July 10, 2018 12:25 Competitive Physics 9.61in x 6.

69in b3146-ch12 page 623

Chapter 12

Lagrangian Mechanics

In the past few chapters, we have examined the quintessence of Newtonian

mechanics. However, in this chapter, we shall adopt a completely different
approach to dynamics via a general principle known as Hamilton’s principle.
In analytical mechanics, dynamics problems are reduced to writing down a
function known as the Lagrangian, performing a few differentiations and then
solving the ensuing differential equations. In fact, Lagrange prided himself
on not including a single diagram in his treatise! We will delve directly into
the heart of this topic, so please bear with the abrupt jump.

12.1 Action and Hamilton’s Principle

The action S of a system along an evolutionary path q(t) between a starting
state at time t1 and an ending state at time t2 is defined as
 t2
S= L(q1 , q2 , . . . , qn , q̇1 , q̇2 , . . . , q̇n , t)dt. (12.1)
t1

L is the Lagrangian and is a function given by

L = T − V, (12.2)
where T and V are the instantaneous kinetic and potential energies of
the system. This seems like an odd combination but some analogies will
be revealed soon. q1 , q2 ,. . . , qn are generalized coordinates which uniquely
describe the state of a system. As their nomenclature implies, they are indeed
quite all-encompassing as they can represent translational coordinates such
as x, angular coordinates such as θ and even time-varying coordinates, such
as those of a rotating coordinate system. For brevity, they are collectively
represented as q which can be seen as an n-dimensional vector. Note that
this vector has little physical meaning — evident from the fact that even

623
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 624

624 Competitive Physics: Mechanics and Waves

the units of each qi can be different. q̇1 , q̇2 , . . . , q̇n are the time derivatives
of the generalized coordinates and are known as generalized velocities —
collectively denoted by q̇.
Next, we move on to the crucial pillar of Lagrangian mechanics.
Hamilton’s principle states that the path q(t) taken by a system between
times t1 and t2 is one that results in a stationary value of the action S.
Notice that S is dependent on the entire function q(t) and is hence, known
as a functional. To determine the condition that results in a stationary value
of S, we turn to the calculus of variations.

12.2 Calculus of Variations

Firstly, we quantify what a stationary value of a functional actually means.
Suppose that q 0 (t) is the desired path that results in a stationary value of
the functional
 t2
S= L[q(t)]dt.
t1

Then, functions q(t)’s in the immediate vicinity of q 0 (t) result in values of S

adopting deviations of second-order and above from the value of S0 produced
by q 0 (t) for any possible deviation. Let us consider the case of a functional
that involves only a single coordinate x(t), its derivative ẋ and t (which is
just a variable that is not necessarily time).
 t2
S[x(t)] = L(x, ẋ, t)dt.
t1

Remember that the end points x(t1 ) and x(t2 ) are fixed at certain values,
as we have predetermined initial and final states.1 Usually if f (x) were a
df
function of x, we would determine points that yield dx = 0 to find stationary
points. Since it is impossible to differentiate a variable with respect to an
entire function, consider the substitution of

x(t) = x0 (t) + αη(t)

ẋ(t) = ẋ0 (t) + αη̇(t),

1
Note that the final state at time t2 is not really well-specified. For example, if you
are analyzing the motion of a simple projectile, there is no clear point in time where the
motion ends. However, Hamilton’s principle states that whatever time you choose as the
final state, the actual path taken by the ball between time t1 and this final time extremizes
the action.
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 625

Lagrangian Mechanics 625

where α is a constant and η(t) is any arbitrary function with η(t1 ) =

η(t2 ) = 0 such that the endpoints of the path are kept fixed at x(t1 ) = x0 (t1 )
and x(t2 ) = x0 (t2 ). Together, αη is known as a variation of x(t), denoted by
δx. Then, S is now a function of α and we can consider its derivative with
respect to α. In order for S to be stationary, we require ∂S∂α = 0.
 t2
∂S ∂
= Ldt.
∂α ∂α t1
As L is not integrated with respect to α, we can bring the partial derivative
into the integral such that it becomes a total derivative.
 t2
∂S dL
= dt
∂α t1 dα
 t2  t2
∂L ∂x ∂L ∂ ẋ
= dt + dt,
t1 ∂x ∂α t1 ∂ ẋ ∂α
t
where the term t12 ∂L ∂t
∂t ∂α dt = 0 has been excluded as α is a time-independent
constant. Now, note that
∂x
= η(t)
∂α
∂ ẋ
= η̇(t).
∂α
Then,
 t2  t2
∂S ∂L ∂L
= ηdt + η̇dt.
∂α t1 ∂x t1 ∂ ẋ
Integrating the second term by parts,
 t2  t2  
∂S ∂L ∂L t2 d ∂L
= ηdt + η − ηdt.
∂α t1 ∂x ∂ ẋ t1 t1 dt ∂ ẋ
Since η(t1 ) = η(t2 ) = 0,
 t2  
∂L d ∂L
− ηdt = 0
t1 ∂x dt ∂x
∂S
in order for ∂α = 0. By the fundamental lemma of the calculus of
variations, if
 t2
M (x, t)η(t)dt = 0
t1
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 626

626 Competitive Physics: Mechanics and Waves

for all η(t)’s that are twice continuously differentiable, then

M (x, t) = 0.
Applied to the situation at hand where η is an arbitrary function, it implies
that
 
d ∂L ∂L
=
dt ∂ ẋ ∂x
for S to be extremized. This is known as the Euler-Lagrange equation (we will
refer to this as the E-L equation for the sake of convenience). Another way
to see why this must be true without the use of the fundamental lemma is to
imagine choosing a η(t) such that it has the same sign at all times with ( ∂L∂x −
d ∂L
dt ∂x ) (it could be this expression scaled down by a certain factor). If the
latter expression were not zero, the integrand would be positive at all times
and, hence, cause the integral to be positive — leading to a contradiction.
For functions that depend on n coordinates q1 , q2 ,. . . , qn , the above can be
easily generalized by letting
qi (t) = qi0 (t) + αi ηi (t),
where qi0 (t) is the correct function for coordinate qi and αi ηi (t) is a variation
along this coordinate. Then one can take
∂S
=0
∂αi
for all 1 ≤ i ≤ n. Since adding a variation to the coordinates other than
qk does not result in a change in the partial derivative of the Lagrangian
with respect to αk , we can use the previous result to deduce that the overall
condition for S(q(t)) to undertake a stationary value is
 
d ∂L ∂L
=
dt ∂ q̇i ∂qi
for all 1 ≤ i ≤ n. We can express this in a rather succinct form
 
d ∂L ∂L
= ,
dt ∂ q̇ ∂q
where a derivative of a function y with respect to q means
⎛ ∂y ⎞
∂q
⎜ ∂y1 ⎟
∂y ⎜ ∂q2 ⎟
=⎜ .. ⎟
⎟.
∂q ⎜⎝ . ⎠
∂y
∂qn
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 627

Lagrangian Mechanics 627

12.3 Equations of Motion

Now that we have understood the requirements for a function to extremize
a functional, let us return to our original question regarding the action that
depends on the path taken by a system
 t2
S= L(q, q̇, t)dt.
t1

To attain a stationary value for S, the E-L equations require that

 
d ∂L ∂L
= . (12.3)
dt ∂ q̇ ∂q
The desired q(t) that extremizes the action is known as an extremal. Let
us see how the Newtonian equations of motion in Cartesian coordinates can
be “recovered” in the case of a single particle. In Cartesian coordinates,
1
L=T −V = m(ẋ2 + ẏ 2 + ż 2 ) − V.
2
Then,
 
d ∂L ∂L
=
dt ∂ q̇ ∂q
∂V
=⇒ mẍ = − ,
∂x
∂V
mÿ = − ,
∂y
∂V
mz̈ = −
,
∂z
∂V
=⇒ mr̈ = − .
∂r
Notice that − ∂V
∂r is simply the net conservative external force on the particle.
Therefore, we retrieve the equation
F = ma.

For a system of particles, our generalized coordinates could be the union

of all (xi , yi , zi ) for each particle i and the F = ma equation would hold
for each particle. Newton’s second law is therefore equivalent to Hamilton’s
principle applied in Cartesian coordinates!
From the above example, it should be clear that there are analogies
between the Newtonian momentum and forces with the terms in the E-L
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 628

628 Competitive Physics: Mechanics and Waves

equations. pi ≡ ∂∂Lq˙i is known as the canonical momentum conjugate

to coordinate qi , while Qi ≡ ∂q ∂L
i
is known as the generalized force
associated with coordinate qi .
What about non-conservative forces? The fact is that from experience,
all of the fundamental forces that we know today can be formulated in terms
of potentials. Non-conservative forces, such as the viscous drag force which
originates from the bombardment of fluid particles on an object, are not
fundamental forces. Therefore, the Lagrangian formulation is still general at
the most fundamental level. However, if we utterly insist on including non-
conservative forces, we can “cheat” a little and add an appropriate expression
on the right-hand side of Eq. (12.3) to represent a non-conservative force.

12.3.1 Different Coordinate Systems

The Lagrangian may not seem particularly enlightening right now but its
utility really shines when it comes to different coordinate systems. If we
were to use Newton’s laws, which are only valid in an inertial frame, in a
rotating frame for example, we would have to modify the laws of motion
to include “fictitious forces.” However, by the Lagrangian formulation, the
fundamental Hamilton’s principle holds in all frames. Concomitantly, the
E-L equations hold for various coordinate systems — we simply have to
express T and V in terms of the different coordinates. This can be seen in
two ways: physically and mathematically. Physically, the path of an extremal
should not depend on the frame of reference. For example, the shortest path
between two points is a straight line, regardless of the frame it is viewed
from. Mathematically, we can show that if the E-L equations hold for a set
of n coordinates x(t) (and we already know that it holds for the Cartesian
coordinate system), it must hold for the same Lagrangian in another set of
N coordinates2 q(t) that is given by

qi (t) = fi (x, t) (12.4)

=⇒ q̇i (t) = f˙i (x, ẋ, t)

for all 1 ≤ i ≤ N . That is, each transformed coordinate is a function of the

previous coordinates and time t. The following proof is just a formality and
can be skipped by readers who simply want to learn how the Lagrangian can
be applied.

2
N is not necessarily equal to n.
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 629

Lagrangian Mechanics 629

Returning to the problem, we wish to prove that

 
d ∂L ∂L
=
dt ∂ q̇ ∂q

given that
 
d ∂L ∂L
= .
dt ∂ ẋ ∂x

Consider
n n
∂L ∂L ∂xi ∂L ∂ ẋi
= · + ·
∂qk ∂xi ∂qk ∂ ẋi ∂qk
i=1 i=1
n   n
d ∂L ∂xi ∂L ∂ ẋi
= · + · .
dt ∂ ẋi ∂qk ∂ ẋi ∂qk
i=1 i=1

For the final term on the right, observe that

⎛ ⎞
N
∂ ẋi ∂ ⎝ ∂xi ∂xi ⎠
= · q̇j +
∂qk ∂qk ∂qj ∂t
j=1
N    
∂ ∂xi ∂ ∂xi
= · q̇j +
∂qj ∂qk ∂t ∂qk
j=1
 
d ∂xi
= .
dt ∂qk

Note that we have used the fact that partial derivatives are interchangeable
in writing the second inequality. Therefore,
n   n  
∂L d ∂L ∂xi ∂L d ∂xi
= · + ·
∂qk dt ∂ ẋi ∂qk ∂ ẋi dt ∂qk
i=1 i=1
 n 
d ∂L ∂xi
= · .
dt ∂ ẋi ∂qk
i=1

d ∂L
As our objective is to show that this is equal to dt ( ∂ q̇k ), we simply have to
prove that

∂xi ∂ ẋi
= .
∂qk ∂ q̇k
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 630

630 Competitive Physics: Mechanics and Waves

Observe that
N
∂xi ∂xi
ẋi = · q̇j +
∂qj ∂t
j=1

∂ ẋi ∂xi
=⇒ = .
∂ q̇k ∂qk

Thus,
 n
  
∂L d ∂L ∂ ẋi d ∂L
= · =
∂qk dt ∂ ẋi ∂ q̇k dt ∂qk
i=1

for all 1 ≤ k ≤ N . This implies that

 
d ∂L ∂L
= .
dt ∂ q̇ ∂q

If the E-L equations are valid for one set of coordinates (for example, Carte-
sian coordinates), they are also valid for all other coordinates of the form
given by Eq. (12.4). A direct corollary of this is that frames related by
Galilean transformations are equivalent as the coordinate transformations
are of the form qi = qi + vi t for some constant vi .

Polar Coordinates

The Lagrangian really distinguishes itself in finding the F = ma equa-

tions of a single particle in polar coordinates — whose transformations
from Cartesian coordinates evidently obey Eq. (12.4). Our objective is
to express T in terms of polar coordinates. To do so, we have to deter-
mine | dr 2
dt | where r is the position vector of the particle. This can be
written as
 2
 dr  2 2 2
  = dr · dr = (dr1 ) + (dr2 ) + (dr3 ) ,
 dt  (dt)2 (dt)2 (dt)2 (dt)2

where dr1 , dr2 and dr3 are three perpendicular infinitesimal length segments
(the usual infinitesimal quantities we integrate over). Let us apply this to an
example.
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 631

Lagrangian Mechanics 631

Spherical Coordinates

The infinitesimal length segments in spherical coordinates are dr1 = dr,

dr2 = r sin θdφ and dr3 = rdθ. Hence, the Lagrangian is
1  
L = m ṙ 2 + r 2 sin2 θ φ̇2 + r 2 θ̇ 2 − V.
2
We wish to evaluate how the change of these coordinates relates to
the conservative forces along the unit vectors in spherical coordinates3
which are
⎛ ⎞ ⎛ ∂V ⎞ ⎛ ⎞
Fr − ∂r1 − ∂V∂r
⎜ ⎟ ⎜ ∂V ⎟ ⎜ ⎟
⎝Fφ ⎠ = ⎝− ∂r2 ⎠ = ⎝− r sin ∂V
θ∂φ ⎠ . (12.5)
Fθ − ∂V
∂r3 − ∂V
r∂θ

Applying the E-L equations ( ∂L

∂q =
d ∂L
dt ( ∂ q̇ )) with respect to generalized coor-
dinates r, φ and θ respectively,
∂V d
− + mr sin2 θ φ̇2 + mr θ̇ 2 = (mṙ) = mr̈,
∂r dt
∂V d  2 2 
− = mr sin θ φ̇ = 2mr sin2 θ ṙφ̇ + 2mr 2 sin θ cos θ φ̇θ̇ + mr 2 sin2 θ φ̈,
∂φ dt
∂V d  2 
− + mr 2 sin θ cos θ φ̇2 = mr θ̇ = 2mr ṙθ̇ + mr 2 θ̈.
∂θ dt
Dividing the second and third equations by r sin θ and r respectively and
applying Eq. (12.5),
Fr = m(r̈ − r sin2 θ φ̇2 − r θ̇ 2 ),
Fφ = m(2 sin θ ṙφ̇ + 2r cos θ φ̇θ̇ + r sin θ φ̈),
Fθ = m(2ṙ θ̇ − r sin θ cos θ φ̇2 + r θ̈).
We have derived such a complicated version of F = ma without the use of
any vectors! This is the slickness of the Lagrangian.

Rotating Coordinate Systems

Consider a rotating coordinate system X, Y , Z that is rotating at a constant

anti-clockwise angular velocity ω around the z-axis, of a fixed coordinate

3
This is just the negative gradient of V in spherical coordinates (the definition of a
conservative force).
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 632

632 Competitive Physics: Mechanics and Waves

frame x, y, z which coincides with the rotating frame at t = 0. Then, the x,

y, z coordinates can be expressed in terms of X, Y and Z as

x = X cos ωt − Y sin ωt,

y = X sin ωt + Y cos ωt,
z = Z.

The first
 two equations  can
 be swiftly obtained from applying the rotation
matrix sin ωt cos ωt to X
cos ωt − sin ωt
Y . In doing so, be cautious that the x and y-axes
rotate at ω clockwise relative to the X and Y -axes.

ẋ = Ẋ cos ωt − ωX sin ωt − Ẏ sin ωt − ωY cos ωt,

ẏ = Ẋ sin ωt + ωX cos ωt + Ẏ cos ωt − ωY sin ωt,
ż = Ż.

The Lagrangian in terms of the fixed coordinate frame is

1  
L = m ẋ2 + ẏ 2 + ż 2 − V.
2
This can be expressed in terms of the rotating coordinates as

1  
L = m Ẋ 2 + Ẏ 2 + Ż 2 + 2ωX Ẏ − 2ω ẊY + ω 2 (X 2 + Y 2 ) − V.
2

Applying the E-L equations ( ∂L

∂q =
d ∂L
dt ( ∂ q̇ )) with respect to X, Y and Z yields

∂V
mω Ẏ − + mω 2 X = mẌ − mω Ẏ ,
∂X
∂V
−mω Ẋ − + mω 2 Y = mŸ + mω Ẋ,
∂Y
∂V
− = mZ̈.
∂Z
∂V
Note that (− ∂X , − ∂V
∂Y , − ∂Z ), where the unit vectors are along the axes of the
∂V

rotating frame, represents the real conservative force F which is invariant

across all frames (as it is a physical vector). Let r be the position vector
of the particle which is independent of the frame of reference. Furthermore,
denote v rot = (Ẋ, Ẏ , Ż) and arot = (Ẍ, Ÿ , Z̈) as the particle’s velocity and
acceleration, as observed in the rotating frame, respectively. Summarizing
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 633

Lagrangian Mechanics 633

the above equations in terms of vectors would yield

marot = F − 2mω × v rot − mω × (ω × r).
Keep in mind that ω is the angular velocity of the rotating frame with
respect to the lab frame and points in the positive z-direction.

12.3.2 Additional Time Derivatives

The Lagrangian describing the evolution of a system is in fact not unique.

Try adding a total time derivative of a function of generalized coordinates
and time df (q,t)
dt to the original Lagrangian of a system. Note that df dt could
simply be a constant too. The action between times t1 and t2 becomes
 t2    t2
 df (q, t)
S = L+ dt = Ldt + f (q2 , t2 ) − f (q1 , t1 ),
t1 dt t1

where q 2 and q 1 are the final and initial generalized coordinates respectively.
Those additional terms on the right are fixed and do not vary — the equa-
tions of motion of this new system thus do not differ from the original one.
This implies that total time derivatives can simply be discarded from the
Lagrangian of a system as they are inconsequential — a neat trick in tidying
up the Lagrangian.

Problem: A particle of mass m is attached to a wall via a massless spring

with spring constant k. In this one-dimensional problem, the x-coordinate
of the wall is constrained to obey X(t) = A cos ωt where A > 0 is a constant
while the particle and the origin lie on opposite sides of the wall. Let x
denote the additional displacement of the particle from the wall, beyond the
rest length of the spring, in the positive x-direction. Show that
ẍ + ω02 x = B cos ωt,

k
where ω0 = m is the natural frequency of this system and B is a constant.
The velocity of the particle is Ẋ + ẋ = ẋ − Aω sin ωt. The Lagrangian of
the particle is thus
1 1
L = m(ẋ − Aω sin ωt)2 − kx2 .
2 2
We can discard the 12 mA2 ω 2 sin2 ωt term since it is a total time derivative,
to obtain
1 1
L = mẋ2 − mAω sin ωtẋ − kx2 .
2 2
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 634

634 Competitive Physics: Mechanics and Waves

 d  ∂L  ∂L

Applying the E-L equation dt ∂ ẋ = ∂x ,

mẍ − mAω 2 cos ωt = −kx

ẍ + ω02 x = Aω 2 cos ωt,

as required by the problem. Observe that this equation of motion describes

a forced oscillation whose general solution can be determined by methods
introduced in the chapter on oscillations.

12.4 Systems with Constraints

In many mechanics problems, there are constraints imposed on a system. For
example, a falling object cannot penetrate the ground and the length of a
rigid rod remains constant. Constraints can be holonomic or non-holonomic.
A holonomic constraint can be expressed algebraically in the following form:

f (q, t) = 0.

An example of a holonomic constraint would be that of rigid bodies —

the preservation of relative distances can be expressed in terms of the gen-
eralized coordinates. Constraints that cannot fulfil the above criterion are
termed non-holonomic — they usually involve time derivatives of generalized
coordinates, and inequalities instead of equalities. Note that the constraint
on a rigid body to roll without slipping along a single direction, despite its
appearance, is holonomic as the time derivatives can be easily removed via
integration.
The Lagrangian formulation can only be conveniently applied to holo-
nomic systems. The two general methods in doing so will be elaborated in
the following section.
But first, there is an often simpler approach to determine whether a sys-
tem is holonomic or non-holonomic, instead of expressing the constraints
algebraically. Define the degrees of freedom (DOFs) of a system as the num-
ber of coordinates of a system that can be varied independently of all other
coordinates (i.e. keep the other coordinates fixed). Let there be N particles
in a system. Then, an unconstrained system has 3N DOFs and requires 3N
coordinates to specify a state. In holonomic systems, the DOFs of a system
are equal to the number of independent coordinates required to uniquely
specify a state of a system. This is due to m holonomic constraints reducing
the DOFs by m. From a mathematical perspective, holonomic constraints
furnish m additional equations in terms of the generalized coordinates and
time such that the 3N coordinates can be expressed in terms of (3N − m)
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 635

Lagrangian Mechanics 635

independent ones. A non-holonomic system, naturally, does not fulfil the cri-
terion that the DOFs are equal to the number of independent coordinates
needed to uniquely define a state.

Example of a Non-holonomic System

One might wonder how such an intuitive requirement can be violated.

Consider the classic example of a sphere on a two-dimensional table. It is not
allowed to translate or rotate about an axis in the vertical direction. Further-
more, it is constrained such that it cannot slip. Then, the sphere only has
2 DOFs — rolling in the x and y directions along the table. Now you might
think that one simply needs two coordinates such as the x and y-coordinates
of the center of the sphere to specify its state. Let us perform an experiment
to convince you otherwise. Place the sphere at the origin initially and paste
a sticker on top of it. Then, consider the following series of movements.

(1) Let the circumference of the sphere be C. Then, roll the sphere to coor-
dinate (C, 0). The sticker is on top of the sphere at this juncture. Then,
roll it to coordinate (C, C). The sticker is still on top.
(2) Roll the sphere directly along the diagonal from the origin to (C, C).
The sticker is no longer on top of the sphere!

In this system, 4 coordinates are in fact necessary to describe a configuration

of the sphere! The non-holonomic constraint originates from the requirement
of rolling without slipping in two dimensions such that the velocities in two
directions are inextricably coupled in an equation which cannot be trivially
integrated to remove the time derivatives. Anyway, non-holonomic systems
will not be considered in this chapter and our analysis of such systems ends
here.

Hamilton’s Principle in Holonomic Systems

We cannot directly apply the previous results — namely the E-L equations
to holonomic systems as the variations may not be consistent with the con-
straints. However, Hamilton’s principle still holds and our modified objective
is to determine a path that extremizes the action, while obeying the con-
straints. There are then two approaches that we can take.
Firstly, we can use m constraint equations to solve for (3N − m) inde-
pendent coordinates which can be used to define the state of a system.
Often, we can even directly define coordinates that satisfy the constraints.
The variations then naturally obey the constraints — the E-L equations can
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 636

636 Competitive Physics: Mechanics and Waves

consequently be directly applied to this judiciously chosen set of indepen-

dent coordinates. The greatest benefit of this approach is that the forces of
constraint, such as normal forces, are completely ignored. As this method
is rather straightforward, let us consider two examples to summarize the
concepts so far.

Problem: A block of mass m lies on a frictionless inclined plane of mass M

and angle of inclination θ. If there is no friction between the plane and the
ground, determine the acceleration of the plane.
Define x to be the horizontal coordinate of the highest tip of the plane.
Take the vertical coordinate to be zero at this tip. Then, define s to be the
distance between the mass m and this tip (we assume that the gradient of
the slope is negative in the positive x-direction). This definition ensures that
m satisfies the constraint of remaining on the plane. The coordinates of m
are thus (x + s cos θ, −s sin θ). The velocity of m is (ẋ + ṡ cos θ, −ṡ sin θ). The
Lagrangian of the combined system comprising m and M is

L=T −V
1 1  
= M ẋ2 + m (ẋ + ṡ cos θ)2 + ṡ2 sin2 θ + mgs sin θ
2 2
1 1
= (M + m)ẋ2 + mṡ2 + mẋṡ cos θ + mgs sin θ.
2 2
d ∂L ∂L
Applying the E-L equation ( dt ( ∂ q̇ ) = ∂q ) with respect to x and s,

(M + m)ẍ + ms̈ cos θ = 0,

ms̈ + mẍ cos θ = mg sin θ.

Solving for ẍ,

mg sin θ cos θ
ẍ = − .
M + m sin2 θ
Problem: Consider a mass m undergoing uniform circular motion at angular
velocity ω in a fixed, hollow inverted cone of half angle θ. Determine the
equations of motion of the particle and hence find l0 , the distance of the
particle from the apex such that this uniform circular motion can occur.
Then, show that this equilibrium is stable and determine the frequency of
small oscillations Ω about l0 when the particle is slightly perturbed in the
direction joining the apex and the particle.
Let the distance of the particle to the apex be l and the azimuthal angular
coordinate of the particle be φ. Then, l sin θ φ̇ is the azimuthal velocity while
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 637

Lagrangian Mechanics 637

l̇ is the other component of velocity along the surface of the cone. Hence,
the Lagrangian of this particle is
1
L = m(l̇2 + l2 sin2 θ φ̇2 ) − mgl cos θ.
2
Applying the E-L equation with respect to l and φ,

m¨l = ml sin2 θ φ̇2 − mg cos θ,

d
(ml2 sin2 θ φ̇) = 0 =⇒ ml2 sin2 θ φ̇ = L,
dt
for some constant L. The Newtonian interpretation of this is simply the
component of the angular momentum of the particle along the symmetrical
axis of the cone. The above are the equations of motion for the particle. To
determine l0 , we set ¨l = 0 and φ̇ = ω. Then,
g cos θ
l0 = .
ω 2 sin2 θ
Now, consider a perturbation from l0 such that l = l0 + ε for some infinitesi-
mal deviation ε. We first rewrite the first equation of motion strictly in terms
of a single variable l with the help of L:
L2
m¨l = − mg cos θ.
ml3 sin2 θ
Keep in mind that
L2
− mg cos θ = 0, (12.6)
ml03 sin2 θ
as this will be used later. We then substitute l = l0 + ε into the previous
equation to get
L2
mε̈ =  3 − mg cos θ.
ml03 sin2 θ 1 + ε
l0

Using the first-order binomial expansion (1 + x)n ≈ 1 + nx,

 
L2 ε
mε̈ = 1−3 − mg cos θ.
ml03 sin2 θ l0
Applying Eq. (12.6) yields
3L2
ε̈ = − ε.
m2 l04 sin2 θ
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 638

638 Competitive Physics: Mechanics and Waves

Note that such cancellations always occur in perturbation problems. The

equilibrium is stable as the particle will undergo simple harmonic motion
about l0 , after a slight deviation, at a frequency
√
3 |L|
Ω= .
ml02 sin θ
Since L = ml02 sin2 θω,
√
Ω= 3 sin θ|ω|.

Lagrange Multipliers

The second method entails directly finding variations that adhere to the
constraints without changing the generalized coordinates. Mathematically,
if there are m holonomic constraints and n generalized coordinates, we wish
to extremize the functional
 t2
S= L(q, q̇, t)dt
t1

under m constraints, with the ith constraint being

fi (q, t) = 0.

Taking the total derivative of the above equation, we obtain the relationship
between the variations of generalized coordinates that is imposed by the ith
constraint.
∂fi ∂fi ∂fi
δq1 + δq2 + · · · + δqn = 0. (12.7)
∂q1 ∂q2 ∂qn
Now, consider the most general variation of the action which depends on
q and q̇ (t cannot be varied):
 t2 n  
∂L ∂L
δS = δqj + δq̇j dt.
t1 ∂qj ∂ q̇j
j=1

The δq̇j term for all j can be integrated by parts, while keeping in mind that
δqj (t1 ) = δqj (t2 ) = 0 (as the end points must be fixed), to obtain
 t2 n   
∂L d ∂L
δS = − δqj dt.
t1 ∂qj dt ∂ q̇j
j=1

Since the right-hand side of Eq. (12.7) is zero, we can add to the above equa-
tion Eq. (12.7), for each 1 ≤ i ≤ m, multiplied by an arbitrary function of
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 639

Lagrangian Mechanics 639

time λi (t) without affecting the value of δS. These variables {λ1 , λ2 , . . . , λm },
collectively referred to as λ, are known as Lagrange Multipliers.
 t2 n    m

∂L d ∂L ∂fi
δS = − + λi (t) δqj dt.
t1 ∂qj dt ∂ q̇j ∂qj
j=1 i=1

We can always find a λ such that the terms in the brackets equate to zero
for j = {1, 2, . . . , m}. Effectively, the first m variations, {δq1 , δq2 , . . . , δqm },
are expressed in terms of the other (n − m) variations. Then,
  m
∂L d ∂L ∂fi
− + λi (t) =0
∂qj dt ∂ q̇j ∂qj
i=1

for 1 ≤ j ≤ m. Furthermore, the variations {δqm+1 , δqm+2 , . . . , δqn } have

been decoupled and thus can be varied independently. This is because the
first m variations have been expressed in terms of them in a manner such
that given {δqm+1 , δqm+2 , . . . , δqn }, one can always tweak {δq1 , δq2 , . . . , δqm }
such that they collectively satisfy the constraints. Since the last (n − m)
variations are free to take on any functions, the term in the brackets for
each m + 1 ≤ j ≤ n must also be zero for S to be stationary (by the
fundamental lemma). All-in-all,
  m
d ∂L ∂L ∂fi
− = λi (t) (12.8)
dt ∂ q̇j ∂qj ∂qj
i=1

for all 1 ≤ j ≤ n! These are the modified E-L equations for systems with
holonomic constraints. There is a neat way to recapitulate the results derived.
In order to solve an extremization problem with holomonic constraints,
we introduce m additional coordinates, λi (t). Then, the above problem can
be solved by finding the stationary values of the new action
 t2  m


S = L(q, q̇, t) + λi (t)fi (q, t) dt,
t1 i=1

with (n + m) generalized coordinates {q, λ}. The modified Lagrangian is

m

L =L+ λi fi .
i=1

To see why this is coherent with the previous results, apply the E-L equations
with respect to a coordinate λi . Since L is independent of λ̇i ,

fi (q, t) = 0.
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 640

640 Competitive Physics: Mechanics and Waves

It can be seen that the constraints are enforced by introducing these addi-
tional terms. Therefore, we can vary q freely in a certain sense now as the
extremals of S  will definitely obey the constraints. Furthermore, it is obvious
that if S  is extremized, then S is also extremized (in a legal manner consis-
 
tent with the constraints) as m i=1 λi (t)fi (q, t) = 0 when S is extremized.
Applying the E-L equations to L , with respect to qj , yields
  m
d ∂L ∂L ∂fi
− = λi .
dt ∂ q̇j ∂qj ∂qj
i=1

Therefore, this method of modifying the Lagrangian is equivalent to the

previous discussion and is valid. There are now (n + m) E-L equations to
solve for (n + m) variables {q, λ}.
Finally, the term on the right-hand side of Eq. (12.8) has a physical
meaning. If there were no constraints, the right-hand side should be zero.
Then, this additional term must be due to the generalized force exerted by
the constraints in this case!
m
∂fi
Qj, constraint = λi . (12.9)
∂qj
i=1

If qj is a particular translational coordinate, Qj, constraint corresponds to the

forces of constraint, such as static friction and normal force, along that coor-
dinate! If qj is an angular coordinate, one would have to modify Eq. (12.8)
∂L
such that the denominator of the partial derivative ∂q j
becomes an infinites-
imal length segment (similar to our derivation of F = ma in spherical coor-
dinates) to obtain that particular component of the forces of constraint.

Problem: Let us consider the trivial example of calculating the tension in a

simple pendulum. Let the fixed end of the string be at the origin and let the
coordinates of the bob be (r, θ) where θ is the anti-clockwise angle between
the string and the negative y-axis, that is pointing vertically downwards.
There is a holonomic constraint r = l where l is the length of the inextensible
string. Then, the modified Lagrangian (we will still use L to represent it) is
1  
L = m ṙ 2 + r 2 θ̇ 2 + mgr cos θ + λ(r − l).
2
The E-L equation with respect to r is

mr̈ = mr θ̇ 2 + mg cos θ + λ.
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 641

Lagrangian Mechanics 641

Now applying the constraint r = l, r̈ = 0,

λ = −mr θ̇ 2 − mg cos θ.

The force of constraint along the r direction, which is simply the tension
force on the bob due to the string in this case, is

∂(r − l)
T = Qr, constraint = λ = λ = −mr θ̇ 2 − mg cos θ,
∂r
where the negative sign indicates that the direction of the force on the bob
is radially inwards.

Problem: A point mass m initially rests on top of an immobile, frictionless

circle of radius R. If it is given a slight displacement, determine the angle
from the vertical at which it loses contact with the circle.
Although the constraint in this problem is technically that the mass
cannot penetrate the circle (i.e. the radial coordinate of m with respect
to the center satisfies r ≥ R), the only relevant regime is where r = R.
Therefore, we shall take r = R as our holonomic constraint as we can
identify the exact moment where it fails (i.e. when the normal force
becomes zero). Define θ to be the angle that the position vector of m
makes with the vertical axis, which is positive upwards. The modified
Lagrangian is
1
L = m(ṙ 2 + r 2 θ̇ 2 ) − mgr cos θ + λ(r − R).
2
This Lagrangian is similar to that in the previous problem, with an additional
negative sign in front of mgr cos θ due to the differing definitions of θ. Then,
the normal force N in this case can be obtained by substituting (π − θ) into
the previous expression for tension, i.e.

∂(r − R)
N =λ = −mr θ̇ 2 + mg cos θ.
∂r
Next, we apply the E-L equation with respect to θ. In doing so, we can treat
r as a constant (r = R) as the Lagrange multiplier is independent of θ such
that its presence does not affect the equation of motion with respect to this
coordinate.

g sin θ = Rθ̈.
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 642

642 Competitive Physics: Mechanics and Waves

d(θ̇ 2 )
Adopting the substitution θ̈ = 2dθ and separating variables,
 θ̇ 2  θ
2 2g
d(θ̇ ) = sin θdθ
0 0 R
2g
θ̇ 2 = (1 − cos θ).
R
Armed with this expression, we can determine the angle θ at which N = 0.
When N = −mr θ̇ 2 + mg cos θ = 0,
g
θ̇ 2 =
cos θ
R
g 2g
cos θ = (1 − cos θ)
R R
2
θ = cos−1 .
3
At this point in time, we are pining for the conservation of energy — a
crucial component of Newtonian mechanics — as it would have drastically
simplified the process in this problem.

12.5 Conserved Quantities and Symmetry*

The conservations of energy, momentum and angular momentum and more
general forms of conserved quantities are actually direct consequences of
Hamilton’s principle. In fact, they are closely related to the symmetries
of a system. This section is a formality in the sense that it simply proves
that the Lagrangian formulation is coherent with several defining aspects of
the Newtonian one. It is often much simpler to directly use the Newtonian
formulation to obtain the conserved quantities.

12.5.1 The Conservation of Energy

If the Lagrangian is not an explicit function of t, we claim that the quantity

 n 
∂L
H≡ · q̇i − L
∂ q̇i
i=1

is conserved (time-independent). In a closed system, the Lagrangian is indeed

time-independent as we should in principle be able to describe all states —
past, present and future — in terms of the generalized coordinates and their
first-order time derivatives.
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 643

Lagrangian Mechanics 643

H is known as the Hamiltonian of the system but is, most of the time,
equivalent to the mechanical energy of the system. Since we will not be
analyzing the Hamiltonian formulation here, we shall just refer to H as the
mechanical energy of the system. To prove the claim above, consider the
total time derivative of the Lagrangian.

n n
dL ∂L ∂L ∂L
= + · q̇i + · q̈i
dt ∂t ∂qi ∂ q̇i
i=1 i=1
n   n
d ∂L ∂L d
= · q̇i + · (q̇i )
dt ∂ q̇i ∂ q̇i dt
i=1 i=1
 n

d ∂L
= · q̇i ,
dt ∂ q̇i
i=1

where we have applied ∂L∂t = 0 as the Lagrangian does not explicitly depend
dL
on t. Shifting dt to the right-hand side,
 n

d ∂L
· q̇i − L =0
dt ∂ q̇i
i=1
n
∂L
=⇒ · q̇i − L = H
∂ q̇i
i=1

for some constant H. How does this obscure-looking term reduce to the
familiar H = T + V ? The answer: Euler’s theorem of homogeneous
functions.

Euler’s Theorem of Homogeneous Functions

Let f (q1 , q2 , . . . , qn ) be a homogeneous function of degree k such that

f (mq1 , mq2 , . . . , mqn ) = mk f (q1 , q2 , . . . , qn ).

We claim that
n
∂f
· qi = kf.
∂qi
i=1
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 644

644 Competitive Physics: Mechanics and Waves

We differentiate the first equation with respect to m to obtain

∂f (mq1 , mq2 , . . . , mqn )

kmk−1 f (q1 , q2 , . . . , qn ) =
∂m
n
∂f (mq1 , mq2 , . . . , mqn ) ∂(mqi )
= ·
∂(mqi ) ∂m
i=1
n
∂f (mq1 , mq2 , . . . , mqn )
= · qi .
∂(mqi )
i=1

Substituting m = 1,
n
∂f
· qi = kf.
∂qi
i=1

Returning to our problem at hand, we need to evaluate

n
∂L
· q̇i .
∂ q̇i
i=1

Assuming that the potential energy V does not depend on the time-
derivative of generalized coordinates — an assumption which is only
invalid in the presence of charges, the Lagrangian of a closed system is of
the form

L = T (q, q̇) − V (q),

where T is quadratic in q̇ (i.e. it is a homogeneous function in q̇ of degree 2).

Therefore,
n n
∂L ∂T
· q̇i = · q̇i = 2T
∂ q̇i ∂ q̇i
i=1 i=1

H = 2T − (T − V ) = T + V.

We have recovered the familiar expression for mechanical energy! Two

assumptions were made in our derivation. Firstly, the Lagrangian is time-
independent. This is equivalent to saying that a translation in time does not
modify the system’s behaviour. Secondly, the potential energy V is inde-
pendent of q̇. When an electromagnetic field is present, this assumption is
invalid and energy is seemingly not conserved.
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 645

Lagrangian Mechanics 645

12.5.2 Cyclic Coordinates

Observe that if the Lagrangian does not depend on a certain coordinate qk

(but possibly q̇k ), it follows from the E-L equations that
 
d ∂L
=0
dt ∂ q̇k
∂L
=⇒ =c
∂ q̇k
is a conserved quantity. Such coordinates qk are known as cyclic coordi-
nates. To illustrate this, consider a particle under the influence of a central
potential in spherical coordinates. Then,
1  
L = m ṙ 2 + r 2 sin2 θ φ̇2 + r 2 θ̇ 2 − U (r).
2
Noticing that L does not depend on φ,
∂L
= mr 2 sin2 θ φ̇
∂ φ̇
is conserved. This is simply the z-component of the angular momentum of
the particle!

12.5.3 Continuous Symmetries

Even in the case where there are no obvious cyclic coordinates, there can
still be certain conserved quantities. Well, perhaps we just did not choose a
convenient coordinate system. Let us consider an instructive Lagrangian of
two particles with coordinates q1 and q2 that move only in a single direction.
1
L = m(q̇12 + q̇22 ) + U (aq1 + bq2 )
2
for some constants a and b. Observe that if we increment q1 by a certain bδ
where δ is a constant infinitesimal quantity and q2 by −aδ,

q1 → q1 + bδ,
q2 → q2 − aδ.

The Lagrangian remains the same! This is a form of continuous symmetry

(continuous because the Lagrangian is invariant after infinitesimal vari-
ations of coordinates). Now, we turn to another topic: finding a quantity
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 646

646 Competitive Physics: Mechanics and Waves

that is conserved in this system. Consider the canonical momenta p1 and p2

conjugate to q1 and q2 respectively. From the E-L equations,

dp1 ∂U (aq1 + bq2 ) ∂(aq1 + bq2 )

= · = aU  (aq1 + bq2 ),
dt ∂(aq1 + bq2 ) ∂q1
dp2 ∂U (aq1 + bq ) ∂(aq1 + bq2 )
= · = bU  (aq1 + bq2 ).
dt ∂(aq1 + bq2 ) ∂q2

Notice that
dp1 dp2
b −a =0
dt dt
=⇒ bp1 − ap2 = c

is a conserved quantity of the system, also known as an integral of motion,

even though there were no cyclic coordinates! Furthermore, we notice that
the coefficients in front of p1 and p2 are strangely identical to that of δ
when varying q1 and q2 ! We shall now explore the deeper reason behind this
coincidence.

12.5.4 Noether’s Theorem

Noether’s theorem elegantly connects the symmetries of a system to con-

served quantities through Hamilton’s principle, which acts as a mediator.
Consider a variation of each coordinate qi in the following manner:

qi → qi + fi (q, t)δ

q̇i → q̇i + f˙i (q, q̇, t)δ,

where fi (q, t) is any arbitrary function in terms of the generalized coordinates

q and t for each coordinate qi and δ is a constant infinitesimal quantity. If
the Lagrangian is preserved to the first order in δ after such an infinitesimal
variation of coordinates,
n n
∂L ∂L ∂qi ∂L ∂ q̇i
0= = · + ·
∂δ ∂qi ∂δ ∂ q̇i ∂δ
i=1 i=1
n   n
d ∂L ∂L ˙
= · fi + · fi
dt ∂ q̇i ∂ q̇i
i=1 i=1
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 647

Lagrangian Mechanics 647

 n

d ∂L
= · fi
dt ∂ q̇i
i=1
n
∂L
=⇒ · fi = Q
∂ q̇i
i=1

is an integral of motion. Note that we have applied the E-L equation in

obtaining the second equality. This relationship is known as Noether’s the-
orem. We see that it elegantly implies that symmetries are closely tied to
conserved quantities and this relationship is enforced by Hamilton’s princi-
ple. In practice, we have to determine fi by trial and error, but this is usually
not too tedious.
Consider the Lagrangian of a two-dimensional harmonic oscillator
1 1
m(ẋ2 + ẏ 2 ) − k(x2 + y 2 ).
L=
2 2
Let us examine the ramifications of the following variation:
x → x − yδ,
y → y + xδ.
Notice that
(x − yδ)2 + (y + xδ)2 = x2 + y 2 ,
where we have neglected second-order terms in δ. Meanwhile, ẋ2 + ẏ 2 remains
unchanged as the variations are time-independent. Therefore, the above vari-
ation is a symmetry of the system. The conserved quantity is then
n
∂L ∂L ∂L
· fi = · −y + · x = m(xẏ − y ẋ).
∂ q̇i ∂ ẋ ∂ ẏ
i=1

This is, again, simply the angular momentum of the oscillating body in
the z-direction. We have merely expressed our system in the cumbersome
Cartesian coordinates instead of the convenient polar coordinates, which
would have directly resulted in a cyclic angular coordinate!

Conservation of Momentum

Now, let us derive two conserved quantities in Newtonian mechanics from

the homogeneity and isotropy of space and time in inertial frames. The
conservation of momentum is a result of the homogeneity of space — which
simply put, means that space is uniform. This causes the Lagrangian to be
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 648

648 Competitive Physics: Mechanics and Waves

invariant when the entire closed system is displaced by a certain distance in

a certain direction. Let there be n particles in total, with the ith particle
having coordinates xi , yi and zi . Then, we can shift the entire system (or our
coordinate axes) such that

xi → xi + δ,
yi → yi ,
zi → zi ,

and the Lagrangian should still remain the same. By Noether’s theorem, the
quantity
n n n
∂L ∂T
= = mẋi = px ,
∂ ẋi ∂ ẋi
i=1 i=1 i=1

which corresponds to the x-component of the total momentum of the system,

is conserved for a closed system. Note that in writing the first equality, the
potential V was assumed to be independent of ẋi . We can repeat the above
process for the y and z directions, to conclude that the total momentum
vector p of a closed system is conserved.

Conservation of Angular Momentum

Due to the isotropy of space in an inertial frame (that is, all directions are
equal), the Lagrangian of a closed system does not change after an infinitesi-
mal rotation about an origin. Note that the position vector of the ith particle,
r i , changes in the following manner after an infinitesimal rotation about the
origin described by the rotation vector δ.

ri → ri + δ × ri .

Assuming that the rotation is solely along the z-direction, δ = δẑ,

xi → xi − yi δ,
yi → yi + xi δ,
zi → zi .
July 10, 2018 12:25 Competitive Physics 9.61in x 6.69in b3146-ch12 page 649

Lagrangian Mechanics 649

The conserved quantity is then

n n n
∂L ∂L
· (−yi ) + · xi = m(xi ẏi − yi ẋi )
∂ ẋi ∂ ẏi
i=1 i=1 i=1
n
= (r i × pi )z
i=1

= Lz ,
which is the z-component of the total angular momentum of the system. We
can then repeat this for rotations about the x and y-axes to conclude that
the total angular momentum L is conserved for a closed system.