Classical Mechanics

Small Oscillations
Dipan Kumar Ghosh
UM-DAE Centre for Excellence in Basic Sciences,
Kalina Mumbai 400098
September 24, 2016

1 Introduction
When a conservative system is displaced slightly from its stable equilibrium position,
it undergoes oscillation. The cause of oscillation is the restoring forces which are called
into play. Restoring forces can do both positive and negative work. When the work done
is positive, the restoring forces change the potential energy into kinetic energy and when
the work done is negative, they change kinetic energy back into potential energy.
For most mechanical systems, when the system is not too far from the equilibrium, the
restoring force is proportional to the displacement (F = kx). Such oscillators are called
linear oscillators. For linear oscillators, the oscillation frequencies are independent of the
amplitude of oscillation. Oscillator motion can be damped in the presence of resistive
forces. Resistive forces extract energy from the oscillator. For low velocities, the resistive
forces are proportional to velocity. Oscillators, whether damped or undamped, can be
driven by external agencies which continuously supply energy to the oscillator to keep it
oscillating. Such oscillators are known as forced or driven oscillators. Driven oscilla-
tors can cause amplitude of oscillation to become very large when the driving frequency
matches the natural frequency of oscillation. This is known as the phenomenon of reso-
nance.

2 Normal Modes
2.1 Equilibrium:
Consider a system with qi  as the generalised coordinates. Since the system is conser-
vative, the forces acting on the system are derivable from a potential energy function
c
©D. K. Ghosh, IIT Bombay 2

V (q1 , q2 ,    , qN ). Lagrange dened equilibrium as a conguration in which all generalised

∂V
forces vanish, i.e. = 0. Clearly, in such a situation, the system will not change its
∂qi
conguration. However, even when Qi = 0, the system may not be stable in the sense
that if it is slightly disturbed from a position of equilibrium, it may not return to the
position of equilibrium. If it does, such a conguration is called one of stable equilibrium
- otherwise the equilibrium is unstable.
Example: Simple Pendulum

θ The potential energy is given by V (θ) =

+mgl(1  cos θ), so that

∂V
x F (θ) =  = mgl sin θ = mgx
∂θ

V=0
The generalised force corresponding to θ in this case is actually the restring torque.
Equilibrium occurs when the restoring torque is zero. There are two such positions, θ = 0
and θ = π.
Let us look at the form of the Lagrangian near these two positions.
1
L = ml2 θ̇2  mgl(1  cos θ)
2
1
Near θ = 0, cos θ ≈ 1  θ2 so that
2
1 1
L = ml2 θ̇2  mglθ2
2 2
1
so that the potential energy is V (θ) = mglθ2 and the corresponding generalised force
2
is mglθ which is of restoring nature. On the other hand, near the second position of
1
equilibrium θ = π, cos θ = cos(π + δθ) ≈  cos δθ = 1 + δθ2 . In this situation,
2
1 1
L = ml2 θ̇2 + mgl(δθ)2
2 2
the corresponding force is anti-restoring, making the equilibrium unstable.
For one dimensional holonomic systems, equilibrium can be either stable on unstable
(leaving out a trivial case of neutral equilibrium where the potential energy function is
spatially at) for which the potential energy has an extremum

∂V
=0 (1)
∂qi
c
©D. K. Ghosh, IIT Bombay 3

for every generalised coordinate qi . Let the position of equilibrium be qi0 . If the position
is one of stable equilibrium, the potential energy has to be minimum. This is because, the
system being conservative, the total energy is constant. If we go away from the position of
minimum potential energy, it leads to an increase in the potential energy and a consequent
decrease in the kinetic energy. Thus the system returns back to the equilibrium position.
For stable equilibrium, we, therefore, have

∂ 2V
>0 (2)
∂qi ∂qj

The converse would be true for an unstable equilibrium.

Without loss of generality, let us shift the equilibrium position to the origin (q1 = q2 =
   , = qN = 0). If the system is disturbed to a conguration qi , we can write,
∑ ( ∂V ) (
1 ∑ ∂ 2V
)
V (q1 , q2 ,   ) = V (0, 0,   ) + qi + qi qj + higher order terms
i
∂qi 0 2 i,j ∂qi ∂qj 0

where the partial derivatives are evaluated at the position of equilibrium and all higher
order terms which involve third order or higher corrections are neglected. If the potential
energy is measured from its minimum value, we choose V (0, 0,   ) = 0. Along with
( )
∂V
=0
∂qi 0

The leading term in the change in potential energy is then

( )
1 ∑ ∂ 2V
>0
2 i,j ∂qi ∂qj 0

for stable equilibrium. Let us write

( )
∂ 2V
Vij =
∂qi ∂qj 0

so that ∑1
V = Vij qi qj (3)
i,j
2

with Vij = Vji .

Now, the kinetic energy of a scleronomic system is, in general, a quadratic in generalised
velocities, and can be written as
1∑
T = Tij q̇i q̇j (4)
2 i,j
c
©D. K. Ghosh, IIT Bombay 4

where the coecients tij are, in general, functions of generalised coordinates. One can
expand tij in a Taylor series about the equilibrium position
∑ ( ∂tij )
Tij (q1 , q2 ,   ) = tij (0, 0,   ) + qk +   
k
∂q k 0

( )
∂tij
It turns out that the quantities and the higher order derivatives are negligibly
∂qk 0
small so that the coecients tij s can be essentially treated as constants having the same
values as they would have in the equilibrium position. Thus around the equilibrium
position, the Lagrangian has the following structure:
1∑
L=T V = (tij q̇i q̇j  Vij qi qj )
2 i,j

The Lagrangian equations of motion

( )
d ∂L ∂L
 =0
dt ∂ q̇k ∂qk

can then be written as

d ∑1 1∑
[tij δik q̇j + tij q̇i δkj ] 
Vij (δik qj + qi δkj ) = 0
dt i,j 2 2 i,j
   
1 ∑ ∑ 1 ∑ ∑
tkj q̈j + tik q̈i  Vik qj + Vik qi = 0 (5)
2 j i
2 j i

Changing the dummy summation index j to i in the rst and the third terms of the above
and using the symmetry of Vij and of tij , we get
∑ ∑
tik q̈i + Vik qi = 0
i i

for each k. We seek a solution to the above equations of the form

qi = Ai eiωt

which gives ∑ 
Vik  ω 2 tik Ai = 0 (6)
i

The equation is a homogeneous equation in Ai s and the condition for existence of the
solution is
det(Vik  ω 2 tik ) = 0
which is a single algebraic equation of n th degree in ω 2 . This equation has n roots
some of which are real and some complex (some of the roots may be degenerate). We are
c
©D. K. Ghosh, IIT Bombay 5

only interested in real roots of the above equation. ωk ’s determined from this equation
are known as characteristic feequencies or eigenfrequencioes.
From physical arguments it is clear that for real physical situations, the roots are real
and positive. This is because the existence of an imaginary part in ω would mean time
dependence of qk and q̇k such that the total energy would not be conserved in time and
such solutions are unacceptable.
We can arrive at the same conclusion mathematically as well. Multiplying (6) with A∗k
and summing over k we get
∑
(Vik  ω 2 tik )A∗k Ai = 0
i,k

so that  ∗
2 i,k Vik Ak Ai
ω =  ∗
i,k tik Ak Ai
Both the numerator and the denominator are real because Vik = Vki and tik = tki . It
is seen that the terms are positive as well because expressing Ai = ai + ibi , we have
∑ ∑
tik A∗i Ak = tik (ai  ibi )(ak + ibk )
i,k i,k
∑
= tik (ai ak + bi bk )
i,k

where the imaginary terms cancel because of symmetry of tik . Thus we have been able
 
to express i,k tik A∗i Ak as a sum of two positive semi-denite terms ( tik ai ak = aT ta is
positive denite).

2.2 Matrix Formulation

Let us rewrite (6) (using symmetry properties of V and T ) as
∑
(Vki  λtki )Ai = 0
i
2
where λ = ω . Let us dene a column vector
 
A1
 
A 
A =  2

AN
The matrices V and t are given by
   
V11 V12    V1N t11 t12    t1N
   
V V22    V2N  t t22    t2N 
V =  21  T =  21 
              
VN 1 VN 2    VN N tN 1 tN 2    tN N
c
©D. K. Ghosh, IIT Bombay 6

Using these, the equation (6) can be expressed as a matrix equation

V A = ω2T A (7)

Note that this equation is not in the form of an eigenvalue equation as V A is not equal to
a constant times A but a constant times T A. (If T is invertible, one can get an eigenvalue
equation T −1 V A = ω 2 IA.
Since we have N homogeneous equations, we have N modes, i.e. N solutions for ω 2 . Let
us denote the k-th mode frequency by ωk2 = λk . Let the vector A corresponding to this
mode be written as  
Ak1
 
 Ak2 
Ak =  
  
AkN
We then have
V Ak = λk T Ak (8)
Taking conjugate of this equation and changing the index k to i, we get

Ãi V = λi Ãi T (9)

where we have used à to denote the transpose of the matrix A.. From (8) we get by
multiplying with Ãk
Ãk V Ak
λk = (10)
Ãk T Ak
From (8) and (9) it follows that

Ãi V Ak = λk Ãi T Ak
Ãi V Ak = λi Ãi T Ak

so that
(λk  λi )Ãi T Ak = 0
Thus, if the eigenvalues are non-degenerate, i.e. if λi 6= λk , we get the orthogonality
condition
Ãi T Ak = 0 (11)
Note that this is dierent from the orthogonality condition on eigen vectors for a regular
eigenvalue equation. Since (7) does not uniquely determine A, we dene normalization
condition as
Ãi T Ai = 1 (12)
Example 1:
Consider two masses m1 = 2m and m2 = m connected by three springs, as shown.
c
©D. K. Ghosh, IIT Bombay 7

k 1= 4k k 2= k k = 2k
3

m 1 = 2m m2= m

We know that a single mass  spring system with mass m and spring constant k has a
natural frequency of oscillation km. Let us consider the system in the gure. Let the
generalised coordinates be displacement of the masses from their equilibrium positions,
the mass m1 being displaced by x1 while the mass m2 by an amount x2 . The central
spring is then compressed or stretched by an amount x2  x1 . Let us attempt to solve the
problem using the force method. The equations of motion for m1 and m2 are

m1 ẍ1 = 4kx1  k(x1  x2 ) (13a)

m2 ẍ2 = 2kx2  k(x2  x1 ) (13b)

The sign of the last term is xed by taking x2 to be large positive so that the force on m1
is in the positive direction, as it ought to be.
We will attempt to solve this pair of coupled equations (13a) and (13b) by using a bit
of guesswork and a bit of luck. This will not work except in cases which show sucient
symmetry. The idea is to nd a linear combination of x1 and x2 so that the coupled
equations become uncoupled. Let us rewrite (13a) and (13b) as

2mẍ1 = 5kx1 + kx2 (14a)

mẍ2 = kx1  3kx2 (14b)

It can be easily seen that

7
m(ẍ1  ẍ2 ) =  k(x1  x2 ) (15a)
2
m(2ẍ2 + ẍ1 ) = 2k(2x1 + x2 ) (15b)

Thus, if we dene new normal coordinates

y1 = x 1  x2
y2 = 2x1 + x2 (16)

the equations for y1 and y2 are uncoupled,

7
mÿ1 =  ky1
2
mÿ2 = 2ky2 (17)
 
The normal coordinates oscillate with independent frequencies 2km and 7k2m,
which are known as the normal modes. These new generalised coordinates execute
c
©D. K. Ghosh, IIT Bombay 8

simple periodic oscillations known as normal oscillations. The new normal coordinates
satisfy
ÿα + ωα2 yα = 0 (18)
With this intuitive background, let us look at the problem more formally.
The Lagrangian of the system is
1 1 1 1 1
L = m1 ẋ21 + m2 ẋ22  k1 x21  k2 (x2  x1 )2  k3 x22 (19)
2 2 2 2 2
Dene t and V matrices
( )
∂ 2T m1 0
t= = (20)
∂ ẋi ∂ ẋj 0 m2
( )
∂ 2V 5k k
V = = (21)
∂xi ∂xj k 3k

Condition for existence of non-trivial solution is

det(Vik  ω 2 tik ) = 0

which gives ∣ ∣
∣5k  2mω 2 k ∣∣
∣ =0
∣ k 3k  mω 2 ∣
which has the solution ω 2 = 7k2m and 2km. The normal modes are now found from
(6)
ω2T A = V A
( )
A1 7k
Writing A = , we get, for ω 2 =
A2 2m
( )( ) ( )( )
7k 2m 0 A1 5k k A1
=
2m 0 m A2 k 3k A2

For this normal mode we have A2 = 2A1 so that (normalizing)

( )
1 1
A= √
6m 2
2k
For ω 2 = , a similar calculation gives
m
( )
1 1
A= √
3m 1

The general solution can then be written as

( ) ( ) ( ) ( ) ( )
x1 1 1 1 1
=A cos ω− t + B sin ω− t + C cos ω+ t + D sin ω+ t
x2 1 1 2 2
c
©D. K. Ghosh, IIT Bombay 9

We can x the constants A, B, C and D by using initial conditions. Suppose we we pull

the two masses aside and release,
x1 (0) = x10 , x2 (0) = x20 , ẋ1 (0) = ẋ2 (0) = 0
we then have
A + C = x10
A  2C = x20
Bω− + Dω+ = 0
Bω−  2Dω+ = 0
which gives
2 1
A = x10 + x20
3 3
B=D=0
1 1
C = x10  x20
3 3
so that
( ) ( )
2 1 1 1
x1 = x10 + x20 cos ω− t + x10  x20 cos ω+ t (22)
3 3 3 3
( ) ( )
2 1 2 2
x2 = x10 + x20 cos ω− t  x10  x20 cos ω+ t (23)
3 3 3 3
It can be checked that,
x1  x2 = (x10  x20 ) cos ω+ t
2x1 + x2 = (2x10 + x20 ) cos ω− t
as expected.
Example 2:
As a second example, consider a pair of identical pendulums consisting of a pair of mass-
less rigid rods of length l each at the end of each of which a mass m is attached. The
mid-points of the rods are connected by a spring √of force constant k. In the absence of
g
coupling, each oscillator has a frequency ω = . As before, let us try to solve the
l
problem by force method.

A B
e1 e2

m m