QFT KleinGordon

Notes On Relativistic Quantum Field Theory
A Course Given By Dr. Tobias Osborne

Part I: The Klein-Gordon Field
Transcribed by Dr. Alexander V. St. John
February 23, 2019
1
Lecture 1: Motivations for (Relativistic) Quantum
Field Theory
We assume that there is a ”Theory of Everything”, called ToE herein, which is
suspected to include a contained description of small and large scale forces in the
universe (e.g., strong nuclear, weak nuclear, electromagnetic, and gravitational
forces). Our classical observations fall into a low energy, large scale, decoherent
approximation of this ToE. The purpose of this collection of notes is to study
(effective) relativistic quantum field theory which is the low energy, large scale
approximation of the ToE, and we are not sure how many ”steps” there are
in between the two, but the relativistic quantum field theory is the closest we
currently come to a ToE. If we lower the energy and lengthen the scale of
our study, we land in a nonrelativistic quantum field theory. A quantum field
theory is then subject to decoherence, as we exist on the classical scale, and
we begin the development of a relativistic quantum field theory with a classical
field theory. In this sense, our purpose is to undo all of the approximations
that nature forces us to take when we run experiments, take measurements,
and devleop theoretical frameworks to describe observed phenomena.
Figure 1: Schematic of the study of field theories in physics.
2
Mathematical Machinery of Relativistic QFT
For a quantum field theory to be relativistic it must be symmetric under the
Poincaré group transformations.
Let the four-vector (x0 , x1 , x2 , x3 ) be the spacetime coordinates in an inertial
reference frame. Then in any other reference frame that an observer chooses,
the following condition is satisfied.
(Note: We adopt the Einstein summation notation, and repeated indices are
summed over, such that µ, ν, ρ, and σ below are summed from 0 to 3.)
ηµν dx0µ dx0ν = ηρσ dxρ dxσ (1)

Where we are using the Minkowski metric
 
1 0 0 0
 0 −1 0 0 
ηµν = gµν =  . (2)
 0 0 −1 0 
0 0 0 −1
Any transformation satisfying the first equation, the relationship between the
coordinates of two reference frames, must be a linear transformation of the form,
denoted by the pair (Λ, a),
x0µ = Λµν xν + aµ (3)
Where Λµν is the 4x4 Lorentz transformation matrix that represents rotations,
and aµ is a constant four-vector that represents spatial translations.
Also note that the Lorentz transformation must satisfy the following condition,
shown in index and matrix notation.
ηµν Λµρ Λνσ = ηρσ (4)

T
Λ ηΛ = η (5)
These transformations (Λ, a) are the elements of the Poincaré group P4 . Let’s
check the group conditions, using equation (??) for calculating the product of
two Poincaré transformations,
product (Λ, a) ◦ (Λ̄, ā) = (Λ̄Λ, Λ̄a + ā)

identity (1, 0) ◦ (Λ, a) = (Λ, a)
inverse (Λ−1 , a−1 ) ◦ (Λ, a) = (1, 0) (6)
¯
associativity (Λ, a) ◦ (Λ̄, ā) ◦ (Λ̄, ā ¯ , ā
¯) = (Λ, a) ◦ (Λ̄, ā) ◦ (Λ̄ ¯)
In index notation, the total effect of the product of two Poincaré transfomations
is written as
x0µ = Λµν xν + aµ (7)

x00µ = Λ̄µρ x0ρ + āµ (8)
3
Quantum mechanical symmetries are representated by unitary or anti-unitary
linear operators as elements of a separable Hilbert space H such that for each
Poincaré transformation (Λ, a) ∈ P4 , there exists a unitary transformation
U (Λ, a) : H → H (9)
The identity unitary transformation, and the product of any two unitary transformations
of elements of the Poincaré group are physically indistinguishable from each
other up to a phase factor eiφ , where φ ∈ R
U (Λ̄, ā)U (Λ, a) = eiφ((Λ,a),(Λ̄,ā)) U (Λ̄Λ, Λ̄a + ā) (10)

U (1, 0) = e iφ
1. (11)
The family of transformations U (Λ, a) that satisfy these equations is called

the projective unitary representations of P4 , and is precisely what we need to
establish a relativistic quantum field theory that includes translations, rotations,
and Lorentz boosts as its transformations.
Consider the time translation subgroup of the Poincaré group
{(Λ = 1, a = (t, 0, 0, 0)) : t ∈ R} ⊂ P4 (12)
And let U be a unitary representation of P4 . Then V (t) = U (1, (t, 0, 0, 0)) is a

one-parameter family of unitary transformations, which is a group homomorphism,
such that V (s)V (t) = V (s + t), and is called a propagator, and is a solution to
the Schrödinger equation, assuming the Hamiltonian Ĥ is self-adjoint and is
stable, such that there at most a finite number of, preferrably zero, negative
eigenvalues
dV (t)
= iĤV (t). (13)
dt
So, having a unitary representation is equivalent to solving the time-dependent
Schrödinger equation, and we require that U (Λ, a) is positive energy, such
that the spectrum of eigenvalues of Ĥ is positive spec(Ĥ) ⊂ R+ , and stable.
Otherwise, the system may be unstable and plunge to more and more negative
eigenvalues.
All single-particle unitary representations of P4 have been classified by Wigner,

and are labelled by their mass m (invariant under Poincaré and Lorentz transformations)
and their helicity/spin s. The bad news is that the universe is comprised of many
particles, creating tension between locality and interactions, rendering thorough
study of single-particle representations pointless. Even for a single particle, as
we increase locality, self-interaction increases, and new particles are created via
tunnelling.
4
Why is Obtaining a Relativistic Quantum Field Theory
Hard?
The reason that constructing a relativistic QFT is considered so difficult is
because there are no nontrivial, only trivial, finite-dimensional unitary representations
of the Poincaré group P4 . The only nontrivial representations of P4 are infinite-
dimensional, and are not easy to work with when constructing a unitary representation.
Consider this in contrast to the, more easily constructed, special orthogonal
group of three-dimensional rotations SO(3) with unitary representations
ˆ ˆ ˆ
U = ei(σx Jx +σy Jy +σz Jz ) . (14)
Where Jˆα are the angular momentum operators. The reason SO(3) is easier to
work with and construct a unitary representation is because SO(3) a compact
group, while P4 is a non-compact group.
5
Lecture 2: Introduction to Classical Field Theory
To build a relativistic QFT, we start with an effective model from a classical
field theory, and make an ”educated guess” to quantize the classical field theory.
The desired relativistic QFT has nothing to do a priori with the classical field
theory. After quantization, the ”educated guess”, we take the limits of low
energy, large scale, and decoherence, and check that we get back the classical
field theory we started with, demonstrating whether the chosen quantization is
correct or incorrect... enough, to some level of approximation.
Fields
A field is a quantity (e.g., density, spin, charge) tht is defined at every point on a
manifold M. Note that a rigorous definition of a field requires the introduction
of vector bundles, of which we will not go so far.
We work on the Minkowski spacetime manifold M = M1,3 = R1 × R3 , with
the field often taken to be two-times differentiable such that φ ∈ C 2 (M, ρ) and
defined as a function from the manifold to some target space ρ
φ:M→ρ (15)
Some target spaces ρ, their associated field type, and example applications and
models include
• ρ = R; scalar field; charge density, magnetization density, Higgs boson

• ρ = Rn ; vector field; electromagnetic field (actually a gauge field), pions
Figure 2: Sketch of the vector field ρ = Rn over the Minkowski space M1,1 .
• ρ = S 2 ; vector field on the surface of a sphere; σ-model, quantum

magnets
• ρ = S 1 × S 1 ; vector field on a torus; Chern-Simons theory, Lie groups
Note that when our target space is the N -dimensional real vector field RN , the
vector field is described as a list N scalar fields {φa (x)}N
a=1 , where x is the
coordinate four-vector.
6
Dynamics of Classical Fields
We restrict this to discussion to classical dynamics generated by Lagrangians,
obtained via the variational principle applied to th action functional, where
the system of scalar fields {φa (x)}N a=1 , and a labels the particle type (e.g.,
charge). The action functional S contains a function of the Langrangian density
L = L (φa , ∂µ φa ). The Lagrangian density is actually a function of higher order
derivatives of the fields φa , but we make the assumption and approximation of
first order derivatives, based on observation
Z
S(Ω) = d4 x L (φa , ∂µ φa ). (16)
Ω
4
Where d x = dx0 dx1 dx2 dx3 and Ω ⊂ M1,3 , as a measurable set, is a region in
(3 + 1)-dimensional spacetime. Typically, we consider the spacetime region as
the entire Minkowski space Ω = M1,3 .
To extract the equations of motion, we suppose that the action S is stationary

under infinitesimal variations of the component scalar fields φa (x) → φa (x) +
δφa (x), which vanish on the spacetime region boundary, such that δφa (x) = 0
on ∂Ω.
Varying the action functional, we obtain N Euler-Lagrange equations of motion

Z
∂L ∂L
δS(Ω) = d4 x δφa + δ(∂µ φa ) (17)
Ω ∂φa ∂(∂µ φa )
Z Z
∂L ∂ ∂L ∂ ∂L
0= d4 x δφa − δφa + d4
x δφa
Ω ∂φa ∂xµ ∂(∂µ φa ) Ω ∂xµ ∂(∂µ φa )
(18)
Z
∂L ∂ ∂L
0= d4 x δφa − µ
δφa + ∂µ Mµ . (19)
Ω ∂φa ∂x ∂(∂ µ φ a )
The last term is a surface term which vanishes on the boundary, since we first
demanded that δφa = 0 vanishes on the boundary, such that ∂µ Mµ = 0 on ∂Ω.
Since S(Ω) is stationary with respect to all variations of the fields δφa and
admissable spacetime regions Ω, the integrand of the remaining term must
also vanish ∀ a = 1, 2, ... , N , and we obtain the N Euler-Lagrange equations
of motion
∂L ∂ ∂L
− + = 0. (20)
∂φa ∂xµ ∂(∂µ φa )
So, with equations of motion in hand, gotten by whatever means, they can be
encoded in the Langrangian density, and then recovered via the Euler-Lagrange
equations of motion. This allows us to discover equations of motion with
certain properties, such as being symmetric under the Poincaré transformations,
by designing Lagrangian densities, which are scalars under these symmetry
7
transformations, apply Euler-Lagrange ”recipe”, and get the equations of motion,
which are guaranteed to be symmetric under the chosen transformations. This
benefit of the action principle makes it easy to design equations of motion with
certain symmetries.
Example: Klein-Gordon field

Consider the Langrangian density
1 2 1 1
L = φ̇ − (∇φ)2 − m2 φ2 (21)
2 2 2
1 2 1 2 2
= (∂µ φ) − m φ (22)
2 2
Where φ̇ = ∂0 φ is the time derivative of the field, and (∇φ)j = ∂j φ for j = 1, 2, 3
are the x, y, z spatial components of the field derivatives.
Apply the Euler-Lagrange equations of motion to obtain the dynamics of the

Klein-Gordon field

∂L ∂ ∂L
− + =0 (23)
∂φa ∂xµ ∂(∂µ φa )
∂
(∂ µ φ) − (−m2 φ) = 0 (24)
∂xµ
∂µ ∂ µ φ + m2 φ = 0 (25)
φ + m2 φ = 0 (26)
Hamiltonian formalism
A better way to guess/build a quantum theory, with the correct calssical limit
determined by the Langrangian density, is the Hamiltonian formalism, where
we calculate conjugate variables and impose canonical (algebraic) commutation
relations.
Suppose that φa (x) is a component field’s canonical position. Then define the
conjugate momentum density for each field to be
∂L
πa (x) = . (27)
∂ φ˙a
In order to computationally study a field, we discretize the space into a regular
lattice with spacing between each lattice site, since representing a continuous
object on a computer would require an infinite amount of data. Discretization
yields generalized coordinates, defined by the field itself, of the form qja (t) =
φa (t, xj ), j ∈ Z. For example, consider the 1 + 1-dimensional Minkowski space
M1,1
8
Figure 3: Schematic of field discretization.
To discretize the action functional and apply the variational principle, replace
the partial derivatives ∂µ φa (t, x), since it is a continuous operation, by applying
Taylor’s theorem to approximate the spatial component of the field as a finite
difference, where should be made as small as possible such that the error
approximation is minimized
a
φa (t, xj + ) − φa (t, xj ) qj+1 − qja
∂x φa (t, xj ) ∼
= = . (28)

Leave the temporal component to be continuous
∂t φa (t, xj ) ∼
= q̇ja . (29)
To obtain the Lagrangian, integrate the density over space, such that only the
time dependence is left
Z
d3 x L (φa , ∂µ φa ) = L(qja (t), q̇ja (t)) = L(t). (30)
The discrete approximation of the Lagrangian is written as
L(t) ∼
X
= δxj L (φa (t, xj ), ∂µ φa (t, xj )). (31)
j
And the discrete conjugate momenta

∂L X ∂L X
paj (t) = a = δxj a = δxj π a (t, xj ). (32)
∂ q̇j j
∂ q̇ j j
Simple Example
As a simple example, consider the Lagrangian density
1
L = (∂µ φ∂ µ φ) (33)
2
1 1
= (∂t φ)2 − (∂x φ)2 . (34)
2 2
9
(RECHECK) Integrate over space d3 x to obtain the Lagrangian, and discretize
using the rules defined above
∞ 2 !
1 X dqj (qj+1 − qj )
L(t) = − . (35)
2 j=−∞ dt
dq
Where δxj = and pj (t) = dtj = π(t, xj )δxj .
As → 0, qj (t) → φ(t, x) and pj (t) → φ(t,˙ x)
Hamiltonian Density
For the discrete approximation, the Hamiltonian H, obtained by integrating the
Hamiltonian density H of space d3 x, is written as
X X
H= paj q̇ja − L = δxj (πa (t, xj )φ̇a (t, xj ) − Lj ) (36)
j j
In the limit as → 0, the Hamiltonian density is
H (t, x) = πa (t, x)φ̇a (t, x) − L (φa (t, x), ∂µ φa (t, x)) (37)
The Hamiltonian density for the Klein-Gordon field is, dropping the field index
and spacetime dependencies from the expression,
1 2 1 1
H = π + (∇φ)2 + m2 φ2 (38)
2 2 2
10
Lecture 3: Symmetries in Classical Field Theory
Suppose that L (φa , ∂µ φa ) is the Lagrangian density for some set of fields
{φa (x)}N
a=1 . Recall that the Lagrangian density is a compact, encrypted of
writing the equations of motion of a system, and the Euler-Lagrange equations
and the principle of least action are used to unpack/decrypt the equations of
motion.
Now consider an infinitesimal continuous transformation of the fields
φ0a (x) = φa (x) + Xa (φa ) (39)
This produces an infinitesimal symmetry in the equations of motion when they

are left invariant under the principle of least action
L → L (φ0a , ∂µ φ0a ) = L (φa , ∂µ φa ) + ∂µ F µ (40)

Z Z
& S[φa ] = d4 xL = d4 x(L + ∂µ F µ ) (41)
Noether’s Theorem
Every continuous symmetry of a Lagrangian implies the existence of a conserved
current j µ (x).
Proof: Let Xa [φa ] = δφa be an arbitrary, infinitesimal change in each field,

such that the infinitesimal change in the Lagrangian density is
δL (φa , ∂µ φa ) = L (φa + δφa , ∂µ (φa + δφa )) − L (φa , δµ φa ) (42)

∼ ∂L ∂L
= δφa + δ(∂µ φa ) (43)
∂φa ∂(∂µ φa )

∂L ∂L ∂L
= − ∂µ δφa + ∂µ δφa (44)
∂φa ∂(∂µ φa ) ∂(∂µ φa )

∂L
δL (φa , ∂µ φa ) = ∂µ δφa (45)
∂(∂µ φa )
Where in line (14), we have kept only first order terms O(δφa ), and have used
the Taylor expansion and added zero to get to line (15). To obtain line (16),
note that the first term in line (15) is equal to zero, since φa (x) obey the Euler-
Lagrange equations.
As defined above, for an infinitesimal transformation, call δφa = Xa [φa ] and for
an infinitesimal symmetry, call δL = ∂µ F µ .
11

∂L
∂µ F µ = ∂µ Xa [φa ] (46)
∂(∂µ φa )

∂L
0 = ∂µ Xa [φa ] − F µ (47)
∂(∂µ φa )
Call the conserved quantity the conserved current

µ ∂L µ
j (x) = Xa [φa ] − F . (48)
∂(∂µ φa )
Now, a conserved current implies the existence of a conserved charge. For any
measurable region in our Minkowski space V ⊂ M, define the integral of the
time-like component of the current as
Z
QV = d3 x j 0 (x) (49)
V
Take V = R3 , and assume that the current vanishes at infinity, such that j → 0
on the boundary ∂V , and take the time derivative
Z
dQR3
= d3 x ∂0 j 0 (x) (50)
dt R3
Z
=− d3 x ∂ k j k (51)
R3
Z
=− jds (52)
∂R3
dQR3
=0 (53)
dt
Example
Consider an active transformation of spacetime coordinates xµ → xµ − µ .
Then each field transforms as
φ0a (xµ ) = φa (xµ + µ ) (54)

µ ν µ
= φa (x ) + ∂ν φa (x ) (55)
And the Lagrangian density transforms as, yielding 4 × 4 = 16 equations from

summing over µ and ν
L (x0µ ) = L (xµ + µ ) (56)

= L (xµ ) + ν ∂ν L (xµ ) (57)
Where the infinitesimal field transformation is Xa [φa ] = ν ∂ν φa (xµ ), and the

infinitesimal symmetry of the Lagrangian density is ∂µ F µ = µ ∂µ L (xµ )
12
Consider the infinitesimal element with basis vector entries [ν̂]µν = δ µν
µ = ν̂ µ = {(1, 0, 0, 0), (0, 1, 0, 0), (0, 0, 1, 0), (0, 0, 0, 1)}. (58)
Apply Noether’s theorem to each of the 4 symmetry terms ν̂ µ , yielding 16
total terms that we assign as the elements T µν of the energy-momentum or
stress-energy tensor
∂L
T µν = j µν = ∂ν φa − δ µν L . (59)
∂(∂µ φa )
Note that each of the ν th columns of the energy-momentum tensor correspond
to one of the four conserved currents and translation in each of the ν th directions
∂µ T µν = 0, ∀ ν. (60)
In a closed system, the corresponding conserved charges, from the columns
(conserved currents) of the energy-momentum tensor, are the total energy and
the momentum in each of the three spatial directions
Z
E = d3 x T 00 (61)
Z
pj = d3 x T 0j . (62)
Example of the Example: Klein-Gordon Field

Consider the Klein-Gordon Lagrangian density L = 12 ∂µ φ∂ µ φ − 21 m2 φ2 . The
energy-momentum tensor has elements of the form
T µν = ∂ µ φ∂ν φ − δ µν L (63)
0
T µν = η νν T µν 0 = ∂ µ φ∂ ν φ − η µν L . (64)
The corresponding conserved charges are
Z Z
1 1
E = d3 x H (x) = d3 x (∂ 0 φ∂ 0 φ − ∂ 0 φ∂ 0 φ + m2 φ2 ) (65)
2 2
Z
pj = d3 x φ̇∂ j φ. (66)
How To Apply Noether’s Theorem

1. Identify the continuous symmetry.
2. Calculate the change in the Lagrangian density.
3. Calculate the change in each of the fields.
4. Work out the conserved currents and charges.
13
Infinitesimal Lorentz Transformations
Consider the transformation xµ → Λµν xν , where Λµν = δ µν + ω µν , and ω µν
is infinitesimal. Next, recall the following property of the group of Lorentz
transformations, restricting the possible values for ω
η = ΛT ηΛ (67)
µν
η = (δ µσ + ω µσ )(δ ντ + ω ντ )η στ (68)
µν νµ
0 =O(ω) ω +ω (69)
This is a linear equation in ω, since we have kept only up to first order terms
in ω, and tells us that ω is an antisymmetric, infinitesimal generator of Lorentz
transformations with six independent variables which define six continuous symmetries
and six conserved currents and charges.
 
0 −α −β −γ
 α 0 −δ − 
ω=  (70)
 β δ 0 −κ 
γ κ 0
The action of this infinitesimal Lorentz transformation on the fields is
φa (x) → φ0a (x) = φa (Λ−1 x) (71)

= φa ((δ − ω)x) (72)
= φa (x − ω µν xν )
µ
(73)
=O(ω) φa (x) − ω µν xν ∂µ φa (x) (74)
Showing that the symmetry is defined by the infinitesimals
δφa = −ω µν xν ∂µ φa (75)
& δL = −ω µν xν ∂µ L = −∂µ (ω µν xν L ) (76)
∂L
& j µω = ω ρ xν ∂ρ φa + ω µν xν L . (77)
∂(∂µ φa ) ν
(CHECK how j to J ) Applying Noether’s theorem tells us that the six independent
conserved currents ∂µ (J µ )ρσ = 0, and conserved charges, are of the form
(J µ )ρσ = xρ T µσ − xσ T µρ (78)
Z
Qjk = d3 x (xj T 0k − xk T 0j ) (79)
Z
Q = d3 x (x0 T 0j − xj T 00 )
0j
(80)
Call Qjk the generators of rotations, and Q0j the generators of boosts of the
Lorentz transformations.
14
Generators
Let f and g be maps from phase space to the real numbers
f, g : RN × RN → R. (81)
Define the Poisson bracket with the pairs of canonical coordinates (qj , pj )
N
X ∂f ∂g ∂f ∂g
{f, g} = − (82)
j=1
∂qj ∂pj ∂pj ∂qj
df
{f, H} = (83)
dt
The field theory version of the Posson bracket is defined with the canonical
coordinate pairs (φ(x), π(x))
Z
3 δF δG δF δG
{F, G} = d x − (84)
δφ(x) δπ(x) δπ(x) δφ(x)
∂f
{f, Qρσ } = ρσ (85)
∂s
Where the Poisson bracket of f and the conserved charges Q generate the
corresponding symmetry transformations. Conserved cahrges also obey the Lie
algebra obeyed by the Poincaré group.
Standard Dogma of Quantization

Basically, put hats on things
• Function f on phase space → linear operator fˆ (observable) on Hilbert

space
• Poisson bracket {f, g} = h → commutator [fˆ, ĝ] = iĥ
• Conserved charge Qρσ → conserved charge operator Q̂ρσ
Where the conserved charge operators generate the Lorentz transformations on

Hilbert space
dÛ
= i[Û , Q̂ρσ ]ωρσ (86)
ds
15
Lecture 4: Field Quantization
More than one quantum field theory can have the same classical field theory
as an effective model, making field quantization not a well-posed problem.
Developing a quantum field theory is therefore built on educated guesses.
Canonical quantization of particles

The standard approach of canonical quantization is to begin with a classical
theory and suppose n classical degrees of freedom, which are used to measure
the canonical coordinate pairs, position qj and momentum pj , for each degree
of freedom, such that the Poisson bracket is defined by {qj , pk } = δjk . The total
energy of the of the system is measured by the classical Hamiltonian, defined
by
n n
X p2j m X
H= + qj [Q]jk qk (87)
j=1
2m 2
j,k=1
Where Q is an n × n symmetric, positive matrix.
For example, consider the quantum harmonic oscillator, and take the naive
approach by basically putting hats on everything. This ends up working for
field quantization, and yields a unitary representation of the Poincaré group.
• Canonical coordinates: (qj , pj )

→ Canonical coordinate operators : (q̂j , p̂j )
• Poisson bracket: {qj , pk } = δjk

→ Commutator: [q̂j , p̂k ] = iδjk
• Hamiltonian: H, as defined above

Pn p̂2j m
Pn
→ Hamiltonian operator : Ĥ = j=1 2m + 2 j,k=1 q̂j [Q]jk q̂k
To diagonalize the Hamiltonian operator, first note that since Q is a symmetric,

positive n×n matrix, there exists an orthogonal matrix O (s.t., OT O = I), such
that OQOT = D, where D is a diagonal matrix, where we call the diagonal
elements {ωi2 }ni=1
Now, transform the canonical coordinates, using the orthogonal matrix, such
16
that the correct commutation relation is still obeyed
n
X
q̂j = [O]jk q̂k0 (88)
k=1
Xn
p̂j = [O]jk p̂0k (89)
k=1
iδjk = [q̂j0 , p̂0k ] (90)
Using the facts that OT = O−1 and OQOT = D, the Hamiltonian becomes
diagonalized
n 2 n
X p̂0 j m X
Ĥ = + qˆ0 l [OT ]jl [Q]jk [OT ]km qˆ0 m (91)
j=1
2m 2
j,k,l,m=1
n 2 n
X p̂0 j 1 X 2 ˆ0 2
Ĥ = + ωk q k (92)
j=1
2m 2
k=1
n
1X 1
Ĥ = ωk (â†k âk + ) (93)
2 2
k=1
Where the annihilation and creation ladder operators that diagonalize the quantum
harmonic oscillator Hamiltonian are defined as
r
mωk 0 i 0
âk = (q̂k + p̂ ) (94)
2 mωk k
r
† mωk 0 i 0
âk = (q̂k − p̂ ) (95)
2 mωk k
Canonical Quantization of Fields

To quantize the Klein-Gordon field, we follow the same seemingly naive approach
of putting hats on everything. In this example fo quantizing a field, the continuous
variable x is used, in contrast to the discrete labels j in the previous example
of the quantum harmonic oscillator
• Canonical coordinates: (φ(x), π(x))

→ Canonical coordinate operators : (φ̂(x), π̂(x))
• Poisson bracket: {φ(x), π(y)} = δ (3) (x − y)

→ Commutator: [φ̂(x), π̂(y)] = iδ (3) (x − y)
Note that this is the equal time Poisson bracket, such that (x − y) is
the spatial three-vector.
17
Also note that this commutator is strange, as it is comprised of ”two
self-adjoint operators and something’s that not even a function”
1
+ 12 m2 φ2 (x)

d3 x π 2 (x) + (∇φ(x))2
R
• Hamiltonian: HKG = 2
→ Hamiltonian operator : ĤKG = 21 d3 x π̂ 2 (x) + (∇φ̂(x))2 + 21 m2 φ̂2 (x)
R
Essentially, replace discrete sums with continuous integrals, by switching to a

continuous label j → x and a continuous dynamical variable qj → qx = φ(x).
Solving the quantum Hamiltonian, by analogy of the canonical quantization of
particles, should be as simple as creating the analog of the Q matrix and its
diagonalization.
Replacing sums by integrals allows the full diagonalization of Q, and, therefore,

the full diagonalization of the Hamiltonian Ĥ = ĤKG , but this does not yet yield
a unitary representation of the Poincaré group or a valid relativistic quantum
field theory. Diagonalizing the Hamiltonian only quantizes a one-parameter
subgroup of the Poincaré group. The conserved currents, charges, and operators
obeying the correct Lie algebra are still needed for a relativistic quantum field
theory.
Diagonalization of the quantum field theory

The diagonalization of a field theory begins with emergence of the Fourier
transform. Replace sums with integrals, and, since the matrix elements are
described by two numbers, let’s define a continuous function in two variables
K(x, y)
X Z
0
q̂j = [O]jk q̂k → φ̂(x) = d3 y K(x, y)φ̂(y) (96)
k
Where K(x, y) is the kernel of the Fourier transform.
Using the Fourier transform is motivated by certain features of symmetric

matrices. Consider the circulant matrix, a type of Toeplitz matrix where each
successive column is a cyclic permutation of the previous column, initialized by
the first column vector, and has the form as an n × n matrix
 
c0 cn−1 . . . c2 c1
 c1 c0 cn−1 c2 
 .. .. 
 
..
 .
 c1 c 0 . . .
 (97)

cn−2 . .. . .. c

n−1

cn−1 cn−2 . . . c1 c0
18
These matrices are diagonalized via the discrete Fourier transform, which is an
n × n unitary matrix, though not orthogonal and may have complex entries
 
1 1 1 ...
1 µ µ2 . . . 
.
 
1  2 4 .. 
U= √  1 µ µ  (98)
n
. .

 .. ..

µjk 
nn
... µ
2πi
Where µ = e n is the nth roots of unity. The elements of the discrete Fourier
2πijk
transform, therefore, have the form √1n e n . Compare this to the continuous
1 ixy
Fourier transform kernel function K(x, y) = 2π e .
For transformations between position and momentum space, make the guess
that the Fourier transform that will diagonalize our quantized Klein-Gordon
Hamiltonian has the form
d3 p ip·x
Z
φ̂(x) = e φ̂p (p). (99)
(2π)3
Where φ̂p (p) is the momentum space wavefunction, and is not Hermitian, such
that φ̂p (p)† = φ̂p (−p). To check if the guess is correct, apply ĤKG to the
transform defined above, and observe whether it is diagonalized or not.
As in the discrete case of diagonalization, we construct ladder operators
d3 p
Z
1
âp eip·x + â†p e−ip·x

φ̂(x) = 3
p (100)
(2π) 2ωp
3
Z r
d p ωp
âp eip·x − â†p e−ip·x

π̂(x) = −i (101)
(2π)3 2
p
ωp = |p|2 + m2 (102)
Check the commutation relation

Z 3 3 0r
−i d pd p ωp0 † 0
[φ̂(x), π̂(x0 )] = 6
[â−p , âp0 ] − [âp , â†−p0 ] ei(p·x+p ·y) (103)
2 (2π) ωp

[âp , â†p0 ] = (2π)3 δ (3) (p − p0 ) · I (104)
= iδ (3) (x − y) (105)
Making this substitution, the quantum Klein-Gordon Hamiltonian is diagonalized
19
d3 pd3 p0 i(p+p0 )·x −1 √
Z Z
ĤKG = 3
d x e ( ωp ωp0 (âp − â†−p )(âp0 − â†−p0 ) (106)
(2π)6 4
−pp0 + m2
+ √ (âp + â†−p )(âp0 + â†−p0 )) (107)
4 ωp ωp0
d3 p
Z
1
= 3
ωp (â†p âp + [âp , â†p ]) (108)
(2π) 2
d3 p
Z
1
= 3
ωp (â†p âp + δpp · I) (109)
(2π) 2
Z 3
∼ d p
= ωp â†p âp (110)
(2π)3
Where the infinite absolute energy shift is tossed to get the last line, since we
only measure energy differences, and ĤKG is diagonalized!
20
Lecture 5: Scalar Quantum Field Theory
By analogy to the classical Klein-Gordon equation and Hamiltonian, a model
for the (equal time, t = 0) quantum Klein-Gordon Hamiltonian was constructed
and diagonalized, via (continuous) Fourier transform and ladder operators,
Z
1
ĤKG = d3 x π̂ 2 (x) + (∇φ̂(x))2 + m2 φ̂2 (x) (111)
2
d3 p
Z
ĤKG = ωp â†p âp (112)
(2π)3
Where
p the zeroth, time, component of the momentum 4-vector p0 = ωp =
|p|2 + m2 depends on the spatial 3-vector p and the constant m2 .
This yields a representation of a one parameter subgroup of the Poincaré group,

namely U ((t, 0, 0, 0)) = e−itĤKG , but a true relativistic quantum field theory
requires the full (projective) unitary representation of the Poincaré group, including
generators for all possible transformation: 10 Lorentz + 4 translation = 14 total
transformations in the Poincaré group.
To quantize, put hats on the conserved charges identified by Noether’s theorem:

Qα → Q̂α . First, consider the generators of spatial translations, namely momentum.
Recall that the classical conserved current T 0j gives these, which is quantized:
pj → p̂j .
Recall the classical energy-momentum tensor for the Klein-Gordon field
1
T µν |KG = ∂ µ φ∂ν φ − δ µν (∂ µ φ∂µ φ − m2 φ2 ) (113)
2
From this, quantize and calculate the conserved charge for temporal translations,
namely the Hamiltonian, and conserved charges for spatial translations, namely
linear momentum.
d3 p
Z Z
ĤKG = d3 x T̂ 00 = ωp â†p âp (114)
(2π)3
d3 p j †
Z Z Z Z
ˆ
p̂j = d3 x T̂ 0j = d3 x φ̇∂j φ̂ = d3 x π̂∂j φ̂p̂j = p âp âp (115)
(2π)3
Note that there are several choices for the ordering of π̂ and φ̂ in the expression
of p̂j matters, and here is written the one that works.
Now check that the four-vector obeys the commuation relations, using the
R d3 p j †
diagonalized momenta p̂j = (2π) 3 p âp âp
{Qα , Qβ }P B = fαβγ Qγ → [Q̂α , Q̂β ] = ifαβγ Qγ (116)

µ ν µ ν
{p , p }P B = 0 → [p̂ , p̂ ] = 0 (117)
21
This confirms a projective unitary representation of the translation subgroup
of the Poincaré group, and now construct the explicit Hilbert space as a Fock
space, since the operators are quantized and diagonalized via ladder operators.
To construct a Fock space, begin by defining the vacuum state, highest weight
vector in the language of representation theory, |Ωi such that the annihilation
operator will completely obliterate it: âp |Ωi = 0, ∀ p, where p = (ωp , p).
The Hilbert space would then be generated via all finite linear combinations of
vectors of the form |p1 p2 . . . pn i = â†p1 â†p2 . . . â†pn |Ωi, but there is a technical
issue of the n-dimensional momentum state vectors actually being improper
vectors that are not normalizable, such that the scalar product needed to
finish the defintion of the Hilbert space will always blow up to infinity, since
hp|qi = (2π)3 δ (3) (p − q). These states are also not preparable by experiment,
since the state vector |p1 p2 . . . pn i represents n delta functions in position-
momentum space.
To create a normalizable state that can be used to define the Hilbert space,
”smear out” the momentum states by defining a smooth (L2 ) function ψ, which
must be Lorentz invariant, though the invariance it is not obvious
d3 p d3 p
Z Z
|ψi = 3
ψ(p) |pi = ψ(p)â†p |Ωi . (118)
(2π) (2π)3
Now, introduce a method to normalize these improper vectors to a new set of
improper vectors that are manifestly, more obviously, Lorentz invariant, and
offer a nice parameterization to make many calculations easier.
Consider the projection operator onto a single particle state, and note that the
integrand and the integral (volume element) are both separately not invariant
d3 p
Z
Isingle = |pi hp| . (119)
(2π)3
Enter a reference frame where this state is invariant by multiplying by one
d3 p
Z
Isingle = X(p) |pi hp| . (120)
(2π)3 X(p)
Where X(p) is a mystery factor to make the integral and integrand invariant.
2ωp
Claim: X(p) = (2π)3 ,
where p here must be the momentum 4-vector, since we
p
are using the zeroth, or time, component p0 = ωp = |p|2 + m2 .
Proof:
R 3
invariant, but d4 p is, such that
R
First,
R 4 observe
R 4 0that d p0µis notµ Poincaré
d p = d p , where p = Λ ν pν + aµ is a Poincaré transformation, and
22
Λµν is the Jacobian of the transformation, and det(Λ) = ±1, ∀ Λ unitary
transformation.
Now notice that pµ pµ = const. = m2 (4-vector length invariant), whose solution

is the dispersion
p relation for a single relativistic particle, and has two branches
p0 = ωp = ± |p|2 + m2 , where |p| is the norm of the momentum (spatial)
3-vector.
Restrict to the positive upper branch, and consider the Poincaré invariant
quantity
Z Z 3
d p
d4 p δ(p20 − |p|2 − m2 )|p0 >0 = |p =ω . (121)
2p0 0 p
Therefore, to make the single particle state projection operator from above
Poncaré invariant, compare terms in the line above to the ”mystery factor”
2ω
expression, proving that X(p) = (2π)p3 .
Thus, the ”delta normalization” of 3-vectors is defined via
2ωp δ (3) (p − q). (122)

And the renormalized, Lorentz invariant momentum 4-vector is built as
2ωp â†p |Ωi .

p p
|pi = 2ωp |pi = (123)
And the Lorentz invariant four-length comes out to be
hp|qi = 2(2π)3 ωp δ (3) (p − q). (124)

Now, to express the operators in terms of the Fock vector space we build on
top of the vectors |pi, and determine the action of the generator of spacetime
translations, the 4-momentum operator, p̂µ on the Hilbert space of momentum
states (improper vectors) |p1 p2 . . . pn i.
This requires some commutation relations with the ladder operator âp in the
following lemma.
Lemma: [ĤKG , âp ] = −ωp âp and [p̂j , âp ] = pj âp .
Next follows the corollary, demonstrating that the operator p̂µ is diagonalized
in this Hilbert space basis, such that the 4-momentum operator annihilates the
vacuum state: p̂µ |Ωi = 0.
Pn
Corollary: p̂µ |p1 p2 . . . pn i = ( j=1 pjµ ) |p1 p2 . . . pn i.
23
Lorentz Invariance in the Heisenberg Picture
So, this operator allows unitary quantum spacetime translations, in the Schrödinger
µ
picture, via the exponentiated Hermitian operator quantity U (a) = e−iaµ p̂ .
Now, to manifest any symmetries that may have not been shown in the Schrödinger
picture, explore Lorentz invariance in the Heisenberg picture, which is also later
helpful in perturbation theory. Real space calculations, at a specific spacetime
location (e.g., (t, x)) are also much easier in the Heisenberg picture than in the
”spread-out” Fourier transformed Schrödinger picture.
To enter the Heisenberg picture, where time is explicitly included, an operator O

is unitarily transformed, and its time evolution is determined via the Hamiltonian
in the Heisenberg equation of motion
OH = eiĤt Oe−iĤt (125)

dOH
= i[Ĥ, OH ]. (126)
dt
In the Heisenberg picture, the commutation relations for the canonical position
and momentum operators become
[φ̂H (t, x), φ̂H (t, y)] = [π̂H (t, x), π̂H (t, y)] = 0 (127)
(3)
[φ̂H (t, x), π̂H (t, y)] = iδ (x − y). (128)
Evolve the canonical position and momentum operators in time via the (spatially
localized) Heisenberg equation of motion
dφ̂(t, x)
= i[ĤKG , φ̂(t, x)] = π̂(t, x) (129)
dt
dπ̂(t, x)
= i[ĤKG , π̂(t, x)] = ∇2 φ̂(t, x) + m2 φ̂(t, x). (130)
dt
Where the second equality is gotten by using integration-by-parts. Substitute
the first equality for π̂(t, x) into the second equality, and combine the second
derivatives of space and time, to show that the canonical field position operator
obeys the Klein-Gordon equation
(∂ µ ∂µ + m2 )φ̂(t, x) = 0. (131)
This completes the development of the unitary representation of spacetime

translations. Rotations and boosts are yet to be integrated into the unitary
representation of the Poincaré group.
24
Lecture 6: Causality in Scalar QFT
Thus far, we have diagonalized the Klein-Gordon Hamiltonian
1
Z Z d3 p
ĤKG = 3 2 2 2 2
d x π̂ (x) + (∇φ̂(x)) + m φ̂ (x) = â† âp (132)
2 (2π)3 p
And used this to construct the unitary operator that allows us to study the
dynamics of Klein-Gordon’s solution to Schrödinger’s equation in the Heisenberg
picture
Û = e−iĤKG t . (133)
We found the spatial solutions φ̂(x), in the Schrödinger picture, for the Klein-
Gordon equation, to be the position field operators
d3 p
Z
1

φ̂(x) = 3
p (134)
(2π) 2ωp
Notation: The spatial 3-vectors of position x and momentum p are no longer

bold-faced, and the spacetime 4-vectors will be bold-faced, such that
x = (x0 , x) = (t, x) = (x0 , x1 , x2 , x3 ) (135)

0 1 2 3
p = (p0 , p) = (ωp , p) = (x , x , x , x ). (136)
These solutions are represented in the Heisenberg picture via
φ̂(t, x) = Û † φ̂(x)Û = eiĤKG t φ̂(x)e−iĤKG t . (137)
We still need to complete the (projective) unitary representation of the Poincaré

group, including translations, boosts, and rotations, since we are doing relativistic
quantum field theory.
The next step here is to check that φ̂(t, x) respects causality, such that if
two spacetime events are space-like separated, then they have no influence on
each other. Note that if the two spacetime events are time-like, there may be
influence.
Recall the commutation relation of the Hamiltonian (dropping subscript ”KG”)

and the ladder operator
[Ĥ, âp ] = −ωp âp =⇒ eiĤt âp e−iĤt = e−iωp t âp . (138)
Substituting φ̂(x) into the Heisenberg picture, and using the above commutation
relation, we have the field oeprator in the Heisenberg picture
d3 p
Z
1

φ̂(t, x) = (139)
(2π)3 2ωp
p
25
Consider the delocalized (”smeared out”) field operator φ̂(t, x) as an observable
that samples the field at a localized spacetime location x = (t, x). The question
is whether this interpretation respects causality.
Consider a (projective) measurement event at (t, x) of the quantum field φ̂(t, x).
This disturbance of the field should fly out at the speed of light along its
spacetime light cone. The fact is that measuring the field causes an instantaneous
disturbance everywhere, but relativity is safe since we can not signal, send
information, faster than the speed of light (outside of the forward light cone
of the measurement event).
Figure 4: Sketch of expected propagation of information (at the speed of light)

due to measurement event of the field φ̂(t, x).
The result is that no information may be transmitted via the field across a
space-like interval ((xA − xB )2 < 0).
Now, we have to agree on which quantities are observable, and a natural guess
is to study the correlation function
h0| φ̂(x0 , x)φ̂(y 0 , y) |0i . (140)
This is unfortunately wrong, since the correlation function has no operational

meaning, and can not be directly measured, since the field operators are not
Hermitian, in general.
Digression: Interference experiment
26
To study correlation functions which can be measured in the lab via experiment,
consider the following interference experiment set up with a Klein-Gordon field
with auxiliary modes of light used to perform measurements and generate results.
Figure 5: Schematic of experiment to measure the correlation of two spacetime

locations and the Klein-Gordon field under measurement.
1. Prepare the field and auxiliary system (left and right) in the vacuum state
|0if ield |0ilef t |0iright . (141)
2. Apply the Hadamard gate (beam splitter) to the left and right auxiliary
states
1
|0if ield √ |0ilef t |1iright + |1ilef t |0iright . (142)
2
3. Create a particle at x or y via the unitary operator
Û (x) ⊗ |1i h1|lef t ⊗ Iright + Û (y) ⊗ Ilef t |1i h1|right (143)
Applied to the state in Step 2, which evolves to the state (dropping ”left”
and ”right” labels)
1
√ Û (y) |0if ield |0 1i + Û (x) |0if ield |1 0i (144)
2
4. Apply a second beam splitter to the auxiliary states to check for interference
in the final state
1 1
Û (x) + Û (y) |0if ield |0 1i + Û (x) − Û (y) |0if ield |1 0i (145)
2 2
27
5. Detect auxiliary states |0 1i and |1 0i.
The probability of measuring one of the states, |0 1i, for example, is

1
P(|0 1i) = h0|f ield Û † (x) + Û † (y) Û (x) + Û (y) |0if ield (146)
4
1 1 h i
= + Re h0|f ield Û † (x)Û (y) |0if ield (147)
2 2
1 1 h i
= + Re h0|f ield e−iφ̂(x) eiφ̂(y) |0if ield (148)
2 2
1 1 h 1 2
i
P(|0 1i) = + Re h0|f ield e−iφ̂(x)+iφ̂(y)+ 2 [φ̂(x),φ̂(y)] |0if ield (149)
2 2
(150)
Where the last line is gotten via the Baker-Campbell-Hausdorff relation, since
the two fields operators do not necessarily commute, and the interference between
the two measurement events is determined by whether the commutator is zero
or nonzero, which is calculated
d3 p
d3 q
Z Z
1
[âp , â†q ]e−ip·x+iq·y + [â†p , âq ]eip·x−iq·y

[φ̂(x), φ̂(y)] =
(2π)3 3
p
(2π) 4ωp ωq
(151)
Z 3
d p 1
= e−ip·(x−y) − eip·(x−y) · I (152)
(2π)3 2ωp
p
[φ̂(x), φ̂(y)] = ∆(x − y) · I (153)
Where the commutation relation [âp , â†q ] = (2π)3 δ (3) (p − q) · I is used, and the
quantity ∆(x − y), the correlation function, which is Lorentz invariant, must
be zero when x and y are space-like separated, such that (x − y)2 < 0.
Consider the space-like separation, (x − y)2 < 0, and enter a reference frame
where x − y = (0, x − y), and the correlation function ∆(x − y) becomes zero!
d3 p
Z
1 1
ip·(x−y) −ip·(x−y)

∆(x − y) = e − e (154)
(2π)3 |p|2 + m2
p
2
d3 p
Z
1 1
= 3
p eip·(x−y) − eip·(x−y) (155)
2 (2π) |p|2 + m2
∆(x − y) = 0 (156)
Where the second line is gotten by applying the Lorentz transformation (x −

y) → −(x − y), which is allowed for in a space-like interval.
28
Figure 6: Sketch of spatial inversion of spacetime location y.
Therefore, information does not travel faster than the speed of light, and the
field operators respect causality when they are space-like separated! Note that
this does not yet prove that the entire theory is causal.
As a side note, consider the case where the two field measurements are time-
like separated, such that (x − y)2 > 0, and enter a reference frame where
x−y = (t, 0, 0, 0). In this reference frame of time-like separation, the correlation
function can not be zero
d3 p
Z
1
e−iωp t − eiωp t

∆(x − y) = 3
p (157)
(2π) 2ωp
Z ∞
1 p
= 2
dE E 2 + m2 e−iEt (158)
4π m
∆(x − y) ∼ e−imt − eimt 6= 0. (159)
Return to consider the (non-physical) two-point correlation function in a space-
like interval
D(x − y) = h0| φ̂(x)φ̂(y) |0i (160)
d3 p 1 −ip·(x−y)
Z
= e (161)
(2π)3 2ωp
d3 p
Z
1 1
= 3
p eip·(x−y) (162)
2 (2π) |p| + m2
2
m 1
= K1 (m|x − y|) (163)
4π 2 |x − y|
D(x − y) ∼ e−m|x−y| 6= 0 (164)
Where K1 (x) denotes the Hankel function.
29
So, when x and y are space-like separated, this correlation function can not be
zero, and can therefore not carry any information, as it would have to travel
faster than the speed of light.
An important application of the correlation function is the Feynman propagator,

which is used later in perturbative expansions for describing interactions.
(
D(x − y), x0 > y 0
∆F (x − y) = (165)
D(y − x), x0 < y 0
∆F (x − y) = h0| T [φ̂(x)φ̂(y)] |0i (166)
Where T [ ] is called the time-ordering operator

(
φ̂(x)φ̂(y), x0 > y 0
T [φ̂(x)φ̂(y)] = (167)
φ̂(y)φ̂(x), x0 < y 0
Another important definition of the Feynman propagator is in terms of complex

variables and contour integrals.
Lemma:
d4 p ie−ip·(x−y)
Z
∆F (x − y) = (168)
(2π)4 |p|2 − m2 + i
d3 p dp0 ie−ip·(x−y)
Z Z
= (169)
(2π)3 C 2π |p|2 − m2 + i
Where the integrand of the contour integral over the complex variable p0 has
two poles at ±i, and the contour is taken along the real p0 axis, and closed in
the upper or lower half-plane, depending on the value of p0 .
30
Figure 7: Poles of the Feynman propagator, where the contour may be taken
in the upper half-plane for t < 0, and in the lower half-plane for t > 0, where
t = x0 − y0 .
Lastly, an observation that the Feynman propagator, which is related thus far
to the two-point correlation function, is also a Green’s function (inverse of a
differential operator) for the Klein-Gordon partial differential equation
(∂02 − ∇2 + m2 )∆F (x − y) = −iδ (4) (x − y) (170)

−1
∼ L̂ · L̂ =I (171)
Where the left-hand side is the product of a linear differential operator, the
Klein-Gordon operator, and its inverse, the Feynman propagator, and the right-
hand side of the equation is, in essence, the identity.
31
Lecture 7: Representing Symmetries in QFT
Here we finish study of the quantum Klein-Gordon field by working out how
the Lorentz group is unitarily represented on the space of states on the Klein-
Gordon field. Recall that for a relativistic quantum field theory, we must have
a (projective) unitary representation of the Poincaré group U (Λ, a) on a Hilbert
space H. Thus far, we have constructed the Hilbert space, with respect to some
norm || · ||, as the space of states gotten by applying creation operators to the
vacuum state
H = span{â†p1 â†p2 . . . â†pn |Ωi}||·|| (172)
Digression: Continuous Groups of Symmetries

The central idea of a group with a continuous manifold structure (e.g., Lie
groups) is to study symmetries close to, localized to, the identity and then
exponentiate to larger, more global, elements. Important continuous operations
on this manifold M include (closed) composition and inverse.
Figure 8: Sketch of a group manifold with identity I and an element of the

manifold g.
Closure: M × M → M (173)
g×h→g·h (174)
Inverse: M→M (175)
−1
g→g (176)
The Poincaré group is an example of a continuous Lie group, and to understand
its structure, consider elements g ∈ M infinitesimally close to the identity I,
such that g − I ∼ O(). These infinitesimal elements are elements of the tangent
space TI M, which is a linear space and is called a Lie algebra. The basis vectors
of TI M are written as xj , j = 1, 2, . . . , dim(M).
32
Examples of Tangent Spaces
1. 1 × 1 unitary matrices U (1) = {φ ∈ C : |φ|2 = 1}
→ T1 U (1) = {z : (1 + z)∗ (1 + z) = 1 to order , s.t. Re[z] = 0)}
2. 3 × 3 Euclidean rotation matrices O(3) = {O ∈ M3 (R) : OT O = I}
→ TI O(3) = {X : (I + X)T (I + X) = I to order , s.t. X + X T = 0}
- Note that including inversions promotes O(3) to SO(3).
- Basis of O(3) =
0 10 0 01
P3 0 0 0

j 1 −1 0 0 , J 2 =
{X = j=1 Xj J : J =
0 00 , J3 = 0 0 1 }
0 00 −1 0 0 0 −1 0
3. 4 × 4 Lorentz transformation matrices G = {Λ : ηµν Λµρ Λνσ = ηρσ }
→ TI G = {ω : ηµν (I + ω)µρ (I + ω)νσ = ηρσ }

= {ω : antisymmetric, s.t. ω µν = −ω νµ }
Note the ”upstairs” covariant indices on ω
Using ”downstairs” contravariant indices
requires multiplication by the metric ηµν as above.

6
X 1
= {ω : change of basis to ω = Ωj J j }
j=1
2
Change indices via the bijection j → (ρ σ), with 0 < ρ < σ ≤ 3
1 → (0 1); 2 → (0 2); 3 → (0 3)
4 → (1 2); 5 → (1 3); 6 → (2 3)
X1
= {ω : with index bijection ω = Ω(ρσ) J (ρσ) } (Exercise)
2
(ρσ)
This bijection allows the basis to be expressed
by the formula [J (ρσ) ]µν = η ρσ δ σν − η σµ δ ρν
Understanding the structure of a linear space infinitesimally close to the identity

yields information about the whole Lie group and the global structure of the
manifold. Also note that multiplication on this linear space is a continuous map
that strongly determines the group structure on the manifold.
To demonstrate how a Lie algebra on the (linear) tangent space produces a Lie
group, and vice versa, consider the element of the tangent space of a manifold
X ∈ TI M, such that I + sX ∼ eX ∈ M, to order .
Define the function g(s) = limn→∞ (I + ns X)n = esX ∈ M. Therefore, every

element of the Lie algebra (tangent space to the identity) determines an element
33
of the Lie group (manifold). To go the other way, and determine the Lie algebra
from the Lie group, apply the logarithm map.
Figure 9: Figure of exponentiation map on Lie group (e.g., O(3)) manifold.

Translation from identity a distance s along the X direction. Resulting position
on manifold is called the exponential of sX.
Algebraic structure on tangent space

The algebraic structure on the tangent space is defined by the group commutator
on the manifold, which is also a group.
[, ] : M × M → M (177)
−1 −1
(g, h) → [g, h] = ghg h , ∀ g, h ∈ M (178)
The group commutator also pushes forward to a mapping of the tangent space
[ , ] : TI M × TI M → TI M (179)
Consider the following group commutator which is an element of the manifold
[I + X, I + δY ] = (I + X)(I + δY )(I − X)(I − δY ) (180)
2 2
= I + δ(XY − Y X) + O( ) + O(δ ) + . . . (181)
∼ I + δ[X, Y ] (182)
Where and δ are small, independent parameters, and the Lie algebra commutator
[X, Y ] is, therefore, an element of the tangent space of the manifold, such that
[X, Y ] ∈ TI M.
34
Examples of Lie algebra commutators
1. U (1): trivial
2. O(3): [J j , J k ] = −jkl J l
3. Lorentz group: [J ρσ , J τ ν ] = η στ J ρν − η ρτ J σν + η ρν J στ − η σν J ρτ
E.g.,
   
0 1 0 0
1 0
0  0 −1 0
[J 01 ]µν  , [J 12 ]µν = 

= 
0 0 0 1
0 0
 
0 0 0 0
 
0 0
 0 0 −1
[J 01 , J 12 ] = −J 13 = 
0 0

0

0 1
Special names for this particular Lie group elements in the Lorentz
transformation
1 j
Generators of boosts (pure boost by exponentiating, s.t. e 2 j K ):
J 0j = K j
Generators of rotations (elements of O(3) ⊂ Lorentz group):
J 12 = J 1 , J 13 = J 2 , J 23 = J 3
Representations of Lie groups

Consider a representation π of the Lie algebra TI M which is a linear map from
the tangent space to the Hilbert space of bounded linear operators, such that
the Lie bracket property is preserved, such that [π(X), π(Y )] = π([X, Y ])
π : TI M → B(H) (183)
This representation is exponentiated to a representation of the Lie group manifold

M
π(g = eX ) = eπ(X) (184)
This shows that one can either try to find matrices that obey the Lie group law,
or, more easily, focus on the Lie algebra (linear space) and find matrices that
obey the Lie bracket property; Lie group representations are often not worked
with directly, but the elements of the Lie algebra can just be exponentiated to
obtain the representation of the Lie group.
To implement a general Lorentz transformation on the Hilbert space of states

allowed in the Klein-Gordon field, Noether’s theorem, as well as the inverse
35
Noether’s theorem, is employed. Noether’s theorem allows conserved currents
to be derived from symmetry transformations. Information in thrown out when
integrating the currents over space d3 x to get the conserved charge, but the
R
time-like component is left alone, and information about the structure of the
symmetry transformation is conserved, which can be gotten back by the inverse
Noether’s theorem.
Recap of Noether’s theorem: For each symmetry of a field, with respect

ρσ
to a coordinate transformation x → expJ , there exists a conserved current
per J ρσ . For example, the group of Lorentz transformations is described by
six independent parameters, the generators of the transformation, associated
with six conserved currents. The conserved currents from the symmetries of the
Lorentz transformation have the form
J (ρσ) = xρ T µσ − xσ T µρ (185)
Where xµ are 4-vector spacetime coordinates, and T µν are elements of the

energy-momentum tensor. The conserved charges are then gotten by integrating
over space Z Z
Q(ρσ) = d3 x J (ρσ) = d3 x (xρ T 0σ − xσ T 0ρ ) (186)
”Noether’s theorem is really just a fancy telescoping series in disguise.”
The arguably more profound statement regarding conserved charges and symmetries
is the inverse Noether’s theorem.
Inverse Noether’s theorem: Conserved charges are the generators, represent

the Lie algebra, of the symmetry transformations from which they came, and
generate canonical transformations, or representations of the symmetries, on
phase space. Classically,
{Q(ρσ) , Q(τ ν) }P B = η στ Qρν − η ρτ Qσν + η ρν Qστ − η σν Qρτ (187)
Propose that we ”just put hats on” the conserved charges, and check that they
obey the Lie algebra of the Lorentz group. It turns out that this works for
free theories, and we have, at least, one representation of the Lie algebra of the
Lorentz group in the context of one, the Klein-Gordon, quantum field.
Z
Q̂µν = d3 x(xµ T̂ 0ν − xν T̂ 0µ ) (188)
Consider the 0j th conserved charge, using the specialized notation, and check
36
that Q̂0j = K̂ j does indeed generate boosts and is time independent.
Z
Q̂ = K̂ = d3 x (xj T̂ 00 − x0 T̂ 0j )
0j j
(189)
Z
K̂ j = −t P̂ j + d3 x xj Hˆ (t, x) (190)
dK̂ j
Z
= −P̂ j + i[Ĥ, d3 x xj Hˆ (t, x)] (191)
dt
0 = P̂ j + i[Ĥ, K̂ j ] (192)
j j
[Ĥ, K̂ ] = −iP̂ . (193)
Where P̂ j is the total field momentum, and Hˆ is the Hamiltonian density, which
does not commute with the Hamiltonian Ĥ. The third line cancels the time
dependence of the two right-hand side terms. This shows that an infinitesimal
shift in time and an infinitesimal boost is equal to an infinitesimal shift in space.
Similarly for the generators of rotation, which are manifestly time independent,
it must be checked that they obey the correct Lie algebra.
Z
Jˆjk = d3 x π̂(x)(xj ∂k − xk ∂j )φ̂(x) (194)
To perform a Lorentz transformation on the Hilbert state space, a unitary

operator is created by putting some parameters Ω in front of the generators
of rotation and exponentiating
1 ˆρσ
Û (Λ) = e− 2 Ωρσ J (195)
There are 9 commutation relations, that must be checked (Exercises), that

yield the full Lie algebra of the Poincaré group.
[Jˆj , Jˆk ] = −ijkl Jˆl (196)

[Jˆj , K̂ k ] = −ijk K̂ l l (197)
j
[K̂ , K̂ ] =k
ijkl Jˆl (198)
[Jˆj , P̂ k ] = −ijkl P̂ l (199)
j k jk
[K̂ , P̂ ] = iδ Ĥ (200)
j j
[K̂ , Ĥ] = iP̂ (201)
[Jˆj , Ĥ] = [P̂ j , Ĥ] = [P̂ j , P̂ k ] = 0 (202)
This now demonstrates how the Klein-Gordon field gives a full (Lie algebra)
representation of the Poincaré group, and proves that to perform a Poincaré
transformation on a state of Klein-Gordon particles, one simply applies a unitary
transformation via exponentiation of the above operators, which are the generators
of transformations.
37
Lecture 8: Interactions in QFT
Thus far, we have studied the Klein-Gordon quantum field, which evolves with
time in the Heisenberg picture via the Klein Hamiltonian HKG , the generator
of time translations
φ̂(t, x) = eiĤKG t φ̂(0, x)e−iĤKG t . (203)
We have obtained a full unitary representation of the Poincaré group for the
Klein-Gordon field by constucting a space of states in terms of the field position
operator φ̂ and the field momentum operator π̂ via the generator of time translation
Ĥ, the generators of spatial translation P̂ j , and the conserved charges Q̂µν . This
is a free theory, where the dynamics of two or more spacetime events evolve
completely independently of each other with no interactions between particles
and field, and is relatively easy to solve.
To attempt to account for interactions, construct a Hilbert space spanned

by states of the form {â†p â†q |0i}, and add a (spatially localized) momentum
distribution Z 3 3
d pd q
|Φ2 i = φx (p)φy (q) · â†p â†q |0i (204)
(2π)6
This states evolves according to the Hamiltonian
|Φ2 (t)i = e−iĤKG t |Φ2 i (205)

Z 3 3
d pd q
= φx (p)φy (q)e−iĤKG t â†p eiĤKG t e−iĤKG t â†q eiĤKG t |0i (206)
(2π)6
Where ĤKG is quadratic in the creation operators â†p , meaning that the quantity
e−iĤKG t â†p eiĤKG t is linear in the creation operators â†p . Therefore, the particles
eveolve independently of each other in this attempted formalism, and there are
no interactions, which is unphysical for an interacting theory.
Desired characteristics of the interactions that we are attempting to describe

are
1. Model physical experiments
2. Maintain Lorentz invariance
3. Local interactions
To fulfill these characteristics, we consider studying models with (classical)
Lagrangian densities of the form
∞
1 1 X λn
L= (∂µ φ(x))(∂ µ φ(x)) − m2 φ2 (x) − φn (x) (207)
2 2 n!
n≥3
38
It will later be shown that Lagrangian densities with n > 4 are irrelevant
to observable physics, and n = 3 leads to instabilities, and neither case is
renormalizable. Therefore, the only relevant interacting scalar quantum field
theory is the n = 4 case
1 1 λ
L= (∂µ φ(x))(∂ µ φ(x)) − m2 φ2 (x) − φ4 (x) (208)
2 2 4!
In a quantum field theory, interactions are handled in several ways
1. Perturbation theory
Expand Hamiltonian in Taylor series in terms of a small parameter
Leads to a solvable model when this parameter is set to zero
Feynman diagrams systematically handle all interactions in infinite series
2. Variational methods
Approximates the system and minimizes error parameters
3. Monte Carlo sampling

4. Exact solutions
Bethe Ansatz in (1 + 1) dimensions
Topological QFT in (2 + 1) dimensions
Supersymmetry in higher dimensions
Large N limit
Perturbation theory
Consider the ”small” addition Ĥint to the free theory Hamiltonian Ĥ0 to make
the full Hamiltonian Ĥ
Ĥ = Ĥ0 + Ĥint (209)
Technically, we demand that ||Ĥint ||∞ << 1, but it often happens that ||Ĥint ||∞ →
∞, where ||Ĥint ||∞ is the largest eigenvalue that dominates the error estimates.
Therefore, we pretend that ||Ĥint ||∞ << 1, and solve the time-dependent
Schrödinger equation
d
i |ψi = Ĥ |ψi (210)
dt
Interaction picture
Enter a new reference frame, the interaction picture, or the Heisenberg
picture, where all states and operators from the Schrödinger picture, denoted
by subscript ”S”, are transformed via
|ψI (t)i = eiĤ0 t |ψS (t)i (211)

O = eiĤ0 t OS e−iĤ0 t (212)
39
The time evolution on the interaction space of states is then
d d
i |ψI (t)i = i eiĤ0 t |ψS (t)i (213)
dt dt
d
= −Ĥ0 eiĤ0 t |ψS (t)i + ieiĤ0 t |ψS (t)i (214)
dt
= −Ĥ0 eiĤ0 t |ψS (t)i + ieiĤ0 t (−iĤ |ψS (t)i) (215)
iĤ0 t iĤ0 t
= (−Ĥ0 e +e (Ĥ0 + Ĥint )) |ψS (t)i (216)
eiĤ0 t + eiĤ0 t (Ĥ0 + Ĥint )) · e−iĤ0 t eiĤ0 t · |ψS (t)i

−Ĥ
= (0 (217)
d
i |ψI (t)i = (Ĥint )I (t) |ψI (t)i (218)
dt
Note: From here we drop the subscript ”I” on the interaction Hamiltonian
(Ĥint )I (t) → Ĥint (t). (219)
The interacting time-dependent solution is
|ψI (t)i = Û (t, t0 ) |ψI (t0 )i . (220)
Where the operator Û (t, t0 ) is the propagator, and satisfies the equation
d
i Û (t, t0 ) = Ĥint (t) Û (t, t0 ) (221)
dt
Integrating this equation with respect to t yields a constraint on the propagator
Z t
Û (t, t0 ) = I − i dt0 Ĥint (t0 )Û (t0 , t0 ) (222)
t0
One way to solve for Û (t, t0 ) is to guess a solution and check if both sides of the
constraint equation are equal.
Another approach is through fixed point iteration

1. Make a guess for Û (t, t0 )
2. Evaluate how wrong it is
3. Minimize error by adding and/or modifying terms to guess
4. Repeat, by substituting the old right-hand side into the new right-hand
side, until the left-hand side and the right-hand side of the constraint
approaach each other
The repeated substitution of the propagator into the constraint equation produces
the Dyson series, where the nth has the form
Z t Z t0 Z t(n−1)
Û (t, t0 ) = (−i)n dt0 dt00 · · · dt(n−1) Ĥint (t0 )Ĥint (t00 ) . . . Ĥint (t(n−1) )
t0 t0 t0
(223)
40
By the triangle inequality and the product inequality, the norm of the nth term
has an upper bound
(t − t0 )n
Z
|| . . . ||∞ ≤ (||Ĥint ||∗∞ )n . (224)
n!
And, since Ĥint (t) is just unitarily rotated from Ĥint
||Ĥint ||∗∞ = sup ||Ĥint (t0 )||∞ = (||Ĥint ||∞ )S . (225)
t0 ∈[t,t0 ]
If ||Ĥint ||∞ << 1, the series has a nonzero radius of convergence.

dt0 Ĥint (t0 )
Rt
−i
Theorem: Û (t, t0 ) = T [e t0
], where T [ ] is the time-ordering operator.
To prove, expand the right-hand side in a Taylor series, apply time-ordering,

and check that the two sides are equal.
Observables in QFT
An important observable in QFT is scattering cross sections in scattering experiments.
Namely, the S-matrix is determined by Green’s functions (n-point correlation
functions)
G(n) (x1 , x2 , . . . , xn ) = hΩ| T [φ̂1H φ̂2H . . . φ̂nH ] |Ωi (226)
Where |Ωi is the vacuum state of the Hamiltonian, and the subscript ”H”
denotes the Heisenberg picture, such that φ̂jH = φ̂(xj ) = φ̂(tj , xj ).
Claim:
h0| T [φ̂1I φ̂2I . . . φ̂nI Ŝ] |0i
G(n) (x1 , x2 , . . . , xn ) = (227)
h0| Ŝ |0i
Where hφ| Ŝ |ψi = limt± →±∞ hφ| Û (t+ , t− ) |ψi, and Ĥ0 |0i = 0.
Proof:
Assume that t1 > t2 > · · · > tn .
Then the right-hand side of the numerator reads
h0| Û (∞, t1 )φ̂1I Û (t1 , t2 )φ̂2I . . . φ̂nI Û (tn , −∞) |0i (228)
= h0| Û (∞, t1 )φ̂1H φ̂2H . . . φ̂nH Û (t0 , −∞) |0i (229)
Where φ̂H (t, x) = Û † (t, t0 )φ̂I (t, x)Û (t, t0 )

Now dealing with
0 0
Û (t0 , −∞) |0i = 0 lim lim eiĤ0 (t−t0 ) e−iĤ(t−t ) e−iĤ0 (t −t) |0i (230)
t →−∞ t→t0
X 0
= 0 lim (|Ωi hΩ| + e−iEn (t0 −t ) |En i hEn |) |0i (231)
t →−∞
n>0
41
Where the nonvanishing terms are written in the eigenbasis
P of RĤ.
Quantum fields have a continuous spectra, such that n>0 ∼ dE.
Invoke the Riemann-Lebesgue Lemma (limk→∞ ψ̂(k) = limk→∞ ψ(x)eikx dx =
R
0), and consider

Z
0
lim dE hE|0i e−iE(t−t0 ) hφ|Ei = 0 (232)
t →−∞
The numerator is now equal to
h0|Ωi hΩ|0i hΩ| φ̂1H . . . φ̂nH |Ωi (233)
Where h0|Ωi hΩ|0i is equal to the denominator and cancels. QED.
Essentially, interacting quantum field theories come down to throwing in a

Taylor series for the S-matrix and φ̂iI , and truncating some terms.
42
Lecture 9: Interactions and Feynman Diagrams
The story so far
• Built a free (non-interacting) relativistic quantum field theory, namely,
the Klein-Gordon field, with Hamiltonian ĤKG .
• Added (Lorentz invaraint) interactions via the Hamiltonian
Z
λ
Ĥint = d3 x φ̂4 (x), λ >> 1 (234)
4!
.
• Used perturbation theory to solve the Hamiltonian Ĥ = ĤKG + Ĥint .
• Studied observable quantities via the n-point correlation function
G(n) (x1 , . . . , xn ) = hΩ| T [φ̂H (x1 ) . . . φ̂H (xn )] |Ωi (235)

.
Where |Ωi is the full, interacting vacuum state, and φ̂H (xi ) = φ̂iH =
φ̂I (xi ) = φ̂iI is the field operator in the Heisenberg, interaction, picture,
where the observables (e.g., field operators) closer to direct observation.
• Claimed and ”proved” that the n-point correlation function can be calculated
in terms of the ground state expectation values of the field operators, and
the scattering S-matrix .
h0| T [φ̂1I . . . φ̂nI S] |0i

G(n) (x1 , . . . , xn ) = (236)
h0| S |0i
Ĥint (t0 )dt0

Rt
Where hφ| S |ψi = limt→∞ hφ| T [e−i −t ] |ψi , ∀ |φi , |ψi
Now, to calculate quantities like the numerator of the n-point correlation function
consider the field operator in the interaction picture (dropping the subscript ”I”)
d3 p
Z
φ̂I (x) = φ̂(x) = p (âp e−ip·x + â†p eip·x ) = φ̂+ (x) + φ̂− (x) (237)
2ωp
p
Where p = (ωp , p), with ωp = p2 + m2 , such that p · x = p0 x0 − p · x, and the
newly defined operators annihilate the ground state, such that
φ̂+ (x) |0i = 0 and h0| φ̂− (x) = 0. (238)
43
For example, consider the time-ordering of the two particle case (note the
notation change for the four-vector in this section x → x)
T [φ̂(x)φ̂(y)]x0 >y0 = φ̂(x)φ̂(y)

= φ̂+ (x)φ̂+ (y) + φ̂+ (x)φ̂− (y)
+ φ̂− (x)φ̂+ (y) + φ̂− (x)φ̂− (y)

= φ̂+ (x)φ̂+ (y) + φ̂− (y)φ̂+ (x) + [φ̂+ (x), φ̂− (y)]
+ φ̂− (x)φ̂+ (y) + φ̂− (x)φ̂− (y)

T [φ̂(x)φ̂(y)]x0 >y0 = φ̂+ (x)φ̂+ (y) + φ̂− (y)φ̂+ (x) + D(x − y) · I
+ φ̂− (x)φ̂+ (y) + φ̂− (x)φ̂− (y)
Then the ground state expectation value of two interacting field operators is
simply the Feynman propagator
h0| T [φ̂(x)φ̂(y)] |0i = ∆F (x − y) (239)

d4 p e−ip·(x−y)
Z
=i , >0 (240)
(2π)4 p2 − m2 + i
(
D(x − y), x0 > y 0
= (241)
D(y − x), x0 ≤ y 0
Wick Contraction and Normal Ordering

Introduce some notation for extracting Feynman propagators from quantities
like the expectation value of time-ordered field operators, called the Wick
contraction. For two field operators, φ̂(x) and φ̂(y), and any three other
operators Â, B̂, and Ĉ, write
φ̂(x)φ̂(y) = ∆F (x − y) · I (242)
Aφ̂(x)B φ̂(y)C = ∆F (x − y) · ÂB̂ Ĉ (243)
Also introduce normal ordering, denoted by N [ ] that sends all ”dagger”

operators to the left. For example,
N [âp â†q âr â†s ] = â†q â†s âp âr (244)
Observe the relationship between time-ordering and normal-ordering using the

Wick contraction
T [φ̂(x)φ̂(y)] = N [φ̂(x)φ̂(y) + ∆F (x − y) · I] (245)
= N [φ̂(x)φ̂(y) + φ̂(x)φ̂(y) (246)
44
A more involved example of the time-ordering of four field operators, where
the only nonzero terms at the end of acting on states will be the ”double
contractions”, since
T [φ̂(x1 )φ̂(x2 )φ̂(x3 )φ̂(x4 )] = N [φ̂1 φ̂2 φ̂3 φ̂4 + ”all possible contractions”]
= N [φ̂1 φ̂2 φ̂3 φ̂4 + φ̂1 φ̂2 φ̂3 φ̂4 + φ̂1 φ̂2 φ̂3 φ̂4
+ φ̂1 φ̂2 φ̂3 φ̂4 + +φ̂1 φ̂2 φ̂3 φ̂4 + φ̂1 φ̂2 φ̂3 φ̂4 + φ̂1 φ̂2 φ̂3 φ̂4
+ φ̂1 φ̂2 φ̂3 φ̂4 + φ̂1 φ̂2 φ̂3 φ̂4 + φ̂1 φ̂2 φ̂3 φ̂4 ]
The ground state matrix elements of T [φ̂1 φ̂2 φ̂3 φ̂4 ] is then

( ( (
h0| T [φ̂1 φ̂2 φ̂3 φ̂4 ] |0i = (
h0|(
N( 1 φ̂2 φ̂3 φ̂4 ] |0i +
[φ̂( ( ( h0|
N[φ̂1 φ̂2 φ̂3 φ̂4 ] |0i +
.
..
+ ∆F (x1 − x2 )∆F (x3 − x4 ) + ∆F (x1 − x3 )∆F (x2 − x4 )
+ ∆F (x1 − x4 )∆F (x2 − x3 )
Where these are the values of the associated Feynman diagrams, which we write
down in a ”reverse” way, extracting the diagram from the calculated value.
Later, we will extract the values from the diagrams.
Figure 10: Feynamn diagrams representing the nonzero values in the four
particle example above.
Now stated in its general form
Wick’s Theorem: T [φ̂1 . . . φ̂n ] = N [φ̂1 . . . φ̂n + ”all possible contractions”].
Proof:
Induct on n, with the base case n = 2 confirmed to be true, and check that the
n − 1 case implies the full n case.
Assume, without loss of generality, that everything is time-ordered, such that
x01 > x02 > · · · > x0n .
Then the left-hand side of Wick’s theorem becomes
T [φ̂1 . . . φ̂n ] = φ̂1 φ̂2 . . . φ̂n . (247)
45
Use the inductive hypothesis for n − 1 case on this equation

− all possible contractions
φ̂1 φ̂2 . . . φ̂n = (φ̂+
1 + φ̂ 1 )N [φ̂ 2 . . . φ̂ n + ]
excluding φ̂1 .

all possible contractions
= φ̂+
1 N [φ̂2 . . . φ̂n + ]
excluding φ̂1 .

− − all possible contractions
+ N [φ̂1 φ̂2 . . . φ̂n + φ̂1 ].
excluding φ̂1 .
Since φ̂−
1 is already normal-ordered.
Focus on the first part of the first term of φ̂1 φ̂2 . . . φ̂n above
φ̂+ + +
1 N [φ̂2 . . . φ̂n ] = N [φ̂2 . . . φ̂n ]φ̂1 + [φ̂1 , N [φ̂2 . . . φ̂n ]]
= N [φ̂+
1 φ̂2 . . . φ̂n ]
+ N [[φ̂+ + +
1 , φ̂2 ]φ̂3 . . . φ̂n + φ̂2 [φ̂1 , φ̂3 ]φ̂4 . . . φ̂n + · · · + φ̂2 φ̂3 . . . φ̂n−1 [φ̂1 , φ̂n ]]
φ̂+ + + + +
1 N [φ̂2 . . . φ̂n ] = N [φ̂1 φ̂2 . . . φ̂n + φ̂1 φ̂2 φ̂3 . . . φ̂n + φ̂1 φ̂2 φ̂3 . . . φ̂n + · · · + φ̂1 φ̂2 φ̂3 . . . φ̂n ].
Where the second equality follows from [φ̂†j , φ̂k ] ∝ I.

Now, focus on the rest of the first term of φ̂1 N [φ̂2 . . . φ̂n ]

+ all possible contractions
φ̂1 N [ ] = [φ̂+ +
1 , N [. . . ]] + N [. . . ]φ̂1
excluding φ̂1 .

all possible contractions + all possible contractions
= N[ +
] + N [φ̂1 ]
including φ̂1 . excluding φ̂1 .
46
Lecture 10: Feynman Rules for ϕ4 Theory
In perturbative, interacting field theories, we must calculate the n-point correlation
function
h0| T [φ̂1 . . . φ̂n S] |0i
G(n) (x1 , . . . , xn ) = (248)
h0| S |0i
Where the scattering matrix elements and the interacting Hamiltonian are
0 0
Rt
hφ| S |ψi = lim hφ| T [e−i −t ĤI (t )dt ] |ψi (249)
t→∞
Z
λ
ĤI (t) = d3 x φ̂4 (x) (250)
4!
The numerator and denominator of the correlation function must be expanded
perturbatively in the small parameter λ.
a0 + λa1 + λ2 a2 + . . .
G(n) = , where aj , bj ∈ C (251)
b0 + λb1 + λ2 b2 + . . .
1 b1 b2
= (a0 + λa1 + . . . )(1 − λ + λ2 + . . . ) (252)
b0 b0 b0
1 b1
=O(λ) (a0 + λa1 )(1 − λ ) (253)
b0 b0
a0 a1 b1 a0
G(n) = + λ( − 2 ) to order λ (254)
b0 b0 b0
This gives the intermediate task of calculating the coefficients of the perturbative
expansion: a0 , a1 , b0 , and b1 .
Consider the n = 2 case

ĤI (t0 )dt0
Rt
(2) h0| T [φ̂(x)φ̂(y)e−i −t ] |0i
G (x, y) = lim Rt (255)
t→∞ −i ĤI (t0 )dt0
h0| T [e −t ] |0i
Expand the exponentials, keeping to order λ, and insert into numerator. The
numerator is then
0 0
Rt
h0| T [φ̂(x)φ̂(y)S] |0i = h0| T [φ̂(x)φ̂(y)e−i −t ĤI (t )dt ] |0i (256)
Z
iλ
= h0| T [φ̂(x)φ̂(y) I − d4 z φ̂4 (z) ] |0i (257)
4!
Z
iλ
= h0| T [φ̂(x)φ̂(y)] |0i − d4 z h0| T [φ̂(x)φ̂(y)φ̂4 (z)] |0i
4!
(258)
Z
iλ
h0| T [φ̂(x)φ̂(y)S] |0i = ∆F (x − y) − d4 z h0| N [φ̂(x)φ̂(y)φ̂4 (z) + all contractions] |0i
4!
(259)
47
The last line is gotten by applying Wick’s theorem, and recall that any terms
with uncontracted operators will evaluate to zero in the vacuum expectation
value (e.g., h0| N [. . . ] |0i), and only the fully contracted terms will contribute,
such as h0| φ̂(x)φ̂(y)φ̂(z)φ̂(z)φ̂(z)φ̂(z) |0i, and h0| φ̂(x)φ̂(y)φ̂(z)φ̂(z)φ̂(z)φ̂(z) |0i.
Now, ”all good physics and math ends in linear algebra, combinatorics, or both”,
and we use combinatorics to calculate how many fully contracted terms we
expect to see in the expansion.
• There are (2n−1)!! full contractions per expansion, where n is the number
of unique particles. In this case, n = 3 for x, y, and z spacetime coordinates.
• There are 2 unique contraction types out of the (2 · 3 − 1)!! = 15 full
contractions.
∆F (x − y)∆2F (z − z)
∆F (x − z)∆F (y − z)∆F (z − z)
• There are 3 contractions of the first type
Connect x → y in 1 way and z → z twice, each in only 1 way.
• There are 12 contractions of the second type
Connect x → z in 4 ways, followed by y → z in 3 ways, and 4 · 3 = 12.
Return to the numerator of the correlation function, and pretending that ∆F (z−
z) is finite, for now,
h0| T [φ̂(x)φ̂(y)S] |0i = ∆F (x − y)

Z
iλ
d4 z h0| N [( (4 (z)
(( all partial
+ − φ̂(x)
((φ̂(y)
(φ̂ + + contractions
all full
] |0i
4! contractions
Z
iλ
= ∆F (x − y) + 3 · − d4 z ∆F (x − y)∆2F (z − z)
4!
Z
iλ
+ 12 · − d4 z ∆F (x − z)∆F (y − z)∆F (z − z)
4!
And in diagrammatic form
Figure 11: Feynman diagram representation of the above integral values for the
ϕ4 interacting theory.
48
Similarly calculate the denominator of the two-point correlation function to
obtain the following
Figure 12: Feynman diagram calculation for the two-point correlation function
for the ϕ4 theory. Note that this ”miraculous” cancellation of divergent terms
(the ”self-interacting figure-eight”) actually follows from a more general result.
The arguments of the correlation function are the external vertices, and the
spacetime coordinates of the interacting Hamiltonian are the internal vertices.
To calculate the value of a diagram, associate a factor, the Feynman propagator,
∆F (x−y) to each
R edge connecting external vertices (e.g., x and y), and associate
the factor −iλ d4 z (dividing by the symmetry factor 4!) to each internal vertex
(e.g., z).
For example,
49
Figure 13: Example Feynman diagram and associated values with 4 external
vertices and 4 internal vertices.
In summary, to calculate G(2) (x, y) to arbitrary order, sum over all possible
diagrams with 2 external vertices, subject to the (canonical) Feynman rules for
ϕ4 theory.
h0| T [φ̂(x)φ̂(y)S] |0i sum of all possible diagrams

G(2) (x, y) = = (260)
h0| S |0i with two external vertices
Figure 14: Canonical Feynman rules for ϕ4 theory.
Note the abscence of the inverse 4! factor in rule number 2, as it is added based
on observation, and not considered a component on the canonical rules.
50
Higher order terms in the expansion of the two-point correlation function become
successively more complicated and often redundant. For example, The self-
interacting ”figure-eight” diagram of an internal vertex can occur in 4! · 8 = 192
ways. The redundancy is encoded in the symmetry factor (see rule 4 above)
of the diagram. Typically, in practice and computation, distinct diagrams are
written down and overcounting is determined via the symmetry factor.
51
Lecture 11: Feynman Rules and Vacuum Bubbles
The Feynman rules demonstrated in the last lecture are specifically for the ϕ4
theory of interacting quantum fields in position space, and allow the calculation
of n-point correlation (Green’s) functions, which are not directly observable,
but related to scattering amplitudes which are directly observable. We did not
rigorously prove, but demonstrated (with n = 2) the feasibility of and accepted
as ”definition”, that the n-point correlation function is equal to the sum of all
possible diagrams with n external vertices, subject to the Feynman rules in
position space. Keep note that we did not consider n > 2 or λ > 1 in the
following equality.
h0| T [φ̂(x1 ) . . . φ̂(xn )S] |0i

G(n) (x1 , . . . , xn ) = (261)
h0| S |0i
sum of all possible diagrams with n external vertices

= (262)
subject to Feynman rules in position space
A ubiquitous issue and deep concern in quantum field theory is the appearance
infinities in calculations. Nature seems to suggest that there are no infinities,
unless one asks the wrong question. (Are there actually any physical quantities
that can be proven to be infinite by experiment, such as the energy levels of the
hydrogen atom or the results of continuous scattering theory?) Many infinities
that appear due to the application of the Feynman rules will be trivially dispelled
(cf., in quantum mechanics when an infinite ground state energy is calculated,
simply apply a shift to make it finite). Vacuum bubbles are diagrams with no
external vertices (e.g., self interactions) that evaluate to infinity, but will cancel
in calculating the n-point correlation function G(n) (x1 , . . . , xn ), as we saw in
the n = 2 case in the last lecture.
Feynman Rules in Momentum Space (ϕ4 Theory)

Remember that the perturbation expansion of the S-matrix and the Green’s
function can be calculated in momentum space via the Feynman rules in momentum
space. The momentum space Feynman propagator has the form
d4 p ie−ip·(x−y)
Z
∆F (x, y) = (263)
(2π)4 p2 − m2 + i
Calculating the perturbation expansion in momentum space may make additional

cancellations more apparent. The Feynman rules for ϕ4 theory in momentum
space are as follows
i
1. For each propagator with momentum p, add a factor of p2 −m2 +i .
2. For each (internal) vertex, add a factor of −iλ.

3. For each external vertex, add a factor of e−ip·x .
52
4. Impose momentum conservation at each vertex.
5. Integrate over undetermined momenta.
6. Divide by symmetry factor.
Figure 15: Diagrammatic representation for the Feynman rules in momentum

space.
Returning to the n = 2 case, a general diagram consists of a product of connected

components and disconnected components. Note that the external vertices (x
and y in the n = 2 case) are always connected, since their degrees are odd. For
example,
53
Figure 16: Typical diagram for n = 2 case.
All possible disconnected pieces form a countable set.
Figure 17: Countable set of all possible disconnected pieces.
Each type of disconnected diagram Dj has a unique value v(Dj ) = vj , which

are all infinite. To interpret this result, place a cutoff on the theory. It must
be checked that the results, after evaluating the diagrams of the expansion, do
not depend on the cutoff imposed.
Suppose that a given diagram D has nj components of type Dj , including the

connected component. The value of the diagram is the product of the values of
the connected and disconnected components.
∞
Y 1 nj
vj = vconnected · v (264)
n
j=1 j
! j
Where nj ! is the symmetry factor.
For the n = 2 diagram example above, we have the product of the connected
piece with disconnected types D1 , D3 , and D2 , respectively, with nj = 1 for
each diagram type j. Therefore the value of the diagram is
v(D) = vconnected · v1 · v3 · v2 (265)
54
Figure 18: Typical diagram for n = 2 case with disconnected pieces labelled by
type.
This allows us to write down a more closed form for the series expansion for
G(2) (x, y), and ditch the ”all possible diagrams” bit. The numerator of G(2) (x, y)
is then
∞
X X Y 1 nj
Numerator(G(2) (x, y)) = vconnected · v (266)
n ! j
j=1 j
connected {nj }
∞
X XY 1 nj
= vconn. v (267)
conn.
n ! j
j=1 j
{nj }
∞ X
X Y 1 nj
= vconn. v (268)
conn. j=1
nj ! j
{nj }
X P∞
Numerator(G(2) (x, y)) = vconn. · e j=1 vj (269)
conn.
Note that we justify these manipulations of not-necessarily-convergent series by

the implemented cutoff, which makes each value finite. Similarly, the denominator
is simply the sum over the same exponential
P∞
vj
Denominator(G(2) (x, y)) = e j=1 (270)
This cancels exactly with the same term in the numerator, pulled out of the
sum over connected parts. Therefore, the 2-point correlation function, which
generalizes to n > 2, is equal to the sum of all connected components subject
to the Feynman rules.

(2) sum over all connected diagrams
G (x, y) = (271)
subject to Feynman rules
55
Figure 19: The sum over all connected diagrams subject to the Feynman rules.
Cutoffs in QFT
Consider the Klein-Gordon Hamiltonian
Z
1
ĤKG = d3 x π̂ 2 (x) + (∇φ̂(x))2 + m2 φ̂2 (x) (272)
2
This assumes an infinitie number of degrees of freedom, one per each point
in spacetime x ∈ M4 . Now, for example, our theory and understanding of
spacetime breaks down below the Planck scale. We can either continue to
work in ignorance, or remember that we are dealing in effective theories that,
hopefully, represent something more fundamental. For example, the Navier-
Stokes equation is an effective theory for quantum chromodynamics (QCD). An
effective theory is a description which explains all observations up to a given
scale s, often expressed in inverse length, and may break down beyond the given
scale.
To impose a cutoff on the Hamiltonian, add a ”fixing” term ĤΛ which is

well-behaved up to the scale Λ, and must match predictions of the original
Hamiltonian Ĥ
Ĥ → Ĥ 0 = Ĥ + ĤΛ (273)
To modify the new Hamitonian for other scales, Wilson’s renormalization
theory may be used, where the ”fixing” Hamiltonian ĤΛ is parameterized by
56
the field operators and their derivatives
X
ĤΛ = P(φ̂, ∇φ̂, ∇2 φ̂, . . . , π̂, ∇π̂, ∇2 π̂, . . . ) (274)
polynomials
= c0 φ̂ + c1 φ̂2 + · · · + d0 π̂ + d1 π̂ 2 + · · · + e1 (∇φ̂)2 + . . . (275)
Only a finite number of the coefficients are nonzero and have an effect on
observations beyond the scale Λ. By effect, the value of the term in the sum over
field operator polynomials is not suppressed by inverse powers of Λ. Such terms
that are not suppressed and have an effect on observations are called relevant
terms. Any cutoff that is imposed must differ only by relevant terms.
For example, the relevant terms up to 4 dimensions is
{φ̂, φ̂2 , ∇2 φ̂, π̂, π̂ 2 , φ̂3 , φ̂4 }. (276)
The simplest cutoff to impose on the momentum states of the free Klein-Gordon
Hamiltonian is
d3 p d3 p
Z Z
†
ĤKG = ωp âp âp → Ĥ KG + ĤΛ = ω â† â
3 p p p
(277)
(2π)3 |p|<Λ (2π)
And the simplest cutoff for ϕ4 interaction is

Z Z
λ
Ĥϕ4 = d3 x φ̂4 (x) → Ĥϕ4 + ĤΛ = d3 p1 d3 p2 d3 p3 d3 p4 (. . . ) (278)
4! |pj |<Λ
The new Hamiltonian for the quantum Klein-Gordon field with ϕ4 interactions,
and an imposed cutoff at scale Λ is then written as
Ĥ → Ĥ 0 = ĤKG + Ĥϕ4 + ĤΛ (279)

d3 p
Z Z
†
= ω â â +
3 p p p
d3 p1 d3 p2 d3 p3 d3 p4 (. . . ) (280)
|p|<Λ (2π) |pj |<Λ
57
Lecture 12: The S-matrix in ϕ4 Theory
Recall that some diagrams evaluate to infinity in the calculation of the time-
ordered n-point correlation function

G(n) (x1 , . . . , x n ) = hΩ| T [φ̂(x1 ) . . . φ̂(xn )] |Ωi = sumwith
over all connected diagrams
n external points/legs
(281)
Example: n = 4 Diagrams
For example, the n = 4 case sums over diagrams like the following
Figure 20: Sum over connected diagrams for n = 4 case.
Diagrams with one loop (first order ) evaluate to infinity, but are easily eschewed.
Two vertex diagrams (last one in figure above) is a much more difficult infinity
to tame, and causes divergences, typically of the form, in momentum space,
d4 p
Z
i
I= (282)
(2π)4 p2 − m2 + i
To attempt to tame the infinity, impose a cutoff at scale Λ
d4 p
Z
i
I → I(Λ) = 4 2 2
. (283)
|p|<Λ (2π) p − m + i
To cope with arbitrary choices of Λ, arbitrary coupling constants can be employed

to compensate and remove possible Λ-dependencies in calculations. Further
explanation can be found in the renormalization theory of scalar particles.
Scattering Theory
Scattering theory is an effective theory for large time limits |t| >> constant.
The probability of a scattering event occuring, the scattering amplitude, in the
58
case of well-collimated beams with incoming momenta kA and kB , and outgoing
momenta pj , j = 1, . . . , n, is related to the quantity
out hp1 p2 . . . pn |kA kB iin = hp1 p2 . . . pn | S |kA kB i . (284)
Figure 21: Schematic of scattering experiment with incoming momenta and

outgoing momenta as described above.
For the free, non-interacting, theory, we have the equation for incoming momenta
related to the free vacuum state, denoted by | i0 ,
2ωkA ωkB â†kA â†kB |0i0 .

p
|kA kB iin = (285)
The task at hand is to calculate the scattering amplitude hp1 p2 . . . pn | S |kA kB i.

Define the S-matrix as
S = I + iT̂. (286)
Where the identity I is the dominant term in small intercation, where almost
nothing happens, and the operator T̂ is the dominant in larger interactions.
Introduce the quantity M to factor out the conservation of momentum in the
scattering process, where all momenta are ”on shell”, such that p0 = ωp .
n
X
hp1 . . . pn | iT̂ |kA kB i = (2π)4 δ (4) ((kA + kB ) − pj ) · iM((kA , kB ) → pf ).
j=1
(287)
Recall that the differential scattering cross section of 2 → 2 particles in the
center of mass frame is
|M|2

dσ
= 2 (288)
dΩ CM 64π 2 ECM
To compute the scattering amplitude hp1 p2 . . . pn | S |kA kB i, take ”on faith”
that for small interactions
|kA kB iint ∝ lim e−iĤt |kA kB i0 . (289)

t→∞
59
And compare this to the result related to the Riemann-Lebesgue lemma
|Ωiint = lim e−iĤt |0i0 . (290)
t→∞
An argument for the plausibility for the second equality of |Ωiint , where Ĥ =
ĤKG + Ĥint , ĤKG |0i0 = 0, and Ĥ |Ωiint = 0, is as follows. Suppose that the
Hamiltonian is parameterized by s ∈ [0, 1], such that Ĥ(s) = ĤKG + sĤint ,
and that the free Hamiltonian ĤKG has a spectral gap of ∆(s) = m. When
s = 0, we have the free, Klein-Gordon Hamiltonian, and when s = 1, we have
the interacting, ϕ4 Hamiltonian. Then it’s plausible that the spectral gap in
the spectrum of eigenvalues of Ĥ(s) will always be nonzero for at least just one
eigenvalue, and the spectrum is adiabatically connected (can’t instantaneously
go from massful to massless, but interactions can be turned on/off gradually).
Figure 22: Schematic of the spectral gap for the parameterized Hamiltonian.
For the proportionality factor of |kA kB iint to be true for small interactions,
which is far more radical, it is required that there are nonzero spectral gaps,
for some s > 0, at each momentum eigenstate of momenta kA and kB , and
there are no other nearby eigenvalues that may be mistaken for the incoming
momenta. In practice, these incoming particles can create bound states, for
example, and make this distinction difficult. If all of this is justifiable, and
we accept the proportionality |kA kB iint ∝ limt→∞ e−iĤt |kA kB i0 , then we can
create equalities and proportionalities, although difficult, to quantities that we
can actually calculate
Ĥint (t0 )dt0
Rt
lim 0 hp1 . . . pn | e−ihatH·2t |kA kB i0 ∝ lim 0 hp1 . . . pn | T [e−i −t ] |kA kB i0
t→∞ t→∞
(291)
This is exactly the same as the argument for the Dyson series expansion for G(n)
with the vacuum states, and, just as in that case, the proportionality difficulty
can be eliminated, and we have the following (not yet justified) equality
Rt 0 0

hp1 . . . pn | iT̂ |kA kB i = lim 0 hp1 . . . pn | T [e−i −t Ĥint (t )dt ] |kA kB i0 connected,
t→∞
amputated
(292)
60
The condition of ”connected, amputated” is analogous to ”connected” as in the
G(n) case.
We can begin to justify this equality through some calculations. Consider the
O(1) term, corresponding to the identity operator in S = I + iT̂, for the 2 → 2
scattering process.
out hp1 p2 |kA kB iin =O(1) 0 hp1 p2 |kA kB i0 (293)
= 2ωp1 ωp2 · 2ωkA ωkB h0| âp1 âp2 â†kA â†kB |0i
p
(294)
6
= 2ωA · 2ωB (2π) (δ(p1 − kA )δ(p2 − kB ) + δ(p1 − kB )δ(p2 − kA ))
(295)
Figure 23: Diagram of the O(1) term as calculated above.
The next term to O(λ), applying Wick’s theorem, and retaining all contractions,
since we are not just working with the vacuum state anymore,
−iλ
Z
hp1 p2 | iT̂ |kA kB i =O(λ) 0 hp1 p2 | T [ d4 x φ̂4I (x)] |kA kB i0 (296)
4!
−iλ
Z
= 0 hp1 p2 | N [ d4 x φ̂4I (x) + contractions
all
] |kA kB i0 (297)
4!
To see what kinds of terms that survive to order λ, consider the interacting
creation field operator interacting with the free four-momentum eigenstate
d3 k
Z
1
φ̂†I (x) |pi0 = √ e−ik·x âk |pi (298)
(2π)3 2ωk
d3 k
Z
1
e−ik·x âk · 2ωp â†p |0i
p
= 3
√ (299)
(2π) 2ωk
d3 k
Z
1 −ik·x 3 (3)
p
= √ e · (2π) δ (k − p) 2ωp |0i (300)
(2π)3 2ωk
φ̂†I (x) |pi0 = e−ip·x |0i . (301)
To deal with momentum eigenstates, such as |kA kB i, include them in extended
contractions by defining
φ̂†I (x)|pi = e−ip·x (302)
hp|φ̂†I (x) = eip·x (303)
61
Now, we can prove (Exercise) an extended version of Wick’s theorem, where the
time-ordered product of field operators in the presence of incoming and outgoing
momentum eigenstates is equal to the sum over all possible contractions, including
contractions of the field operators with the states as well.
X
0 hp1 . . . pn | T [field operators] |k1 . . . kn i0 =
all possible full contractions
including momentum eigenstates
(304)
This allows us to approximate the S-matrix transition amplitudes, justified up

to the assumption about momentum eigenstates in the free theory being related
to momentum eigenstates in the interacting theory.
62
Lecture 13: Feynman Diagram Expansions in ϕ4
Theory
In overview, thus far we have take a progression of reasonable, small steps to
building a relativistic quantum field theory. We established Lorentz invariance
as a symmetry of the theory, which may not quite be a fundamental symmetry
of nature, but has not been violated by any experiments to date. Next, we
established a unitary representation of the Poincaré group (difficult since the
Poincaré group is not compact, unlike rotation groups), where space and time
can transform into other via Lorentz boosts, meaning that energy and momentum
can also exchange roles under Poincaré symmetries. This also causes difficulties
in writing down quantum mechanical Hamiltonians that behave appropriately,
but, fortunately, we can invoke locality in the theory, which is also seemingly
fundamental to nature. Motivated by classical examples, such as the Klein-
Gordon equation, a free theory, we write the relativistic quantum field theory,
namely the quantum Klein-Gordon field, by ”just putting hats” on the operators,
and it worked.
To account for interactions, we added a perturbative ϕ4 term to the quantum

Klein-Gordon Hamiltonian
ĤKG → ĤKG + Ĥϕ4 (305)
Where Ĥϕ4 is Lorentz invariant, but is also an unbounded operator on any

Hilbert space, such that ||Ĥϕ4 || → ∞. Though this is not mathematically
rigorous, our forefathers and foremothers have shown through physical rigor,
as well as time and time again by using tools outside of the realm of their
mathematical applicability, that this is an acceptable perturbative term to
account for field interactions.
Examples of Mathematically Non-Rigorous Applications

• The building block of perturbation theory assumes that the interacting
theory vacuum state is proportional to the free theory vacuum state via
the time-evolution operator of the interacting theory
|Ωi ∝ limt→∞ e−iĤϕ4 t |Ω0 i.
– Even for finite dimensional systems, this limit does not exist and
is oscillatory. The limit can make sense for Hamiltonians with a
continuous spectra with a few particles.
• The momentum eigenstates require preparation of delta functions in the

momenta, and are assumed to be related similar to the vacuum states
|p1 . . . pn i = limt→∞ e−iĤϕ4 t |p1 . . . pn i0 .
63
• The Dyson series, a Taylor expansion of the interacting theory time evolution
operator can not be expected to converge
Rt 0 0
e−iĤϕ4 t = e−iĤKG t − i 0 dt0 e−iĤKG t ĤI eiĤKG t + . . . .
We are often punished by infinities, where we would prefer finite numbers. Some
infinities are trivially removable with no operational consequence, such as the
ground state energy shift in the Klein-Gordon theory, such that E0 → E0 + ∞
. Other infinities can be factored away by rescaling observables, such as in the
∞
case of vacuum bubbles (e.g., ∞ = 1). Yet other infinities just don’t go away,
such as in the diagram of two internal vertices.
Perhaps the choice of interaction terms is the cause of all of these infinities, and
we can impose a cutoff to the theory that will eliminate some infinities. For
example, in the ϕ4 theory
Ĥϕ4 → Ĥϕ4 (Λ) (306)
Where Λ is a cutoff up to some family of models/theories, and the norm of the
cutoff Hamiltonian is proportional to the cutoff, which may be large, but not
infinity, such that ||Ĥϕ4 (Λ)|| ∝ Λ.
For example, consider the quantum harmonic oscillator cutoff, where ||Ĥ(N )|| =
N
∞ N
X 1 X 1 X 1
Ĥ = (n+ ) |ni hn| → Ĥ(N ) = (n+ ) |ni hn|+ (N + ) |ni hn| (307)
n=0
2 n=0
2 2
n>N
The major concern with imposing cutoffs is whether observables depend on the
cutoff or not. To eschew this issue, we allow unobservable parameters, coupling
constants, to shift and absorb all the cutoff dependence.
Renormalization
This practice is called renormalization, and it works. We make the hypothesis
that the Hamiltonian depends on some number of parameters, called coupling
constants, such that Ĥ = Ĥ(z1 , . . . , zn ). The mapping from the coupling
constants to observables is not expected to be, and is usually not, bijective.
For example, Z
λ
Ĥϕ4 (λ, m) = d3 x ∇2 φ̂ + m2 φ̂2 + φ̂4 (308)
4!
This Hamiltonian corresponds to a list of observables Oj (z1 , . . . , zn ), j = 1, 2, . . . ,
with complicated (nonlinear) dependencies with the coupling constants, which
usuaslly exist on a smooth manifold before mapping to observables.
For example, consider the spectral gap O = E1 − E0 , where O = O(z1 +

c, z2 , . . . , zn ), and Ĥ = z1 · I + Ĥ 0 (z2 , . . . , zn ).
64
Add the cutoff dependence to the observables, such that
ĤΛ (z1 , . . . , zn ) → Oexpt. = Oj (z1 , . . . , zn ; Λ) (309)
If ”lucky”, each of the coupling constants can absorb all of the Λ-dependence,
and none of the observables will depend directly on the cutoff
Oexpt. = Oj (z1 , . . . , zn ; Λ) = Oj (z1 (Λ), . . . , zn (Λ)). (310)
If the above is true, then the theory is renormalizable, and is now defined by
a highly overdetermined set of equations, such that there are a finite number
of parameters zi and an infinite list of equations (observables) Oj to solve.
Renormalizable theories effectively have no cutoff, since the remaining infinities
are eliminated by changing the order of limits.
1. Take the limit as the cutoff tends to infinity, Λ → ∞, then compute
observable quantities.
2. Vice versa.
Typical Terms in Scattering Experiments

Scattering experiments generate interaction terms such as the following
Z
λ
0 hp p
1 2 . . . pn | T [ d4 x φ̂4I (x)] |qA qB i0 . (311)
4!
Apply (generalized) Wick’s theorem to calculate the contractions, and sum over
three types of terms (to order λ) encountered. Note that φ̂I (x) = φ̂ in the
following. Wick’s theorem applied to T [φ̂4 (x)] yields a sum over the normal
ordering of all of the following contractions, such that
T [φ̂4 (x)] = fully contracted + partially contracted + uncontracted (312)
Fully contracted : N [φ̂φ̂φ̂φ̂] + other fully contracted
Partially contracted : N [φ̂φ̂φ̂φ̂ + φ̂φ̂φ̂φ̂] + other partially contracted

Uncontracted : N [φ̂φ̂φ̂φ̂].
The diagrammatic contributions to the integrals from each type of contracted

term to the S-matrix have the following forms (for the n = 2 case)
Type 1, fully contracted:
−iλ
Z
d4 x 0 hp1 p2 | φ̂φ̂φ̂φ̂ |qA qB i0 = (313)
4!
65
Figure 24: Diagrammatic contributions of fully contracted terms: a product of
fully connected diagrams, from four copies of the field operator (figure eight),
and delta functions imposing a sort of momentum conservation.
The full contraction is just a C number, making the integral over the inner
product of momentum eigenstates.
Type 2, partially contracted:
−iλ −iλ
Z Z
d4 x 0 hp1 p2 | φ̂φ̂ N [φ̂φ̂] |qA qB i0 = d4 x ( (1) + (2) + (3) ) (314)
4! 4!
The partial contractions have three more types of normal orderings, as labelled
above: (1) + (2) + (3)
1. Right: φ̂φ̂ 0 hp1 p2 | φ̂φ̂ |qA qB i0 + all other contractions to the right.
2. Left: φ̂φ̂ 0 hp1 p2 | φ̂φ̂ |qA qB i0 + all other contractions to the left.
3. Both: φ̂φ̂ 0 hp1 p2 | φ̂φ̂ |qA qB i0 + all other contractions one left and one
right.
66
Figure 25: Diagrammatic contributions of partially contracted terms.
Note that the contraction of the interacting field operators and the momentum
eigenstates, as in the three cases above, contribute incoming/outgoing momenta,
a vertex, and three external legs.
Figure 26: Diagrammatic contributions of contractions between intertacting

field operators and momentum eigenstates.
Type 3, uncontracted:
−iλ −iλ
Z Z
d4 x 0 hp1 p2 | N [φ̂φ̂φ̂φ̂] |qA qB i0 = · 4! · d4 xe−i(qA +qB −p1 −p2 )·x
4! 4!
= −iλ(2π)4 δ (4) (qA + qB − p1 − p2 )
= Diagram below.
67
Figure 27: Uncontracted term contributes diagram that enforces momentum
conservation.
Adding all of the terms together in the full expansion of the interacting component
of the S-matrix, namely 0 hp1 p2 | iT̂ |qA qB i0 , we get a series lke the following
Figure 28: Full diagrammatic expansion of the interacting component fo the

S-matrix.
These are the three types of diagrams, with respect to connectedness, that are
always encountered: fully connected, partially connected, and vacuum bubble
times cully connected.
68
Figure 29: Three types of connectedness encountered in diagrammatic
expansions of the interacting component of the S-matrix.
Vacuum Bubbles and Fully Connected Diagrams

Claim, here without proof, that products of vacuum bubbles and fully connected
diagrams exponentiate, and, further, the only diagrams which contribute to the
S-matrix are the fully connected diagrams. There still exist fully connected
diagrams (integrals) that result in infinities. To quell these infinities, either
impose a cutoff scale, or argue that such a diagram does not contribute to the
S-matrix.
69
Figure 30: An example of a fully connected diagram that results in infinity.
External leg corrections may be used to ”amputate” parts of a diagram that

will remove infinities from the external legs. This process represents a projection
from the free momentum eigenstates to the interacting momentum eigenstates,
just as in the projection of the vacuum state |Ωi. Thus, external leg corrections
factorize the diagram.
|qB i0 → |qB i (315)
Figure 31: Single cuts made from the body of the diagram are called external
leg corrections, and ”amputate” off infinities.
Amputation can not remove all infinities, since it is defined via the external leg
corrections. Self-interactions on internal legs are not removable via amputation.
70
Figure 32: An example of an amputation process where not all infinities can be
removed, since it is not on an external leg.
And the final statement of the Feynman rule for the ϕ4 interaction theory reads
 
sum of all connected and
X  amputated Feynamn diagrams 
iM(2π)4 δ (4) (qA + qB − pf ) =   (316)
with incoming momenta qA , qB
f
and outgoing momenta pf
71

QFT KleinGordon

Uploaded by

Copyright:

Available Formats

QFT KleinGordon

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

QFT KleinGordon

Uploaded by

Copyright:

Available Formats

Notes On Relativistic Quantum Field Theory

A Course Given By Dr. Tobias Osborne

Figure 1: Schematic of the study of field theories in physics.

ηµν dx0µ dx0ν = ηρσ dxρ dxσ (1)

x0µ = Λµν xν + aµ (3)

ηµν Λµρ Λνσ = ηρσ (4)

product (Λ, a) ◦ (Λ̄, ā) = (Λ̄Λ, Λ̄a + ā)

x0µ = Λµν xν + aµ (7)

U (Λ̄, ā)U (Λ, a) = eiφ((Λ,a),(Λ̄,ā)) U (Λ̄Λ, Λ̄a + ā) (10)

The family of transformations U (Λ, a) that satisfy these equations is called

Consider the time translation subgroup of the Poincaré group

{(Λ = 1, a = (t, 0, 0, 0)) : t ∈ R} ⊂ P4 (12)

And let U be a unitary representation of P4 . Then V (t) = U (1, (t, 0, 0, 0)) is a

All single-particle unitary representations of P4 have been classified by Wigner,

• ρ = R; scalar field; charge density, magnetization density, Higgs boson

• ρ = S 2 ; vector field on the surface of a sphere; σ-model, quantum

To extract the equations of motion, we suppose that the action S is stationary

Varying the action functional, we obtain N Euler-Lagrange equations of motion

Example: Klein-Gordon field

Apply the Euler-Lagrange equations of motion to obtain the dynamics of the

The discrete approximation of the Lagrangian is written as

And the discrete conjugate momenta

In the limit as  → 0, the Hamiltonian density is

Now consider an infinitesimal continuous transformation of the fields

φ0a (x) = φa (x) + Xa (φa ) (39)

This produces an infinitesimal symmetry in the equations of motion when they

L → L (φ0a , ∂µ φ0a ) = L (φa , ∂µ φa ) + ∂µ F µ (40)

Proof: Let Xa [φa ] = δφa be an arbitrary, infinitesimal change in each field,

δL (φa , ∂µ φa ) = L (φa + δφa , ∂µ (φa + δφa )) − L (φa , δµ φa ) (42)

Call the conserved quantity the conserved current

φ0a (xµ ) = φa (xµ + µ ) (54)

And the Lagrangian density transforms as, yielding 4 × 4 = 16 equations from

L (x0µ ) = L (xµ + µ ) (56)

Where the infinitesimal field transformation is Xa [φa ] = ν ∂ν φa (xµ ), and the

Example of the Example: Klein-Gordon Field

How To Apply Noether’s Theorem

2. Calculate the change in the Lagrangian density.

3. Calculate the change in each of the fields.

4. Work out the conserved currents and charges.

φa (x) → φ0a (x) = φa (Λ−1 x) (71)

Showing that the symmetry is defined by the infinitesimals

Standard Dogma of Quantization

• Function f on phase space → linear operator fˆ (observable) on Hilbert

• Poisson bracket {f, g} = h → commutator [fˆ, ĝ] = iĥ

• Conserved charge Qρσ → conserved charge operator Q̂ρσ

Where the conserved charge operators generate the Lorentz transformations on

Canonical quantization of particles

Where Q is an n × n symmetric, positive matrix.

• Canonical coordinates: (qj , pj )

• Poisson bracket: {qj , pk } = δjk

• Hamiltonian: H, as defined above

To diagonalize the Hamiltonian operator, first note that since Q is a symmetric,

Canonical Quantization of Fields

• Canonical coordinates: (φ(x), π(x))

• Poisson bracket: {φ(x), π(y)} = δ (3) (x − y)

Essentially, replace discrete sums with continuous integrals, by switching to a

Replacing sums by integrals allows the full diagonalization of Q, and, therefore,

Diagonalization of the quantum field theory

Where K(x, y) is the kernel of the Fourier transform.

In the limit as → 0, the Hamiltonian density is

φ0a (xµ ) = φa (xµ + µ ) (54)

L (x0µ ) = L (xµ + µ ) (56)

Where the infinitesimal field transformation is Xa [φa ] = ν ∂ν φa (xµ ), and the

→ TI G = {ω : ηµν (I + ω)µρ (I + ω)νσ = ηρσ }

[Jˆj , Jˆk ] = −ijkl Jˆl (196)