − θ1 ) + mglθ1 = 0 , (59)

2 2
ml θ̈2 + kl (θ2 − θ1 ) + mglθ2 = 0 . (60)

The usual procedure for solving this system of equation is to make the ansatz
   
θ1 (t) θ1,0 iωt
= e . (61)
θ2 (t) θ2,0

Note that the same frequency ω is used for both components of the solution. We have

θ̈j (t) = −ω 2 θj (t) , (62)

and we can use this to write the system of equations of motion in matrix form:

−ml2 ω 2 + mgl + kl2 −kl2

  
θ1,0
= 0. (63)
−kl2 −ml2 ω 2 + mgl + kl2 θ2,0

This is an eigenvalue problem whose solutions are the characteristic frequencies ω 2 and associ-
ated characteristic vectors that define the normal modes of the system of coupled oscillators.
The characteristic polynomial of the matrix is

−ml2 ω 2 + mgl + kl2 −kl2

 
det
−kl2 −ml2 ω 2 + mgl + kl2

9
 = (−ml2 ω 2 + mgl + kl2 )2 − (−kl2 )2 = 0 . (64)

Moving the final term to the right and taking the square root on both sides, we end up with the
equation
−ml2 ω 2 + mgl + kl2 = ±kl2 . (65)
Its solutions are
g
2 2 g 2k
ω+ =, ω− = + . (66)
l l m
We can determine the characteristic vectors by plugging the frequencies back into Eq. (63). For
2 , we have
ω+
 (+) !
−ml2 gl + mgl + kl2 −kl2

ρ1
0= 2 2 g 2 (+)
−kl −ml l + mgl + kl ρ2
 2  (+) !
kl −kl2 ρ1
= 2 2 (+) , (67)
−kl kl ρ2

which means that the components of the characteristic vector need to satisfy
(+) (+)
ρ1 = ρ2 . (68)

Thus, a normalized solution is  

(+) 1 1
ρ
~ =√ . (69)
2 1
2 , we find
For ω+
!
g k (−)
−ml2 + mgl + kl2 −kl
2


l+ 2m ρ1
0= 2 g k
−kl −ml2 l + 2m + mgl + kl2 (−)
ρ2
 (−) !
−kl2 −kl2

ρ1
= , (70)
−kl2 −kl2 (−)
ρ2

which implies
(−) (−)
ρ1 = −ρ2 , (71)
so a solution for the second characteristic vector is
 
(−) 1 1
ρ
~ =√ . (72)
2 −1

Figure 5 visualizes the two normal modes we have found. For ρ ~ (+) , the two pendula are in sync,
while the masses swing in opposite directions for ρ (−)
~ , as indicated by the signs of the characteristic
vectors.

3.1.2 Normal Coordinates and Normal Modes

In the example discussed in the previous section, we determined the normal modes of a system of
coupled oscillators after deriving the equations of motion. The characteristic vectors associated
with these modes constitute a complete basis for the description of the system — in fact, if we

10
 169 Normal Modes of Vibration

Fig. 5.8 Normal modes of vibration of two equal pendulums coupled by a spring.

Figure 5: Normal modes for two pendula that are coupled by a spring.
The two linearly independent real solutions can be chosen as
make a change of basis from our initial choice ! " of coordinates ! and "
(1) 1 (2) 1 basis vectors to the normal mode
basis, the equations of motion areϱdecoupled = , ϱ = (5.114)
1 and can be solved −1 independently.
The central premise of the Lagrangian formalism, however, is to formulate the problem in the
and choice
most efficient the general solution (5.104)
of generalized takes thefrom
coordinates formthe very beginning (unless we need to deal with
#2
some constraints explicitly). If we can identify the matrix appearing in the equations of motion
(s)
η 1 (t) =
of a system of oscillators C(s)already
ϱ1 cos(ω
in sthe φs ) = C(1) cos(ω
t + Lagrangian, t+φ
we 1can 1)+C
(2)
diagonalize cos(ω φ2 ) , L directly
2 t +express
it and
s=1
in terms of normal coordinates with respect to the characteristic vectors. The Lagrange equations
(5.115)
for the normal coordinates are then particularly easy to derive, because they correspond to simple,
#2
uncoupled oscillators. (s)
η2 (t) = C(s) ϱ2 cos(ωs t + φs ) = C(1) cos(ω1 t + φ1 ) − C(2) cos(ω2 t + φ2 ) .
[...] Consider a general Lagrangian of the form
s=1
1X (5.116)
L= Mjk q̇j q̇k − V (~q) , (73)
2 0, η2 (0) = a, η̇1 (0) = η̇2 (0) = 0, we infer
From the initial conditions η1 (0) = jk

C(1) cos
where ~q = (q1 , . . . , qn ) and M is the mass
(2)
cos φ2 = coordinates
φ1 +inCgeneralized
tensor 0, (5.117a) #5):
(cf. worksheet
(1) (2)
C cos φ1 − C cos φ2 = a , (5.117b)
A
(1) ∂~ri ∂~ rj φ = 0 ,
(2)
(5.117c)
M1jkC= sin φm
−ω 1 i − ω2·C sin
X
=M 2 kj . (74)
(1) i=1 ∂q j ∂q
(2)k
−ω1 C sin φ1 + ω2 C sin φ2 = 0 . (5.117d)
By adding
Let us consider the first twoηand
displacements the lastan
j around twoequilibrium
of Eqs. (5.117) we deduce
position of the system:

C(1) cos φ1qj==−C

q0j(2)+cos
ηj ,φ2 = a/2 , (5.118a) (75)
C(1) sin φ1q̇j==Cη̇(2)
j . sin φ2 = 0 . (5.118b) (76)

Neither
The potential canCbe(1) expanded
= 0 nor as C(2) = 0 is admissible because this would give rise to a
contradiction with (5.118a). Therefore,  sin φ1 = sin φ2 2= 0, whose simplest solution is
X ∂V  (1)
φ1 = φ 2 V=(~ 0, by reason of which C ηj=+−1C(2) =∂ a/2. V Finally, the solution to the
X
3
q ) = V (~q0 ) +  ηj ηj + O(η )
equations of motion is writtenj ∂q j

q~ 2 ∂qj ∂qk q~0
| {z }0 jk
=0a
(t) = ( cos ω1 t − cos ω2 t ) ,
η11X (5.119a)
≡ V (~q0 ) + Vjk2ηj ηk + O(η 3 ) , (77)
2 a
η2 (t)
jk = ( cos ω1 t + cos ω2 t ) , (5.119b)
2
where we with 1 and ω2 given by (5.113).
haveωused that the gradient of the potential vanishes in equilibrium, and introduced the
symmetric matrix
∂2V
Vjk = = Vkj , (78)
∂qj ∂qk

11
which is nothing but the Hessian of the system evaluated at ~q0 . Dropping cubic and higher terms
in η, a general potential therefore looks like a system of coupled oscillators near an equilibrium
configuration. For a system of coupled harmonic oscillators, the higher-order terms vanish and the
expression for the potential is exact.
Introducing
Tjk ≡ Mjk |q~0 = Tkj , (79)
and dropping the constant V (~q0 ) we can write the Lagrangian as a quadratic form,

1 1 1X 1X
L = ~η˙ · T · ~η˙ − ~η · V · ~η = η̇j Tjk η̇k − ηj Vjk ηk . (80)
2 2 2 2
jk jk

Equations of Motion and Generalized Eigenvalue Problem

The partial derivatives are
 
∂L 1X ∂ 1 X ∂ηk ∂ηl
=− (ηk Vkl ηl ) = − Vkl ηl + ηk Vkl
∂ηj 2 ∂ηj 2 ∂ηj ∂ηj
kl kl
1X
=− (δjk Vkl ηl + ηk Vkl δjl )
2
kl
X
=− Vjk ηk , (81)
k

and
∂L X
= Tjk η̇k , (82)
∂ η̇j
k

where we have used the properties of the Kronecker delta and combined terms by using the freedom
to rename summation indices. In terms of the matrices T and V , the Lagrange equations therefore
read
X
(Tjk η̈k + Vjk ηk ) = 0 , j = 1, . . . , n . (83)
k

To identify the normal modes, we make the usual switch to complex coordinates

ηj −→ zj = z0j eiωt , zj ∈ C , (84)

and plug them into the equations of motion to obtain

X
−ω 2 Tjk z0k + Vjk z0k = 0 ,


(85)
k

or in matrix form

V − ω 2 T ~z0 = 0 .


(86)

12
This is, in fact, a generalized eigenvalue problem since T appears in place of the identity
matrix 1. It contains the information about the curvature of the configuration manifold iself
in the vicinity of the equilibrium configuration we expand around. If the configuration manifold
is flat, e.g., because we use a Cartesian basis to describe the sytem, T will be simply a diagonal
matrix with the masses of the oscillators on the diagonal. If T is invertible, we can in principle
convert Eq. (86) into a regular eigenvalue problem by multiplying with T −1 from the left,

T −1 V − ω 2 1 ~z0 = 0 .


(87)

Solving Eq. (86), we obtain the characteristic frequencies ωs2 as the eigenvalues, and the char-
~ (s) . Since T and V are real and symmetric, the ωs2 are guaranteed to be real as
acteristic vectors ρ
well. The characteristic vectors will be real as well, since their entries are just linear combinations
of the original coordinates we used to describe the oscillators. A general oscillation of the system
can then be expanded in the basis of normal modes instead of our initial choice of basis vectors
(denoted here by {~e (1) , . . . , ~e (n) }):
n
X n
X
(j)
~η (t) = ηj (t)~e = ρ (s) ,
ζs (t)~ (88)
j=1 s=1

where ζs are the normal coordinates of the problem.

Determination of the Normal Coordinates

Let us now assume that we have a Lagrangian given by Eq. (80), and introduce normal coordinates
that are related to our initial displacement coordinates by a matrix A that is to be determined:

~η = Aζ~ , ηk , ζk ∈ R. (89)

Plugging this relation into the Lagrangian, we have

1˙ ˙ 1~
L = ζ~ · AT T A · ζ~ − ζ · AT V A · ζ~ . (90)
2 2
If the normal coordinates are to be decoupled, AT T A and AT V A must each be diagonal because
˙
they depend on ζ~ and ζ,
~ respectively, so there can be no cancellations in off-diagonal matrix elements
between the two terms.
To proceed, we first define the inner product induced by T :
  X
~a, ~b ≡ ~a · T · ~b = ak Tkl bl (91)
kl

It generalizes the standard scalar product from a flat space to the geometry of configuration manifold
or, more precisely, the tangent vector spaces to the configuration manifold at the equilibrium
configuration. For Tkl = δkl , we get back the regular scalar product.
The characteristic vectors ρ~ (s) are orthonormal with respect to the new inner product:
  X
(r) (s)
ρ~ (r) , ρ
~ (s) = ρk Tkl ρl = δrs . (92)
kl

13
Let us now define the modal matrix A in terms of the characteristic vectors
 (1) (n) 
ρ1 . . . ρ1
A ≡  ... .. ..  , (93)

. . 
(1) (n)
ρn . . . ρn
or componentwise
(s)
Aks = ρk . (94)
Then
(r) (s) !
X X X
ρk Tkl ρl = Akr Tkl Als = ATrk Tkl Als = δrs , (95)
kl kl kl
which means that T will be represented by the identity matrix in the basis of characteristic vectors:

AT T A = 1 . (96)

For the potential term, the eigenvalue problem implies

(s) (s)
X X
Vkl ρl = ωs2 Tkl ρl . (97)
l l

Defining the diagonal matrix

W = diag ω12 , . . . , ωn2 ,


(98)
we can rewrite this as X X X
Vkl Als = Tkl Als = Tkl Alr Wrs , (99)
kl l lr
or in matrix form as
V A = T AW . (100)
Multiplying from the left by AT and using the fact that T is the identity matrix in the basis of
characteristic vectors, we have
AT V A = A T
| {zT A} W = W , (101)
=1
so the modal matrix also renders V diagonal.
Plugging our results for the transformed matrices back into Eq. (90), the Lagrangian becomes
1˙ ˙ 1 1˙ ˙ 1 1 X 2 
L = ζ~ · AT T A · ζ~ − ζ~ · AT V A · ζ~ = ζ~ · ζ~ − ζ~ · W · ζ~ = ζ̇s − ωs2 ζs2 , (102)
2 2 2 2 2 s

and the Lagrange equations yield the decoupled equations of motion

ζ̈s + ωs2 ζs = 0 , s = 1, . . . , n . (103)

as desired. The normal coordinates can now be determined from original ansatz (89) by mutiplying
it with AT T from the left-hand side and using AT T A = 1. We obtain

ζ~ = AT T ~η (104)

using the coefficient matrix T and the modal matrix A obtained from the characteristic vectors of
our generalized eigenvalue problem.
[Add remark about additional generalized forces / cf. homework.]

14
3.1.3 Example: A Linear Triatomic Molecule
We now demonstrate the introduction of normal coordinates in the Lagrangian for the example
of a linear triatomic molecule, which we describe as a central mass M that is connected to two
equal masses m via identical springs with spring constants k. The spring potential can be viewed
as the harmonic approximation to the (classical) intermolecular interaction close to an equilibrium
configuration, i.e.,

1 X ∂ 2 V 

V (|xi − xj |) = V0 + ∆xi ∆xj + O(∆x3 ) . (105)
2 ∂xi ∂xj |xi,0 −xj,0 |
ij

Referring to the coordinates introduced in Fig. 6, we have

k k
V = (x2 − x1 − l)2 + (x3 − x2 − l)2 , (106)
2 2
where l is the natural length of the spring. If we expand the squares in V , we get
k 2
x1 + 2x22 + x23 + 2l2 − 2x1 x2 + 2lx1 − 2lx2 − 2x2 x3 − 2lx2 + 2lx3 ,


V = (107)
2
which cannot be written as a quadratic form in the coordinates due to the presence of terms like
lxi . Thus, we first have to introduce coordinates that directly measure the displacement out of
equilibrium:
η1 = x1 + l , η2 = x2 , η3 = x3 − l (108)
(other choices are possible as long as they allow us to write L as a quadratic form.) In terms of the
new coordinates, the Lagrangian of the system can be written as
1 1 k k
L = m(ẋ21 + ẋ23 ) + M ẋ22 − (x2 − x1 − l)2 − (x3 − x2 − l)2
2 2 2 2
1 1 k k
= m(η̇1 + η̇3 ) + M η̇2 − (η2 − η1 ) − (η3 − η2 )2 ,
2 2 2 2
(109)
2 2 2 2
or
1 1
L = ~η˙ · T · ~η˙ − ~η · V · ~η (110)
2 2
with the matrices    
m 0 0 k −k 0
T = 0 M
 0 , V = −k 2k −k  . (111)
0 0 m 0 −k k
Note that we have distributed terms of the form 2ηi ηj to the upper and lower off-diagonal parts of
V so that the matrix is symmetric.
Now we solve the generalized eigenvalue problem for V − ω 2 T , starting with the determination
of the characteristic polynomial:

k − mω 2
 
−k 0
0 = det  −k 2k − M ω 2 −k 
0 −k k − mω 2
= k − mω 2 (2k − M ω 2 )(k − mω 2 ) − (−k)2 − (−k) (−k)(k − mω 2 )

   

= k − mω 2 (2k − M ω 2 )(k − mω 2 ) − 2k 2

 

= k − mω 2 2k 2 − k(2m + M )ω 2 + mM ω 4 − 2k 2

 

15
 that
∂V ∂V
0= = −k (x2 − x1 − ℓ) , 0= = k (x2 − x1 − ℓ) − k (x3 − x2 − ℓ) , (5.149)
∂x1 ∂x2

m M m

x1 x2 x3 x
0
Fig. 5.9 Model for the study of vibrations of a symmetric linear triatomic molecule.

Figure 6: Coordinates for the linear triatomic molecule

= ω 2 k − mω 2 −k(2m + M ) + mM ω 2 .

 
(112)

We can immediately read off the solutions

 
k k 2m
ω12 = 0, ω22 = , ω32 = 1+ . (113)
m m M

Plugging these frequencies into the eigenvalue equation, we can determine the characteristic
vectors. For ω12 = 0, we have   
k −k 0 ρ1
−k 2k −k  ρ2  = 0 , (114)
0 −k k ρ3
and the first and third rows give us the equations

ρ1 = ρ2 , ρ2 = ρ3 . (115)

Thus, our first characteristic vector is

~ (1) = C1 1 1 1


ρ (116)

with some normalization constant C1 .

Analogously, ω22 = k/m yields
 k     
k − km m −k 0 ρ1 0 −k 0 ρ1
 −k k
2k − M m −k  ρ2  = −k k(2 − M
m ) −k  ρ2  = 0 , (117)
k −k
0 −k k − mm ρ3 0 0 ρ3

and we obtain the equations

−kρ2 = 0 , (118)
m 
−k(ρ1 + ρ3 ) + k 2 − ρ2 = 0 . (119)
M
Thus, our second characteristic vector is


1
~ (2)
ρ = C2  0  . (120)
−1

16
 175 Normal Coordinates
175 Normal Coordinates

~ (1) :
ω12 , ρ
Normal Coordinates
Mode 1
Mode 1
ω22 , ρ
~ (2) :

Mode 2 Mode 1 Mode 2 Mode 3 Mode 3

g. 5.10 Normal modes ~of(3)theFig.

ω32 , ρ : symmetric Normal
5.10 linear modesmolecule.
triatomic of the symmetric linear triatomic molecule.

Mode 2 ProceedingMode 3 we find

similarly
Proceeding similarly we find
modes of the symmetric linear triatomic molecule. ⎛ 1 ⎞ ⎛ ⎛ ⎞⎞ ⎛ ⎞
11 1
Figure 7: Normal modes of(3)
the linear triatomic molecule
ϱ(2) = B ⎝ 0 ⎠ , ϱ ϱ(2)= = C⎝ ⎝ 0 ⎠⎠, . ϱ(3) = (5.159)
B −2m/M C ⎝ −2m/M ⎠ .
eding similarly we find we plug ω2 = k/m (1−1
Finally,
1
−1
+ 2m/M ) into the eigenvalue equation and obtain 1
3
The normal⎛modes 1
⎞ of vibration are
The normal
⎛

sketchedmodes
⎞
1 −k inofFig. 5.10. Only
vibration the
 modes
are sketched  in 2Fig. and 5.10.
3 Only the m
k − k 1 + 2m

0 ρ1
correspond
(2)
ϱ = B to
⎝ vibrations
⎠ of
(3)the molecule,
00 = , ϱcorrespond
M
−k = C 2k−2m/M ⎝ to the
− kvibrations
M mode
⎠ number
.2m of the−k

1 being
molecule,
(5.159) a rigid
 ρ2the uniform
 mode number 1 being a
(k) m 1+ M
translatory motion. −1 The vectors 0 ϱ aremotion.
translatory associated
1 −k Thewith kdistinct
vectors − k ϱ1(k)+frequencies,ρ3 so they
are associated
2m
withare distinct frequenc


M
automatically orthogonal in the inner product (5.124), as will be generally proved in the
 2m
normal modes of vibration are sketched Mautomatically
in M Fig. 5.10.orthogonal
Only 1 the modes in the 2inner and product
3 (5.124), as will be generall
 
−k −k 0 ρ
next section. The normalisation
spond to vibrations of the molecule, the
= −k −k
modecondition
−k
number is ρ
The1normalisation
being a rigid uniform
0 next−k section. condition is

m
  2 
% −k 2mM 2 ρ3
atory motion.ϱ(s)
TheTϱvectors (k)
= ϱ .Tare associated T with distinct
% frequencies, Mϱso2 they = are
T (s) (s) (s) (s) 2 (s) 2
(s)
kℓ ϱk ϱℓ = (s)1 ⇒(s)m(ϱ 1 +Tϱ3ϱ (s) ) +(s) =2 1, 2, 3(s)2(5.160) (s)2
1 , s (s)
matically orthogonal in the inner product (5.124),
ϱ Tϱ as =
will be generally =
kℓ k ℓ proved in the
ϱ 1 ⇒ m(ϱ1 + ϱ3 ) + (121) Mϱ2 = 1 , s =
kℓ
ection. Theandnormalisation condition
Thus, the components
the normalised vectors
ofisthe characteristic vector
can be chosen as
kℓ must satisfy the equations
(s)2 and(s) the2 normalised vectors can be chosen as
% (s) (s) (s)⎛
2 2m ⎞
Tϱ(s) = Tkℓ ϱk ϱℓ = 1 ⇒ m(ϱ1 + ϱ3 ) + Mϱ2 − = 1 ρ11,−sρ= 2 =1,02, , 3 (5.160) ⎛ ⎞ (122)
(1) 1 M 1
kℓ ϱ =√ ⎝ ⎠
1 M, (1) 1 (5.161a)
he normalised vectors can be chosen as 2m +
−k(ρ M1 + ρ3 ) − ρϱ2 = 0= , √ ⎝ 1 ⎠ , (123)
1 m 2m + M
⎛ ⎞ ⎛ −ρ⎞ 2m 1
1 −
(1) 1 1 2
M
ρ 3 = 0 ,
⎛ ⎞
(124)
ϱ =√ ⎝(2)1 ⎠ , 1 (5.161a) 1
and we have 2m + Mϱ 1= √
⎝ 0⎠,
1 ⎝ (5.161b)
2m −1

1ϱ
 (2)
=√ 0⎠,
⎛ ⎞ (3) 2m  2m
1 ~ = C3 −⎛
ρ 
M . ⎞ −1 (125)
(2) 1 ⎝ 1 1
ϱ =√ 0(3)⎠ , 1 (5.161b) ⎛ ⎞
2m ϱ = √ with different frequencies, ⎝ −2m/M ⎠ . (5.161c) 1
1
(k)
The vectors ρ −1
are associated 2m(1 + 2m/M) so they are guaranteed to be or-
thonormal in the inner product ⎛ induced by T⎞ ϱ(3) 1= √ condition yields ⎝ −2m/M ⎠ .
. The normalization
With the(3)help of the modal 1(s) (s)matrix
 1   2m(1 + 2m/M) 1
⎝ ~ =.m ρ1 +⎞ρ3 (5.161c)
(s)⎠
(s) (s)2 (s)2 (s)2
ϱ =√ ρ
~ , ρ ~ =ρ ~ −2m/M
⎛· T (1)
·ρ + M ρ2 = 1 , (126)
2m(1 + 2m/M) With the help ⏐ ofϱ ⏐the modal
(2) ϱ (3) matrix
1ϱ ⏐
A=⎝⏐ ⏐ ⏐ ⎠, ⎛ (1) (2) ϱ(5.162)
(3)
⎞
the help of the modal matrix ⏐ ⏐ 17 ⏐ ϱ
⏐ ϱ
⏐ ⏐
⎛ (1) (2) (3)
⎞' ' '
A = ⎝⏐ ⏐ ⏐ ⎠,
ϱ
⏐ ϱ
⏐ ϱ
⏐ ⏐ ⏐ ⏐
the normal coordinates ⎝ ⏐ ⏐ ⏐ ⎠ T
A = ⏐ are⏐ given⏐ by ζ ,= A Tη, whence, for example, (5.162)' ' '
which allows us to determine the normalization constants Cs :
1 1 1
C1 = p , C2 = √ , C3 = p . (127)
2m(1 + M/2m) 2m 2m(1 + 2m/M )

Assembling the modal matrix A from the characteristic vectors, we have

 
√ 1 1 √ 1
1+M/2m 1+2m/M 
1 √ 1 √ 2m/M
(1) (2) (3) −


A= ρ ~ ρ
~ ρ
~ =√  1+M/2m 0 
(128)
1+2m/M

2m  1 1

√ −1 √
1+M/2m 1+2m/M

and we can compute the normal coordinates

√ 1 √ 1 √ 1
 
mη1
  
ζ1 1+M/2m 1+M/2m 1+M/2m
1 
ζ~ = AT T · ~η ⇒ ζ2  = √

 1 0 −1  M η2 
 (129)
2m
− √ 2m/M
 
ζ3 √ 1 √ 1
1+2m/M 1+2m/M 1+2m/M mη3

The three normal modes of the molecule are illustrated in Fig. 7. The mode (ω12 , ρ ~ (1) ) is
actually not vibrational, but corresponds to a rigid uniform translation of the entire molecule. This
is reflected in the normal coordinate ζ1 : carrying out the matrix-vector product in Eq. (129) and
plugging in the definition of the displacement coordinates in terms of the initial coordinates xi , we
have
mη1 + M η2 + mη3 mx1 + M x2 + mx3 √
ζ1 = √ = √ = 2m + M X , (130)
2m + M 2m + M
where X is the center of mass coordinate of the molecule. The equation of motion for the normal
coordinate ζ1 therefore becomes

ζ̈1 + ω12 ζ1 = 0 ⇒ Ẍ = 0 , (131)

|{z}
=0

which is of course solved by

X(t) = X0 + V0 t . (132)
In mode (ω22 , ρ
~ (2) ), the outer masses move in opposite direction from each other while the center
mass stays fixed, while in mode (ω32 , ρ ~ (3) ) the outer masses move in sync in the opposite direction
as the center mass.

18
4 Group Exercises

Problem G29 – Two Masses on a Circle

Consider two identical masses m that can move on a circular horizontal track of radius R (see
figure). Each of the masses is connected to a fixed point by identical springs with constant 2k, and
a spring with constant k connects the masses to each other.
1. Construct the Lagrangian for the system in terms of
the (counterclockwise) angular displacements φ1 and
φ2 of the two masses from their equilibrium positions,
as shown in the diagram. k m
Hint: Distances on the circular track can be expressed
in terms of arc lengths.
φ1
2. Determine the normal modes, i.e., characteristic fre-
quencies and vectors (vectors do not need to be nor-
φ2
malized). Sketch and interpret your solutions.

3. Now the fixed point is released, so that the system 2k

m
can rotate freely on the circular track. The track itself
remains at rest. How will the characteristic frequencies 2k
change qualitatively as a result? @
I
Hint: A calculation is not necessary, but if you explic- @
fixed point
itly want to check, you can use that two springs con-
nected “in series” can be replaced by a single spring
with constant keff = ( k11 + k12 )−1 .

Problem G30 – Vertical Oscillators

Consider two equal masses m that are suspended from the ceiling using identical springs with
constants k and unstretched length l. Assume that the masses are only allowed to move vertically.

1. Find the Lagrangian for the two masses in terms of

the absolute positions of the two masses, as well as in
terms of displacements of the system out of equilib-
rium. What is the advantage of the latter?
m ~
g
2. Show that the characteristic frequencies of the normal
modes are √
2 (3 ± 5)k
ω± = . (133)
2m m

19