Electromagnetism and Quantum

Theory
Luis Fernando Alday

Part1: Electromagnetism
This half of the course is a short introduction to the mathematical description of the salient
points of the theory of electromagnetism. I will assume that you have some familiarity
with the physical phenomena of electromagnetism which the theory aims to encompass and
account for. For many of you this will have come from physics courses in school, and can
be reinforced by consulting a book like Dun (see below). Our main concern is with the
formalism: The language of vector operators and the integral theorems of Gauss, Green
and Stokes are precisely tailored to handle electromagnetic theory. Once one has the theory
written that way, there are various tantalising hints of how to carry it forward into the
four-dimensional space-time formalism of special relativity - but that is another course.
Synopsis
Maxwells equations in vacuum with sources. Lorentz force law. Plane waves and polariza-
tion. Electrostatics and Magnetostatics. Energy density and the Poynting vector. Scalar and
vector potentials. Gauge invariance. Maxwells equations in the Lorentz gauge. The wave
equation for potentials.
Reading list
One good source, with extensive discussion of the experimental background, which I shall of
necessity by omitting is:
W.J. Dun, Electricity and Magnetism, McGraw-Hill (2001), Fourth Edition, Chapters 1-
4,7,8,13.
Classic texts on electromagnetism:
R. Feynman, Lectures in Physics, Vol.2. Electromagnetism, Addison Wesley.
J.D. Jackson, Classical Electrodynamics, John Wiley.

Largely based on notes by Paul Tod.

1 Time independent Electromagnetic Theory
1.1 Electrostatics
This is the subject of static electricity, which weve all played with: recall rubbing a pen on
your cat and picking up bits of paper. We interpret this as charging up or adding charge to
the pen. As greeks didnt have pens, they used amber. The greek word for amber is electron
and this is where the subject gets its name from.
We will assume the following facts about charge:
there exist charges in nature;
charge can be positive or negative;
charge cannot be created or destroyed;
For a more sophisticated standpoint, we may know that charge is discrete, and that electrons
have charge -1 and protons charge +1 in suitable units, but for our purposes we want to
think of charge as continuous. As a mathematical idealization, we represent charge by a real
function, the charge density (x, y, z, t), with the property that the total charge in a region
V of space is
Q =
_
V
dV =
___
V
dxdydz. (1)
Other idealizations are possible, and we shall sometimes use them. For example, it may be
convenient to think of charges residing on a surface, something two-dimensional, and then
charge surface density (x, y, t) is charge per unit area, so that the charge on a piece S of
surface is
Q =
_
S
dS; (2)
similarly, for a one-dimensional distribution of charge, charge line density is charge per unit
length on some curve. Finally, a point charge corresponds to a nite, nonzero charge localized
at a point.
How do charges interact with each other? An important experimental fact about point
charges is the
Coulombs Law: Given two point particles P, P

with charges Q, Q

at positions r, r

, the
electric force on P due to P

is

F =
1
4
0
QQ

r r

[r r

[
3
(3)
where
0
is a constant and accounts for the correct units.
2
We treat this law as an experimental fact. To see how one might have discovered it, note
that:
it is an inverse-square law, acting along the line joining the particles;
it is proportional to the product of the charges, so that like charges repel and unlike
charges attract.
It is convenient to think of this law as follows: P

generates an electric eld to which P

responds.
Put P

at the origin and dene the electric eld at r to be

E(r) =
1
4
0
Q

r
r
3
(4)
Then P, which has charge Q, when placed in this eld is subject to (or feels) the force

F = Q

E (5)
Replace P

by various particles P
1
, P
2
, ..., P
n
, at points r
1
, r
2
, ..., r
n
with charges Q
1
, Q
2
, ..., Q
n
.
The forces will add as vectors so if we dened the electric eld now as

E(r) =
1
4
0
n

i=1
Q
i
(r r
i
)
[r r
i
[
3
, (6)
then, as before, P placed at r is subject to the force

F(r) = Q

E(r) (7)
We want to explore (6). Note rst that it is the gradient of another function:

E = , =
1
4
0
n

i=1
Q
i
[r r
i
[
. (8)
3
This function is called the electric potential. In particular, therefore

E = 0, (9)
so the force (7) is a conservative force in the language of mechanics (familiar to you from
Mods). We also have

E =
2
= 0, except at r = r
i
(10)
Consider the diagram
P
1
P
3
P
2
S
1
S
3
S
2
V
large sphere S
i.e. S is a large sphere containing all P
i
, S
i
is a small sphere containing only P
i
and

V is the
region in between.
Then

E = 0 on

V , so
_
S

E dS
n

i=1
_
S
i

E dS =
_

EdV = 0 (11)
the rst two are surface integrals (with dS pointing away from the center of the corresponding
sphere), while the last one is a volume integral. Make sure you understand the signs in this
equation. Let us focus in the integral over S
1
, taking r
1
= 0 for simplicity we obtain:
_
S
1

E dS =
_
S
1
1
4
0
_
Q
1
r
r
3
n + terms integrating to zero
_
dS
=
1
4
0
Q
1
_
1
r
2
r
2
sin dd
=
1

0
Q
1
,
4
in the second step we used n =
r
r
. Each term in the sum (11) can be calculated like this, so
we obtain:
_
S

E dS =
n

i=1
_
S
i

E dS =
1

0
(Q
1
+ Q
2
+... +Q
n
) =
1

0
total charge inside S (12)
This is the Gausss Law in its integral form:
Gausss Law: The ux of

E out of V =
1

0
total charge in V .
If we have a smoothed out charge density instead of point charges, Gausss law would read
_
S

E dS =
1

0
_
V
dV, (13)
where V is the volume inside the closed surface S. We will assume this to be true.
Now, by Divergence theorem we have
_
V
_

E
1

_
dV = 0 (14)
but if this is to hold for all possible regions V , then

E =
1

0
, (15)
which is the dierential version of Gausss Law.
This together with

E = implies Poissons equation

2
=
1

0
(16)
What is the solution for this equation? remember that for point particles we had (see (8))
=
1
4
0
n

i=1
Q
i
[r r
i
[
(17)
so we might guess that the solution of (16) is
(r) =
1
4
0
___
V
(r

)
[r r

[
dx

dy

dz

(18)
where V is the region in which ,= 0. In fact, this is the correct guess.
An example
The main problem of Electrostatics is to obtain

E given the charge distribution. We have
solved that problem with (18): Given the density , integrate this to nd and then
5

E = . However, in simple cases we can go straight to

E by using (4). Here is an
example:
Total charge Q is spread out uniformly round a plane circular wire of radius a. Find the
electric eld at a point P on the axis of the circle, at a distance b from the centre.
L
P
b
a

k
e
unit vectors
We can do this one directly from the inverse square law:

E =
q
4
0
r
r
3
.
Cut the circular wire into elements, each of legth ad; then each contains charge equal to
Q
2
d and so contributes
Q
2
d
1
4
0
e
L
2
, (with e as in the gure) to

E at P. Adding these up
around the circle leads, by symmetry, to a vector along

k. Note, from the diagram, that
e

k = cos , so that

E(p) = E

k, with E a scalar and given by

E =
_
Q
2
d
1
4
0

cos
L
2
=
Q
4
0
cos
L
2
=
Q
4
0
b
(a
2
+ b
2
)
3/2
which is the answer.
It is trickier to nd

E for a point P

not on the axis.

and the Potential Energy
Let us think about Newtons equations of motion for a charged particle of mass m and charge
Q subject to the force (7)
ma =

F = Q

E = Q (19)
Then
d
dt
_
1
2
mv v
_
= mv a = Qv = Q
3

i=1

x
i
dx
i
dt
= Q
d
dt
(20)
6
so that
c =
1
2
m[v[
2
+ Q = constant (21)
In the motion of the charged particle, the kinetic energy changes with time, but the sum
of the kinetic energy and Q is constant; thus Q is the potential energy of the particle of
charge Q in the electric eld with potential .
1
1.2 Magnetostatics
Frequently, people nd magnetism a little more mysterious than electricity. Rather than
thinking of fridge magnets, you should think of electromagnets: charges in motion give rise
to electric currents and these produce magnetic elds; many of you will have visualized
these elds in experiments by sprinkling iron lings on sheets of cardboard transverse to a
current-carrying wire. (The fridge magnet, or any permanent magnet, derives its magnetism
from microscopic currents.) The subject corresponding to electrostatics, which is produced
by time-independent charges, is magnetostatics, which is produced by time-independent, or
steady, currents.
+ v
+ v
+ v
+ v
+ v
Given a collection of point charges Q
i
in motion with velocities v
i
, we dene the corresponding
electric currents as:

j =

Q
i
v
i
This is a vector eld made by adding up elementary contributions. By analogy with charge
density as a smoothed-out distribution of point charges, we introduce a vector eld, the
current density

J(x, y, z, t), so that the total electric current in a region V is the integral
_
V

JdV .
In this section we shall usually be thinking of steady currents, but there is one thing to deal
with rst, namely the mathematical expression for the physical observation than charge is
conserved.
Conservation of charge
Consider a region V with surface S with charges moving through.
1
Note that this is completely analogue to the usual (gravitational) potential energy, the mass mis replaced
by the charge Q and the potential gh is replaced by the potential .
7
V
S
The total charge inside V is Q =
_
V
dV , so that
dQ
dt
=
_
V

t
dV allowing

t
,= 0 for the moment
= rate of increase of Q
= rate charge goes in rate charge goes out
=
_
S

J dS
=
_
V

JdV by divergence theorem
an so
_
V
_

t
+

J
_
dV = 0 (22)
this is to be true for all regions V , then

t
+

J = 0 (23)
which is the charge conservation equation.
For the rest of this section, we suppose none of the quantities of interest depends on time.
For this time independent situation (23) reduces to

J = 0.
Magnetic Field Strength
Like the electric eld

E, the magnetic eld strength

B can be dened from an experimentally
established force law. Analogously to (7), a charged particle with charge Q, moving with
velocity v in a magnetic eld

B is subject to a force

F = Qv

B (24)
8
F
B
v
This serves to dene

B, but note a peculiar feature of this force law, that the force is
orthogonal to the velocity. In particular, therefore v

F = 0 and the force does no work.
Forces between two parallel current-carrying wires
1
I = I k
1 2
I = I k
2
k
i
j a
lines of force
To see (24) in action, let us consider the force between two parallel current-carrying wires.
What can we say about the magnetic eld

B generated by a single wire? let us choose coordi-
nates such that the wire is along the z direction and let us use cylindrical-polar coordinates.
It can be experimentally veried that if we take a charge Q and move it around the wire, the
charge is subject to a force. This force vanishes, i.e. there is no force, if the charge moves
along the direction, while there is a force if the charge moves in any other direction. This
means that

B has a component only in the direction
2
; using the symmetry in and z we
can write

B = B(R)e

, where R is the distance to the wire.

Now suppose we have parallel wires, wire 1 carrying current I
1
and wire 2 carrying current
I
2
. Each wire gives rise to a B-eld and the eld from one exerts a force on the current in
the other. Using (24), and thinking of the currents as charged particles in motion, we have:
Force per unit length on wire 2 due to wire 1 is

F
12
=

I
2

B
1
.
2
This is consistent with the following fact: if you sprinkle iron lings on a cardboard held at right angle
to a current-carrying wire, you can see that the magnetic lines of force are concentric circles, centered on the
wire.
9
where

B
1
means

B at wire 2 due to current in wire 1, and from the geometry

I
2
= I
2

k.
Similarly, the force on wire 1 due to wire 2 is

F
21
=

I
1

B
2
.
where

B
2
is

B at wire 1 due to wire 2, and I
1
= I
1

k. These forces must be equal in magnitude

and opposite, due to Newtons third law:

I
2

B
1
=

I
1

B
2
.
From what was said about the direction of

B, we have that at wire 2,

B
1
=

jB
1
in terms of
some magnitude B
1
, and in wire 1,

B
2
=

jB
2
in terms of some magnitude B
2
. Therefore
[

F
12
[ = I
2
B
1
= [

F
21
[ = I
1
B
2
.
It is an experimental fact that this magnitude is
[

F[ =

0
2a
I
1
I
2
in terms of a constant
0
(which, as with
0
is needed to get the right dimensions).
From this experimental fact we deduce that, at wire 2,

B =

0
I
1
2a

j
At a general point, in terms of cylindrical polar coordinates (R, , z) with the wire along the
z-axis this is
=

0
I
2
1
R
e

which we can write in cartesian coordinates as

B =

0
I
2
_

y
R
2
,
x
R
2
, 0
_
, R
2
= x
2
+ y
2
(25)
This is the magnetic eld due to an innite straight wire with constant current, which can
be though of as the elementary magnetic eld, much as (4),

E =
Q
4
0
r
r
3
gives the elementary
electric eld.
From (25), for R ,= 0, we calculate

B = 0,

B = 0 (26)
Our aim for the next few pages is to calculate the

B due to current in an arbitrary wire,
much as the integral (18) together with the relation

E = gives

E for an arbitrary
10
charge distribution. We achieve this aim with (32) below. As a step towards this aim, we
note that, when is a horizontal circle of radius R centered on a straight wire
_

B d

=

0
I
2
_
1
R
e

Rd (27)
=
0
I (28)
Even though we have derived this for a straight wire, the result is much more general:
Amperes Law:
_

B d

=
0
total current through .

I
Compare amperes law to Gauss Law in the rst form we had.
For a current density instead of a wire, we obtain an integral version of Amperes Law:
_

B d

=
0
_

J dS spans
=
_

B dS by Stokes theorem
and for this to be true for all we obtain the dierential version of Amperes law:

B =
0

J (29)
(recall that

J = 0 from charge conservation and time independence, and we need this for
(29) to make sense).
By looking at (26) we have

B = 0, which is to be contrasted with (15):

E =
1

0
.
Thus we interpret (26) as saying there are no magnetic charges.
The magnetic potential
11
It follows from (9) that, in a simply-connected region, there exists a function such that

E = . It is a less familiar fact that

Claim: if

B = 0 in a suitable region, then there exists in that region a vector eld

A
such that

B =

A.
You will prove this on problem sheet 2.

A is the magnetic potential (also called vector
potential, in which case is called the scalar potential).
Note that a change

A

A + , for any scalar function , leaves

B unchanged. We may
exploit this freedom to impose another condition, namely

A = 0
for suppose F =

A ,= 0 and change

A

A+ , then this would change F F +
2
;
now, choose such that
2
= F, so that now

A = 0.
Equation (29) can be turned into an equation for

A as follows:

B = (

A) = (

A)
2

A
=
0

J (30)
with the choice

A = 0, then we have

A =
0

J (31)
Remember that this is true for the particular choice

A = 0. This has the vector form of
the Poissons equation, and weve solved Poissons equation before: recall (18) and compare
with (16). So we solve (31) by

A =

0
4
___

J(r

, t)
[r r

[
dx

dy

dz

This is for a volume distribution of current. If we have just a line current

J = I

t, where I is
constant and

t is the tangent (of unit length) to a curve L (where the curve L is the wire),
then instead

A =

0
I
4
_
L

td
[r r

()[
12
t
I
where parametrizes where we are along the wire.
From

A we calculate

B as

B =

A =

0
I
4
_
L

_

t
[r r

[
_
d
=

0
I
4
_
L

t (r r

)
[r r

[
3
d
or

B =

0
I
4
_
L
d

(r r

)
[r r

[
3
(32)
Where d

= d

t. This is now

B at all points of space due to an arbitrary wire, and so
generalizes (25). We can use it to obtain an expression for the force between two arbitrary
wires, by chopping the second wire into elementary pieces.
P
1
P
2
r
I
1
I
2
At P
2
on wire 2
d

F = I
2
d

B
1
so

F =
_
I
2
d

B
1
=

0
I
1
I
2
4
__
d

2
(d

1
(r
2
r
1
))
[r
2
r
1
[
3
(33)
13
where r
1
= r
1
(
1
) belongs to wire 1 and r
2
= r
2
(
2
) belongs to wire 2. This is the Biot-Savart
Law for the force between two arbitrary current-carrying wires (more precisely, the force on
wire 2 done by wire 1). Its interest is largely theoretical, as it is rather hard to use.
1.3 The story so far
Weve studied time-independent electricity (or electrostatics) and time-independent mag-
netism (or magnetostatics), and found these to be governed by the following system of
equations

E =
1

0


B = 0

E = 0

B =
0

J (34)
with the understanding that

E
t
= 0 =

B
t
=

J
t
=

t
.
The electric and magnetic eld strengths themselves can be dened from the force laws (7)
and (24), which we combine as follows: a particle with charge Q moving with velocity v in
a combination of an electric eld

E and a magnetic eld

B is subject to the force given by

F = Q(

E +v

B).
In addition we have one equation which we expect to be valid also in the time-dependent
case, namely the charge conservation equation

t
+

J = 0. (35)
The problem now is how to extend all these equations to the time-dependent case, i.e. how
to put back the terms that are zero in the time-independent case. This will be the subject
of the second part of this course.
14
2 Time Dependent Electromagnetism: Maxwells Equa-
tions
Experimental evidence suggests two qualitative facts:
A time-varying

B produces an

E.
A time-varying

E produces a

B.
This experimental evidence can be summarized quantitatively in the following two integral
laws:
_

E d

=
d
dt
_

B dS Faradays law of induction (36)

_

B d

=
_

J dS +
_

E
t
dS (37)
Where spans . Apply Stokes theorem to both to obtain
_

E +

B
t
_
dS = 0 (38)
_

B
0

J
0

E
t
_
dS = 0 (39)
and if these hold for all we obtain the
Maxwell equations:

E +

B
t
= 0;

B =
0

J +
1
c
2

E
t

B = 0,

E =
1

0
(40)
with c
2
=
1

0
.
These are the Maxwells equations, the basic equations of the new theory of electromagnetism,
in which electricity and magnetism are merged.
Note that
Two of the equations always have zero on the right; the other two contain the sources
and

J.
15
If all time-derivatives are set to zero, we recover the equations of the previous section.
The system obtained by setting and

J to zero is called source-free Maxwell equations.
Regarding and

J as given, there are eight equations for six unknowns, so there should
be a consistency condition.
To see what the consistency condition is, calculate:

t
=
0

t
(

E) =
0

E
t
_
=
1

B
0

J
_
=

J
which we recognize as the charge conservation equation. From this point of view the charge
conservation equation is a consistency condition for the Maxwells equations. As an exercise,
show that there is a second consistency condition but that it is automatically satised.
Energy of the Electromagnetic Field
There is a deductive way to approach this question, but we shall rely on intuition and
guesswork. We want an energy density (energy-per-unit-volume) which is something like
1/2mv v.
After playing around with Maxwell equations for a while, we are lead to consider
c =
1
2

0
[

E[
2
+
1
2
0
[

B[
2
(41)
Then, using the Maxwell equations, we calculate
c
t
=
0

E
t
+
1

B
t
=
1

E
_

B
0

J
_

B

E
=
_
1

E

B
_

E

J
For the source-free case, we set

J = 0 so that the last term vanishes. Then, this has the
form of a conservation equations
c
t
+

P = 0 (42)
16
Remember that c is the energy density.

P
1

E

B is called the Poynting vector and has
the interpretation of the rate of energy ow, or momentum density.
If we now reinstate the sources, this equation has a name:
Poyntings theorem
c
t
=

P

E

J
We integrate Poyntings theorem over a volume V with surface S to obtain an energy balance
equation:
d
dt
_
V
cdV =
_
S

P dS
_
V

E

JdV
(1) (2) (3)
where each term has the following interpretation:
1. Rate of increase of electromagnetic energy.
2. Rate of energy ow into V .
3. Rate of work done by the eld on sources.
To justify the interpretation of (3), consider a single charge. Remember that work is the
product of force and displacement dW =

F d

, so the rate of work is

dW
dt
=

F v. For a single
charge

F = Q

E and Qv is by denition the current

j, so that
dW
dt
=

E

j. This satisfactory
interpretation of the third term reinforces the interpretation of c as energy density.
Potentials
The introduction of potentials in the time-dependent case is similar to that in the time-
independent case. Assume we are interested in a suitable (simply-connected, etc) region of
space, then the Maxwell equations imply, rst:

B = 0

A such that

B =

A;
next


B
t
=


A
t
, so that

E +

B
t
=
_

E +


A
t
_
= 0,
whence
such that

E +


A
t
=
17
i.e.

B =

A (43)

E =


A
t
(44)
The freedom to modify

A and changes slightly: if

A

A+, then

B is unchanged, but
we need

t
to leave

E unchanged.
This is a gauge transformation and can be exploited, as before, to simplify potentials.
Electromagnetic Waves
For the rest of this section, we show that the source-free Maxwell equations admit wave
solutions, and use these solutions to recover some elementary properties of optics. The
realization that light is an electromagnetic phenomenon was a triumph of nineteenth century
science.
We start from the Source-free Maxwell equations:

B = 0,

E = 0 (45)

E +

B
t
= 0,

B
1
c
2

E
t
= 0
we would like to decouple the equations for

E and

B. Calculate:
(

E) = (

E)
2

E =
2

E
=

B
t
=

t
(

B) =
1
c
2

E
t
2
so that
1
c
2

E
t
2
=
2

E (46)
which is the wave-equation with c = (
0

0
)
1/2
as the wave speed.
Similarly (exercise)
1
c
2

B
t
2
=
2

B (47)
We want to solve these wave equations, so think rst about a scalar version
1
c
2

2
F
t
2
=
2
F =

2
F
x
2
+

2
F
y
2
+

2
F
z
2
(48)
18
and try F = f(t

K x) with constant and

K. Then
F
t
= f

, F =

Kf

2
F
t
2
=
2
f

,
2
F = [

K[
2
f

Hence, F = f(t

K x) is a solution for any (twice-dierentiable) f provided [

K[
2
=

2
c
2
.
Note that F is constant on surfaces
t

K x = const
at a xed t this is the equation of a plane, so these solutions are called planewaves. If f is
cos or sin, then they are called harmonic waves with a single frequency, or monochromatic
plane waves.
Now try

E = e
1
cos +e
2
sin , = t

K x (49)

B =

b
1
cos +

b
2
sin ,
with constant vectors

K e
i
and

b
i
. Imposing the source-free Maxwell equations in order:

E = sin e
1
+ cos e
2

= sin

K e
1
cos

K e
2
= 0
so we need

K e
1
=

K e
2
= 0 and then

K

E = 0 (50)
In the same way we also have

B = 0

K

b
1
=

K

b
2
= 0 and then

K

B = 0 (51)
Next

E +

B
t
= 0
so

K e
1
sin

K e
2
cos

b
1
sin +

b
2
cos = 0
i.e.

b
i
=

K e
i
(52)
19
and so

B =

K

E (53)
Finally

B
1
c
2

E
t
= 0
so

b
1
sin

K

b
2
cos +

c
2
e
1
sin

c
2
e
2
cos
or

c
2
e
i
=

b
i
(54)

c
2

E =

K

B (55)
From this and (52) we deduce:

2
c
2
e
i
=

b
i

K
= (e
1

K)

K
= [

K[
2
e
i
so that [

K[
2
=

2
c
2
as expected and (52) implies [

b
i
[ = [e
i
[[

K[, e
i

b
i
= 0, so
[e
i
[ = c[

b
i
[, e
i

b
i
= 0 (56)
These are transverse waves (unlike e.g. sound). The waves travel in the direction of

K, while

E and

B lie in wave fronts, transverse to the direction of propagation, and are orthogonal
to each other.
K
E, e
i
B, b
i
direction of wave
To simplify the expressions, take

K in the z-direction,

K x = Kz. Look in the plane z = 0
so that = t. Then

E = e
1
cos t +e
2
sin t (57)

B =
1

K e
1
cos t +
1

K e
2
sin t (58)
20
To discuss the phenomenon of polarisation, we distinguish some particular cases:
(i) e
1
, e
2
proportional, say to

i, then

E = E

i cos(t + ) (59)

B =
1
c
E

j cos(t + ) (60)
for some constant ;
x
y
c B
E
a wave like this is said to be lineary polarised;

E and

B oscillate parallel to two xed
orthogonal directions.
(ii) e
1
, e
2
orthogonal, of equal length, e.g. e
1
= E

i, e
2
= E

j, then

E = E(

i cos t +

j sin t) (61)

B =
E
c
(

i sin t +

j cos t); (62)

a wave like this is said to be circularly polarised;

E and

B rotate at a constant rate about
the direction of propagation.
t
x
y
E
cB
21
The direction is anti-clockwise and the wave is left circularly polarized if e
1
e
2

K > 0, and
the rotation is clockwise and the wave is right circularly polarized if e
1
e
2

K < 0.
(iii) the general case can be regarded as a combination of waves with dierent polarizations.
Epilogue
What comes next in this story? The question how does one transform the Maxwell equations
between frames of reference which are moving uniformly relative to each other? Evidently,
waves in one frame should be waves in the other, but the big surprise is the experimental fact
that the speed of light in vacuum is the same for all observers in uniform relative motion!.
Since the speed of light is made from constants appearing in Maxwells equations, this at
least suggests that Maxwells equations are the same in all frames. How all this works is the
province of Special Relativity: rst time and space t together as space-time, then t and
r t together so that vectors have four components instead of three, now and

A form a
4-vector and so do and

J, while

E and

B t together as an anti-symmetric 4 4 matrix.
See B7.2b for more...
22
Part 2: Quantum Mechanics
Synopsis
The mathematical structure of quantum mechanics and the postulates of quantum mechan-
ics. Commutation relations. Poissons brackets and Diracs quantization scheme. Heisen-
bergs uncertainty principle. Creation and annihilation operators for the harmonic oscillator.
Measurements and the interpretation of quantum mechanics. Schroedingers cat. Spin 1/2
particles. Angular momentum in quantum mechanics. Particle in a central potential. The
hydrogen atom.
Reading list
The main book followed for this part of the course is:
K. C. Hannabuss, Introduction to Quantum Mechanics (Oxford University Press, 1997).
Classic texts on Quantum Mechanics:
L. I. Schi, Quantum Mechanics (3rd edition, Mc Graw Hill, 1968)
L D Landau and E.M. Lifshitz, Quantum Mechanics Non-Relativistic Theory: Volume 3.
23
3 The Mathematical Structure of Quantum Mechanics
(Hannabuss chapter 6)
3.1 Reviewing Part A Course
We shall recall what you learned there and add one new idea:
a particle is described by its wave function:
(x, t) = (x, y, z, t) in 3 dim
= (x, t) in 1 dim
which is complex and continuous (and usually twice-dierentiable in x).
the function [[
2
is interpreted as the probability density of position provided that
_
=
_
[[
2
= 1 (63)
i.e. provided is normalized. (Well usually omit the range of integration when its
intended to be all space). For this interpretation to make sense we at least need
_
[[
2
< , i.e. is normalizable, (64)
for then a constant multiple of is normalized. Recall also the denition of the
probability current.

j
i
2m
(

) (65)
for a particle of mass m moving in a real potential V (x) changes in time according
to the Schrodinger equation:
i

t
= H
_

2
2m

2
+ V (x)
_
(66)
This is motivated by the Hamiltonian from Classical Mechanics
H =
1
2m
[ p[
2
+ V = E (67)
which gives the total energy E, together with the substitution
p i. (68)
24
the Schrodinger equation is linear, so the solutions of (66) form a complex vector space,
and given a normalizable solution, we can normalize it (multiplying it by an appropriate
constant) and still have a solution.
there exist separable solutions of (66):
(x, t) = (x)T(t) (69)
with
T(t) = e

iEt

(70)
and
H = E (71)
which is an eigenvalue equation for the allowed energies. Solutions like (69) are called
stationary states even though

t
,= 0, because

t
= 0 =

j. Correspondingly (71) is
called the stationary state Schrodinger equation.
in many examples (e.g. particle in a box, harmonic oscillator,...) (71) has a countably
innite, orthonormalizabled set of solutions
n
: n N with corresponding real
eigenvalues E
n
, so that, without loss of generality
_

j
=
ij
. (72)
These provide a basis for the general solution of (66):
(x, t) =

k
c
k

k
(x)e
iE
k
t/
(73)
If we know the wave function at the initial time (x, 0), then we can determine the
constants c
k
:
c
k
=
_

k
(x)(x, 0) (74)
Hence, the wave function at initial time determines the wave function at all times
(as it should, since the Schrodinger equation is a rst order equation in t). Now the
normalization of becomes
1 =
_
[[
2
=
_
_

c
j

j
e
iE
j
t/
__

c
k

k
e
iE
k
t/
_
=

[c
k
[
2
NEW IDEA: Now comes the new idea: since they sum to one, it is tempting to think
of the c
j
as the probabilities that the particle is in state j, and we shall. Since the
state j has energy E
j
this implies
P(E = E
j
) = [c
j
[
2
,
25
writing P for probability, and then the average value, or more properly the expectation,
of the energy is
E(H) =

j
E
j
[c
j
[
2
, (75)
note that this coincides with the usual denition of expectation
H

=
_

H =
_

i

t
=
_
_

c
j

j
e
iE
j
t/
__

c
k
E
k

k
e
iE
k
t/
_
=

E
j
[c
j
[
2
where we have used the denition from the Part A course for the expectation of a
function f of position (x) in the state :
E(f) f(x)

=
_
f(x)[(x, t)[
2
dx =
_

f (76)
3.2 States and Observables
Having nished the review, we start this part of the course by seeking a more abstract setting
for the above. We suppose that, in quantum mechanics,
a system can take up one of a number of states which correspond to elements of a
complex vector space 1, called the Hilbert space, which has a Hermitian inner product
that we will write u[v for u, v 1
proportional vectors dene the same state (so states actually correspond to one-dimensional
subspaces or rays in 1).
an element 1 is normalized if
[[[[
2
[ = 1
(this still leaves a freedom e
i
without changing the state.)
observables are self-adjoint linear operators A : 1 1 (recall the adjoint A

of a
linear operator A is dened by
A

[ = [A, for all , 1

and A is self-adjoint if A

= A)
26
the mot important example of an 1 is complex-valued functions on R
3
with
[
_
R
3

(77)
then the position operator

X = X
i
, i = 1, 2, 3 and the momentum operator

P =
P
i
, i = 1, 2, 3 are dened (as we know) by
X
j
= x
j

P
j
= i

x
j
(78)
Are these in fact self-adjoint? For P
j
we calculate
[P
j
=
_

(i

x
j
) =
_
(i

x
j
) =
_ _
i

x
j
_
= P
j
[
on integration by parts, assuming dierentiability of , and the vanishing of the
integrated term, which would follow from normalizability of , . Thus the P
j
are
self-adjoint, and it is trivial to check that the X
j
are. (Note however, that the P
j
are
only dened on a subspace of 1 (normalizable functions), and that both X
i
and P
i
may spoil normalizability- we will note but pass over this kind of technical issue.)
3.3 Statistical Issues
Following section 3.1, we will dene the expectation of an observable A in a (normalizable)
state to be
E

(A)
[A
[[[[
2
(79)
then e.g. for a function F(

X) and a normalized ,
E

(F) = [F =
_

F (80)
just as before. From the abstract denition (79) we obtain
1. E

(I) = 1 for the identity operator I;

2. E

(A) is real for self-adjoint A;

3. E

(A) 0 for positive A (i.e. A for which [A 0 for all );

4. E

(A + B) = E

(A) +E

(B) for all , C

27
All but the second are easy to see; for the second note that
[A = A[ = [A

= [A
for self-adjoint A.
Next dene the dispersion

(A) of an observable A in a state as

(A) =
_
E

(A
2
) (E

(A))
2
_
1/2
(81)
This is like standard deviation, so it measures the spread of the measurements of A about
the average value.
Proposition

(A) = 0 i is an eigenvector of A.
Proof
Suppose is normalized, then for any real ,
E

((A I)
2
) = [(A I)
2
= (A I)[(A I)
since A I is self-adjoint
= [[(A I)[[
2
Also, by linearity
E

((A I)
2
) = E

(A
2
) 2E

(A) +
2
so put = E

(A) (which is allowed, since this is real) to nd

(

(A))
2
= [[(A I)[[
2
, = E

(A)
Thus

(A) = 0 iif (A I) = 0 iif A =

as required.
To make contact with earlier formulae, suppose 1 has a (countable) basis of normalized
eigenvalues of A, say
n
, with
A
n
=
n

n
,
then for any normalized 1 we have an expansion =

c
n

n
with c
n
=
n
[, so that
E

(A) = [A =

n
c
n
[A
n
=

n
c
n
[
n
=

n
[c
n
[
2
,
and we recover (75).
28
Observables are going to be the things we can measure, and the results of these measurements
are the eigenvalues of the corresponding linear operator, but we cant assume that every
observable of interest has normalizable eigenvectors. For example, consider

P, then the
eigenvalue equation is

P i =

k
for a constant vector

k, so the eigenfunction is const e
i

kx/
, which isnt normalizable. As
another example, suppose

X =a
for eigenvalue a. Then
(x a)(x) = 0
and vanishes at every point except x = a. Continuity would then force to vanish
everywhere (though if you are Dirac, you might invent the delta function at this point).
As an example of dispersion, we revisit the treatment of a particle in a box from Part A. If
the box occupies 0 x a, then the Hamiltonian and normalized eigenfunctions were
H =
1
2m
P
2
,
n
(x) =
_
2
a
sin
nx
a
Now calculate
E

n
(P) =
_
a
0

n
(x)(i

n
(x))dx = 0
writing prime for d/dx, so that on average the particle moves neither left nor right, while
E

n
(P
2
) = E

n
(2mH) = 2mE
n
=
n
2

2
a
2
Thus the dispersion of P is not zero - the particle is jiggling about.
3.4 Time Evolution
In this abstract setting we will write
t
to indicate a time-dependent element of 1, then
the time-evolution, following the Schrodinger equation, is determined by a Hamiltonian H,
which is an observable, by
i

t
t
= H
t
(82)
29
Formally this can be solved by the expression
3

t
= exp
_

_
t
0
Hdt
_

0
(83)
where exp of an operator is dened by
exp(A) =

k=0
1
k!
A
k
and we will just note that there may be convergence issues.
To prove (83), suppose for simplicity that H has no t-dependence so that
_
t
0
Hdt = Ht
Then introduce
U
t
exp
_

iHt

_
(84)
so that (83) is just

t
= U
t

0
(85)
Claim:
U
0
= I

dU
t
dt
=
i

HU
t
=
i

U
t
H,
U
s
U
t
= U
s+t
so in particular U
t
= (U
t
)
1
.
Since H is self-adjoint,
U

t
= exp(
iH

) = U
t
= (U
t
)
1
and so U
t
is unitary.
The rst is clear; the second follows from dierentiating (84); for the third, consider
W
t
= U
s
U
t
U
s+t
,
3
The Hamiltonian H could depend on time, but for the following expression to be true we will assume
that Hamiltonians at dierent times always commute. This will always be true for the examples seen in this
course.
30
then W
0
= 0 and
dW
t
dt
=
i

H(U
s
U
t
U
s+t
) =
i

HW
t
this is a rst-order linear ODE so has a unique solution, which must therefore be W
t
= 0.
Then the fourth is clear. Now (85) is equivalent to
U
t

t
=
0
(86)
and we will prove it in this form. Dierentiating (86) we obtain
d
dt
U
t

t
=
d
dt
_
exp(
iHt

)
t
_
= exp(
iHt

)(
d
t
dt
+
i

H
t
)
which vanishes by (82). So U
t

t
is independent of t and we can evaluate it att = 0, obtaining
(86).
Thus time-evolution following from the Schrodinger equation (82) is a sequence of unitary
transformation (85). In particular
[[
t
[[
2
=
t
[
t
= U
t

0
[U
t

0
=
0
[U

t
U
t

0
=
0
[
0
= [[
0
[[
2
(87)
i.e. unitary evolution preserves the norm.
3.5 Measurements
When we measure an observable A the result is one of its eigenvalues with some probability.
If we measure it again at once, then we should get the same answer with probability one.
Thus the dispersion is zero, and the state just after a measurement must be an eigenstate of A
with the measured eigenvalue. There may be more than one eigenstate with such eigenvalue,
so we are led to make Luders Postulate:
Suppose a measurement of A in a state gives the result ; let Q

be the projection onto the

subspace of 1 spanned by the eigenvectors of A with eigenvalues ; then immediately after
the measurement has become Q

.
(Recall that a projection is a linear operator with Q
2
= Q). We say the measurement has
collapsed the wave function. The norm changes in this collapse so that it is not unitary.
4 The Commutation Relations
(Hannabuss chapter 7)
31
4.1 The Commutation Relations
Recall, in one dimension, the denitions of position and momentum operators as
X = x, P = i

then PX = i(x)

while XP = x(i

) so that
(PX XP) = i (88)
Generally, we dene the commutator of observables A, B as
[A, B] AB BA, (89)
then we have
[P, X] = iI, (90)
and more generally, in three-dimensions
[P
i
, X
j
] = i
ij
I, (91)
For completeness, note also
[X
i
, X
j
] = [P
i
, P
j
] = 0 (92)
Commutators have an algebra which includes the following properties:
1. [A, B] = [B, A]
2. [A, BC] = [A, B]C + B[A, C]
3. linearity in A and B, e.g. [A, B + C] = [A, B] + [A, C]
4. The Jacobi identity: [A, [B, C]] + [B, [C, A]] + [C, [A, B]] = 0.
5. For self-adjoint A, B , [A, B] is not self-adjoint, but i[A, B] is.
I leave the proofs as an exercise!
4.1.1 Aside: Similarity with Poisson Brackets
(Read this only if you know what a Poisson Bracket is)
The commutation relations between X
i
and P
i
strongly resemble the classical Poisson Bracket
relations for generalized coordinates and momenta. Recalling that the Poisson bracket is
dened by
32
f, g =

j
_
f
p
j
g
x
j

f
x
j
g
p
j
_
(93)
we have
p
i
, x
j
=
ij
, x
i
, x
j
= p
i
, p
j
= 0 (94)
Furthermore, the above properties of the commutator also paralell those of the Poisson
bracket.
The similarity with Poisson brackets led Dirac to suggest that each function classical f (for
instance, the coordinate x or the momentum p) should be related in quantum theory by an
operator Q(f), in such a way that for any pair of functions f, g the corresponding operators
satisfy
[Q(f), Q(g)] = iQ(f, g) (95)
This idea doesnt work simultaneously for all operators, and sometimes one faces ambiguities.
For instance, to the classical function px
2
we could assign PX
2
or X
2
P or PXP, and so on.
These dier from each other, since P and X do not commute, but their dierence is small,
namely, proportional to .
4.2 Heisenbergs Uncertainty Principle
We will start with a
Lemma
Suppose A, B, C are self-adjoint with
[A, B] = iC
and is real. Then
1. (A iB)

(A iB) = A
2
+ C +
2
B
2
;
2. for normalized 1
[[(A iB)[[
2
= E

(A
2
) + E

(C) +
2
E

(B
2
);
3. for any normalizable 1
E

(A
2
)E

(B
2
)
1
4
(E

(C))
2
(96)
with equality iif (A iB) = 0 for some real .
33
Proof
The rst one is obvious (just multiply); for the second
[[(A iB)[[
2
(A iB)[(A iB) = [(A iB)

(A iB),
and use the rst; now the LHS is non-negative so that the discriminant of the RHS as a
quadratic in must be non-positive, which gives the third.
Corollary: Heisenbergs Uncertainty Principle
For normalized ,

(P)

(X)
1
2
(97)
with equality i
= const. exp
_

2
(x )
2
_
(98)
for some real and complex .
Proof
Set A = P E

(P)I, B = X E

(X)I so these are self-adjoint and

[A, B] = [P, X] = iI
so take C = I. Note that
E

(A
2
) = (

(P))
2
, E

(B
2
) = (

(X))
2
, E

(C) = ,
then the result follows from (96). For equality we need , with
(A iB) = 0.
Introduce = E

(P) iE

(X), then
(A iB) = (i
d
dx
ix ) = 0
which is a separable ODE leading to
log =
1
2
(x
2
2ix + const.)
which is (98) with = i/.
We shall call a state with wave-function like (98) a minimum uncertainty state. You saw an
example in Part A, the ground state of the harmonic oscillator which had

0
(x) = const. exp(
mx
2
2
)
34
4.3 Simultaneous Measurability
Why is (97) called the Heisenberg Uncertainty principle? It implies that a reduction in the
dispertion of P,

(P), must be accompanied by an increase in the dispersion of X and

vice versa - there is a fundamental limit on reducing the common uncertainty. Thus P and
X cannot be simultaneously be measured to arbitrarily high accuracy, and the origin of this
is in the nonvanishing of the commutator. The same will hold for any pair of observables
which dont commute.
What if [A, B] = 0? For a nite-dimensional 1 there will then be a basis of common
eigenvectors, and on a common eigenvector A and B can be simultaneously measured exactly,
i.e. the dispersions vanish. For an innite-dimensional 1 one can do less. If A, B are self-
adjoint introduce:
1
A
= span of eigenvectors of A.
1
B
= span of eigenvectors of B.
1
AB
= span of common eigenvectors.
Then Claim: If AB = BA then 1
AB
= 1
A
1
B
Then we can say the following:
Corollary
If 1has an orthogonal basis of eigenvectors of A (so 1
A
= 1) and [A, B] = 0 then 1
AB
= 1
B
so that A, B can be simultaneously diagonalized on 1
B
.
We shall see below, in section 5, some important application of this observation. There we
shall have operators which commute with the Hamiltonian, so that energy eigenstates (i.e.
eigenstates of the Hamiltonian) can simultaneously be eigenstates of these other operators.
A result which can usefully be given here concerns operators which do not commute with
the Hamiltonian. We have
Proposition
For any observable A write A = [A then
d
dt
A =
A
t
+
i

[H, A] (99)
This follows rapidly from (66) by dierentiating [A. For an A with non explicit time-
dependence the rst term is zero, so that the rate of change of A is determined by the
commutator [H, A].
4.4 The Harmonic Oscillator Revisited
Now we come to the pay-o for this abstraction. Recall that the harmonic oscillator has
Hamiltonian
H =
1
2m
P
2
+
1
2
m
2
X
2
(100)
35
Inspired by section 4.2 above, dene
a

= P imX (101)
This operator is not self-adjoint and in fact its adjoint is
a
+
= (a

= P +imX (102)
These operators have the following important properties:
Lemma
1.
[a

, a
+
] = 2mI
a
+
a

= 2m(H
1
2
I)
a

a
+
= 2m(H +
1
2
I)
2.
[[a

[[
2
= 2m(E

(H)
1
2
)[[[[
2
[[a
+
[[
2
= 2m(E

(H) +
1
2
)[[[[
2
3.
[H, a

] = a

[H, a
+
] = a
+
Proof: 1 is straightforward; for 2 write a

[a

= [a
+
a

and use 1; for 3 use 1.

Proposition: The Harmonic Oscillator done by Algebra
Suppose H from (100) has a normalized eigenvector 1 then
1. there is a normalizable ground state
0
with energy eigenvalue E
0
=
1
2
;
2. there are normalizable excited states
n
= (a
+
)
n

0
for each positive integer n, with
energy eigenvalue E
n
= (n +
1
2
);
3. if we take P = i
d
dx
and X = x then
0
= const. exp(
mx
2
2
) and the
n
, n = 0, 1, 2, ...
are all the eigenstates.
36
Proof
Suppose the given normalized eigenvector satises
H = E
then
Ha

= (Ha

H + a

H) = (a

H + [H, a

]) = (E )a

by part 3 of the Lemma. That means that a

is also an eigenvector of H, but with

eigenvalue (E ). i.e. a

lowered the eigenvalue by . Also

[[a

[[
2
= 2m(E

(H)
1
2
)[[[[
2
= 2m(E
1
2
)[[[[
2
by part 2 of the lemma, so that this new eigenvector is normalizable.
We can then apply a

again. We repeat this process obtaining normalizable eigenstates

(a

)
k
with eigenvalue E
k
= E k. However, this process must stop as
[[(a

)
k
[[
2
(E (k 1/2)),
which cannot be negative. Thus the process will stop when we reach
0
with E = E
0
=
1
2

and
[[a

0
[[
2
= 0, i.e. a

0
= 0
so that further lowering gives zero. (At this point we dont know the dimension of ker a

,
but we do know that any element of it has E = E
0
=
1
2
). Now we start to raise: set

1
= a
+

0
then
H
1
= (Ha
+
a
+
H + a
+
H)
0
= ( +
1
2
)a
+

0
= (1 + 1/2)
1
Inductively, raising will give normalizable states
n
= a
+

n1
with eigenvalues E
n
= (n +
1/2). This process can be repeated indenitely and we have a chain of eigenstates, labelled
by the positive integers, arising from each candidate
0
. Every eigenstate must be in one of
these chains, since the only elements of ker a

are lowest-energy states so that a

+
and a

are one-to-one going up and down the chain.

As regards
0
if we make the usual identication P = i
d
dx
and X = x, then we have
0 = a

0
= (P imX)
0
(i
d
dx
imx)
0
= 0
from which

0
= const. exp
_

mx
2
2
_
This gives a unique normalized ground state
0
, i.e. in this case dim( ker a

) = 1, and
every other state is obtained by raising this one.
37
4.5 Uniqueness of the Commutation Relations
This is a technical aside: the algebraic approach of the previous subsection started just from
the commutation relations
[P, X] = i
We only used the identication P = i
d
dx
and X = x to nd
0
(x) and hence
n
(x).
What freedom do we have in making this identication for P and X? If we had instead
proceeded by assuming that ker(a

) was one-dimensional, spanned by say, then the nor-

malizable eigenstates would be

n
= (a
+
)
n
. Now we would have the Hilbert space 1

spanned by these and there would be an isomorphism of Hilbert spaces dened by relating
the bases:

n

n
(n)
Under this isomorphism the abstract P, X acting on 1

would map to the operators i

d
dx
, x
acting on our previous Hilbert space of functions. Thus the identication P = i
d
dx
and
X = x is unique up to isomorphism.
5 Angular Momentum
(Hannabus chapter 8)
5.1 Angular Momentum in Quantum Mechanics
Motivated by the denition of angular momentum from classical mechanics:

L = x mv = x p (103)
we dene in quantum mechanics the observables
L
1
= X
2
P
3
X
3
P
2
L
2
= X
3
P
1
X
1
P
3
(104)
L
3
= X
1
P
2
X
2
P
1
We call this

L orbital angular momentum. More briey
L
i
=
ijk
X
j
P
k
(105)
38
using the Einstein summation convention (repeated indices are automatically summed over
their whole range) and in terms of the alternating symbol dened by

ijk
= 1 if ijk is an even permutation of 123
= 1 if ijk is an odd permutation of 123
= 0 otherwise
(so in particular (a

b)
i
=
ijk
a
j
b
k
)
Properties
1. [L
i
, P
j
] = i
ijk
P
k
;
2. [L
i
, X
j
] = i
ijk
X
k
;
3. [L
i
, L
j
] = i
ijk
L
k
Proof
Well just do the rst of these. One can do them in an unsophisticated way, calculating e.g.
[L
1
, P
2
] and appealing to symmetry or in a more sophisticated way as follows:
[L
i
, P
j
] = [
imn
X
m
P
n
, P
j
] =
imn
[X
m
, P
j
]P
n
= i
imn

mj
P
n
= i
ijn
P
n
as required. This needs some familiarity with the Einstein summation convention, a skill
worth adquiring.
We shall call any set

J = (J
1
, J
2
, J
3
) of observables which satisfy
[J
i
, J
j
] = i
ijk
J
k
(106)
an angular momentum operator, and we will study them in the abstract.
More Properties
If [J
i
, A
j
] = i
ijk
A
k
and [J
i
, B
j
] = i
ijk
B
k
then
[J
i
,

A

B] = 0 (107)
Proof
[J
i
,

A

B] = [J
i
, A
j
B
j
] = [J
i
, A
j
]B
j
+ A
j
[J
i
, B
j
] = i
ijk
(A
k
B
j
+ A
j
B
k
) = 0
where the last step follows because
ijk
changes sign under switching a pair of indices and
(A
k
B
j
+ A
j
B
k
) does not.
We deduce in particular that

P

P ,

P

X,

X

X,

L

L and similar expressions all commute

with

L.
39
5.2 Ladder Operators
We have just seen that
[J
2
, J
i
] = 0 (108)
writing J
2
for

J

J. We dene the ladder operators
J

= J
1
iJ
2
(109)
and note that (J
+
)

= J

assuming the J
i
are self-adjoint.
Properties
1. [J
2
, J

] = 0,
2. J
+
J

= J
2
J
2
3
+J
3
,
J

J
+
= J
2
J
2
3
J
3
,
[J
+
, J

] = 2J
3
,
3. [J
3
, J

] = J

1 is done and 2 and 3 follow rapidly from the denitions.

Now, much as with the algebraic treatment of the Harmonic oscillator, we will nd eigen-
vectors of J
2
and J
3
. They commute so we will be looking for simultaneous eigenvectors.
Assume these exist in some 1, so there is a with
J
2
=
2
, J
3
= m,
for real , m. Consider the new states

+
= J
+
,

= J

,
then
J
2

= J
2
J

= (J
2
J

J
2
+ J

J
2
) =
2

(110)
i.e. since [J
2
, J

] = 0,

is an eigenvector of J
2
with the same eigenvalue. This is true for

+
too.
Next
J
3

= J
3
J

= (J
3
J

J
3
+ J

J
3
) = (m1)

(111)
so the J
3
eigenvalue is lowered by one for

(and raised by one for

+
).
Next calculate
[[

[[
2
= J

[J

= [J
+
J

= ( m
2
+ m)
2
[[[[
2
(112)
40
which must be non-negative, as
[[
+
[[
2
= J
+
[J
+
= [J

J
+
= ( m
2
m)
2
[[[[
2
(113)
which also must be non negative.
Putting these last two results together, we have
m(m1) and equality i J

= 0 (114)
m(m+ 1) and equality i J
+
= 0
This is enough to obtain the results below.
5.3 Representation of the Angular Momentum
Here representations means as operators acting on some explicit vector space, so we suppose
we have a nite-dimensional, complex vector space 1 of eigenvectors of J
2
:
J
2
=
2
for all 1.
J
3
will have an eigenvector in 1 say with
J
3
= m
We can obtain more eigenvectors of J
3
by raising and lowering with J

, when m will go up
and down in integer steps. This cant be repeated indenitely, since (114) implies
+
1
4
m
2
m +
1
4
= (m
1
2
)
2
so that [m[ cant be arbitrarily large. Thus there will be a maximum m, call it j with
corresponding
j
, and further raising must give zero:
J
+

j
= 0
By (113), if the LHS vanishes, we nd
= j(j + 1)
We lower
j
in integer steps and this will stop at a value of m at which further lowering
gives zero. By (112) this will happen when
0 = m
2
+ m = j(j + 1) m
2
+ m = (j + m)(j m + 1)
41
i.e. at m = j since m = j + 1 is not allowed (m must be less than its maximum value
which was j)
Thus we lower in integer steps from m = j to m = j, which requires 2j to be a non-negative
integer.
Summarizing:
the eigenvalues of J
2
are j(j + 1)
2
where j = 0,
1
2
, 1,
3
2
, 2, ...;
for each xed j, the eigenvalues of J
3
are m with m = j, j + 1, ..., j 1, j;
assuming no degeneracies, the eigenspace on which J
2
= j(j +1) has dimension 2j +1.
we can choose an orthonormal basis
m
whith
J
3

m
= m
m
J

m
= c

m1
c

m
= [(j m)(j m + 1)]
1/2
We have proved all of this except the last part which comes from (113) and (112).
Important example: j =
1
2
(also called the spin representation)
Now 2j +1 = 2 so 1 is 2-dimensional. The allowed values of m are m =
1
2
so write

for

1/2
, then
J
3

=
1
2

(115)
and
J
+

+
= J

= 0, J
+

=
+
, J

+
=

(116)
We can introduce a matrix formalism by taking as basis

+
=
_
1
0
_
,
+
=
_
0
1
_
Then (115) and (116) x the form of the operators to be
J
3
=
1
2

_
1 0
0 1
_
, J
+
=
_
0 1
0 0
_
, J

=
_
0 0
1 0
_
so also
J
1
=
1
2

_
0 1
1 0
_
, J
2
=
1
2

_
0 i
i 0
_
and one may check all the commutation relations. These formulae bring to our attention the
Pauli spin matrices:
42

1
=
_
0 1
1 0
_
,
2
=
_
0 i
i 0
_
,
3
=
_
1 0
0 1
_
(117)
which are Hermitian, trace zero and dene the J
i
for j = 1/2 via
J
i
=
1
2

i
(118)
from which many algebraic relations follow.
5.4 Orbital Angular Momentum and Spherical Harmonics
Let us focus in the particular case of orbital angular momentum

J =

L =

X

P. We seek
a space of functions for

L to act on. As we will shortly see, we need j to be an integer. To
justify this claim recall
L
3
= X
1
P
2
X
2
P
1
= i(x

y
y

x
) = i

(119)
in terms of spherical polar coordinates. Now the eigenvalue equation becomes
L
3
= i

= m (120)
so that
= F(r, )e
im
(121)
Now the important point: if this is genuinely a function, i.e. single-valued, then we need m
to be an integer, so j is also an integer. (We also see that j = 1/2 must be associated with
something, not a function, which changes sign under rotation by 2.) To pursue this line
further, lets calculate the other operators in spherical polar coordinates
L
+
= e
i
_

+ i cot

_
L

= e
i
_

i cot

_
(122)
L
2
=
2
_

2

2
+ cot

+
1
sin
2

2
_
We will nd the eigenfunctions explicitly, so write
m
for the one with
L
2

m
= ( + 1)
2

m
, L
3

m
= m
m
(123)
then the highest m-value is and

satises
43
L
3

= i

, L
+

= 0 (124)
The rst is solved by

= F()e
i
, where we are only looking at the , dependence (the
r-dependence will be reinstated in the next section), and then the second is
L
+

= e
i
_
dF
d
(cot )F
_
e
i
= 0
whence F = const. (sin )

, so

= const. (sin )

e
i
. (125)
The constant can be xed by normalization. Once we have this we obtain the other
m
by
lowering with L

. We shall write Y
m
(, ) for the corresponding functions. Since they arose
as eigenfunctions we know at once that

1
m
1
[

2
m
2
=
_
2
0
_

0
Y

1
m
1
Y

2
m
2
sin dd = 0, if (
1
, m
1
) ,= (
2
, m
2
). (126)
5.5 The Hydrogen Atom Again
With a better understanding of angular momentum we can do a more thorough treatment
of the hydrogen atom. The time-independent Schrodinger equation for a particle of mass m
and charge e (the electron) in the electric eld of a xed point of charge Ze (the nucleus,
so Z = 1 for the hydrogen atom) is

2
2m

Ze
2
4
0
r
= E (127)
and the Laplacian in spherical coordinates is

2
=

2
r
2
+
2
r

r
+
1
r
2
_

2

2
+ cot

+
1
sin
2

2
_
(128)
From (122), we recognize that the angular derivatives are proportional to L
2
, so that we can
separate = F(r)Y
m
(, ). We then obtain

2
=
_
1
r
(rF)

( + 1)
r
2
F
_
Y
m
(129)
writing prime for d/dr.
We simplify the Schrodinger equation by renaming quantities:
E =

2
2m
, =
2m

2
Ze
2
4
0
, (130)
44
Further, we write F(r) in terms of a new function:
F(r) = f(r)e
r
(131)
We nd that f satises the following equation
f

+
_
2
r
2
_
f

_
2
r
+
( + 1)
r
2
_
f = 0 (132)
We seek a series solution
f =

n=0
a
n
r
n+c
, with a
0
,= 0, (133)
and substitute into (132) to obtain

_
a
n
(n + c)(n + c 1)r
n+c2
+
_
2
r
2
_
a
n
(n + c)r
n+c1

_
2
r
+
( + 1)
r
2
_
a
n
r
n+c
_
= 0
(134)
the lowest power is r
c2
and the coecient is
a
0
(c(c 1) + 2c ( + 1))
This must vanish and a
0
,= 0 so we obtain the indicial equation:
(c(c 1) + 2c ( + 1)) = (c )(c + + 1) = 0 (135)
The root c = ( + 1) is not allowed as this would give unbounded at r = 0, thus c = .
For the general power r
n+c1
we obtain the recurrence relation:
a
n+1
((n + c + 1)(n + c) + 2(n + c + 1) ( + 1)) = a
n
(2(n +c) + 2 ), (136)
which determines all terms in the series from a
0
. If the series continues to innity, then for
large n
a
n+1
a
n

2
n
and so
f e
2r
Tracking back through (131), we see that this will give a non-normalizable . So normal-
izability requires the series to terminate and from the recurrence relation there must be a
non-negative n

for which the RHS vanishes, i.e.

2(n

+ + 1) = (137)
45
where we have also substituted for c.
Set n = n

+ + 1, which is a positive integer, then the series for f takes the form
f = f
n
= r

k=0
a
k
r
k
, (138)
which is a polynomial of degree n 1, with corresponding
=
n
=

2n
. (139)
The wave function is

nm
= const. f
n
(r)e

n
r
Y
m
(, ) (140)
and the energy eigenvalue is
E = E
n
=

2
n
2m
=

2
8m
1
n
2
(141)
The integers n, , m, which label the state, are known respectively as the principal quantum
number, the azimuthal quantum number and the magnetic quantum number. The energy
eigenvalue notices only n, not or m. Any spherically symmetric potential will ignore m,
which only comes into E when spherical symmetry is given up, for example by putting the
atom in a magnetic eld (whence the name), but spherical potentials other than the pure
Coulomb (i.e. V
1
r
) will have appearing in E.
We can count the degeneracy of an energy eigenvalue as follows: given n = n

+ + 1 we
have 0 n 1 and for each there are 2 + 1 choices for m, so that the degeneracy is
n1

=0
(2 + 1) = n
2
6 A Charged Particle in a Magnetic Field
(this section is not part of the course and so not examinable, but it makes a link back to the
material on electromagnetism, and introduces some interesting new ideas.)
To connect the second half of the course with the rst, we briey consider the quantum
version of a classical charged particle of mass m and charge e moving in a electromagnetic
eld. In the case of an electric eld, with electric potential say , we know how to do this:
simply consider the Hamiltonian
H =
1
2m
[

P[
2
+ V (

X)
with V (

X) = e(

X). The hydrogen atom is just like this.

46
In the case of a magnetic eld, recall that the force does not work on the particle, so does not
contribute a potential energy term: something new is needed. To motivate what this is, and
because it gives some insight into how classical mechanics can be recovered from quantum
mechanics, look at (99):
d
dt
A =
A
t
+
i

[H, A] (142)
With the Hamiltonian above, this will give
d
dt

X =
i

[H,

X] =
i

[
1
2m
[

P[
2
,

X] =
1
m

P,
and
d
dt

P =
i

[H,

P] =
i

[V (

X),

P] = V
Combining these two we obtain
m
d
2
dt
2

X = V (143)
which closely parallels the classical equations of motion for a particle in a potential, which
read
m
d
2
dt
2
x = V (144)
In other words, the expectation values satisfy the classical equations of motion. Now we
consider the Hamiltonian
H =
1
2m
_
(P
1
+
1
2
eBY )
2
+ (P
2

1
2
eBX)
2
+ P
2
3
_
(145)
which is similar to one introduced on one of the questions sheets, where we found the energy
levels without the term P
2
3
. What classical problem does this corresponds to? We nd
d
dt
X =
i

[H, X] =
1
m
P
1
+
1
2
eBY V
1
(146)
where the right-hand-side serves to dene the velocity V
1
. Similarly
d
dt
Y =
i

[H, Y ] =
1
m
P
2

1
2
eBX V
2
(147)
while
d
dt
Z =
i

[H, Z] =
1
m
P
3
V
3
(148)
Then
47
d
dt
P
1
=
i

[H, P
1
] =
1
2
eBV
2

d
dt
P
2
=
i

[H, P
2
] =
1
2
eBV
1

d
dt
P
3
= 0
so that
m
d
2
dt
2
X = eBV
2

m
d
2
dt
2
Y = eBV
1
(149)
m
d
2
dt
2
Z = 0
This system corresponds to the classical equation
m
d
2
dt
2
x = ev

B (150)
with

B = B

k, which is the equation of motion of a classical particle of mass m and charge

e moving in the magnetic eld

B, which is uniform and in the z-direction

B = (0, 0, B)
To see how to get (145) note that this magnetic eld has vector potential

A =
1
2
B(y, x, 0) (151)
so that (145) may be obtained by starting with the free particle Hamiltonian H =
1
2m
[

P[
2
and making the replacement

P

P e

A (152)
It is a straightforward calculation to see that this is the correct prescription more generally
so that one could, for example, go on to calculate the energy levels of a hydrogen atom in a
uniform magnetic eld. We wont do that, but to end we will note a new interpretation to
gauge transformations that this gives. Recall a gauge transformation was a change in

A:

A =

A + (153)
48
which leaves the magnetic eld unchanged. This should therefore lead to no physical change,
but it will change the term
1
2m
[

P e

A[
2
in H. The resolution of this puzzle is to make a
simultaneous change in the wave function:


= exp(ie/) (154)
This change doesnt aect the probability density or probability current

j, which are the
physical observable quantities, and also
[

P e

A[
2

= exp(ie/)[

P e

A[
2
(155)
so that the term exp(ie/) factors out of the Schrodinger equation, and solutions will
transform to solutions.
49