Part III quantum mechanics preparatory workshop

notes
Sean Seet Roland Bittleston Kai Roehrig

Michaelmas 2021

In this workshop we will revise the major lessons of QM before you get
started with the Part III theoretical physics courses. We will do this by
recapping the main ideas, and then working through a series of examples.
This is designed to refresh knowledge that may have been lost or gone rusty.
You will not find anything here beyond standard QM material, it is however
presented in a way that is meant to maximise the applicability to Part III
(mainly QFT) and may differ from the way you learnt it in your previous
QM courses.

The topics we will touch are roughly:

ˆ Hilbert Space, Observables, etc.

ˆ Dirac (Bra-Ket) Notation, Schrödinger & Heisenberg Picture

ˆ Probabilities & Measurement

ˆ Harmonic Oscillator, Spin Algebra, Ladder Operators

ˆ Perturbation Techniques

Furthermore, you are encouraged to bring any questions or specific problems

about QM to the live session so we can talk about them in detail.

Please get in touch with comments and suggestions, or if you find errors
in this script, at sxes2@[Link].

1
Units & Notation
Since most formulas and expressions in QM and QFT have many factors of
c and ℏ, it is very convenient to work in a units such that c = ℏ = 1. In
other words, we measure all physical quantities in relation to the fundamen-
tal constants c, ℏ, etc., and appropriate combinations thereof. This is called
natural units. At the level of our formulas, you can imagine we rescaled
all coordinates, parameters, functions, etc. by a combination of the natural
constants (in the fashion of a coordinate change) to obtain new, unit-free
coordinates, parameters, functions. This is sometimes called ‘nondimension-
alization’.

We sometimes use ∇ ⃗ (read nabla) to mean the the following. For a function
⃗ )i = ∂xi f , the components of the gradient df . For a vector field ⃗v (x)
(∇f
⃗ × ⃗v .
we write its divergence as ∇ · v and its curl as ∇

Recommended Material
ˆ PQM lecture notes by David Skinner, [Link]/user/dbs26 (!)

ˆ Past PQM example sheets, [Link]/user/examples (!)

ˆ Griffiths, Introduction to Quantum Mechanics

ˆ Dirac, The Principles of Quantum Mechanics

ˆ Weinberg, Lectures on Quantum Mechanics

ˆ For QFT: Nair, QFT – A Modern Perspective

1 States, observables and measurements

The central equation in quantum mechanics is, of course, the Schrödinger
equation
(−i∂x )2
 
i∂t Ψ(t, x) = + V (x) Ψ(t, x) (1)
2m
describing the time evolution of the function Ψ(t, x) in terms of a partial
differential equation. It involves the potential energy V (x)Ψ(t, x) and the
1 2
kinetic energy − 2m ∂x Ψ(t, x). The sum of potential and kinetic energy is
called the Hamiltonian, and is already in classical mechanics responsible for
p2
the time evolution of a system. Recall that H(x, p) = 2m + V (x) and the
equation of motion for any quantity f (x, p) is

d ∂f ∂H ∂f ∂H
f = {f, H}PB = − (2)
dt ∂x ∂p ∂p ∂x

2
using the Poisson bracket {·, ·}PB between the two functions H and f . In
words we can say that the Hamiltonian function H generates time evolu-
tion by acting on other functions via the Poisson bracket {·, H}PB . So in
particular we get
p
ẋ = ṗ = −∂x V (x) (3)
m
describing a classical trajectory γ(t) = (x(t), p(t)) in phase space. Note that,
given two solutions γ1 (t) and γ2 (t), the sum γ1 + γ2 is not necessarily also
a solution of the equations of motion. In fact, this can generically only be
the case if the system eq. (3) is linear, i.e. the potential is at most quadratic.

Of course we know from Schrödinger’s (thought) experiments with his cat

that if two states of a system are physical, i.e. they solve the fundamental
equation of motion of nature, then so is any linear combination of the two.
Say the fundamental equation of motion of nature is given by some differ-
ential operator L acting on a function f describing the current state of a
system: L(f ) = 0. The previous comments imply that L must be a linear
operator and hence its solution space be a vector space! By now we are of
course already talking about the Schrödinger equation (1) with the operator
given by
(−i∂x )2
L = −i∂t + + V (x) . (4)
2m
We also know that the functions Ψ(t, x) on which this operator acts have a
very special interpretation, their mod-square ρ(t, x) = |Ψ(t, x)|2 is a proba-
bility distribution on configuration space. It gives the probability of finding
a particle at given point (or rather in a given subset) of configuration space.
For this last sentence to make sense, the total probability of finding the
particle at any point of configuration space needs to be unity, or at the very
least finite, so that we can normalise Ψ if need be1
Z Z
dx ρ(t, x) = dx |Ψ(t, x)|2 < ∞ . (5)
R R

It will turn out to be very rewarding to notice that the solutions of eq. (1)
have the analytic structure of a Hilbert space. Let’s recall the definition.
A Hilbert space H is a complex vector space with an inner product ⟨ψ|ξ⟩
between vectors, which satisfies

1. ⟨ψ|ξ⟩ = ⟨ξ|ψ⟩∗

2. ⟨ψ|α1 ξ1 + α2 ξ2 ⟩ = α1 ⟨ψ|ξ1 ⟩ + α2 ⟨ψ|ξ2 ⟩, with α1 , α2 ∈ C

3. ||ψ||2 = ⟨ψ|ψ⟩ ≥ 0 and ⟨ψ|ψ⟩ = 0 iff |ψ⟩ = 0

1
We loosen the condition of unit total probability so that the solution space remains a
vector space. This has to be taken into account later.

3
Clearly these conditions are very sensible and may seem nearly too obvious
to even mention. You will however soon encounter very important physical
theories which violate some of these conditions, and dealing with this issue
will take a good deal of work.

In QM, the two relevant Hilbert spaces are the familiar Cn , and the space
of complex valued functions on Rd with the inner product
Z
⟨ψ|ξ⟩ = dd x ψ ∗ (x)ξ(x) . (6)

The latter example is infinite dimensional, and we think of every function

ψ(x) as a vector in this infinite dimensional space. The passage from a finite
to infinite dimensional vector space can roughly be thought of as making the
index i of a vector ψi continuous, so the component form together a func-
tion ψ(x). However, since there are infinitely many coordinates, we have
to include a new requirement: for a vector to be sensible, it better have a
finite length, i.e. ⟨ψ|ψ⟩ < ∞, which we recognize as eq. (5). We denote the
Hilbert space of functions satisfying this constraint by L2 (Rd ).

From your linear algebra course you should be familiar with the concept
of a basis and a change of basis. In short, a vector is independent of a basis,
but the components which we use to describe the vector are not. A basis
is specified by a set of basis vectors, i.e. a linearly independent spanning
set of vectors, and the components of any vector in that basis are simply
the projection of the vector on the basis elements. The components of a
vector in two different bases are related by a matrix, whose entries are the
inner products of basis elements. In the case of complex functions, you
should be familiar with the two most common bases. The position space
and momentum space basis. The components of a vector |ψ⟩ are given by

ψ(x) = ⟨x|ψ⟩ and ψ̃(p) = ⟨p|ψ⟩ (7)

in the respective bases. We put the tilde over the Fourier transform of ψ
to emphasise that it is a different function, though related, to ψ. Since we
claimed that the set of basis elements for position space |x⟩ and momentum
space |p⟩ form a basis, they had better satisfy the completeness relation:
Z Z
dd x |x⟩⟨x| = 1 and dd p |p⟩⟨p| = 1 . (8)

Finally, if ψ(x) and ψ̃(p) are the components of the same vector, just in
expressed in different bases, then there should be a linear transformation
relating the two. Indeed, using eq. (8), we find
Z Z
ψ̃(p) = ⟨p|ψ⟩ = ⟨p|1|ψ⟩ = d x ⟨p|x⟩⟨x|ψ⟩ = dd x ⟨p|x⟩ψ(x)
d
(9)

4
which we recognise as the Fourier transform of ψ(x) given that the matrix
elements of the change of basis are ⟨p|x⟩ ∝ e−ip·x . Hence taking the Fourier
transform of a function corresponds simply to changing basis in this Hilbert
space.

Q 1.1 Use the canonical commutation relation [x̂, p̂] = i to show that

eiap̂ x̂e−iap̂ = x̂ + a (10)

and hence, without loss of generality,

e−iap̂ |x⟩ = |x + a⟩ . (11)

From this deduce that ⟨p|x⟩ ∝ e−ixp and finally fix the normalization by
unitarity.

Observables
While the state of a physical system is represented by a vector in a Hilbert
space, the possible observables, i.e. measurable quantities, are represented
by operators (think infinite dimensional matrices). When the system is in
a given state ψ, the expected value of an observable O, represented by the
operator Ô, is defined by

⟨ψ|Ô|ψ⟩
⟨Ô⟩ψ = . (12)
⟨ψ|ψ⟩
If this quantity is supposed to be measurable in the real world, it had better
be a real number. Setting the imaginary part of ⟨Ô⟩ψ to zero leads us to
the requirement Ô = Ô† , i.e. that the operator representing an observable
be Hermitian, or self-adjoint.

Hermitian operators acting on a Hilbert space have a few properties that

are very handy for calculations, and also crucial for their interpretation as
observables. As soon as we know that an operator is Hermitian, we know
that

ˆ its eigenvectors |ψi ⟩ form a complete basis, 1 = i |ψi ⟩⟨ψi |,

P

ˆ its eigenvectors (eigenspaces) are mutually orthogonal ⟨ψi |ψj ⟩ = δij ,

ˆ its eigenvalues λi are real.

Q 1.2 Prove the second and third property of Hermitian operators above.

The most fundamental observable of a system is of course its total energy,

represented by the Hermitian operator Ĥ, the Hamiltonian. Further ob-
servables are position, momentum, spin, etc, and all these quantities are

5
represented by Hermitian operators. There are also many important non-
Hermitian operators, a particularly significant class are the Unitary opera-
tors, which preserve the norm on H.

Often it is convenient to define an operator in terms of is spectral decom-

position, that is, its eigenvectors and eigenvalues, e.g. the position operator
x̂ is defined in the position basis as
Z
ˆ = dd x ⃗x|x⟩⟨x| .
⃗x (13)

These spectral decompositions are very useful in explicit computations.

Finding the matrix elements of an operator in a given basis works just as
in usual linear algebra, the only difference being that the indices on a ma-
trix have now become continuous parameters. Hence matrix elements are
functions (or more generally distributions) in two variables, e.g. O(x, y) =
⟨x|Ô|y⟩.

Q 1.3 For |ψ⟩ ∈ L2 (R), write down the vectors |ψ⟩, x̂|ψ⟩, p̂|ψ⟩, as well as
the operators x̂, p̂, in both position and momentum basis.

Note that in a given basis, we can understand

R d the ‘matrix elements’ of
an operator as an integral kernel (Ôψ)(x) = d y O(x, y)ψ(y).

Q* 1.4 What are ρ(x) = ⟨ψ|δ (3) (x̂ − x)|ψ⟩ and ⃗j(x) = 2m
1
⟨ψ|{p⃗ˆ , δ (3) (x̂ − x)}|ψ⟩
in terms of the position space wave function ψ(x)? Can you write ∇ · j(x)
in a similar way? Here {A, B} = AB + BA is the anti-commutator.

Another way of defining an operator is to declare how it acts on a set of

basis vectors. Again for the position operator, we can define it in terms of
ˆ|x⟩ = ⃗x|x⟩. Naturally, the definition
its action on the position space basis, ⃗x
in terms of its spectrum is a special case of this.

Q 1.5 Working in position basis show that p̂ is Hermitian, i.e. that ⟨ψ|p̂|ξ⟩ =
(⟨ξ|p̂|ψ⟩)∗ for arbitrary |ξ⟩, |ψ⟩ ∈ L2 (R).

Incidentally, we can now use the basis independent notation in our

Hilbert space, which is also known as Dirac notation, to write the Schrödinger
equation (1) compactly as

i∂t |Ψ⟩ = Ĥ|Ψ⟩ , (14)

1 2
where Ĥ = 2m p̂ + V (x̂).

Q* 1.6 Recall that in 3d a particle with charge e moving in a time indepen-

dent magnetic field has Hamiltonian H(x, p) = 2m 1 ⃗
p − eA(x))
(⃗ 2 , where the
⃗
electromagnetic potential A determines the magnetic field via B ⃗ =∇ ⃗ × A.
⃗

6
 ⃗ = −∂t A
(The electric field is given by E ⃗ and hence vanishes since A
⃗ is time
independent.)
¨ = e(E
a) Derive the classical equation of motion F⃗ = m⃗x ⃗ + ⃗x˙ × B).
⃗

b) What is the momentum p⃗A = m⃗x˙ in terms of ⃗x and p⃗. Calculate the
Poisson bracket {(pA )i , (pA )j }PB .
c) What is the quantum Hamiltonian in terms of p⃗ˆ and A(x̂)?
⃗ Is there a
normal ordering ambiguity in promoting the functions to operators? (Hint:
remember the Coulomb gauge fixing condition ∇ · A = 0.)
d) After canonically quantizing ( [x̂i , p̂j ] = iδ ij ), what is p⃗ˆA in position repre-
sentation? Show that the commutator becomes [(p̂A )i , (p̂A )j ] = ieϵijk Bk (x̂).

This is a simple example of a system with a gauge symmetry, which will

become very important in QFT and GR and you will learn a lot about it
over the next terms. The non-vanishing of the commutator [(pA )i , (pA )j ]
indicates that the particle moves in a curved space. In this example the
curvature does not reside in (position-)space, but rather in an ‘internal’
space. See e.g. Nair or Nakahara for more.

Probabilities
Given that we can calculate the expectation value of an observable in a state
ψ using eq. (12), it is natural to ask what the outcome of a single measure-
ment will be. The probabilistic nature of QM suggests that this question
is ill-posed. We can only sensibly ask for the probability of measuring some
outcome. Recall that the expectation is in general a sum over all possible
outcomes, weighted by their probability. Inserting the spectral decompo-
sition of the operator  in terms of its eigenvalues ai and (orthonormal)
eigenvectors |i⟩ yields
X X |⟨i|ψ⟩|2
λi p(λi ) = ⟨Â⟩ψ = ai . (15)
∥ψ∥2
i i

We recognise that the probability of measuring a given eigenvalue ai is given

by
|⟨i|ψ⟩|2
p(ai ) = = ⟨|i⟩⟨i|⟩ψ = ⟨PA (i)⟩ψ (16)
∥ψ∥2
where we defined the projection operator on the ith eigenstate of Â.

Q 1.7 Check that PA (i) is indeed a projection operator (i.e. P 2 = P ) and

observable (i.e. P † = P ). Construct the projection operator projecting on
two states PA ({i, j}), then on an arbitrary subset of the spectrum of Â.

7
 If we want to compute the probability to find an eigenvalue within a
certain range of the spectrum, we likewise take the expectation value of the
projection operator onto that part of the spectrum.
Q 1.8 a) Show that ⟨ψ|ψ⟩ is constant in time using eq. (14) (with general
Ĥ) and the product rule for differentiation. What property of Ĥ is crucial?

b) Using the Dirac notation form for ∇ · j, evaluate ∂t ρ + ∇ · j using the

1 2
Schrödinger equation (14) with a Hamiltonian Ĥ = 2m p̂ + V (x̂). What is
⃗
the interpretation of ρ, j, and the differential equation they satisfy?

c) What is the relation between a) and b)?

d) What happens if the Hamiltonian is not Hermitian? How can we in-

terpret this?
Now that we know the probability for each possible outcome of a mea-
surement, we can go ahead and actually perform it! A crucial part of (the
Copenhagen interpretation of) QM is that measurements cause the wave
function to collapse. Suppose we measure A, and observe a value ai (corre-
sponding to the ith eigenvalue of Â). The measurement will cause the wave
function to collapse

|ψ⟩ −→ PA (i)|ψ⟩ = |i⟩⟨i|ψ⟩ . (17)

This is an instant projection. It represents a non-continuous, non-unitary

evolution. (There exist other interpretations of QM which avoid wavefunc-
tion collapse, see e.g. decoherence, however there exist no generally accepted
resolutions of the so called measurement problem.) Note that before and
after the measurement the time evolution is given by the Schrödinger equa-
tion.

Uncertainty
An immediate consequence of the representation of observables by operators
(again, think matrices) is that observables no longer necessarily commute,
e.g. the commutator between the position and momentum operator is

[x̂, p̂] = i (18)

and is important enough to carry the name canonical commutation relation.

(It is called canonical because is follows from replacing the Poisson bracket
in classical mechanics by the commutator

{x, p}PB = 1 → −i[x̂, p̂] = 1 (19)

which is an ad hoc, yet canonical, procedure.)

8
Q 1.9 Using the position basis expressions for x̂ and p̂ find the position
space expression for the commutator [p̂, f (x̂)], for a function f ∈ C ∞ (R),
by letting it act on an arbitrary vector. Compare this with what you get
without choosing a basis, by writing f as a Taylor series (assuming that it
is analytic).

The effect of this is that the uncertainty in different observables may

become correlated, even if they were classically uncorrelated. Recall that
the uncertainty (variance) for the value of an observable A is given by
2
σA (ψ) = ⟨(Â − ⟨Â⟩ψ )2 ⟩ψ = ⟨Â2 ⟩ψ − ⟨Â⟩2ψ = ⟨Â20 ⟩ψ (20)

where for convenience we introduced the shorthand Â0 = Â − 1⟨Â⟩ψ . The

interpretation of this uncertainty is that some properties of the system sim-
ply do not have an exact value, but are smeared out over an interval σA
around the expected value, depending on the state.2

Q 1.10 Show that the uncertainty of the observable A in the state ψ satisfies
2 (ψ) ≥ 0, with σ 2 (ψ) = 0 iff ψ is an eigenvector of Â.
σA A

Now, it follows from the positivity of the norm (∥ · ∥ ≥ 0) that the

product of the uncertainties of two observables obeys
 2  2
2 2 1 1
σA (ψ)σB (ψ) ≥ ⟨{Â0 , B̂0 }⟩ψ + ⟨[Â, B̂]⟩ψ , (21)
2 2i

the Schrödinger uncertainty relation. Note that the symmetric part {Â0 , B̂0 },
is in direct analogy with the classical version of this relation. The quantum
contribution comes from the commutator [Â, B̂], and has no classical ana-
logue. This becomes particularly interesting when  and B̂ are conjugate,
so that the commutator is i. Then we can write a weaker, but more practical
bound
σA (ψ)σB (ψ) ≥ 1/2 (22)
where the right hand side is now independent of the state ψ.

Q* 1.11 Show eq. (21) by considering the function f (λ) = ⟨|Â0 + λB̂0 |2 ⟩ψ
and its minimum, for λ ∈ C.
2
In particular, it doesn’t mean that we are not able, technologically or otherwise, to
measure the precise value – the system just does not have a fixed value for A. The right
picture is that of a wave, it simply does not make sense to ask for its ‘location’ to higher
accuracy than its wavelength.

9
2 Unitarity, the Heisenberg picture and perturba-
tion theory
Unitary transformations
Many computations become much easier if we choose a good basis for the
Hilbert space. The basis independence of the result is guaranteed if we
only allow basis changes that preserve the norm ∥ · ∥. This amounts to the
requirement of unitarity of the operator U implementing the basis change

U †U = 1 . (23)

Unitary operators on our Hilbert space are the analogue of orthogonal ma-
trices on Rn .

Now, the Schrödinger equation (14) says that the evolution of a physical
system is given by the operator

Û (t) = exp(−itĤ) (24)

which is sensibly called the time evolution operator.

Q 2.1 a) Check that |Ψ(t)⟩ = Û (t)|ψ⟩ solves the Schrödinger equation (14).
b) Check that the evolution operator is unitary if Ĥ is Hermitian.
c) Check that the evolution operator for evolving by a time t1 + t2 can be
obtained by first evolving by t1 and then by t2 .

In other words, the action of Û (t) on states generates time translations.

Q 2.2 a) What is the probability to find a system in an eigenstate |j⟩ of an

observable A at time T > 0 if it was in the eigenstate |i⟩ at time 0? (Hint:
what is ψ(t) for T > t > 0?)

b) What if |i⟩ is an energy eigenstate?

Q* 2.3 Use the result from the last question to compute the probability
p̂2
density for a free particle, Ĥ = 2m , initially in a position eigenstate at
x = 0, to be measured at position x at time t. (You may have to go to
imaginary time t → −itE to compute an integral and go back to real time
tE → it.)

There is in general a close relationship between unitary operators and

transformations. Note that for every Hermitian operator  there is an as-
sociated family of unitary operators ÛA (θ) = exp(−iθÂ). Physically, ÛA (θ)
generates a shift by θ in the observable conjugate to A, e.g. the momentum
operator p̂ generates translations of x̂.

10
Q 2.4 Show that the operator Ûp (a) = exp(−iap̂) indeed generates a shift
of the observable x̂ by the amount a in the sense that Ûp (a)|x⟩ = |x + a⟩ as
well as
e−ia[p̂,·] f (x̂) = Ûp (a)f (x̂)Ûp (a)† = f (x̂ − a) . (25)
(Hint: you can show that the operator relations hold when acting on a com-
plete set of basis vectors, and hence, by linearity, on all of the Hilbert space.)
The exponential of the commutator action is defined as
(ia)2 (ia)3
e−ia[p̂,·] f (x̂) = f (x̂)−ia[p̂, f (x̂)]+ [p̂, [p̂, f (x̂)]]− [p̂, [p̂, [p̂, f (x̂)]]]+· · · .
2 3!
(26)
As observed earlier, the operator ÛA is unitary, i.e. preserves total proba-
bility, if  is Hermitian, i.e. is an observable.

This is again in direct analogy with classical mechanics, where every ob-
servable (function on phase space), generates a canonical transformation on
the phase space via the Poisson bracket

O → ea{f,·}PB O
a2 a3
= O + a{f, O}PB + {f, {f, O}PB }PB + {f, {f, {f, O}PB }PB }PB + · · ·
2 3!
(27)
for any function O(x, p). Canonical transformations are coordinate transfor-
mations on phase space which preserve the symplectic (or Poisson bracket)
structure. This fits in place nicely with the statement that unitary operators
effect basis changes on Hilbert space.

If you want to learn more about this perspective on quantum mechanics

you should look up geometric quantization, e.g. in chapter 20 of Nair’s
‘modern perspective on QFT’, or in ‘Symplectic geometry and geometric
quantization’ by Blau.

Heisenberg Picture
Extending the line of argument from the previous section, we could ask for
the time evolution of observables directly. Using the Schrödinger equation
(14) we find
d
⟨O⟩Ψ = −i⟨[O, Ĥ]⟩Ψ (28)
dt
where the time dependence sits purely with the state |Ψ(t)⟩ and O can be
any operator. It is not hard to see that you get the same answer if instead
of using the Schrödinger equation, you used the Heisenberg equation
d
O(t) = −i[O(t), Ĥ] (29)
dt

11
for the time dependence of an operator, with the states frozen at a fixed
point in time.
d d
Q 2.5 Check that both a) (14) with dt O = 0 and b) eq. (29) with dt |Ψ⟩ = 0
lead to the right equation of motion for expectation values eq. (28).
It is not difficult to see that the solution of the Heisenberg equation is
given by
O(t) = Û (t − t0 )† O(t0 )Û (t − t0 ) (30)
which could of course be obtained directly by inspection of the matrix ele-
ment ⟨Ψ|O|Ψ⟩, and stripping the time evolution operator off the state, onto
the operator.

In summary, we can either view the states as evolving in time and the
operators as being constant

|Ψ(t)⟩ = Û (t)|ψ⟩ , O(t) = O(0) , (31)

or vice versa

|Ψ(t)⟩ = |ψ⟩ , O(t) = Û (t)† O(0)Û (t) . (32)

Since these two formulations describe the same dynamics, the difference be-
tween them is completely unobservable. They are merely different pictures
of the same underlying physics. They carry the names Schrödinger and
Heisenberg picture respectively.

A common trick in QFT is to work in an intermediate framework, called

the interaction picture. In this picture the evolution due to the simpler part
of the Hamiltonian (usually the quadratic bit) is put into the operators, and
the evolution due to the more complicated parts (the interactions) are put
into the states.
1 2
Q 2.6 Using Ĥ = 2m p̂ +V (x̂), determine the commutation relation [x̂(t), p̂(t)]
from the requirement that the operators x̂(t) and p̂(t) satisfy an equation
analogous to the classical equation of motion eq. (3). Note the close simi-
larity to the classical computation involving the Poisson bracket.

Perturbation Techniques
There are not many quantum systems that admit an exact analytic solution
(finding the eigenstates and their energies), so in general a more widely
applicable method for extracting the physics from a Hamiltonian is needed.
One way is to split the Hamiltonian Ĥ in question into a part that we can
solve, Ĥ0 , and treat the remainder as a perturbation.

Ĥ = Ĥ0 + ĤI . (33)

12
Of course this split is arbitrary, but typically Ĥ0 will describe a particle
with either no, or quadratic potential. For this to be useful, there needs to
be some sense in which the contribution from ĤI is much smaller than that
from Ĥ0 , so we expand the quantities of interest in a series.

In QFT you will be mainly concerned with calculating transition ampli-

tudes. Recall from Q 2.2 that the probability to measure a particle in state
|i⟩ at time 0 and then in state |j⟩ at time t is
Pi→j (t) = |⟨j| exp(−itĤ)|i⟩|2 (34)
where |i⟩ are the normalized eigenvectors of some observable, e.g. position
eigenstates, momentum eigenstates or energy eigenstates.

To make use of the split into a known Hamiltonian and a perturbation

we need to use the operator identity
Z 1
exp(A + ϵB) = eA + ϵ ds e(1−s)A BesA + O(ϵ2 ) (35)
0
or equivalently
Z t
−itĤ0 ′ ′
exp(−it(Ĥ0 + ĤI )) = e + dt′ e−i(t−t )Ĥ0 ĤI e−it Ĥ0 + O(ĤI2 ) . (36)
0

A mnemonic for this is that the perturbation ĤI is inserted at every possible
point along the interval of evolution [0, t], dressed by evolution under the
free H0 before and after.
Q* 2.7 Show the above relation e.g. by expanding both sides in their Taylor
expansions and counting powers of A and B. (Hint: you may need the
binomial coefficient.)
This expansion is particularly useful if the states of interest are eigen-
states of the unperturbed system Ĥ0 .
Q* 2.8 Assuming that |i⟩, |j⟩ are energy eigenstates of Ĥ0 with different
energy eigenvalues, insert eq. (36) into the transition amplitude and find the
leading order correction, expressed in terms of the matrix element ⟨i|ĤI |j⟩.
(Assume ⟨i|j⟩ = 0.)
The result is important enough to carry the name Fermi’s golden rule.
You should find
sin(ωt/2)2
Pi→j (t) = |⟨j|ĤI |i⟩|2 (37)
(ω/2)2
with the frequency of oscillation given by ω = Ej − Ei . As an example,
consider the Hamiltonian
   
E 0 0 δ
Ĥ = Ĥ0 + ĤI = + (38)
0 −E δ̄ 0

13
where |δ| ≪ E.

Q 2.9 Find the energies and eigenstates for δ = 0. Then use the result
of Q 2.8 to find the transition amplitude between the two energy levels to
leading order in δ.

You will have noticed that the computation is tremendously simplified

when computing transitions between eigenstates of the unperturbed system.
Recall that the Schrödinger and Heisenberg picture are different, yet equiv-
alent, assignments of time evolution between the state and the operators.
We can now make use of the above observation by introducing a half-way
picture. Since we know the evolution of the state |Ψ(t)⟩ by the unperturbed,
exactly solvable Hamiltonian Ĥ0 , we absorb this into the definition of the
state, i.e.
|Ψ(t)⟩I = exp(itĤ0 )|Ψ(t)⟩S , (39)
where |Ψ⟩I and |Ψ⟩S are the state in the interaction and Schrödinger pic-
tures respectively. Only the interaction term ĤI contributes to the time
dependence of the state in the interaction picture. Comparison with the
Schrödinger equation yields that in this picture the state and the operators
have to satisfy the equations
d
i∂t |Ψ⟩I = ĤI |Ψ⟩I and OI = i[Ĥ0 , OI ] . (40)
dt
Q 2.10 Verify this.

For the next two terms you will be working mainly in this picture.

3 Tensor products, harmonic oscillator and spin

Tensor Products & Identical particles
When combining two physical systems, represented by Hilbert spaces H1
and H2 , and Hamiltonians Ĥ1 and Ĥ2 respectively, we get a system with
Hilbert space
H = H1 ⊗ H2 (41)
the tensor product of the individual spaces, and a Hamiltonian

Ĥ = Ĥ1 ⊗ 1 + 1 ⊗ Ĥ2 . (42)

If bases of the Hilbert spaces H1 and H2 are given by {|i⟩}i∈I and {|j⟩}j∈J
respectively, then a basis of the tensor product is given by

{|i, j⟩ = |i⟩ ⊗ |j⟩}i∈I,j∈J . (43)

14
A general element of the combined Hilbert space can be expressed as a linear
combination of these X
|ϕ⟩ = ci,j |i⟩ ⊗ |j⟩ (44)
i,j

where the ci,j are complex coefficients. Clearly, the number of basis ele-
ments and hence the dimension of the combined Hilbert space is dim H =
dim H1 · dim H2 . (The generalisation to combining several physical systems
follows analogously.)

For the purpose of calculations you can think of the several states and op-
erators living in different worlds; they behave within their Hilbert spaces as
you are used to, e.g.
     
 ⊗ B̂ |ψ⟩ ⊗ |ξ⟩ = Â|ψ⟩ ⊗ B̂|ξ⟩ (45)

and
(⟨ψ| ⊗ ⟨ξ|) |ψ ′ ⟩ ⊗ |ξ ′ ⟩ = ⟨ψ|ψ ′ ⟩⟨ξ|ξ ′ ⟩ .


(46)
In particular the energy of a state |ψ⟩ ⊗ |ξ⟩ is given by
 
Ĥ1 ⊗ 1 + 1 ⊗ Ĥ2 (|ψ⟩ ⊗ |ξ⟩) = (E1 (ψ) + E2 (ξ)) (|ψ⟩ ⊗ |ξ⟩) (47)

where ψ and ξ are eigenstates of Ĥ1 and Ĥ2 respectively. Often the triv-
ially acting part is omitted, so we abbreviate Ĥ1 ⊗ 1 + 1 ⊗ Ĥ2 as simply
Ĥ1 + Ĥ2 and use the index to remind us which subspace the operator acts on.

It is crucial to observe that not every state in the combined system is a

product of two individual states, rather, it is in general a superposition of
such combined states. This leads to the phenomenon known as entangle-
ment, but we won’t have time to dwell on this further in these notes.

Important examples of combined systems occur for constituents describing

the same type of particle, e.g. H1 ≃ H2 ≃ H and Ĥ1 = Ĥ2 = Ĥ. Then we
find a new requirement that any state has to fulfil. Consider the operator
S, called swap operator, which we define by its action on the basis elements
as
S :H⊗H→H⊗H
(48)
|i⟩ ⊗ |j⟩ 7→ |j⟩ ⊗ |i⟩ .
We can now demand a state in H ⊗ H be invariant under the action of S,

S|ψ⟩ = α|ψ⟩ (49)

where α is some phase (|α| = 1). Recall that two states |ψ1 ⟩ and |ψ2 ⟩ are
equivalent, i.e. describe the same physical state, if they differ only by a

15
phase, |ψ2 ⟩ = α|ψ1 ⟩. (This ambiguity may seem innocuous, but has very far
reaching and delicate consequences.) Hence we included an arbitrary phase,
α, in eq. (49) so as to not lose generality.

Experiments find that in nature only two possibilities are realised, α = 1

and α = −1. Note that in both cases S 2 |ψ⟩ = |ψ⟩.3 Furthermore, particles
with integer spin (bosons) are found to have α = 1 while particles with
half-integer spin (fermions) have α = −1.

So if we are combining identical bosons, the product state has to satisfy

S|ψ⟩ = |ψ⟩ (50)

while for fermions it has to satisfy

S|ψ⟩ = −|ψ⟩ . (51)

In other words, the combined state space is just the (anti-)symmetric part of
the tensor product. They are spanned by the (anti-)symmetric combinations
of basis vectors
|i, j⟩ = |i⟩|j⟩ + |j⟩|i⟩ for i ≤ j (52)
and
|i, j⟩ = |i⟩|j⟩ − |j⟩|i⟩ for i < j (53)
respectively, e.g. a two-fermion system can never be in a state |ψ⟩ ⊗ |ψ⟩.
This is the famous Pauli exclusion principle.

Q 3.1 What is the dimension of the state space for a spin- 12 particle? What
is the dimension of the state space of N distinct spin- 21 particles? What
if they are identical fermionic spin- 12 particles? What if they are identical
bosonic spin- 12 particles?

This phase α = ±1 has important consequences for the thermodynamic

properties of the full system and is hence not just a technicality.

Harmonic oscillator
Ladder operators are one of the many avatars of complex methods in physics.
Let’s start by writing the classical Hamiltonian for the harmonic oscillator
not using two real coordinates x, p, but instead one complex coordinate
r
mω i
z=x + p√ , z̄ = z ∗ , (54)
2 2mω
3
Using QFT one can in fact prove that this has to be the case, if their interactions are
to make sense. Particles for which α is not ±1, called anyons, are possible in 2d and can
be realised in condensed matter lab experiments.

16
in terms of which the Hamiltonian becomes

H(x, p) → H(z, z̄) = ωz z̄ = ω|z|2 . (55)

To quantize we promote these variables to operators (z → a, z̄ → a† ) and

write the Hamiltonian as the symmetric product
ω ω
Ĥ = {a, a† } = (aa† + a† a) . (56)
2 2

Q 3.2 Show that ⟨ψ|Ĥ|ψ⟩ ≥ 0 for any state |ψ⟩ by using the property ∥·∥2 ≥
0 of Hilbert spaces (and ω > 0).

Q 3.3 Write the operators a, a† in position space and compute their anti-
commutator {a, a† }. Do you recognise the result? Also compute their com-
mutator [a, a† ] without using position representation, instead using the canon-
ical commutation relations eq. (18).

Q 3.4 Show ⟨ψ|Ĥ|ψ⟩ > 0 for any state |ψ⟩ by refining your answer to Q 3.2.
Find a lower bound for ⟨Ĥ⟩ψ and give a condition on any |ψ⟩ saturating this
bound.

Q 3.5 Show that a, a† act on eigenstates of Ĥ by shifting their eigenvalue by

±1. Since the spectrum of Ĥ is bounded from below, argue that there must
be a state which is annihilated by a.

In a general system, the state which is annihilated by (all) lowering operators

is defined to be the vacuum and given the symbol |0⟩ or |Ω⟩. Note that this
is not the zero element of the Hilbert space, i.e. not ⃗0. It is some non-
zero vector, and often a great deal of effort goes just into figuring out what
exactly the vacuum of a system is.

Q 3.6 a) Use the commutation relation [a, a† ] to calculate the norm of the
states |n⟩ = cn (a† )n |0⟩. Calculate the overlap ⟨n|m⟩. Finally choose cn such
that ⟨n|n⟩ = 1 and write down ⟨n|m⟩ in terms of the normalized states.
b) Show that you can realize the commutation relation for a, a† via a → ∂a† ,
by showing an f (a† )|0⟩ = f (n) (a† )|0⟩ for any polynomial f .
c) Repeat a) by realising the commutation relation as in b).

Q* 3.7 Consider the coherent state |λ⟩ = exp(λa† )|0⟩ for some λ ∈ C.
Show that it is an eigenstate of the annihilation operator a. Calculate the
overlap ⟨λ|κ⟩, e.g. by realising the a, a† commutation relation via a → ∂a† .
Calculate the average energy ⟨Ĥ⟩λ = ⟨λ|H|λ⟩/⟨λ|λ⟩.

Here are some questions on perturbations of the harmonic oscillator.

You may find it useful to recall Fermi’s golden rule, eq. (37).

17
Q 3.8 Calculate to leading order the transition probability between different
levels of the harmonic oscillator, Ĥ0 = ω2 {a, a† }, under the anharmonic
perturbation ĤI = ϵx̂3 . (Hint: express x̂3 in terms of a, a† .)
Q* 3.9 Show that the correction to the transmission probability at next-to-
leading order is quartic in the perturbation. (You are not required to find
this correction.)

Spin
Spin is an abstract dynamical property of particles and can be thought of
by analogy with the familiar angular momentum. Though tempting, it is
not quite correct to think of a particle spinning around itself like a top. The
analogy is however very deep, which is captured in the statement that the
spin and angular momentum operators obey the same algebra. Recall that
the ith component of the angular momentum operator is defined as
ˆ × p⃗ˆ)i = ϵijk x̂j p̂k ,
L̂i = (⃗x (57)

from which we can verify the commutation relations (angular momentum

algebra)
[L̂i , L̂j ] = iϵijk L̂k . (58)
Note that the definition of L̂i in terms of position and momentum gives
it a natural interpretation as the quantum operator associated to classical
angular momentum L ⃗ = ⃗x × p⃗.

Q 3.10 When promoting the calssical angular momentum to an operator,

is there a normal ordering ambiguity in the definition? Use the canonical
commutation relation to verify eq. (58).
As we mentioned before, an observable (Hermitian operator) generates
transformations on states. In order to get a feel for what this operator does:
Q* 3.11 Calculate the action of exp(iα· L̂) on a scalar function f (x). (Hint
1: go to a basis for R3 in which α⃗ looks simple. Hint 2: you should find
that f (x) → f (R(α)x), where R(α) is the rotation matrix around the vector
⃗ /∥α∥ by an angle ∥α∥.)
α
We now create a new set of three operators, Ŝi , called spin operators and
declare they obey the same commutation relations as the angular momentum
operators eq. (58), i.e.
[Ŝi , Ŝj ] = iϵijk Ŝk (59)
ˆ
⃗ i = Ŝi . So far, this is an abstract
for the components of the spin vector (S)
algebra. If we want to describe an actual physical system and calculate
its properties, we’ll have to choose an appropriate representation for this
algebra.

18
Q 3.12 Verify that Ŝi = 21 σi , where the three matrices
     
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 = , (60)
1 0 i 0 0 −1

are the Pauli matrices, satisfy eq. (59).

Q 3.13 Calculate the value of the spin s for this representation using the
relation
⃗ˆ · S
Ŝ 2 = S ⃗ˆ = s(s + 1)1 . (61)
What is the Hilbert space that this operators act on? Write down a basis
and a generic element of it.

Having a spin- 21 representation and the corresponding space of physical

states at hand, we go on to define the Hamiltonian operator. Let’s couple
the spin to an constant external magnetic field B⃗ by taking

⃗ˆ · B
X
Ĥ = S ⃗ = Ŝi Bi , (62)
i

where B⃗ is some constant vector. For convenience, let’s rotate our measuring
apparatus such that its z-axis is parallel to the magnetic field, in other words,
⃗ = bêz .
take B

Q 3.14 Find the normalized eigenvectors and eigenvalues of the Hamilto-

nian eq. (62) and denote them by | ↑ ⟩ and | ↓ ⟩ with energy eigenvalues E±
respectively. Given a state |ψ⟩ = c1 | ↑ ⟩ + c2 | ↓ ⟩ at time zero, what is |Ψ(t)⟩?
i.e. calculate the time evolution operator U (t) = exp(−itĤ) for eq. (62) and
apply it to |ψ⟩.

Recall that physical states in a quantum theory are described by rays in

the Hilbert space. A ray is uniquely determined by a normalized element
of the Hilbert space modulo phase. In a spin-1/2 system any normalized
element of the Hilbert space can be expressed in the form |ψ⟩ = c1 | ↑ ⟩+c2 | ↓ ⟩
for c1 = sin(ϑ/2) and c2 = cos(ϑ/2) eiφ , up to an overall phase.

Q* 3.15 By determining the redundancy in the parametrization of physical

states by ϑ and φ determine the physical state space of a spin-1/2 system.
Express |Ψ(t)⟩ from question Q 3.14 in this form and determine ϑ(t) and
φ(t).

⃗
Q 3.16 Calculate the average spin S(t) ⃗ˆ Ψ(t) in the state Ψ(t). (Hint:
= ⟨S⟩
look at Sz and S⃗⊥ = (Sx , Sy ) separately, and express S ⃗⊥ (t) = R(t)S ⃗⊥ (0)
using a time dependent rotation matrix R.)

19
 Observe the rotation of the spin expectation value around the axis of the
magnetic field, called precession. Given the nice behaviour of the expecta-
tion value, we can go further and ask how the operator itself evolves (in the
Heisenberg picture).

Q* 3.17 Calculate the time evolution of the operator S ⃗ˆ in the Heisenberg

picture for the Hamiltonian in eq. (62). (Hint 1: you can use the spin algebra
to turn the Heisenberg equation into a linear coupled ODE. Hint 2: separate
⃗ˆ into Ŝz and S
S ⃗ˆ⊥ as above.)

Q* 3.18 Repeat Q 3.14 without making the simplification B ⃗ = bêz . If you

choose the right basis for the Hilbert space, this exercise requires nearly no
additional computation.
Q* 3.19 Compute the matrix exponential exp(i⃗ α · ⃗σ ) for any vector α⃗ . In
addition to the algebra eq. (59), you’ll need to use {σi , σj } = 2δij . A handy
way to combine the two is in the equation σi σj = δij + iϵijk σk .

4 The Feynman Propagator

Define the time ordered product of two arbitrary operators in the Heisenberg
picture as
T (O1 (t)O2 (t′ )) = θ(t − t′ )O1 (t)O2 (t′ ) + θ(t′ − t)O2 (t′ )O1 (t) (63)
with the Heaviside step function θ(x) = (sign(x) + 1)/2.
Q* 4.1 Compute the Feynman propagator for the harmonic oscillator, de-
fined as
GF (t, t′ ) = ⟨0|T (x̂(t)x̂(t′ ))|0⟩ . (64)
Notice how it is a function of t − t′ , which reflects the time translation
symmetry of the system. The Feynman propagator is important because
it is a fundamental solution to the equation of motion for the harmonic
oscillator.
Q* 4.2 Show that GF satisfies the equation of motion with delta function
source term

d2
 
i
2
+ ω GF (t, t′ ) = − δ(t − t′ ) .
2
(65)
dt m
The Feynman propagator still knows about the canonical commutation re-
lation of x̂, p̂.
˙
Q* 4.3 Using p̂(t) = mx̂(t) and taking an appropriate limit of the Feynman
propagator, show that
⟨0|[x̂, p̂]|0⟩ = i . (66)

20