SPA5304 Physical Dynamics Lecture 23-24

David Vegh
(figures by Masaki Shigemori)

6 March 2019

1 Small Oscillations

1.1 Example: Double pendulum

• 2 DoFs


x1 = l1 sin φ1
y1 = l1 cos φ1

x2 = x1 + l2 sin φ2
y2 = y1 + l2 cos φ2

1 1 1 1
T = m1 (ẋ21 + ẏ12 ) + m2 (ẋ22 + ẏ22 ) = m1 l12 φ̇21 + m2 (l12 φ̇21 + l22 φ̇22 + 2l1 l2 φ̇1 φ̇2 cos(φ1 − φ2 ))
2 2 2 2
V = −m1 gy1 − m2 gy2 = −g(l1 (m1 + m2 ) cos φ1 + l2 m2 cos φ2 )

For simplicity, let us set

m1 = m2 = m, l1 = l2 = l
Then,
1
T = ml2 φ̇21 + ml2 φ̇22 + ml2 φ̇1 φ̇2 cos(φ1 − φ2 )
2
V = −2mgl cos φ1 − mgl cos φ2

1
 • Equilibrium positions are found by solving

∂V
∂φ1 = 2mgl sin φ1 

⇒ φ1 = φ2 = 0
∂V 
∂φ2 = mgl sin φ2 

Other solutions, such as φ1 = φ2 = π are clearly unstable.

• The Hessian
∂2V

∂φ21
= 2mgl cos φ1 




  
2 0
2

∂ V
∂φ22 = mgl cos φ 2 ⇒ Vij = mgl

 0 1



∂2V

∂φ1 ∂φ2=0 

• Kinetic energy

1 1
T = ml2 φ̇21 + ml2 φ̇22 + ml2 φ̇1 φ̇2 cos(φ1 − φ2 ) ≡ aij (φ1 , φ2 )φ̇i φ̇j
2 2
From this we can read off aij :

2ml2 ml2 cos(φ1 − φ2 )

 
aij =
ml2 cos(φ1 − φ2 ) ml2

and finally,  
2 2 1
Tij = aij (φ1 = 0, φ2 = 0) = ml
1 1
1
Note: be careful with the 2 when reading off aij !

• Let us now follow the procedure. The secular equation:

det(ω 2 T − V) = 0

    
2 2 2 1 2 0
det ω ml − mgl =0
1 1 0 1

2(ω 2 − g/l) ω2
 
det =0
ω2 ω 2 − g/l

2(ω 2 − g/l)2 − ω 4 = 0

g g2
ω4 − 4 ω2 + 2 2 = 0
l l

g √
ω2 = (2 ± 2)
l

2
 So the normal frequencies are √
2 g
ω(1) = (2 − 2)
l
2 g √
ω(2) = (2 + 2)
l
• Plug these back in to find amplitudes A
(ω 2 Tlj − Vlj )Aj = 0

2(ω 2 − g/l) ω2
  
A1
=0
ω2 ω 2 − g/l A2
Since the determinant of the matrix vanishes, the two equations are not independent. We only need to solve
one of them.
2
• Plug in ω(1)
g √g g √
2 2) −
(2 − A1 + (2 − 2)A2 = 0
l l l
√
2A1 − A2 =0
   
A1 1
=c √ , c∈C
A2 2
This is the first normal mode that we have found.
Solution
η1 = Re[ceiω(1) t ]


√
η2 = Re[ 2ceiω(1) t ]


2
• Plug in ω(2)
g √
g g √
2 2) −
(2 + A1 + (2 + 2)A2 = 0
l l l
√
2A1 + A2 = 0
   
A1 1
=c √ , c∈C
A2 − 2
This is the second normal mode that we have found.
Solution
η1 = Re[ceiω(2) t ]


√
η2 = Re[− 2ceiω(2) t ]


3
1.2 Example: Linear triatomic molecule in 1D

• 3 DoFs
• Modeled with spring forces (e.g. CO2 ).

1 2 2 1 2

T = 2 m(ẋ1 + ẋ3 ) + 2 M ẋ2
1 2 1
V = 2 k(x2 − x1 − l) + 2 k(x3 − x2 − l)2
where l is the rest length of the springs.
Equilibrium positions are found by solving
 ∂V

 0 = ∂x 1
= −k(x2 − x1 − l)




∂V
0 = ∂x 2
= k(x2 − x1 − l) − k(x3 − x2 − l) = k(2x2 − x1 − x3 )




∂V

0 = ∂x = k(x3 − x2 − l)

3

Thus,
x2 − x1 = l, x 3 − x2 = l
Note that only relative positions are determined.

• Hessian:
∂2V ∂2V ∂2V
= k, = 2k, =k
∂x21 ∂x22 ∂x23
∂2V ∂2V ∂2V
= −k, = 0, = −k.
∂x1 ∂x2 ∂x1 ∂x3 ∂x2 ∂x3
Thus,
−1 0
 
1
Vij = k  −1 2 −1 
0 −1 1
The matrix is independent of the x’s.

 
m 0 0
Tij =  0 M 0 
0 0 m

• Let us now again follow the procedure. The secular equation:

det(ω 2 T − V) = 0

4
 mω 2 − k
 
k 0
2
det  k M ω − 2k k =0
2
0 k mω − k
(mω − k) (M ω − 2k) − 2k (mω 2 − k) = 0
2 2 2 2

• Solutions:
2
ω(0) =0
2 k
ω(1) =
m
 
2 k 2m
ω(2) = 1+
m M

• The amplitudes:
mω 2 − k
 
k 0 A1
 k M ω 2 − 2k k   A2  = 0
2
0 k mω − k A3

2
• Plug in ω(0) =0

−1
  
1 0 A1
k  −1 2 −1   A2  = 0
0 −1 1 A3
Thus,
   
A1 1
 A2  = c  1 
A3 1

η1 = η2 = η3 = Re[cei×0×t ] = const.

2 k
• Plug in ω(1) = m

  
0 1 0 A1
M
m −2
k 1 1   A2  = 0
0 1 0 A3
Thus,
   
A1 1
 A2  = c  0 
A3 −1
√k
η1 = −η3 = Re[cei mt ], η2 = 0

5
 k 2m
2


• Plug in ω(1) = m 1+ M

2m


−k
  
k 1+ M k 0 A1
k 2m


 k Mm 1+ k   A2  = 0
M
2m


0 k k 1+ M −k A3
Thus,
   
A1 1
 A2  = c  − 2m 
M
A3 1

2m
q
η1 = η3 = Re[ce
i k
m (1+ 2m
M )t
], η2 = − η1
M

If m  M , then the middle one doesn’t move.

1.3 Example: Pendulum with moving suspension point

• 2 DoFs

 
xM = x xm = x + l sin φ
yM = 0 ym = l cos φ

1 2 2
) + 12 m(ẋ2m + ẏm
2 M 2 m 2
+ l2 φ̇2 + 2lẋφ̇ cos φ)

T = 2 M (ẋM + ẏM )= 2 ẋ + 2 (ẋ
V = −mgym = −mgl cos φ

6
 • Equilibrium positions
∂V ∂V
= 0, = mgl sin φ = 0
∂x ∂φ
This gives φ = 0 and x =arbitrary.

• The T and V matrices:

 
M + m ml
T =
ml ml2
 
0 0
V=
0 mgl

• secular equation

ω 2 (M + m) ω 2 ml
 
det =0
ω 2 ml ω 2 ml2 − mgl

• The solutions are

2
ω(0) =0
This is again an overall translation.

2 g m
ω(1) = 1+
l M

ω 2 (M + m) ω 2 ml
  
A1
2 =0
ω ml ω ml2 − mgl
2
A2

ml
A1 = − A2
M +m
 
ml
q
i g 1+ m t
x(t) = xarbitrary − Re A2 e l ( M )
M +m
 q 
i gl (1+ M
m
) t
φ(t) = Re A2 e

If m  M , then x = xarbitrary : the problem reduces to a simple pendulum.

7
2 Hamiltonian mechanics
So far w ehave been using the Lagrangian formalism of mechanics, in which the system is described by
generalized coordinates ~q and generalized velocities ~q˙.
We now want to introduce the Hamiltonian formalism, in which the system is described by generalized
coordinates ~q and generalized momenta p~.
This formalism is even more general: it treats ~q and p~ on equal footing.

2.1 Legendre transformation

• It is a procedure to go from (q, q̇) to (q, p).

• Very often used in thermodynamics.
∂L
Consider one degree of freedom. Let us eliminate q̇ in terms of p ≡ ∂ q̇ . This gives a function

q̇ = q̇(q, p, t)

Now go from the Lagrangian L(q, q̇, t) to the Hamiltonian H(q, p, t) by

H = pq̇ − L

The right hand side involves q̇ which we need to reexpress in terms of p. Then we get H = H(q, p, t).

2.2 Hamilton’s equations

Consider a small change in H
dH = d(pq̇ − L) = dpq̇ + pdq̇ − dL
Here
∂L ∂L ∂L
dL = dq + dq̇ + dt
∂q ∂ q̇ ∂t
So  
∂L ∂L ∂L ∂L ∂L ∂L
dH = dpq̇ + pdq̇ − dq − dq̇ − dt = q̇dp + p − dq̇ − dq − dt
∂q ∂ q̇ ∂t ∂ q̇ ∂q ∂t
| {z }
0

Thus,
∂L ∂L
dH = q̇dp − dq − dt
∂q ∂t
On the other hand, H = H(q, p, t) gives

∂H ∂H ∂H
dH = dq + dp + dt
∂q ∂p ∂t

8
Since q, p, t are independent, comparing the coefficients gives
∂H ∂L ∂H ∂H ∂L
=− , = q̇, =− .
∂q ∂q ∂p ∂t ∂t
These simply followed from the definition of p and H.
∂L
Let us now also use the Euler-Lagrange equation ṗ = ∂q to get Hamilton’s equations

∂H
q̇ =
∂p

and
∂H
ṗ = −
∂q

• Note that q and p appear symmetrically (up to a sign).

• These are first-order differential equations for two variables, in contrast to the Euler-Lagrange equation
which is a second-order equation for a single variable q.
• The 2d space of (q, p) is called the phase space. (Previously we have defined the configuration space
which is the 1d space of q.)
• The Hamiltonian is the energy of the system: the Noether invariant associated to time-translation.
• Given the initial position (q, p) in pahse space at t = t0 , H(q, p) tells us how to evolve in t:
   ∂H 
q̇ ∂p
=
ṗ − ∂H
∂q

For instance, if H = 12 (p2 + q 2 ), then    

q̇ p
=
ṗ −q