Skript QFT2
Skript QFT2
Skript QFT2
Timo Weigand
Literature
This is a writeup of my Master programme course on Quantum Field Theory I (Chapters 1-6) and
Quantum Field Theory II. The primary source for this course has been
which I urgently recommend for more details and for the many topics which time constraints have
forced me to abbreviate or even to omit. Among the many other excellent textbooks on Quantum
Field Theory I particularly recommend
as further reading. All three of them oftentimes take an approach different to the one of this course.
Excellent lecture notes available online include
Special thanks to Robert Reischke1 for his fantastic work in typing these notes.
5
6 CONTENTS
or, even better, by modifying the Schrdinger equation altogether such as to make it symmetric in
~
t and the spatial derivative . However, the central insight underlying the formulation of Quantum
Field Theory is that this is not sufficient. Rather, combining the principles of Lorentz invariance and
Quantum Theory requires abandoning the single-particle approach of Quantum Mechanics.
In any relativistic Quantum Theory, particle number need not be conserved, since the relativistic
dispersion relation E 2 = c2 ~p2 + m2 c4 implies that energy can be converted into particles and
vice versa. This requires a multi-particle framework.
Quantum Field Theory (QFT) solves both these problems by a radical change of perspective:
9
10 CHAPTER 1. THE FREE SCALAR FIELD
The fundamental entities are not the particles, but the field, an abstract object that penetrates
spacetime.
Before developing the notion of an abstract field let us try to gain some intuition in terms of a mechan-
ical model of a field. To this end we consider a mechanical string of length L and tension T along the
x-axis and excite this string in the transverse direction. Let ( x, t ) denote the transverse excitation of
the string. In this simple picture ( x, t ) is our model for the field. This system arises as the continuum
limit of N mass points of mass m coupled by a mechanical spring to each other. Let the distance of
the mass points from each other projected to the x-axis be and introduce the transverse coordinates
qr (t ), r = 1, . . . , N of the mass points. In the limit 0 with L fixed, the profile qr (t ) asymptotes
to the field ( x, t ). In this sense the field variable x is the continuous label for infinitely many degrees
of freedom.
We can now linearise the force between the mass points due to the spring. As a result of a simple
exercise in classical mechanics the energy at leading order is found to be
N !2
X 1 dqr (t ) T
E= m + k(q2r qr qr1 ) + O(q3 ), k= . (1.4)
r =0
2 dt L
in terms of the mass density of the string and a suitably defined characteristic velocity c. Note that
the second term indeed includes the nearest neighbour interaction because
!2 !2
( x, t ) ( x + x, t ) ( x, t ))
' lim (1.6)
x x0 x
where k = kc/L. We are now dealing with a collection of infinitely many, decoupled harmonic
oscillators Ak (t ).
1.2. CLASSICAL SCALAR FIELD: LAGRANGIAN FORMULATION 11
In a final step, we quantise this collection of harmonic oscillators. According to Quantum Mechanics,
each mode Ak (t ) can take energy values
The idea of Quantum Field Theory is to adapt this logic to particle physics and to describe a particle
as the quantum of oscillation of an abstract field - just like in solid state physics we think of a
quasi-particle as the vibrational excitation of a solid. The only difference is that the fields are now
more abstract objects defined all over spacetime as opposed to concrete mechanical fields of the type
above.
As a familiar example for a field we can think of the Maxwell field A ( x, t ) in classical electrody-
namics. A photon is the quantum excitation of this. It has spin 1. Similarly we assign one field to
each particle species, e.g. an electron is the elementary excitation of the electron field (Spin 1/2). We
will interpret the sum over harmonic oscillator energies as an integral over possible energies for given
momentum,
X Z
E= hk (nk + 1/2) dp h p (n p + 1/2). (1.9)
k =1
A single particle with momentum p corresponds to n p = 1 while all others vanish, but this is just a
special example of a more multi-particle state with several n pi , 0. In particular, in agreement with
the requirements of a multi-particle framework, at fixed E transitions between various multi-particle
states are in principle possible. Such transitions are induced by interactions corresponding to the
higher order terms in the Hamiltonian that we have discarded so far. As a triumph this formalism also
solves the problem of causality, as we will see.
where we have included the mass m in the definition of qi (t ). In a first step replace
qi ( x ) ( x), (1.11)
( x)
qi (t ) , (1.12)
t
thereby substituting the label i = 1, ...N by a continous coordinate ~x xi with i = 1, 2, 3. For the
moment we consider a real scalar field i.e. ( x) = ( x) which takes values in R, i.e.
: x ( x ) R. (1.13)
We will see that such a field describes spin-zero particles. Examples of scalar particles in nature are
the Higgs boson or the inflaton, which cosmologists believe to be responsible for the exponential ex-
pansion of the universe during in inflation.
To set up the Lagrange function we first note that in a relativistic theory the partial time derivative can
only appear as part of
( x ) ( x ) . (1.14)
x
Thus the Lagrange function can be written as
Z
L= d3 x L(( x), ( x)), (1.15)
While, especially in condensed matter physics, also non-relativistic field theories are relevant, we fo-
cus on relativistic theories in this course.
h = c = 1. (1.17)
Then L has the dimension mass4 , i.e. [L] = 4, since [S ] = 0 and [d4 x] = 4.
The next goal is to find the Lagrangian: In a relativistic setting L can contain powers of and1
, which is the simplest scalar which can be built from . The action in this
case is Z " #
1
S = d x V () + O( () ) ,
4 n m
(1.18)
2
where
1 1 1
= 2 ()2 (1.19)
2 2 2
1 Note that the only remaining option is a total derivative and will therefore not alter the equations of motion under
the usual assumptions on the boundary terms.
1.2. CLASSICAL SCALAR FIELD: LAGRANGIAN FORMULATION 13
and the last type of terms consists of higher derivative terms with m 2 or mixed terms with n 1.
Notice that
the signature for the metric is, in our conventions, (+, , , ), such that the sign in the action
is indeed chosen correctly such that the kinetic term appears with a positive prefactor;
V (( x)) = a + b( x) + c2 ( x) + d3 ( x) + . . . . (1.20)
V ()
|=0 = 0, V (0 ) = V0 (1.21)
By a field redefinition we ensure that the minimum is at 0 ( x) 0 and expand V (( x)) around this
minimum as
1
V (( x)) = V0 + m2 2 ( x) + O(3 ( x)). (1.22)
2
Here we used that the linear terms vanish at the extremum and the assumption that we are expanding
around a minimum implies m2 > 0. The constant V0 is the classical contribution to the ground state or
vacuum energy. Since in a theory without gravity absolute energies are not measurable, we set V0 = 0
for the time being, but keep in mind that in principle V0 is arbitrary. We will have considerably more
to say about V0 in the quantum theory in section (1.8).
Therefore the action becomes
Z " #
1 1 2 2
S = d x () m + ... .
4 2
(1.23)
2 2
We will find that m2 , the prefactor of the quadratic term, is related to the mass of the particles and that
the omitted higher powers of as well as the terms O(n ()m ) will give rise to interactions between
these particles.
As an aside note that a negative value of m2 signals that the extremum around which we are expanding
the potential is a maximum rather than a minimum. Therefore m2 < 0 signals a tachyonic instability:
quantum fluctuations will destabilise the vacuum and cause the system to roll down its potential until
it has settled in its true vacuum.
We will start by ignoring interaction terms and studying the action of the free real scalar field theory
Z " #
1 1 2 2
S = d x () m , = .
4 2
(1.24)
2 2
The equations of motion are given by the Euler-Lagrange equations. As in classical mechanics we
derive them by varying S with respect to and subject to |boundary = |boundary = 0. This
14 CHAPTER 1. THE FREE SCALAR FIELD
yields
L(( x), ( x)) L(( x), ( x))
Z " #
!
0 = S = d x 4
( x) + ( x)
( x) ( ( x))
(1.25)
L(( x), ( x)) L(( x), ( x))
Z " #
4
= d x ( x) + ( x) .
( x) ( ( x))
Integrating by parts gives
L(( x), ( x)) L(( x), ( x))
Z " #
4
d x ( x) + boundary terms. (1.26)
( x) ( ( x))
Since the boundary terms vanish by assumption, therefore the integrand has to vanish for all variations
( x). This yields the Euler-Lagrange equations
By inserting (1.24) into (1.27) we find the equations of motion for the free scalar field
L L
= = = m2 , (1.28)
( )
i.e. the Klein-Gordon equation
(2 + m2 )( x) = 0. (1.29)
Note that (1.29) is a relativistic wave equation and that it is solved by
q
E ~p := ~p2 + m2 E p (1.31)
Every continuous symmetry in the above sense gives rise to a Noether current j ( x) such
that
j ( x) = 0 (1.33)
L
j = X F (1.37)
( )
we therefore have
L L
!
j = X (1.38)
( )
off-shell. Note that the terms in brackets are just the Euler-Lagrange equation. Thus, if
we use the equations of motion, i.e. on-shell, j = 0.
Every continuous symmetry whose associated Noether current satisfies ji (t, ~x) 0 suffi-
ciently fast for |~x| gives rise to a conserved charge Q with
Q = 0. (1.39)
Indeed if we take Z
Q= d3 x j0 (t, ~x), (1.40)
R3
16 CHAPTER 1. THE FREE SCALAR FIELD
by assumption of sufficiently fast fall-off of ji (t, ~x). We used that j = 0 in the first
step.
The technical assumption ji (t, ~x) 0 for |~x| is really an assumption of sufficiently fast fall-off
of the fields at spatial infinity, which is typically satisfied. Note that in a finite volume V = const., the
quantity QV = dV j0 (t, ~x) satisfies local charge conservation,
R
V
Z Z
QV = dV ~j = ~j d~s. (1.42)
V V
We now apply Noethers theorem to deduce the canonical energy-momentum tensor: Under a global
spacetime transformation x x + a scalar field ( x ) transforms like
( x ) ( x ) = ( x ) ( x ) + O ( 2 ) . (1.43)
| {z }
X ()
L
T = L with T = 0 on-shell. (1.46)
( )
R
The conserved charges are the energy E = d3 x T 00 associated with time translation invariance and
R
the spatial momentum Pi = d3 x T 0i associated with spatial translation invariance. We can combine
them into the conserved 4-momentum
Z
P = d3 x T 0 (1.47)
In general, T may not be symmetric - especially in theories with spin. In such cases it can
be useful to modify the energy-momentum tensor without affecting its conservedness or the
associated conserved charges. Indeed we state as a fact that the Belinfante-Rosenfeld tensor
BR := T + S (1.48)
can be defined in terms of a suitable S = S such that BR is symmetric and obeys
BR = 0.
In General Relativity (GR), there exists yet another definition of the energy-momentum tensor:
With the metric replaced by g and
Z
S = d4 x gLmatter (g , , ), (1.49)
which is obviously symmetric and it the object that appears in the Einstein equations
1
R + Rg = 8G ( H ) . (1.51)
2
In fact one can choose the Belinfante-Rosenfeld tensor such that it is equal to the Hilbert energy-
momentum tensor.
L
pi ( t ) = . (1.52)
qi (t )
To quantise in the Schrdinger picture we drop the time dependence of qi and pi and promote them to
self-adjoint operators without any time dependence such that the fundamental commutation relation
[qi , p j ] = ii j (1.54)
18 CHAPTER 1. THE FREE SCALAR FIELD
L
(t, ~x) := (1.55)
(t, ~x)
where H is the Hamiltonian density. For the scalar field action (1.24) one finds
and therefore
Z " #
1 1
H= d3 x 2 (t, ~x) ( )( ) + m2 2 (t, ~x)
2 2
Z " #
1 2 1 1 2 2 (1.58)
= d x (t, ~x) + ((t, ~x)) + m (t, ~x) .
3 2
2 {z } 2
| 2
= 12 2 (t,~x)
Note that as in classical mechanics one can define a Poisson bracket which induces a natural sym-
pletic structure on phase space. In this formalism the Noether charges Q are the generators of their
underlying symmetries with the respect to the Poisson bracket (see Assignment 1 for details).
We now quantise in the Schrdinger picture. Therefore we drop the time-dependence of and
and promote them to Schrdinger-Picture operators ( s) (~x) and ( s) (~x). For real scalar fields we get
self-adjoint operators ( s) (~x) = (( s) (~x)) with the canonical commutation relations (dropping ( s)
from now on)
[(~x), (~y)] = i(3) (~x ~y) , [(~x), (~y)] = 0 = [(~x), (~y)]. (1.59)
couples the degrees of freedom at ~x and ~x + ~x. To arrive at a description in which the harmonic
oscillators are decoupled, we must diagonalise the potential. Now, a basis of eigenfunctions with
1.5. MODE EXPANSION 19
respect to is ei~p~x . Thus the interaction will be diagonal in momentum space. With this motivation
we Fourier-transform the fields as
d3 p
Z
(~x) = ( ~p)ei~p~x ,
(2)3
(1.61)
d3 p
Z
(~x) = ( ~p)e ,i~p~x
(2)3
where ( ~p) = (~p) ensures that (~x) is self-adjoint. To compute H in Fourier space we must
insert these expressions into (1.58). First note that
!2
d3 p
Z Z Z
1 1
d x((~x)) =
3 2 3
d x 3
ei~p~x ( ~p)
2 2 (2)
(1.62)
d3 pd3 q
Z Z
1
= 3
d x (~p ~q)ei( ~p+~q)~x ( ~p)(~q).
2 (2)6
Thanks to the important equality
Z
d3 x ei( ~p+~q)~x = (2)3 (3) ( ~p + ~q), (1.63)
d3 p 2
Z Z
1 1
d3 x((~x))2 = ~p ( ~p)(~p), (1.64)
2 2 (2)3 | {z }
|( ~p)|2
d3 p 1
Z " #
1 2
H= |( ~p)| + p |( ~p)| ,
2 2
(1.65)
(2)3 2 2
The next step is to solve these oscillators in close analogy with the quantum mechanical treatment of
a harmonic oscillator of frequency with Hamiltonian
1 2
H= + 2 q2 . (1.66)
2
The quantum mechanical definition of ladder operators a, a such that
r
1
q = (a + a ) , =
i(a a ) (1.67)
2 2
and with commutation relation [a, a ] = 1 can be generalised to field theory as follows: By taking
into account that
( ~p) = (~p) , ( ~p) = (~p), (1.68)
20 CHAPTER 1. THE FREE SCALAR FIELD
Solving for ( ~p) and ( ~p) and plugging into (1.61) yields
d3 p
Z
1
(~x) = a ( ~
p ) ei~p~x
+ a
( ~
p ) e i~p~x
,
(2)3 2 p
p
(1.70)
p
r
d3 p
Z
(~x) = 3
(i) a( ~p)ei~p~x a ( ~p)ei~p~x .
(2) 2
d3 p p
Z h i
H= a( ~p)a ( ~p) + a (~p)a(~p) 2
(2)3 4
d3 p p
Z h i
= a( ~p)a ( ~p) + a (~p)a(~p) (1.75)
(2)3 2
d3 p p
Z h i
= a ( ~p)a( ~p) + (2)3 (3) ( ~p ~p) + a (~p)a(~p) .
(2)3 2
1.6. THE FOCK SPACE 21
d3 p
Z
H= p a ( ~p)a( ~p) + H , (1.76)
(2)3
[ H, a( ~p)] = p a( ~p),
(1.78)
[ H, a ( ~p)] = p a ( ~p).
to be
d3 p i
Z
i
P = p a ( ~p)a( ~p) + pi , (1.80)
(2)3
with Z
1
pi = d3 p pi (3) (0) 0. (1.81)
2
We combine H and P into the 4-momentum operator
d3 p
Z
P = p a ( ~p)a( ~p) + p , (1.82)
(2)3
[ P , a ( ~p)] = p a ( ~p),
(1.83)
[ P , a( ~p)] = p a( ~p).
Since P is self-adjoint it has eigenstates with real eigenvalues. Let |k i be such an eigenstate
with
P |k i = k |k i . (1.84)
22 CHAPTER 1. THE FREE SCALAR FIELD
and similarly
This means that a(~q) and a (~q) are indeed ladder operators which respectively subtract and add
4-momentum q to or from |k i.
Next we observe that the Hamiltonian H = P0 given by (1.76) is non-negative, i.e. h|H|i
0 states |i because
d3 p
Z
h|H|i = 3
p h|a ( ~p)a( ~p)|i + H h|i 0. (1.87)
(2)
Otherwise successive action of a(~q) would lead to negative eigenvalues of H. |0i is called the
vacuum of the theory. It has 4-momentum
H , = 0
P |0i = p |0i =
. (1.89)
0 , =i
We interpret the divergent constant H given in (1.77) as the vacuum energy. A more thorough
discussion of the significance of the divergence will be given later. For now we should note that
in a theory without gravity absolute energy has no meaning. We can hus discard the additive
constant p by defining
d3 p
Z
P := P p = p a ( ~p)a( ~p), (1.90)
(2)3
with P |0i = 0. From now on we only work with P and drop the tilde.
with p = ( E p , ~p) and E p = ~p2 + m2 . Since this is the relativistic dispersion relation for
p
a single particle with mass m we interpret a ( ~p) |0i as a 1-particle state with energy E p and
momentum ~p.
1.7. SOME IMPORTANT TECHNICALITIES 23
So much about the formalism. To get a better feeling for the objects we have introduced, let us recap
what we have done: We have started with the assertion that spacetime - in our case R1,3 - is filled with
the real scalar field (~x), which we have taken to be a free field with Lagrangian L = 21 ()2 12 m2 2 .
This field is interpreted as a field operator, i.e. in the Schrdinger picture at every space point ~x the
object (~x) represents a self-adjoint operator that acts on a Hilbert space. This Hilbert space possesses
a state of lowest energy, the vacuum |0i. The vacuum corresponds to the absence of any excitations of
the field (at least on-shell). If one pumps energy E p and momentum ~p into some region of spacetime
such that the relativistic dispersion relation E p = ~p2 + m2 holds, a particle a ( ~p) |0i is created as
p
an excitation of (~x). In particular, for the free theory the parameter m in L is interpreted as the mass
of such a particle. Since the underlying field ( x) is a scalar field, the associated particle is called
scalar particle. This realizes the shift of paradigm advertised at the very beginning of this course that
the fundamental entity in Quantum Field Theory is not the particle, but rather the field:
The field (~x) is the property of spacetime that in the presence of energy and momentum
( E p , ~p) a particle of energy ( E p , ~p) can be created.
In particular, this naturally gives rise to a multi-particle theory. The particles are just the excitations
of the field and transitions with varying particle number can occur as long as the kinematics allows it.
A first new, non-trivial conclusion we can draw from the formalism developed so far is the following
special case of the spin-statistics theorem:
Scalar particles obey Bose statistics.
All we have to show that the N-particle wavefunction is symmetric under permutations. This follows
immediately from the commutation relations, specifically the second line in (1.73):
|~p1 , ..., ~pi , ..., ~p j , ..., ~pn i ' a ( ~p1 )...a ( ~pi )...a ( ~p j )...a ( ~pN ) |0i
= a ( ~p1 )...a ( ~p j )...a ( ~pi )...a ( ~pN ) |0i (1.93)
' |~p1 , ..., ~p j , ..., ~pi , ..., ~pn i,
where we have not fixed the normalization of the Fock state yet.
Then q q
h~q|~pi = 2E p 2Eq h0| a(~q)a ( ~p) |0i . (1.95)
To compute the inner product of two such states we use an important trick: Move all as to the right
and all a to the left with the help of
This gives rise to terms of the form a(~q) |0i = 0 = h0| a ( ~p). Therefore
Note that, as in Quantum Mechanics, momentum eigenstates are not strictly normalisable due to the
appearance of the delta-distribution, but we can form normalisable states as wavepackets
Z
|fi = d3 p f ( ~p) |~pi . (1.98)
d3 p 1
Z
11particle = |~pi h~p| . (1.99)
(2)3 2E p
R 3
One should notice that (d2p)3 2E1 p is a Lorentz-invariant measure. This, in turn, is part of the motiva-
tion for the normalisation of the 1-particle momentum eigentstates. To see this we rewrite the measure
as
d3 p 1 d3 p 2
Z Z
=
(2)3 2E p (2)4 2E p
(1.100)
d4 p
Z
= 2( p m )( p ),
2 2 0
(2)4
where we used that (ax) = 1a ( x) and that
( p2 m2 ) = ( p0 E p )( p0 + E p ) (1.101)
in the last step. (1.100) is manifestly Lorentz-invariant: First, d4 p det()d4 p under a Lorentz-
transformation, and since the determinat of a Lorentz-transformation is 1, it follows that d4 p is
Lorentz-invariant. Moreover the sign of p0 is unchanged under a Lorentz transformation.
d3 p 1 i~p~x
Z
|~xi = e |~pi (1.103)
(2)3 2E p
because then
d3 q 1 i~q~x
Z
h~x|~pi = 3
e h~q|~pi = ei~p~x , (1.104)
(2) 2E q
where we used (1.97). Note that
d3 p
Z
1 i~p~x
|~xi = e a ( ~
p ) + ei~p~x
a ( ~
p ) |0i = (~x) |0i . (1.105)
(2)3 2E p
p
In other words, the field operator ( x) acting on the vacuum |0i creates a 1-particle position eigenstate.
The divergent factor (2)3 (3) (0) is interpreted as follows: We know that
Z
d3 x ei~p~x = (2)3 (3) ( ~p), (1.107)
R3
The divergence of (3) (0) is rooted in the fact that the volume of R3 is infinite, and the cor-
responding divergent factor in E0 arises because we are computing an energy in an infinite
volume. This divergent factor thus results from the long-distance (i.e. small energy) behaviour
of the theory and is an example of an infra-red (IR) divergence. Generally in QFT, IR diver-
gences signal that we are either making a mistake or ask an unphysical question. In our case,
the mistake is to consider the theory in a strictly infinite volume, which is of course unphysical.
One can regularise the IR divergence by instead considering the theory in a given, but finite
volume. What is free of the IR divergence is in particular the vacuum energy density
d3 p
Z
E0 1
0 = = p. (1.109)
V R3 2 (2)3
26 CHAPTER 1. THE FREE SCALAR FIELD
which goes to infinity due to the integration over all momenta up to p . This is an ultra-
violet (UV) divergence. The underlying reason for this (and all other UV divergences in QFT)
is the breakdown of the theory at high energies (equivalently at short distances) - or at least a
breakdown of our treatment of the theory.
To understand this last point it is beneficial to revisit the mechanical model of the field (t, x) as an
excitation of a mechanical string as introduced in section 1.1. Recall that the field (t, x) describes the
transverse position of the string in the continuum limit of vanishing distance between the individual
mass points at position qi (t ) which were thought of as connected by an elastic spring. However,
in reality the string is made of atoms of finite, typical size R. A continuous string profile (t, x) is
therefore not an adequate description at distances R or equivalently at energies E ' 1/R
resolving such small distances. Rather, if we want to describe processes at energies E , the
continuous field theory (t, x) is to be replaced by the more fundamental, microscopic theory of
atoms in a lattice. In this sense the field (t, x) gives merely an effective description of the string valid
at energies E . Extrapolation of the theory beyond such energies is doubtful and can give rise to
infinities - the UV divergences.
In a modern approach to QFT, this reasoning is believed to hold also for the more abstract relativistic
fields we are considering in this course. According to this logic, QFT is really an effective theory
that eventually must be replaced at high energies by a more fundamental theory. A necessary
condition for such a fundamental theory to describe the microscopic degrees of freedom correctly
is that it must be free of pathologies of all sort and in particular be UV finite.3 At the very least,
gravitational degrees of freedom become important in the UV region and are expected to change the
qualitative behaviour of the theory at energies around the Planck scale MP ' 1019GeV.
Despite these limitations, in a good QFT the UV divergences can be removed - for all practical
purposes - by the powerful machinery of regularisation and renormalisation. We will study this
procedure in great detail later in the course, but let us take this opportunity to very briefly sketch the
logic for the example of the vacuum energy density:
The first key observation is that in the classical Lagrangian we can have a constant term V0 of
dimension mass4 ,
1 1
L = ()2 m2 2 V0 , V0 = V|min , (1.111)
2 2
corresponding to the value of the classical potential at the minimum around which we expand.
We had set V0 0 in our analysis because we had argued that such an overall energy offset
is not measurable in a theory without gravity. However, as we have seen the vacuum energy
3 To date, the only known fundamental theory that meets this requirement including gravity is string theory.
1.8. ON THE VACUUM ENERGY 27
density really consists of two pieces - the classical offset V0 and the quantum piece H . Thus,
let us keep V0 for the moment and derive the Hamiltonian
Z ! Z
1 2 1 2 2 1
H= d x + m + () + d3 x V0 .
3 2
(1.112)
2 2 2
Next, we regularise the theory: Introduce a cutoff-scale and for the time being only allow for
energies E . If we quantise the theory with such a cutoff at play, the overall vacuum energy
is now
H |0i = VR3 (0 () + V0 ) |0i (1.113)
with
Z
1
q
0 () = 2
dp p p2 + m 2 . (1.114)
(2)2
0
Note that the momentum integral only runs up to the cutoff in the regularised theory.
V0 = V0 () = 0 () + , finite as . (1.115)
To summarize, we define the quantum theory as the result of quantising the classical Lagrangian
L = 12 ()2 12 m2 2 V0 () with V0 () = 0 () + and taking the limit at the very
end. Since appears only in the classical - or bare - Lagrangian, but in no observable, physical
quantity at any of the intermediate stages, we can safely remove it at the end. In this sense, the theory
is practically defined up to all energies.
This way to deal with UV divergences comes at a prize: We lose the prediction of one observable per
type of UV divergence as a result of the inherent arbitrariness of the renormalisation step. In fact,
above we have chosen V0 () such that H |0i = VR3 |0i. It is only in the absence of gravity that
the vacuum energy is unobservable and thus is irrelevant. More generally, the value of the vacuum
energy must now be taken from experiment - is an input parameter of the theory rather than a predic-
tion. While we have succeeded in removing the divergence by renormalising the original Lagrangian,
28 CHAPTER 1. THE FREE SCALAR FIELD
the actual value of the physical observable associated with the divergence - here the vacuum energy
density - must be taken as an input parameter from experiment or from other considerations. In a
renormalisable QFT it is sufficient to do this for a finite number of terms in the Lagrangian so that
once the associated observables are specified, predictive power is maintained for the computation of
all subsequent observables.
It is then a simple matter to repeat the programme of quantisation, e.g. by quantizing 1 and 2 as
before and rewriting everything in terms of the complex field ( x). At the end of this rewriting we can
forget about 1 and 2 and simply describe the theory in terms of the complex field ( x). The details
of this exercise will be provided in the tutorials so that we can be brief here and merely summarise the
main formulae.
The fields ( x) and ( x) describe independent degrees of freedom with respective conjugate
momenta
L
(t, ~x) = = (t, ~x),
(t, ~x)
(1.120)
L
(t, ~x) =
= (t, ~x).
(t, ~x)
The Hamiltonian H is
Z
H= d3 x (~x) (~x) + (~x)(~x) L
Z (1.121)
= d x (~x)(~x) + (~x)(~x) + m (~x)(~x) .
3 2
These fields are promoted to Schrdinger picture operators with non-vanishing commutators
h i
[(~x), (~y)] = i(3) (~x ~y) = (~x), (~y) (1.122)
d3 p
Z
1
(~x) = a ( ~
p ) ei~p~x
+ b
( ~
p ) e i~p~x
(2)3 2E p
p
(1.123)
d3 p
Z
1
(~x) = 3
p b( ~p)ei~p~x + a ( ~p)ei~p~x ,
(2) 2E p
where the mode operators a( ~p) and b( ~p) are independent and a ( ~p) and b ( ~p) describe the
respective conjugate operators. A quick way to arrive at this form of the expansion is to plug
the mode expansion of the real fields 1 and 2 into (1.118). This identifies
1
a = (a1 + ia2 ),
2
(1.124)
1
b = (a1 + ia2 ).
2
In particular this implies that
[a( ~p), a (~q)] = (2)3 (3) ( ~p ~q) = [b( ~p), b (~q)], (1.125)
d3 p
Z
P =
p a ( ~
p ) a ( ~
p ) + b
( ~
p ) b ( ~
p ) . (1.126)
(2)3
both of energy E p = ~p2 + m2 . I.e. both states have mass m, but they differ in their U (1)
p
The Lagrange density is invariant under the global continuous U (1) symmetry
( x) ei ( x), (1.128)
where is a constant in R. Recall that the unitary group U ( N ) is the group of complex N N
matrices A satisfying A = A1 . The dimension of this group is N 2 . In particular ei U (1).
According to Noethers theorem there exists a conserved current (see Ass. 3)
j = i( ) (1.129)
and charge
d3 p
Z Z
Q= 3
d xj = 0
3
a ( ~p)a( ~p) b ( ~p)b( ~p). (1.130)
(2)
We interpret a ( ~p) |0i as a particle of mass m and charge 1 and b ( ~p) |0i as a particle with
the same mass, but positive charge, i.e. as its anti-particle. For the real field the particle is
its own anti-particle. Note that the term charge 1 so far refers simply to the eigenvalue of
the Noether charge operator Q associated with the global U (1) symmetry of the theory. That
this abstract charge really coincides with what we usually call charge in physics - i.e. that it
describes the coupling to a Maxwell type field - will be confirmed later when we study Quantum
Electrodynamics.
that alternatively quantum operators can be described in Heisenberg picture (HP), where the time-
dependence is carried by the operators A(H ) (t ) and not the states. The HP operator A(H ) (t ) is defined
as
(S ) (S )
A(H ) (t ) = eiH (tt0 ) A(S ) eiH (tt0 ) , (1.132)
where A(S ) is the corresponding Schrdinger picture operator and H (S ) is the Schrdinger picture
Hamilton operator. At the time t0 the Heisenberg operator and the Schrdinger operator coincide,
i.e. A(H ) (t0 ) = A(S ) . We will set t0 0 from now on. The definition (1.132) has the following
implications:
H (H ) (t ) = H (S ) t.
The time evolution of the Heisenberg picture operators is governed by the equation of motion
d (H )
A (t ) = i[ H, A(H ) (t )]. (1.133)
dt
The Schrdinger picture commutators translate into equal time commutators, e.g.
(H ) (H )
[ qi (t), p j (t)] = i i j . (1.134)
where used the standard relation [ A, BC ] = [ A, B]C + B[ A, C ] together with (1.136). This gives
(t, ~x) = (t, ~x) (1.140)
t
as expected. One can similarly show that
(t, ~x) = i[ H, (t, ~y)]
t (1.141)
= 2 (t, ~x) m2 (t, ~x).
( 2 + m2 ) ( x ) = 0 (1.142)
with Pi = d y (y) i (y). Indeed this equation can be explicitly confirmed by evaluating the
R 3
This, in fact, is simply the transformation property of the quantum field ( x) under translation.
Let us now compute the mode expansion for the Heisenberg field. From the mode expansion for the
Schrdinger picture operator we find
d3 p
Z
1 iHt
(t, ~x) =
(H )
e a ( ~
p ) eiHt i~p~x
e + e iHt
a ( ~
p ) e iHt i~p~x
e . (1.145)
(2)3 2E p
p
To simplify this we would like to commute eiHt through a( ~p) and a ( ~p). Since [ H, a( ~p)] = a( ~p) E p ,
we can infer that
Ha( ~p) = a( ~p)( H E p ) (1.146)
Thus
eiHt a( ~p) = a( ~p)ei(HE p )t , (1.148)
d3 p
Z
1
(H )
(t, ~x) ( x) = a ( ~
p ) e ipx
+ a
( ~
p ) e ipx
, (1.149)
(2)3 2E p
p
with
In other words, the coefficient of eipx in the mode expansion of the Heisenberg field corresponds to
the annihilator and the coefficient of eipx to the creator. Note that indeed this mode expansion solves
the operator equation of motion (1.142).
For later purposes we also give the inverted expression
Z
i
a(~q) = p d3 x eiqx 0 ( x),
2Eq
Z
(1.151)
i
a (~q) = p
3
d xe iqx
0 ( x ) ,
2Eq
where
u( x) 0 v( x) := u( x)0 v( x) (0 u( x))v( x),
1.11.1 Commutators
For causality to hold we need two measurements at spacelike distance not to affect each other. This is
guaranteed if any two local observables O1 ( x) and O2 (y) at spacelike separation commute, i.e.
!
[O1 ( x), O2 (y)] = 0 for ( x y)2 < 0. (1.152)
By a local observable we mean an observable in the sense of quantum mechanics (i.e. a hermitian
operator) that is defined locally at a spacetime point x, i.e. it depends only on x and at most on a
local neighborhood of x. Any such local observable is represented by a local (hermitian) operator, by
which we mean any local (hermitian) expression of the fundamental operators ( x) and ( x) such
as products or powers series (e.g. exponentials) of the operators. Indeed such operators are local in
the sense that they depend only on a local neighborhood of the spacetime point x.
To check for (1.152) we thus need to compute the commutator
not just at equal times x0 = y0 , but for general times. In Fourier modes we have
d3 p d3 q
Z Z
1 1
( x y) = 3
p 3
p
(2) 2E p (2) 2Eq (1.154)
h i
a( ~p)eipx + a ( ~p)eipx , a(~q)eiqy + a (~q)eiqy .
34 CHAPTER 1. THE FREE SCALAR FIELD
d3 p 1 ip( xy)
Z
( x y) = e e ip(yx)
. (1.155)
(2)3 2E p
R 3
Now assume ( x y)2 < 0. Since (d2p)3 2E1 p is Lorentz invariant (see the discussion in section 1.7.2)
we can apply a Lorentz transformation such that ( x0 y0 ) = 0. Indeed this can be always be achieved
if two points are at spacelike distance. This gives
d3 p 1 i~p(~x~y)
Z
ei~p(~x~y) ,
( x y) = 3
e (1.156)
( xy)2 <0 (2) 2E p
which is evidently zero because we can change the integration variable from ~p to ~p in the second
term. Consequently, also commutators of the form [ x ( x), (y)] = x [( x), (y)] vanish for x and
y at spacelike distance. Therefore for all local operators we have
Remark:
In the above we have been able to derive (1.157) because we are working with a free theory, for which
the free mode expansion implies [( x), (y)] = 0 for ( x y)2 < 0. In an interacting theory, such a
derivation may not be possible since in this case a free mode expansion is no longer available. More
generally, therefore, one must postulate [( x), (y)] = 0 for ( x y)2 < 0 in form of an axiom of QFT.
As we have seen, locality in Quantum Field Theory refers to the fact (or requirement) that local
operators commute at spacelike distances. By contrast, the states on which these operators act, i.e.
the elements of the Hilbert space, do exhibit non-local behaviour just as in non-relativistic Quantum
Mechanics. Indeed the fact that in QFT local operators commute is not at odds with the existence of
entangled states, known from Quantum Mechanics. Consider e.g. an entangled 2-spin state | 21 , 12 i
| 12 , 12 i. Even though the state is entangled, the spin operator S 1 for measurement of the spin of particle
1 at x and S 2 for measurement of the spin of particle 2 at y commute. In particular the expectation
value of S 2 is not changed by measuring S 1 . Therefore by measuring the expectation value of S 2 we
cannot determine if S 1 was measured and vice versa and the two measurements do not affect each
other in this sense. More generally, the states of the Quantum Field Theory are non-local objects. We
will understand this better when discussing the Schrdinger Representation of states in the context of
path integral quantization in QFT 2. The states are represented as functionals of the field configuration
and are necessarily dependent on non-local information.
1.11. CAUSALITY AND PROPAGATORS 35
1.11.2 Propagators
Consider now the probability amplitude for a particle emitted at y to propagate to x (where it can be
measured). This is given by the propagator
because |xi = ( x) |0i is a position eigenstate. In particular, unlike in the non-relativistic expression
(1.3), the time-evolution operator need not be inserted by hand because |xi contains information about
the time variable x0 . By a mode expansion we find
"
d3 p 1 d3 q 1
D( x y) = h0| a (~q ) e iqx
+ a
(~q ) e iqx
(2)3 2E p (2)3 2Eq
p p
(1.159)
a( ~p)eipy + a ( ~p)eipy |0i .
Using a( ~p) |0i = 0 = h0| a (~q) we find that the only contribution is due to the term
eiqx eipx h0| a(~q)a ( ~p) |0i = eiqx eipx (2)3 (3) (~q ~p). (1.160)
This yields
d3 p 1 ip( xy)
Z
D( x y) = h0| ( x)(y) |0i = e . (1.161)
(2)3 2E p
We see that D( x y) is non-zero even for x and y at spacelike distance. On the other hand, in view
of (1.156) this cannot affect causality. So how can this be consistent? What saves the day is the
observation that
[( x), (y)] = D( x y) D(y x). (1.162)
We can therefore interpret the commutator as describing 2 physical processes, whose quantum prob-
ability amplitude apparently cancel each other for ( x y)2 < 0:
Indeed if x and y are at spacelike distance from each other, there is no Lorentz invariant notion of
whether ( x0 y0 ) is bigger or smaller than zero. Therefore both processes can occur. In the ex-
pression for [( x), (y)] these two processes cancel each other in the sense of a destructive quantum
mechanical interference.
Even more interestingly we can consider a complex scalar field, which contains the mode operators
according to
( x) a( ~p), b ( ~p),
(1.163)
( x) a ( ~p), b( ~p).
36 CHAPTER 1. THE FREE SCALAR FIELD
For a complex scalar field [( x), (y)] = 0 for all x, y. More interesting is the commutator
which vanishes for ( x y)2 < 0 while being in general non-zero otherwise. The first term corresponds
to a particle which travels from y to x, while the second term corresponds to an anti-particle travel-
ling from x to y. Again both processes cancel each other in the expression for the commutator. The
important conclusion is that the field formalism saves causality in the QM sense even though the QM
probability for a propagation x y itself is non-zero if ( x y)2 < 0. This is why a single-particle
approach must fail. The field commutators, on the other hand, know about processes of all possible
particles and anti-particles. Thus it is the intrinsic nature of Quantum Field Theory as a multi-particle
framework which is responsible for causality.
where
( x )(y)
if x0 y0 ,
T ( x )(y) =
(1.166)
(y)( x )
if y0 > x0
is the time-ordered product. The Feynman-propagator can be written as
DF ( x y) = ( x0 y0 ) h0| ( x)(y) |0i +(y0 x0 ) h0| (y)( x) |0i
| {z } | {z }
D( xy) D(yx)
Z 3
d p 1 ip( xy)
= ( x y )
0 0
e (1.167)
(2)3 2E p
p0 =+ E p
d3 p 1 ip( xy)
Z
+ (y x )
0 0
e .
(2)3 2E p
p0 =+ E p
we close the contour in the lower half-plane such as to pick up the residue at p0 = + E p and
It is important to note that this result holds for any contour that encloses the poles as drawn. In
particular, since x0 y0 > 0 the integral along the lower half-plane asymptotically vanishes if we
choose to deform the contour to infinity, corresponding to R . For the second term in (1.168) we
can similarly write
1 iE p ( x0 y0 )
(y0 x0 ) e
2E p
0 0 0 (1.171)
eip ( x y )
I
0 0 1 0
= (y x ) dp .
2i C2 ( p0 E p )( p0 + E p )
This time C2 runs counter-clockwise, but picks up the pole at p0 = E p , which again yields an overall
minus.
Now, as stressed above both expressions hold for any R > E p , but if R , then the integral
in the lower-/upper halfplane each vanishes due to the appearance of ( x0 y0 ) and (y0 x0 ),
respectively. It is at this place that the time ordering becomes crucial.
We can therefore evaluate both integrals for R , add them with the help of 1 = ( x0 y0 ) +
(y0 x0 ) and arrive at
d4 p
I
i
D F ( x y) = 4 2 2
eip( xy) , (1.172)
C (2) p m
38 CHAPTER 1. THE FREE SCALAR FIELD
p0 = E p + i /E p and p0 = E p i p /E p . (1.173)
1.11. CAUSALITY AND PROPAGATORS 39
This must be combined with taking the limit 0 after performing the integral. With this understood
2
and using furthermore ( p0 ( E p i /E p ))( p0 + ( E p i /E p )) = p0 E 2p + 2i + 2 /E 2p =
p2 m2 + i with = 2i i /E p , the Feynman propagator can be written as the integral
d4 p
Z
i
D F ( x y) = 4 2 2
eip( xy) (1.174)
(2) p m + i
with the p0 integration along the real axis and with the limit 0 after performing the integral.
Direct computation reveals that DF ( x y) is a Greens function for the Klein-Gordon equation
(2 + m2 )( x) = i(4) ( x) (1.176)
d4 p d4 p ipx
Z Z
( x) = ( p)eipx , (4) ( x) = e (1.177)
(2)4 (2)4
and noting that the Klein-Gordon equation becomes an algebraic equation for the Fourier transforms,
i
(p2 + m2 ) ( p) = i ( p) = . (1.178)
p2 m2
DR ( x y) = ( x0 y0 )[ D( x y) D(y x)]
, (1.179)
( x0 y0 ) h0| [( x), (y)] |0i
while avoiding both poles in the lower half-plane yields the the advanced Greens function
Figure 1.5: The retarded contour. Figure 1.6: The advanced contour.
Note that DR and DA appear also in classical field theory in the context of constructing solutions to
the inhomogenous Klein-Gordon equation, where they propagate the inhomogeneity forward (DR )
and backward (DA ) in time. In particular the classical version of causality is the statement that DR and
DA vanish if ( x y)2 < 0, which we proved from a different perspective before.
By contrast DF ( x y) is the solution corresponding to the contour described previously in this section.
It does not appear in classical field theory. The reason is that DF ( x y) propagates positive frequency
modes eipx forward and negative frequency modes eipx backward in time (see (1.168)). Remember
that DF ( x y) is non-vanishing, even for x and y at spacelike distance. Finally there is a fourth
prescription which has no particularly important interpretation in classical or quantum field theory.
Chapter 2
2.1 Introduction
So far we have considered a free scalar theory
1
L0 = ()2 V0 () (2.1)
2
with V0 () = 1
2 m20 2 .
The theory is exactly solvable: The Hilbert space is completely known as the Fock space of
multi-particle states created from the vacuum |0i.
Interactions are described in QFT by potentials V () beyond quadratic order. We can think of V ()
as a formal power series in ,
1 2 2 1 1
V () = m0 + g 3 + 4 + ..., (2.2)
2 {z } |
| 3! 4!
{z }
V0 Vint
This yields the decomposition of the full Hamiltonian as H = H0 + Hint . Introducing interaction
terms leads to a number of important changes in the theory:
The Hilbert space is different from the Hamiltonian of the free theory.
This is true already for the vacuum, i.e. the full Hamiltonian H has a ground state |i
different from the ground state |0i of the free Hamiltonian H0 :
41
42 CHAPTER 2. INTERACTING SCALAR THEORY
The mass of the momentum eigenstates of H does no longer equal the parameter m0 that
appears in L0 .
Bound states may exist in the spectrum.
The dream of Quantum Field Theorists is to find the exact solution of a non-free QFT, i.e. find the
exact spectrum and compute all interactions exactly. This has only been possible so far for very
special theories with a lot of symmetry, e.g. certain 2-dimensional QFTs with conformal invariance
(Conformal Field Theory), or certain 4-dimensional QFTs with enough supersymmetry.
However, for small coupling parameters such as g and in V () we can view the higher terms as
small perturbations and apply perturbation theory. Note that depending on the mass dimension of
the couplings it must be specified in what sense these parameters must be small, but e.g. for the
dimensionless parameter this would mean that 1 for perturbation theory to be applicable.
Before dealing with interactions in such a perturbative approach, however, we will be able to establish
a number of non-trivial important results on the structure of the spectrum and interactions in a non-
perturbative fashion.
We take a first look at the spectrum of an interacting real scalar field theory in a manner valid for
all types of interactions and without relying on perturbation theory. As a consequence of Lorentz
invariance the Hamiltonian and the 3-momentum operator must of course still commute, [ H, P ~ ] = 01 ,
~ in the
and can thus be diagonalised simultaneously. By | ~p i we denote such an eigenstate of H and P
full theory such that
H | ~p i = E p () | ~p i ,
(2.5)
~ | ~p i = ~p | ~p i .
P
Each | ~p i is related via a Lorentz boost with the corresponding state at rest, called |0 i. We can have
the following types of | ~p i:
1-particle states |1 ~p i with E p = ~p2 + m2 and rest mass m. Remember that m is no more
p
identical to m0 in L0 .
2- and N-particle states formed out of 1-particle and the bound states. In this case we take ~p to
be the centre-of-mass momentum of the multi-particle state.
1 This is simply the statement that the momentum operator is conserved - cf. (1.137).
2.2. KLLN-LEHMANN SPECTRAL REPRESENTATION 43
All these states are created from the vacuum |i. The crucial difference to the free theory is, though,
that ( x) cannot simply be written as a superposition of its Fourier amplitudes a( ~p) and a ( ~p) because
it does not obey the free equation of motion, i.e.
(2 + m2 ) , 0. (2.6)
Rather,
(2 + m2 ) = j (2.7)
for a suitable current j. Thus, acting with on |i does not simply create a 1-particle state as in the
free theory.
We will make frequent use of the completeness relation of this Hilbert space,
XZ d3 p 1
1 = |i h| + | ~p i h ~p | . (2.8)
(2) 2E p ()
3
Here the formal sum over includes the sum over the 1-particle state, over all types of bound states
R 3
as well as over all multi-particle states, while the integral (d2p)3 E 1() refers to the centre-of-mass
p
momentum of a state of species . In particular, since specifying a multi-particle state requires spec-
ifying the relative momenta of the individual states (in addition to the centre-of-mass momentum ~p),
the sum over is really a sum over a continuum of states.
Our first goal is to compute the interacting Feynman-propagator
h| T ( x)(y) |i (2.9)
and to establish a physical interpretation for it. Even though we cannot rely on the mode expansion of
the field any more, we will be able to make a great deal of progress with the help of two tricks.
First we insert 1 between ( x) and (y)2 . We first ignore time ordering. Then, without loss of
generality we can assume that
h| ( x) |i = 0, (2.10)
because since
P P
( x) = eix
(0)eix
(2.11)
h| ( x) |i = h| (0) |i x. (2.12)
2 Whenever we do not know what to do, and Fourier transformation is not the answer, we insert a 1.
44 CHAPTER 2. INTERACTING SCALAR THEORY
Now
h| ( x) | ~p i = h| eiPx (0)eiPx | ~p i , (2.14)
and with h| eiPx = h| and eiPx | ~p i = | ~p i eipx we find
The next trick is to relate | ~p i to |0 i by a Lorentz boost. To this end we investigate the trans-
formation behaviour of a scalar field under a Lorentz transformation
x 7 x0 = x. (2.16)
In the spirit of Quantum Mechanics the action of the Lorentz group is represented on the Hilbert
space in terms of a unitary operator U () such that all states transform like
|i 7 |0 i = U () |i . (2.17)
What is new to us is that also the transformation of the field is determined in terms of U ().
More precisely, the scalar field transforms as
( x) 7 0 ( x0 ) = ( x( x0 )) (2.18)
To see this we start with the familiar transformation of a classical scalar field under a Lorentz
transformation given by
7 0 ( x0 ) = ( x). (2.20)
We now need to find the analogue of this equation for operator-valued fields. The analogue
of the classical value of ( x) is the matrix element h| ( x) |i evaluated on a basis of the
Hilbert space. The transformed field 0 ( x) then corresponds to transformed matrix elements
h0 | ( x) |0 i. Thus the classical relation (2.20) translates into
h0 | ( x0 ) |0 i = h| ( x) |i , (2.21)
| {z }
h|U 1 ( x0 )U|i
| ~p i = U 1 |0 i . (2.22)
which is to be compared with the free scalar result (1.161). Including time ordering, we can peform
the same manipulations for the integral as in the free theory and therefore conclude
XZ d4 p i 2
h| T ( x)(y) |i = e ip( xy)
h| ( 0 ) |0 i . (2.25)
(2)4 p2 m2 + i
which has a typical form like in Figure 2.1. It is crucial to appreciate that the 1-particle states leads
to an isolated -function peak around M 2 = m2 . Therefore below M 2 (2m)2 or M 2 m2bound the
spectral function takes the form
( M 2 ) = 2 ( M 2 m2 ) Z. (2.29)
This identifies the 1-particle state as the first analytic pole at m2 . This is an important observation:
The mass-square m2 of the particle is the location of the lowest-lying pole of the Fourier
transformed propagator.
I.e. computation of the propagator gives us, among other things, a way to read off the mass of the
particle. Note furthermore that
The field-strength renormalisation Z was 1 in free theory because (0) just creates the free particle
from vacuum. In an interacting theory
h| (0) |10 i = Z < 1 (2.32)
because creates not only 1-particle states and thus the overlap with the 1-particle states is smaller.
In fact, by exploiting the properties of the spectral function one can prove formally that
In the asymptotic past, t , the in-states |i, ini are described as distinct wave-packets
corresponding to well-separated single particle states. Being far apart for t , they travel
freely as individual states. This is a consequence of locality of the interactions, which we
assume in the sequel.
As these states approach each other, they start to "feel each other", interact and scatter into the
final states | f i.
2.3. S-MATRIX AND ASYMPTOTIC IN/OUT-STATES 47
For t these final states are again asymptotically free and well-separated 1-particle states.
The concept of free asymptotic in/out states is formalised by the so-called in- and out-fields in and
out with the following properties:
The in-state |i, ini is created from the asymptotic vaccum |vac, ini by action of in as t .
We will see that
|i, ini has E = p2 + m2 with m the value of the 1-particle pole in the Feynman propagator of
p
the full interacting theory. In particular m , m0 . Therefore, in is a free field obeying the free
Klein-Gordon-equation, but with the full mass m , m0 ,
(2 + m2 )in = 0. (2.34)
It is thus possible to expand in in terms of ain ( ~p) and ain ( ~p) so that
d3 p
Z
1
in ( x) = ain ( ~
p ) e ipx
+ a in ( ~
p ) e ipx
, (2.35)
(2)3 2E p
p
where p0 = ~p2 + m2 .
p
in satisfies the following relation to the interacting field : Asymptotically for t the
above logic suggests identifying in ( x) with ( x), at least in weak sense that their their matrix
elements with a basis of the Hilbert space must agree in a suitable manner. We therefore make
the ansatz
C in (2.36)
in the weak sense only, i.e.
h| |i C h| in |i (2.37)
for all |i and |i as t . With this input one can show (see Examples Sheet 5 for the proof)
that
h1 ~p | (0) |i = C h1 ~p | in (0) |i . (2.38)
Since by construction h1 ~p | in (0) |i = 1 this identifies C = Z with Z the wavefunction
renormalisation of the full theory as defined in (2.30). Thus
h| |i Z h| in |i (2.39)
as t . Note that this is really true only in this weak sense. What does not hold in an
interacting field theory is that operator products (i.e. powers of ( x)) approach corresponding
products of in ( x).3
3 Ifthis were to hold, then the field theory would be free: Indeed the assumption that ( x)(y) Z in ( x)in (y) and
similarly for (y)( x) implies that Z = 1 by exploiting the commutation relations for the fields.
48 CHAPTER 2. INTERACTING SCALAR THEORY
Our considerations can be summarised as follows: in/out are free fields with single particle states of
p
energy E p = p2 + m2 , with the mass m of the full theory. We can think of switching off all inter-
actions of the theory as t except for self-interactions of the field. This leads to mass m , m0
and Z , 1. We will be able to understand what is meant by these self-interactions very soon when
discussing the resummed propagator, and it will become clear then that indeed it is imperative to take
into account m , m0 and Z , 1 for in/out .
The Hilbert spaces of asymptotic in- and out-states are isomorphic Fock spaces. Thus there exists
an operator S which maps the out-states onto the in-states, i.e.
S is unitary, i.e S = S 1 ,
in ( x) = S out ( x)S 1 ,
so that | h f , in| S |i, ini |2 it the probability for scattering from the initial states to the final states.
for scattering of asymptotic in- and out-states of definite momenta. One can think of this as the build-
ing block to describe scattering of asymptotically localised wave-packets | fin i = d3 p f ( ~p) |pin i.
R
We first use the definition of |qi , ini in terms of the creation operator of the in-field
d3 k
Z
1 ~ ikx ~ ikx
in ( x) = a in ( k ) e + a in ( k ) e (2.43)
(2)3 2Ek
given by q
|qi , ini = 2Eqi ain (qi ) |i . (2.44)
2.4. THE LSZ REDUCTION FORMULA 49
(cf. (1.151)), where the integral can be evaluated at arbitrary time t. This allows us to trade |qi i
by in ( x) as follows:
q
hp1 , ..., pn , out|q1 , ..., qr , ini = 2Eq1 hp1 , ..., pn , out| ain (q~1 ) |q2 , ..., qr , ini
1
Z (2.46)
= d3 x eiq1 x 0 hp1 , ..., pn , out| in (t, ~x) |q2 , ..., qr , ini x0 =t .
i
Since t is arbitrary, we can take t because in that limit we can make use of the relation
or more generally
lim hp1 , ..., pn , out| in (t, ~x) |q2 , ..., qr , ini = lim Z 1/2 hp1 , ..., pn , out| ( x) |q2 , ..., qr , ini .
t t
(2.48)
This leads to
Our next aim is to let ( x) act from the right on the out-states in order to annihilate one of the
states. To this end we need to relate the above matrix element with t to a matrix element
in the limit t , where we can re-express ( x) by out ( x). We can do so by exploiting that
for all functions f (t, ~x)
Zt f
Z Z
( lim lim ) d x f (t, ~x) = lim
3
dt d3 x f (t, ~x), (2.50)
t t t f t
ti ti
| {z }
d4 x 0 f ( x)
R
i.e. Z Z Z
lim d x f (t, ~x) = lim
3
d x f (t, ~x)
3
d4 x 0 f ( x). (2.51)
t t+
with Z " #
1
B= d4 x Z 1/2 0 eiq1 x 0 hp1 , ..., pn , out| ( x) |q2 , ..., qr , ini (2.53)
i
and
Z
1 iq1 x 1/2
A = lim d3 x e 0 Z hp1 , ..., pn , out| ( x) |q2 , ..., qr , ini
t i | {z }
for t: hp1 ,...,pn ,out|out ( x)|q2 ,...,qr ,ini
q
= hp1 , ..., pn , out| aout (~q1 ) |q2 , ..., qr , ini 2Eq1 . (2.54)
Altogether
q
hp1 , ..., pn , out| q1 , ..., qr , ini = hp1 , ..., pn , out| aout (~q1 ) |q2 , ..., qr , ini 2Eq1 (2.55)
Z " #
+i d4 x Z 1/2 0 eiq1 x 0 hp1 , ..., pn , out| ( x) |q2 , ..., qr , ini .
With respect to the spatial variable we can integrate two times by parts,
Z Z
d4 x m2 2 eiq1 x h...i = d4 x eiq1 x m2 2 h...i, (2.59)
since boundary terms at spatial infinity vanish. To justify this recall that the momentum eigen-
states are to be thought of as convoluted with a wavefunction profile as in | fin i = d3 p f ( ~p) |pin i
R
so that the full states are really localised in space. Altogether this gives
hp1 , ..., pn , out| q1 , ..., qr , ini =
Xn
2E pk (2)3 (3) ( ~pk ~q1 ) hp1 , ..., pk , ..., pn , out| q2 , ..., qr , ini (2.60)
k =1
Z
+ i Z 1/2 d4 x1 eiq1 x1 1 + m2 hp1 , ..., pn , out| ( x1 ) |q2 , ..., qr , ini .
2.4. THE LSZ REDUCTION FORMULA 51
Now we repeat this for all remaining states. First consider replacing hp1 | by out as follows:
We would like to repeat the previous logic and transform this into a sum of two terms, one
of which being the disconnected term hp2 , ..., pn , out| ( x1 )ain ( ~p1 ) |q2 , ..., qr , ini. However, we
need to be careful with the ordering of operators as we cannot simply commute ain ( ~p1 ) through
( x1 ). This is where the the time-ordering symbol T comes in. Namely, observe that for finite
values of x10
Zt f
Z
y0 hp2 , ..., pn , out| T (y1 )( x1 ) |q2 , ..., qr , ini
1/2 ip1 y1
dy01 3
lim iZ d y1 e
t f y1
0 1
ti ti
" Z #
= lim i Z 1/2 3
d y1 e ip1 y1
y0 hp2 , ..., pn , out| (y1 )( x1 ) |q2 , ..., qr , ini (2.62)
y01 1
" Z #
lim i Z 1/2 d3 y1 eip1 y1 y0 hp2 , ..., pn , out| ( x1 )(y1 ) |q2 , ..., qr , ini .
y01 1
| {z }
hp2 ,...,out|( x1 )ain ( ~p1 )|q2 ,...,ini 2E p1 disconnected term
Note the different ordering of (y1 )( x1 ) in both terms. Indeed since the limit x10
appears outside of the correlator in eq. (2.60), the time-ordering symbol precisely yields these
orderings as y01 . The term on the lefthand-side of (2.62) is, as before,
Z
i Z 1/2 d4 y1 eip1 y1 y1 + m2 hp2 , ..., pn , out| T (y1 )( x1 ) |q2 , ..., qr , ini . (2.63)
This can be repeated for all in- and out states to get the Lehmann-Symanzik-Zimmermann
reduction formula
hp1 , ..., pn , out| q1 , ..., qr , ini hp1 , ..., pn , in| S |q1 , ..., qr , ini
= ( disconnected terms)+
n+r Z Z
+ i Z 1/2 d4 y1 ...d4 yn d4 x1 ...d4 xr (2.64)
Pn Pr
ei( k=1 pk yk l=1 ql xl )
y1 + m2 ... x1 + m2 ... h| T (y1 )...(yn )( x1 )...( xr ) |i .
This formula reduces the computation of the S -matrix to the computation of time-ordered
correlation functions of the full interacting theory.
52 CHAPTER 2. INTERACTING SCALAR THEORY
In terms of the Fourier transformed quantities it reads as follows. First note that
d4 p 2
Z
y + m (y) =
2
p + m 2 i py
e ( p). (2.65)
(2)4
We can plug this into (2.64), perform the integrals d4 yk ei( pk pk )yk = (2)4 (4) ( pk pk ) (and
R
R d4 q
similarly for xl , where we define accordingly y + m2 ( x) = (2 ) 4 q 2 + m2 eiqx (q))
and arrive at
n+r Y n Y r
hp1 , ...pn | S |q1 , ..., qr i connected = iZ 1/2 p2k + m2 q2l + m2
k =1 l=1 (2.66)
h| T ( p1 )...( pn )(q1 )...(qr ) |i .
Note that the p1 , . . . , pn and q1 , . . . qr which appear on both sides of this equation are on-shell
since these correspond to the physical 4-momenta of the out- and incoming 1-particles states.
Therefore p2k m2 = 0 = q2l m2 . In order for hp1 , ...pn | S |q1 , ..., qr i connected to be non-zero,
the correlation function appearing on the right must therefore have a suitable pole structure such
Q
as to cancel precisely the kinematic factors nk=1 p2k + m2 r 2 2
Q
l=1 ql + m .
We have arrived at a very precise prescription to compute the connected piece of the S-matrix element
hp1 , ...pn | S |q1 , ..., qr i :
connected
Compute the Fourier transformation of the correspondingqtime-ordered (n + rq)-correlation
function and take all momenta pk and ql on-shell, i.e. pk = ~pk + m and ql = ~p2l + m2 .
0 2 2 0
The result will be a sum of terms each a function of the momenta, which are distinguished by
the structure of their poles.
The connected S -matrix element (times (i Z )n+r )) is the residue with respect to
Qn 1 Qr 1
k=1 p2 m2 l=1 q2 m2 .
k l
2.5. CORRELATORS IN THE INTERACTION PICTURE 53
H = H0 + Hint . (2.69)
There are two ways how to do this, either by performing a path-integral computation, or by computing
in the Interaction Picture. The path-integral formalism is reserved for the course QFT II. In the sequel
we consider the latter approach. Our strategy is to reduce the computation of the full correlator to a
calculation in terms of
This is achieved in the Interaction Picture ( Dirac picture). Let (t, ~x) denote the Heisenberg Picture
field of the full interacting theory and fix some reference time t0 . Then we define the Interaction
Picture operators
I (t, ~x) =eiH0 (tt0 ) (t0 , ~x)eiH0 (tt0 )
(2.70)
I (t, ~x) =eiH0 (tt0 ) (t0 , ~x)eiH0 (tt0 ) .
The motivation behind this definition is that I (t, ~x) satisfies the free Klein-Gordon-equation
with mass m0 as in H0 . One can see this as follows: The defintion (2.70) implies that
Here we are using that (t0 , ~x) = i[ H, (t0 , ~x)] = i[ H0 , (t0 , ~x)] because H and H0 differ only by
powers in (t, ~x).4
Therefore
t I (t, ~x) = I (t, ~x) (2.73)
and likewise
Here it is crucial to appreciate that m0 appears because the commutator involves only H0 and the
computation of i[ H0 , (t0 , ~x)] proceeds as in the free theory.
Equ. (2.71) implies that I (t, ~x) enjoys a free mode expansion of the form
d3 p
Z
1
I ( x) = a I ( ~
p ) e ipx
+ a I ( ~
p ) e ipx
, p0 = E ~p = ~p2 + m20 . (2.76)
(2)3 2E p
p
Furthermore it is easy to see that the interaction picture fields and, as a result of (2.73), also the modes
satisfy the free-field commutation relations
[I (t, ~x), I (t, ~y)] = i(3) (~x ~y), [aI ( ~p), aI (~q)] = (2)3 (3) ( ~p ~q). (2.77)
as in the free theory. This can be seen e.g. by noting that H0 = eiH0 (tt0 ) H0 eiH0 (tt0 ) ( H0 )I
and therefore we can replace in H0 all fields ( x) by I ( x) so that all free field results carry over.
Consequently by the same arguments as in the free theory there must exist a vacuum state annihilated
by all aI ( ~p) and by H0 . This identifies this state as the unique vacuum of the free theory, H0 |0i = 0,
and therefore
aI ( ~p) |0i = 0. (2.79)
At t , t0 , the Heisenberg Picture (t, ~x) and the Interaction Picture I (t, ~x) relate as
which yields
(t, ~x) = U (t, t0 ) I (t, ~x)U (t, t0 ), (2.81)
where
is the time-evolution operator. Note that, since H and H0 do not commute, this is not just eiHint (tt0 ) .
The logic is now to replace the Heisenberg Picture operators ( x) in the correlator (2.68) by the
Interaction Picture operators I ( x) because these obey a free-mode expansion. To further evaluate
the resulting expression, we first derive a useful expression for U (t, t0 ) and second establish a relation
between the vacuum |i of the interacting theory as appearing in (2.68) and the free vacuum |0i on
which the interaction picture modes act.
2.5. CORRELATORS IN THE INTERACTION PICTURE 55
i U (t, t0 ) = HI (t )U (t, t0 ) (2.83)
t
As in Quantum Mechanics we solve the differential equation (2.83)by rewriting it as an integral equa-
tion,
Zt
1
U (t, t0 ) = HI (t0 )U (t0 , t0 )dt0 + U (t0 , t0 ). (2.85)
i
t0
The latter can be solved iteratively with initial value U (0) (t, t0 ) = 1,
Zt
1
U (1)
(t, t0 ) =1 + dt1 HI (t1 ) U (0) (t1 , t0 ),
i | {z }
t0 =1
(2.86)
Zt
1
U (2) (t, t0 ) =1 + dt2 HI (t2 )U (1) (t2 , t0 ),
i
t0
and so on at each iteration. The exact solution is given by the n-th iteration for n ,
!n Z Z1 Zn1 t t t
X 1
U (t, t0 ) = 1 + dt1 dt2 ... dtn HI (t1 ) HI (t2 )...HI (tn ), (2.88)
n=1
i
t0 t0 t0
solves equ. (2.83). Note that the HI (ti ) under the integral are time-ordered.
The solution can be simplified further. Consider e.g. the second term and observe that it can be
rewritten as
Zt Zt1 Zt Zt
1
dt1 dt2 HI (t1 ) HI (t2 ) = dt1 dt2 T ( HI (t1 ) HI (t2 )) . (2.89)
2
t0 t0 t0 t0
56 CHAPTER 2. INTERACTING SCALAR THEORY
and intgral b) is in fact the same as integral a). To see this rotate the square in the t1 t2 plane over
which we integrating by 90 , which gives
Zt Zt2
1
b) = dt2 dt1 HI (t2 ) HI (t1 ) a) . (2.91)
2
t0 t0
From the series expression one can verify that the time-evolution operator has the following properties:
5 You must not confuse time ordering T and time T over the next pages.
2.5. CORRELATORS IN THE INTERACTION PICTURE 57
Note that we are assuming here that h| |0i , 0, which is guaranteed at least for small perturbations
Hint in H.
To bring this to a form involving the time-evolution operator we shift T T + t0 = t0 (T )) and
write for eiH (T +t0 ) |0i
eiH (t0 (T )) eiH0 (T t0 ) |0i = U (t0 , T ) |0i , (2.98)
| {z }
=|0i since H0 |0i= 0
because
h i1
U (t0 , T ) = U (T , t0 )1 = eiH0 (T t0 ) eiH (T t0 )
(2.99)
= eiH (T +t0 ) eiH0 (T t0 ) .
So the vacuum is
1
|i = lim eiE (t0 (T )) h| 0i U (t0 , T ) |0i . (2.100)
T (1i )
Finally we can compute h| T ( x)(y) |i. Suppose first that x0 y0 t0 : Then h| T ( x)(y) |i
becomes
h| ( x)(y) |i = lim (eiE (T t0 ) h0| i)1
T (1i )
h0| U (T , t0 ) U ( x0 , t0 ) I ( x)U ( x0 , t0 )
| {z }
( x)
1
U (y0 , t0 ) I ( x)U (y0 , t0 ) U (t0 , T ) |0i eiE (t0 (T )) h| 0i (2.102)
| {z }
(y)
1
= lim | h0| i|2 eiE 2T
T (1i )
For x0 y0 t0 we arrive at
h0| U (T , x0 ) I ( x)U ( x0 , y0 ) I (y)U (y0 , T ) |0i
h| ( x)(y) |i = lim . (2.104)
T (1i ) h0| U (T , T ) |0i
The nominator is
h0| T I ( x) I (y) U (T , x0 )U ( x0 , y0 )U (y0 , T ) |0i , (2.105)
| {z }
U (T ,T )
where the time-ordering symbol takes care of order. Similar conclusions are obtained for x0 y0 .
Therefore altogether
RT
h0| T I ( x) I (y)ei T dtHI (t) |0i
h| T ( x)(y) |i = lim RT . (2.106)
T (1i )
h0| T ei T dtHI (t) |0i
This same reasoning goes through for higher n-point correlators. Our master formula for computing
correlation function becomes
RT
i I ( xi ) e
Q i dtHI (t )
Y h0| T T |0i
h| T ( xi ) |i = lim RT . (2.107)
T (1i ) dtHI (t )
i h0| T ei T |0i
This formula can be applied to concrete interactions, i.e.
4
L = L0 + Lint , e.g. Lint = ( x ), (2.108)
4!
where has mass-dimension 0. The interaction picture Hamiltonian is then
Z
HI = d3 x I (t, ~x)4 (2.109)
4!
and we can expand the time-evolution perturbatively order by order in if 1 by expanding the
exponential
4 4
RT
lim ei d x 4! I ( x) .
R
lim ei T dtHI (t) = (2.110)
T (1i ) T (1i )
As we will see the (1 i ) prescription for the boundaries of the integral will pose no problems (cf.
discussion after (2.133)).
There are basically two remaining problems:
We will solve these problems by exploiting the action of the creation and annihilation operators on
the vacuum.
2.6. WICKS THEOREM 59
d3 p d3 p
Z Z
1 1
I = a I ( ~
p ) eipx
+ a ( ~p)eipx . (2.111)
3 (2) 2E p I
3
p p
(2) 2E p
| {z } | {z }
=: +
I ( x)
=: I ( x)
+
I ( x) |0i = 0 = h0| I ( x).
(2.112)
To do so we definie normal-ordering:
: a ( ~p1 )a( ~p2 )a ( ~p3 ) : = a ( ~p1 )a ( ~p3 )a( ~p2 ) = a ( ~p3 )a ( ~p1 )a( ~p2 ). (2.113)
It is obvious that
h0| : O : |0i = 0 (2.114)
for every non-trivial operator O , c1 for c C. We begin with h0| T I ( x) I (y) |0i and drop the
subscript "I" from now on. There are two cases to consider, either x0 y0 or y0 x0 . If x0 y0 , then
T ( x)(y) = ( x)(y) and
The first three terms are already normal-ordered and the last term can be put in normal-ordered form
by commuting the fields through each other,
We notice that the last two terms are a C-number c and with
(0)
h0| T ( x)(y) |0i = h0| : ( x)(y) : |0i + h0| c |0i = c DF ( x y) (2.118)
| {z }
0
60 CHAPTER 2. INTERACTING SCALAR THEORY
we find that
(0)
T (( x)(y)) = : ( x)(y) : + DF ( x y). (2.119)
(0)
We use the notation DF ( x y) for the free Feynman propagator in the sequel to emphasize that this
object is defined in terms of the free mass parameter m0 appearing in the Lagrangian. If we define the
contraction
(0) (0)
( x ) (y) = D F ( x y) = D F (y x ), (2.120)
Note that by definition : c := c for a C-number, which explains the notation : ( x)(y) :.7 This
generalises to higher products. E.g. one can show by direct computation that
T (( x1 )( x2 )( x3 )) T (1 2 3 )
= : 1 2 3 : + : 1 2 3 : + : 1 2 3 : + : 1 2 3 : (2.122)
(0) (0) (0)
= : 1 : DF ( x2 x3 )+ : 3 : DF ( x1 x2 )+ : 2 : DF ( x1 x3 ).
T (1 ... N ) = : 1 ... N :
X
+ : 1 ... i ... j ... N :
1i< jN
X
+ : 1 ... i ...k ... j ...l ... N : (2.123)
1i<k< j<lN
!
+ 2 more contractions
+ ...
Indeed one can prove this by induction and summarize Wicks theorem as
h0| T 1 ...2N +1 |0i = 0, because there is always an odd number of normal-ordered fields re-
maining and
(0) (0) (0)
h0| T 1 ...2N |0i = DF ( x1 x2 ) DF ( x3 x4 )...DF ( x2N1 x2N ) + (all other contractions).
P
First, draw 1 point for all xi in ( xi ) and connect these by lines in all possible ways. This gives,
in the present case, three distinct diagrams displayed in Figure 2.2. Note that these are distinct
because we distinguish between x1 , x2 , x3 , x4 .
(0)
Then, to each line between i and j we attach a factor of DF ( xi x j ).
(0)
We then multiply all DF ( xi x j ) and sum over all distinct diagrams.
In interacting theories also contractions involving fields at the same spacetime point occur. For exam-
ple consider the propagator in 4 -theory up to first order in :
R
d4 z4 (z)
h0|T ( x)(y)ei 4! |0i =
Z i (2.126)
h0| T ( x)(y) |0i + h0| T ( x)( x) d4 z 4 (z) |0i + O(2 ).
4!
The O(0 )-term corresponds to a straight line between x and y, i.e. the propagation from x to y, which
(0)
is assigned a factor DF ( x y). The term to first order in is
Z
i
d4 z ( x)(y) (z) (z)(z) (z) (2.127)
4!
alalalaaaaaaaaaaaaaaaaaaaa
Z alalalaaaaaaaaaaaaaaaaaaa
+ d4 z( x)(y)(z) (z) (z)(z) . (2.128)
62 CHAPTER 2. INTERACTING SCALAR THEORY
The graphical representation of these two different contributions is given in Figure 2.3. Note the
appearance of the point z, which is integrated over. The first term is proportional i 18 , while the
second term has a factor of i 12 .
To give an interpretation of these combinatorial factors of 1/8 and 1/2, we observe that they coincide
with 1 over the order ( number of elements) of the symmetry group of the diagrams. These are called
the symmetry factors. The symmetry factors are given by the number of ways one can exchange
components of the diagram without changing the diagram, where by components we mean either the
two ends of a line starting and ending on the same point, entire lines between points or internal points
(vertices - see below).
For the two diagrams these are in turn:
The single loop at z is symmetric under the exchange of the out- and in-going end of the loop, but
the direction must be consistent (i.e. one out- and one ingoing arrow). Therefore the symmetry
group is G = Z2 and its order is |G| = 2.
The double loop is separately symmetric under exchange of the out- and in-going end of both
loops, which gives a Z2 Z2 as well as under exchange of the two loops as a whole. Therefore
the symmetry group is G = Z2 Z2 Z2 and its order is |G| = 8.
One can show in general that if a Feynman diagram with symmetry group G always carries a combi-
1
natorial factor |G| . This is because if some symmetry remains the factor of n!1 in the interaction term
n! n is only partially cancelled by counting the various contractions that yield the same diagram. In
the above examples the symmetry group was due to the ambiguity in assigning arrows to the lines at a
vertex. In addition, if k vertices are identical, the symmetry group includes the group of permutations
of these k vertices of order |S k | = k!.
All other points - here z - over which is integrated over are internal points.
For a n interaction, at an internal point n lines meet. Such points are called vertices.
To each line between points yi and y j (both external and internal) we associate a free propagator
(0) (0)
D F ( yi y j ) = D F ( y j yi ) , (2.131)
where we stress again that the superscript reminds us to take the Feynman propagator of the
free theory.
Multiply all factors, Feynman propagators etc. and divide by the symmetry factor of the dia-
gram.
d4 p
Z
(0) i
D F ( x y) = eip( xy) (2.132)
(2) p m20 + i
4 2
associate an arrow and momentum p in that direction. This means an arrow from x to y corresponds
(0) (0) (0)
to the propagator DF ( x y). Obviously the direction is arbitrary because DF ( x y) = DF (y x),
64 CHAPTER 2. INTERACTING SCALAR THEORY
R
but let us make one such choice for each line. We can then perform the integral d4 z at each vertex
explicitly. E.g. for the 4 -theory 4 lines meet at a vertex, and if, say, the momenta p1 and p2 point
into that vertex and p3 , p4 point out of it, this integral yields
Z
d4 z eip1 z eip2 z eip3 z eip4 z = (2)4 (4) ( p1 + p2 p3 p4 ). (2.133)
This implements momentum conservation at each vertex. The delta-function will effectively eliminate
R
one integral d4 pi per vertex.8
RT
Note at this point that the formal prescription that we integrate limT (1i ) T in (2.107) translates
R
into a corresponding prescription for d4 z. We should therefore view this integral as a complex
contour integral. From the fact that such integrals can be deformed in the complex plane as long as no
poles are hit it follows that the (1 i ) prescription does not affect the result.
The momentum-space Feynman rules9 for the computation of the n-point function
m
d4 z n (z)
Y R
h0| T ( xi )ei n! |0i (2.134)
i=1
at order k are:
To each line between points yi and y j (both external and internal) we associate a free propagator
(0)
D F ( yi y j ) (2.135)
(0)
with one choice of direction and to each such DF (yi y j ) we associate directed momentum p
from yi to y j and a factor
i
. (2.136)
p2 m20 + i
For each external point we multiply a factor of eipx for momentum pointing out of the external
point, or eipx for momentum pointing into the point.
8 Theelimination of the zi -integration works this way only if the vertex is connected to at least one other point, either
internal or external. This is, for example, not the case for the second diagram in Fig. 2.3. More generally, one overall factor
of d4 zi = VolR1,3 = (2)4 (4) (0) remains for diagrams not connected to any of the external points. We will see in the
R
but rather that we give an equivalent set of rules for the computation of the correlator where the integral over the vertex
positions has been performed explicitly.
2.8. DISCONNECTED DIAGRAMS 65
d4 p
Integrate over each momentum
R
(2)4
and divide by the symmetry factor.
Consider the first diagram at order , make a consistent choice of momenta and assign eikx for in-going
arrows at external points, eikx for out-going arrows and i(4) (k1 + k2 k3 k4 ) at the vertex due
to momentum consvervation. We end up with
4 Z !
Y i
h0| T ( xk ) d z (z) |0i
4 4
k =1
4! Diagram
4 Z
Y d4 k j ik1 x1 ik2 x2 ik3 x3 ik4 x4
= e e e e (2.138)
j=1
(2)4
4
Y i
(i)(2)4 (4) (k1 + k2 k3 k4 ) .
k2
j=1 j
m20 + i
As we will see soon, in the context of the LSZ formalism we actually need not the correlation function,
but rather its Fourier transform. E.g. we will need expressions of the form
Z Z Z Z
d4 x1 eip1 x1 d4 x2 eip2 x2 d4 x3 eip3 x3 d4 x4 eip4 x4 h0| ... |0i
Diagram
= (2) ( p1 + p2 q3 q4 )(i)
4 (4)
(2.139)
i i i i
2 2 2 2 2 2
.
p1 m0 + i p2 m0 + i p3 m0 + i p4 m0 + i
case is not connected to any external point and therefore called disconnected piece (or disconnected
diagram). A disconnected piece contains only internal points. By contrast, the part of the diagram
which is connected to at least one external point is called partially connected diagram.
According to the Feynman rules disconnected pieces appear as overall factors of the Feynman dia-
gram. If we sum up all Feynman diagrams that contribute to
N d4 x
R
h0| T ( x1 )...( xk )ei N! |0i (2.140)
the result factorises into the sum of all partially connected diagrams in the above sense multiplied
by the sum of all disconnected diagrams. If you dont believe this, just draw a few pages of Feynman
diagrams relevant for, say, the 4-point function in 4 -theory and convince yourself that the infinite
sum over all relevant diagrams can be organized this way.
Let {V j } denote the set of all individual disconnected pieces (i.e. Vi is a Feynman diagram containing
only internal points which itself is connected as a diagram). Then the sum over all diagrams not
connected to any of the external points can be organized as a sum over all Vi , where each Vi appears
ni -times with ni = 0, 1, ..., . If some Vi appears ni -times we must divide its contribution by ni !
to account for the symmetry factor from interchanging identical Vi . For instance, the sum over all
disconnected diagrams contains
1+ 8+ 88 + 888 + ... + arbitrary products and number of loops.
1
2!
1
3!
Denoting by Vi also the value of the corresponding diagram, we have that the sum over all discon-
nected diagrams is
YX 1 Y P
(Vi )ni = eVi = e i Vi . (2.141)
i n =0
ni ! i
i
This is called exponentiation of disconnected pieces. The final result for (2.140) is therefore of the
form X P
( at least partially connected pieces) e i Vi . (2.142)
Now consider the full correlator
R
i i e
Q i HI dt
Y h0| T |0i
h| T i |i = R . (2.143)
i h0| T ei HI dt
|0i
The denominator contains no external points and thus yields precisely
R P
h0| T ei HI dt
|0i = e i Vi
the partition function. (2.144)
This cancels the factor from the nominator and thus
n
Y X
h| T i |i = (all partially connected diagrams with n external points). (2.145)
i
Note that the partition function plays an important role also in Statistical Mechanics. In computing
the partition function in that context you will encounter many times the same reasoning that organizes
the sum over all its contributions in a form similar to the one above. In fact, when discussing the path
integral approach to Quantum Field Theory in QFT II we will see that this is no coincidence.
2.9. 1-PARTICLE-IRREDUCIBLE DIAGRAMS 67
so X
iE 2T T (1i ) = Vi log | h| 0i|2 (2.147)
| {z }
finite number
and therefore P
i Vi
E = lim (2.148)
T (1i ) 2T
because the finite term plays no role in the limit we are taking. E is the vacuum energy of the
interacting theory in the scheme where E0 - the vacuum energy of the free theory - has been set to
0 by renormalisation of the original Lagrangian after adding a term VolE0 3 to L. In terms of the
R
corresponding energy density this means
P P
E i j Vj j Vj
= lim =i . (2.149)
VolR3 T (1i ) 2T VolR3 (2)4 (4) (0)
Note that the factor of (2)4 (4) (0) in the denominator cancels a corresponding factor appearing in
R
the computation of each of the Vi from an uncancelled overall d4 z - see footnote 8. This factor
represents the familiar IR divergence of the vacuum energy from integration over spacetime. As
E
we will see later when actually computing loops of this type, Vol 3
is also UV divergent because
R
of divergent momentum integration for the vacuum loops. This is no surprise as already VolE0 3 was
R
divergent. We organise these divergences order by order in perturbation theory and can renormalise
them away by counter-terms, i.e. by adding to the classical Lagrangian terms of form10
(1) (2)
L L 1 V0 2 V0 ..., (2.150)
Some of the first few Feynman diagrams are shown in Figure 2.5.
10 Infact, due to renormalisation of the other operators order by order in this is a bit of an oversimplification, but we
will come to this in more detail later in the course.
68 CHAPTER 2. INTERACTING SCALAR THEORY
To organise the sum over Feynman diagrams it is useful to define the concept of 1 particle-irreducible
(1PI) Feynman diagrams. These are diagrams out of which one cannot produce 2 separate non-trivial
diagrams (diagrams containing more than just one line) by cutting a single line. E.g. in Figure 2.5
only the 4th diagram is not 1PI (one can cut it in 2 by cutting the line between the two loops). It is
standard to introduce the notation
X
1PI =
(all non-trivial 1PI diagrams), (2.152)
where it is understood that we do not attach external points to both ends (i.e. no factors eipx or
i/( p2 m20 + i ) from the left or right of
1PI.
2 2
We now define iM ( p ) to be the value of
1PI, where p2 denotes the in- and outgoing momentum.
Clearly computing iM 2 ( p2 ) can be a hard task, and the result will be a complicated function of p2 .
In particular this quantity may be divergent (again due to integration over internal momenta). We will
compute iM 2 ( p2 ) explicitly for an electron in Quantum Electrodynamics, in which case it is called
self-energy, later in the course and find ways to deal with its divergence systematically. Irrespective
of the outcome for iM 2 ( p2 ), the sum over all diagrams contributing to h| T ( x)(y) |i can
be organised as a geometric series in 1 PI diagrams. Let us compute the Fourier transform DF ( p2 )
defined such that
d4 p ip( xy)
Z
D F ( x y) = e D F ( p2 ) . (2.153)
(2)4
DF ( p2 ) can be expanded as
D F ( p2 ) = ff f f f f f
i
p2 m2 +i
0
+
i
1PI
p2 m2 +i
0
i
p2 m2 +i
0
+
1PI
1PI
+ ... (2.154)
The fact that we are not drawing points at the two ends of diagrams symbolizes that we are omitting
the factors of eipx or eipy as these are, by definition, not part of DF ( p2 ). The Fourier transformed
2.9. 1-PARTICLE-IRREDUCIBLE DIAGRAMS 69
propagator is then
i i i
D F ( p2 ) = 2
+ 2
iM 2 2
( p ) + ...
2
p m0 + i 2
p m0 + i p m20 + i
2
2
2
M (p )2 2 2
M ( p )
i
= 2 2
1 + 2 2
+ 2 2
+ ...
(2.155)
p m0 + i p m + i p m0 + i
| {z0 }
q
i 1
= 2 .
p m20 + i 1
M 2 ( p2 )
p2 m20 +i
D F ( p2 ) =
i
p2 (m20 + M 2 ( p2 )) + i
fpf (2.156)
iZ
DF ( p2 ) = + terms regular at m2 (2.157)
p2 m2 + i
for some Z. I.e. Z is by definition the residue (in the above sense) of DF ( p2 ) at its first analytical pole
m2 . This way we can compute
perturbatively to given order in . This is a beautiful result because we can now understand quantita-
tively why the mass of the 1-particle momentum eigenstates in an interacting theory differs from m0 :
The reason are the self-interactions of the field, which are resummed as above to shift the pole of the
full propagator from m0 to m. In particular this picture justifies our assertion made in the context of
asymptotic in- and out-states that the asymptotic states behave as free particles, but with fully renor-
malised mass m , m0 : By sending the particles infinitely far apart from each other we effectively
switch off the interactions between the different particles, but we cannot switch off the interactions
of the asymptotic particles with themselves (or rather of the field with itself). These are precisely the
1PI contributions to DF ( p2 ) and thus the in-and out-states do have the fully resummed mass m2 , m20 .
70 CHAPTER 2. INTERACTING SCALAR THEORY
n Z
Y r Z
Y
d4 yk eipk yk d4 xl eiql xl h| T (y1 )...( x1 )... |i
k =1 l=1
(2.158)
Y i Z 1/2 Y i Z 1/2
= hp1 ...| S |q1 ...i connected ,
k
p2k m2 + i l
2 2
ql m + i
where all ql and pk are on-shell. Now, recall that hp1 ...| S |q1 ...i connected itself cannot be proportional
to p2k m2 or q2l m2 because otherwise it is zero on-shell and thus no scattering occurs; likewise
hp1 ...| S |q1 ...i connected cannot contain any factors of ( p2k m2 )1 or (q2l m2 )1 because then it would
be divergent on-shell, in contradiction with its definition as a quantum mechanical amplitude. As a
result only those Feynman diagrams are relevant with exactly (n + r ) poles at m2 in the above sense.
It is not hard to see that these are precisely the ones that contribute to the fully connected correlation
function. A fully connected correlation function has the structure displayed in Figure 2.6.
Figure 2.6: Fully connected correlation function with r in and n out states.
Here xfpf denotes the fully resummed propagator from lines connected to external points.
The fully resummed propagator would appear if we computed to all orders in perturbation theory.
More realistically, to a given finite order in perturbation theory we should think of the propagator as
computed to suitable order.
of scattering amplitudes is
n + r
hp1 , ..., pn | S |q1 , ..., qr i connected = Z (2.159)
YZ YZ
d4 yk eipk yk d4 xl eiql xl h| T (y1 )...( x1 )... |i |fully connected |Amputated
k l
Note that the wavefunction renormalisation factor Z itself is of the form 1 + O() in perturbation
theory, so to leading order in the coupling constant the Z-factors play no role as only O() diagrams
xfxfpfx
can be fully connected.
x1
x2 pfx y1
y2
As an example we consider again the 4 -theory to leading order. The only fully connected diagram
at O() is the first diagram in Figure 2.4. We make a consistent choice of ingoing versus outgoing
momenta and recall the result
h|T ( x1 )( x2 )(y1 )(y2 ) |i =
4 Z
Y d4 k j ik1 x1 ik2 x2 ik3 y1 ik4 y2
= e e e e
j=1
(2)4 (2.160)
4
Y i
(i)(2)4 (4) (k1 + k2 k3 k4 ) .
k2
l=1 l
m20 + i
and then we amputate by discarding all propagators from external lines. The result is
hp1 , p2 | S |q1 , q2 i connected, O() = (i)(2)4 (4) (q1 + q2 p1 p2 ), (2.162)
4
where we have discarded the factor Z because, as discussed, it does not contribute to the leading
order result.
Draw the relevant fully connected Feynman diagrams with (n + r ) external points to given order
in .
Assign ingoing momenta ql and outgoing momenta pk and label momenta of internal lines with
k j.
d4 k j
Integrate over all internal momenta
Q R
j (2)4
and divide by the symmetry factor.
n + r
Sum up all diagrams and multiply by Z to given order in .
The connection between correlation functions and S -matrix elements gives Feynman diagrams an
intuitive physical meaning: In Figure 2.8 two particles come in and interact at Z1 to form two so-
called "virtual" particles, which in turn join again at Z2 and form two outgoing states. The obvious
f
interpretation of the Feynman diagram elements is therefore:
P
eipx is the wavefunction for a momentum-eigenstate.
probability amplitudes for all possible processes - called "channels" - with the same macro-
scopic result. In particular, this will lead to QM interference between the different channels.
Intermediate particles, e.g. those running in the loop as k1 and k2 in Figure 2.8, are called virtual
because they are generally off-shell: This means that these particles do in general not satisfy the
relation ki2 m20 = 0. E.g. for the loop-diagram above we integrate over all internal momenta (where
the delta-functions at the vertices implies that some of these integrals become trivial). In the present
diagram momentum conservation at the first vertex implies that
q1 + q2 = k1 + k2 k2 = q1 + q2 k1 , (2.163)
where k1 is free and thus integrated over. For most values of k1 we have k12 m20 , 0. We can
think of this as a quantum-mechanical violation of the energy-momentum relation because E 2j ,
2
k~j + m20 for virtual particles. In Quantum Mechanics this is allowed for sufficiently short times - in
fact this precisely what is meant by the energy-time uncertainty relation arising in quantum mechanical
perturbation theory. Note, however, that 4-momentum conservation is guaranteed at each vertex due
to the factors
X
(2)4 ingoing momenta outgoing momenta . (2.164)
On assignment 7 will work out a detailed a comparison of the LSZ approach to scattering as com-
pleted in this chapter and a more naive approach based on quantum mechanical Interaction Picture
perturbation theory.
2.11 Cross-sections
S -matrix elements are in general of the form
Assume now that |ii < {| f i} (so that in particular the term f i is irrelevant), then
X h i2 2
P|ii{| f i} = (2)4 (4) ( p f pi ) M f i . (2.167)
| f i{| f i} | {z }
=(2) ( p f pi ) (2) (0)
4 ( 4 ) 4 (4)
| {z }
=V 1,3
R
We define the transition rate as the probability normalised per spacetime volume,
P|ii{| f i}
fi = (2.168)
unit time unit volume
and therefore
X 2
fi = (2)4 (4) ( p f pi )M f i . (2.169)
| f i{| f i}
N
d3 kn 1
Z
1 Y X
M 2 ,
X
fi =
4 (4)
( )
2 pi kn fi (2.170)
(2)3 2En
N! n=1
i n
1
where the symmetry factor N! accounts for the indistinguishability of the N identical particles.
F
Now, a typical scattering experiment is of the form
where A,B denote the respective number densities (i.e. number of particles per volume). The number
(#) of scattered particles is proportional11 to lA lB d2 x a ( x) B ( x). The factor of proportionality has
R
If A and B are constant and both beams overlap over the area A, then
# of scattered particles
= . (2.172)
lA l B A B A
This suggests the following intuitively clear interpretation of the meaning of : The cross-section
is the effective area of the beam B that participates in the scattering.
Consider now a 2 N scattering process such that all out-going states |k j i are momentum eigen-
states. The initial states are the momentum eigenstates |pA i and |pB i. Let us go to the rest frame of
11 Here we are assuming that all particles of the beam hitting the target get to interact with the target particles with equal
probability. For simplicity we are also assuming that the particle density varies only in the directions x1 , x2 transverse to the
beam.
2.11. CROSS-SECTIONS 75
the target A called lab frame. Suppose for simpilcity that the beam B hits the target A over its entire
area A and that the densities of beam and target do not vary much. Then
# of events A B lA l B A
= = A B ~vLB , (2.173)
Volume Time Volume Time
where the relevant interaction volume is is just Volume = A lA and the relevant interaction time is
what it takes for the beam to pass over a given slice in the target orthogonal to the beam direction, i.e.
(L)
Time = ~
lB
vB . Sometimes it is also useful to think in terms of the beam flux F defined as
B ~vLB =: F . (2.174)
we must replace
hpA |pA i
A = 2E A and b 2E B . (2.178)
VR3
( L)
Since we are in the rest frame of A we have 2E A = 2m. Therefore in the lab frame (~vA = 0)
QM probability ( L)
fi = = 4 m pB . (2.179)
Volume Time
Note that
( L) ( L) ( L) ( L)
4mpB = 4 E A E B |~vB | = 4E A E B |~vA ~vB | (2.180)
( L) ( L)
in the lab frame. Finally, one can show that 4 E A E B |~vA ~vB | is invariant under Lorentz boosts
in direction A B. Therefore f i = 4 E A E B |~vA ~vB | is the correct general expression for the
transition rate valid in any frame 12 . To conclude, the differential cross-section is
(2)4 X 2
d N pa + pB
(4)
d = ki M f i , (2.181)
4E A E B |~vA ~vB | i
12 For a more formal, but also considerably more complicated proof see Peskin-Schrder p.102 108.
76 CHAPTER 2. INTERACTING SCALAR THEORY
where
N
d3 kn 1
Z
1 Y
d N . (2.182)
N! n=1 (2)3 2En
For example consider 2 2 scattering as shown in Figure 2.9. On Assignment 7 we show that if all
s = ( p1 + p2 )2 = ( p3 + p4 )2 (2.184)
appears. Note that in 4 -theory |M|2 = 2 to first order in , i.e. for the maximally localised,
pointlike interaction given by the first Feynman diagram in Figure 2.4. Therefore
d 1
(2.185)
d s
to first order in . This is a famous result, which in fact holds more generally:
The differential cross-section for hard scattering off a pointlike target (i.e. a target with
no substructure of length l 1/ s) falls off as 1/s.
This characteristic behaviour as observed in deep inelastic scattering experiments with hardons was a
crucial clue to the the parton structure of hadrons as you will surely recall from your particle physics
course.
Chapter 3
x 7 x0 = x , S O(1, 3). (3.1)
Quite generally a field a ( x) transforms as a representation1 of the Lorentz group S O(1, 3) - or, more
precisely, as we will see momentarily, of its double cover Spin(1, 3). This means that the classical
field a ( x) is a map
a : R1,3 V, (3.2)
x 7 a ( x), a = 1, . . . , dim(V ) (3.3)
with V a vector space such that the specific transformation behaviour of the field a ( x) under (3.1) is
given by
a ( x) 7 R()ab b (1 x0 ) = R()ab b ( x). (3.4)
Here for every element S O(1, 3) the object R() is an automorphism (an invertible linear map)
acting on V in a manner compatible with the group action of S O(1, 3). Specifying the representation
in which the field transforms amounts to assigning such an automorphism R to every element
S O(1, 3), where compatibility with the group action of S O(1, 3) means that2
1 The concept of Lie groups, Lie algebras and their representations has been discussed in detail in the course on Quantum
Mechanics. A good book in the present context is: Urbantke, Sexl: Relativity, Groups, Particles. As a quick reminder of the
main points see e.g. http://www.thphys.uni-heidelberg.de/$\sim$weigand/Skript-QM2011/skript.pdf.
2 In short, a representation R of a group G is a group homomorphism from G to Aut(V ) for some vector space V.
77
78 CHAPTER 3. QUANTISING SPIN 21 -FIELDS
The vector space V is also called representation space. Note that the index a refers to the components
of the fields with respect to a basis of the vector space V - called the representation space -, while ,
are spacetime indices. These two are in general completely different objects.
For a real (or complex) scalar field ( x) the representation space V is just R (or C) and
R() = 1 . (3.6)
The scalar field is said to transform in the trivial or scalar representation and describes parti-
cles with spin 0.
A vector field A ( x) (e.g. the gauge potential of electro-magnetism) transforms in the vector
representation: The representation space V is identified with spacetime R1,3 itself (or rather
its tangent space) and
R() = . (3.7)
Since here V = R1,3 , we have in this case that a, b = , . A field in the vector representation
describes particles with spin 1 (as we will see later in this course when quantising such vector
fields).
Spin 1/2 particles are described by fields in the spinor representation. To find this representation
our starting point will not be the irreducible representations of the Lorentz group S O(1, 3), but of
This is the algebra of infinitesimal Lorentz transformations connected to the identity. This is analo-
gous to the precedure applied in quantum mechanics to find spin 21 representations of the algebra of
spatial rotations so(3) ' su(2).
Since the metric must be invariant under a Lorentz transformation in the sense that
= , (3.10)
Then
i
= A M A (3.12)
2
for some A with summation over A understood. We introduce the double-index notation A =
with , = 0, 1, 2, 3 such that
( M ) = ( M ) (3.13)
is antisymmetric in . Therefore
i
= ( M ) . (3.14)
2
So an infinitesimal Lorentz transformation is
i
= ( M ) . (3.15)
2
A finite transformation is given by applying the infnitesimal version N times for N in the sense
!N
1 i
( M ) = lim ( M )
N N2 (3.16)
i
= e 2 ( M ) .
This is what is meant by saying that the M are the generators of the Lie group S O(1, 3), or equiva-
lently form a basis of the Lie algebra so(1, 3).
The structure of the Lie algebra so(1, 3) is encoded in the commutation relations of its basis elements.
To find these one expands the defining relation of the underlying Lie algebra S O(1, 3)
= (3.17)
[ M , M ] = i ( M + M M M ) . (3.18)
This defines the structure constants of the Lie-algebra so(1, 3). Recall that in general a basis T A of a
Lie algebra satisfies X
[T A , T B ] = i f AB C T C , (3.19)
C
f AB C = f BA C
f AB D f CD E + f CA D f BD E + f BC D f AD E = 0. (3.20)
We view the commutator (3.18) as the defining relation for abstract objects M . An n-dimensional
representation of the Lie algebra so(1, 3) is an assignment that associates to each M an invertible
map
acting on an n-dimensional vector space V such that the same relation (3.18) holds for the represen-
tation matrices (R )ab . The indices a, b = 1, . . . , n refer to a basis ea of V in the sense that v V is
expanded as v = va ea . Note again that in general a and b are unrelated to and .
Consider as a special case the vector representation by choosing V = R1,3 (viewed as the tangent
space to spacetime). In this case we do identify a, b , and set
(J ) := i( ). (3.22)
One can check that this indeed satisfies the so(1, 3) relations (3.18) and that
i
= (J ) . (3.23)
2
Therefore
i
= e 2 J . (3.24)
We can use this result to deduce the matrix representation of spatial rotations by an angle around an
axis ~n. The corresponding can be written as
i j = i jk nk . (3.25)
( )A B ( ) BC + ( )A B ( ) BC = 2 1AC . (3.30)
This implies
if =
=
. (3.31)
if ,
Therefore
(0 )2 = 1, (i )2 = 1. (3.32)
i
(S )A B := [ , ]A B (3.33)
4
S , S k = i k S + S k S k k S ,
h i
(3.34)
therefore (3.18) holds. Thus we have constructed a representation of so(1, 3) from the Clifford alge-
bra:
One can prove this by direct computation. To this end note that
0 if = i
= ( ).
S = (3.35)
i if ,
2
2
The claim then follows with the help of the anti-commutation relations (3.29) after some algebra as
worked out in the tutorial.
Now we want to find an explicit representation of the Clifford algebra. Since it is useful to know
the result for an arbitrary number of spacetime dimensions let us give the general result valid for
Cliff(1, d 1). The problem is therefore to find n n-matrices
subject to (3.29). This will then also give a representation of the Lorentz algebra so(1, d 1). In the
tutorials we will prove the following famous theorem:
Let us now specialise to d = 4 as is relevant for QFT in four spacetime dimensions. In this case
n = 4. One choice of called chiral or Dirac representation is
0 12 0 i
=
0 , =
i , (3.38)
12 0 i 0
with
{i , j } = 2i j . (3.40)
Given an invertible matrix U, every other choice U U 1 is also a representation of Cliff(1, 3). The
complex vector space on which ( )A B acts is called the space of Dirac spinors A , A = 1, ..., n with
n = 4 for d = 4.
Now, from the above it is clear that the A also form a representation of so(1, 3) because the
induce a representation of so(1, 3). More precisely, a Dirac spinor A transforms under a Lorentz
transformation
i
= e 2 J (3.41)
as
A
i
A 7 e 2 S B . (3.42)
B
3.2. THE DIRAC SPINOR REPRESENTATION 83
x 7 x0 = x, (3.43)
A ( x) 7 [S ()] A B B (1 x0 ) = [S ()] A B B ( x) (3.44)
with
A
i
[S ()]A B = e 2 S . (3.45)
B
The indices A = 1, ..., n, with n = 4 for Cliff(1, 3) are called spinor indices. Even though in 4
dimensions A happens to run from 1, . . . , 4, we must not cunfuse them at any time with spacetime
indices , .
Now that we found a new type of representation of the Lorentz algebra so(1, 3) we would like to
give it a physical interpretation. The important claim is that
i j = i jk nk (3.46)
and i A
2 i j S i j
S [] = e . (3.47)
B
Now,
i j k
i 1
i i j 0 0
S ij = [ , ] = 2 = 2 (3.48)
4 0 2i i j 0 2
1 k
with (i jk) cyclic permutations of 123. The latter equations follows from the properties of the Pauli
matrices. So
ei 2 ~n ~
0
S [] = . (3.49)
ei 2 ~n ~
0
What is crucial is the factor of 12 , which indeed indicates a transformation as a spin 21 under the
subalgebra so(3) so(1, 3) of spatial rotations. In particular consider a rotation by = 2 around
~nT = (0, 0, 1),
i3 0
S [] = exp
= 1. (3.50)
0 i3
Therefore
A 7 A (3.51)
under a rotation by 2.
84 CHAPTER 3. QUANTISING SPIN 21 -FIELDS
While we have found a representation of the Lorentz-algebra so(1, 3), this does not give a representa-
tion of the Lorentz-group S O(1, 3), but rather of its double cover Spin(1, 3). The latter is isomorphic
to S L(2, C), the group of complex 2 2 matrices with determinant 1 (not to confuse with S U (2), the
group of complex unitary such matrices). In the tutorial we will investigate this isomorphism and the
corresponding description of spinors.
A Dirac spinor forms a representation of Spin(1, 3) ' S L(2, C), not of S O(1, 3).
The reason is that S [ = 1] , 1, which is incompatible with the group law of S O(1, 3). In Spin(1, 3)
by contrast, a rotation around 2 does not correspond to the 1 because it is defined as the double cover
of S O(1, 3).3 This is the relativistic version of the statement familiar from Quantum Mechanics
that the j = 21 spinor representation of the algebra of spatial rotations so(3) does not furnish a
representation of the Lie group S O(3) but only of its double cover S U (2).
We now would like to know how ( x)( x) behaves under a Lorentz transformation , under which
x 7 x0 = x, (3.53)
( x) 7 S () (1 x0 ), (3.54)
( x) 7 ( x ) S ().
1 0
(3.55)
In order for ( x)( x) to transform as a Lorentz scalar, we would need S () = S 1 (). However,
S () = e[ 2 S ] = e 2 (S )
i i
(3.56)
and i
i h i
i h i
(S ) = [ , ]
, = , .
= (3.57)
4 4 4
It is not possible to pick all hermitian at the same time since
(0 )2 = 1 real eigenvalues,
(3.58)
(i )2 = 1 imaginary eigenvalues.
(0 ) = 0 , (i ) = i , (3.59)
( ) = 0 0 , (3.60)
:= 0 . (3.63)
It transforms as
So ( x)( x) is a scalar quantity. Furthermore one can show that ( x) ( x) transforms as a vector
because
( x) ( x) 7 (1 x0 )S 1 () S ()(1 x0 ) (3.66)
and
S 1 () S () = , (3.67)
as will be proven in the tutorials. This logic can be repeated for tensors of rank n.
We can now build an action for ( x), where we follow the principle of simplicity. The simplest
Lorentz scalars built from ( x) which include non-trivial dynamics (i.e. at least one derivative has to
appear) are
( x) ( x), ( x)( x). (3.68)
So we take as the action for the free classical Dirac spinor field
Z
S = d4 x ( x) [i m] ( x), (3.69)
where
86 CHAPTER 3. QUANTISING SPIN 21 -FIELDS
This is the simplest action we can build. Interestingly - unlike in for the scalar field - it is possible to
construct a Lorentz invariant action with only one derivative. Furthermore note that ( x) has mass
dimension [mass]3/2 .
Since ( x) is complex we treat ( x) and ( x) as independent when deriving the equations of motion.
Varying with respect to ( x) yields the Dirac equation
(i m)( x) = 0, (3.70)
which is linear in the derivatives. In order to see that this indeed describes a field with mass |m| we
note that
0 = (i + m)(i m)
= ( m2 )
1 (3.71)
= { , } + m2
2 {z }
|
=
and thus obtain the Klein-Gordon equation (1.29) for the spinor field ( x). Loosely speaking:
p
Dirac equation = Klein-Gordon equation. (3.72)
This is a consequence of the particular manner how we constructed a representation of the Lorentz
algebra from the Clifford algebra. Equ. (3.71) also justifies the relative factor of i in the Dirac action,
for which we can take m to be real.
e 2 ~~
1
0
7 .
+ 1
~
~
(3.75)
0 e 2
3.4. CHIRALITY AND WEYL SPINORS 87
transform separately. Irrespective of the concrete representation the reducibility of the Dirac spinor
repsentation as a representation of Spin(1,3) can be seen as follows: Define
5 = i0 1 2 3 (3.77)
following from the properties of the Clifford-algebra. Now consider the orthogonal projection opera-
tors
1
P := (1 5 ) (3.79)
2
with the properties
(P )2 = P , P+ P = 0 = P P+ , 1 = P+ + P . (3.80)
Defining
:= P (3.81)
yields
P = 0. (3.82)
0 12 0 i
12 0
=
0 , =
i , =
5 (3.83)
12 0 i 0 0 12
we find
u 0
= , + = . (3.84)
0 u+
The 2-component objects u are called Weyl-spinors. It is the Weyl spinors that form irreducible
representations of Spin(1,3).
It is instructive to rewrite the Dirac action in this language. The Dirac action decomposes as fol-
lows: Since 0 5 = 5 0 we have
0 P = P 0 (3.85)
and therefore
= (P ) 0 = P 0 = P . (3.86)
88 CHAPTER 3. QUANTISING SPIN 21 -FIELDS
This means
= P P = 0,
(3.87)
= P , 0
and similary (using in addition that P = P )
= P , 0
(3.88)
= 0.
The Dirac action can then be written as
Z
S = d4 x (i m)
Z (3.89)
= d4 x + i + + i m(+ + + ) ,
h i
or in Weyl-spinor notation as
Z
d4 x u+ i u+ + u i u m(u+ u + u u+ ) .
h i
S = (3.90)
Here
(12 , i ) and = (12 , i ). (3.91)
From the decomposed Dirac equation we can draw the following important conclusions about the
underlying physics:
If m = 0, u+ and u decouple and describe independent degrees of freedom subject to the Weyl
equations
i u+ ( x) = 0, i u ( x) = 0. (3.92)
1
Both u+ and u transform in the s = 2 representation of SU(2) Spin(1, 3), in the sense that
under spatial rotations
x 7 R x, (3.93)
2i ~n~
u ( x) 7 e u (R1 x). (3.94)
u ( x) = u ( p)eipx (3.96)
and finds
1
h u ( p) = u ( p), (3.97)
2
where u+ corresponds to right-handed and u to left-handed spinors.
3.5. CLASSICAL PLANE-WAVE SOLUTIONS 89
If m , 0, u+ and u do not decouple. In this sense the full 4-component Dirac spinor is
needed to describe massive spin- 21 fields. Note for fields of mass m , 0 it is impossible to
define a Lorentz invariant notion of helicity because massive particles travel with a velocity
v < c and it is always possible to find a Lorentz frame in which the particle moves into the
opposite direction. This causes a change in the helicity.
( x) = u( ~p)eipx (3.98)
with p = ( E p , ~p) and E p = ~p2 + m2 . Here we have used that the dispersion relation p2 m2 = 0
p
is satisfied because ( x) obeys the Klein-Gordon equation (3.71). Plugging this ansatz into the Dirac
equation yields
( p m)u( ~p) = 0. (3.99)
( p )( p ) = ( p0 )2 pi p j |{z}
i j = m2 . (3.103)
i j
The general solution can be parametrised by introducing some 2-component Weyl spinor 0 and writ-
ing
p 0
1
u( ~p) = m
. (3.104)
0
We make a conventional choice
0 = p
p
(3.105)
with = 1. Here M denotes the matrix whose eigenvalues are the square root of those of M.
With this choice we find the so-called positive frequency solution
p
( x) = u( ~p)e ipx
, u( ~p) = p . (3.106)
p
90 CHAPTER 3. QUANTISING SPIN 21 -FIELDS
Likewise
( x) = v( ~p)eipx (3.107)
is a solution if
( p + m)v( ~p) = 0. (3.108)
and therefore
s s0 = ss0 . (3.111)
1 1
3 1 = 1 . (3.112)
2 2 2 2
Correspondingly u s and v s with s = 21 describe spinors with spin 12 in direction x3 . In the tutorial
we will convince ourselves of the important identities
and
X
u s ( ~p)u s ( ~p) = p + m,
s
X (3.114)
v s ( ~p)v s ( ~p) = p m.
s
Let us consider the following example: Suppose we have chosen coordinates such that p = ( E, 0, 0, p3 ).
According to the above, the Dirac spinor solution u 1 ( ~p) with spin 21 along x3 is given by
2
p 1 E p3 1
p
u 1 ( ~p) = 2 = p 2 . (3.115)
2 p 1 E + p3 1
2 2
3.6. QUANTISATION OF THE DIRAC FIELD 91
If m = 0, then E = p3 and
0
0
u 1 ( ~p) = 2E u+ ( ~p) (3.116)
2 1
0
1 0
since 1 = . The solution correspnding to 1 = is
2 0 2 1
0
E + p3 1 m=0
p
1
u 1 ( ~p) = p 2 2E u ( ~p). (3.117)
2 E p3 1 0
2
0
1
h u+ ( ~p) = u+ ( ~p) right-handed,
2 (3.118)
1
h u ( ~p) = u ( ~p) left-handed.
2
L
A = = iA . (3.120)
A
where u s and v s are the spinor-valued solutions to the classical equation (3.70) corresponding to the
ansatz (3.98). The quantum operators a s and b s are independent because the field is complex. This
expansion guarantees that the corresponding Heisenberg fields with ei~p~x 7 eipx satisfy the Dirac
equation as an operator equation.
We are now careful with operator orderings and proceed without specifying the commutation rela-
tions. The Hamiltonian density is
H = L = (i j j + m). (3.122)
Now,
XZ d3 p 1
(i j + m) =
j
3
p ( j p j + m)u s ( ~p)a s ( ~p)ei~p~x
s
( 2 ) 2E p (3.123)
+ ( j p j + m)v s ( ~p)bs ( ~p)ei~p~x
because
k
j ei~p~x = j eipk x = ip j ei~p~x . (3.124)
Now use that v s and u s solve the classical equation of motion, i.e.
So far no re-ordering of operators has been performed. Now, the naive guess would be to impose, at
the next step, canonical commutation relations as for the scalar field,
h i ?
A (~x), B (~x0 ) = i AB (3) (~x ~x0 ),
h i ?
A (~x), B (~x0 ) = AB (3) (~x ~x0 ), (3.130)
h i ?
A (~x), B (~x0 ) = 0.
3.6. QUANTISATION OF THE DIRAC FIELD 93
We will now show why this is wrong. The commutation relations would imply for the modes
h i
ar ( ~p), as (~q) = (2)3 , rs , (3) ( ~p ~q),
h i
br ( ~p), bs (~q) = (2)3 , rs , (3) ( ~p ~q), (3.131)
()
ar ( ~p), b s (~q) = 0.
Indeed we have
XZ d3 p d3 q 1 h
ur ( ~p)u s (~q) ei( ~p~x~q~y)
i
[(~x), (~y)] =
3 (2)3
p a r ( ~p ) , a
s ( ~
q )
( 2 ) 4E p E q
r,s
(3.132)
i( ~p~x~q~y)
h i
+ br ( ~p), b s (~q) vr ( ~p)v s (~q)e
.
Then, using s u s ( ~p)u s ( ~p) = p + m and s v s ( ~p)v s ( ~p) = p m, the commutator relations
P P
(3.130) of the field would follow from the commutator relations (3.131) of the modes. Note that
appear on same footing in the commutator of the field. This ordering is the reason for the crucial
relative minus sign in (3.131). If this were correct, then we could reorder the Hamiltonian to find
d3 p
Z Xh i
H= 3
E p as ( ~p)a s ( ~p) bs ( ~p)b s ( ~p) + (2)3 (0) . (3.134)
(2) s
[ H, a s ( ~p)] = E p a s ( ~p),
(3.135)
[ H, as ( ~p)] = E p as ( ~p)
and also
[ H, b s ( ~p)] = E p b s ( ~p),
(3.136)
[ H, bs ( ~p)] = E p bs ( ~p)
because the minus sign in the Hamiltonian cancels with the sign we pick up in the commutator. So as
in the scalar theory we would interpret
and define |0i such that a s ( ~p) |0i = b s ( ~p) |0i. The creation operators give positive energy modes by
construction.
While at first sight everything looks fine, in actuality this whole construction is in deep conflict with
unitarity. The reason is that due to the minus sign in the commutator
h i
br ( ~p), bs (~q) = (2)3 rs (3) ( ~p ~q) (3.139)
94 CHAPTER 3. QUANTISING SPIN 21 -FIELDS
the b-mode excitations are negative norm states. To see this note that
h0| [br ( ~p), bs (~q)] |0i = (2)3 rs (3) ( ~p ~q) h0| 0i. (3.140)
0 > (2)3 (3) (0) = h0| [br ( ~p), br ( ~p)] |0i = h0| br ( ~p)br ( ~p) |0i = ||br |0i ||2 . (3.141)
The origin of this problem lies in the fact that the signs in the spinor theory conspire such that to
establish a commutation relation of the schematic form [, ] 1 we must impose
If instead we impose a relation symmetric in and this minus sign for the b-mode relation would
not occur. The task is therefore to promote the classical Poisson-bracket relations not to operator
commutation relations, but to an analogous bracket which is symmetric in both entries. The simplest
such bracket is the anti-commutator. It turns out that this procedure is successful.
a a bb we now find
P
Starting from H
d3 p
Z Xh i
H= 3
E p as ( ~p)a s ( ~p) + bs ( ~p)b s ( ~p) (2)3 (0) . (3.145)
(2) s
3.6. QUANTISATION OF THE DIRAC FIELD 95
The divergent vacuum energy has opposite sign compared to a scalar theory. In theories with scalars
and spin- 12 fields cancellations in the vacuum energy are indeed possible.4
Since for this Hamiltonian the anti-commutation relations still imply the commutation relations (3.135)
and (3.136), the vacuum is again defined by a s ( ~p) |0i = 0 = b s ( ~p) |0i. From this vacuum we define
the Fock space of a- and b-mode excitations. Let us start with the a-modes, which we will call the
particle sector.
1-particle states
The state |~p, si := 2E p as ( ~p) |0i is a 1-particle state with momentum ~p, energy E p = ~p2 + m2
p p
N-particle states
The state
N q
Y
|p1 , s1 ; ...; pN , sN i = 2E pi as1 ( ~p1 )...asN ( ~pN ) |0i (3.147)
i=1
is an N-particle momentum eigenstate. This allows us, in complete analogy to the scalar field, to state
the following theorem
1
The wavefunction of N-particle states of spin 2 particles is anti-symmetric under particle
exchange.
Indeed, if we exchange two particles, we pick up a minus sign due to the anti-commutation relations:
1
Spin 2 particles obey Fermi-statistics, i.e. they are fermions.
No two fermionic states of exactly the same quantum numbers are possible.
4 More generally, scalars and spinors contribute with opposite signs in loops and theories with supersymmetry, i.e. with
an equal number of bosonic and fermionic degrees of freedom, therefore have a chance to exhibit better UV properties.
96 CHAPTER 3. QUANTISING SPIN 21 -FIELDS
The Pauli exclusion principle is again a result of the anti-commutation relation because
Excitations bs ( ~p) |0i describe the corresponding anti-particles. Indeed the Lagrangian
L = (i m) (3.150)
7 ei , 7 ei , R. (3.151)
The associated conserved Noether current will be found in the tutorial to take the form
j = , (3.152)
d3 p X
Z Z Z !
Q= 3
d xj = 0
d x =
3
a s ( ~p)a s ( ~p) b s ( ~p)b s ( ~p) ,
(3.153)
(2)3 s
after dropping a normal ordering constant. The charge acts on the 1-particle state as follows:
Finally, a careful analysis of the angular momentum operator via Noethers theorem reveals that
but
J x3 bs ( ~p = 0) |0i = s bs ( ~p = 0) |0i , (3.156)
with s = 21 , For details of the derivation see Peskin-Schrder, page 61. This shows that bs ( ~p = 0) |0i
has spin (in x3 -direction) s, while as ( ~p = 0) |0i has spin + s.
5 For a general proof see Weinberg (1, 5.7).
3.7. PROPAGATORS 97
3.7 Propagators
As for the scalar fields, we now move to the Heisenberg picture by considering the time-dependent
free fields (with free mass denoted by m0 to avoid confusion)
XZ d3 p 1 h i
( x) = 3
p a s ( ~p)u s ( ~p)eipx + bs ( ~p)v s ( ~p)eipx ,
s
(2) 2E p
(3.157)
d3 p
Z
X 1 h i
( x ) = b s ( ~p ) v
s ( ~
p ) e ipx
+ a
s ( ~
p ) u
s ( ~
p ) e ipx
,
(2)3 2E p
p
s
which, as noted already, satisfy the free Dirac equation as an operator equation.
and compute
XZ d3 p 1 d3 q 1
Z
S ( x y) =
(2)3 E p (2)3 Eq
p p
s,r
{a s ( ~p), ar (~q)} u s ( ~p)ur (~q)eipx eiqy (3.159)
| {z }
sr (2)3 (3) ( ~p~q)
+ {bs ( ~p), br (~q)}v s ( ~p)vr (~q)eipx eiqy .
The identities
X X
u s ( ~p)u s ( ~p) = p + m0 and v s ( ~p)v s ( ~p) = p m0 (3.160)
s s
imply that
d3 p 1 h
Z i
S ( x y) = 3
( p + m0 )eip( xy) + ( p m0 )eip(yx) (3.161)
(2) 2E p
Here
d3 p 1 ip( xy)
Z
(0)
D ( x y) = e (3.163)
(2)3 2E p
is the propagator from the scalar theory with mass m0 . In particular this implies that S ( x y) = 0
for ( x y)2 < 0. This in turn guarantees that [O1 ( x), O2 (y)] = 0 for ( x y)2 < 0 for Oi ( x) any
local expression of fermion bilinears . Since all physical observables are bosonic this establishes
causality of the Dirac theory.
98 CHAPTER 3. QUANTISING SPIN 21 -FIELDS
Note the crucial minus sign. It is required because if ( x y)2 < 0, we have
because S ( x y) = 0 for ( x y)2 < 0. Now, for ( x y)2 < 0 the question of whether x0 y0 or
x0 < y0 depends on the Lorentz frame we have chosen. To arrive at a Lorentz frame independent
definition of the time-ordering symbol T the expression for T (( x)(y)) for x0 y0 and y0 x0
must agree.
The Feynman propagator is
while
h0| T ( x)(y) |0i = 0 = h0| T ( x)(y) |0i . (3.167)
Now by the usual tricks one evaluates
with
d4 p
Z
(0) i
D F ( x y) = eip( xy) . (3.169)
(2) p2 m20 + i
4
This gives
d4 p i( p + m0 ) ip( xy)
Z
S F ( x y) = e . (3.170)
(2)4 p2 m20 + i
We define normal-ordered products as expressions with all creation operators to the left of all anni-
hilation operators, where, unlike in the scalar theory, each exchange of two operators induces a minus
sign, e.g.
: b s ( ~p)ar (~q)bv (~k) : = (1)2 ar (~q)bv (~k)b s ( ~p) (3.174)
with
( x)(y) = h0| T ( x)(y) |0i = S F ( x y) (3.176)
and
( x)(y) = 0 = ( x)(y) . (3.177)
Direct computation confirms that Wicks theorem goes through, with the understanding that we in-
clude the minus signs from operator exchanges. For instance
T (1 2 3 ) = : 1 2 3 : + : 1 2 3 :
+ : 2 3 : . (3.178)
| 1{z }
:1 3 2 : =S f ( x1 x3 ):2 :
In particular Y Y
h0| T ( xi ) ( x j ) |0i , 0 (3.180)
i j
only for equal numbers of and fields. Physically this just reflects charge conservation.
To compute a 2n-point function of this type, we draw the corresponding Feynman diagrams, but
now
Be sure to always draw the arrow from x j to xi in order to account for the correct sign in S F ( xi x j ).
Apart from an overall sign (which is typically unimportant because we will eventually take the square
of the amplitude), the relative signs between the diagrams (which are important due to interference)
equal the number of crossing lines. For instance, for
h0| T ( x1 )( x2 )( x3 )( x4 ) |0i (3.181)
S F ( x1 x2 )S ( x3 x4 ) + (1)1 S F ( x1 x4 )S ( x3 x2 ). (3.182)
1 3 1 3
2 4 2 4
We will examine interacting spin 21 fields in great detail in the context of Quantum Electrodynamics.
Another example of an interesting interacting theory is Yukawa theory, which couples a spin 12 field
to a real boson via a cubic coupling. Its form is given in the tutorial.
In this section we only briefly summarise the logic behind the computation of scattering amplitudes
with spin 12 . As in the scalar theory, in the presence of interactions we define asymptotic in- and
out-fields satisfying the free Dirac equation with mass m , m0 , where m0 is the mass in the free Dirac
action. We then express the creation and annihilation modes by the in- and out-fields, e.g. for the
3.9. LSZ AND FEYNMAN RULES 101
in-fields
Z
1
ain,s (~q) = p d3 x u s (~q)eiqx 0 in ( x),
2Eq
Z
1
ain,s (~q) = p d3 x in ( x)0 eiqx u s (~q),
2Eq
Z (3.183)
1
bin,s (~q) = p d3 x in ( x)0 eiqx v s (~q),
2Eq
Z
1
bin,s (~q) = p d3 x v s (~q)eiqx 0 in ( x).
2Eq
Using these one can perform exactly the same LSZ reduction procedure as in the scalar field case to
extract the S -matrix. In this process we make heavy use of the equations
where now m is the fully renormalized physical mass. Consider incoming fermions |q, s, +i and anti-
fermions |q0 s0 , i and outgoing fermions hp, r, +| and anti-fermions hq0 , r0 , |. The final result for the
S-matrix element is
h...( p, r, +)...( p0 r0 , )....| S |...(q, s, +)...(q0 , s0 , )...i connected
n n0 Z Z Z Z
21 21
= (iZ ) (iZ ) d x... d x ... d y... d4 y0
4 4 0 4
(3.185)
0 0 0 0
ei(qx+q x pyp y +...) ur ( ~p)(i y m)...v s0 (~q0 )(i x0 m)...
h| T ...(y0 )...(y)...( x)...( x0 )... |i (i x m)u s (~q)...(i y m)vr0 ( ~p0 ).
Thus to compute the S-matrix we compute the Fourier transform of the amputated fully connected
associated Feynman diagram, where for each external particle we include
Alternatively, the appearance of these spinor polarisations can also be deduced by the Interaction
picture procedure discussed in the tutorials. We will exemplify this for Yukawa theory in the tutorial
and later in the course for Quantum Electrodynamics.
102 CHAPTER 3. QUANTISING SPIN 21 -FIELDS
Chapter 4
~
~ E~ = ,
~ E~ = B ,
t (4.1)
~
~ = ~j + E ,
~ B ~ B
~ = 0,
t
where the sources are subject to local charge conservation, i.e.
~ ~j = 0.
+ (4.2)
t
By virtue of the inhomogeneous Maxwell equations and Helmholtzs theorem the fields E~ and
~ can locally be expressed as
B
~
~ A ,
E~ = ~ A.
~=
B ~ (4.3)
t
~ is redundant because
This description in terms of the scalar potential and the vector potential A
E~ and B
~ are invariant under a gauge transformation, i.e. a transformation
( x) ( x) + ( x ),
t (4.4)
~ ( x) A
A ~ ( x ).
~ ( x)
To establish a Lorentz invariant formulation we introduce the 4-vector gauge potential A and
the 4-current j as
A = , j = . (4.5)
~
A ~j
103
104 CHAPTER 4. QUANTISING SPIN 1-FIELDS
F = A A , (4.6)
Using the field strength tensor the inhomogenous Maxwell equations can be written as
F = j . (4.8)
The homogenous Maxwell equations are automatically satisfied because they were used to ex-
press E~ and B
~ in terms of the potentials. Indeed they correspond to the Bianchi identity
[ F] = 0, (4.9)
where [ ] denotes all cyclic permutations. Note that j = 0 follows as a consistency condition
because F = 0 since we contract the symmetric tensor with the anti-symmetric
tensor F .
A ( x) A ( x) + ( x). (4.10)
The Maxwell equations follow as the equation of motion of A from the action
Z !
4 1
S = d x F F A ( x) j ( x) . (4.11)
4
Note that S is gauge invariant if and only if j = 0. The equation of motion can be rewritten
as
A ( A ) = j . (4.12)
A = 0. (4.13)
4.2. CANONICAL QUANTISATION OF THE FREE FIELD 105
This partially fixes the gauge in Lorenz gauge, but we are still free to perform a residual gauge
transformation
A A + with = 0 (4.14)
without violating the Lorenz gauge condition. In Lorenz gauge the equation of motion is
A = j . (4.15)
Lorenz gauge can be implemented by adding a Lagrange multiplier term in the action:
Z !
1
S = d x F F ( A) A j .
4 2
(4.16)
4 2
Therefore we now have two equations of motion, namely for A
A (1 ) ( A) = j (4.17)
and for
A = 0, (4.18)
where by equation of motion for we mean that the variation of S with respect to is propor-
tional to A.
0 = 0, (4.21)
( A , ) are no good canonical variables. Instead quantisation starts from the gauge fixed Lagrangian
1
L = F F ( A)2 , (4.22)
4 2
with
= F0 0 ( A). (4.23)
We could proceed for a general Lagrange multiplier , but for simplicity we set
1
L = A A together with A = 0. (4.25)
2
Since the Lagrangian multiplier has been integrated out by setting = 1, its equation of motion
A = 0 must now be imposed by hand as a constraint.
The extra constraint A = 0 is a consequence of the underlying gauge symmetry of the system
and will ensure that a consistent quantization is possible despite the wrong sign for = 0.
= A . (4.27)
We quantise the system by promoting A and to Heisenberg picture fields with canonical equal-
time commutators
[ A (t, ~x), (t, ~y)] = i (3) (~x ~y) (4.28)
and therefore
[ A (t, ~x), A (t, ~y)] = i (3) (~x ~y),
(4.29)
[ A (t, ~x), A (t, ~y)] = 0 = [ A (t, ~x), A (t, ~y)].
As observed above there is an odd minus sign for = = 0. Despite this issue we proceed and
consider the mode expansion
3
d3 p
Z
1 X h
i
A ( x) = ( ~
p , ) a ( ~
p ) e ipx
+ a ( ~
p ) eipx
, (4.30)
(2)3 2E p =0
p
p2 = 0 E p = |~p|. (4.31)
Furthermore the vectors ( ~p, ), = 0, 1, 2, 3 are the 4 linearly independent real polarisation vectors
whose definition depends on the value of the lightlike vector p (satisfying p2 = 0). Our conventions
to define these are as follows: Let n denote the time axis such that n2 = 1.
4.2. CANONICAL QUANTISATION OF THE FREE FIELD 107
( ~p, i) for i = 1, 2 are called transverse polarisation vectors. They are defined via ( ~p, i) n =
0 = ( ~p, i) p and
( ~p, i) ( ~p, j) = i j . (4.32)
( ~p, 3) is called longitudinal polarisation and is defined via ( ~p, 3) n = 0 = ( ~p, 3) ( ~p, i)
for i = 1, 2 and ( ~p, 3)2 = 1. Since p2 = 0 it can therefore be expressed as
p n( p n)
( ~p, 3) = . (4.33)
p+n
So altogether
( ~p, ) ( ~p, 0 ) = ,0 . (4.34)
We stress that this basis of polarisation vectors depends on the concrete momentum vector p with
p2 = 0. Consider e.g. a momentum vector p = (1, 0, 0, 1)T , then
1 0 0 0
0
1
0
0
( ~p, 0) = , ( ~p, 1) = , ( ~p, 2) = , ( ~p, 3) = . (4.35)
0 0 1 0
0 0 0 1
and
[a ( ~p), a0 ( ~p0 )] = 0 = [a ( ~p), a0 ( ~p0 )]. (4.37)
after dropping the vacuum energy. This leads to the following commutation relations (valid )
We define again the vacuum |0i such that a ( ~p) |0i = 0 and the 1-particle states
q
|~p, i := 2E p a ( ~p) |0i (4.41)
as the states of momentum ~p and polarisation . The corresponding particles are called photons.
The previous analysis seems to suggest that the theory gives rise to 4 independent degrees
of freedom per momentum eigenstate, but from classical electrodynamics we only expect 2
transverse degrees of freedom.
h~p, 0| ~q, 0i h0| [a( ~p, 0), a (~q, 0)] |0i = (2)3 (3) ( ~p ~q). (4.42)
Such negative norm states are called ghosts and spoil unitarity.
0 = [ A (t, ~x), A (t, ~y)] = A0 (t, ~x), A (t, ~y) = i0 (3) (~x ~y).
! h i
(4.43)
The idea of the Gupta-Bleuler formalism is to implement A = 0 not at the level of operators, but
directly on the Hilbert space, i.e. as a defining constraint on the so-called physical states. The naive
guess would be to require
!
A |i = 0 (4.44)
A ( x ) = A + ( x ) + A ( x ) , (4.45)
with
d3 p
Z
1 X
+
A ( x) = 3
p ( ~p, )a ( ~p)eipx ,
(2) 2|~p|
(4.46)
d3 p
Z
1 X
A ( x ) = ( ~p, )a ( ~p)eipx
(2)3 2|~p|
p
But this is still too strong because then not even the vacuum |0i would be such a physical state - after
all
A+ |0i = 0, but A |0i , 0. (4.48)
suffices to guarantee
So in the spirit of Ehrenfests theorem the classical relation A = 0 is realised as a statement about
the expectation value h| A |i = 0 in the quantum theory.
To summarise: Out of the naive Fock space we define the physical Hilbert space by
Hphys A+ |i = 0. (4.51)
It suffices to construct the physical 1-particle states of definite momentum since Hphys is spanned by
the tensor product of these. Consider a state |~p, i, i.e. 1 photon of polarisation , where we define a
general polarisation 4-vector
= 0 ( ~p, ).
X
(4.52)
,0
Therefore
h~q, | ~p, 0 i = (2)3 2|~p|(3) ( ~p ~q) 0 . (4.54)
!
The physical state condition A+ |~p, i = 0 implies that p = 0 because
p
d3 q
Z
2|~p| X
A+ |~p, i =
X
( iq ) (~q, ) eiqx
a (~q ) a ( ~p) |0i
(2)3 2|~q|
p
| {z } (4.55)
=(2)3 (3) ( ~p~q)|0i
P
= ip |0i .
Therefore
|~p, i Hphys p = 0. (4.56)
where
|~p, S i describes 1 combined timelike and longitudinal degree of freedom of zero norm, k |~p, S i k =
0.
and |~p, S i (a0 ( ~p) a3 ( ~p)) |0i. Up to now, the Gupta-Bleuler procedure has eliminated the neg-
ative norm states and left us with 3 polarisation states, but in fact one can prove the following theorem:
Such a zero-norm state that decouples from all physical processes is called spurious, hence the sub-
script S . The meaning of this decoupling of null states is as follows:
Without loss of generality we take p = (1, 0, 0, 1)T and |~p, S i (a0 ( ~p) a3 ( ~p)) |0i and
confirm h~p, S | H |~p, S i = 0 by noting the structure of the Hamiltonian
X
H ai ai a0 a0 (4.62)
i
and the relative minus sign in the commutation relations for timelike and spacelike modes.
Furthermore the spurious states decouple in the sense that |~p, S i has zero overlap with |~p, T i
because S T = 0.
The decoupling statement becomes actually non-trivial in the presence of interactions: As long
as the interactions respect gauge invariance, a spurious state |~p, S i decouples from the S -matrix
as an external (in or out) state. This follows from the Ward identities as will be discussed later.
4.3. GUPTA-BLEULER QUANTISATION 111
The conclusion is that only the 2 transverse polarisations are physically relevant as external states.
These have positive norm.
This does not mean that spurious states play no role at all:
Spurious states do appear as internal states in S -matrix processes and are important for consis-
tency of the amplitudes.
Spurious states are required to establish Lorentz-invariance of the theory because T may pick
up a spurious component S by going to a different Lorentz frame.
Let |S i denote any multi-photon state constructed entirely out of spurious photons with polarization
S . Since it decouples in the above sense, we can add and substract it without affecting any physical
properties of a state. This establishes an equivalence relation on Hphys :
|1 i |2 i if | s i : |1 i = |2 i + |S i . (4.63)
This is the analogue of the residual gauge symmetry in classical theory: Indeed, one can show that1
h s | A ( x) | s i = ( x) (4.64)
Therefore:
A A + with = 0 (4.66)
in Lorenz gauge.
Our starting point was not an arbitrary massless vector field A ( x), but the very specific gauge po-
tential that arises in Maxwell-theory. More generally we might ask: Given a general massless vector
field A ( x), how can we quantise it?
The definitive treatment of this question can be found in Weinberg, Vol. I, Chapter 8.1, which we
urgently recommend. Following this reference, the arguments are:
By Lorentz invariance alone, any massless vector field A ( x) must describe precisely two he-
licity or polarization states (the two transverse degrees of freedom we found above).
1 See Itzykson/Zuber, p.132 for a proof.
112 CHAPTER 4. QUANTISING SPIN 1-FIELDS
On general grounds one can show that Lorentz vector fields describing two polarization states
transform under a Lorentz transformation as
A ( x) A (1 x) + ( x, ) (4.67)
The Lagrangian L = 14 F F is the unique Lorentz invariant and gauge invariant Lagrangian
for a massless free vector field (i.e. up to quadratic order).
This proves the general statement:
Massless vector field theories must be gauge theories.
F + 2 A = 0. (4.69)
One can show that this Lagrangian is the unique Lorentz invariant Lagrangian for a free massive spin-
1 field (without any spin-0 components - see Weinberg I, 7.5 for a proof). The mass term explicitly
breaks gauge invariance. So it is not possible to arrange for A = 0 by gauge-fixing. However, the
Proca equation implies
0 = F +2 A A = 0. (4.70)
| {z }
=0
Thus, if m , 0, the constraint A = 0 arises classically as a consequence of the equations of motion,
not of gauge invariance. In the quantum theory it can indeed be justified to impose A = 0 as
an operator equation.2 The physical Hilbert space now exhibits 3 positive-norm degrees of freedom
corresponding to the polarisations
Since no residual gauge transformation is available, there is no further decoupling of one degree of
freedom. To summarise:
2 See Weinberg I, 7.5 for details. The main difference to the massless theory is that 0 0 now poses no problems
because, unlike in the massless case, we can solve A0 for the spacelike degrees of freedom and simply proceed with the
quantisation of ( Ai , i ). In more sophisticated terms, the system is amenable to quantisation with Dirac constraints, see
again Weinberg.
4.5. COUPLING VECTOR FIELDS TO MATTER 113
The procedure for quantising the Proca action breaks down if we set 0. A framework
where a smooth limit m 0 is possible is provided by the Stckelberg Lagrangian3
1 1
L = F F + 2 A A (A)2 . (4.72)
4 2 2
S = S 0A [ A] + S int [ A, ] + S matter
rest
[] (4.73)
that describes the coupling of this matter sector to a vector field A ( x) with free action (prior to
coupling) S 0A [ A] and interaction terms S int [ A, ]. Naively, we would think that we can simply write
down all possible Lorentz invariant terms in S int [ A, ] involving A and and then organize these
as a series in derivatives and powers of fields, as we have done when writing down the most general
interaction for a scalar field. This time, however, S int [ A, ] is subject to an important constraint: It
must be chosen such that the successful decoupling of negative norm states and zero norm states (in
the case of a massless vector) in the free vector theory is not spoiled by the interaction. More precisely,
if we denote by
M = ( k ) M (4.74)
the scattering amplitude involving an external photon of polarisation vector (k) and momentum k,
then consistency of the interaction requires - at the very least - that
k M = 0. (4.75)
3 Cf. Itzykson, Zuber, p. 136 ff. for details. Note that a priori this action does include spin-0 components in agreement
with the above claim that the Proca action is the most general action describing spin-1 degrees of freedom only.
114 CHAPTER 4. QUANTISING SPIN 1-FIELDS
This is equivalent to the requirement that external photons of polarisation (k) = k decouple from
the interactions. As we will see this is necessary and also sufficient to ensure that if we start with
a physical photon state of positive norm, no negative or zero norm states (in the case of a massless
vector theory) are produced via the interaction.
Now, one can (and we will somewhat later in this course) prove very generally the following two
important theorems:
1.) The decoupling of unphysical photon states with polarisation (k) = k , i.e. equ. (4.75), is
equivalent to the statement that
S int [ A, ]
= j (4.76)
A
for j the conserved current associated with a global continuous U (1) symmetry of the full action
S under which
( x) ( x) e ( x), R (4.77)
infinitesimally. Here we have rescaled the symmetry parameter by a coupling constant e to com-
ply with later conventions. In particular, j = 0 on-shell. This is the statement that the vector
theory must couple to a conserved current. That coupling to a conserved current is equivalent to
(4.75) is ensured by the Ward identities to be discussed in detail later in this course.
2.) If the vector theory is massless, i.e. if S 0A [ A] is invariant under the gauge symmetry
A ( x) A ( x) + ( x), (4.78)
then (4.76) is equivalent to the statement that the full action S is invariant under the combined
gauge transformation
( x) ( x) e ( x) ( x) A ( x) A ( x) + ( x), (4.79)
where in particular ( x) now transforms under a local symmetry since we have promoted the
constant appearing in (4.77) to a function ( x). This process of promoting the global continuous
symmetry (4.77) to a combined gauge symmetry as above is called gauging.
As indicated, for pedagogical reasons we postpone a proof of both these assertions and first exemplify
the consistent coupling of a massless vector theory to matter via gauging by discussing interactions
with a Dirac fermion and a complex scalar theory.
The only available vector Noether current of this action is due to the a priori global U (1) symmetry
where R is the symmetry parameter and as above we have introduced a dimensionless coupling
constant e.
With the normalisation conventions of eq. (1.37), the corresponding Noether current is
j = e and j = 0 (4.82)
on-shell for ( x). As prescribed by the above theorem we proceed by gauging this global U (1)
symmetry. This means that we promote the global U (1) symmetry to a local one, i.e. we promote the
constant R to a function ( x). Now the kinetic term is no longer invariant because
(i m) 7 eie( x) (i m0 )eie( x)
= (i ( ie ( x)) m0 ) (4.83)
= (i m0 ) + [ ( x)] j ( x).
However, we observe that the interaction term
(i m0 ) A j , j = e (4.84)
Thus the interaction is gauge invariant off-shell and fully consistent. One can rewrite the interaction
in terms of the covariant derivative
D := + ieA . (4.86)
Let us stress that global and gauge symmetries are really on very different footings:
Note furthermore that, by construction, the gauge symmetry (4.87) reduces for ( x) = = const. to
the global symmetry
eie , A A, (4.91)
from which, in turn, charge conservation follows. In particular the appearance of a combined U (1)
gauge symmetry of the system matter plus gauge fields requires an underlying global U (1) symmetry
of the matter system.
The Noether current of the free theory associated with the global U (1) symmetry eie with
R is jfree = ie ( ( )).
The naive guess for the coupling to the gauge sector would therefore be
Lnaive
int = A jfree . (4.93)
However - unlike in the fermionic case - this coupling does not exhibit off-shell gauge invariance
under
eie( x) , A A + ( x). (4.94)
Let us instead follow the general route of replacing the usual derivative by the covariant derivative
D ( x) = ( + ieA ). (4.95)
The last term quadratic in A is required for gauge invariance and was missed in the naive guess (4.93).
To see what had gone wrong in the naive guess (4.93) note that for constant gauge parameter R
the combined gauge transformation (4.94) reduces to a U (1) global symmetry of the full action (4.96).
One can check that the Noether current associated with this global U (1) symmetry of (4.96) is just
j = 2e2 A + ie ( ) = ie( D ( D ) )
(4.98)
The Feynman propagator for the gauge field in Feynman gauge ( = 1) can easily be computed
as
d4 p i ip( xy)
Z
( 0 )
h0| T A ( x) A (y) |0i = DF ( x y) = e (4.101)
m2 =0 (2)4 p2 + i
0
by plugging in the mode expansion for the quantised spin 1 field in the Heisenberg picture and
proceeding as in the scalar field case.
This is proven most easily in path-integral quantisation as will be discussed in detail in the
course QFT II.
gg
Graphically we represent the propagator of a gauge field as follows:
y x (0)
DF ( x y).
To determine the Feynman rules we must go through the LSZ analysis for gauge fields. By
arguments similar to the ones that lead to the appearance of the spinor polarisation in the spin
1/2 case one finds that external photon states |~p, i come with polarisation factors
( ~p, ) for ingoing
|~p, i
(4.103)
( ~p, ) for outgoing
This leads to the following Feynman rules of QED:
To compute the scattering amplitude iM f i defined in equ. (2.165) of a given process we draw all
relevant fully connected, amputated Feynman diagrams to the given order in the coupling constant e
and read off iM f i as follows:
gGg
(with the arrows denoting fermion number flow) and carries a factor ie ,
faf
each internal photon line carries a factor pi
2 +i ,
each internal fermion line with the arrow denoting fermion (as opposed to anti-
i(p+m0 )
fermion) number flow carries a factor p2 m20 +i
,
1
each outgoing fermion of spin s carries a factor u s ( ~p)Ze2 ,
1
each outgoing anti-fermion of spin s carries a factor vs ( ~p)Ze2 .
The overall sign of a given diagram is easiest determined directly in the interaction picture. If
2 diagrams are related by the exchange of n fermion lines, then the relative sign is (1)n . If we
are only interested in |iM f i |2 this is often enough to determine the cross-section.
We will now give some important examples of leading order QED processes:
Electron scattering
e e e e . (4.104)
To leading order in the coupling e this process receives contributions from the two following fully
connected, amputated diagrams:
q, r q0 , r0 q, r q0 , r 0
ie
t t0
ie
p, s p0 , s0 p, s p0 , s0
in the second diagram. Note the relative minus sign between both diagrams due to the crossing fermion
line!
120 CHAPTER 4. QUANTISING SPIN 1-FIELDS
Electron-positron annihilation
e+ e 2. (4.108)
p, s p0 , p, s p0 ,
ie
t t0
ie
q, r q 0 , 0 q, r q0 , 0
t = p p0 , t 0 = p q0 , q0 = p p0 + q, p0 = p q0 + q. (4.110)
The process
e + e e + e (4.111)
Compton scattering
Figure 4.4: Compton scattering at leading order.
Non-linearities
As an interesting new quantum effect, loop diagrams induce an interaction between photons. For
example, at one loop two photons scatter due to a process of the form
Figure 4.5: Photon scattering is not allowed in classical Electrodynamics, but it is in QED.
Such effects are absent in the classical theory, where light waves do not interact with each other due to
the linear structure of the classical theory. It is a fascinating phenomenon that such QED effects break
the linearity of classical optics. In high intensity laser beams, where quantum effects are quantitatively
relevant, such non-linear optics phenomena can indeed be observed.
To derive V (~x) the key idea is to consider the non-relativistic limit of the electron scattering process
e + e e + e (4.113)
p p0
k k0
This is then compared to the non-relativistic scattering process |~pi |~p0 i of two momentum eigen-
states off a potential V (~x).
In non-relativistic quantum mechanics, the scattering amplitude A(|~pi |~p0 i) for scattering of
an incoming momentum eigenstate |~pi to an outgoing momentum eigenstate |~p0 i in the presence
of a scattering potential V is computed in the interaction picture as
R Z
i dt0 VI (t0 )
0 0
A(|~pi |~p i) 1 = h~p | e |~pi 1 i h~p |0
dt0 VI (t0 ) |~pi + . . . ,
where we take the potential to be constant in time. As derived in standard textbooks on Quantum
Mechanics this equals to first order
This result goes by the name Born approximation and reads more explicitly in position space
Z
d3 r V (~r ) ei( ~p~p )~r .
0
A(|~pi |~p i) 1 = (2)( E p E p0 )(i)
0
(4.115)
In the the non-relativistic limit we have m0 |~p| and m0 |~k| and thus approximate p
(m0 , ~p) + O( ~p2 ). Then
p r
p
0 m0 r
= m0 r
p
ur ( ~p) = p
p (4.118)
p r 0 m0 r r
and
2m0 rr0
if = 0,
ur0 ( ~p ) ur ( ~p)
0
(4.119)
0 otherwise.
The factor (2m)2 is due to the different normalisation of momentum eigenstates in QFT and in
QM and must therefore be neglected in comparing the respective expressions for the amplitude.
e2
Z
d3 r V (~r ) ei( ~p~p )~r =
0
. (4.121)
|~p ~p0 |2 i
Since V is positive the potential for scattering of two electrons is repulsive as expected. If we replace
one e by e+ and consider instead the process
e + e+ e + e+ , (4.124)
However, a careful analysis of the scattering amplitude in the interaction picture shows that one picks
up one relative minus sign in the amplitude. Indeed, establishing the absolute sign of the amplitude
will be the task of Assignment 11. Thus we confirm the expected result that the potential mediated
by exchange of a spin-1 particle yields an attractive potential for oppositely charged particles and a
repulsive potential for identically charged particles.
The Coulomb potential due to exchange of a massless spin-1 particle leads to a long-range
force as the potential dies off only like 1/r. If on the other hand the exchanged spin-1 particle
is massive, the interaction is short-ranged. To see this suppose the photon has mass . All that
changes is a mass term in the phton propagator. Thus
d3 q e2 e2 r
Z
V (~r ) = ei~q~r
= e . (4.126)
(2)3 ~q2 + 2 i 4r
indeed highly relevant in particle physics and explains why in everyday physics at distances
bigger than (100 GeV)1 only the electromagnetic force can be experienced.6
The concept of forces mediated by exchange bosons is not restricted to spin-1 theories.
If we replace the vector boson of QED by a scalar boson of mass we arrive at Yukawa
theory defined by
1 2 1 2 2
L= + (i m0 ) e . (4.127)
2 2 |{z}
Yukawa int.
The Feynman rules are very similar to those in QED except for two important changes: First, the
interaction vertex carries no factor and second the scalar boson propagator must be modified
from
i i
2 . (4.128)
p + i
2 2 p 2 + i
This amounts to a crucial sign change in the non-relativistic limit because
i00 i
. (4.129)
|~p|2 2 + i |~p|2 2 + i
Combining these two changes yields the universally attractive Yukawa potential
e2 r
V (r ) = e . (4.130)
4r
Perturbative gravity can be understood at the level of field theory as a theory of massless spin-2
particles, the gravitons, whose exchange likewise yields a universally attractive force. Since the
gravitational potential is of the form 1/r (at least for conventional Einstein gravity), gravitons
must be massless.7
To summarise the potential induced by the exchange of bosons of different spin acts on fermions ( f )
and anti-fermions ( f) as follows:
ff f f f f
Spin 0 (Yukawa theory) attractive attractive attractive
Spin 1 (Gauge theory) repulsive attractive repulsive
Spin 2 (Gravity) attractive attractive attractive
6 The remaining Standard Model forces mediated by the eight massless gluons of S U (3) gauge theory is also short-
ranged, but this is because of confinement.
7 Current constraints imply that the graviton mass must be smaller than 1020 eV, see
http://arxiv.org/pdf/0809.1003v5.pdf.
126 CHAPTER 4. QUANTISING SPIN 1-FIELDS
Chapter 5
Quantum Electrodynamics
5.1 QED process at tree-level
d
As an example for a typical tree-level QED process we compute the differential cross-section d for
the scattering e e . We describe both the electron/positron and the muon/anti-muon by a
+ +
Dirac spinor field of respective free mass me and m which we couple to the U (1) gauge field.
k
p
q
p0 k0
e+ +
In this protypical example and in many similar processes one proceeds as follows:
where q = p + p0 = k + k0 . Thus
ie2
iM = v s0 ( p0 ) u s ( p)ur (k) vr0 (k0 ). (5.2)
q2
127
128 CHAPTER 5. QUANTUM ELECTRODYNAMICS
Often we do not keep track of the polarisation states of in and out states but are interested in the
unpolarised amplitude-square
1 X 1 X X X 2
iM( p, s; p0 , s0 k, r; k0 , r0 ) , (5.3)
2 s 2 s0 r r0
where 12 s 21 s0 averages over the initial state polarisation as is appropriate if the incoming beam is
P P
P P
not prepared in a polarisation eigenstate. The sum r r0 over the final state polarisations is required
in addition if the detector is blind to polarisation. With
(v s0 u s ) = us 0 v s0 = u s v s0 (5.4)
| {z }
=0
one finds
1 X 1 e2 X
0
|M|2 = ( u s v s )( v s0 u s )(vr 0 ur )(ur vr 0 ) . (5.5)
4 Spins 4 q4
s,s0 ,r,r0
u s v s0 = (u s )A A B v s0B ,
X
(5.7)
A,B
1 X 1 e2 X A
|M|2 = 4 u s ( p)D u s ( p)A B v s0 ( p0 ) B v s0 ( p0 )C C D
4 Spins 4 q s,s0 ,r,r0
vr0 (k0 )H vr0 (k0 )E EF ur (k)F ur (k)G GH
(5.8)
1 e2
= 4 tr ( p me ) ( p + me )
4q
tr ( k0 + m ) ( k m ) .
In order to proceed further we need to evaluate the traces over the matrices.
1 Recall that the indices , are completely unrelated to the spinor indices, c.f. section 3.2.
5.1. QED PROCESS AT TREE-LEVEL 129
tr = 0, because:
tr = tr 5 5 = tr5 5 = tr 5 5 = tr . (5.9)
|{z}
1
tr = 4 , because
tr5 = 0 since
tr5 = tr0 0 5 = tr0 5 0 = tr5 . (5.13)
tr5 = 4i , because the result must be antisymmetric in all indices and in partic-
ular
tr5 0 1 2 3 = i tr 5 5 = 4i. (5.15)
tr ( p me ) ( p0 + me )
h i
= 4p p0 +
(5.17)
m2e 4 + tr(odd number of s) me ...
= 4 p p0 + p p0 ( p p0 + m2e )
h i
130 CHAPTER 5. QUANTUM ELECTRODYNAMICS
and similarly
h i h i
tr ( k0 + m ) ( k m) = 4 k0 k + k0 k (k k0 + m2 ) . (5.18)
Since the ratio of m and me is about 200 we can drop me . Then altogether after a few cancellations
1 X e4 h i
|M|2 = 8 4 ( p k)( p0 k0 ) + ( p0 k)( p k0 ) + ( p p0 )m2 . (5.19)
4 Spins q
k
p
p0
k0
Figure 5.2: Centre-of-mass frame, p in z-direction.
Introducing the angle between p0 and k and taking the relativistic limit, i.e. m2e |~p|2 , yields
E E , k = E , ~k z = |~k| cos .
p = , p0 =
~ (5.20)
Ez Ez k
Therefore
q2 = ( p + p0 )2 = 2p p0 = 4E 2 ,
p k = E 2 E|~k| cos = p0 k0 , (5.21)
p k0 = E 2 + E|~k| cos = p0 k.
m2 m2 2
1 X
|M| = e 1 + 2 + 1 2 cos .
2
4
(5.22)
4 Spins E E
5.1.5 Cross-section
To this end recall the general formula for a 2-2 scattering event. For this we had derived the expression
(2)4
d = d2 (4) ( p + p0 k k0 )|M f i |2 (5.23)
4E p E p0 |v p v p0 |
5.1. QED PROCESS AT TREE-LEVEL 131
with
d3 k d3 k 0 1
d2 = 3 3
. (5.24)
(2) (2) 2Ek 2Ek0
Note that no factor of 1
2 is required here since + and are distinguishable. In the c.o.m. frame we
generally have
d|~k||~k|2 d
d2 (2)4 (4) ( p + p0 k k0 ) = ( Ecom Ek Ek0 ) (5.25)
(2)3 4Ek Ek0
q q
with Ek = |~k|2 + m2k and Ek0 = |~k|2 + m2k0 and
|~k| ~k| 1
~ |
d|k|( Ecom Ek Ek0 ) = + . (5.26)
Ek E k0
Then, altogether
d 1 |~k|
= |M f i |2 . (5.27)
d 4E p E p0 |v p v p0 | 162 Ecom
Note that, if all 4 particles had equal masses and the outstates were indistinguishable, this would
correctly reduce to the famous expression encountered earlier
d 1 1 1
= |M f i |2 (5.28)
d 2! 642 s
2 . Applied to the present case (5.27) yields, with
with s = Ecom
1
E p = E p0 = E = Ecom , |v p v p0 | = 2, (5.29)
2
the differential cross-section
d 1 1 |~k| 1 X
= |M f i |2 . (5.30)
d 2
322 Ecom Ecom 4 Spins
e2 1
:= (5.31)
4 137
leaves us with
s
m2 m2 m2 2
d 2
= 2
1 2 1 + 2 + 1 2 cos (5.32)
d 4Ecom E E E
for the differential cross-section. Thus for the total cross-section
2
r
42 m 1 m
= 1 + 2 1 +
. (5.33)
3Ecom E 2 E2
The first term reflects the purely kinematic and thus universal energy dependence, while the second
term represents the correction due to specifics of the QED interaction and is thus characteristic of the
concrete dynamics involved in this process.
132 CHAPTER 5. QUANTUM ELECTRODYNAMICS
Consider a QED amplitude M(k) involving an external photon of momentum k with k2 = 0 and
polarisation (k). We thus can write the scattering amplitude as
M ( k ) = ( k ) M ( k ) . (5.34)
k M (k) = 0. (5.35)
To appreciate its significance recall from Gupta-Bleuler quantisation that the constraint
A+ |phys i = 0 (5.36)
implies k = 0 for states |k, i Hphys . This left 2 positive norm transverse polarisations T and 1
zero norm polarisation s = k. (5.35) proves that the unphysical zero-norm polarization state |k, s i
decouples from the S -matrix as an external state, as claimed.
(5.35) follows from a more general from of the Ward identity. Consider a theory with arbitrary spin
fields a ( x) with a global continuous symmetry
a ( x) a ( x) + a ( x), R, (5.37)
j = 0 (5.38)
holds on-shell. Then in the quantum theory the general Ward-Takahashi-identity is a statement
about current conservation inside a general n-point function:
0 = h| T j ( x)a1 ( x1 )...an ( xn ) |i
n
X (5.39)
+i h| a1 ...a j ( x j )a j ( x j )(4) ( x x j )...an ( xn ) |i ,
j=1
For definiteness and simplicity consider a scalar theory. For the free theory, the classical equa-
tion of motion is
S L L
:= = (2 + m2 )( x) = 0. (5.40)
( x) ( x) ( ( x))
n
S X
h0| T ( x1 )...( xn ) |0i = i h0| T ( x1 )...( x j )(4) ( x x j )...( xn ) |0i . (5.43)
( x) j=1
L L S
!
j = ( x) ( x) (5.44)
( x) ( ( x)) ( x)
In interacting theories the reasoning goes through - all that changes is that the equations of
motion involve extra polynomial terms due to the interactions, which however do not alter
the conclusions. This yields the corresponding statements about the full correlation functions
h| ... |i.
The Schwinger-Dyson equation and the Ward-identity show that the classical equation of motion and
current conservation hold inside the correlation functions only up to so-called contact terms, which
are precisely the extra terms we pick up if the insertion point x of the operator S( x) or j ( x) coin-
cides with the insertion point x j of one of the other fields inside the correlator.
Caveat:
It is important to be aware that the n-point correlator h| ... |i involves in general divergent loop-
diagrams to be discussed soon. The Ward identity only holds if the regularisation required to define
these divergent intergrals respects the classical symmetry. We will see that for QED such regulators
can be found. On the other hand, if no regulator exists that respects the Ward-identity for a classical
134 CHAPTER 5. QUANTUM ELECTRODYNAMICS
symmetry, this symmetry is anomalous - it does not hold at the quantum level. Such anomalies will
be studied in great detail in QFT II.
Following LSZ, the scattering amplitude M(k) involving an external photon |k , (k)i and n
further particles is given by
Z
1/2
h f | ii = iZA (k) d4 x eikx 2x ... h| T A ( x) |{z}
... |i (5.46)
(n other fields)
1
L = A A A j + (i m0 ), j = e . (5.47)
2
The classical equation of motion for A is
2 A = j (5.48)
and therefore
2x h| T A ( x)... |i = h| T j ( x)... |i + contact terms, (5.49)
where the contact terms include (n 1) fields inside h| ... |i. Being correlators only of (n 1)
fields, these contact terms cannot have precisely n poles in the momenta of the n fields and thus
do not contribute to h f | ii according to the LSZ formalism.
For = k we find
Z Z
k d4 x eikx h| T j ( x)... |i = i d4 x ( eikx h| T j ( x)... |i
Z (5.51)
= i d4 x eikx h| T j ( x)... |i ,
where we used that surface terms vanish (because, as in LSZ, we are really having suitable
wave-packets in mind). According to (5.39) this is
Also in this case the contact terms do not contribute to h f | ii because we are essentially trad-
ing one A ( x) field against one matter field ( x) inside the correlator so that the resulting pole
structure is not of the right form, in and (5.35) is proven.
5.2. THE WARD-TAKAHASHI IDENTITY 135
S = S 0A [ A] + S int [ A, ] + S matter
rest
[] (5.53)
with
S int [ A, ]
= j (5.54)
A
the Noether current associated with a global continuous symmetry of the full interacting action S .
For a massless vector theory, which, as a free theory, must always be gauge invariant, (5.54)
is in fact equivalent to invariance of the theory under combined gauge transformations of the
vector and the matter sector.
To prove that (5.54) implies gauge invariance, our point of departure is the existence of a global
continuous symmetry ( x) ( x) + ( x) with R constant, which leaves the full ac-
tion S invariant. If we perform instead a local transformation ( x) ( x) + ( x)( x) with
varying ( x), then S is in general no longer zero. Since it vanishes for constant , S must
be proportional (to first order) to ( x). In fact, Lorentz invariance implies that there ex-
ists a 4-vector j ( x) such that S = j ( x) = ( j ) ( x). This identifies j with
R R
j , the conserved current (because for constant, the Noether current has the property that
S int
( j ) = ( S ) S - see equ. (5.44)). Now, since by assumption A = j we
R R
R S
have ( Aint + j ) ( x) = 0. Writing S = S 0A [ A] + S int [ A, ] + S matter
rest [], the fact that S 0 [ A]
A
is gauge invariant for a massless vector theory implies that this is just the variation of S with
respect to the combined gauge transformation
To prove the other direction, suppose (5.55) leaves the action S 0A [ A] + S int [ A, ] + S matter
rest []
invariant (where again S 0A [ A] is separately gauge invariant). Then there exists some conserved
j ( x) such that ( S
A + j ) ( x) = 0, and we conclude (5.54).
R
int
To conclude, massless vector theories are consistent if and only if the vector couples to a
conserved current. This is the necessary and consistent condition for gauge invariance and for
the Ward identities.
If we couple a massive vector, whose action is never gauge invariant, to a conserved current as
in (5.54), then k M = 0 still holds. The fact that in a massive vector theory the Ward identities
are still satisfied provided that theory couples as in (5.54) is crucial for its consistency: Recall
that in a massive vector theory (k2 > 0), the negative norm states are the states
|k , i with = k . (5.56)
136 CHAPTER 5. QUANTUM ELECTRODYNAMICS
Restriction to with k = 0 removes these. This is only justified in the interacting theory
as long as no dangerous states |k , = k i are produced, as is guaranteed thanks to k M = 0
for couplings to conserved currents.
To evaluate the consequences of the Ward identities we can take, without loss of generality, the photon
4-momentum to be
1
0
k = k (5.57)
0
1
and consider the corresponding basis of polarisation vectors
1 0 0 0
0
1
0
0
(k, 0) = , (k, 1) = , (k, 2) = , (k, 3) = . (5.58)
0 0 1 0
0 0 0 1
The Ward identity k M (k) = 0 then implies M0 (k) = M3 (k). In physical applications we often
need to sum over the two transverse polarisation vectors = 1, 2 of an external photon involved in a
scattering process. By the Ward identity this becomes
2 2
| (k, )M (k)|2 = (k, ) (k, )M (k)(M (k)) = |M1 (k)|2 + |M2 (k)|2
X X
=1 =1
= |M (k)|2 + |M2 (k)|2 + |M3 (k)|2 |M0 (k)|2
1
Thus in sums over polarisations of the above type we can replace 2=1 (k, ) (k, ) by . This
P
will be used heavily in deriving the Klein-Nishina formula for Compton scattering in the tutorials.
So far we have merely shown that the zero-norm states with polarization k decouple from the S-
matrix and are thus not produced in scattering experiments. But what about the status of timelike
polarisation states, which correspond to the even more dangerous negative norm ghosts? Even if we
declare Hphys not to contain them, we must prove that no such states are created as outgoing states
from physical in-states in scattering processes. Otherwise the interactions would render the theory
inconsistent.2
2 Note that this is only a question for massless vector fields. For massive vector fields the negative norm states are just
the photons with polarization k (as k2 > 0) and these obviously decouple by the Ward identity.
5.3. RADIATIVE CORRECTIONS IN QED - OVERVIEW 137
In fact, it is again the Ward identity that guarantees that no ghosts are created as external states in
interactions. The argument relies on the relation
2
(k, ) (k, )M (k)(M (k)) = M (k)(M (k))
X
(5.60)
=1
derived above and is sketched as follows: It suffices to show that the restriction of the S-matrix S to
the space of transverse polarisations is unitary. This means that we lose no information by considering
only transverse in and out-states and thus guarantees that no unphysical states can be produced out
of transverse incoming states. Let P denote the projector onto these states of transverse polarisations.
The relation (5.60) amounts to the statement
S PS = S S . (5.61)
Since S S = 1, this implies that ( PS P) ( PS P) = P. This is precisely the statement that the restric-
tion of S to the state of transverse polarisation is unitary.
d4 p ip( xy) p + m0
Z
S F ( x y) = e . (5.62)
(2)4 p2 m20 + i
| {z }
pmi
0 +i
fafpf
f f|af f|aff|af
Taking into account corrections due to QED interactions we have instead
h| T ( x)(y) |i = y x+ + +...
(5.63)
y x.
f f
Let
A 1PI
B i(6 p)AB (5.64)
fpf
denote the amputated 1PI diagram. By Dyson resummation the full propagator then takes the form
d4 p ip( xy)
Z
i
y x= e , (5.65)
4 p m0 (6 p) + i
P
(2)
P
where (6 p) is called self-energy of the electron.
138 CHAPTER 5. QUANTUM ELECTRODYNAMICS
gpg
Corrections to the photon propagator
The Fourier transform of the photon propagator, denoted by , derives by Dyson resum-
g g
mation from the 1-PI diagram
1PI
i (q2 ) = self-energy of the photon or vacuum polarisation. (5.66)
Loop corrections modifiy the cubic interaction vertex. We will find it useful to define an effective
vertex by summing up all loop-corrections. Schematically,
p p0 + +... = p p0 ' ie ( p, p0 ).
We will compute these corrections to 1-loop order. The diagrams will exhibit
ultraviolet (UV) divergences from integrating the momenta of particles in the loop up to infinity
and
infra-red (IR) divergences if the diagram contains massless particles - i.e. photons - running in
the loop.
IR divergences in loop-diagrams cancel against IR divergences from radiation of soft, i.e. low-
energy photons and thus pose no conceptual problem.
UV divergences require regularisation of the integral and can be absorbed in a clever definition
of the parameters via renormalisation.
faf f|af
At 1-loop order the electron propagator takes the form
h| T ( x)(y) |i = y x + y x (5.67)
| {z }
d4 p ip( xy)
R
= e (I)
(2)4
f|af
where the integrand (I) is given by the diagram
k
(I) p p. (5.68)
5.4. SELF-ENERGY OF THE ELECTRON AT 1-LOOP 139
The photon in the loop carries momentum p k. By means of the Feynman rules,
i ( p + m0 ) i( p + m0 )
(I) = 2
( i 2 (6 p )) . (5.69)
p2 m0 + i p2 m20 + i
The amputated 1-loop contribution corresponds to omitting the two outer fermion propagators and
is thus given by
d4 k i(6 k +mo ) i
Z
i2 (6 p) = (ie) 2
4
2 2
. (5.70)
(2) k m0 + i ( p k)2 + i
It will turn out that the integral is divergent near k = 0 if p 0. This is an IR divergence. A
careful analysis reveals that it will cancel in all amplitudes against similar such IR divergences from
other diagrams involving soft photons and thus poses no harm. For the present discussion we could
just ignore it, but for completeness we introduce a ficticious small photon mass to regulate the IR
divergence:
d4 k i(6 k +mo )
Z
i
i2 (6 p) = (ie) 2
2 . (5.71)
(2) 4 k m0 + i ( p k) 2 + i
2 2
x is called a Feynman parameter. Applying this identity yields in the present case
Z1
1 1
= dx
AB ( x(( p k)2 2 + i ) + (1 x)(k2 m20 + i ))2
0
Z1
1
= dx (5.73)
(k 2 2xk p + xp2 x2 (1 x)m20 + i + x2 p2 x2 p2 )2
0
Z1
1
= dx ,
( l2 + i )2
0
Z1
d4 l (6 k +m0 )
Z
2
i2 (6 p) = e dx . (5.74)
(2)4 (l2 + i )2
0
140 CHAPTER 5. QUANTUM ELECTRODYNAMICS
The term in the numerator can be simplified with the help of the gamma-matrix identity
k = (2 )k = k (2 ) (5.75)
Remark: For more general loop integrals one makes use of the identity
Z1
1 X (n 1) !
= dx1 ...dxn ( xi l) , (5.79)
A1 ...An i
( x1 A1 + ... + xn An )n
0
d4 l d4 l E
Z Z
1 1
= i(1)m . (5.81)
(2)4 (l2 + i )m (2)4 (l2E + i )m
Since we wont need the i any longer we can omit it at this stage. We can now peform the integral as
a spherical integral in R4 .
5.4. SELF-ENERGY OF THE ELECTRON AT 1-LOOP 141
Momentum cutoff: Isolate the UV divergence as a divergence in the upper momentum limit
by writing
Z Z
. . . = lim .... (5.83)
0 o
Following this procedure we can evaluate the integral in an elementary fashion as
Z
|lE |3
Z
d|lE |
I4 = d4 lim
(2)4 (|lE |2 + )2
0
Z
d|lE |2 |lE |2
Z
1
= d4 lim
2 (2)4 (|lE |2 + )2 (5.84)
0
#2
Z "
1 1
= d4 lim + log(|lE |2 + )
2 (2) 4 + |lE |2 l2 = 0
2
Z " ! #
1 1
= d4 lim log 1 .
2 (2)4
The loop leads to a log-divergence as . The problem is that this regularisation procedure
is not consistent with the Ward identities, as we will see when computing the photon propagator,
and is therefore not a useful regularisation method in QED.
Dimensional regularisation (dimReg) is probably the most common method. We first evaluate
the integral in d dimensions; writing d = 4 then isolates the divergence as a pole in as
0.
Let us therefore consider
Z
dd l E |lE |d1
Z
1 dd
Id = = d|lE | . (5.85)
(2)d (l2E + )2 (2)3 (|lE |2 + )2
0
and substitute
x= |lE |2 = (5.90)
|lE +
|2 x
with
dx = d|lE |2 . (5.91)
(|lE |2 + )2
Then the integral (5.89) becomes
!2d/2 Z1
1 1
= dx x1d/2 (1 x)d/21 . (5.92)
2
0
where the last identity is non-trivial and proven again in textbooks on complex analysis.
We can rewrite (5.92) in terms of B(, ) with = 2 d/2 and = d/2. Together with
(2) = 1 we arrive at
!2d/2 Z1 !2d/2
1 1 1 1
dx x 1d/2
(1 x ) d/21
= (2 d/2)(d/2). (5.94)
2 2
0
5.4. SELF-ENERGY OF THE ELECTRON AT 1-LOOP 143
Therefore
!2d/2
dd lE 1 1 1
Id = = (2 d/2) . (5.95)
(2) (lE + )
d 2 2 (4) d/2
We now apply this to Id with d = 4 and note that we encounter a singularity from (0). To
isolate the singularity we define
d = 4 (5.96)
and would like to write I4 lim0 I4 . However I4 is dimensionless because
whereas I4 has dimension [mass] . If we introduce a compensating mass scale M, we can
write
!/2
1 1
I4 = lim M I4 = lim M (/2)
0 42/2
!/2 (5.98)
1 4 M 2
= lim 2 (/2) .
0 4
Finally we use
where M 2 = 4e M 2 .
The final result for the electron self-energy at 1-loop is
Z1 !
2 M2
i2 (6 p) = (i) dx ((2 /2)m0 (1 /2) x 6 p) + log , (5.101)
2
0
0
where M is some as yet arbitrary mass-scale. We will understand its significance soon.
f|af f|af
the loop - here a photon - of mass , i.e. we compute
k k
p p . (5.102)
| {z } | {z }
i i
involves involves
( pk)2 2 +i ( pk)2 2 +i
144 CHAPTER 5. QUANTUM ELECTRODYNAMICS
Subtracting the second diagram subtracts the divergence because for infinite momenta the mass
of the particle in the loop becomes irrelevant and both diagrams asymptote to the same divergent
value. If the contribution of the subtracted vanishes and we recover the actual diagram.
The divergence reappears therefore as a divergence in as .
Subtracting both diagrams yields
i2 (6 p) = lim (computation with computation with )
= ...
(5.103)
Z1
x2
!
i2 (6 p) = lim (i) dx(2m0 x 6 p) log ,
2
0
where the omitted steps are elementary and similar to the integration performed in (5.84). We
can rewrite this by introducing again an arbitrary mass scale M as
Z1 !!
x2 M 2
!
i2 (6 p) = lim (i) dx(2m0 x 6 p) log + log (5.104)
2 M2
0
Comparing (5.101) and (5.104) we note that the momentum-dependent terms are in both regu-
larisation approaches the same, while the 1 -divergence in dimReg corresponds to a logarithmic
divergence in PV or in momentum cutoff regularisation.
f (6 p) :=6 p m0 1 (6 p) (5.107)
around 6 p 0 = m 1. Then
d f
f (6 p) = f (6 p 0 ) + (6 p 6 p 0 ) + O (6 p 6 p 0 )2
d 6 p
6 p0
! (5.108)
d(6 p)
= 0 + (6 p m 1) 1 + O (6 p m)2 .
d6 p
6 p=m1
5.5. BARE MASS M0 VERSUS PHYSICAL MASS M 145
Therefore
" # !1
1 d(6 p)
res = 1 =: Z2 (5.109)
6 p m0 (6 p)
d6 p
6 p=m 6 p=m
and thus
d(6 p)
Z21 = 1 . (5.110)
d 6 p
6 p=m
(6 p) = 2 (6 p) + O(2 ). (5.111)
Then to order
m m0 2 (6 p= m) = 0 (5.112)
and thus
m := m mo = 2 (6 p= m) + O(2 ). (5.113)
Note that if we trade m by m0 in 2 (6 p= m), the error will be O(2 ) and thus we can also write
m := m mo = 2 (6 p= m0 ) + O(2 ). (5.114)
In dimReg we find
Z1 2
M 2
m m0 = dx (2 x) + ( x 1) + log
m0 (5.115)
2 2 2 2 2
(1 x) m0 + x
0
and in PV
Z1
2 2
!
x M
m m0 = m0 dx(2 x) log + log . (5.116)
2 M2 (1 x)2 m20 + x2
0 | {z }
!
x2
log
(1x)2 m2 + x2
0
Note that m m0 > 0: This is because what we are computing is literally the self-energy of the
electron, i.e. the mass shift due to the (positive!) energy stored in its own electric field. Recall from
classical electrodynamics that also classically this quantity is divergent due to the pointlike structure
of the electron. Not surprisingly, the divergence remains in QED.3
Likewise we can compute Z2 to order by evaluating (5.110) at order , the result being
Z1
x2 x(1 x)m2
" ! #
Z2 = 1 + dx x log + 2(2 x ) + O(2 ). (5.117)
2 (1 x)2 m2 + x2 (1 x)2 m2 + x2
0
3 Indeed, the fundamental reason for this divergence is because in QFT the fundamental objects are still pointlike, i.e.
they have no substructure. In string theory, on the other hand, the fundamental objects do have an intrinsic substructure (as
1-dimensional strings instead of points) and correspondingly this theory is free of UV divergences.
146 CHAPTER 5. QUANTUM ELECTRODYNAMICS
Only m, the physical mass, is a physical observable. Namely m is the rest mass of an electron
as measured in experiments.
By contrast, m0 , the so-called bare mass, is merely a parameter that appears in the Lagrangian
and per se cannot be measured directly. Rather the Lagrangian produces for us, via the Feynman
rules, measurable quantities, the scattering amplitudes, which depend on m0 .
This suggests the following solution to the divergence of m: The divergence can be absorbed in the
definition of m0 by interpreting the equation (5.113) for m as an equation for m0 in terms of the
measured physical mass m and the cutoff or . Concretely in PV regularisation (to be specific)
Z1
x 2
+ O(2 ) = m0 ().
m0 = m 1 dx(2 x) log
(5.118)
2 (1 x)2 m20 + x2
0
This means that we take the parameter m0 in the Lagrangian to be divergent. We compute scattering
amplitudes etc. in terms of this divergent object m0 () and at the end plug in the above equation
for m0 () to express everything via m. If through this procedure all physical quantities in the end
are independent of , the theory is said to be renormalisable. Indeed QED is renormalisable.We will
discuss how one can see this more systematically later.
Rather than being systematic here, let us exemplify this in the following trivial example: Compute
the mass m at 1-loop. Well, this means we take the equation for m in terms of m0 at 1-loop and plug
in (5.118). Obviously we find m = m. This demonstrates that absorbing the divergence in the bare
mass comes at a price: We lose predicitivity for the physical mass. In other words, we must now take
m directly from experiment. In a renormalisable theory we retain, however, predicitivity for all but a
finite number of quantities.
g g
The building block to compute radiative corrections to the photon propagator is the 1PI amputated
diagram
1PI
:= i (q). (5.119)
q q
q (q) = 0, (5.120)
5.6. THE PHOTON PROPAGATOR 147
Finally, i (q) cannot have an analyic pole at q2 = 0,5 because this would require a single-
particle massless intermediate state (whose propagator vanishes at zero momentum), but no
such intermediate states occur for the 1 PI diagram relevant for the photon propagator.
Figure 5.3: Possible diagrams: The first loop carries only massive particles. In the second diagram the photon in the loop is
massless, but it does not arise as a single-particle due to the accompanying electron line.
Therefore,
i (q) = q2 q q (q2 ),
(5.122)
such that (q2 ) is regular at q2 = 0. It is useful to define the projection operator onto momenta
orthogonal to q ,
q q
P (q) = 2 (5.123)
q
with
q P (q) = 0 and P (q) P (q) = P (q). (5.124)
gpg gg g g g g g
By Dyson resummation the Fourier transform of the full propagator takes the form
=
i
+
1PI
q q
!
+ 1PI
i q q
1PI
+ ...
(5.125)
= 2 2 + 2 2 ,
q (1 (q ))
2 q q q
where the final result follows with the help of (5.124) as will be discussed in the tutorial. Note that
gpg
this holds in Feynman gauge = 1. In general gauge we would get6
q q i q q
!
i
= 2 + . (5.126)
q (1 (q2 )) q2 q2 q2
4 In fact, this argument applies to the fully resummed propagator rather than to (q). However, given the relation
between both via Dyson resummation it is not hard to see that the Ward identity carries over to q (q) = 0 because it
must hold order by order in the coupling constant.
5 To avoid confusion, think of computing the diagram for finite cut-off or > 0. The statement is that apart from the
This identifies the -dependent term as pure gauge. We can omit it by going to Landau gauge = 0,
gpg
in which
!
i q q
2 .
= 2 (5.127)
q (1 (q2 )) q
Landau
Since (q2 ) is regular at q2 = 0 (again in the sense that for finite cutoff, there is no analytic
pole) the pole at q2 = 0 is unaffected by the radiative corrections. Thus the photon remains
massless as required by gauge invariance. This is a consequence of the Ward identity, which
had lead to this form of i (q).
1
Z3 = . (5.128)
1 (0)
To obtain the result at order we procced in a similar manner as for 2 (6 p). We denote by 2 (q2 ) the
1-loop contribution to (q2 ) (just like 2 (6 p) denotes the 1-loop contribution to (6 p)). To compute
gcg
2 (q2 ) we must consider the diagram
(5.129)
q q
and bring the result into the form (5.121) with 2 (q2 ) instead of (q2 ).
This computation is most conveniently carried out in dimensional regularization. We merely quote
the final result
Z1 !
2
2 2 M
2 ( q2 ) = dx x(1 x) + log , (5.130)
0 0
This would violate q (q) = 0 and thus the Ward identity. This shows, as claimed, that momen-
tum cutoff regularisation breaks gauge invariance and is thus not useful in QED. Note that from our
derivation of the Ward identities, it may not be completely obvious why cutoff regularization leads to
a breakdown of the Ward identities. Next term we will get to know a very simple derivation of the
Ward identities in the path integral quantization approach and interpret the breakdown of the Ward
identities as due to the fact that the cutoff-regularised measure of the path integral is not invariant
under the U(1) symmetry of the classical action.
5.7. THE RUNNING COUPLING 149
q q
Taking into account the appearance of e0 at each vertex, it is clear that the amplitude involves a factor
of
(ie0 )2 i( q q /q2 )
. (5.132)
1 (q2 ) q2
We can absorb the correction term (1 (q2 ))1 into the coupling and define an effective coupling
e0
e ( q2 ) : = p . (5.133)
1 (q2 )
This obviously depends on q2 , the energy transferred by the photon in the scattering process. When
measuring the physical charge of an electron, we therefore need to specify the energy scale at which
the measurement is performed. Let us define the physical coupling or renormalised charge as the
effective charge as measured at q2 = 0, i.e.
e0 p
e := lim e(q2 ) = p = e0 Z3 . (5.134)
2
q 0 1 (0)
Note that (0) is a divergent constant. E.g. at 1-loop in perturbation theory it takes the form in
dimensional regularisation
Z1
2
2 2 M
(0) = 2 (0) + O( ) =
2
dx x(1 x) + log 2 + O(2 ).
(5.135)
m0
0
Note that at this order in , m0 and m can be exchanged (because the difference is itself of O()).
We proceed as we did when defining the physical electron mass and absorb the divergence into the
definition of the bare coupling e0 . This can be done because only e is a physical observable, whose
finite value we take from experiment. We define the bare coupling as
q
e0 = e 1 (0) e0 () or e0 ( ), (5.136)
where 1 2 (0) = 1
1+ 2 (0)
+ O(2 ). Thus
e2
e2 ( q2 ) = + O(2 ), (5.138)
1 (2 (q2 ) 2 (0))
where e2 e2 (q2 = 0) is the physical charge at q2 = 0. We conclude that the effective coupling at
q2 , 0 is
e2
e2 (q2 ) = + O ( 2 ) , (5.139)
1 2 (q )
2
Z1
m2
!
2
2 (q ) =
2
dx x(1 x) log 2 . (5.140)
m x(1 x)q2
0
The effective coupling of QED increases logarithmically with the energy scale at which the experiment
is performed. The physical interpretation of this is that the charge of an electron is screened by virtual
e+ e pairs so the effective charge decreases at long distance corresponding to small energy. Thus the
QED vacuum appears like a polarisable medium. This explains the name vacuum polarisation for
the radiatively corrected photon propagator.
Note that from this analysis, the effective coupling increases indefinitely as we increase the energy.
This phenomenon is called the Landau pole of QED and casts doubt on the validity of the theory at
arbitrarily high energies. We will find in QFT II that suitable non-abelian gauge theories such as QCD
exhibit precisely the opposite phenomenon: they become asymptotically free and are thus perfectly
well-defined in the UV. One proposed solution to the Landau pole problem of QED is therefore that
the electromagnetic gauge group, which is part of the Standard Model, might in fact emerge as part of
a unified non-abelian gauge group, whose dynamics takes over at high energies.
p0 q
ieo ( p0 , p)
p
amputated
For later purposes we point out that in diagrams with external fermions this resummed vertex is
dressed with suitable factors of Z2 . For instance the amplitude associated with the diagram
5.8. THE RESUMMED QED VERTEX
p0
p
q q
k0
k
151
is of the form
i( q q /q2 )
iM u( p0 ) (ie0 Z2 ( p0 , p)) u( p) u(k0 ) (ie0 Z2 (k0 , k)) u(k). (5.141)
q2 (1 (q2 )
p0 p0
q q
+ + O(2 )
p p
p
q u( p0 ) ( p0 , p)u( p) = 0, (5.142)
S q
( p0 , p ) = F 1 ( q2 ) + i F 2 ( q2 ) , (5.143)
2m0
where F1 (q2 ) and F2 (q2 ) are called form factors and S = 4i [ , ]. The form factors can be
computed perturbatively via loop-integrals. We merely quote the following results:
Z1
F1 (q ) =
2
dx dy dz ( x + y + z 1)
2
0 (5.145)
z2
" #
1 2 2 2
log + (1 x)(1 y)q + (1 4z + z )m0 ,
with = xyq2 + (1 z)2 m20 + 2 z. Here shows up to regulate the IR divergences that will
cancel eventually.
152 CHAPTER 5. QUANTUM ELECTRODYNAMICS
For F2 (q2 ) = 0 + F2 (q2 ) + O(2 ) one finds in PV regularisation
Z1
2m20 z(1 z)
F2 (q ) =
2
dx dy dz ( x + y + z 1) , (5.146)
2
0
which is finite. This finiteness of F2 (q2 ) persists to all orders in perturbation theory.
In the limit p0 p, i.e. q 0, we expect the loop corrections to vanish, i.e. as far as the vertex
corrections are concerned e.g.
p0
k0 k0
p0
p k
k p
Note the appearance of the factor Z2 due to the external fermion legs, as pointed out before. We can
prove (5.147) in two ways:
S q
!
Z2 ( p + q, p) = (1 + Z2 + O(2 )) 1 + F1 (q2 + O(2 ) + i
F2 (q2 )
2m0
(5.148)
i
= (1 + Z2 + F1 (q ) + O( )) +
2 2 2
S q F 2 ( q ) ,
2m0
F1 (0) = Z2 . (5.149)
In fact one can prove (5.147) non-perturbatively via the Ward identities. One defines a quantity
Z1 by
lim ( p + q, p) = Z11 . (5.150)
q0
As we will prove on Assignment 12, the Ward identities imply that Z1 = Z2 and thus
lim Z2 ( p + q, p) = . (5.151)
q0
In turn this proves that the relation (5.149) must hold order by order in perturbation theory.
5.8. THE RESUMMED QED VERTEX 153
~ (~x), e ~
V (~x) = ~ B ~ = g S, (5.153)
2m
where ~ represents the magnetic moment of the electron and S~ denotes the quantum mechanical spin
operator. The Land factor comes out as
The value g = 2 follows already from relativistic Quantum Mechanics, in which the Dirac equation is
interpreted as an equation for the wavefunction of the electron. Crucially, F2 (0) yields QED loop cor-
rections to g = 2. These can be computed order by order in perturbation theory and are in impressive
agreement with experiment.
154 CHAPTER 5. QUANTUM ELECTRODYNAMICS
For a systematic treatment of UV divergent diagrams, it suffices to consider all amputated, 1PI UV-
divergent diagrams. All divergent diagrams in a QFT are given either by these diagrams or possibly
by diagrams containing these as subdiagrams.
In QFT II we will find a simple way to classify the divergent 1PI amputated diagrams in a given QFT.
Applied to QED, this classification will prove that in QED the UV divergent 1PI amputated diagrams
are precisely the three types of diagrams which we have studied in the previous sections:
p+q
g g faf faf
1PI
,
1PI
,
p
1PI
q (5.155)
q q
!
i (q ) = i 2 2 (q2 ).
2
(5.156)
q
(1)
2 (q2 ) = c0 log + finite O(q2 ). (5.157)
M
That is, the UV divergence appears at order (q2 )0 and is characterized by a constant coefficient
(1)
c0 independent of q2 .
The remaining two diagrams at 1-loop order have the following structure:
p p
(1) (1)
i2 (6 p) = ao mo log M + a1 6 p log M + finite terms
and
q
(1)
p0 i2 ( p0 , p) = b0 log M + finite
Therefore the UV divergences are specified, at 1-loop order, by altogether 4 divergent constants.
5.9. RENORMALISED PERTURBATION THEORY OF QED 155
In QFT II we will argue that this structure persists to all orders in perturbation theory, i.e.
i(6 p) = a0 m0 log + a1 6 p log + finite O(6 p 2 )
M M
(1) (2) (1) (2)
a0 = a0 + a0 +... a1 = a1 + a1 + . . .
|{z} |{z}
O() O(2 )
(q2 ) = c0 log + finite O(q2 ) + ... (5.158)
M
(1) (2)
c0 = c0 + c0 + ...
i2 ( p0 , p) =b0 log + finite O(( p0 p)2 )
M
(1) (2)
b0 = b0 + b0 + ...
Thus at each order in perturbation theory, we encounter 4 constants that multiply UV divergent terms
in the above 1PI diagrams. One can absorb these 4 divergent constants order by order in perturbation
theory by a procedure called renormalisation.
There are two different, but equivalent ways to perform this procedure, which we now discuss in the
context of QED.
We compute the above three 1PI amputated UV divergent amplitudes to a given order in pertur-
bation theory as functions of e0 and m0 and a cutoff or . We keep the cutoff finite (or
non-zero) for the time being so that all computations are perfectly well-defined.
and also
Z2 = Z2 (m0 , e0 , ), the wavefunction renormalisation of the electron
(5.161)
Z3 = Z3 (m0 , e0 , ), the wavefunction renormalisation of A .
The renormalisation step amounts to interpreting (5.160) as an equation for the bare mass m0
and the bare coupling e0 in terms of the finite physical quantities m and e, i.e. we write
Thus 2 linear combinations of the 4 UV divergent constants are now contained in m0 and e0 .
We plug the expression for e0 and m0 back into L. The resulting renormalized Lagrangian is
now cuttoff dependent,
L = L(e0 (m, e, ), m0 (m, e, )) (5.163)
and in particular divergent if we take . This is not a problem because the Lagrangian
per se has no physical meaning - only physical observables computed from L must be finite to
make sense.
The remaining 2 linear combinations of divergent constants which are not contained in m0 and
e0 are contained in the expressions for Z2 , Z3 . Even though these do not appear in L in the
present formulation, they enter the computation of scattering amplitudes via the Feynman rules.
We now compute a given observable from L in perturbation theory to given order as functions
of e0 and m0 , Z2 , Z3 and . The UV divergent terms in cancel in all final expressions and all
observables are finite expressions of m and e plus terms in which vanish as we take 0.
For this to work it is crucial that we compute the observable to the same order in perturbation
theory to which we have computed (5.160) and (5.161). E.g. if we evaluate (5.160) and (5.161)
at 1-loop order, then we must compute all remaining scattering amplitudes to 1-loop order as
well.
The cancellation of works because in all amplitudes we have just the correct numbers of Z2 , Z3 etc.
so that all divergences drop out. Rather than give a general proof we demonstrate this for one of the
diagrams contributing to Compton scattering, including radiative corrections:
Z21/2 Z21/2
p p
Z31/2 Z31/2
1 i
(Z31/2 ie0 Z2 ) (Z 1/2 ie0 Z2 ). (5.164)
Z2 6 p m0 (6 p) 3
e0 Z31/2 = e is finite.
5.9. RENORMALISED PERTURBATION THEORY OF QED 157
( p + q, p)Z2 is also finite. To see this recall that divergence in ( p + q, p) arises as the
q2 -independent term as parametrized in (5.158) and is thus already contained in limq2 0 ( p +
q, p). Since Z2 is independent of q2 it thus surfices to consider limq2 0 ( p + q, p)Z2 . But as
discussed around equ. (5.151) this is finite to all orders in perturbation theory by means of the
Ward identities.
Explicitly, this can be confirmed perturbatively from
S q
( p + q, p) = F1 (q2 ) + i F2 (q2 ). (5.165)
2m0
Concerning the second term, F2 (q2 ) is finite as function of m0 and e0 . To a given order in
perturbation theory we can replace e0 by e and m0 by m as the difference is relevant only at the
next order and thus the second term is finite, order by order in perturbation theory. Concerning
the first term,
F1 (q2 ) = 1 + F1 (q2 ) + O(2 ) (5.166)
with
F1 (q2 ) = F1 (0) + f (q2 ), (5.167)
where F1 (0) carries all UV divergences, while f (q2 ) is finite as a function of m0 . Furthermore
we have
Z2 = 1 + Z2 + O(2 ) (5.168)
and
Z2 ( p + q, p) = (1 + Z2 + F1 (0) + f (q2 ) + O(2 ) + ...finite). (5.169)
| {z }
=0
Therefore the divergence has cancelled out, because F1 (0) = Z2 . This persists to all orders.
Crucially, the terms analytic at 6 p= m do not contain any divergence in the cutoff. This follows
from the Taylor expansion (5.108), according to which these terms are given by the second
derivative of (6 p), together with the fact that in i(6 p) all terms quadratic in 6 p and higher are
UV finite. The Z2 factors in the first term cancel and the entire expression is finite.
Finally let us outline the generalization and consequences of this renormalisation procedure:
Consider a general QFT. The theory is called renormalisable if only a finite number of re-
summed amputated 1PI diagrams is UV divergent.
158 CHAPTER 5. QUANTUM ELECTRODYNAMICS
Suppose the renormalisable QFT contains m different fields and suppose that the UV divergent
1PI diagrams give rise to n divergent constants order by order in perturbation theory. Then (n
m) of these constants can be absorbed in the definition of (n m) unphysical parameters, the
so-called bare couplings. This procedure requires specifying the outcome of (n m) physical
observables as external input. The remaining m constants can be absorbed in the definition of
the kinetic terms of the m fields without reducing the predictability of the theory further.
However, the price to pay for the appearance of the UV divergences in the first place is that
the (n m) observables cannot be computed by the theory even in principle!For instance, QED
cannot make any prediction whatsoever for the absolute value of the electron mass or the charge
at q2 = 0.7 As a result the renormalized QFT necessarily contains free parameters that must be
fitted to experiment. Another example for such an observable for which no prediction can be
made in a QFT with divergent partition function is the vacuum energy (cosmological constant).
From a modern and widely accepted point of view (introduced by K. Wilson), a non-UV finite,
but renormalisable QFT is an effective theory: The UV divergences hint at a breakdown of
the theory at high energies, where it does not describe the microscopic degrees of freedom
correctly. Renormalisation hides our ignorance about the true physics at high energies in
the (n m) observables and we can fit the theory to experiment as one typically does with
a phenomenological model.
If we want to go beyond this and describe a truly fundamental (as opposed to effective) theory,
we need a theory that is UV finite even before renormalisation. As of this writing the only
known theory with this property which also includes gravity is string theory. It has indeed no
free parameters.
1
L = F 2 + (i 6 m0 ) e0 A . (5.171)
4
7 However, once we take e(q2 = 0) from experiment, QED does predict the logarithmic running of the effective charge
as a function of q2 .
5.9. RENORMALISED PERTURBATION THEORY OF QED 159
fpf
= + analytic terms,
q2
(5.172)
iZ2
= + analytic terms.
6 p m
We can absorb Z2 and Z3 into the fields by renormalising the field strengths as
A =: Z31/2 Ar ,
(5.173)
=: Z21/2 r
such that no factor Z2 and Z3 appears in the propagators of r and Ar ,
i + . . . i
h| T Ar Ar |i + ..., h| T r r |i +... (5.174)
q2 6 p m
1
L = Z3 Fr2 + Z2 r (i 6 m0 )r Z2 Z31/2 e0 r r A . (5.175)
4
To compute an amplitude, we apply the Feynman rules but with m0 replaced by Z2 m0 and e0
replaced by Z2 Z31/2 e0 , and no factors of Z21/2 or Z31/2 for external particles in the S-matrix. It is
important to appreciate that
Renormalising the field strengths does not change any physics if we modify the Feyn-
man rules accordingly.
where
e0
3 = Z3 1, 2 = Z2 1, m = Z2 m0 m, 1 = Z2 Z31/2 1 =: Z1 1. (5.177)
e
(1) (2)
This is only a rewriting of L in the form (5.175) by adding and subtracting Lr . Lr contains
the so-called counterterms. Note that we have defined Z1 by Z1 e = e0 Z2 Z31/2 . At this stage m
and e are just arbitrary parameters to be fixed soon.
The Feynman rules associated with the form (5.176) of the Lagrangian are now as follows:
(1)
Associated with Lr are the usual Feynman rules, but with the correct couplings as appearing
160 CHAPTER 5. QUANTUM ELECTRODYNAMICS
(1)
gag
in Lr :
faf
= (in Feynman gauge),
q2 + i
i
= in terms of m, not m0 , (5.178)
6 p m + i
= ie , in terms of e, not e0 .
(2)
The counterterms in Lr give rise to additional diagrams (counterterm diagrams). Their
(2)
structure becomes evident if we view the terms in Lr as extra couplings leading to amputated
diagrams. In deriving the Feynman rules note that a derivative in position space will give
rise to a factor of ip in the momentum space Feynman rules.
gives rise to
fafxfaf i(6 p 2 m ). (5.180)
ie 1 . (5.181)
1 1
3 Fr Fr = 3 Ar ( 2 + ) Ar (5.182)
4 2
gives rise to
i( q2 q q )3 . (5.183)
We compute diagrams with the above rules and stress again that no factors of Z2 and Z3 appear
for external particles because these are already contained in the counterterms.
The above procedure is merely a reorganisation of perturbation theory. The result for an amplitude is
the same irrespective of whether the computation is performed
either starting from L as given in (5.171) with the original Feynman rules including all Z-factors
as before,
(1) (2)
or from Lr + Lr in terms of Ar and r , including counterterm diagrams, and hence without
Z-factors for external particles.
5.9. RENORMALISED PERTURBATION THEORY OF QED 161
fpf
The fully resummed propagator of the renormalised electron field takes the form
i
= . (5.184)
r 6 p m r (6 p)
f f
At 1-loop level, i.e. at order in perturbation theory,
(1)
1PI
|1loop = ir (6 p)|1loop = i2 (6 p) + i(6 p 2 m )| . (5.185)
(1) (1)
Here i2 (6 p) is computed from Lr in terms of m and e, i.e. it has the form (5.101) with e0
(2)
and m0 replaced by e and m. The term i(6 p 2 m ) is due to the counterterm present in Lr .
As we will see momentarily, m and 2 depend on (as do the remaining counterterms 1 and
3 ). At 1-loop level only terms up to and including order are to be included.
gpg
q q
i + q 2
= 2 . (5.186)
r q (1 r (q2 ))
gg
At 1-loop level, r (q2 ) is computed via
2 (1)
r |1loop = ir (q)|1loop = i(q q q )( 2 (q ) 3 | )
2
1PI
(5.187)
(1) (1)
with 2 (q2 ) computed from Lr and 3 | ) the counterterm expanded up to order .
fpf gpg
Finally the 4 counterterm couplings are fixed by the renormalisation conditions as follows:
Two more conditions arise by specifying the meaning of e and m. This is arbitrary in principle.
One possible choice (out of infinitely many) is that m represents the physical mass and e the
physical charge measured at q = 0.
162 CHAPTER 5. QUANTUM ELECTRODYNAMICS
fpf
Combining all of these four conditions translates into the following equations:
!
! i
= + terms analytic at m. (5.190)
r 6 p m
Therefore we find
!
r (6 p) = 0 m is the physical mass
(5.191)
6 pm
and
d
r (6 p)
1 = 1 residue is 1. (5.192)
d6 p 6 p=m
Therefore
ier ( p, p) = ie .
!
(5.194)
gpg
Note that (5.192) has already been used. The last condition is
!
! i
= 2 + terms analytic at q = 0 (5.195)
r q
and thus
!
1 r (0) = 1. (5.196)
The conditions (5.191), (5.192), (5.194) and (5.196) are the 4 renormalisation conditions. We can
solve these perturbatively, order by order in perturbation theory. At 1-loop order we find for
condition (5.191)
(1) !
2 (m) (m 2 m ) = 0 (5.197)
and therefore
(1)
m = m 2 2
Z1 (5.198)
M2
!
2
= m 2 dx 2 m 1 x m + log ,
2 2 2
0
conditon (5.192)
d (1)
2 = (6 p) , (5.199)
d6 p 2 m
condition (5.194)
(1)
1 = F1 (0), (5.200)
5.9. RENORMALISED PERTURBATION THEORY OF QED 163
condition (5.196)
Z1
M2
!
(1) 2 2
3 = 2 (0) = dx x(1 x) + log 2 . (5.201)
m
0
e and m are finite quantities to be taken from experiment - the physical coupling and mass.
When computing an amplitude, we take into account both the Feynman rules (5.178) and the
counterterms. Since we have computed the counterterms only to order O() it is only consistent
to compute all other diagrams to that same order, where we must take into account that the i
are of order O(). Thus, as in the computation of the 1PI diagrams, a loop diagram is always
accompanied by a counterterm diagram, but to order no counterterm diagrams appear inside
a loop (as this would be order 2 ).
The Feynman rules (5.178) give rise to a divergent expression in terms of m, e, . The counter
term coefficients i (m, e, ) will precisely cancel these divergences. All physical amplitudes
are therefore independent of in the end, so we can take .
If we wish to compute a quantity to order 2 we first need to compute the divergent 1PI terms
to order 2 , and impose again the renormalisation conditions. In this computation we must take
into account, in addition to usual 2-loop diagrams from (5.178), also 1-loop diagrams on top of
O() counterterms, as well as diagrams involving only counterterms with coefficients expanded
to order 2 . We will discuss this more systematically in a 2-loop example in QFT II.
The perturbative cancellation of divergences works only because the counterterms are of the
same form as the terms in the bare Lagrangian, i.e. they do not introduce any qualitatively new
interactions. This is guaranteed because at each order in perturbation theory, no qualitatively
new UV divergent 1PI diagrams arise, but the same 1PI diagrams merely receive higher-loop
contributions. This leads to a readjustment of the counterterm coefficients order by order such
as to absorb the new divergences.
164 CHAPTER 5. QUANTUM ELECTRODYNAMICS
The renormalisation conditions (5.191) and (5.194) are arbitrary: We could define m and e to be
any function of the physical mass m or e, thereby changing the i accordingly. This ambiguity
will cancel in all final amplitudes.
Conditions (5.192) and (5.196) are fixed by the Feynman rules, which do not contain any Z-
factors. However, it is possible to change (5.192) and (5.196) if at the same time we modify
the Feynman rules accordingly. I.e. we can include arbitrary Z-factors for external states in the
Feynman rules, provided we change (5.192) and (5.196). This reflects our freedom in normal-
ising the fields in a manner consistent with the Feynman rules.
5.10 Infrared divergences
Loop diagrams with massless particles may exhibit infrared divergences from integration over loop
momenta k 0.
Indeed, we had introduced a small fictitious photon mass to regulate the IR divergences.
These IR divergences cancel in all cross-sections against another source of IR divergences from
radiation of soft photons (Bremsstrahlung). Consider e.g.
As k 0 these processes are divergent. This is the case already in classical electrodynamics
and called infrared catastrophe of electrodynamics. The reason why this divergence is not
that catastrophic after all is that no detector can measure a photon below a certain threshold.
Thus we can include all possible Bremsstrahlung photons for k 0 to a given process (because
as k 0 the soft photons cannot be measured. It now so happens that the resulting infrared
divergences precisely cancel the IR divergences from the above loop diagrams oder by order in
, in the cross-section . For details we refer to Peskin-Schrder 6.1, 6.4, 6.5.
Chapter 6
A consistent Lorentz invariant quantum theory of free massless spin-1 field A ( x) must be a
gauge theory - see our discussion around equ. (4.67) based on Weinberg I, 8.1.
At the level of interactions consistency requires that A ( x) couples to a conserved current j ( x).
This is equivalent to the gauging of a global symmetry in the matter sector - see our discussion
around (5.54) .
L = (i 6 m), (6.1)
In a local QFT it is natural to consider local symmetries, i.e. to promote this to a local transfor-
mation
( x) 7 eie( x) ( x) =: U ( x)( x). (6.2)
The logic behind such a modification is that a symmetry transformation at spacetime point y far
away from x should not affect the field at x. Note that in QFT global symmetries, though, from
this perspective, unnatural, are of course fully consistent. By contrast, it is conjectured that in
presence of gravity all symmetries must be local and that no global symmetries exist. In any
case we are motivated to consider the consequences of the transformation (6.2).
1
n ( x) := lim [( x + n ) ( x)] (6.3)
0
165
166 CHAPTER 6. CLASSICAL NON-ABELIAN GAUGE THEORY
is not a good object with respect to (6.2) because in (6.3) two objects with very different trans-
formation behaviour under (6.2) appear, i.e.
To define a better notion of derivative we introduce the object C ( x, y) - the so-called comparator
or Wilson line - such that under (6.2)
1
n D ( x) := lim [( x + n ) C ( x + n, x)( x)] . (6.6)
0
D ( x) 7 U ( x) D ( x). (6.7)
Let us now construct the Wilson line starting from the requirement (6.5), which implies that
under (6.2)
C (y, x) 7 U (y)C (y, x)U 1 ( x). (6.8)
Furthermore we impose C (y, y) = 1 for obvious reasons. Note that for the U (1) symmetry
under consideration U ( x) is a pure phase, and it thus suffices to take C (y, x) as a pure phase.
We will soon generalize this.
i
A ( x) 7 U ( x) A ( x)U 1 ( x) + U ( x)U 1 ( x). (6.13)
e
The vector field A ( x) is therefore a direct consequence of the existence of a local symmetry. It
is called a connection. A ( x) is a local field and thus has dynamics in its own right.
D ( x) 7 U ( x) D ( x) (6.14)
we have
and thus
[ D , D ] 7 U ( x)[ D , D ]U 1 ( x). (6.16)
The term
1
F := [ D , D ] = A A + ie[ A , A ]. (6.18)
ie
F = A A (6.19)
where g R takes the role of e, a R and T a form a basis of the Lie algebra Lie( H ), i.e. the algebra
of infinitesimal group transformations. Viewed as an abstract Lie algebra, Lie( H ) is determined by
the commutation relations
[T a , T b ] = i f abc T c , (6.22)
168 CHAPTER 6. CLASSICAL NON-ABELIAN GAUGE THEORY
where a sum over c is understood. Without proof we state here that the structure constants f abc are
totally antisymmetric in the indices a, b, c and thus in particular invariant under cyclic permutations
of a, b, c.
For example:
The Lie group H = U (1) has dim( H ) = 1 and its generator is simply T a T R. Therefore
[T , T ] = 0 and H is an abelian Lie algebra.
The Lie group H = S U ( N ), viewed as an abstract group, is defined as the group of volume-
element preserving linear transformations on CN which leave the sesqui-linear form
N
X
C 3 u, v C : (u, v) 7
N
ui vi (6.23)
i=1
by assigning h H a matrix V (h) as above. Lie( H ) is then the algebra of T a CN,N such that
The dimension of Lie( H ) is N 2 1. For instance, the generators of H = S U (2) are typically
normalised to be
1
T a = a (6.26)
2
with a = 1, 2, 3 and a the Pauli matrices. The structure constants of S U (2) are then
Consider now a matter Lagrangian with matter fields ( x) transforming in a unitary representation of
H such that L is invariant under global transformations
1 ( x)
The Lagrangian
N
X
L = (i 6 m) = i (i 6 m)i (6.30)
i=1
By the dimension of a representation we mean the dimension of the vector space in which the
matter field takes its value. The (complex) dimension of the fundamental representation of
S U ( N ) is thus N.
Take H = S U ( N ), but ( x) now in the adjoint representation, i.e. we consider now a spinor
field valued in the Lie algebra Lie(H) viewed as a vector space. This means that, with spinor
indices suppressed,
n o
x : ( x) i j ( x) CN,N ( x) = ( x), tr ( x) = 0 (6.32)
and the transformation behavior is given by the adjoint action of the Lie group H on its Lie
algebra,
R(h) := V (h)V (h)1 V (h)i j jk V 1 (k) ,
kl
(6.33)
R (h) := V (h) V (h) = V (h)V (h).
1 1
L = tr (i 6 m) tr [i j (i 6 m) jl ] i j (i 6 m) ji . (6.34)
where cyclicity of the trace was used to go from the second to the third line.
170 CHAPTER 6. CLASSICAL NON-ABELIAN GAUGE THEORY
We can now repeat all the steps involved in the gauging of U (1) in this more general setting. For
definiteness we work with a Dirac spinor field in the fundamental representation of S U ( N ). Consider
the gauge transformation
( x) 7 U ( x)( x) (6.36)
with
dim
X (H )
U ( x) = exp ig
( x) T V (h( x)).
a a
(6.37)
a=1
Now it turns out that the compensator must take values in the representation of H. It can therefore be
expanded as
dim (H )
Aa ( x)T a n + O( 2 ).
X
C ( x + n, x) = 1 ig (6.38)
a=1
is then an N N matrix-valued vector field. The covariant derivative takes the form
X
D ( x) = ( x) + ig Aa ( x)T a ( x), (6.40)
a
i
A ( x) 7 U ( x) A ( x)U 1 ( x) + U ( x)U 1 ( x), (6.41)
g
1
The field strength F = ig [ D , D ] F ( x)a T a is
F ( x) 7 U ( x) F ( x)U 1 ( x) (6.46)
tr ( F F ) 7 tr UF |{z}
U 1 U F U 1 = tr F F U 1 U ) = tr( F F ).
(6.47)
=1
defines the so-called Yang-Mills Lagrangian. The crucial difference to U (1) gauge theory is that the
Yang-Mills gauge field exhibits cubic and quartic self-interactions which are contained in the term
1 X a a
F F . (6.50)
4 a
Diagrammatically, these interactions between the gauge bosons are of the form
The representations of fermionic matter are given as follows: The SM is a chiral theory, i.e.
left- and righthanded fermion fields L PL and R PR transform in different representa-
tions. The Standard Model comprises 3 families of
172 CHAPTER 6. CLASSICAL NON-ABELIAN GAUGE THEORY
The interactions are - apart from the Yang-Mills interactions that follow from the above representa-
tions - given by
and the gauge invariant Yukawa interactions, very roughly of the form i i , which yield
fermion masses if hi , 0.
Chapter 7
for which we now derive a path integral representation as follows: The idea is to partition the transition
time T := tF tI into N + 1 intervals of length
T
t = (7.5)
N+1
1 Richard Feynman, 1948.
173
174 CHAPTER 7. PATH INTEGRAL QUANTISATION
and to factorise
eiHT = eiHt . . . eiHt . (7.6)
p2
Suppose first that H ( p, q) = f ( p) + V (q), e.g. with f ( p) = 2m , and use the Baker-Campbell-
Hausdorff formula
eA+ B = eA eB e 2 [A,B]+... .
1
(7.9)
Then we have
eiHt = ei f ( p)t eiV (q)t e 2 [ f ( p),V (q)]t
1 2 +O(t3 )
. (7.10)
1 1
hqk+1 | pk i = eipk qk+1 and hpk | qk i = eipk qk (7.12)
2 2
and taking the limit t 0 we obtain for hqk+1 | eiHt |qk i the expression
Z
dpk i( f ( pk )+V (qk ))t i(qk+1 qk ) pk
e e . (7.13)
2
with
1
qk = (qk + qk+1 ). (7.15)
2
Thus, altogether the transition amplitude is
Z N
dp0 Y dpk dqk i PkN=0 [ pk qk+t1 qk H ( pk ,qk )]t
hqF , tF | qI , tI i = lim e , (7.16)
N 2 k=1 2
7.1. PATH INTEGRAL IN QUANTUM MECHANICS 175
is called Weyl-ordered. Thus the above derivation of the path integral is, by definition, correct for
quantum theories with Weyl-ordered Hamiltonian. As we have seen this trivially includes the case
H ( p, q) = f ( p) + V (q). On the other hand, if H ( p, q) is not Weyl-ordered, then on the right hand
side of (7.18) extra terms appear. This is understood in the sequel - we take the right hand side of
(7.18) as an abbreviation for the inclusion of these terms if necessary.
A note on rigor
A much more detailed and very careful justification of the above procedure leading to equ. (7.18) can
be found e.g. in Chapter 1 of Brown, Quantum Field Theory.
q(Z
tF )=qF RtF
i dtL( p,q)
hqF , tF | qI , tI i = Dq(t )Dp(t )e tI (7.21)
q(tI )=qI
176 CHAPTER 7. PATH INTEGRAL QUANTISATION
Rt
with t F dtL( p, q) =: S [ p, q] and L( p, q) pq H ( p, q). Note that here L( p, q) is really a function
I
of q and p, both of which we integrate over separately.
except that the prefactor of the quadratic terms does not yet have a positive part. However, as will be
discussed in more detail below, it is mathematically justified to perform an analytic continuation of the
integrand by rotating t to the lower complex half-plane. This means that t is replaced by t (1 i )
with small and positive. We then evaluate the integral for t (1 i ) and take = 0 at the very end
of all computations. With this in mind, (7.23) becomes
r
m
(q j+1 q j )2 im m 2
e i 2t
ei 2t (q j+1 q j ) C. (7.25)
2t
Then m2 q2 V (q) = L(q, q) remains and we obtain the Feynman-Kac-formula
RtF q(Z
tF )=qF RtF
N Z
Y i dtL(q,q) i dtL(q,q)
hqF , tF | qI , tI i = lim C N +1 dqk e tI Dq(t )e tI . (7.26)
N
k =1
q(tI) =qI
h i
counts all possible continuous paths from qI to qF weighted by exp hi S . This explains the name path
integral. In the classical limit S [q, q] h and due to the strongly oscillating phase of the integrand,
the right-hand side is dominated by paths for which the action becomes stationary,
S
= 0. (7.28)
q
This way one recovers the classical dynamics as the saddle-point approximation to the full quantum
amplitude.
7.1. PATH INTEGRAL IN QUANTUM MECHANICS 177
For N finite, we may wonder if each of the integrals (7.14) underlying (7.16) is convergent.
For a typical Hamiltonian H ( p, q) which is bounded from below, convergence of (7.14) follows
by analytic continuation to Euclidean time. Indeed we may view the parameter t in (7.14)
as along the (w.l.o.g. positive) real axis in the complex-plane. We can then rotate H ( p, q)t
to the negative imaginary time-axis by a Wick rotation. Note that the dependence on p and q
is unaffected. This Wick rotation is possible in Quantum Mechanics because for an analytic
function H ( p, q)t we do not encounter any poles in the complex time-plane along the way.
Wick rotation then effectively replaces
t i (7.29)
The limit N and the ensuing interpretation of (7.16) as an integral over all continuous
p2
paths from qI to qF an be made mathematically rigorous - at least for H = 2m + V (q) - by
directly passing to the Euclidean theory:
theories such as the harmonic oscillator the path integral in this form can be solved and convergent
expressions can be obtained in a very explicit manner. This will be demonstrated in the tutorials.
Crucial in this context is again a suitable
q normalization of the position space path integral by carefully
im
taking into account the factor C = 2t in the measure in (7.26).
of the Heisenberg operator qH (t1 ) for tF > t1 > tI . We can transform this to the Schrdinger picture as
and obtain
hqF | eiH (tF t1 ) qS eiH (t1 tI ) |qI i . (7.33)
We proceed as before in the computation of the transition amplitude by factorising the time evolution
operator and inserting the identity multiple times. The operator qS now acts on |qiK hq|K correspond-
ing to the time t1 ,
qS |qiK hq|K t = q(t1 ) |qiK hq|K t . (7.34)
1 1
q(Z
tF )=qF
q(Z
tF )=qF
hqF , tF | T qH (t1 )qH (t2 ) |qI , tI i = DqDp q(t1 )q(t2 )eiS [q,p] . (7.38)
q(tI )=qI
We will need above expressions with initial and final states given by |i - the vacuum of the interacting
theory. To compute this expression most simply we use the same trick employed in the context of
7.2. THE PATH INTEGRAL FOR SCALAR FIELDS 179
the interaction picture derivation of correlation functions in QFT I. Consider an eigenbasis of the
Hamiltonian with
H |ni = En |ni and H |i = E |i (7.39)
and suppose without loss of generality that E = 0 (otherwise we shift H by a constant). Then
|qI , tI i we can write
X X
|qI , tI i = eiHtI |qI i = eiHtI |ni hn| qI i = |i h| qI i + eiEn tI |ni hn| qI i. (7.40)
n |ni,|i
Let us now replace tI tI (1 i ) and take tI , with small and positive. In this limit all terms
with |ni , |i are exponentially suppressed. Thus
for all |qI i , |qF i with non-zero overlap with |i. Therefore
RT
R i dtL( p,q) Q
Y Dq(t )Dp(t )e T i q ( ti )
h| T qH (ti ) |i = lim . (7.42)
T (1i ) RT
i R i dtL( p,q)
Dq(t )Dp(t )e T
Note that
where
1 2 1
H (( x), ( x)) = ( x) + (( x))2 + V (( x)). (7.44)
2 2
Note that all non-hatted quantities are classical fields. One defines the quantum fields H (t, ~x) in the
Heisenberg picture and its eigenstates via
Our task is thus to compute the transition amplitudes between two such states
Introduce an infrared (IR) cutoff by considering the theory in a finite volume V as well as an
ultraviolet (UV) cutoff by discretizing V. I.e. we consider the theory on a finite 3-dimensional
lattice L with lattice spacing a and parametrise the lattice sites by the discrete vectors
where the sum over ~j is a suitable sum over nearest neighbors such as to reproduce the spatial
derivative.
In the end we remove the IR and the UV cutoff in space. It is this continuum limit which in-
troduces both IR and UV divergences. These must be taken care of by carefully implementing
renormalisation as we will see in more detail later. In a properly renormalized theory, all physi-
cal quantities computed via the path integral (in particular correlation functions) are finite even
after taking the continuum limit.
This is a formal expression meant in the above sense. Inserting the specific Hamiltonian (7.43) in the
exponent shows that the integral is Gaussian, since . The Gaussian integral can be performed
explicitly as in the finite-dimensional case. A rigorous derivation can be given by discretizing the path
integral and taking the continuum limit in the end. Altogether we have
where the extra factors picked up upon performing the Gaussian integral are, by definition, included
in the measure. Since we will normalize the relevant expressions in a suitable manner momentarily
7.2. THE PATH INTEGRAL FOR SCALAR FIELDS 181
with arbitrary boundary conditions with non-zero overlap with |i. This yields the path-integral
master formula for the computation of quantum correlation functions
Y
G ( x1 , ..., xn ) h| T H ( xi ) |i
i
RT
R
D( x)( x1 )...( xn )ei T d xL(( x))
4 (7.52)
= lim RT ,
D( x)ei T d x L(( x))
R 4
T (1i )
where the denominator ensures again correct normalisation. Note in particular that all universal con-
stant prefactors from the Gaussian integral cancel as promised.
xE = ( x1 , x2 , x3 , x4 = ix0 ). (7.53)
It is the suppression of the path integral by eS E which is responsible for its improved convergence
properties. The correlation functions in the Euclidean theory are called Schwinger functions, as op-
posed to the correlation functions in Minkowski space, called Wightman functions. This way, one can
182 CHAPTER 7. PATH INTEGRAL QUANTISATION
compute a correlation function of a theory in R1,3 by first computing its associated Schwinger func-
tion in the Euclidean theory and then Wick rotating back to R1,3 - provided one does not encounter
any poles in the complex xE -plane as one undoes the Wick rotation.
The Osterwalder-Schrader theorem gives conditions under which a mathematically precise defini-
tion of a Lorentzian QFT (in the sense that it satisfies the so-called Wightman axions) via analytic
continuation of the Euclidean path integral is possible. In particular, the Schwinger functions must
satisfy a property known as reflection positivity. Unfortunately, to date the only interacting field the-
ories for which the Osterwalder-Schrader conditions have been rigorously proven to be satisfied are
certain theories in 2 and 3 dimensions.3 These theories can be viewed as mathematically well-defined
quantum field theories. Establishing such a definition of Quantum Yang-Mills theory in R1,3 is a 1-
million-dollar problem.
On wave functionals
We can conclude that there are two equivalent ways to define a quantum field theory starting from a
classical theory:
Canonical formalism:
First promote the classical field ( x) to a quantum operator H ( x). Then define the Hilbert
space of states and derive the correlators in the interaction picture as given by
RT
i I ( xi )e T
Q i dtHI (t )
Y h0| T |0i
h| T H ( xi ) |i = lim RT . (7.55)
T (1i ) i T dtHI (t )
i h0| T e |0i
This is the Gell-Mann-Low formula, which we derived in section 2.5 in QFT I.
Alternatively we can begin not with the concept of a quantum operator, but with the concept
of a physical state as a wave functional of the classical fields. This is a generalization of the
well-familiar correspondence between states and wave functions (as functions on the space of
classical observables) in Quantum Mechanics. Consider for instance the Schrdinger represen-
tation in Quantum Mechanics, where the correspondence between a state localised at q0 and its
wavefunction is
|q0 , ti (q(t )) = (q(t ) q0 (t )). (7.56)
Similarly, in Quantum Field Theory we define a state |0 (~x, t )i via a functional ((~x, t )) (as
a map from the space of classical observables - the classical fields - to C) as
The next step is then to define the concept of a quantum operator H ( x) by specifying its matrix
elements with a basis of states as
(~x,tZ
2 )=2 ( ~
x)
Rt
i 2 d4 x[H ]
h1 (~x), t1 | H ( x) |2 (~x), t2 i = D( x)D( x)e t1
(7.58)
(~x,t1 )=1 (~x)
and more generally the correlators G ( x1 , ..., xn ) via path integral of classical fields. In particular
the path integral is a convenient way to express the vacuum wave functional as the overlap
Z Rt
0 i tI dt d x[H ]
0 3
0 [] = h(~x), t| i = lim D(~x, t )D(~x, t )e
0
, (7.59)
tI (1i )
where (~x, t0 )|t0 =t = (~x, t ), but (~x, tI ) is arbitrary (with non-zero overlap with |i).4
i2
Z Z Z !
iS []
Z [ J ] = De 4
1 + i d x( x) J ( x) + d x1 d x2 ( x1 )( x2 ) J ( x1 ) J ( x2 ) + ...
4 4
2!
Z Z Z
iS []
= De + i d x J ( x) D ( x)eiS []
4 (7.62)
i2
Z Z
+ d x1 d x2 J ( x1 ) J ( x2 ) D ( x1 )( x2 )eiS [] + ...
4 4
2!
and thus
X in
Z[J]
Z Z
= d4 x1 ... d4 xn J ( x1 )...J ( xn )G ( x1 , ..., xn ). (7.63)
Z [0] n=0
n!
To solve this formula for G ( x1 , ..., xn ) we need the tools of functional calculus.
We can think of ( x) as an element of an -dimensional vector space and consider the following
formal analogies:
4 For a detailed treatment of wave functionals we refer e.g. to Chapter 10 of Hatfield, QFT of Point Particles and Strings.
184 CHAPTER 7. PATH INTEGRAL QUANTISATION
A vector is given as u ui . We can map two real vectors to R via the inner product
ui vi uT v u v. (7.64)
vi = Ai j u j ( A v)i . (7.65)
jk = ik .
Ai j A1 (7.67)
In this case the vector is the classical field ( x), with the dependence on the continuous pa-
rameter x formally representing the previous index i. The inner product becomes
Z
dx ( x)( x) . (7.68)
K ( x, y) = ( x y), (7.70)
To set up a differential calculus we first consider a function of a finite-dimensional vector and perform
a Taylor expansion
f
f (ui + hi ) = f (ui ) + hi + O(h2 ). (7.72)
ui
f
This defines the derivative ui . Formally we can generalise this definition of a derivative to a functional
F [( x)] by expanding
F
Z
F [( x) + h( x)] = F [( x)] + dx h( x) + O(h2 ). (7.73)
( x)
7.4. FREE SCALAR FIELD THEORY 185
The quantity F( x) is called functional derivative. It obeys the same rules as the finite-dimensional
derivatives. In particular it is useful to consider the analogies
u j = i j (y) = ( x y), (7.74)
ui ( x)
Z
( u j k j ) = i j k j = k i dy (y)(y) ( ) = ( x ). (7.75)
ui ( x) ( x)
( x)
We will oftentimes leave the actual integrals implicit to save some writing. The last equation then
reads
ei = i( x)ei . (7.77)
( x)
! !
1
G ( x1 , ..., xn ) = ... Z [ J ] (7.78)
Z [0] iJ ( x1 ) iJ ( xn ) J =0
In the path integral formula of Z [ J ] the contour is taken along t (1 i ) in order to project the initial
and final states to the vacuum state. For the free theory one can prove that alternatively one can shift
not the time variable into the complex plane, but make the replacement
Since this prescription works for the free theory, it does so also for perturbative interactions as long
as the interaction does not deform the theory too far away from the free theory. Non-perturbatively,
however, the replacement (7.80) cannot in general be proven to serve our purpose of projecting the
186 CHAPTER 7. PATH INTEGRAL QUANTISATION
initial and final states to the vacuum and one must stick with the prescription t t (1 i ).
With (7.80) the partition function for the free theory takes the form
Z
Z0 [ J ] = D e 2 d x d y ( x)K ( x,y)(y)+i d x ( x) J ( x)
i
R R
4 4 4
Z (7.81)
= D e 2 (K)+iJ ,
i
where
K ( x, y) = (4) ( x y)(2x + m20 i ). (7.82)
Inserting K ( x, y) yields
(2x + m20 i ) K 1 ( x, z) = (4) ( x z). (7.84)
Since the Feynman propagator is a Greens function for the Klein-Gordon equation, we can just quote
our result from QFT I,
d4 p
Z
1 i
K ( x, z) = iDF ( x z) with DF ( x) = eipx . (7.85)
(2) p m20 + i
4 2
Note that in this approach the i-term arises automatically as a consequence of (7.80). Thus,
Z
D e 2 DF
1 1 +iJ
Z0 [ J ] = (7.86)
1
= Z0 [0]e 2 (iJ )DF (iJ ) (7.88)
1
R
d4 x d4 y iJ ( x) DF ( xy)iJ (y)
Z0 [ 0 ] e 2 .
We notice that Z
D0 e 2 (K )
1 0 0
Z0 [ 0 ] = (7.89)
for M symmetric and positive definite (or more generally for Re(i ) > 0, where i are the eigenvalues
of M). Formally this generalizes to
const.
Z0 [0] = p , (7.91)
det(iK )
where det(iK ) = det[(4) ( x y)(i2x + im2 + )] is called a functional determinant. Note that
Z0 [0] is a divergent quantity, but as will be discussed in the tutorials it can be defined rigorously by a
regularisation procedure, e.g. on a lattice with IR and UV cutoff. For finite cutoffs, Z0 [0] is finite. All
physical quantities are computed in the regularized theory and the expression Z0 [0] cancels in all such
expressions because the Greens functions derive from Z [ J ]/Z [0]. In the end one takes the continuum
limit, and the formal divergence of det(iK ) poses no problems.
of the free theory by differentiating the exponential with respect to J and setting J = 0. To compute
a 2n-point function one expands the exponential precisely to n-th order, while the 2n + 1-functions
vanish. In this spirit the 2-point function is computed as
Z
1
G0 ( x1 , x2 ) = d4 x d4 y iJ ( x) DF ( x y)iJ (y) = DF ( x1 x2 ), (7.93)
iJ ( x1 ) iJ ( x2 ) 2
confirming that the Greens function of the DAlembert operator gives the 2-point correlator in the
free theory. Feynman diagrams are a graphical way to organise the combinatorics of expanding the
exponential and taking the functional derivatives. For example the result for
G0 ( x1 , x2 , x3 , x4 ) =
iJ ( x1 ) iJ ( x2 ) iJ ( x3 ) iJ ( x4 )
Z Z
11 4 4 1
d x d y iJ ( x) DF ( x y)iJ (y) d4 u d4 v iJ (u) DF (u v)iJ (v)
2! 2 2
(7.94)
can be expressed graphically as
x1 x2 x1 x2 x1 x2
+ +
x3 x4 x3 x4 x4 x3
where a line between xi and x j denotes as always a factor of DF ( xi x j ). Note that interactions only
occur if there is a dot on crossing lines, i.e. there is no interaction in the third diagram. As one can
convince oneself, the combinatorial factors work out correctly.
188 CHAPTER 7. PATH INTEGRAL QUANTISATION
Then
Z
D eiS 0 []+i d4 x Lint ()+iJ
R
Z[J] =
Z
d4 x Lint () iS 0 []+iJ
R
= D ei e
Z (7.96)
R
= D e i d4 x Lint (i J ) iS 0 []+iJ
e
Z
R
= ei d4 x Lint (i J ) D eiS 0 []+iJ .
for any functional F []. Inserting our previous result (7.88) for D eiS 0 []+iJ gives the important
R
result
R R
4 4
Z [ J ] = ei d xLint (i J ) Z [ J ] = Z [0]ei d xLint (i J ) e 2 (iJ )DF (iJ ) .
1
0 0 (7.98)
To prove the equivalence of (7.98) and (7.99), we consider two functionals F and G and note that
iJ
" # " #
F G []eiJ
=F G e . (7.100)
iJ
Since the derivatives commute this is
"#
G F eiJ = G F [iJ ]eiJ . (7.101)
iJ iJ
In particular this implies that
" #
G F [iJ ] = G iJ
F [iJ ]e |=0 = F G []eiJ |=0 . (7.102)
iJ iJ
The equivalence of (7.98) and (7.99) follows from this if we identify
" #
1
F = e 2 DF ,
(7.103)
i d4 xLint ()
R
G [] = e .
7.5. PERTURBATIVE EXPANSION IN INTERACTING THEORY 189
Perturbation theory can be set up from both expressions (7.98) or (7.99). Starting from (7.98) is close
in spirit to our discussion of the free theory as will be demonstrated in the tutorials. In the sequel we
take the approach via (7.99) and make the following definition:
1
DF
hF []i0 B e F [] .
2 (7.104)
= 0
G ( x1 , ..., xn ) follows from (7.105) either by expanding eiJ and using (7.63), or directly as
Z [ J ]
G ( x1 , ..., xn ) = ...
iJ ( x1 ) iJ ( xn ) Z [0] J =0
R
i d4 xLint +iJ (7.106)
...
iJ ( x1 ) iJ ( xn )
e
0
= D R 4 E .
ei d xLint
0
J =0
We can perform the functional derivative in the nominator and arrive at the final expression
R
D 4 E
( x1 )...( xn )ei d xLint
0
G ( x1 , ..., xn ) = D R 4 E . (7.107)
i d xLint
e
0
To understand the meaning of this expression we must familiarize ourselves more with the operation
(7.104). First, let us compute
1
h( x1 )( x2 )i0 B e 2 DF ( x1 )( x2 )=0 .
(7.108)
Expanding the exponential and keeping precisely terms of quadratic order in yields
Z
1
d4 x d4 y D F ( x y) ( x1 )( x2 )=0 = DF ( x1 x2 ) ( x1 )( x2 ) . (7.109)
2 ( x) (y)
Thus
h( x1 )( x2 )i0 = h0| T ( x1 )( x2 ) |0i , (7.110)
where |0i is the free vacuum. Generalising this reasoning gives a simple and very direct proof of
Wicks theorem, which will be presented in the exercises. We will end up with the result known from
QFT I:
h1 ...2n+1 i0 = 0.
190 CHAPTER 7. PATH INTEGRAL QUANTISATION
R
With Wicks theorem at hand we go back to G ( x1 , ...xn ) and expand exp[i d4 xLint ] in (7.107) to
recover exactly the same Feynman rules as in operator language from the Gell-Mann-Low formula
in the interaction picture. We refer to QFT I for a precise statement of these rules in position and in
R
momentum space. The special case exp[i d4 xLint ] = 1 gives the free theory.
Note that ( xi ) in (7.107) or iJ( x ) in (7.106) corresponds to an external point xi in the Feynman
i
diagram. Thus Z [0] contains no external points and therefore represents the partition function since
P
Vi
Z [0] = e i . (7.111)
Z[J]
Here Vi is the value of a vacuum bubble labeled by i. Consequently Z [0]
, i.e. the generating functional
of the Greens functions, contains no vacuum bubbles.
L = number of loops.
L is the number of unfixed momentum integrals. From the momentum space Feynman rules we recall
that
one linear combination of these delta-functions merely corresponds to overall momentum con-
servation.
The perturbative expansion in loops corresponds to an expansion in h. To see this, we note that the
exponent in the path integral is - reinstating h -
Z 1 iZ
i i
S = d x ( + m ) +
4 2 2
d4 x Lint . (7.113)
h h 2 h
Each internal line carries a propagator, i.e. a factor of
!1
2 + m2
h, (7.114)
h
7.6. THE SCHWINGER-DYSON EQUATION 191
and each vertex carries a factor of h1 . For a given scattering process with fixed number E of external
lines, expansion of a scattering amplitude in L therefore gives I V = L 1 extra powers of h (which
come in addition to factors of h from the external particles, whose power is fixed for given E). Taking
into account these extra factors, in most field theories the classical scattering amplitude, i.e. O(h0 ),
corresponds to the tree-level diagram, but exceptions to this rule of thumb exist.5
n
Note furthermore that for Lint = n! , a loop expansion with fixed E is an expansion in , because
E + 2I = nV (7.115)
since both ends of each internal line and one end of each external line end in an n-point vertex. With
L = I V + 1 this gives
(n 2)V = 2L + ( E 2). (7.116)
So for E fixed, an expansion in L corresponds to an expansion in V.
To see this we consider a general theory with arbitrary fields collectively denoted by ( x) defined by
Z
Z[J] = D ei(S []+J ) . (7.117)
( x) 0 ( x) = ( x) + ( x) + O( 2 ) (7.118)
Z (7.120)
= D ei(S [ ]+ J )
0 0
S
Z " ! #
i(S []+J )
= D e 1+i + J + O( ) .
2
5 Themost famous one is in fact the formulation of General Relativity as the quantum field theory of a spin-2 particle,
where the classical potential is recovered from scattering amplitudes at 1-loop order in perturbation theory.
192 CHAPTER 7. PATH INTEGRAL QUANTISATION
In the second line we simply relabeled the integration variable and in the third we used invariance of
the measure. Thus
i(S []+J ) S
Z " #
0 = D e + J
(7.121)
S
Z "Z ! #
i(S []+J )
D e 4
d y + J (y) (y) .
(y)
For the special transformation (y) = (4) (y x), our assumption (7.119) does indeed hold, and
therefore
S
Z !
D + J ( x) ei(S []+ J) = 0. (7.122)
( x)
Equ. (7.122) is the celebrated Schwinger-Dyson equation. It states that the classical equation of
motion (here in presence of a source J),
S
0= + J, (7.123)
holds as an operator equation in the quantum theory, i.e. it holds inside the path integral (with no
further field insertion at the same spacetime point).
If we act on (7.122) with ni=1 iJ( x ) we arrive at
Q
i
n
S
Z ! X
+ J ( x) ( x1 )...( xn ) + ( x1 )...( xi1 )(i( x xi ))( xi+1 )...( xn )
D
( x) i=1
ei(S []+ J) = 0. (7.124)
Taking J = 0 gives
n n
S Y X
h| T ( xi ) |i = i h| T ( x1 )( xi1 )( x xi )( xi+1 )...( xn ) |i . (7.125)
( x) i=1 i=1
In this form the Schwinger-Dyson equation shows that the equations of motion hold inside a correla-
tion function up to contact terms on the right-hand side.
As a simple corollary we derive the Ward-Takahashi identities: To this end start from (7.121) again
and consider a transformation
+ (7.126)
with
(y) = (4) (y x)( x) (7.127)
(with no integration over the x-variable on the left) holds off-shell. Here j ( x) is the Noether current
associated with the global continuous symmetry. If this transformation leaves the measure invariant,
then (7.121) yields Z
0= Dei(S []+J ) ( j ( x) + J ( x)( x)) . (7.129)
Qn
Taking i=1 iJ ( xi ) at J = 0 gives the Ward-Takahashi identity:
n n
Y X
h| T j ( xi ) |i = i h| T ( x1 )...( xi1 )( x)(4) ( x xi )( xi+1 )...( xn ) |i .
i=1 i=1
(7.130)
While we have seen this equation already in QFT1, the above path integral derivation indeed clarifies,
as promised, that the Ward identities only hold for symmetries that leave the measure invariant. If this
is not the case, the symmetry is said to be anomalous and current conservation (up to contact terms)
does not hold at the quantum level. We will learn more about anomalies later in this course.
To conclude this chapter we go back to the Schwinger-Dyson equation in its form (7.122) and use the
identity
F []eiJ = F eiJ (7.131)
iJ
S
for F [] = to rewrite (7.122) as
S
+ J Z [ J ] = 0. (7.132)
= iJ
The Schwinger-Dyson equation in this form is a fundamental identity for Z [ J ] that beautifully encodes
the structure of the correlation functions. To see this we introduce the abbreviations
( xi ) i , J ( xi ) Ji (7.133)
and
1 X 1 (m)
S [] = i DF1i j j + Y i ...im (7.134)
2 m3
m! in ...im 1
as shorthand for Z
1
d4 xi d4 x j ( xi )(4) ( xi x j )(2i + m20 )( x j ) (7.135)
2
and similarly for the interaction terms. To evaluate (7.132), i.e.
S
Z [ J ] = Ji Z [ J ], (7.136)
i i =
iJi
we note that
S
X 1
= DF1i j + Yiim1 ...im1 ... . (7.137)
i i = iJ j ( m 1 ) ! iJ iJ
i1 im1
iJi m
194 CHAPTER 7. PATH INTEGRAL QUANTISATION
Now multiply both sides by ( DF )ki using ( DF )ki D1
F = k j to find ij
X 1 (m)
Z [ J ] + ( DF )ki Yii1 ...im1 ... Z [ J ] = ( DF )ki Ji Z [ J ]. (7.138)
iJk m3
( m 1 ) ! iJ i1 iJ im1
n
f
i1 i2
...
Ji xi , ( D F ) i j = i
j, Yin ...im =
im
Thus the Schwinger-Dyson equation has the graphic representation
(7.140)
But this just states how the full 1-point function can be understood in terms of full higher n-point func-
tions and bare interactions as encoded in S []: The lefthand side gives the amplitude for a particle
starting at xk to couple to the vacuum via a full quantum process, represented by
Z . According to
the righthand-side, it can either be absorbed by a classical source J at xi , where we must sum over all
possible insertion locations xi , or it can decay first into two or more particles by a tree-level process
before these in turn couple to the full quantum process. The numerical factors are the appropriate
symmetry factors.
In fact, this interpretation of the Schwinger-Dyson equation suggests a very elegant derivation of the
key concepts of Quantum Field Theory in a manner completely reverse to the approach we have taken
in this course: Start with the concept of a free particle as an object that propagates freely in spacetime,
and define the propagator DF i j as the quantum probability amplitude for a free particle to travel from
i to j. Next introduce the action S [] as in equ. (7.134) as nothing but a symbolic means to encode
all perturbative bare interactions Yi1 ...im , i.e. all the ways how an interacting particle can split up into
other particles along its way. Then the Schwinger-Dyson equation in its form (7.140) is an obvious
property of quantum probability amplitudes in the sense described in the previous paragraph. Taking
(7.140) as a starting point and going backwards leads to (7.132) as an equation for Z [ J ], which can be
solved by making the Fourier ansatz Z [ J ] = D Z [] eiJ and solving the corresponding equation
R
More details on this approach can be found in the QFT notes by Cvitanovic and the book by Banks.
and thus
Z[J] Z[J]
= eiW [ J ] , i.e. iW [ J ] = ln . (7.145)
Z [0] Z [0]
To see (7.144) one observes that forming products of all fully connected Greens functions indeed
gives all (potentially partially connected) ones. The factor 1/n! prevents overcounting and is neces-
sary to match the numerical factors in (7.143). For later purposes we define
( x1 , ...xn ) = .... iW [ J ] (7.146)
iJ ( x1 ) iJ ( xn )
and thus
( x1 , ...xn ) J =0 = G (c) ( x1 , ..., xn ). (7.147)
6 Note that the sum starts at n = 1 so that W [0] = 0.
196 CHAPTER 7. PATH INTEGRAL QUANTISATION
Assume that the relation between J and J is invertible such that for a given function ( x) there
exists a unique J ( x) with the property
h| ( x) |i J = ( x). (7.151)
Then we define the 1PI effective action [] to be the Legendre transform of W [ J ], i.e.
Z
[] B W [ J ] J W [ J ] d4 x0 ( x0 ) J ( x0 ). (7.152)
formation of WE [ J ] in the Euclidean theory. All computations can be performed there and the
final results can be Wick rotated back in the end.
Our claim is now that [] is the generating functional for the 1PI connected amputated Greens
functions n ( x1 , . . . , xn ),
Z
X 1
i[] = d4 x1 ...d4 xn ( x1 )( xn ) n ( x1 , , xn ). (7.156)
n=0
n!
1) n=2:
We have
2 ( x1 , x2 ) = i[] = (iJ ( x2 )),
( x1 ) ( x2 ) ( x1 )
(7.160)
2 ( x1 , x2 ) = iW [ J ] = ( x2 )
iJ ( x1 ) iJ ( x2 ) iJ ( x1 )
( x2 )
Z Z
dx2 2 ( x1 , x2 )2 ( x2 , x3 ) = dx2 (iJ ( x3 ))
iJ ( x1 ) ( x2 )
(7.161)
J ( x3 )
= = ( x1 x3 ).
iJ ( x1 )
Therefore
2 ( x1 , x2 ) = 1
2 ( x1 , x2 ). (7.162)
(c)
G2 ( x1 , x2 ) = 1
2 ( x1 , x2 ). (7.163)
We will come back to the interpretation of this relation after the treating the case n = 3.
198 CHAPTER 7. PATH INTEGRAL QUANTISATION
2) n=3
To proceed further it is useful to systematise our computations as follows: First by means of the chain
rule,
(y) 2 iW
Z Z
= d4 y = d4 y
iJ ( x) iJ ( x) (y) iJ ( x)iJ (y) (y)
(7.164)
Z
= d y 2 ( x, y)
4
.
(y)
Next, we introduce the graphical representation
( x1 , ..., xn )
, n ( x1 , ..., xn )
W (7.165)
n n
with n legs (denoted as |n ) respectively. For example our result 2 ( x, y) = 2 ( x, y)1 is graphically
represented as
x f f
W
y =
x f f
1
y (7.166)
and
iJ ( x)
x f f
W
|
y
(y)
{z }
. (7.167)
Integrate over y
Finally, by definition adds a further line to n ( x1 , ...xn ) and ( x) adds another line to n ( x1 , ..., xn ).
iJ ( x)
f f
Now we compute graphically
f f f f
3 ( x, y, z) y W
z
iJ ( x)
f f f f
=x W
y W
z (7.168)
()
1
=x W
y
z .
()
To compute ( x, z)1
() 2
we generalize the identity
! !
d d
M ()1 = 1
MAC M () 1
MDB (7.169)
d AB d CD
Thus
()
y f f f f f f
W
z=y
W
W
z, (7.171)
7.8. THE 1PI EFFECTIVE ACTION 199
where the subscript indicates an extra leg. Putting everything together this just means
y y
x z = x z
Here full circles indicate
. Thus 3 gives indeed the amputated
W , while empty circles indicate
f f fpf
1PI connected 3-point function, as external legs merely carry corrections due to the fully resummed
propagator
G2 = W
| J =0 = (7.172)
This can be generalised (grapically) to higher n and the claim (7.156) is proven inductively for all
n ( x1 , . . . , xn ) with n 3.
Let us come back to the interpretation of 2 ( x1 , x2 ), which is a bit special in the following sense:
Namely, the difference between 2 ( x1 , x2 ) and n ( xi ), n 3 is that n ( xi ), n 3 is the sum of the
tree-level vertex plus of the 1- and higher-loop 1PI amputed corrections, while 2 ( x1 , x2 ) consists of
the 1- and higher loop 1PI amputated diagrams minus the tree-level propagator.
f f f f f f
(c)
This can be seen by noting that relation (7.163), i.e. G2 2 = 1, reads diagrammatically
x W
y=x W
W
y. (7.173)
Up to an important overall minus sign this is of the same form as the identity for n = 3 and higher n.
Note that evaluating (7.173) at tree-level gives
x f f fy= x
2 |tree
y. (7.174)
We can then attach two external lines to 2 ( x1 , x2 ) and consider the resulting object
f f f f f f
f f f
x1 x2 = x1 2 |tree x2 + x1 2 |loop x2
(7.175)
= x1 x2 + x1 2 |loop x2 .
To identify 2 |loop we go to Fourier space. Let us define 2 ( p2 ) as
d4 p
Z
2 ( x1 , x2 ) = 2 ( p2 )eip( x1 x2 ) . (7.176)
(2)4
200 CHAPTER 7. PATH INTEGRAL QUANTISATION
f
In momentum space, by factoring out the external lines from the diagrams, we see that
i p2 m20 i
x1 x2
f f
2
p2 m0 i p2 m20
(7.177)
i
2
i
x1 iM 2 x2 2 2
i 2 ( p ) | loop
p m0 p m20
2
and thus
p2 m20
2 ( p2 ) = i2 ( p2 )|loop . (7.178)
i
(c)
On the other hand, consider the Fourier transform of the fully resummed Feynman propagator G2 ( x1 , x2 )
via
(c)
G 2 ( x1 , x2 ) = =
Z
fpfd4 p (c) 2 ip( x1 x2 )
(2)4 2
G ( p )e , (7.179)
(c)
where G2 ( p2 ) is given by Dyson resummation as
(c) i
G2 ( p2 ) = . (7.180)
p2 m20 M 2 ( p2 )
Here iM 2 ( p2 ) is the value of all amputated 1PI diagrams (i.e. without external legs) evaluated at
(c)
1-loop and higher. Relation (7.163), i.e. G2 2 = 1, implies
In view of (7.178), we identify 2 ( p2 )|loop = iM 2 ( p2 ). Together with (7.175) this justifies the above
claim that 2 ( x1 , x2 ) consists of the 1- and higher loop 1PI amputated diagrams minus the tree-level
propagator.
This interpretation of [] can also be seen in many different ways, one of them being the following:
Define the object W [ J, h] via 8 Z
D e h ([)+ J)
i
eiW [ J,h] B (7.182)
D e h (S []+ J) .
i i
contrast, we had e h W [ J ] = Z 1[0]
R
8 By
7.9. () AS A QUANTUM EFFECTIVE ACTION AND BACKGROUND FIELD METHOD 201
and expand
X
W [ J, h] = hL1 W L [ J ]. (7.183)
loops L
Now consider the formal limit h 0, in which the righthand side becomes just W 0 /h. By the
stationary phase argument, the right-hand side of (7.182) is given by the value of the integrand at the
saddle-point, i.e. by
e h ([]+ J)
i
(7.184)
where the righthand side is a functional of J only. But this is just the inverse Lgendre transformation
that defined [] = W [ J ] J with 0
= J. Thus W [ J ] = W [ J ]. Now, W [ J ] is the full quantum
object, and we thus find that
Z
1
De h (S []+ J) = e h ([]+ J) ,
i i i
e h W [ J ] = (7.187)
Z [0]
where J and [] are related via = J. We can interpret this result as follows:
The integral 1
R
Z [0]
D is responsible for the quantum fluctuations.
Replacing S [] by [] (with such that
R
= J) and working at tree-level (i.e. no D)
anymore) gives the full result.
The equation
= J, i.e. =0 for J=0 (7.189)
S
is the quantum effective equation of motion replacing the classical = 0 in the full quantum
theory.
Consider again Z
D e h (S []+ J) ,
i i
e h W [ J ] = (7.190)
202 CHAPTER 7. PATH INTEGRAL QUANTISATION
R 1
R
where D Z [0]
D is the normalised measure such that
Z
De h (S []+ J) J =0 = 1.
i
(7.191)
If we plug in the relation W [ J ] = [] + J and = J we arrive at
Z
D e h (S [] ()) .
i
h []
i
e = (7.192)
Since ( x) is the vacuum expectation value of the quantum operator ( x), we view f ( x) as the
quantum fluctuation around the background ( x). The relation
Z
D f e h (S [+ f ] f )
i
e h [] =
i
(7.194)
shows that in order to determine [] we must integrate out the quantum fluctuation f ( x).
To evaluate this integral we expand
S 1 2 S
S [ + f ] = S [] + f + f 2 f + higher terms (7.195)
2
and with = S + h K and thus
S K
f = f h f (7.196)
we obtain
2S
S []+ 12 f f h K
Z i 3
h [] f +O( f )
i
e = D f e h 2 . (7.197)
Thus altogether
!
2S
f f h K
Z i 1 3
f +O( f )
[] = S [] i h ln D f e h 2 2 S [] + hK []. (7.198)
Since K [] appears on both sides, one has to solve this equation perturbatively. In particular the 1-loop
correction to is due to the lowest fluctuation terms in the integrand of D f . Since K [] is by itself
R
already 1-loop, the relevant piece is only the quadratic one and
2
12 f hi S2 f
Z
(1-loop)
K = i ln D f e . (7.199)
At 2-loop order we then use the 1-loop result for K [] together with higher-order terms. As an example
consider 4 -theory with classical action
4
Z " #
1
S = d x ( + m )
4 2 2
(7.200)
2 4!
7.10. EUCLIDEAN QFT AND STATISTICAL FIELD THEORY 203
and compute [0 ] up to 1-loop order with 0 =const. The computation will be done in the tutorials.
If we define the effective potential V (0 ) via
[0 ] = VolR1,3 V (0 ), (7.201)
and expand
V (0 ) = V (0) (0 ) + V (1-loop) (0 ) + . . . , (7.202)
the result is
1 2 2
V (0) = m 0 + 40 (7.203)
2 4!
and V (1-loop) (0 ) is given by the Coleman-Weinberg-Potential
2 2
2 i
Z 4 m + k
1 d k 2 0
V (1-loop) (0 ) =
4
ln
2 2
. (7.204)
2 (2) i m k i
The denominator is the contribution from the vacuum bubbles of the free theory, which must be
removed with our normalisation of the path-integral.
so that * Y +
1 Y h
T ( xi ) = Z [ J ] | J =0 . (7.207)
i
Z [0] i iJ ( xi )
Our conventions for the effective action are furthermore
i Z[J]
e h W [ J ] = , [ ] = W [ J ] J . (7.208)
Z [0]
As discussed, we can pass from Lorentzian to Euclidean signature via a Wick rotation replacing
x0 ix4E . In Euclidean space R4 with
with Z !
1 1
S E [] B 4
d xE (E )2 + m2 2 Lint (7.211)
2 2
and
4
2
Z X
J = d x E ( x E ) J ( x E ),
4
( E ) =
2
i . (7.212)
xE
i=1
Furthermore * Y +
1 Y h
T ( x E i ) = ZE [ J ] (7.213)
i
ZE [0] i J ( xE i )
and
1 ZE [ J ] E
, E [] B WE [ J ] + J ,
e h WE [ J ] B = J. (7.214)
Z E [0]
An executive summary of what Quantum Field Theory is about can be given as follows (working
in Euclidean signature):
We start with a classical field ( x) and associate to it the classical action S E []. The field ( x)
arises as the continuum limit description of a system of N harmonic oscillators.
In the classical limit, formally h 0, the field ( x) takes a definite value given by the classical
equation of motion
S E
= 0. (7.215)
For h finite, quantum fluctuations arise. These are encoded in ZE [ J ], where the path integral
takes into account all quantum fluctuations of the field.
What we can compute in the quantum theory are quantum expectation values or correlation
functions. In particular the quantum expectation value of the field in the presence of a source J
is Z
1
D ( x)e h (S E []J ) .
1
( x) J := h( x)i J = h| ( x) |i J = (7.216)
Z [0]
With our definition of WE [ J ] we can compute this as
WE
( x) J = . (7.217)
J ( x)
Remarkably, this is formally exactly the same structure as in classical statistical mechanics or more
precisely in its continuum limit, classical statistical field theory, in 3 spatial dimensions:
Start with a classical field describing some classical observable, e.g. with s(~x) defined as the
magnetisation density of a 3D ferromagnet with energy
Z Z
1
d3 x H (~x) = d3 x (s(~x))2 . (7.220)
2
We can think of this as the continuum limit of a system of spins sn with nearest neighbour
interactions.
At temperature T = 0, the system is in its ground state given by minimizing the energy (in this
case given by s(~x) constant).
For finite T there are thermal fluctuations of s(~x). These are encoded in the partition function
Z
Ds(~x)e d x(H ( s)H (~x) s(~x)) ,
R
3
Z [H ] = (7.221)
where = (kB T )1 and H (~x) is an external source coupling to s(~x) (in the present case of a
ferromagnet this would be a magnetic field).
1
F [ H ] = lnZ, (7.223)
i.e. Z R
d3 x(HHs)
eF [ H ]
= Ds(~x)e . (7.224)
Then
F
M (~x) := hs(~x)iH = . (7.225)
H (~x)
G[ M ] = F [HM ] + M HM , (7.226)
viewed as a functional of M. Then the average magnetisation is determined via the equation
G
= H. (7.227)
M
206 CHAPTER 7. PATH INTEGRAL QUANTISATION
Due to this formal analogy, many problems in statistical field theory are solved in exactly the same
manner as in Euclidean QFT. In particular the structure of perturbation theory and important concepts
concerning symmetry breaking and renormalisation agree.
This can be summarized as follows:
4D QFT 3D statistical theory
( x E ) s(~x)
quantum fluctuation thermal fluctuation
h kB T
h1 WE [ J ]
e e [H ]
F
WE
J = J (~x)
M = HF(~x)
E G
=J M =H
and
{A (t, ~x), B (t, ~y)} = 0 = {A (t, ~x), B (t, ~y)}. (7.229)
We seek a path integral representation for the transition function h(~xF ), t f | (~xI ), tI i where the state
|(~x), ti is defined via
A (t, ~x) |(~x), ti = A (t, ~x) |(~x), ti . (7.230)
Here the non-hatted expression A (t, ~x) represents a classical field. We face the immediate problem
that consistency with (7.229) requires that the classical field A (t, ~x) satisfy
Thus A (t, ~x) cannot take values in C - rather we need the notion of anti-commuting numbers.
Consider first the case of a finite number of degrees of freedom by putting the theory, as usual, on
a lattice of finite lattice spacing corresponding to a transition
and let us omit the spinor indices for the time being. Then we search for classical objects i (t ) with
the property
i (t ) j (t ) = j (t )i (t ). (7.233)
i j = j i (7.234)
as a map from
V V 7 2 V, (7.235)
n
M
A= k V = C V 2 V ... n V. (7.236)
k =0
1 1
a + ai i + ai j i j + ... + ai1 ...in i1 ...in (7.237)
2 n!
with completely antisymmetrc coefficients ai j = a ji etc. By abuse of notation, the i are called
Grassmann numbers - though calling them numbers makes things look more exotic than they
are as the i should really be viewed simply as elements of the abstract Grassmann algebra A. In
particular they are nilpotent,
i2 = 0. (7.238)
Note that a general element of A can always be written as the sum of an even and an odd element.
Two elements A, B A of definite degree sA and sB satisfy
A B = (1) sA sB B A. (7.239)
Thus even elements are commuting and therefore sometimes called bosonic, whereas odd elements
are dubbed fermionic due to their anti-commuting nature.
208 CHAPTER 7. PATH INTEGRAL QUANTISATION
f : A C, (7.240)
1
f () = a + ai i + ... + ai ...i i i , (7.241)
n! 1 n 1 n
one defines differentiation with respect to i as follows:
i j = i j , j a = 0 a C;
i (a1 f1 ( ) + a2 f2 ( )) = a1 1 f1 () + a2 2 f2 () (linearity);
f1 () f2 () = f1 () f2 () + (1) s f1 () f2 ( ) . (7.242)
i i i
For general f1 () we decompose f1 () into its even and odd part and then apply the graded
Leibniz rule by linearity.
For example
( j k ) = i j k ik j . (7.243)
i
Next we define an integral I [ f ()] as a functional of f (). Our applications to the path integral will
make it evident that it is useful to define this functional such that it shares two key properties of the
bosonic integral (in one variable for simplicity)
Z
I [ f ( x)] = f ( x)dx, (7.244)
namely
In this case translation invariance just means invariance under a shift of the integration variable
+ c and implies, together with linearity,
Z
d c = 0 c C. (7.246)
7.11. GRASSMAN ALGEBRA CALCULUS 209
d = 1 such that
R
We normalise
Z
d[a + b] = b = (a + b). (7.247)
di d j = d j di (7.249)
Z
dn i1 ...in = i1 ...in (7.250)
with i1 ...in the totally anti-symmetric -tensor normalised as 1...n = 1. In particular for f () =
a + ai i + ... + n!1 ai1 ...in i1 ...in we find
Z
1
dn f ( ) = ai ...i i ...i = a123...n . (7.251)
n! 1 n 1 n
Only those terms contribute for which the Grassmann integral is saturated such that each di is paired
with precisely one i .
Note that the above definitions ensure that upon integration by parts no boundary terms are picked up,
Z
dn f () = 0. (7.252)
i
As an application we compute the Jacobian that arises for a linear change of Grassmann integration
variables. To begin with, if A is an (n n)-matrix, then we can define 0 = A and find
Z Z Z
d f ( ) =
n 0
d f ( A) =
n
dn a12...n A1i1 ...Anin i1 ...in (7.253)
Z
= a12...n A1i1 ...Anin i1 ...in = det A dn f (). (7.254)
| {z }
=detA
To saturate the Grassmann measure we pick up the term to mth order in the expansion of the expo-
nential and find
Z
1
I= m dn Ai1 i2 ...Ain1 in i1 ...in1 in
2 m!
(7.258)
1
= m i1 ...in Ai1 i2 ...Ain1 in B Pf ( A),
2 m!
where Pf ( A) is the so-called Pfaffian. Note the difference in the definition of the Pfaffian and of the
determinant,
1
det A = i1 ...in j1 ... jn Ai1 j1 ...Ain jn . (7.259)
n!
One can show with standard methods of linear algebra that (Pf ( A))2 = det A and take the positive
sign,
Pf ( A) = det A. (7.260)
Complexification
So far the coefficients a C, but i were treated as real, or equivalently as one anti-commuting degree
of freedom (which is the fermionic counterpart of 1 bosonic real number). One can also consider
complex i and i , viewed as independent degrees of freedom. E.g. one can build such complex
Grassmann variables i , i = 1, . . . , n, out of real Grassmann variables j , j = 1, ..., 2n via
Alternatively, we simply declare i and i to denote independent objects and define the complex
conjugate of a product via
(i j ) = j i . (7.262)
Real case:
r
(2)n
Z
12 xi Ai j x j
dn x e = (bosonic integration),
det A (7.265)
Z
2 i A i j j
1
d e
n
= det A (fermionic integration).
Complex case:
(2)n
Z
dn z dn z ezi Bi j z j = (bosonic),
det B
Z (7.266)
i Bi j j
d d e
n n
= det B (fermionic).
Note the change in the order of dn and dn in the last term in order to accomodate for the (1)
sign in the exponent.
ignoring the spinor components A and taking i = 1. We then seek the states |, ti such that
(t ) |, ti = (t ) |, ti (7.268)
with (t ) a complex Grassmann number. Working at fixed time t and suppressing the t-dependence
we must find the Hilbert space acted upon by and with anti-commutation relations
{, } = 0 = { , }, {, } = 1. (7.269)
In analogy with the analogous bosonic relations [a, a ] = 1, [a, a] = 0 = [a , a ] this system is called
a fermionic harmonic oscillator. We assume (or alternatively show by exploiting boundedness of
the Hamiltonian) the existence of a ground state
and define the state |1i as |0i = |1i. By means of the anti-commutation relations
|1i = |0i = 0,
(7.271)
|1i = |0i = (1 ) |0i = |0i .
We now claim that the sought-after eigenstate |i with |i = |i takes the form
To avoid confusion note that the unhatted is a complex Grassmann number, while the hatted ,
are annihilation and creation operators.
Indeed,
Since |1i = |0i is a Grassmann odd object, we have |1i = |1i . Equivalently, this follows from
the relation = . Either way we find
|i = |i (7.276)
is given by
|i = |0i |1i = e |0i (7.277)
The state |i is the fermionic analogue of a bosonic coherent state |i defined as the eigenstate of the
annihilator,
a |i = |i with |i = ea |0i .
(7.278)
The complex conjugate state h| defined via h| = h| is simply
We will also need the inner product of two fermionic coherent states
and thus
h| 0 i = e
0 .
(7.281)
Finallly the completeness relation in terms of the coherent states is given by
Z
d d |i e h| = 1
(7.282)
hF , tF | I , tI i = hF | eiH (t f tI ) |I i (7.284)
between two fermionic coherent states. We assume that the structure of the Hamiltonian is
H = M H ( , ) (7.285)
for some bosonic, i.e. Grassmann even, M. In fact this will be sufficient to treat the field theoretic
Dirac fermion system we are interested in. As in the derivation of the bosonic path integral we partition
the time evolution operator according to
N tF tI
eiH (tF tI ) = lim eiHt , t = (7.286)
N N
and insert suitable factors of the identity, this time given by
Z
dj d j | j i e j j h j | .
1j = (7.287)
Thus
N
Z Y
dj d j hN +1 | eiHt |N i eN N hN | ... |1 i e1 1 h1 | eiHt |0 i
hF , tF | I , tI i = lim
N
j=1
(7.288)
with |0 i |I i and |N +1 i |F i. Now
The factor e j+1 j+1 cancels with a corresponding factor e j+1 j+1 from the insertion of the identity,
N
Z Y
(j+1 ( j+1 j )itH (j+1 , j )) .
PN
dj d j eN +1 N +1 e
hF , tF | I , tI i = lim j=0 (7.290)
N
j=1
and eN +1 N +1 can be absorbed into the measure to arrive at the fermionic path integral for the Quantum
dt ( i
Z Rt
t H )
i F
hF , tF | I , tI i = D D e tI
. (7.292)
To generalise this result to the Quantum Field Theory of a Dirac fermion we consider the classical
action Z Z
S = d x ( x) (i m0 ) ( x) + Lint
4
d4 x L (7.293)
with
L = ( x)i0 0 ( x) + ( x) ii i m0 ( x) + Lint (7.294)
Note that we can integrate over as opposed to without cost. To project initial and final states to
the vacuum |i, the same trick of replacing
m0 m0 i (7.297)
works with tI and tF . With this understanding we define the generating functional for
fermionic correlation functions as
Z (7.298)
= DDeiS [,]+i+i ,
where the external sources , are Grassmann-valued classical fields. Then for instance
1
h| T ( x1 )( x2 ) |i = Z [, ]|=0= , (7.299)
Z [0, 0] i( x1 ) i( x2 )
where special attention must be paid to the various minus signs.
The same methods as in the bosonic theory can be applied to deduce from this the fermionic Feynman
rules. For example, as will be derived in the tutorials, in the free theory completing the square yields
d4 p i( p + m0 i ) ip( xy)
Z
S F ( x y) = e (7.301)
(2)4 p2 m20 + i
7.12. THE FERMIONIC PATH INTEGRAL 215
is a solution to
Thus
h0| T ( x)(y) |0i = S F ( x y). (7.303)
In QFT I we claimed without proof the following properties of Quantum Electrodynamics (QED):
f fg g
The structure of UV divergences is encoded in the following three types of diagrams:
1PI
, 1PI
,
The four constants multiplying the divergences in these diagrams can be absorbed order by
order in perturbation theory in the counterterms of the renormalized Lagrangian such that all
physical amplitudes are finite. For this recall chapter 5.9 of QFT I.
While we demonstrated the second point at 1-loop order in QED, a number of open questions remain:
1. Why are only the above diagrams relevant as far as the structure of UV divergences is concerned,
and how do these diagrams generalise to other QFTs?
2. How does one prove finiteness of the renormalised theory beyond 1-loop order in perturbation
theory?
217
218 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
nV = 2I + E. (8.3)
Now, naively the structure of momentum integrals in the UV for such a diagram is
R d R
d k1 ... dd kL
(8.4)
k12 ...k2I
R
since each loop gives a momentum integral dd k and each internal line contributes a propagator
scaling like (k2 m2 )1 . Let us define D to be the difference of the power of the momentum in
nominator and denominator. In the above theory
D = d L 2 I. (8.5)
then according to this naive analysis the amplitude M for the process associated with a diagram scales
as follows with :
If D 0, then a diagram may still be actually finite if a sufficient amount of symmetry constrains
the form of the amplitude or leads to cancellations among infinite terms. For example, super-
symmetry leads to such cancellations, and the theory with maximal supersymmetry, N = 4
Super-Yang-Mills theory in 4D, is even finite to all orders.
If D < 0, a diagram may be actually divergent if the divergence is due to a simpler divergent
subdiagram.
8.1. SUPERFICIAL DIVERGENCE AND POWER COUNTING 219
Tree-level diagrams have D = 0, but are finite.
1. The following diagram has D = 0 and is superficially and actually logarithmically divergent,
as we will see later.
2. The following diagram has D = 2 and is superficially and actually finite.
3. The following diagram still has D = 2, but is actually divergent. However, a closer look
reveals that the divergence is due to diagram 1, which appears as a divergent subdiagram with
D = 0.
Conclusions: Except for the issue of divergent sudiagrams, all divergences come at most from the
superficially divergent diagrams.
Given its importance, it is useful to derive an explicit expression for D. In the theory under consider-
ation (n in d dimensions) we have
D = d L 2 I = d ( I V + 1) 2I (8.7)
has mass dimension zero (in natural units with h = 1 = c) and therefore
d2 (n) d2
[] = , [0 ] = n + d. (8.10)
2 2
Consider a diagram with E external lines. To determine the mass dimension of the associated am-
plitude, we note that it would receive a tree-level contribution M(E ) = i(E ) if there were a term
(E ) E in L. Thus
d2
[ M( E ) ] = [ ( E ) ] = d E . (8.11)
2
Furthermore, if M(E ) has superficial degree D, then
(n)
M ( E ) ( o ) V D (8.12)
unless one of the 3 exceptions is at work, where V is the number of n-point vertices. Therefore
(n) d2
V [ 0 ] + D = d E (8.13)
2
and thus !
d2 d2 (n) d2
D = d dn V E = d [ 0 ] V E. (8.14)
2 2 2
The appearance of the mass dimension of the coupling constant is very important as we will see in a
second.
p
E = 0:
fpf
E = 2:
E = 4:
8.2. RENORMALISABILITY AND BPHZ THEOREM 221
Since D is independet of L, superficial divergences appear for these amplitudes at every order
in perturbation theory.
From D = d [(n) ]V d2
2 E (and analogous formulae for other theories) we draw the following
important conclusion:
A theory is
Ignore the issue of divergent subgraphs for a moment. Then if a theory is (power-counting)
renormalisable, at each order in perturbation theory only a finite number of divergent dia-
grams, parametrized by a finite number of divergent constants, appear. One can absorb these
divergences order by order in the counterterms of the renormalized Lagrangian such that all
physical amplitudes are finite.
The counterterms create new Feynman diagrams relevant at the next order. These will cancel
the divergences of the divergent subdiagrams (if present) at the next order.
1 Due to Bogoliubov, Parasiuk, Hepp, Zimmermann.
222 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
All of this leads to a perturbative adjustment of a finite number of counterterms in the renor-
malised Lagrangian, and thus predictivity is maintained.
The non-trivial aspect of this theorem concerns the complete cancellation of divergent subdiagrams by
lower-order counterterms. Rather than presenting a full proof, which can be found in Zimmermanns
1970 article (see also Kugos book), we will demonstrate this in the next section in the context of
renormalisation at two-loop order. For simplicity we do this for 4 theory.
p
the propagator with E = 2, D = 2:
fafpfaf
the 4-point amplitude with E = 4, D = 0:
Now comes a simple, but important observation regarding dimensional analysis that we will make
frequent use of: In a theory with a dimensionless coupling constant the dimension of the amplitude
agrees with D, see eq. (8.12). We can therefore parametrise the second amplitude with D = 2 in
fafpfaf
terms of the momentum cutoff as
This is to be viewed as a Taylor expansion in p with dimensionless constants a, b, c. Note that due to
the symmetry p p of L in momentum space b 0.
By the same logic, the 4-point amplitude with D = 0 can be parametrized (with d dimensionless) as
8.3. RENORMALISATION OF 4 THEORY UP TO 2-LOOPS 223
with Z the wavefunction renormalisation and rewrite L as
1 1 1 1
L = ( r )2 m2 2r 4r + Z ( r )2 m 2r 4r (8.18)
2 2 4! 2 2 4!
where Z = Z 1, m = m20 Z m2 and = 0 Z 2 . We henceforth compute with the renormalised
F FxF
Feynman rules
p p p
i
p2 m2 +i
, i ( p 2 Z m )
i, i
and no factors of Z for in- and out-going states in the computation of scattering amplitudes. To
FpF
define the meaning of m and we impose the following renormalisation conditions:
p p
p3
!
= i
p2 m2
+ terms regular at p2 = m2
p1
p2 = i at s = 4m2 , t = u = 0
p4
amp.
Here s, t, u are the Mandelstam variables, which read with the above convention for the arrow of the
momenta,
s = ( p1 + p2 )2 , t = ( p1 p3 )2 , u = ( p1 p4 )2 . (8.19)
fpf
Now
i
= , (8.20)
p2 m2 M 2 ( p2 )
where iM 2 ( p2 ) is the amplitude of the 1PI amputated diagrams. Thus the first renormalisation
condition implies:
! d !
M 2 ( p2 ) p2 =m2 = 0, 2
M 2 2
( p ) p2 =m2 = 0. (8.21)
dp
Note that the renormalisation conditions are arbitrary and define the renormalisation scheme.
224 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
f f
1PI
= + fxf
f f
and thus2
1
Z
d4 k i 2 (1) (1)
1PI
= (i) + i ( p Z m ) iM 2 2
( p ) 1-loop
. (8.22)
2 (2)4 k2 m2 + i
We compute the integral by dimensional regularisation. The result (see exercise sheet 5 for a deriva-
tion) is
Z
d kd
i 1 1 d2
= . (8.23)
(2)d k2 m2 (4)d/2 (m2 )1d/2
For d = 4 and with the help of
2 1
1 + = e 2 1 + + O( 3 ) (8.24)
2 2
we get
+ 1 log(m2 ) (e 4) 2 ,
(m2 )1/2 1 1
= i 2 1 + 2 (4)2/2 e 2 = i
2 2
2 162 m
where we used
2
(m2 )/2 = e 2 m = 1 log(m2 ) + O( 3 ). (8.25)
2
Note that the expected quadratic divergence a2 corresponds, in dimensional regularization, to the
divergent term
i 2 1
m . (8.26)
162
More precisely, for the integral with a UV momentum cutoff, the divergence would be proportional to
m2 + 2
2 m2 log . (8.27)
m2
As a rule of thumb, in a theory with dimensionless coupling constant, the order in corresponding
to a 1 -pole in dimensional regularization agrees with the mass dimension of the prefactor: Here
[ m2 ] = 2 and thus the divergence is expected to scale as 2 .
2 The 1
factor of 2 in front of the integral is a symmetry factor.
8.3. RENORMALISATION OF 4 THEORY UP TO 2-LOOPS 225
Altogether
i 1 1 d2
2 (1) (1)
iM 2 ( p2 ) = + i p Z m (8.28)
2 4d/2 (m2 )1d/2
and the renormalisation conditions imply
d ! (1) !
2 2
M ( p ) = 0 Z = 0 (8.29)
dp2 p2 =m2
and
1 d2
! ( 1 ) !
M 2 ( p2 ) = 0 m =
. (8.30)
p2 =m2 2(4)d/2 (m2 )1d/2 d4
Note that at 1-loop M 2 ( p2 ) 0 p2 and the first non-trivial corrections to the propagator appear at
the 2-loop level. This, however, is a specialty of 4 theory in four dimensions.
The computation of the 4-point amplitude at 1-loop level involves the following diagrams,
p3 p3 p1 p3
p3 p1
p1 p p3 p1
= p1 i + p4 + p2 + +
p2 2
p4 p4 p2 p4 p2 p4
p3
p1
p2 (1) + O ( 3 )
p4
p1 p3
d4 k
R
= 12 (i)2 i i
(2)4 k2 m2 (k+ p)2 m2
B (i)2 iV ( p2 )
p4
p2
where p = p1 + p2 . Then
(1)
iM( p1 p2 p3 p4 ) = i + (i)2 [iV ( s) + iV (t ) + iV (u)] i + O(3 ). (8.31)
Therefore
(1) ! (1) (1)
= 2 V (4m2 ) + 2V (0) ( s) + (u + t ). (8.33)
226 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
The computation of V ( p2 ) proceeds again via dimensional regularisation and is performed on exercise
sheet 5, with the result
1 (
Z 1
2 1
V(p ) = dx 2
2 (4)2/2 0 [m x(1 x) p2 ]
Z 1 " # (8.34)
1 2
= dx + log(4) log(m x(1 x) p .
2 2
322 0
Then
2 1
Z " #
(1) 6
= 2 2
dx 3 + 3 log(4) log(m (1 4x(1 x))) 2 log(m ) . (8.35)
322 0
Z1
i2 m2 x(1 x) s
" !
iM( p1 p2 p3 p4 ) = i dx log 2
322 m x(1 x)4m2
0 (8.36)
m2 x(1 x)t m2 x(1 x)u
! !#
+ log + log .
m2 m2
This is manifestly finite and depends on two measurable input parameters m and .
by the 1-loop counterterms that we have already included.
at 2-loop-level.
(1)
the 1-loop counterterm to the propagator in the second diagram cancels the extra divergence in
the first diagram. In fact, as a speciality of 4 -theory, we have seen that, at 1-loop level,
8.3. RENORMALISATION OF 4 THEORY UP TO 2-LOOPS 227
+ =0
+
plus the corresponding t-channel diagrams
+...
plus the u-channel diagrams
+...
s t+u
We consider only the s-channel, since t-and u-channel work completely analogously. The im-
portant point to notice is that all momentum independent divergences can be absorbed into
228
CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
(2)
One must therefore show that the momentum dependent divergences cancel among the five s-
channel diagrams. Following Peskin-Schrder, they can be grouped into
group 1:
p1
p3 p1 p3 p1 p3
+ (1), s + (1), s
p4
p2 p2 p4 p2 p4
group 2:
and group 3:
Note that group 1 contains only the s-channel 1-loop counter terms, while in group 2 and 3 the
combined (t + u)-channel counter terms appear. The divergences of the diagrams of group 1
are very easy to deal with: In terms of our previous definition
p1 p3
= (i)2 iV ( p2 )
p2 p4
8.3. RENORMALISATION OF 4 THEORY UP TO 2-LOOPS 229
with p = p1 + p2 = p3 + p4 we have
p1
p3
2
= (i)3 iV ( p2 )
p4
p2
and
p1 p3 p1 p3
(1), s = (1), s = (i)iV ( p2 )2 iV (4m2 )
p2 p4 p2 p4
(1)
For the counter term diagram we used our result that i ( s) = 2 iV (4m2 ) . Adding both
contributions up and completing the square leads to
2 2
(i)3 V ( p2 ) V (4m2 ) + V (4m2 ) (8.37)
(2)
s
2
Note that 2 contains a divergence of the form 2 . More generally, higher loop-order diver-
gences give higher powers in 1/.
d4 k
Z
3 i i
2
= (i) iV ( p3 + k ) (8.39)
(2)4 k2 m2 (k + p)2 m2
with p = p1 + p2 = p3 + p4 . This diagram exemplifies the appearance of a nested singu-
larity, which appears when 2 divergent subdiagrams share a common propagator. In this case,
so-called non-local divergences appear, for which the prefactor of the singularity is not a sim-
ple polynomial in p, but a more complicated function in p.
Integrals of this kind would keep us from the street for a while. We do not explicitly perform
the computation, which is straightforward, but a little lengthy, and merely quote the result from
Peskin-Schrder, Chapter 10: When the dust has settled, the amplitude can be written as the
sum of 2 terms:
Term 1 has only a momentum independent singularity, which can be absorbed into the
2-loop counter-term.
Term 2 is given by
Z1
i3 2
" #
1
dy + log(4) log(m2 y(1 y) p2 ) (8.40)
2(4)4
0
2
log(m2 y(1 y) p2 ) (8.41)
is the non-local singularity advertised above. The factor of 2/ is due to the regime in
momentum space where only 1 of the subdiagrams diverges, and the log( p2 ) represents
the finite QFT process around this from the other subdiagram.
Explicit computation confirms that these singularities are cancelled by the counter-term
diagram
(1), t + u
This completes the proof that at 2-loop order, all divergences, including those of divergent subdia-
grams, can be absorbed into the counter-terms. A formalization of this procedure leads to the BPHZ
theorem. More on the ideas behind its proof can be found e.g. in Kugo.
8.4. RENORMALISATION OF QED REVISITED 231
2
p
In particular, since [e] = 0 the theory is renormalisable. The following amplitudes have D 0:
gpg fafpfaf
1. The vacuum energy with D = 4;
3. The 1-photon amplitude pg with D = 3;
Amplitudes 3. and 4. vanish to all orders due to the discrete symmetry of charge conjugation:
LQED is invariant under
j j and A A (8.44)
with j . Now, each external photon couples via its current j , e.g.
232 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
Diagram 5. is non-zero, but actually finite as a consequence of gauge symmetry. This can be
shown by exploiting the Ward identities.
Note that diagram 5. is responsible for the non-linearity of QED due to the scattering of photons
with each other induced by loop effects.
Thus as claimed in QFT I, as far as the UV divergences are concerned it suffices to consider the
diagrams 2. (the contribution of 1. is easily absorbed into the vacuum energy term). But again
symmetry lowers the actual degree of divergence for the first two diagrams of 2. to logarithmic ones:
Ffpfaf
By dimensional analysis, the fermion propagator, having D = 1, can be Taylor-expanded as
p
= A0 + A1 6 p + A2 p2 + ... (8.45)
with
A0 = a0 , A1 = a1 log , A2 and higher: finite as (8.46)
and a0 , a1 dimensionless. But then the mass shift of the electron would be proportional to .
Instead this shift must be proportional to the electron mass me itself. This is because if me 0,
then the theory possesses chiral or axial symmetry, because the Lagrangian can be written as
L = L i L + R i R e j A (8.47)
and is invariant under independent rotations of the left- and right-chiral spinors
L ei L , R ei R . (8.48)
This symmetry, however, forbids a mass term, which must be of the form me R L . Thus A0
must contain an explicit power of me . By dimensional analysis there is no more room for a
factor of , and A0 can at worst be of the form
A0 = a0 me log , (8.49)
GgpgG
The photon propagator is, by dimensional analysis, of the form
q q
= B0 + B1 q + B2 q2 + ... (8.50)
with
B0 = b0 2 , B1 = b1 , B2 = b2 log (8.51)
gpg
and bi dimensionless. But by the Ward identities we know that
= q2 q q (q2 )
(8.52)
and thus b0 = b1 = 0.
8.5. THE RENORMALISATION SCALE 233
to identify with the mass m of because our renormalisation conditions have been
d
M 2 ( p2 ) p2 =m2 = 0 = 2 M 2 ( p2 ) p2 =m2 (8.53)
dp
and
= i at scale s = 4m2 , t = u = 0
More generally, the renormalisation scale can be an independent scale since the renormalisation
conditions are arbitrary. Since this observation is so crucial for all that follows we would like to
illustrate it very explicitly. To this end let us consider the massless theory
1 0
L= ()2 40 (8.54)
2 4!
in d= 4. Note that this theory has no classical scale since m0 = 0. However, as we will now see, the
scale-invariance of the classical theory is broken in the quantum theory due to the appearance of the
renormalisation scale. For instance, we can impose the renormalisation conditions
1. M 2 ( p2 ) = 0 at p2 = 2 , i.e. for spacelike momenta,
d
2. dp2
M 2 ( p2 ) = 0 at p2 = 2 ,
p1
p3
= i at ( p1 + p2 )2 = ( p1 + p3 )2 = ( p1 + p4 )2 = 2
p4
p2
i.e. Z iZ
d4 x eipx h| T 0 ( x)0 (0) |i p2 =2 = 2 p2 =2 . (8.56)
p
In particular, the wavefunction renormalisation factor Z used in = Z 1/2 0 is not the residue of the
propagator at the physical mass p2 = 0, but has the meaning as given in the previous equation.
Now, to fully appreciate the appearance of the renormalisation scale , consider the renormalised
connected 4-correlator
G (4) ( x1 , .., x4 ) = h| T ( x1 )...( x4 ) |i connected . (8.57)
4
h iY i
G (4) ( p1 , .., p4 ) = i + (i)2 (iV ( s) + iV (t ) + iV (u)) i + O(3 ) . (8.58)
p2
i=1 i
Note the product 4i=1 pi2 from the external legs because we are computing the 4-point correlator, not
Q
i
the scattering amplitude iM( p1 p2 p3 p4 ). Renormalisation condition 3. implies
2 Z1
2 d2
3
= (i)2 3V (2 ) = dx
2(4)d/2 ( x(1 x)2 )2d/2
0 (8.59)
32 2
" #
= log + finite terms
2
2(4)2
in dimensional regularization, whereas in cutoff regularisation the term in brackets would be of the
form
2
log 2 + finite. (8.60)
The crucial observation is that G (4) ( p1 , ..., p4 ) is finite, but
(At higher loop, there will also be an implicit -dependence via Z as we will see below.)
(0)
Y
Gn ( x1 , ..., xn ) = h| T 0 ( xi ) |i conn. (8.61)
i
8.6. THE CALLAN-SYMANZYK (CS) EQUATION 235
computed from the bare Feynman rules in terms of the parameters 0 , m0 appearing in the bare La-
(0)
grangian L0 as well as the cutoff (or ). There is no explicit dependence of Gn ( x1 , ..., xn ) on the
renormalisation scale , i.e.
(0) (0)
Gn ( x1 , ..., xn ) = Gn ( x; 0 , m0 ; ) (8.62)
d (0)
Gn ( x; 0 , m0 ) , m fixed = 0. (8.63)
d 0 0
Here it is important that we do not vary with respect to 0 and m0 , but only state that no explicit
dependence on is given. The reason we stress this is that in bare perturbation theory, 0 and m0 must
themselves depend on the cutoff in such a way as to cancel the divergences, and this definition of the
bare parameters in turn involves a renormalisation scale (cf the discussion in section 5.9.1). However,
this dependence on is only implicit via 0 and m0 , and as long as we do not include variation with
respect to these bare parameters, equation (8.63) is correct.
Let us compare this with the renormalised correlator
Y
Gn ( x1 , ..., xn ) = h| T ( xi ) |i conn. (8.64)
i
with
( xi ) = Z 1/2 0 ( xi ) (8.65)
computed via the renormalised Feynman rules including counter-terms. As demonstrated above,
Gn ( xi ) depends on the renormalisation parameters , m and the renormalisation scale , but not on
the cutoff,
Gn ( x1 , ..., xn ) = G ( x; , m; ). (8.66)
Now, going from the bare Lagrangian to the renormalised Lagrangian is only a reformulation of the
same theory. Thus the bare and the renormalized correlators must agree up to the rescaling of fields
= Z 1/2 0 , i.e.
(0)
Gn ( x1 , ..., xn ) = Z n/2Gn ( x1 , , xn ). (8.67)
To avoid confusion, we stress that if we compare scattering amplitudes iM, the amplitudes computed
in bare and renormalised perturbation theory agree completely since we compensate for the rescaling
of by omitting factors of Z 1/2 for external states in LSZ. This is not so in the transition from Gn to
(0)
Gn , and thus the extra factor Z n/2 appears.
d n/2
Z Gn ( x; , m; ) ,m fixed = 0, (8.68)
d 0 0
236 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
where the extra factor of has been included for later convenience. This can be further evaluated with
the help of the chain rule. In terms of the quantities
d
B ,m ,
d 0 0
d
m2 B m2 ,m , (8.69)
d 0 0
1 dZ d
B ,m = log Z ,m
2 Z d 0 0 2 d 0 0
While the higher order terms in , m2 , will depend on the renormalisation scheme, to
leading order in the coupling constant the answer turns out to be universal.
Before embarking on a more systematic analysis, consider massless 4 theory in the renormalisation
scheme of section (8.5):
The 2-point function satisfies
" #
+ + 2 G2 ( p) = 0. (8.72)
At 1-loop order, which for the 2-point function is O(), we have seen that as a speciality of
4 -theory
M 2 ( p2 ) = 0, (8.73)
i.e.
G2 1loop = 0 = G2 1loop . (8.74)
Thus = O(2 ).
3 Cf. renormalisation condition 3. in the previous section, which defines at scale .
8.7. COMPUTATION OF -FUNCTIONS IN MASSLESS THEORIES 237
the momentum independent quadratic divergence can be absorbed into the counter-term m . Thus in
the sequel we only keep the logarithmic divergences and the counter-term Z .
G 2 ( p) = ff
The 2-point function takes the form
i i
+ 1 loop +
p2 2
fxf
!
i
(1)
+ ...
i i (8.82)
= 2
+ 2 A log 2 + finite 2 + 2 ip2 Z 2 + ...
p p i p p p p
with
2
Z = A log + finite (8.83)
2
such that G2 ( p) is finite. We observe that
4 The modification to a fermionic theory with dimensionless couplings is straightforward.
238 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
As we will show below, () = O(2 ). Therefore at leading order the CS equation implies that
Z + 2 = 0 (8.84)
and thus
1
= Z = A. (8.85)
2
where the external legs only contribute to G, not to the scattering amplitude. This can be
parametrized as
Y i 2 X 2
Gn =
2
i iB log 2
i + ( i )
Ai log 2
Z i
, (8.87)
p
i
pi i
pi
P
where i extends over all n external legs. To solve the CS equation
n
X
+ i Gn = 0
+ (8.88)
i=1
we note that
only , Zi contribute to Gn and that
all terms except i are of order 2 . So to leading order
Y i
Gn = ((i) + O()) . (8.89)
i
p2i
To leading order in therefore
X X1
Zi + () + Z = 0. (8.90)
i
2 i
i | {z }
= i
Since
2 2
= B log + finite, Zi = Ai log + finite (8.91)
2 2
the -function becomes
1 X X
() = + Zi = 2B Ai . (8.92)
2 i i
To leading order the finite terms do not enter the expressions for () and (). Since the different
renormalisation schemes only differ by the finite terms, the 1-loop results are independent of the
scheme. At next-to-leading order, the -functions are scheme-dependent.
8.8. THE RUNNING COUPLING 239
As an application we compute the 1-loop -function in QED. Recall our conventions for the counter-
terms appearing in the renormalisation of QED as discussed in QFT 1:
3 = Z3 1 photon renormalisation
(8.93)
2 = Z2 1 fermion renormalisation
and
= ie1
e2 e3
2 = 3 = + O ( e 3
) , = + O(e4 ) (8.96)
122 122
in Feynman gauge.
iM( p1 p2 p3 p4 ) = i at s = t = u = 2 . (8.97)
The so-defined coupling is therefore really a function of the renormalisation scale , = (), and
this () is the effective coupling relevant for processes with typical momenta of order .
To appreciate this latter point, recall that iM( p1 p2 p3 p4 ) in the above theory is
2
" #
s t u
iM = i i log 2 + log 2 + log 2 + O(3 ) (8.98)
322
with = () as in (8.97).
240 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
For momenta s, t, u 2 , the loop-corrections are small, and the tree-level result i() is a
good approximation. In this sense, () is a measure for the strength of the interaction. This is
what we call the effective coupling for processes at momentum scale .
At scales |s|, |t|, |u| 2 (or 2 ), the 1-loop correction becomes big due to the appearance of
a so-called large log. We see that perturbation theory is an expansion not only in , but also in
2
log p
2
with p a typical momentum/energy scale of the process.
Now, the evolution of this coupling () as we vary is precisely what is encoded in the -function
through
d d
() = () = (). (8.100)
d d log
The change in () as we change is called renormalisation group (RG) flow for reasons that will
become clear in section (8.10). The above RG flow equation describes the running of the coupling in
the above sense.
Being a first-order differential equation, (8.100) uniquely specifies () in terms of the value of at a
references scale . Suppose therefore that = ()= as determined by one measurement at .
Then (8.100) is solved by
() Z
d0
Z
= d log 0 = log . (8.101)
( 0 )
Observe that according to this analysis, () increases as increases and even seems to diverge at
= exp 2
3
.
8.8. THE RUNNING COUPLING 241
Alternative RG bevaviours
Depending on the sign of the -function there are 3 qualitatively different types of behavior for the
RG flow of the coupling as well as hybrids thereof. Let us first turn to the extreme cases:
On the other hand, the theory is perturbatively well-defined in the IR, where is small. For
instance, if () 0 as 0 as would be the case e.g. for () = c x with c > 0, then
() 0 as 0. The coupling thus approaches an IR fixed point, and the theory becomes
free in the IR. The fixed point = 0 is therefore called Gaussian fixed point.
Perturbatively () > 0 for 4 theory in d = 4 and for U (1) gauge theory in d = 4, but the
non-perturbative status is less clear.6
Apart from these three extreme cases, it can happen that () interpolates between 1., 2., 3., i.e.
changes sign and thus hits zero at some value.
5 In the less severe case, where as one might consider it as a matter of taste whether or not one should
call the theory well-defined.
6 A numerical analysis via lattice field theory seems to suggest that the Landau pole in both cases is physical and not just
If () > 0 to lowest order in , but higher order terms, relevant for bigger and thus (since we
started with () > 0) for larger , have a negative sign, then () becomes 0 at = :
*
()
where is here the suitably rescaled (dimensionless) quartic scalar coupling. Thus for d < 4
the theory approaches the celebrated Wilson-Fisher fixed point > 0 in the IR.
Near a non-trivial fixed point , the 2-point function of a scalar theory in d = 4 behaves as
!1 ( )
1
G 2 ( p) = C 2 . (8.105)
p
This can be proven by integrating the Callan-Symanzyk equation and using approximate scale - inde-
pendence near the fixed point. For a derivation we refer to the tutorials. The propagator (8.105) looks
like the propagator of the free theory,
1
G2 ( p)free 2 , (8.106)
p
except for the different power (1 ). Consequently is called anomalous dimension of since
near , the propagator G2 ( p) behaves like in a free theory with a field of mass dimension (1
( )). More generally, such power-law behaviour of the propagator is a sign of scale-invariance (as
realised near = where = 0).
where Odi is some local operator7 of mass dimension di , and treat it as a perturbation. Then define i
via the generalised Callan-Symanzyk equation:
" #
0 = + + n + i Ci Gn ( x; , Ci ; ). (8.108)
Ci
Compared to we have scaled out one power of Ci in i Ci Ci so that i is dimensionless. Clearly
i 0 as 0 and Ci 0.
Now, the coupling Ci has mass dimension
[Ci ] = d di , (8.109)
Ci = gi ddi . (8.110)
7 By this we mean any product of the fundamental fields or their derivatives.
244 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
" #
0 = + () + n + (i + di d )gi Gn ( x; , gi ; ), (8.111)
| {z } gi
=:i
" #
0 = + () + n + i Gn ( x; , gi ; ). (8.112)
gi
gi gi ( ) for (8.116)
in a perturbative theory.
gi () gi ( ) (8.117)
in a perturbative theory.
Non-renormalisable couplings become irrelevant in the IR: Even if they are present at higher
scales, they become small for smaller energy processes. On the other hand, super-renormalisable
couplings become relevant in the IR.
But it is important to keep in mind that if the coupling is big, then i cannot be neglected in (8.113).
In principle, this can change the qualitative behaviour of the flow.
8.10. WILSONIAN EFFECTIVE ACTION & RENORMALISATION SEMI-GROUP 245
For example, as already noted, in 4 theory in d dimensions the rescaled dimensionless coupling
behaves as
2
() = (d 4) + . (8.118)
162
For d < 4 is super-renormalisable, but for big values of due to the second term () becomes
positive and changes even the sign of the -function.
The cutoff is merely a way to regulate divergent integrals without any physical meaning.
This comes at the cost of losing predictability for a number of physical masses and coupling.
In a renormalisable theory, only a finite number of such physical couplings must be taken as
input parameters from experiment, and we end up with a well-defined theory.
The so-called Wilsonian approach gives a rather different interpretation to renormalisation, inspired
by analogies with condensed matter and statistical physics:
For instance, consider as our field the magnetisation density s(~x) of a metal in 3D statistical
field theory. This s(~x) is the continuum limit of the spin degrees of freedom si at lattice site i of
the atomic lattice. There is an intrinsic cutoff given by the lattice spacing a:
At distances smaller than the lattice spacing a, no concept of continuous magnetisation exists
- the continuous field theory description is strictly speaking invalid. The appearance of diver-
gences in bare perturbation theory as we take a 0 reflects this mismatch of the field theory
with the microscopic degrees of freedom.
Nonetheless, via renormalisation, one can take the continuum limit, but at the cost of having to
take the couplings as input parameters. These parametrise our ignorance of the true physics at
distances a from the perspective of the field theory.
Applied to relativistic QFT, one would similarly argue that beyond some high energy scale 0 ,
QFT fails to capture the microscopic degrees of freedom. In particular, all matter gravitates,
246 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
and the effects of gravity become non-negligible at the (reduced) Planck scale
1
Mpl = = 2.43 1018 GeV. (8.119)
8G N
The observed UV divergences in bare perturbation theory result from the breakdown of our
QFT beyond this scale. Still, in a renormalisable and asymptotically free (or at least UV safe)
theory one can take the cutoff 0 , but at the cost of parametrising our ignorance of the
true degrees of freedom in a finite number of parameters.
In the modern Wilsonian interpretation of Quantum Field Theory, we should view the QFT
as an effective theory. E.g. in a non UV-finite QFT it is impossible to predict the vacuum
energy V0 - this is one of the input parameters. To explain the mystery of Dark Energy, we need
a more fundamental, UV finite theory. To date, the only known theory that is UV finite and
asymptotes to a weakly coupled effective Quantum Field Theory in the IR is String Theory. In
String Theory, the concept of pointlike particles is abandoned in the UV: All particles are in fact
excitations of a 1-dimensional string of length ` s . At distances above ` s we can approximate the
theory by a point particle theory, i.e. a QFT as we have studied it. Thus the string length is the
intrinsic cutoff of the low-energy effective QFT, and at distances smaller than ` s the theory flows
to a rather different (in particular non-local) description without UV divergences and without
free dimensionless input parameters.
After this general outline we now demonstrate quantitatively how the high energy degrees of freedom
are indeed captured in the effective coupling constants at lower energies. This will give a physical
meaning to the running of the couplings. In view of the above motivation let us start with a bare
Lagrangian in d Euclidean dimensions, e.g.
1 1
L= ()2 + m2 2 + 4 (8.120)
2 2 4!
and impose a momentum cutoff 0 : In path integral quantisation, this means we restrict the measure
- written in Fourier space - to the Fourier modes (k) with 0 |k| 0 , i.e. define the theory via
Z R
d
Z = [D]0 e d xE (L+ J) , (8.121)
where [D]0 |k|0 d(k). Note the specific combinations of minus signs and absence of factors
Q
of i in the Euclidean theory. This partition function enables us to compute amplitudes up to a typical
energy scale 0 of external particles. Suppose now that we are interested in processes at typical
energy scales below
= b0 0 < b < 1. (8.122)
As we saw in the previous picture, loop computations will give rise to large log-corrections if 0 .
To avoid these, we can integrate out the degrees of freedom between and 0 once and for all
and work with a path integral defined only in terms of the remaining degrees of freedom below . To
compute this new effective path integral we first split each Fourier mode as
where
0
if 0 |k| b0
(k) =
(8.124)
(k)
if b0 |k| 0
and
(k)
if 0 |k| b0
(k ) =
(8.125)
0
if b0 |k| 0 .
Taking J = 0 for brevity, we get
Z
Z = D D e d x[ 2 (+) + 2 m (+) + 4! (+) ]
R 1
d 2 1 2 2 4
Z
4
= D e d x( 2 + 2 m + 4! )
R 1
d 2 1 2 2
(8.126)
Z
4 3 2 2
D e d x[ 2 + 2 m + 4! + 6 + 4 + 6 ] ,
R 1
d 2 1 2 2 3
where terms are omitted due to orthogonality of different momentum modes. The partition function
is of the form Z Z
Z = [D]b0 eS [] [D]b0 |k|0 eS [,] . (8.127)
To this end we view all terms in S [, ] except the kinetic term as a perturbation, which is justified for
1 and m = b0 . (8.129)
Then perturbatively
4 3 2 2 3
Z Z " Z 1 #
S [,] dd x 12 2
R
D e = De 1 d x m + + + + + . . .
d 2 2
2 4! 6 4 6
h[...]i.
(8.130)
This is computed via Wicks theorem in perturbation theory, where we contract only pairs of , and
in the resulting propagator only momenta b0 |k| appear. Let us sketch how this works:
d 2 2 d 2
* Z + Z Z !
dd x 21 2
R
d x = De d x ( x) ( x) 2
4 4
Z
1
= dd x 2 ( x) ( x)( x)
2 2 | {z } (8.131)
R dd p
eip( xx) 12
=
(2)d p b0 |p|0
Z !
1 2
= d x ( x )
d
2
248 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
with
dd p 1 1 bd2 d2
Z
= = . (8.132)
2 (2)d p2 b0 |p|0 (4)d/2 (d/2) d 2
Note that in the Euclidean theory no factor of i appears in the propagator. Graphically one
represents the -propagator with two extra lines:
( x)(y) x y
2 ( x)2 (y)
( x) ( x)
Together with n-fold powers of that same term (which arise from expanding the exponential in
e[S (,)] ) this just gives a correction to the mass term present already in eS [] .
2
1 2 2 2 2
D E
4 4 4 +
The first diagram is just the square of the mass correction, as advocated. The second term is
schematically of the form
Z Z
dd x dd y 2 ( x)2 (y) ' dd x(4 ( x) + c 2 ( x)(( x))2 + . . . (8.133)
The first summand provides a contribution to the 4 coupling in S [], while the second gives a
new, derivative interaction not present in our original action.
3. Likewise, also higher order interaction terms in appear, e.g. the contraction
1 3 3
D E
4 6 6
This way, all possible interactions are created in S eff which are consistent with the symmetries of the
bare Lagrangian (here just ), and we find
Z
b0
Z = [D]b0 eS eff (8.134)
with
Z Z "
1 1 1
d
d x Lb
eff
0
= d
d x (1 + Z ) ()2 + (m2 + m2 ) 2 + ( + ) 4 (8.135)
2 2 4!
#
+ C ( )4 + D 6 + ... . (8.136)
b0
The object S eff is called Wilsonian effective action. The coefficients Z, m, , C, D can be
systematically computed in perturbation theory. They receive contributions only from the momentum
regime b0 |k| 0 . When we compute correlators at scales below = b0 via the Wilsonian
effective action (8.134), only momenta |k| = b0 appear in the loops since all effects of the
modes with b0 < |k| < 0 are already encoded in Leff .
The Wilsonian effective action is not to be confused with the quantum effective action []. As
discussed in section 7.9, the latter includes all quantum effects, i.e. computation with [] at tree-
b0
level (no loops!) gives the full quantum theory. By contrast, in the Wilsonian effective action S eff
only the quantum effects due to modes from b0 < |k| < 0 are included, and in computations with
b0
S eff loops with momenta 0 |k| = b0 must still be performed.
For better comparison of Lb 0 0
eff with the original Lagrangian, which we denote now by L , we bring
the kinetic term to standard form by rescaling
0 = (1 + Z )1/2 . (8.137)
and
1 02 02 0 04
Z " #
1
S 0 eff = d x ( ) + m + + C ( ) + D + ...
d 0 2 0 0 4 0 06
(8.139)
2 2 4!
for
Note that in our case C = D = 0 in the original L0 . Thus the effect of integrating out the degrees
of freedom with b0 < |k| < 0 amounts to renormalising the couplings in the Lagrangian as above.
250 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
The successive application of this operation gives rise to the renormalisation group. This is really
only a semi-group, since this procedure only allows us to lower the cutoff by integrating out degrees
of freedom, i.e. there does not exist an inverse element.
The running of the couplings in the Wilsonian picture is to be interpreted as the evolution of the
cutoff dependent couplings in the effective action as we change the cutoff. To make a connection to
our previous picture we must identify the cutoff with the renormalisation scale .
I.e. the effective couplings () are defined by specifying observables computed from the effective
action with cutoff . This replaces our previous renormalisation condition fixing ().
With this identification, the fundamental form of the running couplings agree to leading order in per-
turbation. This is all we can hope for, because, as always, beyond leading order the specific scale
dependence of the coupling depends on the specific renormalisation scheme. E.g. for 4 theory in
d = 4 carrying out the Wilsonian renormalisation procedure as sketched above gives
32 1
Z 1-loop = 0, 1-loop = 2
log . (8.141)
16 b
With b = /0 < 1 this gives for small |1 b|
3 2
( ) = ( 0 ) + (0 ) log , (8.142)
162 0
in agreement with our previous results for the -function of 4 -theory. More generally, for a dimen-
R
sionful coupling dd x Cdi Odi with
In fact, this is just as expected by dimensional analysis. As in our previous approach it is useful to
define a dimensionless coupling, but now in terms of ,
gi () = Ci ()di d , (8.144)
dominates and thus if Ci is non-renormalisable ([Ci ] < 0), the initial value of gi (0 ) is irrelevant at
scales 0 and
Ci () = O di /2 [Ci ] [Ci ] (8.147)
for 1. This agrees with our conclusions from the Callan-Symanzik approach: Non-renormalisable
couplings become irrelevant in the IR: Even if they are present at higher scales, they become small
for smaller energy processes. On the other hand, super-renormalisable couplings become relevant
in the IR.
Conclusion
In the modern Wilsonian interpretation, a QFT is defined via an intrinsic cutoff scale 0 beyond
which it does not capture the microscopic degrees of freedom any more. For particle physics
one would assume that at the very least
which is the scale where quantum gravity effects become non-negligibly strong such that it is
no longer justified to ignore them.
Let us define our theory at this intrinsic cutoff by a Lagrangian L0 . We can think of the
couplings Ci (0 ) as being computed by whatever more fundamental theory takes over at scales
above 0 . In the spirit of an effective description, we view these Ci (0 ) as parametrising the
physics beyond 0 . At scales 0 , all possible non-renormalisable couplings are important,
because the underlying microscopic degrees of freedom become important at 0 and it takes a
lot of parameters to describe these in QFT. I.e. all couplings consistent with the symmetries of
the theory will appear in L0 .
with couplings
For processes at our scales 0 we compute with the effective action Leff
Ci () arising by RG flow starting from Ci (0 ). Crucially, as long as the theory is weakly
coupled up to 0 , the non-renormalisable couplings at 0 are of order
whatever they are originally at 0 . Since the microscopic degrees of freedom become important
only at high scales, it takes only a few parameters - the renormalisable and super-renormalisable
ones - to effectively approximate the physics at 0 . This explains why we only observe
renormalisable couplings at scale .
Despite the physical meaning of 0 (the intrinsic cutoff) we can formally take 0 pro-
vided all couplings remain finite, i.e. no Landau poles arise. This was our original motivation in
doing renormalisation, but from a Wilsonian perspective we now understand that our theory re-
mains an effective theory to the extent that all couplings are really input parameters and cannot
be computed from first principles without knowing the underlying theory.
252 CHAPTER 8. RENORMALISATION OF QUANTUM FIELD THEORY
For scalar theories, in particular 4 in d = 4, there remains a problem because of the RG flow of the
scalar mass:
m2 () = m2 (0 ) + (number)(0 )(2 20 ) + ..., (8.150)
which is the special case of (8.145) for d di = 2. At scales 0 , the dominant term is given
by 20 , and thus m2 () is of order (0 )20 . This is much bigger than we naturally expect in an
effective theory at scale . To arrive at a value for m2 () of order 2 instead, it would appear one
must carefully fine-tune the tree-level m2 (0 ) such as to almost precisely cancel the loop corrections
of order 20 .
For example, in the Standard Model of Particle Physics, the Higgs particle is a scalar field with
mH 126 GeV. But if we take 0 1019 GeV, then
m2H
1032 . (8.151)
20
The necessary fine-tuning to achieve this is considered technically unnatural and gives rise to the
so-called hierarchy problem: What stabilises the mass of a scalar at hierachically smaller mass than
Mpl ?
Basically there are 3 alternative types of solutions to the fine-tuning problem that are being discussed
in the modern literature:
1. If the inherent cutoff for the Standard Model were not MPl. , but at, say, 1 TeV, the fine-tuning
would be dramatically reduced. This would imply that new physics takes over at 0 1 TeV.
Candidates for such new dynamics are supersymmetry (SUSY) or large extra dimensions. This
reasoning motivates the search for such new physics at LHC.
2. Alternatively we could consider accepting the fine-tuning, which is not logically impossible -
for instance, but not necessarily, motivated by the so-called anthropic principle (which would
state that only values of mH 1 TeV could give rise to structure formation as we observe it and
thus, by definition, only a universe in which mH is so fine-tuned can be observed). Clearly this
is hard to make precise, but it is a logical possibility in principle.
3. Yet another approach would be to completely revise our notion of naturalness, even though it is
unclear at present what it should be replaced by.
It is important to appreciate that such a hierarchy problem does not arise for fermions and vectors
because their masses are stabilised by extra symmetry, namely gauge symmetry for vectors and chiral
symmetry for fermions. Small masses for these fields are technically natural in the sense that once
we choose them small at tree-level, they remain small including loop-corrections. This is guaranteed
whenever setting the mass to zero implies an enhancement of symmetry - such as in the present case
gauge symmetry for massless vector bosons and chiral symmetry for massless fermions.
Chapter 9
Quantisation of Yang-Mills-Theory
In terms of a basis T a of the Lie algebra, i.e. in terms of a complete set of generators of the Lie group
H, can be written as X
= a T a a T a . (9.2)
a
The Ta satisfy the defining relations
[T a , T b ] = i f abc T c = [T b , T a ] (9.3)
A Lie algebra possesses a symmetric bilinear from, the so-called Killing form,
ab = T a T b = T b T a R (9.5)
which is invariant under the adjoint action of the Lie group H on its Lie algebra, i.e.
hT a h1 hT b h1 = T a T b h. (9.6)
253
254 CHAPTER 9. QUANTISATION OF YANG-MILLS-THEORY
[T a , T c ] T b = T a [T c , T b ]. (9.8)
If H is compact as a manifold, then the invariant from can be shown to be positive definite, and one
can normalise it as
1
T a T b = ab . (9.9)
2
The Killing form ab = T a T b and its inverse can be used to raise and lower Lie-algebra indices.
Having said this, we will from now on put all Lie algebra indices upstairs with summation over
repeated indices understood. Then the structure constants f abc are totally antisymmetric (as will be
shown in the exercises) and invariant under cyclic permutations.
In the sequel we will typically think of T a as given in the defining matrix representation, i.e. as an
N N matrix acting on CN for some N. E.g. for H =SU(k), N = k and the T a are represented as
hermitian traceless k k matrices. In such a matrix representation we can identify
T a T b trT a T b . (9.10)
A ( x) Aa ( x)T a (9.11)
U ( x) = eig( x) (9.12)
as
i
A ( x) 7 UA U 1 + ( U )U 1 . (9.13)
g
1 a a
F ( x) = [ D , D ] F T , (9.17)
ig
9.2. GAUGE FIXING THE PATH INTEGRAL 255
where we think of ig1 [ D , D ] as acting on a Lie( H )-valued field. Plugging in the above definition of
the adjoint covariant derivative yields
F ( x) = A A + ig[ A , A ] (9.18)
or in components
a
F ( x) = Aa Aa g f abc Ab Ac . (9.19)
The field strength transforms in the adjoint representation of H as
F 7 UF U 1 (9.20)
D F + D F + D F = 0. (9.21)
D F = 0 F + ig[ A , F ]. (9.23)
The commutator introduces cubic and quartic interactions which will be discussed in detail later on.
In canonical quantisation, we would start by defining the conjugate momenta to A , but while
L
i = F 0i (9.24)
Ai
one finds
L
0 =
= 0. (9.25)
A0
This is a result of the fact that A0 appears without a time derivative in L - it is a non-dynamical
field. Variation of the action with respect to A0 gives
Di F 0i = 0, (9.26)
quantization with constraints for the simple case of U (1) theory is in fact the Gupta-Bleuler
quantisation method discussed in QFT1, but for non-abelian gauge theories things are more
complicated due to the more involved form of the constraints. We will not start with canoni-
cal quantization, but rather begin with path-integral quantization and return to the problem of
canonical quantization afterwards.
( K ) = 0 ( x). (9.28)
In fact, this is a consequence of gauge invariance. Recall that an infinitesimal gauge transfor-
mation acts as
A 7 A + + ig[ A , ]. (9.29)
Since the coupling g enters only the interactions, the transformation A 7 A + + O(g)
must leave the kinetic term invariant, which is nothing other than the statement (9.28). Note
that this problem is therefore independent of whether we consider abelian or non-abelian gauge
theory.
We will now begin quantization of Yang-Mills theory in the path-integral approach. Apart from pro-
viding a general method to compute correlation functions of gauge invariant operators, this will result
in a formalism in which we can also perform canonical quantization of the theory as required for the
definition of the (physical) Hilbert space.
The key idea is to cure the above problem of non-invertibility of the kinetic term by removing the
redundancy in the space of gauge field configurations due to gauge invariance. Let
and denote by
i
Ah ( x) B hA h1 + ( h)h1 (9.31)
g
a gauge transformation on Ah with h H. Then fields related by a gauge transformation are physically
equivalent, A ( x) Ah ( x). This is because A satisfies the equation of motion if and only if so does
Ah .
Given A ( x), the gauge equivalent field configurations are said to lie in the same orbit under gauge
transformations h H, denoted
{Ah ( x) h H}. (9.32)
9.2. GAUGE FIXING THE PATH INTEGRAL 257
The physically inequivalent degrees of freedom are then counted by the quotient space
n o
A/H B A Ah : A A, h H , (9.33)
i.e. they are given by all A -field configurations modulo gauge transformations. Our aim is now to
define the path integral not on A, but on the physical space A/H as
Z
d[ A]eiS [A] . (9.34)
A/H
with d[h] a measure on H. If we have found such a measure, we can factor out the integration over
the gauge orbit and only pick one representative per orbit as shown below.
for some function F which maps the Lie( H )-valued connection A to some other Lie( H )-valued field.
For example we could take F ( A) B A or generalisations thereof to be discussed later. Ideally, we
would like that given A A
such as to fix one representative per gauge orbit.2 Then on the lefthand side of (9.35) we can insert 1
given by Z h i
1= d[h] F ( Ah ) M ( Ah ) (9.38)
H
2 Indeed we will assume this here. In actuality, this condition is not satisfied because of a residual gauge symmetry.
Consequently, there are several gauge equivalent field configurations that satisfy F ( A) = 0. These are called Gribov
copies. One must restrict integration to a fundamental region where the gauge fixing is unique. This restriction indeed has
physical consequences, e.g. it modifies the form of the propagator in the deep infra-red.
258 CHAPTER 9. QUANTISATION OF YANG-MILLS-THEORY
H A
For all that follows we make the non-trivial assumption that we can find a DA that satisfies
DA = DAh . (9.42)
If no such measure exists, the gauge symmetry is said to be anomalous and no consistent quantization
procedure can be found. We will discuss such anomalies later in this course. Given (9.42) we have
Z Z Z h i
DA = d[h] DAh F ( Ah ) M ( Ah ). (9.43)
A H A
Relabelling Ah A yields
Z Z Z
DA = d[h] DA [ F ( A)] M ( A), (9.44)
A H A
| R
{z }
d[ A]
A/H
which defines our desired measure on A/H. This procedure can be repeated with any gauge invariant
integrand. In particular, since S [ A] = S [ Ah ] by classical gauge invariance, this results in
Z Z Z
DA eiS [A] = d[h] DA [ F ( A)] M ( A) eiS [A] . (9.45)
A H A
R
The factor H d[h] is just an overall factor Vol(H) that will cancel against the same factor in the
denominator in the expression of correlation functions and can thus be ignored. Thus we define the
partition function Z Z
Z= DA [ F ( A)] M ( A)eiS [A] d[ A]eiS [A] . (9.46)
A A/H
It remains to compute M ( A): By assumption, there exists one h = h0 such that F [ Ah0 ] = 0. Note that
the measure d[h] on H satisfies
Z Z
d[h] f (h) = d[h] f (gh) g H (9.47)
H H
9.3. FADDEEV-POPOV GHOSTS 259
This receives contributions only near h = id. Thus we can express the integral as one over infinitesi-
mal group elements by expanding h = ei with Lie( H ),
Z Z
d [ F ( A )] M ( A )
h i
1= d[h] F ( A ) M ( A ) =
h h
(9.49)
H Lie( H )
F ( A )
(D ). (9.52)
A
Finally we can replace A A as we did before because DA and S [ A] are gauge invariant. Thus we
end up with
Z
Z' DA [ F ( A)] detFP eiS [A] (9.53)
A
where F ( A) is a Lie( H )-valued field, i.e. F ( A) = F a ( A)T a and thus likewise B( x) = Ba ( x)T a
is a Lie( H )-valued auxiliary field called Nakanishi-Lautrup field with suitably normalised
measure DB.
Thus altogether
Z Z Z Z Z
DA [ F ( A)] detFP e iS [ A]
= DA DB D c Dc eiS [A,B,c,c] , (9.61)
where Z !
1 a 2
S [ A, B, c, c] = d x F + B ( x) F ( A) + c ( x)(FP c( x)) .
4 a a a a
(9.62)
4
The Grassmann-valued fields ca ( x) and ca ( x) are called Faddeev-Popov ghosts and anti-ghosts re-
spectively. They transform as scalar fields under SO(1, 3), but have fermionic statistics due to their
Grassmannian nature. Thus they violate the spin-statistics theorem. The underlying reason is that they
violate unitarity since their Fock space does not have a positive definite norm, as will be discussed in
detail later. This explains the name ghost, which generally describes a state with non-positive norm.
We shall think of the ghost fields as unphysical, negative degrees of freedom which, as we will also
see explicitly later, cancel unphysical polarisations of the Yang-Mills field.
F ( A( x)) = f ( x) (9.63)
9.3. FADDEEV-POPOV GHOSTS 261
for some Lie( H )-valued field f ( x). In fact, one can even average over all possible f ( x) with an
arbitrary measure - because all results are independent of how precisely we fix the gauge. A convenient
way to do this is to replace
Z R
2i d4 x f a ( x) f a ( x)
[ F ( A)] D f [ F ( A) f ] e (9.64)
where in the last step we are omitting an overall numerical factor from the Gaussian integral that will
cancel in all correlation functions. Furthermore we abbreviate F ( A( x))a F a ( x) (not to be confused
with the Yang-Mills field strength Fa ( x)). Thus there are two alternative final results for the gauge
with
Z !
1 a 2
S [ A, B, c, c] = d4 x ( F ) + Ba F a + Ba Ba + ca (FP c)a . (9.69)
4 2
Note that in (9.68) Ba ( x) has no kinetic term (hence its name auxiliary field). Its equation of motion
is
1
Ba ( x) = F a ( A( x)). (9.70)
R
Integrating out Ba ( x), i.e. replacing Ba by its equation of motion and omitting DB, gives (9.67)
from (9.68).
F ( A) = A (9.71)
262 CHAPTER 9. QUANTISATION OF YANG-MILLS-THEORY
and
F
FP = D = D . (9.72)
A
Then e.g. (9.67) becomes Z Z Z
Z= DA Dc Dc eiS [A,c,c] (9.73)
with Z " #
1 a 2 1
S [ A, c, c] = d x ( F ) ( A ) + c ( D c) .
4 a 2 a a
(9.74)
4 2
The value of the so-called gauge-fixing parameter is arbitrary, and we will see that all physical
amplitudes are independent of this gauge choice.
1 a 2
L( A, B, c, c) = ( F ) Ba Aa + Ba Ba + ca D ca . (9.76)
4 2
Indeed our approach will be to view (9.76) as the proper Lagrangian that should serve as our starting
point also for canonical quantization.
Starting from (9.76) gives the canonically conjugate fields
L L
a
A = = ( F a )0 , ac = a = c a , (9.77)
Aa c
L L
aB = = Aa0 , ac = a = ca g f abc Ab0 cc . (9.78)
Ba c
Thus we can view this as the missing commutation relation for Aa0 by effectively treating Ba as the
momentum canonically conjugate to Aa0 :
h i
Aa0 (t, ~x), Bb (t, ~y) = i ab (3) (~x ~y). (9.82)
1 2
L0 = Aa Aa Ba Aa + Ba Ba + c a ca , (9.85)
4 2
where the tilde stresses the nature as asymptotic fields. In the so-called Feynman gauge corresponding
to = 1 all fields enjoy a free mode expansion (see the tutorials for a proof). In particular we can
expand the asymptotic gauge field as
3
d3 k
Z
1 X a ~ h
a ~ ikx a ~ ikx
i
Aa ( x) = ( k, ) a ( k ) e + a ( k ) e , (9.86)
(2)3 2Ek =0
Furthermore one can integrate out Ba by replacing it by its equation of motion Ba = Aa (for
= 1) and combine (9.79) and (9.82) into the commutation relations
h i
Aa (t, ~x), Ab (t, ~y) = i ab (3) (~x ~y), (9.88)
264 CHAPTER 9. QUANTISATION OF YANG-MILLS-THEORY
i.e.
~ b ~0 3 (3) ~ ~ 0
h i
aa
(k ), a0 (k ) = (2) (k k ).
ab 0 (9.89)
As in QED the resulting asymptotic Fock spaces contains negative norm states for time-like and and
zero-norm states for longitudinal polarisations. Furthermore
d3 p
Z
1 a
ca ( x) = 3
p c ( ~p)eipx + ca ( ~p)eipx (9.90)
(2) 2E p
(with a similar expansion for c), and the anti-commutation relations for the asymptotic ghost modes
read
n o n o
ca ( ~p), cb ( ~p0 ) and ca ( ~p), cb ( ~p0 ) ab (3) ( ~p ~p0 ). (9.91)
This yields zero-norm excitations in the ghost sector (see again the tutorials for more details).
In view of the appearance of negative and zero-norm states in the asymptotic Fock space - not to
mention the problem that the ghost fields are non-physical degrees of freedom in the first place - the
question arises: What is the physical Hilbert space?
2. Show that time-evolution does not lead out of Hphys , i.e. show that the S -matrix acts as a unitary
operator on Hphys .
In principle we are free to choose any criterion for Hphys as long as it leads to a positive-definite norm
- after all this is part of the definition of the theory. However, in almost all cases the S-matrix will not
respect this criterion. On the other hand, if the criterion for Hphys is related to a symmetry of the full
interacting theory, then invariance of the physical Hilbert space under time-evolution follows because
the S -matrix respects that symmetry by assumption.
As a preparation for our definition of Hphys we must therefore first determine the symmetries of the
the full interacting, gauge-fixed Lagrangian
1
L = F a F a + Ba Ba + Ba Aa + ca ( D c)a . (9.92)
4 2
The first term is gauge invariant, but the gauge fixing term Ba Aa and the ghost Lagrangian cD c
break gauge invariance as a result of the partial derivatives. This is of course as it must be - after all
our desire to fix the gauge was what lead to the above Lagrangian.
However (9.92) possesses a residual global fermionic symmetry - the BRST symmetry3 - acting as
follows: First define a Grassmann odd operator S such that
i
S A = D c = ( c + ig[ A , c]), S c = g{c, c}, S c = B, S B = 0. (9.93)
2
3 Due to Becchi, Rouet, Stora and Tyutin.
9.5. BRST SYMMETRY AND THE PHYSICAL HILBERT SPACE 265
Note that
the assertion that S is Grassmann odd means that, by definition, S obeys a graded Leibniz rule,
e.g.
S (cA) = (S c) A c(S A), (9.94)
1. S is nilpotent,
S 2 = 0, (9.95)
1
L = F a F a S ( x) (9.96)
4
for ( x) = ca Aa + 2 ca Ba .
Note that properties (9.95) and (9.96) immediately imply that the action (9.92) is invariant under
application of S , i.e.
S L = 0, (9.97)
c( x) and by (9.95) S 2 = 0.
The proof of (9.95) is worked out on the exercise sheets and requires the Jacobi identity. Property
(9.96) follows simply by noting that
S ca Aa = (S c)a ( A )a ca (S A )a = Ba ( A )a + ca ( D c)a ,
(9.98)
S ca Ba = ((S c)a Ba ca S Ba ) = (Ba Ba ca 0) . (9.99)
2 2 2
Finally, the BRST symmetry transformation is defined as
= S (9.100)
with any of the fields A, c, c, B. Here is a global (i.e. not spacetime-dependent) Grassmann
variable. Clearly 2 = 0 because S 2 = 0 and
L = 0. (9.101)
266 CHAPTER 9. QUANTISATION OF YANG-MILLS-THEORY
By Noethers theorem one can construct the associated classical Noether current JBRST with
JBRST = 0 (9.102)
Q2BRST = 0. (9.105)
implies
h i
Q, X = iS X. (9.107)
If X is fermionic, then instead
h i
iS X = Q, X = QX X Q = ( QX + X Q), (9.108)
h i
so {Q, X} = iS X. Let X be bosonic. Then S X = i Q, X is fermionic and
n o h i
0 = S 2 X = Q, [ Q, X ] = Q( QX X Q) + ( QX X Q) Q = Q2 , X . (9.109)
So either Q2 1 or Q2 = 0. But Q2 1 is not possible because symmetry of the theory under BRST
transformations implies that also the vacuum must be invariant,
Q |i = 0, (9.111)
in contradiction with Q2 1.
Conservation of the classical Noether charge, QBRST = 0, translates into the relation
h i
H, QBRST = 0. (9.112)
9.5. BRST SYMMETRY AND THE PHYSICAL HILBERT SPACE 267
A = A , c = c , c = c , B = B , (9.114)
h| i = 0 Im( Q) (9.116)
because
h| i = h| Q Q |i = h| Q2 |i = 0. (9.117)
Furthermore
h| i = 0 |i Im( Q), |i Ker( Q). (9.118)
Thus Q-exact states are null and orthogonal to Q-closed states.
As a result, Q-closed states differing only by Q-exact elements have the same norm. To see this
let |1 i, |2 i Ker( Q) such that
|1 i |2 i = Q |i . (9.119)
Then
h1 | 1 i = h2 |2 i + h| Q Q |i + h2 | Qi + h| Q |2 i = h2 |2 i . (9.120)
After this digression, we come back to our physical problem. In order to fully define quantum Yang-
Mills-theory, we must define what we mean by the physical state space Hphy , and then prove that this
definition is stable under time-evolution and implies a positive-norm Hilbert space.
Our definition of Hphys is guided by the following two observations:
1. The space of states on which the time-evolution operator is independent of the specific choice
of gauge-fixing condition is given by Ker( QBRST ), i.e. by |i such that QBRST |i = 0.We will
prove this below.
2. Within this space, the BRST-exact states |i Im( QBRST ), have zero overlap with all other
states - as follows from previous discussion, in particular (9.118).
Now, invariance of time-evolution under changes of gauge fixing condition is obviously required for
the theory to make sense, so we require for sure
!
QBRST |phys i = 0 (9.124)
for physical states. Furthermore, states with zero overlap with all other states can never be measured
and are therefore unphysical. With this motivation we define
Ker( QBRST )
Hphys = Cohom( QBRST ) . (9.125)
Im( QBRST )
Before exploring Hphys further, we prove 1. from above: The time-evolution operator is
(L)dt
R R
U = T ei Hdt
= ei , (9.126)
where stands symbolically for all fields and their conjugates and L can be written as
1 a 2
L= F S , (9.127)
4
with S generating the BRST transformations. Now, only depends on the explicit choice of gauge
fixing condition F ( A) = 0. So a general transition amplitude h| U |i ' h| (1 + i ( L) |i
R
F h| U |i = h| (iF S ) |i = h| iS F |i . (9.128)
h i
Since by definition iS X = QBRST , X , this is
= h| ( QF F Q) |i . (9.129)
9.5. BRST SYMMETRY AND THE PHYSICAL HILBERT SPACE 269
It remains to evaluate the physical state condition explicitly and to check the existence of a positive-
definitene norm on Hphys , as well as to check unitarity of the S -matrix.
We can only do this explicitly for the asymptotic in/out-states, for which we now how they are created
by the action of the free asymptotic in/out-fields A, B, c, c . Since B A by its equation of motion
it suffices to consider 1-particle excitations of A , c, c . Let us define the states
3
|Aa (~k)i B a (~k, )aa ~
X
(k ) |i ,
=0 (9.131)
|ca (~k)i B c (k) |i
a ~
and similarly for |ca (~k)i, with a , c , c the creation operators for the asymptotic fields A, c, c , as
introduced in section 9.4. Now, for these free asymptotic fields (for which essentially g 0), the
BRST transformations take the simple form
S A = c,
S c = 0,
(9.132)
S c = B = A,
S B = 0,
in Feynman gauge = 1. Thus
for some numerical factors , , 0. First, we note that nil potency Q2BRST = 0 implies that
and thus
k2 = 0. (9.137)
First evaluate the condition QBRST |i = 0 if |i Hphys : Applied to (9.135) this implies that a
physical state |i (with non-vanishing k ) cannot have any c-excitations. Furthermore, in view
of (9.133), a state |i B |A (~k)i with polarisation vector is physical only if
k = 0. (9.138)
270 CHAPTER 9. QUANTISATION OF YANG-MILLS-THEORY
Next analyse our identification |i 0 if |i = QBRST |i: (9.133) implies |ca (~k)i is BRST
trivial, while (9.135) implies that |A (k)i is BRST trivial for = k .
To summarise:
A physical state |i Hphys = Cohom( QBRST ) is of the form
|i = |A (~k)i (9.139)
Unitarity of S -matrix:
Since L = L , we know that the S -matrix is unitary on the entire Fock space with basis |i HFock ,
i.e.
X
h| S |i h| S |i = h| 1 |i . (9.140)
|i
!
X
hT | S |T i hT | S |T i = hT | 1 |T i (9.141)
|T iHphys
h i
for |T i , |T i Hphys transversely polarised states. Since H, QBRST = 0 we know that
so S does not lead out of Ker(Q). But S can and in fact does produce BRST exact states in Im(QBRST )
- we will see this explicitly in examples. Still, these are null, i.e. have zero overlap with all other
states in Ker(Q). Therefore using (9.140)
X
hT | 1 |T i = hT | S |i h| S |T i
|i
X (9.143)
= hT | S |T i hT | S |T i .
|T iHphys
In the last line we first decomposed HFock into Ker( QBRST ) and its orthogonal complement and sec-
ond used that only the projection of |i to the transversely polarized physical space Cohom( QBRST )
Ker( QBRST ) has non-trivial overlap with S |T i because the latter lies inside Ker( QBRST ), and states
within Im( QBRS T ) are null and orthogonal on each other. Thus the S -matrix is unitary.
All of the above applies also to the special case of U (1) gauge theory. There, the c c system
4 In the sequel S denotes the S-matrix, not the BRST transformation operator. We hope this causes no confusion.
9.5. BRST SYMMETRY AND THE PHYSICAL HILBERT SPACE 271
does not couple to A since D c = c as the ghosts transform in the adjoint representation, which
for gauge group U (1) is trivial. Also, the constraint
QBRST |i = 0 (9.144)
In YM theory, imposing (9.140) on the full Hilbert space is not adequate since unlike the full operator
QBRST it does not commute with the interacting Hamiltonian H and is thus not stable under time-
evolution.