This action might not be possible to undo. Are you sure you want to continue?
Pierre van Baal
InstituutLorentz for Theoretical Physics,
University of Leiden, P.O.Box 9506,
2300 RA Leiden, The Netherlands.
Fall, 1998
Corrected version
January 2001
Copyright P. van Baal
No part of the notes and problems should be
reproduced or made electronically available
by others without permission of the author.
Table of contents
0. Introduction 1
1. Motivation 2
2. Quantisation of ﬁelds 4
3. EulerLagrange equations 8
4. Treelevel diagrams 13
5. Hamiltonian perturbation theory 17
6. Path integrals in quantum mechanics 20
7. Path integrals in ﬁeld theory 27
8. Perturbative expansion for ﬁeld theory 33
9. The scattering matrix 37
10. Cross sections 43
11. Decay rates 45
12. The Dirac equation 47
13. Plane wave solutions of the Dirac equation 52
14. The Dirac Hamiltonian 55
15. Path integrals for fermions 57
16. Feynman rules for vector ﬁelds 70
17. Quantum Electrodynamics  QED 74
18. NonAbelian gauge theories 82
19. The Higgs mechanism 88
20. Gauge ﬁxing and ghosts 90
21. The Standard Model 93
22. Loop corrections and renormalisation 99
Index to Problems 110
0
0 Introduction
Field theory is most successful in describing the process of scattering of particles in the con
text of the Standard Model, and in particular in the Electromagnetic and Weak Interactions.
The Large Electron Positron (LEP) collider operated 1989 till 2000. In a ring of 27 km in
diameter, electrons and positrons were accelerated in opposite directions to energies of ap
proximately 45 GeV. This energy is equivalent to half the mass (expressed as energy through
E = mc
2
) of the neutral Z
o
vector boson mass, which mediates part of the weak interactions.
The Z
o
particle can thus be created in electronpositron annihilation, at the regions where
the electron and positron beams intersect. As a Z
o
can be formed out of an electron and
its antiparticle, the positron, it can also decay into these particles. Likewise it can decay
in a muonantimuon pair and other combinations (like hadrons). The crosssection for the
formation of Z
o
particles shows a resonance peak around the energy where the Z
o
particle
can be formed. The width of this peak is a measure for the probability of the decay of this
particle. By the time you have worked yourself through this course, you should be able to
understand how to calculate this crosssection, which in a good approximation is given by
σ =
4πα
2
e
E
2
/27
(E
2
−M
2
Z
o)
2
+ M
2
Z
oΓ
2
Z
o
, (0.1)
expressed in units where ¯ h = c = 1. α
e
=
e
2
4π
∼ 1/137.037, is the ﬁnestructure constant,
E is twice the beam energy, M
Z
o the mass and Γ
Z
o the decay rate (or width) of the Z
o
vector boson. The latter gets a contribution from all particles in which the Z
o
can decay, in
particular from the decay in a neutrino and antineutrino of the three known types (electron,
muon and tau neutrinos). Any other unknown neutrino type (assuming their mass to be
smaller than half the Z
o
mass) would contribute likewise. Neutrinos are very hard to detect
directly, as they have no charge and only interact through the weak interactions (and gravity)
with other matter. With the data obtained from the LEP collider (the ﬁgure is from the
ALEPH collaboration) one has been able to establish that there are no unknown types of
light neutrinos, i.e. N
ν
= 3, which has important consequences (also for cosmology).
0
5
10
15
20
25
30
35
0.95
1
1.05
88 89 90 91 92 93 94 95
ﬁg. 1
The main aim of this ﬁeld theory course is to give the student a working knowledge
and understanding of the theory of particles and ﬁelds, with a description of the Standard
Model towards the end. We feel that an essential ingredient of any ﬁeld theory course has
1
to be to teach the student how Feynman rules are derived from ﬁrst principles. With the
path integral approach this is feasible. Nevertheless, it is equally essential that the student
learns how to use these rules. This is why the problems form an integral part of this course.
As Julius Wess put it during his course as a Lorentz professor at our Institute “you won’t
become a good pianist by listening to good concerts”.
These lecture notes reﬂect the ﬁeld theory courses I taught in the fall of 1992 at Utrecht,
and 1993, 1994, 1996, 1998 and 2000 at Leiden. I owe much to my teachers in this ﬁeld,
Martinus Veltman and Gerard ’t Hooft. As I taught in Utrecht from ’t Hooft’s lecture notes
“Inleiding in de gequantiseerde veldentheorie” (Utrecht, 1990) it is inevitable that there is
some overlap. In Leiden I spent roughly 25% longer in front of the classroom (3 lectures of
45 minutes each for 14 weeks), which allowed me to spend more time and detail on certain
aspects. The set of problems, 40 in total, were initially compiled by KarelJan Schoutens
with some additions by myself. In their present form, they were edited by Jeroen Snippe.
Of the many books on ﬁeld theory that exist by now, I recommend the student to consider
using “Quantum Field Theory” by C. Itzykson and J.B. Zuber (McGrawHill, New York,
1980) in addition to these lecture notes, because it oﬀers material substantially beyond the
content of these notes. I will follow to a large extent their conventions. I also recommend
“Diagrammatica: The path to Feynman diagrams”, by M. Veltman (Cambridge University
Press, 1994), for its unique style. The discussion on unitarity is very informative and it has
an appendix comparing diﬀerent conventions. For more emphasis on the phenomenological
aspects of ﬁeld theory, which are as important as the theoretical aspects (a point Veltman
often emphasised forcefully) I can recommend “Field Theory in Particle Physics” by B. de
Wit and J. Smith (NorthHolland, Amsterdam, 1986). For path integrals, which form a
crucial ingredient of these lectures, the book “Quantum mechanics and path integrals” by
R.P. Feynman and A.R. Hibbs (McGrawHill, New York, 1978) is a must. Finally, for an
introduction to the Standard Model, useful towards the end of this course, the book “Gauge
theories of weak interactions”, by J.C. Taylor (Cambridge Univ.Press, 1976), is very valuable.
1 Motivation
Field theory is the ultimate consequence of the attempts to reconcile the principles of rel
ativistic invariance with those of quantum mechanics. It is not too diﬃcult, with a lot of
hindsight, to understand why a ﬁeld needs to be introduced. Although this is not an attempt
to do justice to history  and perhaps one should spare the student the long struggle to arrive
at a consistent formulation, which most likely has not completely crystalised yet either  but
the traditional approach of introducing the concept is not very inspiring and most often
lacks physical motivation. In the following discussion I was inspired by “Relativistic Quan
tum Theory” from V.B. Berestetskii, E.M. Lifshitz and L.P. Pitaevskii (Pergamon Press,
Oxford, 1971). The argument goes back to L.D. Landau and R.E. Peierls (1930).
An important consequence of relativistic invariance is that no information should propa
gate at a speed greater than that of light. Information can only propagate inside the future
light cone. Consider the Schr¨odinger equation
i¯ h
∂Ψ(x, t)
∂t
= HΨ(x, t) . (1.0)
,
x
ct
N
Ψ
ﬁg. 2
Relativistic invariance should require that Ψ(x, t) = 0 for all (x, t) outside the light cone of
the support N
Ψ
= {xΨ(x, 0) = 0} of the wave function at t = 0.
2
Naturally, a ﬁrst requirement should be that the Schr¨odinger equation itself is relativis
tically invariant. For ordinary quantum mechanics, formulated in terms of a potential
H =
p
2
2m
+ V (x) , (1.1)
this is clearly not the case. Using the relation E
2
= p
2
c
2
+ m
2
c
4
the most obvious attempt
for a relativistically invariant wave equation would be the KleinGordon equation
−¯ h
2
∂
2
Ψ(x, t)
∂t
2
= −¯ h
2
c
2
∂
2
Ψ(x, t)
∂x
2
+ m
2
c
4
Ψ(x, t) . (1.2)
However, for this equation the usual deﬁnition of probability density is not conserved
∂
t
d
3
xΨ
∗
(x, t)Ψ(x, t) = 0 . (1.3)
As this is a consequence of the fact that the equation is second order in time, this can be
easily remedied, it seems, by taking the “square root” of the KleinGordon equation
i¯ h
∂Ψ(x, t)
∂t
=
−¯ h
2
c
2
∂
2
∂x
2
+ m
2
c
4
Ψ(x, t) (1.4)
We shall show that this, however, violates the principle of causality, i.e. the wave function
propagates outside of its light cone, which is unacceptable. Nevertheless, we will learn some
thing important from that computation, namely that negative energies seem unavoidable
when trying to localise wave functions within the light cone of N
Ψ
. But ﬁrst we will provide
a simple heuristic argument based on the uncertainty relation.
From the uncertainty principle ∆x∆p > ¯ h/2 and the bound on the speed involved in any
measurement of the position, it follows that precision of a measurement of the momentum
is limited by the available time ∆t∆p > ¯ h/c. Only for a free particle, where momentum is
conserved, such a measurement would be possible, but in that case, of course, the position is
completely undetermined, consistent with the plane wave description of such a free particle
(the light cone of N
Ψ
would in that case indeed give us no constraint). More instructive is
to look at how accurately we can determine the position of a particle. As the momentum is
bounded by the (positive) energy (p ≤ E/c) and as the maximal change in the momentum
is of the order of p itself, we ﬁnd that ∆x > ¯ h/p ≥ ¯ hc/E, which coincides with the limit set
by the De Broglie wavelength.
If we take this serious, that is position can in principle not be measured with arbitrary
accuracy, the notion of a wave function looses its meaning. On the other hand, if we would
like to localise the particle more accurate than within its De Broglie wavelength, it seems
to require an uncertainty in momentum that can only be achieved by allowing for negative
energy states. But negative energy states will be interpreted as antiparticles, and once
antiparticles are introduced, which can annihilate with particles, particle number is no longer
conserved and we likewise loose the notion of position of a particle. Only a free particle, as
a plane wave, seems to be compatible with relativistic invariance.
We will now verify by direct computation that localising the wave function within the light
cone will indeed require negative energy states. We consider ﬁrst the positive square root
of the KleinGordon equation and solve the Schr¨odinger equation for the initial condition
Ψ(x, 0) = δ
3
(x). From this we can solve any initial condition by convolution. As the
Schr¨odinger equation is ﬁrst order in time, the initial condition uniquely ﬁxes the wave
3
function for all later times and there will be a unique answer to the question whether the
wave function vanishes outside the light cone (i.e. for t > x). Problem 1 asks you to
investigate this in the simpler case of one, instead of three, spatial dimensions. For the
latter we simply give the result here, using the fact that in Fourier space the solution is
trivial. Computing Ψ(x, t) thus requires just some skills in performing Fourier integrals.
Ψ(x, t) =
d
3
p
(2π¯ h)
3
e
i p·x/¯h
e
−it
√
p
2
c
2
+m
2
c
4
/¯h
=
p
2
dp sin(θ)dθ
(2π)
2
¯ h
3
e
ipr cos(θ)/¯h
e
−it
√
p
2
c
2
+m
2
c
4
/¯h
=
1
2π
2
r¯ h
2
pdp sin(pr/¯ h)e
−it
√
p
2
c
2
+m
2
c
4
/¯h
=
−i
2π
2
r
∂
2
∂r∂t
∞
0
dp
cos(pr/¯ h)
√
p
2
c
2
+ m
2
c
4
e
−it
√
p
2
c
2
+m
2
c
4
/¯h
. (1.5)
We now introduce
p = mc sinh(u) , mcr/¯ h = z cosh(v) , mc
2
t/¯ h = z sinh(v) ,
z
2
= m
2
c
2
(r
2
−c
2
t
2
)/¯ h
2
, (1.6)
such that (the last identity simply being the deﬁnition of the modiﬁed Bessel function K
o
)
Ψ(x, t) =
−i
4π
2
rc
∂
2
∂r∂t
∞
−∞
du cos(z sinh(u) cosh(v))e
−iz sinh(v) cosh(u)
=
−i
8π
2
rc
∂
2
∂r∂t
∞
−∞
du
e
−iz sinh(u+v)
+ e
−iz sinh(u−v)
=
−i
2π
2
rc
∂
2
∂r∂t
∞
0
du cos(z sinh(u)) ≡
−i
2π
2
rc
∂
2
∂r∂t
K
o
(z) . (1.7)
Outside of the light cone, z is real (r
2
> c
2
t
2
) and Ψ(x, t) is purely imaginary. It decays
exponentially, but does not vanish! Inside the light cone we ﬁnd by analytic continuation (see
e.g. appendix C of “Relativistic Quantum Fields” by J.D. Bj¨ orken and S.D. Drell (McGraw
Hill, New York, 1965)) the following explicit expression
Ψ(x, t) =
1
4πrc
∂
2
∂r∂t
iY
o
(mc
√
c
2
t
2
−r
2
/¯ h) −sign(t)J
o
(mc
√
c
2
t
2
−r
2
/¯ h)
, r
2
< c
2
t
2
.
(1.8)
If we want to insist on locality, i.e. Ψ(x, t) = 0 for x > ct and want to stay as close as
possible to the solutions of the Schr¨odinger equation, we could take the real part of Ψ as the
wave function. It satisﬁes the KleinGordon equation, but not its positive square root. Ψ
∗
is a solution of the negative square root of the KleinGordon equation, and corresponds to
a negative energy solution. Apparently, localisation is only possible if we allow for negative
energy solutions.
2 Quantisation of ﬁelds
As position is no longer a quantum observable, but free particles do not seem to be in
contradiction with relativistic invariance, we can try to introduce such a free particle as a
quantum observable. This observable is hence described by a plane wave
ϕ
k
(x, t) = e
−i(kot−x·
k)/¯h
, (2.1)
4
which satisﬁes the KleinGordon equation
−¯ h
2
∂
2
ϕ(x, t)
∂t
2
= −¯ h
2
c
2
∂
2
ϕ(x, t)
∂x
2
+ m
2
c
4
ϕ(x, t) , (2.2)
where k
0
=
c
2
k
2
+ m
2
c
4
is the energy of the free particle. By superposition of these plane
waves we can make a superposition of free particles, which is therefore described by a ﬁeld
ϕ(x, t) = (2π¯ h)
−
3
2
d
3
k ˜ ϕ(
k, t)e
i
k·x/¯h
. (2.3)
It satisﬁes the KleinGordon equation if the Fourier components ˜ ϕ(
k, t) satisfy the harmonic
equation
−¯ h
2
∂
2
˜ ϕ(
k, t)
∂t
2
= (c
2
k
2
+ m
2
c
4
) ˜ ϕ(
k, t) ≡ k
2
o
(
k) ˜ ϕ(
k, t) . (2.4)
Its solutions split in positive and negative frequency components
˜ ϕ(
k, t) = ˜ ϕ
+
(
k)e
−ikot/¯h
+ ˜ ϕ
−
(
k)e
ikot/¯h
(2.5)
The wave function, or rather the wave functional Ψ(ϕ), describes the distribution over the
various free particle states. The basic dynamical variables are ˜ ϕ(
k). These play the role the
coordinates used to play in ordinary quantum mechanics and will require quantisation. As
they satisfy a simple harmonic equation in time, it is natural to quantise them as harmonic
oscillators. The Hamiltonian is then simply the sum of the harmonic oscillator Hamiltonian
for each
k, with frequency ω(
k) ≡ k
0
(
k)/¯ h.
i¯ h
∂Ψ(ϕ)
∂t
= H Ψ(ϕ) =
¸
k
H(
k) Ψ(ϕ) ,
H(
k) =
1
2
˜ π(
k)
2
+
1
2
ω(
k)
2
 ˜ ϕ(
k)
2
, ˜ π(
k) ≡
¯ h
i
∂
∂ ˜ ϕ(
k)
. (2.6)
In a ﬁnite volume with periodic boundary conditions, the integral over the momenta is
replaced by a sum as the momenta are in that case discrete,
k = 2πn¯ h/L, n ∈ ZZ
3
. Like for
the harmonic oscillator, we can introduce annihilation and creation operators
a(
k) =
1
2¯ hω(
k)
ω(
k) ˜ ϕ(
k) + i˜ π(
k)
,
a
†
(
k) =
1
2¯ hω(
k)
ω(
k) ˜ ϕ
∗
(
k) −i˜ π
∗
(
k)
, (2.7)
and express the ﬁeld operator (the equivalent of the coordinates) in terms of these creation
and annihilation operators. To give the ﬁeld operator its time dependence we have to invoke
the Heisenberg picture, which gives ϕ(x, t) = e
iHt/¯h
ϕ(x, 0)e
−iHt/¯h
. Using the well known
fact that e
iHt/¯h
a(
k)e
−iHt/¯h
= e
−iω(
k)t
a(
k) and e
iHt/¯h
a
†
(
k)e
−iHt/¯h
= e
iω(
k)t
a
†
(
k), which is a
consequence of [a(
k), H] = ¯ hω(
k)a(
k) and [a
†
(
k), H] = −¯ hω(
k)a
†
(
k), we ﬁnd
ϕ(x, t) = L
−
3
2
¸
k
¯ h
2k
o
(
k)
a
†
(
k)e
−i(
k·x−kot)/¯h
+ a(
k)e
i(
k·x−kot)/¯h
. (2.8)
5
In an inﬁnite volume we replace L
−
3
2
¸
k
by (2π¯ h)
−
3
2
d
3
k. Note that in the Heisenberg
picture positive energy modes behave in time as e
iEt/¯h
. Apparently we can identify (up to a
factor) ˜ ϕ
−
(
k) with a
†
(−
k) and ˜ ϕ
+
(
k) with a(
k), which is compatible with ˜ ϕ
∗
(
k) = ˜ ϕ(−
k),
required to describe a real ﬁeld (complex ﬁelds will be discussed in problem 5).
The Hilbert space is now given by the product of the Hilbert spaces of each
k separately
{n
k
} >=
¸
k
n
k
>=
¸
k
a
†
(
k)
n
k
n
k
!
0
k
> , (2.9)
with n
k
the occupation number, which in ﬁeld theory is now interpreted as the number of
free particles of momentum
k, a deﬁnition that makes sense as the energy of such a state
is n
k
k
o
(
k) above the state with zero occupation number (the “vacuum”). It is the property
of the harmonic oscillator, that its energy is linear in the occupation number, which makes
the ﬁeld theory interpretation in terms of particles possible. The annihilation operator in
this language therefore removes a particle (lowering the energy by the appropriate amount),
which consequently can be interpreted as the annihilation of the removed particle with an
antiparticle (described by the annihilation operator). For a real scalar ﬁeld, a particle is its
own antiparticle and this description is perhaps somewhat unfamiliar. But for the complex
ﬁeld of problem 5, the Fourier component with negative energy is independent of the one with
positive energy, hence describing a separate degree of freedom, namely that of an antiparticle
with opposite charge.
Interactions between the particles are simply introduced by modifying the KleinGordon
equation to have nonlinear terms, after which in general the diﬀerent Fourier components
no longer decouple. Field theory thus seems to be nothing but the quantum mechanics of an
inﬁnite number of degrees of freedom. It is, however, its physical interpretation that crucially
diﬀers from that of ordinary quantum mechanics. It is this interpretation that is known as
second quantisation. We were forced to introduce the notion of ﬁelds and the interpretation
involving antiparticles, when combining quantum mechanics with relativistic invariance. We
should therefore verify that indeed it does not give rise to propagation of information with
a speed larger than the speed of light. This is implied by the following identity, which for
the free scalar ﬁeld will be veriﬁed in problem 6
[ϕ(x, t), ϕ(x
′
, t
′
)] = 0, for (x −x
′
)
2
> (t −t
′
)
2
c
2
. (2.10)
It states that the action of an operator on the wave functional at a given spacetime point
is independent of the action of the operator at an other spacetime point, as long as these
two points are not causally connected. Due to the description of the time evolution with
a Hamiltonian, which requires the choice of a time coordinate it remains to be established,
however, that these equations are covariant under Lorentz transformations. We will resolve
this by using the path integral approach, in which the Lorentz invariance is intrinsic, but
which can also be shown to be equivalent to the Hamiltonian formulation.
Before preparing for path integrals by discussing the action principle, we would ﬁrst like
to address a simple physical consequence of the introduction and subsequent quantisation
of ﬁelds. It states that empty space (all occupation numbers equal to zero) has nevertheless
a nontrivial structure, in the same way that the ground state of a Hydrogen atom is non
trivial. Put diﬀerently, empty space is still full of zeropoint ﬂuctuations, which are, however,
only visible if we probe that empty space in one way or another. Also, formally, as each
6
zeropoint energy is nonzero, the energy of the vacuum in ﬁeld theory seems to be inﬁnite
E
0
=
¸
k
k
2
c
2
+ m
2
c
4
= · · ·? (2.11)
However, (as long as gravity is left out of our considerations) one is only sensitive to dif
ferences in energy. If we probe the vacuum, its energy can only be put to zero for one
particular value of the probe. The dependence of the vacuum energy on the probe can be
used to discover the nontrivial structure of the vacuum.
A famous and elegant method for probing the vacuum was introduced by Casimir (Proc.
Kon. Ned. Acad. Wet., ser. B51 (1948) 793), who considered using two conducting plates in
empty space. The energy of the vacuum is a function of the distance between the two plates,
which gives a force. Strictly speaking we should discuss this in the situation of the quantised
electromagnetic ﬁeld (see Itzykson and Zuber, par. 324), but the essential ingredient is
that Fourier components of the ﬁeld are aﬀected by the presence of the conducting plates.
We can also discuss this in the context of the simple scalar ﬁeld we have introduced before,
by assuming that the ﬁeld has to vanish at the plates. For simplicity we will also take the
mass of the scalar particles to vanish. If furthermore we use periodic boundary conditions
in the two other perpendicular directions over a distance L, then one easily veriﬁes that the
force per unit area on the conducting plates is given by
F
L
(x) = −dE
o
(x)/dx = −
1
2L
2
d
dx
¸
n∈ZZ
2
∞
¸
k=1
2π¯ hcn
L
2
+
π¯ hck
x
2
, (2.12)
where due to the vanishing boundary conditions the Fourier modes in the x
1
direction,
perpendicular to the conducting walls, are given by sin(πkx
1
/x) with k a positive integer,
whereas the quantisation of the momenta in the other two directions is as usual.
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
,
>
>
>
>
>
>
>
>.
`
·
L
x
L
ﬁg. 3
One can now formally take the inﬁnite volume limit
F(x) = lim
L→∞
F
L
(x) = −
c
8π
2
¯ h
2
d
dx
∞
¸
k=1
d
2
p
p
2
+
π¯ hk
x
2
. (2.13)
The integral and the sum are clearly divergent, but as Casimir observed, in practise no
conducting plate can shield a ﬁeld perfectly and especially for high frequency the boundary
conditions should be modiﬁed. One can mimic this by artiﬁcially cutting oﬀ the integral and
sum at high momenta. We would not expect the physical result to depend on the details of
how we do this, as otherwise we could use this experiment in an ingenious way to learn how
nature behaves at arbitrarily high energies. Indeed Casimir’s careful analysis showed that
7
the result is independent of the cutoﬀ function chosen. It is an important example of what
we will later recognise as renormalisability of ﬁeld theory. Since the result is insensitive to
the method of regularisation (only an overall constant contribution to E
o
(x) depends on it,
but that is not observable, as we argued before), we can choose a convenient way to perform
the calculation. Details of this will be provided in problem 2. The method of calculation
is known as dimensional regularisation, where one works in an arbitrary dimension (n = 2)
and then analytically extends the result to n = 2. We will ﬁnd that
F(x) = lim
n→2
−
c
8π
2
¯ h
2
d
dx
∞
¸
k=1
d
n
p
p
2
+
π¯ hk
x
2
= lim
n→2
(n + 1)ζ(−n −1)π
3n/2
¯ hc
8π
Γ(−(n + 1)/2)
Γ(−1/2)
x
−(n+2)
. (2.14)
in which ζ(i) ≡
¸
∞
k=1
k
−i
is the Riemann ζ function. It can be analytically extended to odd
negative arguments, where in terms of Bernoulli coeﬃcients ζ(1 − 2i) = −B
2i
/(2i). Also
Γ(−
1
2
) = −
3
2
Γ(−
3
2
) is ﬁnite, and we simply ﬁnd that
F(x) = −
π
2
¯ hc
480x
4
. (2.15)
Please note that we have disregarded the space outside the conducting plates. Imposing
also periodic boundary conditions in that direction, one easily ﬁnds that the region outside
the plates contributes with F(L − x) to the force, and vanishes when L → ∞. Therefore,
the eﬀect of the zeropoint ﬂuctuations in the vacuum leads to a (very small) attractive
force, which was ten years later experimentally measured by Sparnaay (Physica, 24 (1958)
751). Another famous example of the inﬂuence of zeropoint ﬂuctuations is the Lamb shift
in atomic spectra (hyperﬁne splittings), to be discussed at the end of the last section.
3 EulerLagrange equations
The KleinGordon equation in Lorentz covariant form (x ≡ (ct, x, y, z) ≡ (x
0
; x))
g
µν
∂
µ
∂
ν
ϕ(x) + m
2
ϕ(x) = 0 , g
µν
=
¸
¸
¸
¸
1 ⊖
−1
−1
⊖ −1
¸
, (3.1)
can be derived by variational calculus from an action principle
S =
d
4
x L(ϕ, ∂
µ
ϕ, x) , L(ϕ, ∂
µ
ϕ, x) =
1
2
(∂
µ
ϕ)
2
−V (ϕ) ,
(∂
µ
ϕ)
2
≡ ∂
µ
ϕ∂
µ
ϕ = g
µν
∂
µ
ϕ∂
ν
ϕ , V (ϕ) =
1
2
m
2
ϕ
2
. (3.2)
We assume the ﬁeld to be given at the boundary of the domain M of integration (typically
assuming the ﬁeld vanishes at inﬁnity) and demand the action to be stationary with respect
to any variation ϕ(x) →ϕ(x) + δϕ(x) of the ﬁeld,
δS(ϕ) ≡ S(ϕ + δϕ) −S(ϕ) =
M
d
4
x
∂
µ
ϕ∂
µ
δϕ −
∂V (ϕ)
∂ϕ
δϕ
=
M
d
4
x
−δϕ(∂
µ
∂
µ
ϕ +
∂V (ϕ)
∂ϕ
)
+
∂M
d
µ
σ(δϕ∂
µ
ϕ) = 0 , (3.3)
8
where d
µ
σ is the integration measure on the boundary ∂M. The variation δϕ is arbitratry,
except at ∂M, where we assume δϕ vanishes, and this implies the EulerLagrange equation
∂
µ
∂
µ
ϕ +
∂V (ϕ)
∂ϕ
= 0 , (3.4)
which coincides with the KleinGordon equation. We can also write the EulerLagrange
equations for arbitrary action S(ϕ) in terms of functional derivatives
δ
−
S
δ
−
ϕ(x)
=
δS
δϕ(x)
−∂
µ
δS
δ∂
µ
ϕ(x)
= 0 , (3.5)
where δ
−
stands for the total functional derivative, which is then split according to the explicit
dependence of the action on the ﬁeld and its derivatives (usually an action will not contain
higher than ﬁrst order spacetime derivatives). Please note that a functional derivative has
the property δϕ(x)/δϕ(y) = δ
4
(x−y), which is why in the above equation we take functional
derivatives of the action S and not, as one sees often, of the Lagrangian density L.
The big advantage of using an action principle is that S is a Lorentz scalar, which
makes it much easier to guarantee Lorentz covariance. As the action will be the starting
point of the path integral formulation of ﬁeld theory, Lorentz covariance is much more easy
to establish within this framework (there are instances where the regularisation, required
to make sense of the path integral destroys the Lorentz invariance, like in string theory.
Examples of these anomalies will be discussed later for the breaking of scale invariance and
gauge invariance). It is now simple to add interactions to the KleinGordon equation, by
generalising the dependence of the “potential” V (ϕ) to include higher order terms, like
V (ϕ) =
1
2
m
2
ϕ
2
+
g
4!
ϕ
4
, (3.6)
which is known as a scalar ϕ
4
ﬁeld theory. Later we will see that one can not add arbitrary
powers of the ﬁeld to this potential, except in two dimensions.
As in classical ﬁeld theory, we can derive from a Lagrangian with ϕ(x) and ˙ ϕ(x) ≡
∂ϕ(x)/∂t as its independent variables, the Hamiltonian through a Legendre transformation
to the canonical pair of variables π(x) (the “momentum”) and ϕ(x) (the “coordinate”)
π(x) =
δS
δ ˙ ϕ(x)
, H =
H(x)d
3
x =
(π(x) ˙ ϕ(x) −L(x)) d
3
x . (3.7)
The classical Hamilton equations of motion are given by
˙ ϕ(x) =
δH
δπ(x)
, ˙ π(x) = −
δH
δϕ(x)
+ ∂
i
δH
δ∂
i
ϕ(x)
. (3.8)
For the KleinGordon ﬁeld we simply ﬁnd
H =
1
2
π(x)
2
+
1
2
(∂
i
ϕ(x))
2
+
1
2
m
2
ϕ
2
(x) , (3.9)
and in problem 5 one will see that this Hamiltonian coincides with eq. (2.6), if we substitute
for ϕ(x) eq. (2.8). For an interacting scalar ﬁeld one ﬁnds
H =
1
2
π(x)
2
+
1
2
(∂
i
ϕ(x))
2
+ V (ϕ(x)) , (3.10)
which perhaps explains why V is called the potential.
9
It is well known that the Hamiltonian equations imply that H itself is conserved with
time, provided the Lagrangian (or Hamiltonian) has no explicit timedependence
dH
dt
=
d
3
x
˙ π(x)
δH
δπ(x)
+ ˙ ϕ(x)
δH
δϕ(x)
= 0 . (3.11)
Conservation of energy is one of the most important laws of nature and it is instructive to
derive it more directly from the fact that L does not depend explicitly on time. We deﬁne
the Lagrangian L as an integral of the Lagrange density L over space, L ≡
d
3
x L, such
that
dL
dt
=
d
3
x
∂
t
ϕ(x)
δS
δϕ(x)
+ ∂
t
(∂
µ
ϕ(x))
δS
δ∂
µ
ϕ(x)
=
d
3
x ∂
µ
∂
t
ϕ(x)
δS
δ∂
µ
ϕ(x)
.
(3.12)
The last term contains a total derivative, which vanishes if we assume that the ﬁeld is time
independent (or vanishes) at the boundary of the spatial integration domain. The above
equation becomes now
dL
dt
=
d
dt
d
3
x ˙ ϕ(x)
δS
δ ˙ ϕ(x)
=
d
dt
d
3
x ˙ ϕ(x)π(x) , (3.13)
which can also be expressed as
d
dt
d
3
x ( ˙ ϕ(x)π(x) −L) ≡
dH
dt
= 0 . (3.14)
In the same fashion one proves conservation of momentum in case the Lagrangian does
not explicitly depend on space (∂L/∂x
i
= 0)
0 =
dL
dx
i
=
d
3
x ∂
µ
∂
i
ϕ(x)
δS
δ∂
µ
ϕ(x)
=
d
dt
d
3
x π(x)∂
i
ϕ(x) . (3.15)
The conserved momentum is hence given by
P
i
=
d
3
x π(x)∂
i
ϕ(x) . (3.16)
Both conservation of momentum and energy are examples of conservation laws that are
consequence of symmetries (translation and time invariance). They can be derived as the
space integral of the time component of a conserved current or tensor
∂
µ
J
µ
(x) = 0 , ∂
µ
T
µν
(x) = 0 . (3.17)
In problem 3 these quantities will be deﬁned for a charged scalar ﬁeld, where J
µ
(x) can be
identiﬁed with the current, whose time component is the charge density. Indeed the total
charge is conserved. Assuming the current to vanish at spatial inﬁnity one easily ﬁnds
d
dt
d
3
x J
o
(x) =
d
3
x ∂
i
J
i
(x) = 0 (3.18)
The underlying principle is described by the Noether theorem, which implies that if the
Lagrangian L is invariant under ϕ → ϕ
Λ
, where Λ is a parameter (such as a shift of the
coordinates or a phase rotation of a complex ﬁeld), then the following current is conserved
J
µ
(x) =
δS
δ(∂
µ
ϕ(x))
∂ϕ
Λ
(x)
∂Λ
. (3.19)
10
The proof is simple and uses the EulerLagrange equations to substitute ∂
µ
{δS/δ(∂
µ
ϕ(x))}
for δS/δϕ(x)
0 =
dL(ϕ
Λ
)
dΛ
=
δS
δϕ(x)
∂ϕ
Λ
(x)
∂Λ
+
δS
δ(∂
µ
ϕ(x))
∂(∂
µ
ϕ
Λ
(x))
∂Λ
= ∂
µ
δS
δ(∂
µ
ϕ(x))
∂ϕ
Λ
(x)
∂Λ
+
δS
δ(∂
µ
ϕ(x))
∂(∂
µ
ϕ
Λ
(x))
∂Λ
= ∂
µ
J
µ
(x) (3.20)
We here considered the invariance under a global symmetry, but important in nature are
also the local symmetries, like the gauge invariance related to local changes of phase and the
general coordinate invariance in general relativity. Particularly with the latter in mind we
demand therefore that the action S (and not just L) is invariant under ϕ(x) →ϕ
Λ(x)
(x), with
Λ an arbitrary function of spacetime. This actually leads to the same conserved currents
in case L is also invariant. The same computation as above, still using the EulerLagrange
equations, shows that
0 =
δS
δΛ(x)
= ∂
µ
J
µ
(x) . (3.21)
As an important example we will discuss how this construction leads to conservation of the
energymomentum tensor, using general coordinate invariance, which is the local version
of translation invariance. For this we have to make the action invariant under such local
coordinate redeﬁnitions. As long as indices are contracted with the metric tensor g, L will
be invariant under general coordinate transformations, due to the transformation property
¯ x
µ
= x
µ
+ ε
µ
(x) , ¯ g
µν
(¯ x) =
∂¯ x
µ
∂x
α
∂¯ x
ν
∂x
β
g
αβ
(x) . (3.22)
For global translation invariance, ε
µ
is constant and eqs. (3.14) and (3.15) can be easily
generalised to show that the energymomentum tensor, T
µν
= ∂
µ
ϕ∂
ν
ϕ −g
µν
L, is conserved
(eq. (3.17)). For ε
µ
not constant, we note that the integration measure d
4
x is not a scalar
under general coordinate transformations, but the associated Jacobian can be easily absorbed
by
√
−det g, where the determinant is applied to the 4×4 matrix g
µν
. For a scalar ﬁeld this
leads to the following invariant action
S =
d
4
x
−det g (
1
2
g
µν
∂
µ
ϕ∂
ν
ϕ −V (ϕ)) (3.23)
For the original coordinates x of Minkowski spacetime the metric is given as in eq. (3.1), in
particular
−det g(x) = 1, and by expanding ¯ g to ﬁrst order in ε
µ
(x) we ﬁnd
S =
d
4
¯ x (1 −∂
λ
ε
λ
(x)) [
1
2
g
µν
∂
µ
ϕ(¯ x)∂
ν
ϕ(¯ x) −V (ϕ(¯ x))] +∂
α
ε
µ
(x)∂
µ
ϕ(¯ x)∂
α
ϕ(¯ x) . (3.24)
Now observe that g
µν
(x) is constant, such that the term independent of ε is a function of
¯ x, integrated over ¯ x, which is simply the action itself, as ¯ x now plays the role of a dummy
integration variable. The term linear in ε therefore has to vanish, but note that it only
involved the variation of the metric under the general coordinate transformation. Hence,
0 =
d
4
x ε
µ
(x)∂
α
(g
µα
L(x) −∂
µ
ϕ(x)∂
α
ϕ(x)) ≡ −
d
4
x ε
µ
(x)∂
α
T
αµ
(x) , (3.25)
which implies conservation of the energymomentum tensor (T
oo
= H). From the fact that
δg
µν
= −∂
µ
ε
ν
− ∂
ν
ε
µ
, δg
µν
= −g
µα
δg
αβ
g
βν
and δ
√
−det g =
1
2
g
µν
√
−det gδg
µν
, we derive
the identity
T
µν
(x) ≡ −2
δS
δg
µν
(x)
. (3.26)
11
In taking the derivative with respect to the metric, it is important that any Lorentz vector
(like the derivative ∂
µ
ϕ) or tensor appears in the Lagrangian with its indices down. Further
more, the result is to be evaluated for Minkowski space. Eq. (3.26) always gives a symmetric
energymomentum tensor and from the derivation it is clear that the result holds not only
for a simple scalar ﬁeld, but for any other bosonic ﬁeld theory (fermions form an exception,
see problem 23) like the one for the electromagnetic ﬁeld, which we discuss now.
The ﬁeld is given by the tensor F
µν
(x), with E
i
(x) = −F
0i
(x) its electric and B
i
(x) =
−
1
2
ε
ijk
F
jk
(x) its magnetic components. In terms of the vector potential A
µ
(x) one has
F
µν
(x) = ∂
µ
A
ν
(x) −∂
ν
A
µ
(x) . (3.27)
This already implies one of the Maxwell equations (through the socalled Jacobi or integra
bility conditions)
∂
µ
F
νλ
+ ∂
ν
F
λµ
+ ∂
λ
F
µν
= 0 . (3.28)
Written as ε
µνλσ
∂
ν
F
λσ
= 0 they are easily seen (resp. for µ = 0 and µ = i) to give
div
B = 0 , ∂
0
B + rot
E =
0 . (3.29)
The dynamical equations determining the ﬁelds in terms of the currents, or the sources,
J
µ
= (cρ;
J) are given by
∂
µ
F
µν
=
1
c
J
ν
or div
E = ρ , rot
B −∂
0
E =
J . (3.30)
We have chosen HeavisideLorentz units and in the future we will also often choose units
such that ¯ h = c = 1.
These Maxwell equations follow from the following action
S
em
(J) =
d
4
x (−
1
4
F
µν
(x)F
µν
(x) −A
µ
(x)J
µ
(x)) . (3.31)
We note, as is well known, that the equations of motion imply that the current is conserved.
With Noether’s theorem this makes us suspect that this is caused by a symmetry and indeed
it is known that under the gauge transformation
A
µ
(x) →A
µ
(x) + ∂
µ
Λ(x) (3.32)
the theory does not change. Our action is invariant under this symmetry if and only if the
current is conserved. This gauge symmetry will play a crucial role in the quantisation of the
electromagnetic ﬁeld.
An example of a conserved current can be deﬁned for a complex scalar ﬁeld. Its action
for a free particle is given by
S
0
=
d
4
x
∂
µ
ϕ
∗
(x)∂
µ
ϕ(x) −m
2
ϕ
∗
(x)ϕ(x)
. (3.33)
It is invariant under a phase rotation ϕ(x) →exp(ieΛ)ϕ(x) and from Noether’s theorem we
deduce that
J
µ
(x) ≡ ie(ϕ(x)∂
µ
ϕ
∗
(x) −ϕ
∗
(x)∂
µ
ϕ(x)) (3.34)
is conserved, see problem 3. We can extend this global phase symmetry to a local symmetry
if we couple the scalar ﬁeld minimally to the vector potential
S =
d
4
x
−
1
4
F
µν
(x)F
µν
(x) + (D
µ
ϕ)
∗
(x)D
µ
ϕ(x) −m
2
ϕ
∗
(x)ϕ(x)
,
D
µ
ϕ(x) ≡ ∂
µ
ϕ(x) −ieA
µ
(x)ϕ(x) . (3.35)
12
This guarantees the combined invariance under a local gauge transformation
ϕ(x) →exp(ieΛ(x))ϕ(x) , A
µ
(x) →A
µ
(x) + ∂
µ
Λ(x) . (3.36)
which makes the covariant derivative D
µ
ϕ(x) of the scalar ﬁeld transform as the scalar ﬁeld
itself, even for local phase rotations. Note that we can write this action also as
S = S
em
(J) + S
0
+
d
4
x e
2
A
µ
(x)A
µ
(x)ϕ(x)
2
, (3.37)
with J as given in eq. (3.34). We leave it as an exercise to show how the action of the electro
magnetic ﬁeld can be generalised to be invariant under general coordinate transformations
and to derive from this the energymomentum tensor. The result is given by
S
em
(J = 0) = −
1
4
d
4
x g
µλ
g
νσ
F
µν
F
λσ
−det g , T
µν
=
1
4
g
µν
F
λσ
F
λσ
−F
µλ
F
ν
λ
. (3.38)
4 Treelevel diagrams
In general, in the presence of interactions, the equations of motions can not be solved exactly
and one has to resort to a perturbative expansion in a small parameter. We discuss the scalar
case ﬁrst, as it is as always the simplest. We add to the Lagrangian density L a socalled
source term, which couples linearly to the ﬁeld ϕ (compare the driving force term for a
harmonic oscillator)
L =
1
2
(∂
µ
ϕ)
2
−V (ϕ) −J(x)ϕ(x) . (4.1)
For sake of explicitness, we will take the following expression for the potential
V (ϕ) =
1
2
m
2
ϕ
2
(x) +
g
3!
ϕ
3
(x) . (4.2)
The EulerLagrange equations are now given by
∂
µ
∂
µ
ϕ(x) + m
2
ϕ(x) +
1
2
gϕ
2
(x) + J(x) = 0 . (4.3)
If g = 0 it is easy to solve the equation (describing a free particle interacting with a given
source) in Fourier space. Introducing the Fourier coeﬃcients
˜
J(k) =
1
(2π)
2
d
4
x e
ikx
J(x) , ˜ ϕ(k) =
1
(2π)
2
d
4
x e
ikx
ϕ(x) (4.4)
it follows that
(−k
2
+ m
2
) ˜ ϕ(k) +
˜
J(k) = 0 or ϕ(x) =
d
4
y G(x −y)J(y) , (4.5)
where G is called the Green’s function, as it is the solution of the equation
(∂
µ
∂
µ
+ m
2
)G(x −y) = −δ
4
(x −y) (4.6)
Explicitly, it is given by the following Fourier integral
G(x −y) =
d
4
k
(2π)
4
e
−ik(x−y)
k
2
−m
2
+ iε
. (4.7)
13
Please note our shorthand notation of k
2
for k
µ
k
µ
and k(x −y) for k
µ
(x
µ
−y
µ
). A Green’s
function is not uniquely speciﬁed by its 2nd order equations, but also requires boundary con
ditions. These boundary conditions are, as we will see, speciﬁed by the term iε. Because of
the interpretation of the negative energy states as antiparticles, which travel “backwards” in
time, the quantum theory will require that the positive energy part vanishes for past inﬁnity,
whereas the negative energy part will be required to vanish for future inﬁnity. Classically
this would not make sense, and we would require the solution to vanish outside the future
light cone. The eﬀect of the iε prescription is, to shift the poles on the real axes to the
complex k
0
plane at k
0
= ±((
k
2
+ m
2
)
1
2
− iε). In section 5 we will see that this will imply
the appropriate behaviour required by the quantum theory.
Now we have found the solution for the free ﬁeld coupled to a source, we can do pertur
bation in the strength of the coupling constant g.
∂
µ
∂
µ
ϕ(x) + m
2
ϕ(x) + J(x) = −
1
2
gϕ
2
(x) (4.8)
can be solved iteratively by substituting a series expansion for ϕ(x),
ϕ(x) = ϕ
0
(x) + gϕ
1
(x) + g
2
ϕ
2
(x) +· · · . (4.9)
Obviously we have
ϕ
0
(x) =
d
4
y G(x −y)J(y) , (4.10)
whereas ϕ
1
(x) will be determined by the equation
∂
µ
∂
µ
ϕ
1
(x) + m
2
ϕ
1
(x) = −
1
2
ϕ
2
0
(x) . (4.11)
We can therefore interpret the righthand side as a source (up to a minus sign) and this
allows us to solve ϕ
1
(x) using the Green’s function
ϕ
1
(x) =
1
2
d
4
y G(x −y)ϕ
2
0
(y) =
1
2
G(x −y)d
4
y
G(y −z)G(y −w)J(z)J(w)d
4
zd
4
w .
(4.12)
This looks particularly simple in Fourier space
˜ ϕ
1
(k) =
1
2(2π)
2
1
k
2
−m
2
+ iε
d
4
p
˜
J(p)
˜
J(k −p)
(p
2
−m
2
+ iε)((k −p)
2
−m
2
+ iε)
. (4.13)
It is clear that this can be continued iteratively, e.g.
∂
µ
∂
µ
ϕ
n
(x) + m
2
ϕ
n
(x) = −
1
2
n−1
¸
i=0
ϕ
i
(x)ϕ
n−1−i
(x) . (4.14)
which is solved by
ϕ
n
(x) =
1
2
d
4
y G(x −y) (2ϕ
0
(y)ϕ
n−1
(y) + 2ϕ
1
(y)ϕ
n−2
(y) +· · ·) . (4.15)
Here we have written out the terms in the sum explicitly to indicate that all terms occur
twice and are the product of two diﬀerent terms, except for the term ϕ1
2
(n−1)
(y)
2
at n odd,
which occurs once. In Fourier space one ﬁnds
˜ ϕ
n
(k) =
1
2(2π)
2
1
k
2
−m
2
+ iε
d
4
p (2 ˜ ϕ
0
(p) ˜ ϕ
n−1
(k −p) +2 ˜ ϕ
1
(p) ˜ ϕ
n−2
(k −p) +· · ·) . (4.16)
14
By induction it is now easy to prove that
ϕ(x) =
¸
diagrams
×
,
+
1
2
>
>
×
×
,
+
1
2
>
>
>
>
×
×
×
,
+
1
8
>
>
>>
>>
×
×
×
×
,
+
1
2
>
>
>
>
>
>
×
×
× ×
,
+ · · ·
=
¸
diagrams
g
#vertices
N(diagram)
¸
{iv}
dx
iv
¸
<i,j>
G(x
i
−x
j
)
¸
{ks}
J(x
ks
) . (4.17)
Here the index i
v
runs over all vertices and sources (so it does not label the four spacetime
components of a single point, frequently it will be assumed that it is clear from the context
what is meant), whereas k
s
runs only over positions of the sources. The expression < i, j >
stands for the pairs of points in a diagram connected by a line (called propagator).
The Feynman rules to convert a diagram to the solution are apparently that each line
(propagator) between points x and y contributes G(x − y) and each cross (source) at a
point x contributes J(x). Furthermore, for each vertex at a point x we insert
d
4
x and
a power of the coupling constant g. Finally each diagram comes with an overall factor
1/N(diagram), being the inverse of the order of the permutation group (interchange of
lines and vertices) that leaves the diagram invariant (which is also the number of ways the
diagram can be constructed out of its building blocks). We have derived these rules for
the case that λ = 0, such that only three point vertices appear. All that is required to
generalise this to the arbitrary case with npoint vertices, is that each of these come with
their own coupling constant (i.e. λ for a fourpoint vertex). This is the reason why these
vertices are weighed by a factor 1/n! in the potential and hence by a factor 1/(n − 1)! in
the equations of motion (to be precise, if V (ϕ) = g
n
ϕ
n
/n!, the equation of motion gives
∂
2
µ
ϕ(x) +J(x) = −g
n
ϕ
(n−1)
(x)/(n −1)!, and the factor (n −1)! is part of the combinatorics
involved in interchanging each of the n −1 factors ϕ in the interaction term.
coordinate space momentum space table 1
>
>
x
≡ g
d
4
x
>
>
k
1
`
k
2
k
3
> >.
≡
g
(2π)
2
δ
4
(
¸
i
k
i
) vertex
x y
≡ G(x −y)
k
≡
d
4
k
1
k
2
−m
2
+iε
propagator
×
x
≡
d
4
xJ(x)
×
k
,
≡
˜
J(k) source
It is straightforward to translate these Feynman rules to momentum space, by inserting
the Fourier expansion of each of the terms that occur. Each propagator, which carries a
momentum k is replaced by a factor 1/(k
2
−m
2
+iε) and
d
4
k, each source with momentum
k ﬂowing in the source by a factor
˜
J(k), each vertex by a factor of the coupling constant
(i.e. g
n
for an npoint function), a factor 1/(2π)
2
(for an npoint function a factor (2π)
4−2n
)
and a momentum conserving delta function. To understand why momentum is conserved at
each vertex we use that in the coordinate formulation each vertex comes with an integration
over its position. As each line entering the vertex carries a Green’s function that depends
15
on that position (this being the only dependence), we see that a vertex at the point x gives
rise to
d
4
x
¸
α
G(x −x
α
) →
d
4
x
¸
α
d
4
k
α
e
−ikα(x−xα)
(2π)
4
(k
2
α
−m
2
+ iε)
= (2π)
4
¸
α
d
4
k
α
e
ikαxα
(2π)
4
(k
2
α
−m
2
+ iε)
δ
4
(
¸
α
k
α
) . (4.18)
Conventions in the literature can diﬀer on how the factors of i (which will appear in the
quantum theory) and 2π are distributed over the vertices and propagators. Needless to say
that the ﬁnal answers have to be independent of the chosen conventions.
As a last example in this section we will look again at the electromagnetic ﬁeld (whose
particles are called photons). In Fourier space the equations of motion are given by
(−k
2
δ
ν
µ
+ k
µ
k
ν
)
˜
A
µ
(k) =
˜
J
ν
(k) . (4.19)
Unfortunately the matrix −k
2
δ
ν
µ
+ k
µ
k
ν
has no inverse as k
µ
is an eigenvector with zero
eigenvalue. This is a direct consequence of the gauge invariance as the gauge transformation
of eq. (3.32) in Fourier language reads
˜
A
µ
(k) →
˜
A
µ
(k) + ik
µ
˜
Λ(k) . (4.20)
The component of A
µ
in the direction of k
µ
is for obvious reasons called the longitudinal
component, which can be ﬁxed to a particular value by a gauge transformation. Fixing the
longitudinal component of the electromagnetic ﬁeld (also called photon ﬁeld) is called gauge
ﬁxing, and the gauge choice is prescribed by the gauge condition. An important example is
the socalled Lorentz gauge
∂
µ
A
µ
(x) = 0 or k
µ
˜
A
µ
(k) = 0 . (4.21)
Because of the gauge invariance, the choice of gauge has no eﬀect on the equations of motion,
because the current is conserved, or k
µ
˜
J
µ
(k) = 0. The current (i.e. the source) does not
couple to the unphysical longitudinal component of the photon ﬁeld. It stresses again the
importance of gauge invariance and its associated conservation of currents.
To impose the gauge ﬁxing, we can add a term to the Lagrangian which enforces the gauge
condition. Without such a term the action is stationary under any longitudinal variation
δA
µ
(x) = ∂
µ
Λ(x) of the vector ﬁeld, and the added term should be such that stationarity in
that direction imposes the gauge condition. For any choice of the parameter α = 0 this is
achieved by the action
S =
d
4
x
−
1
4
F
µν
(x)F
µν
(x) −
1
2
α(∂
µ
A
µ
(x))
2
−A
µ
(x)J
µ
(x)
. (4.22)
Indeed, the variation δA
µ
(x) = ∂
µ
Λ(x) in the longitudinal direction leads to the equation
−α
d
4
x ∂
µ
∂
µ
Λ(x)∂
ν
A
ν
(x) = 0 , (4.23)
which implies the Lorentzgauge (assuming vanishing boundary conditions for the vector
potential at inﬁnity).
The equations of motion for this action now yield
∂
µ
F
µν
(x) + α∂
ν
∂
µ
A
µ
(x) = J
ν
(x) , (4.24)
16
or in Fourier space
(−k
2
δ
ν
µ
+ (1 −α)k
µ
k
ν
)
˜
A
µ
(k) =
˜
J
ν
(k) , (4.25)
which is invertible, as long as α = 0. The result is given by
˜
A
µ
(k) = −
δ
µ
ν
−
k
µ
kν(1−α
−1
)
k
2
+iε
k
2
+ iε
˜
J
ν
(k) . (4.26)
This is consequently the propagator of the electromagnetic ﬁeld (in the Lorentz gauge), also
simply called the photon propagator. Like in the scalar case it can be used to perform a
perturbative expansion for the classical equations of motion.
Note that the photon propagator simpliﬁes dramatically if we choose α = 1, but all ﬁnal
results should be independent of the choice of α and even of the choice of gauge ﬁxing all
together. This is the hard part in gauge theories. One needs to ﬁx the gauge to perform
perturbation theory and then one has to prove that the result does not depend on the choice
of gauge ﬁxing. In the quantum theory this is not entirely trivial, as the regularisation can
break the gauge invariance explicitly. Fortunately, there are regularisations that preserve the
gauge invariance, like dimensional regularisation which we already encountered in section 2
(in discussing the Casimir eﬀect). In the presence of fermions the situation can, however, be
considerably more tricky. Some diﬀerent choices of gauge ﬁxing will be explored in problems
8 and 9.
5 Hamiltonian perturbation theory
We consider the Hamiltonian for a free scalar particle coupled to a source. We will see
that the source can be used to create particles from the vacuum in the quantum theory,
and it forms an important ingredient, like for the derivation of the classical perturbation
theory of the previous section, in deriving scattering amplitudes and cross sections. Also the
Green’s function will reappear, but now with a unique speciﬁcation of the required boundary
conditions following from the time ordering in the quantum evolution equations.
For the Lagrangian
L =
1
2
∂
µ
ϕ∂
µ
ϕ −
1
2
m
2
ϕ
2
− ¯ εJϕ , (5.1)
the Hamiltonian is given by
H =
1
2
π
2
+
1
2
(∂
i
ϕ)
2
+
1
2
m
2
ϕ
2
+ ¯ εJϕ , (5.2)
where ¯ ε is a small expansion parameter. We will quantise the theory in a ﬁnite volume V =
[0, L]
3
with periodic boundary conditions, such that the momenta are discrete,
k = 2πn/L.
ϕ(x, t = 0) =
¸
k
1
2V k
o
(
k)
a(
k)e
i
k·x
+ a
†
(
k)e
−i
k·x
,
π(x, t = 0) = −i
¸
k
k
o
(
k)
2V
a(
k)e
i
k·x
−a
†
(
k)e
−i
k·x
. (5.3)
The Hamiltonian is now given by H(t) = H
0
+ ¯ εH
1
(t), and we work out the perturbation
theory in the Schr¨odinger representation. We have
H
0
=
¸
k
k
0
(
k)(a
†
(
k)a(
k) +
1
2
) ,
H
1
(t) =
V
d
3
x J(x, t)ϕ(x) =
¸
k
1
2k
0
(
k)
˜
J(
k, t)
a(−
k) + a
†
(
k)
, (5.4)
17
here
˜
J(
k, t) is the Fourier coeﬃcient of J(x, t), or
J(x, t) =
1
√
V
¸
k
˜
J(
k, t)e
i
k·x
. (5.5)
Let us start at t = 0 with the vacuum state 0 >, which has the property that a(
k)0 >= 0
for all momenta, then it follows that
d
dt
Ψ(t) >= −iH(t)Ψ(t) > , (5.6)
which can be evaluated by perturbing in ¯ ε.
Ψ(t) > ≡ e
−iH
0
t

ˆ
Ψ(t) > , 
ˆ
Ψ(t) >=
∞
¸
n=0
¯ ε
n

ˆ
Ψ
n
(t) > ,
d
dt

ˆ
Ψ
n
(t) > = −ie
iH
0
t
H
1
(t)e
−iH
0
t

ˆ
Ψ
n−1
(t) > . (5.7)
Actually, by transforming to 
ˆ
Ψ(t) > we are using the interaction picture, which is the
usual way of performing Hamiltonian perturbation theory known from ordinary quantum
mechanics. These equations can be solved iteratively as follows

ˆ
Ψ
1
(t) > = −i
t
0
dt
1
e
iH
0
t
1
H
1
(t
1
)e
−iH
0
t
1
0 > ,

ˆ
Ψ
2
(t) > = −i
t
0
dt
1
e
iH
0
t
1
H
1
(t
1
)e
−iH
0
t
1

ˆ
Ψ
1
(t
1
) >
= −
t
0
dt
1
e
iH
0
t
1
H
1
(t
1
)
t
1
0
dt
2
e
iH
0
(t
2
−t
1
)
H
1
(t
2
)e
−iH
0
t
2
0 > ,

ˆ
Ψ
n
(t) > = −i
t
0
dt
1
e
iH
0
t
1
H
1
(t
1
)e
−iH
0
t
1

ˆ
Ψ
n−1
(t
1
) >= · · · . (5.8)
Please note the time ordering, which is essential as H
1
(t) does not commute with H
1
(t
′
) for
diﬀerent t and t
′
. We can for example compute the probability that at time t Ψ(t) > is still
in the ground state (whose energy we denote by E
0
, which will often be assumed to vanish)
< 0Ψ(t) > = e
−iE
0
t
< 0
ˆ
Ψ(t) >= e
−iE
0
t
{1 −i¯ ε
t
0
dt
1
< 0H
1
(t
1
)0 >
− ¯ ε
2
t
0
dt
1
t
1
0
dt
2
< 0H
1
(t
1
)e
i(H
0
−E
0
)(t
2
−t
1
)
H
1
(t
2
)0 > +O(¯ ε
3
)} . (5.9)
It is simple to see that the term linear in ¯ ε will vanish, as the vacuum expectation values of
the creation and annihilation operators vanish, i.e. < 0a
†
0 >=< 0a0 >= 0. To evaluate
the remaining expectation value in the above equation we substitute H
1
in terms of the
creation and annihilation operators (see eq. (5.4))
< 0H
1
(t
1
)e
i(H
0
−E
0
)(t
2
−t
1
)
H
1
(t
2
)0 >=
¸
k, p
˜
J( p, t
1
)
˜
J(
k, t
2
)
4k
0
( p)k
0
(
k)
< 0
a(− p) + a
†
( p)
e
i(H
0
−E
0
)(t
2
−t
1
)
a(−
k) + a
†
(
k)
0 >
=
¸
k
˜
J(−
k, t
1
)
˜
J(
k, t
2
)
2k
0
(
k)
e
−ik
0
(
k)(t
1
−t
2
)
. (5.10)
18
Combining these results we ﬁnd
< 0Ψ(t) > e
iE
0
t
= 1 − ¯ ε
2
¸
k
t
0
dt
1
t
1
0
dt
2
˜
J(−
k, t
1
)
˜
J(
k, t
2
)
2k
0
(
k)
e
−ik
0
(
k)(t
1
−t
2
)
+O(¯ ε
3
)
= 1 −
1
2
¯ ε
2
¸
k
t
0
dt
1
t
0
dt
2
˜
J(−
k, t
1
)
˜
J(
k, t
2
)
2k
0
(
k)
e
−ik
0
(
k)t
1
−t
2

+O(¯ ε
3
). (5.11)
Especially the last identity is useful to relate this to the Green’s function we introduced
in the previous section. Using contour deformation in the complex ω plane we ﬁnd
∞
−∞
dω
e
iωt
ω
2
−k
2
0
(
k) + iε
= −
2πi
2k
0
(
k)
e
−ik
0
(
k)t
. (5.12)
This can be shown as follows. When t > 0, we can deform the contour of integration to the
upper halfplane (where e
iωt
decays exponentially) and only the pole at ω = ω
−
≡ −k
0
(
k)+iε
contributes, with a residue 2πie
−ik
0
(
k)t
/(−2k
0
(
k)) (see the ﬁgure). Instead, for t < 0 the
contour needs to be deformed to the lower halfplane and the pole at ω = ω
+
≡ k
0
(
k) − iε
contributes with the residue 2πie
ik
0
(
k)t
/(−2k
0
(
k)) (note that the contour now runs clockwise,
giving an extra minus sign).
,
,
ω
+
,
ω
−
t > 0
,
,
ω
+
,
ω
−
t < 0
ﬁg. 4
This means that we can rewrite eq. (5.11) as
lim
t→∞
< 0Ψ(t) > e
iE
0
t
= 1 −
i
2
¯ ε
2
¸
k
∞
−∞
dω
˜
J(−ω, −
k)
˜
J(ω,
k)
ω
2
−
k
2
−m
2
+ iε
+O(¯ ε
3
) , (5.13)
where
˜
J(ω,
k) =
1
√
2π
dt
˜
J(
k, t)e
iωt
=
1
√
2πV
IR×V
d
3
xdt J(x)e
ikx
. (5.14)
The last expression should be replaced by (2π)
−2
IR
4 d
4
xJ(x)e
ikx
in case the volume is inﬁ
nite. It is important to note that we have chosen J(x) = 0 for t < 0. Equivalently we can
start at t = −∞ and integrate the quantum equation of motion up to t = ∞. We have to
require that J(x) vanishes suﬃciently rapidly at inﬁnity.
In an inﬁnite volume we therefore ﬁnd for what is known as the vacuum to vacuum
amplitude of the scattering matrix
lim
t→∞
< 0Ψ(t) > e
iE
0
t
= 1 −
i
2
¯ ε
2
d
4
k
˜
J(−k)
˜
J(k)
k
2
−m
2
+ iε
+O(¯ ε
3
)
= 1 −
i
2
¯ ε
2
d
4
xd
4
y J(x)G(x −y)J(y) +O(¯ ε
3
) . (5.15)
where G(x − y) is exactly the Green’s function we introduced in the previous section. The
socalled iε prescription, which is equivalent with specifying the boundary conditions, has
19
therefore been derived from the timeordering in the Hamiltonian evolution of the system
and is thus prescribed by the requirement of causality. Note that we can use the diagrams
introduced in the previous section to express this result (taking E
0
= 0 from now on) as
< 0Ψ(t) >= 1 −
i
2
× ×
¯ εJ ¯ εJ
+O(¯ ε
3
) , (5.16)
where the factor of a half is a consequence of the symmetry under interchanging the two
sources.
For a complex scalar ﬁeld, ϕ and ϕ
∗
are independent and we need to introduce two
sources by adding to the Lagrangian −ϕJ
∗
−ϕ
∗
J (see problem 17). It is not too diﬃcult to
show that in this case
< 0Ψ(t) >= 1 −i
× × >
¯ εJ ¯ εJ
∗
+O(¯ ε
3
) , (5.17)
without a factor of one half because the sources J and J
∗
are independent and cannot be
interchanged. This is why in this case the propagator has a direction.
6 Path integrals in quantum mechanics
For simplicity we will start with a onedimensional Hamiltonian
H =
ˆ p
2
2m
+ V (ˆ x) , ˆ p =
¯ h
i
∂
∂x
, (6.1)
where we have indicated a hat on top of operators, to distinguish them from numbervalued
coordinate and momentum. We wish to study the timeevolution operator exp(−iHT/¯ h).
In the coordinate representation its matrix is given by
< x
′
e
−iHT/¯h
x > , (6.2)
where x > is the position eigenfunction. We will also need the momentum eigenfunction
p >, i.e. ˆ pp >= pp >, whose wave function in the coordinate space is given by
< xp >=
e
ipx/¯h
√
2π¯ h
. (6.3)
Indeed, one veriﬁes that
¯ h
i
∂
∂x
< xp >= p < xp > . (6.4)
An important ingredient in deriving the path integral expression will be the completeness
relations
ˆ
1 =
dx x >< x and
ˆ
1 =
dp p >< p . (6.5)
For arbitrary N we can use this to write
< x
′
e
−iHT/¯h
x > = < x
′

e
−iHT/N¯h
N
x >=
· · ·
< x
′
e
−iHT/N¯h
x
N−1
> dx
N−1
< x
N−1
e
−iHT/N¯h
x
N−2
> dx
N−2
< x
N−2
 · · · · · · e
−iHT/N¯h
x
2
> dx
2
< x
2
e
−iHT/N¯h
x
1
> dx
1
< x
1
e
−iHT/N¯h
x > . (6.6)
20
We will now use the socalled Trotter formula
e
−i(A+B)/N
= e
−iA/N
e
−iB/N
(1 +O(N
−2
)) (6.7)
for two operators A and B. This can be seen by expanding the exponents, and the error
term is actually of the form [A, B]/N
2
. (One can also use the CampbellBakerHausdorﬀ
formula, which will be introduced later). With the Hamiltonian of eq. (6.1) this can be used
to write for N →∞
e
−iHT/N¯h
= e
−iˆ p
2
T/2mN¯h
e
−iV (ˆ x)T/N¯h
. (6.8)
By inserting the completeness relation for the momentum we can eliminate the operators
< x
i+1
e
−iHT/N¯h
x
i
> =
dp
i
< x
i+1
p
i
>< p
i
e
−iHT/N¯h
x
i
>
≈
dp
i
< x
i+1
p
i
>< p
i
e
−iˆ p
2
T/2mN¯h
e
−iV (ˆ x)T/N¯h
x
i
>
=
dp
i
< x
i+1
p
i
>< p
i
e
−ip
2
i
T/2mN¯h
e
−iV (x
i
)T/N¯h
x
i
>
=
dp
i
e
ip
i
(x
i+1
−x
i
)/¯h
2π¯ h
e
−i(
p
2
i
2m
+V (x
i
))T/N¯h
. (6.9)
This can be done for each matrix element occurring in eq. (6.6). Writing ∆t = T/N, x
N
= x
′
and x
0
= x we ﬁnd
< x
′
e
−iHT/¯h
x > = lim
N→∞
· · ·
N−1
¸
i=1
dx
i
N−1
¸
j=0
dp
j
N−1
¸
i=0
< x
i+1
p
i
>< p
i
e
−iH∆t/¯h
x
i
>
= lim
N→∞
dp
0
2π¯ h
N−1
¸
i=1
dx
i
dp
i
2π¯ h
exp
¸
i∆t
¯ h
N−1
¸
i=0
p
i
(x
i+1
−x
i
)
∆t
−
p
2
i
2m
−V (x
i
)
¸
. (6.10)
It is important to observe that there is one more p integration than the number of x inte
grations.
The integrals in the path integral are strongly oscillating and can only be deﬁned by
analytic continuation. As parameter for this analytic continuation one chooses the time
t. For ∆t = T/N ≡ −iT /N = −i∆τ the Gaussian integral over the momenta is easily
evaluated
∞
−∞
dp
i
exp
¸
ip
i
(x
i+1
−x
i
)
¯ h
−
p
2
i
∆τ
2m¯ h
¸
=
2πm¯ h/∆τ exp
¸
−
1
2
m
(x
i+1
−x
i
)
2
∆τ¯ h
¸
. (6.11)
which leads to
< x
′
e
−iHT/¯h
x >= lim
N→∞
mN
2πT ¯ h
1
2
N
N−1
¸
i=1
dx
i
exp
¸
−
∆τ
¯ h
N−1
¸
i=0
m(x
i+1
−x
i
)
2
2∆τ
2
+ V (x
i
)
¸
(6.12)
or after substituting T = iT we ﬁnd
< x
′
e
−iHT/¯h
x >= lim
N→∞
mN
2πiT¯ h
1
2
N
N−1
¸
i=1
dx
i
exp
¸
i
∆t
¯ h
N−1
¸
i=0
m(x
i+1
−x
i
)
2
2∆t
2
−V (x
i
)
¸
(6.13)
This is the deﬁnition of the path integral, but formally it will often be written as
< x
′
e
−iHT/¯h
x >=
x(T)=x
′
x(0)=x
Dx(t) exp[iS/¯ h] , S =
T
0
dt
1
2
m˙ x
2
(t) −V (x(t))
¸
, (6.14)
21
since the discretised version of the action with x
j
≡ x(t = j∆t) is precisely
S
discrete
= ∆t
N−1
¸
i=0
1
2
m
(x
i+1
−x
i
)
2
∆t
2
−V (x
i
)
. (6.15)
It is important to note that the continuous expression is just a notation for the discrete
version of the path integral, but formal manipulations will be much easier to perform in
this continuous formulation. Furthermore, the integral is only deﬁned through the analytic
continuation in time.
However, if we integrate over x
N
= x
0
this analytically continued path integral, with
T = −iT , has an important physical interpretation
dx < xe
−HT /¯h
x >= Tr(e
−βH
)
β=T /¯ h
. (6.16)
It is the quantum thermal partition function (the Boltzmann distribution) with a tempera
ture of ¯ h/kT . In the continuous formulation we therefore have
Tr(e
−T H/¯h
) =
x(T )=x(0)
Dx(τ) exp[−S
E
/¯ h] (6.17)
in which S
E
is the socalled Euclidean action
S
E
=
T
0
dτ
1
2
m(dx(τ)/dτ)
2
+ V (x(τ))
¸
. (6.18)
It is only in this Euclidean case that one can deﬁne the path integral in a mathematically
rigorous fashion on the class of piecewise continuous functions in terms of the socalled
Wiener measure
dW
x,x
′ (T ) ≡
x(T )=x
′
x(0)=x
Dx(τ) exp(−
T
0
1
2
m˙ x(τ)
2
dτ/¯ h) , (6.19)
meaning that this measure is independent of the way the path is discretised.
,
`
`
`
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'`
`
`
`
`
'
'
'
'
'
'
'
'
'
`
`
`
`
`
/
/
/
/
/
t
∆t
x
ﬁg. 5
For more details on this see “Quantum Physics: A functional integral point of view”, by J.
Glimm and A. Jaﬀe (2nd ed., Springer, New York, 1987).
We will now do an exact computation to give us some conﬁdence in the formalism. To be
speciﬁc, what we will compute is the quantum partition function for the harmonic oscillator,
where V (x) =
1
2
mω
2
x
2
.
Z
N
≡
mN
2πT ¯ h
1
2
N
∞
−∞
· · ·
N−1
¸
i=0
dx
i
exp
¸
−
∆τ
¯ h
N−1
¸
i=0
1
2
m
x
i+1
−x
i
∆τ
2
+
1
2
mω
2
x
2
i
¸
. (6.20)
22
Note that we have now N integrations, because we also integrate over x
0
= x(0) = x(T ) = x
N
to implement the trace. The path involved is thus periodic in time, a general feature of the
expression for the quantum partition function in terms of a path integral. We now rescale
y
i
= x
i
m
∆τ¯ h
1
2
, ˜ ω = ω∆τ , (6.21)
to obtain the simple result
Z
N
=
∞
−∞
· · ·
N−1
¸
i=0
dy
i
√
2π
exp
¸
−
N−1
¸
i=0
1
2
(y
i+1
−y
i
)
2
+
1
2
˜ ω
2
y
2
i
¸
. (6.22)
We can diagonalise the quadratic term by using Fourier transformation
y
k
=
1
√
N
N−1
¸
ℓ=0
b
ℓ
e
2πikℓ/N
, b
∗
ℓ
= b
N−ℓ
, b
∗
0
= b
0
. (6.23)
It is easy to verify that the Jacobian for the change of variables y
i
→b
ℓ
is 1, and one obtains
a result that must look familiar from the classical small oscillations problem for a ﬁnite
number of weights connected by strings,
Z
N
=
∞
−∞
· · ·
N−1
¸
ℓ=0
db
ℓ
√
2π
exp
¸
−
1
2
N−1
¸
ℓ=0
4 sin
2
(πℓ/N) + ˜ ω
2
b
ℓ

2
¸
. (6.24)
Note that if b
ℓ
is complex, we mean by db
ℓ
= dReb
ℓ
dImb
ℓ
. The integral can now easily be
evaluated
Z
N
=
N−1
¸
ℓ=0
4 sin
2
(πℓ/N) + ˜ ω
2
−
1
2
. (6.25)
We can convert the product to a sum using a Laplace transform. We start with the identity
log(λ/µ) = −lim
aց0
∞
0
ds s
a−1
(e
−sλ
−e
−sµ
) , (6.26)
such that
log[Z
N
(˜ ω)/Z
N
(˜ ω
0
)] =
1
2
lim
aց0
∞
0
ds s
−1+a
(Q(s, ˜ ω) −Q(s, ˜ ω
0
)) . (6.27)
We read oﬀ, from the deﬁnition of Z
N
, that Q is a sum of exponentials
Q(s, ˜ ω) ≡
N−1
¸
ℓ=0
exp
−s
4 sin
2
(πℓ/N) + ˜ ω
2
≡ Ne
−s(˜ ω
2
+2)
f
s
(0) , (6.28)
where
f
s
(x) =
1
N
N−1
¸
ℓ=0
e
2s cos(2π(ℓ+x)/N)
. (6.29)
This is a periodic function with period 1 (f
s
(x + 1) = f
s
(x)) and its discrete Fourier coeﬃ
cients can be computed exactly
˜
f
s
(k) =
1
0
dx e
2πikx
f
s
(x) =
1
N
N
0
dx e
2πikx
e
2s cos(2πx/N)
=
1
2π
2π
0
dθ e
iNkθ
e
2s cos(θ)
≡ I
Nk
(2s) . (6.30)
23
Please note that we have exchanged the sum over ℓ with extending the integration of x to
the interval [0, N]. The last identity is one of the deﬁnitions of the modiﬁed Besselfunction,
see e.g. “Handbook of mathematical functions”, by M. Abramowitz and I. Stegun (Dover,
New York, 1978). The advantage of these manipulations is that the Laplace transform of
this Besselfunction is know (see the same reference)
∞
0
ds e
−λs
I
ν
(s) =
[λ +
√
λ
2
−1]
−ν
√
λ
2
−1
, (6.31)
and as we can express f
s
(x) as a sum over these Besselfunctions
f
s
(x) =
¸
k∈ZZ
e
−2πikx
I
Nk
(2s) , (6.32)
this allows us to evaluate eq. (6.27). For technical reasons it is easier to compute the variation
of the free energy with the frequency, where the free energy F is deﬁned as
Z
N
(˜ ω) = exp[−βF(˜ ω)] . (6.33)
We therefore ﬁnd (using β = T /¯ h = N∆τ/¯ h)
∂
∂˜ ω
F(˜ ω) = ¯ hω
¸
k∈ZZ
∞
0
ds e
−(2+˜ ω
2
)s
I
Nk
(2s)
=
¯ hω
2
(
1
2
˜ ω
2
+ 1)
2
−1
¸
k∈ZZ
¸
1
2
˜ ω
2
+ 1 +
(
1
2
˜ ω
2
+ 1)
2
−1
−Nk
. (6.34)
The geometric series is of course easily summed, but to make the result more transparent
we introduce the scaled eﬀective frequency Ω
ωT /N = ω∆τ = ˜ ω ≡ 2 sinh(
1
2
Ω) (6.35)
and using the identity
1
2
˜ ω
2
+ 1 = cosh(Ω), one easily ﬁnds that
∆τ
¯ h
∂
∂Ω
F(Ω) =
1
2
¸
k∈ZZ
e
−NΩk
=
1
1 −e
−NΩ
−
1
2
= −
1
N
∂
∂Ω
log
e
−ΩN/2
1 −e
−NΩ
. (6.36)
The last identity can be seen as the free energy of the harmonic oscillator with the frequency
ΩN/T = Ω/∆τ, as it can also be written as
∆τ
¯ h
∂
∂Ω
F(Ω) = −
1
N
∂
∂Ω
log
∞
¸
n=0
e
−(n+
1
2
)ΩN
. (6.37)
Amazingly, even at ﬁnite N the euclidean path integral agrees with the quantum partition
function of a harmonic oscillator, but with a frequency that is modiﬁed by the discretisation,
see eqs. (6.21) and (6.35). It is trivial to check now that the limit N → ∞ is well deﬁned
and gives the required result, since
lim
N→∞
ΩN/T = ω . (6.38)
In general the exact ﬁnite N path integral is no longer of a simple form. Nevertheless,
one can evaluate this exact expression in relatively simple terms (which will verify the above
24
result along a diﬀerent route, see also problem 10). So from now on we will take the potential
arbitrary and in a sense we follow the derivation of the path integral in the reverse order.
Z
N
(x
′
, x; T ) =
dp
0
2π¯ h
N−1
¸
i=1
dx
i
dp
i
2π¯ h
exp
¸
∆τ
¯ h
N−1
¸
i=0
ip
i
(x
i+1
−x
i
)
∆τ
−
p
2
i
2m
−V (x
i
)
¸
=
dp
0
N−1
¸
i=1
dx
i
dp
i
N−1
¸
j=0
< x
j+1
p
j
>< p
j
 exp
−
∆τ ˆ p
2
2m¯ h
exp
−
∆τV (ˆ x)
¯ h
x
j
>
= < x
′

exp
−
∆τ ˆ p
2
2m¯ h
exp
−
∆τV (ˆ x)
¯ h
¸
N
x > . (6.39)
This means that we can deﬁne an eﬀective Hamiltonian by
e
−H(N)T /¯h
≡
exp
−
∆τ ˆ p
2
2m¯ h
exp
−
∆τV (ˆ x)
¯ h
¸
N
. (6.40)
But this Hamiltonian is not hermitian as one easily checks from the above expression, since
under conjugation the order of the exponents containing the kinetic and potential terms is
reversed. This can be corrected in two ways
e
−H
1
(N)∆τ/¯h
≡ exp
−
∆τ ˆ p
2
4m¯ h
exp
−
∆τV (ˆ x)
¯ h
exp
−
∆τ ˆ p
2
4m¯ h
,
e
−H
2
(N)∆τ/¯h
≡ exp
−
∆τV (ˆ x)
2¯ h
exp
−
∆τ ˆ p
2
2m¯ h
exp
−
∆τV (ˆ x)
2¯ h
, (6.41)
leading to two equivalent expressions for the ﬁnite N path integral
Z
N
(x
′
, x; T ) = < x
′
 exp
−
∆τ ˆ p
2
4m¯ h
exp (−H
1
(N)T /¯ h) exp
∆τ ˆ p
2
4m¯ h
x >
Z
N
(x
′
, x; T ) = < x
′
 exp
∆τV (ˆ x)
2¯ h
exp (−H
2
(N)T /¯ h) exp
−
∆τV (ˆ x)
2¯ h
x > . (6.42)
In particular the partition function is given by
Z
N
=
dx Z
N
(x, x, T ) = Tr(e
−H
1
(N)T /¯h
) = Tr(e
−H
2
(N)T /¯h
) . (6.43)
It is actually not too diﬃcult to show that there exists a unitary transformation U, such
that UH
1
U
†
= H
2
, which shows that both choices are indeed physically equivalent.
In principle we can now compute H
i
(N) for ﬁnite N as an expansion in 1/N, by using
the socalled CampbellBakerHausdorﬀ formula
e
A
e
B
= e
F(A,B)
, F(A, B) = A+B+
1
2
[A, B] +
1
12
[A, [A, B]] +
1
12
[B, [B, A]] +· · · , (6.44)
which is a series in multiple commutators of the, in general noncommuting, operators A and
B. It can be derived by expanding the exponentials, but in the mathematics literature more
elegant constructions are known, based on properties of Lie groups and Lie algebras. These
objects will be discussed in sect. 18. For the harmonic oscillator, working out the products
of the exponential can be done to all orders and one ﬁnds (see problem 10 for details)
∆τH
i
(N) =
ˆ p
2
2M
i
+
1
2
M
i
Ω
2
ˆ x
2
, (6.45)
25
with Ω deﬁned as in eq. (6.35) and the eﬀective masses M
i
deﬁned by
M
1
=
2mtanh(
1
2
Ω)
∆τΩ
, M
2
=
msinh(Ω)
∆τΩ
. (6.46)
One can now explicitly verify that (cmp. eq. (6.37))
Tr(exp(−H
1
(N)T /¯ h)) = Tr(exp(−H
2
(N)T /¯ h)) = exp(−F(Ω)T /¯ h) . (6.47)
Now we have seen that, at least for some examples, the limit of increasingly ﬁner dis
cretisation is in principle well deﬁned, we can think of generalisation to an arbitrary number
of dimensions (n) (for ﬁeld theory even to an inﬁnite number of dimensions).
< x
′
e
−iHT/¯h
x > ≡
Dx(t) exp
¸
i
¯ h
T
0
dt L(x(t),
˙
x(t))
¸
= lim
N→∞
d
n
p
0
(2π¯ h)
n
N−1
¸
i=1
d
n
x
i
d
n
p
i
(2π¯ h)
n
exp
¸
i
∆t
¯ h
N−1
¸
i=0
p
i
·
(x
i+1
−x
i
)
∆t
−H( p
i
, x
i
)
¸
= lim
N→∞
mN
2πiT¯ h
1
2
nN
N−1
¸
i=1
d
n
x
i
exp
¸
i
∆t
¯ h
N−1
¸
i=0
L(x
i
, (x
i+1
−x
i
)/∆t)
¸
. (6.48)
We have purposely also given the expression that involves the path integral as an integral
over phase space, as it shows that the Gaussian integration over the momenta eﬀectuates
the Legendre transform
i p ·
˙
x −i
p
2
2m
−iV (x) = −i
( p −m
˙
x)
2
2m
+ i
m
˙
x
2
2
−iV (x) , (6.49)
which is equivalent to the stationary phase approximation for the momentum integration
δ
δp
( p ·
˙
x −H( p, x)) =
˙
x −
∂H
∂ p
=
˙
x −
p
m
= 0 . (6.50)
An interesting other example of the path integral is the case of the interaction of a charged
particle with a magnetic ﬁeld. In that case one has for the Hamiltonian
H(
ˆ
p,
ˆ
x) =
(
ˆ
p −e
A(
ˆ
x))
2
2m
+ V (
ˆ
x) . (6.51)
Now, however, the matrix element < p
i
 exp(−iH∆t/¯ h)x
i
> will depend on the speciﬁc
ordering for the position and momentum operators in H. Diﬀerent orderings diﬀer by terms
linear in ¯ h, or
A(
ˆ
x) ·
ˆ
p =
ˆ
p ·
A(
ˆ
x) + i¯ h∂
i
A
i
(
ˆ
x) . (6.52)
So, by chosing the socalled Coulomb gauge ∂
i
A
i
(x) = 0, the problem of the operator ordering
disappears. We leave it as an exercise to verify that the action, obtained from the Legendre
transform, is given by
S =
T
0
dt
1
2
m
˙
x
2
−V (x) + e
˙
x ·
A(x)
. (6.53)
Under a gauge transformation A
i
(x) →A
i
(x) +∂
i
Λ(x), one ﬁnds that the action changes to
S
Λ
= S + e{Λ(x(T)) −Λ(x(0))}. Using the path integral this means that
< x
′
 exp(−iH
Λ
T/¯ h)x >= exp(ieΛ(x
′
)/¯ h) < x
′
 exp(−iHT/¯ h)x > exp(−ieΛ(x)/¯ h) .
(6.54)
26
Since it is easily shown that H
Λ
= exp(ieΛ(
ˆ
x)/¯ h) H exp(−ieΛ(
ˆ
x)/¯ h), this proves that the
path integral derived from eq. (6.53) has the correct properties under gauge transformations,
despite the fact that the derivation was performed by ﬁrst going to the Coulomb gauge.
As long as the Hamiltonian is quadratic in the momenta, the stationary phase approx
imation for the momentum integral is exact. However, also for the coordinate integrals we
can use the stationary phase approximation (exact for a harmonic oscillator), which is related
to the WKB approximation in quantum mechanics. It gives a way of deﬁning an expansion
in ¯ h, where in accordance to the correspondence principle, the lowest order term reproduces
the classical time evolution. Indeed the stationary phase condition
δ
−
S
δ
−
x
i
(t)
=
δS
δx
i
(t)
−
d
dt
δS
δ ˙ x
i
(t)
= 0 (6.55)
is precisely solved by the classical solutions, x
cl
(t), with x
cl
(0) = x and x
cl
(T) = x
′
. We
expand around these solutions by writing
x(t) = x
cl
(t) + q(t) , q(0) = q(T) =
0 , (6.56)
such that
S(x) = S(x
cl
) +
1
2
dt
′
dt q
i
(t)
δ
−
2
S(x
cl
)
δ
−
q
i
(t)δ
−
q
j
(t
′
)
q
j
(t
′
) +O(q
3
) . (6.57)
There is no term linear in q
i
(t), as this term is proportional to the equations of motion, or
equivalently to the stationary phase condition. For the simple Lagrangian L =
1
2
m
˙
x
2
−V (x)
one has
δ
−
2
S(x
cl
)
δ
−
q
i
(t)δ
−
q
j
(t
′
)
= −δ(t −t
′
)
mδ
ij
d
2
dt
2
+ M
ij
(t)
, M
ij
(t) ≡
∂
2
V (x)
∂x
i
∂x
j

x=x
cl
(t)
. (6.58)
For the harmonic potential, V =
1
2
mω
2
x
2
, where the stationary phase approximation is
exact, i.e. there are no O(q
3
) corrections. Introducing q(t), however, splits the action in a
classical piece, that depends on the boundary conditions, and a quantum piece described by a
harmonic oscillator action for the ﬂuctuations around the classical path, that is independent
of the boundary conditions and the classical path
S(x) = S(x
cl
) +
T
0
dt (
1
2
m
˙
q
2
−
1
2
mω
2
q
2
) . (6.59)
In practical situations one splits from the action the part quadratic in the coordinates and
velocities, and considers the rest as a perturbation. In that case x
cl
is the classical solution
of the quadratic part only. As this can always be solved exactly, and as nonquadratic path
integrals can rarely be computed explicitly, this will be the way in which we will derive the
Feynman rules for the quantum theory, in terms of which one can eﬃciently perform the
perturbative computations.
7 Path integrals in ﬁeld theory
For a scalar ﬁeld in a ﬁnite volume V = [0, L]
3
the Hamiltonian is given in the Fourier
representation by (see eq. (2.6))
H =
¸
k=2πn/L
1
2
˜ π(
k)
2
+
1
2
(
k
2
+ m
2
) ˜ ϕ(
k)
2
+ V ( ˜ ϕ) + ˜ ϕ(
k)
˜
J(−
k, t)
. (7.1)
27
As ϕ(x) is real we have ˜ ϕ
∗
(
k) = ˜ ϕ(−
k). It is customary to write the term quadratic in the
ﬁelds (the mass term) explicitly, such that the potential V (ϕ) only contains the interaction
terms. If we like we could split the Fourier modes in their real and imaginary components
(the cos(x·
k) and sin(x·
k) modes). Or even simpler is to use Dirichlet boundary conditions,
i.e. ϕ(x) = 0 at the boundaries of the volume, such that the Fourier modes are given by
¸
j
sin(πn
j
x
j
/L) (with n
j
> 0), with real coeﬃcients. In either case, for V (ϕ) = 0 the
Hamiltonian simply describes an inﬁnite set of decoupled harmonic oscillators, which can
be truncated to a ﬁnite set by introducing a socalled momentum cutoﬀ 
k ≤ Λ. In this
case we know how to write the path integral, even in the presence of interactions. The
introduction of a cutoﬀ is called a regularisation. The ﬁeld theory is called renormalisable
if the limit Λ → ∞ can be deﬁned in a suitable way, often by varying the parameters in a
suitable way with the cutoﬀ. The class of renormalisable ﬁeld theory is relatively small. For
a ﬁnite momentum cutoﬀ the path integral is nothing but a simple generalisation of the one
we deﬁned for quantum mechanics in n dimensions, or in the absence of interactions
Z = lim
N→∞
¸
k
(2πi∆t)
−N/2
N−1
¸
j=1
¸
k
d ˜ ϕ
j
(
k) exp
¸
i∆t
N−1
¸
j=0
¸
k
 ˜ ϕ
j+1
(
k) − ˜ ϕ
j
(
k)
2
2∆t
2
(7.2)
−
1
2
(
k
2
+ m
2
) ˜ ϕ
j
(
k)
2
− ˜ ϕ
j
(
k)
˜
J(−
k, j∆t)
¸
≡
D˜ ϕ(
k, t) exp
¸
i
T
0
¸
k
1
2

˙
˜ ϕ(
k, t)
2
−
1
2
(
k
2
+ m
2
) ˜ ϕ(
k, t)
2
− ˜ ϕ(
k, t)
˜
J(−
k, t)
¸
dt
¸
.
One of course identiﬁes ˜ ϕ
j
(
k) = ˜ ϕ(
k, t = j∆t) and performing the Fourier transformation
once more, we can write
T
0
dt
¸
k
1
2

˙
˜ ϕ(
k, t)
2
−
1
2
(
k
2
+ m
2
) ˜ ϕ(
k, t)
2
− ˜ ϕ(
k, t)
˜
J(−
k, t)
¸
=
T
0
dt
V
d
3
x
1
2
(∂
t
ϕ(x, t))
2
−
1
2
(∂
i
ϕ(x, t))
2
−
1
2
m
2
ϕ
2
(x, t) −ϕ(x, t)J(x, t)
¸
=
V ×[0,T]
d
4
x
1
2
∂
µ
ϕ(x)∂
µ
ϕ(x) −
1
2
m
2
ϕ
2
(x) −ϕ(x)J(x)
¸
. (7.3)
The last expression is manifestly Lorentz invariant apart from the dependence on the bound
ary conditions on the ﬁelds (which should disappear once we take L and T to inﬁnity). This
will allow us to perform perturbation theory in a Lorentz covariant way (things are some
what subtle as any ﬁnite choice of the momentum cutoﬀ does break the Lorentz invariance,
and there are some theories where this is not restored when removing the cutoﬀ, i.e. taking
the limit Λ →∞). This achieves a substantial simpliﬁcation over Hamiltonian perturbation
theory. It is now also trivial to reintroduce the interactions, by adding the potential term
to the Lagrange density, and we ﬁnd in yet another shorthand notation for the measure
the following expression for the path integral (implicitly assuming that the boundary values
ϕ(x, 0) and ϕ(x, T) are ﬁxed, prescribed functions)
Z =
Dϕ(x) exp
i
V ×[0,T]
d
4
x
1
2
∂
µ
ϕ(x)∂
µ
ϕ(x) −
1
2
m
2
ϕ
2
(x) −V (ϕ) −ϕ(x)J(x)
¸
.
(7.4)
In principle a path integral should be independent of the discretisation used in order
to deﬁne it. For the euclidean path integral, one particular way that is used quite often is
28
the lattice discretisation, where instead of a momentum cutoﬀ one makes not only time but
also space discrete. This means that the ﬁeld now lives on a lattice and its argument takes
the values ja where j ∈ ZZ
4
and a is the socalled lattice spacing, which in the end should
be taken to zero. By suitably restricting the components of j, with appropriate boundary
conditions on the ﬁelds, one keeps space and time ﬁnite, V = a
3
M
3
and T = aN. This leads
to an integral of the form
(2πa)
−NM
3
/2
¸
j
dϕ
j
exp
¸
−a
4
¸
j
1
2
¸
µ
(ϕ
j+eµ
−ϕ
j
)
2
a
2
+
1
2
m
2
ϕ
2
j
+ V (ϕ
j
) + J
j
ϕ
j
¸
¸
.
(7.5)
where e
µ
is a unit vector in the µ direction, and ϕ
j
is identiﬁed with ϕ(aj). In a sense,
the momentum cutoﬀ Λ is similar to the spacetime cutoﬀ 1/a. The lattice formulation is
very suitable for numerically evaluating the path integral, whereas the momentum cutoﬀ
is suitable for performing perturbation theory around the quadratic approximation of the
action. For the latter we will compute, using the path integral, the same quantity as was
calculated in section 5, using Hamiltonian perturbation theory.
In the presence of a source, the Hamiltonian depends on time and the evolution operator
has to be written in a way that takes the time ordering into account. As the time evolution
operator U(t) satisﬁes the Schr¨odinger equation
i
d
dt
U(t) = H(t)U(t) , U(0) = 1 , (7.6)
its solution can be written as (note the absence of 1/n!)
U(t) = Texp
−i
t
0
H(t)dt
≡
∞
¸
n=0
(−i)
n
t
0
dt
1
t
1
0
· · ·
t
n−1
0
dt
n
H(t
1
) · · · H(t
n
) . (7.7)
For convenience we introduce the notation
U(t
2
, t
1
) ≡ Texp
−i
t
2
t
1
H(t)dt
, t
2
> t
1
, (7.8)
which satisﬁes the property that
U(t
3
, t
2
)U(t
2
, t
1
) = U(t
3
, t
1
) , t
3
> t
2
> t
1
. (7.9)
In section 5 we calculated the matrix element < 0U(T)0 > to second order in the source
(from now on we put ¯ ε = 1). The Lagrangian relevant for the path integral evaluation is
given by L =
1
2
∂
µ
ϕ∂
µ
ϕ−
1
2
m
2
ϕ
2
−Jϕ, with J(x) = 0 for t < 0 and t > T (and for x / ∈ [0, L]
3
).
The vacuum 0 > is the state where all
k oscillators are in their ground state. It turns out
that we do not need an explicit expression for this vacuum wave functional, denoted by
Ψ
0
({ ˜ ϕ(
k)}) =< { ˜ ϕ(
k)}0 >. We have
< 0U(T)0 >=
¸
p
d ˜ ϕ( p)d ˜ ϕ
′
( p) < 0{ ˜ ϕ
′
( p)} >< { ˜ ϕ
′
( p)}U(T){ ˜ ϕ( p)} >< { ˜ ϕ( p)}0 >,
(7.10)
where { ˜ ϕ( p)} plays the role of x and { ˜ ϕ
′
( p)} the role of x
′
. The path integral expression for
the evolution operator therefore becomes
< { ˜ ϕ
′
( p)}U(T){ ˜ ϕ( p)} >=
Dϕ(x) exp
i
T
0
V
L(ϕ(x))d
4
x
=
D˜ ϕ(k) exp
¸
¸i
¸
k,k
0
1
2
(k
2
−m
2
) ˜ ϕ(k)
2
− ˜ ϕ(k)
˜
J(−k)
¸
¸
. (7.11)
29
We have here also performed the Fourier transformation with respect to time, thereby con
verting the path integral measure D˜ ϕ(
k, t) to the multiple integral over the (discrete tempo
ral) Fourier components D˜ ϕ(k), exactly as was done in one dimension, see eq.(6.23), hence
we also ﬁnd a unit Jacobian for this change of variables. Obviously our notations are such
that k ≡ (k
0
,
k) and k
2
= k
2
0
−
k
2
. If we take the limit of space and time to inﬁnity (L →∞
and T → ∞), the sums over k can be converted into integrals. Finally we note that the
oscillatory integrals occurring in the path integral can be dampened by replacing m
2
by
m
2
−iε as this leads to replacement exp(i
L(ϕ)d
4
x) →exp(i
L(ϕ)d
4
x −ε
ϕ
2
d
4
x). This
prescription also allows us to make the analytic continuation to imaginary time, and coin
cides with the prescription derived for the propagator in the Hamiltonian formulation, so
that causality is also properly implemented in the path integral approach. To recover the
result obtained in the Hamiltonian approach, we simply split oﬀ a square
< { ˜ ϕ
′
( p)}U(T){ ˜ ϕ( p)} >=
D˜ ϕ(k) exp
¸
¸i
¸
k,k
0
1
2
(k
2
−m
2
+ iε)
˜ ϕ(k) −
˜
J(k)
k
2
−m
2
+ iε
2
¸
×exp
−
i
2
¸
k

˜
J(k)
2
k
2
−m
2
+ iε
. (7.12)
We can now shift the integration of ˜ ϕ(k) over
˜
J(k)/(k
2
−m
2
+iε) and introduce the Green’s
function in coordinate space (eq. (4.7)) to get
< { ˜ ϕ
′
( p)}U(T){ ˜ ϕ( p)} >=
= exp
−
i
2
d
4
xd
4
y J(x)G(x −y)J(y)
D˜ ϕ(k) exp
¸
¸i
¸
k,k
0
1
2
(k
2
−m
2
+ iε) ˜ ϕ(k)
2
¸
= exp
−
i
2
d
4
xd
4
y J(x)G(x −y)J(y)
< { ˜ ϕ
′
( p)}e
−iH(J=0)T
{ ˜ ϕ( p)} > . (7.13)
In the last step it is crucial to note that the source is taken to vanish for t ≤ 0 and for t ≥ T,
as otherwise the shift we performed in the ﬁeld would have changed the boundary values.
Using eq. (7.10) we now obtain the remarkable result that
< 0U(T)0 >=< 0e
−iH(J=0)T
0 > exp
−
i
2
d
4
xd
4
y J(x)G(x −y)J(y)
(7.14)
to all orders in the source J. Even at the level of a noninteracting scalar ﬁeld theory,
this demonstrates the dramatic simpliﬁcations that can arise from using the path integral
method for calculating quantum amplitudes. One particularly feature that is noteworthy
in the path integral calculation, is that the part of the Lagrangian that is quadratic in the
ﬁelds represents the inverse propagator. This is no accident and is in general the way the
(lowest order) propagator is directly read oﬀ from the Lagangian, since the quadratic part of
the action is the starting point of the perturbative expansion. But before we will derive the
Feynman rules from the perturbative expansion, it will be useful to emphasise that the time
ordering, playing such an important role in the Hamiltonian formulation, is automatically
implemented by the path integral. Furthermore, it will be helpful to understand in more
detail how the source can be used to create and annihilate particles, as this will be our tool
to write down the matrix elements of the evolution operator (the socalled scattering matrix,
or for short Smatrix) with respect to the basis speciﬁed by particle number and momentum
(the socalled Fock space), see eq. (2.9).
30
In the Hamiltonian formulation we consider
< 0 ˆ ϕ(x
′
, t
2
) ˆ ϕ(x, t
1
)0 >, t
2
> t
1
, (7.15)
where the ﬁeld operator (cmp. eq. (2.8)) is given by
ˆ ϕ(x, t) = e
iHt
ˆ ϕ(x)e
−iHt
, (7.16)
such that
< 0 ˆ ϕ(x
′
, t
2
) ˆ ϕ(x, t
1
)0 >=
< 0e
−iH(T−t
2
)
ˆ ϕ(x
′
)e
−iH(t
2
−t
1
)
ˆ ϕ(x)e
−iHt
1
0 >
< 0e
−iHT
0 >
, (7.17)
where we have made use of the fact that the vacuum 0 > is assumed to be an eigenstate of
the Hamiltonian H (without a source term). The normalisation by < 0e
−iHT
0 > is hence a
rather trivial factor. We can even write a similar expression in the presence of the source. In
perturbation theory this would not be needed, but it is useful from a general point of view
to consider this situation too.
In the presence of a time dependent source one can write
< 0U(T) ˆ ϕ(x
′
, t
2
) ˆ ϕ(x, t
1
)0 >=< 0U(T, t
2
) ˆ ϕ(x
′
)U(t
2
, t
1
) ˆ ϕ(x)U(t
1
)0 > . (7.18)
In this case the ﬁeld operators are of course given by
ˆ ϕ(x, t) = U(t)
†
ˆ ϕ(x)U(t) , (7.19)
where in general U(t) depends on the source J. It is now trivial, but a bit tedious, to convert
this matrix element to a path integral. One ﬁrst writes the product of the operators as a
product of matrices in a suitable representation (e.g. the ﬁeld representation { ˜ ϕ( p)} >).
Each of the matrices for the three evolution operators can be written as a path integral,
excluding the integral over the initial and ﬁnal ﬁeld components. The matrix product involves
an integral over the ﬁnal ﬁeld component of the matrix to the right, which is also the initial
ﬁeld component for the matrix to the left. Without the insertion of the ﬁeld operators
ˆ ϕ(x), this would describe the fact that U(t
3
, t
2
)U(t
2
, t
1
) = U(t
3
, t
1
) in the path integral
formulation simply means that one glues the paths in U(t
3
, t
2
) and U(t
2
, t
1
) together by
integrating over ˜ ϕ(
k, t
2
). With the ﬁeld operator sandwiched between the two U matrices
one simply includes its eigenvalue in the integrand over the paths, since the ﬁeld operator
(or its Fourier components) is diagonal on the states { ˜ ϕ( p)} >. The ﬁnal result can be
written as
¸
p
d ˜ ϕ( p)d ˜ ϕ
′
( p) < 0{ ˜ ϕ
′
( p)} > {
Dϕ(x) exp(i
T
t
2
L d
4
x)ϕ(x
′
, t
2
) exp(i
t
2
t
1
L d
4
x) ×
ϕ(x, t
1
) exp(i
t
1
0
L d
4
x) } < { ˜ ϕ( p)}0 > , (7.20)
where we implicitly assumed that the boundary conditions for the ﬁeld ϕ in the path integral
is in momentum space given by ˜ ϕ( p, t = 0) = ˜ ϕ( p) and ˜ ϕ( p, t = T) = ˜ ϕ
′
( p). The way the
time ordering in the path integral is manifest is now obvious. Note that the ﬁeld expectation
values can also be written in terms of derivatives with respect to the sources, which is
particularly simple to derive in the path integral formulation
< 0U(T) ˆ ϕ(x
′
, t
2
) ˆ ϕ(x, t
1
)0 >= −
δ
2
δJ(x
′
, t
2
)δJ(x, t
1
)
< 0U(T)0 > . (7.21)
31
Since we assumed that 0 > is an eigenstate of H (i.e. at J = 0), < 0U
J=0
(T) =< 0e
−iE
0
T
(with E
0
the vacuum energy) and we can bring the trivial phase factor e
−iE
0
T
to the other
side by normalising with < 0U
J=0
(T)0 >, as in eq. (7.17).
< 0 ˆ ϕ(x
′
, t
2
) ˆ ϕ(x, t
1
)0 >
J=0
=
¸
− < 0U(T)0 >
−1
δ
2
< 0U(T)0 >
δJ(x
′
, t
2
)δJ(x, t
1
)
¸
J=0
. (7.22)
where in the path integral formulation one has
< 0U(T)0 >=
¸
p
d ˜ ϕ( p)d ˜ ϕ
′
( p) < 0{ ˜ ϕ
′
( p)} >
Dϕ(x) exp(i
T
0
L d
4
x) < { ˜ ϕ( p)}0 >,
(7.23)
with the boundary conditions as listed below eq. (7.20).
To study the role the source plays in creating and annihilating particles, we will calculate
both in the Hamiltonian and in the path integral formulations the matrix element
< pU(T)0 > ,  p >≡ a
†
( p)0 > . (7.24)
Hence  p > is the oneparticle state with momentum p. Using the result of eqs. (5.7) and
(5.8), which is equivalent to the result Ψ(t) >= U(t)0 >, we ﬁnd for this matrix element
in lowest nontrivial order
< pU(T)0 > = −i
T
0
dt < 0a( p)e
i(t−T)H
0
H
1
(t)e
−itH
0
0 >
= −i
¸
k
T
0
dt < 0a( p)e
i(p
0
( p)+E
0
)(t−T)
˜
J(
k, t)
(a(−
k) + a
†
(
k))
2k
0
(
k)
e
−iE
0
t
0 >
= −i
e
−i(E
0
+p
0
( p))T
2p
0
( p)
T
0
dt
˜
J( p, t)e
ip
0
( p)t
= −i
√
πe
−i(E
0
+p
0
( p))T
p
0
( p)
˜
J(p) . (7.25)
To write down the path integral result, we ﬁrst express the annihilation operator in terms
of the ﬁeld. From eq. (5.3) we ﬁnd
a( p) + a
†
(− p) =
2p
0
( p)
V
d
3
x ˆ ϕ(x)e
−i p·x
, (7.26)
such that (using < 0a
†
(− p) = 0)
< pU(T)0 >=
2p
0
( p)
V
d
3
x e
−i p·x
< 0 ˆ ϕ(x)U(T)0 > . (7.27)
We now use eqs. (7.14), (7.19) and an obvious generalisation of eq. (7.21), such that
< 0 ˆ ϕ(x)U(T)0 > = < 0U(T) ˆ ϕ(x, T)0 >= i
δ
δJ(x, T)
< 0U(T)0 >
= ie
−iE
0
T
δ
δJ(x, T)
exp
−
i
2
d
4
xd
4
y J(x)G(x −y)J(y)
.(7.28)
We evaluate this to linear order in the source J, using that in a ﬁnite volume the Green’s
function is given by
G(x −y) =
1
V
¸
k
dk
0
2π
e
−ik(x−y)
k
2
−m
2
+ iε
, (7.29)
32
such that
< pU(T)0 > = e
−iE
0
T
2p
0
( p)
V
d
3
x e
−i p·x
d
4
y G(x −y)J(y) +O(J
3
)
= e
−iE
0
T
p
0
( p)
π
dp
0
˜
J(p)e
−ip
0
T
p
2
−m
2
+ iε
+O(J
3
) . (7.30)
Note that in the last step we integrate over p
0
as a dummy variable, which in the expression
for the Green’s function above is called k
0
 this renaming is just for ease of notation. Also,
x
0
= T is assumed. For the p
0
integration we need the analytic behaviour of
˜
J(p) for
imaginary p
0
in order to see if we are allowed to deform the integration contour such that
only one of the poles in the integrand contribute. Since
˜
J(p) =
T
0
dt e
ip
0
t
˜
J( p, t)/
√
2π,
˜
J(p)
will vanish for Imp
0
→ ∞, whereas e
−ip
0
T
˜
J(p) will vanish for Imp
0
→ −∞. In eq. (7.30)
the p
0
integration can therefore be deformed to the lower halfplane in a clockwise fashion
giving a minus sign and a residue from the pole at p
0
= p
0
( p) ≡
√
p
2
+ m
2
, which yields the
result
< pU(T)0 >= −i
√
πe
−i(E
0
+p
0
( p))T
p
0
( p)
˜
J(p) exp
−
i
2
d
4
xd
4
y J(x)G(x −y)J(y)
. (7.31)
To linear order in the source J this coincides with the result of eq.(7.25). Again, the path
integral trivially allows an extension to arbitrary order in the source, as indicated.
For later use, we will also consider the matrix element
< 0U(T) p > =
2p
0
( p)
V
d
3
x e
i p·x
< 0U(T) ˆ ϕ(x, 0)0 >
= i
2p
0
( p)
V
d
3
x e
i p·x
δ< 0U(T)0 >
δJ(x, 0)
. (7.32)
The analogue of eq. (7.30) becomes
< 0U(T) p > = e
−iE
0
T
2p
0
( p)
V
d
3
x e
i p·x
d
4
y G(x −y)J(y) +O(J
3
)
= e
−iE
0
T
p
0
( p)
π
dp
0
˜
J(− p, p
0
)
p
2
−m
2
+ iε
+O(J
3
) , (7.33)
in which case x
0
= 0 is assumed. Now we must deform the contour for the p
0
integration
to the upper halfplane in a counterclockwise fashion such that the residue at the pole
p
0
= −p
0
( p) contibutes. This gives, analogously to eq. (7.31),
< 0U(T) p >= −i
√
πe
−iE
0
T
p
0
( p)
˜
J(−p) exp
−
i
2
d
4
xd
4
y J(x)G(x −y)J(y)
. (7.34)
Apart from the trivial diﬀerence of the factor exp(−ip
0
T), we see that the amplitude for
the annihilation of a oneparticle state is proportional to
˜
J(p) whereas it is proportional
(exactly with the same factor) to
˜
J(−p) for the creation of a oneparticle state (in both
cases p
0
= p
0
( p)).
33
8 Perturbative expansion in ﬁeld theory
As we have seen in the previous section (
Dϕ(x) where relevant includes groundstate factors)
Z(J, g
n
) ≡< 0U(T)0 >=
Dϕ(x) exp(i
d
4
x L(ϕ)) (8.1)
will play the role of a generating functional for calculating expectation values of products of
ﬁeld operators, which will now be studied in more detail. In general the Lagrange density
for a scalar ﬁeld theory is given by
L(ϕ) = L
2
(ϕ) −V (ϕ) −J(x)ϕ(x) , (8.2)
where L
2
(ϕ) is quadratic in the ﬁelds, hence for a scalar ﬁeld
L
2
(ϕ) =
1
2
∂
µ
ϕ(x)∂
µ
ϕ(x) −m
2
ϕ
2
(x)
, V (ϕ) =
g
3
3!
ϕ
3
(x) +
g
4
4!
ϕ
4
(x) +· · · . (8.3)
As mentioned before, it is customary to not include the mass term in the potential V ,
such that V describes the interactions. We can add the interaction as an operator, when
evaluating the path integral for the quadratic approximation L(ϕ) = L
2
(ϕ) −ϕ(x)J(x),
Z(J, g
n
) =
Dϕ(x) exp
i
d
4
x {L
2
(ϕ) −ϕ(x)J(x)}
exp
−i
d
4
x V (ϕ)
= < 0U
2
(T)Texp
−i
d
4
x V ( ˆ ϕ)
0 > . (8.4)
We can now use the fact that
Dϕ(x)
¸
j
ϕ(x
(j)
) exp
i
d
4
x (L
2
(ϕ) −ϕ(x)J(x))
=
¸
j
iδ
δJ(x
(j)
)
Dϕ(x) exp
i
d
4
x (L
2
(ϕ) −ϕ(x)J(x))
=
¸
j
iδ
δJ(x
(j)
)
exp
−
i
2
d
4
xd
4
y J(x)G(x −y)J(y)
, (8.5)
to ﬁnd a somewhat formal, but in an expansion with respect to the coupling constants g
n
well deﬁned expression for the fully interacting path integral
Z(J, g
n
) = exp
−i
d
4
x V (
iδ
δJ(x)
)
exp
−
i
2
d
4
xd
4
y J(x)G(x −y)J(y)
≡ exp
−i
d
4
x V (
iδ
δJ(x)
)
Z
2
(J) , (8.6)
We have assumed the vacuum energy to be normalised to zero, in absence of interactions, such
that Z(J = g
n
= 0) = 1. Equivalently, Z(J, g
n
) is synonymous with Z(J, g
n
)/Z(J = g
n
= 0).
We now deﬁne G
J
as
G
J
≡ log Z(J, g
n
) , (8.7)
where the dependence on the coupling constants in G
J
is implicit. We will show that
G
J
can be seen as the sum of all connected diagrams. A diagram is connected if it can
not be decomposed in the product of two diagrams that are not connected. Note that at
34
J = 0, iG
J
/T equals the energy of the ground state as a function of the coupling constants,
normalised so as to vanish at zero couplings.
The diﬀerent diagrams arise from the expansion of
exp
−i
d
4
x V (
iδ
δJ(x)
)
= exp
¸
−i
d
4
x
¸
ℓ=3
g
ℓ
ℓ!
iδ
δJ(x)
ℓ
¸
(8.8)
in powers of g
ℓ
. Each factor
g
ℓ
ℓ!
(iδ/δJ(x))
ℓ
will represent an ℓpoint vertex, with coordinate
x, which is to be integrated over. As we saw in the derivation of the classical equations
of motion, the integral over x in the Fourier representation gives rise to conservation of
momentum at the vertex. Using
ϕ(x) =
1
(2π)
2
d
4
k e
−ikx
˜ ϕ(k) , J(x) =
1
(2π)
2
d
4
k e
−ikx
˜
J(k) , (8.9)
we have for each vertex in the Fourier representation
−ig
ℓ
d
4
x
iδ
δJ(x)
ℓ
= −ig
ℓ
(2π)
4
δ
4
(
ℓ
¸
j=1
k
(j)
)
ℓ
¸
j=1
d
4
k
(j)
(2π)
2
iδ
δ
˜
J(k
(j)
)
. (8.10)
Note that factors of 2π are dropping out in the identities
exp
−
i
2
d
4
xd
4
y J(x)G(x −y)J(y)
= exp
−
i
2
d
4
k
˜
J(k)
˜
J(−k)
k
2
−m
2
+ iε
,
d
4
x ϕ(x)J(x) =
d
4
k ˜ ϕ(k)
˜
J(−k) . (8.11)
In the quantum theory we have to keep track of the factors i. Compared to the Feynman
rules of section 4, the propagator will come with an extra factor −i. A vertex will now
carry a factor ig
ℓ
((2π)
2
/i)
2−ℓ
(in a ﬁnite volume this becomes ig
ℓ
(
√
2πV /i)
2−ℓ
), see the table
below. To compute the vacuum energy there is an overall factor i, since E
0
T = iG
J=0
=
i log Z(J = 0, g
n
). The same factor of i applies for using the treelevel diagrams to solve
the classical equations of motion. It is easy to see that these Feynman rules give identical
results for these treelevel diagrams, as compared to the Feynman rules introduced in section
4. The factors of i exactly cancel each other.
coordinate space momentum space table 2
>
>
x
≡ i
ℓ−1
g
ℓ
d
4
x
>
>
k
1
`
k
2
k
3
> >.
≡ i
ℓ−1
(2π)
4−2ℓ
g
ℓ
δ
4
(
¸
i
k
i
) vertex
x y
≡ −iG(x −y)
k
≡
d
4
k
−i
k
2
−m
2
+iε
propagator
×
x
≡
d
4
xJ(x)
×
k
,
≡
˜
J(k) source
We note that the propagator connected to a source comes down whenever a derivative
in the source acts on Z
2
(J), see eq. (8.7). When this derivative acts on terms that have
already come down from previous derivatives, one of the sources connected to a propagator
35
is removed and this connects that propagator to the vertex associated to δ/δJ(x). As any
derivative is connected to a vertex, the propagator either runs between a vertex and a source
> >
×
, between two vertices
>>
> >
, or it connects two legs of the same vertex
¯
. The
possibility of closed loops did not occur in solving the classical equations of motion, and is
speciﬁc to the quantum theory.
To prove that G
J
only contains connected diagrams we write
G
J
= log {exp(X(iδ/δJ)) exp(Y (J))} 1 , (8.12)
with
X(iδ/δJ(x)) ≡ −i
d
4
x V (δ/δJ(x)) , Y (J) ≡ −
i
2
d
4
xd
4
yJ(x)G(x −y)J(y) . (8.13)
As J(x) and δ/δJ(x) form an algebra (similar to the algebra of ˆ x and ˆ p in quantum mechanics,
however, generalised to inﬁnite dimensions), also X and Y are elements from the algebra, and
we can express G
J
in a sum of multiple commutators using the CampbellBakerHausdorﬀ
formula (see eq. (6.44)).
G
J
=
X + Y +
1
2
[X, Y ] +
1
12
[X, [X, Y ]] +
1
12
[Y, [Y, X]] +· · ·
1 . (8.14)
Due to the multiple commutators, all components that do not commute are connected.
However, if the components would commute they would not contribute to the commutators.
This is even true if we do not put the derivatives with respect to the source to zero, once
they have been moved to the right (this is why we consider the action on the identity).
The only thing that remains to be discussed is with which combinatorial factor each
diagram should contribute. This is, as in section 4, with the inverse of the order of the
permutation group that leaves the topology of the diagram unchanged. These combinatorial
factors are clearly independent of the spacetime integrations and possible contractions of
vector or other indices. We can check them by reducing the path integral to zero dimensions,
or ϕ(x) →ϕ and Dϕ(x) →dϕ. In other words, we replace the path integral by an ordinary
integral. As an example consider
Z(J, g) = C
dϕ exp
i{
1
2
ϕMϕ −
g
3!
ϕ
3
−ϕJ}
= exp
¸
−i
g
3!
i∂
∂J
3
¸
exp
−
i
2
JM
−1
J
. (8.15)
The constant C is simply to normalise Z(J = g
n
= 0) = 1. Expanding the exponents we get
in lowest nontrivial order
Z(J = 0) = 1 −
g
2
2(3!)
3
i∂
∂J
6
−
i
2
JM
−1
J
3
· · ·
= exp
¸
i
g
2
2
4
(3!)
3
∂
∂J
6
JM
−1
J
3
+O(g
3
)
¸
= exp
i
5g
2
24M
3
+O(g
3
)
= exp
1
12
`
+
1
8
` `
+O(g
3
)
. (8.16)
In the last term, the numerical factors in front of the diagrams indicate the combinatorial
factors (for the ﬁrst diagram a factor 2 from interchanging the two vertices and a factor 3!
36
from interchanging the three propagators, for the second diagram the latter factor is replaced
by 4 as we can only interchange for each vertex the two legs that do not interconnect the
two vertices). The Feynman rules for this simple case are that each vertex gets a factor −g
(in zero dimensions there are no factors 2π) and each propagator gets a factor −i/M. In
problem 13 the exponentiation is checked for Z(J) to O(g
2
) and O(J
5
) (giving the simplest
nontrivial check).
We will now show how the number of loops in a diagram is related to the expansion in
¯ h. We can expect such a relation, as we have shown at the end of section 6 that the ¯ h →0
limit is related to the classical equations of motion, whereas we have shown in section 4 that
these classical equations are solved by tree diagrams. If we call L the number of loops of a
diagram, we will show that
Z(J, g
n
) =
Dϕ exp
i
¯ h
(L(ϕ) −ϕJ)
≡ exp(G
J
/¯ h) , G
J
=
∞
¸
L=0
¯ h
L
G
L,J
, (8.17)
where G
L,J
is the sum of all connected diagrams with exactly L loops. This means that a
loop expansion is equivalent with an expansion in ¯ h. To prove this we ﬁrst note that due to
reinstating ¯ h the source term will get an extra factor 1/¯ h, the propagator a factor ¯ h and the
coupling constants g
n
are replaced by g
n
/¯ h. A diagram with V
n
npoint vertices, E external
lines (connected to a source) and P propagators has therefore an extra overall factor of
¯ h
−E
¯ h
P
¸
n=3
¯ h
−Vn
. (8.18)
We can relate this to the number of loops by noting that the number of momentum in
tegrations (i.e. the number of independent momenta) in a diagram equals the number of
loops plus the number of external lines, minus one for the overall conservation of energy and
momentum, i.e. L+E−1. On the other hand, the number of momentum integrations is also
the number of propagators minus the number of delta functions coming from the vertices,
i.e. P −
¸
n=3
V
n
. Hence
L = 1 +P −E −
¸
n=3
V
n
, (8.19)
which implies that the total number of ¯ h factors in a diagram is given by L−1. In the next
sections we will often consider socalled amputated diagrams, where the external propagators
connected to a source are taken oﬀ from the expressions for the diagram. If we do not count
these external propagators, eq. (8.19) has to be replaced by L = 1 + P −
¸
n=3
V
n
, as there
are exactly E such external propagators.
9 The scattering matrix
We would like to compute the amplitude for the transition of n incoming particles at t = T
in
to ℓ outgoing particles at t = T
out
in the limit where T
out
→ ∞ and T
in
→ −∞. The
diﬀerence with quantum mechanics is that the particle number is no longer conserved.
out
< p
1
, p
2
, · · · , p
ℓ

k
1
,
k
2
, · · · ,
k
n
>
in
≡< p
1
, p
2
, · · · , p
ℓ
U(T
out
, T
in
)
k
1
,
k
2
, · · · ,
k
n
> . (9.1)
In terms of creation and annihilation operators this can be written as
out
< p
1
, p
2
, · · · , p
ℓ

k
1
,
k
2
, · · · ,
k
n
>
in
=
< 0a( p
1
)a( p
2
) · · · a( p
ℓ
)U(T
out
, T
in
)a
†
(
k
1
)a
†
(
k
2
) · · · a
†
(
k
n
)0 > . (9.2)
37
From eqs. (7.26), (7.28) and (7.32) we know how to implement these creation and annihilation
operators on the generating functional Z(J, g
n
)
ˆ a
†
( p) = i
2p
0
( p)
V
d
3
x e
i p·x
δ
δJ(x, t = T
in
)
= i
2p
0
( p)
δ
δ
˜
J( p, t = T
in
)
,
ˆ a( p) = i
2p
0
( p)
V
d
3
x e
−i p·x
δ
δJ(x, t = T
out
)
= i
2p
0
( p)
δ
δ
˜
J(− p, t = T
out
)
. (9.3)
This implies the following identity for the scattering matrix
out
< p
1
, p
2
, · · · , p
ℓ

k
1
,
k
2
, · · · ,
k
n
>
in
=
ℓ
¸
i=1
ˆ a( p
i
)
n
¸
j=1
ˆ a
†
(
k
j
) exp(G
J
)
J=0
. (9.4)
In principle, this allows us to calculate the scattering, taking T
in
→−∞ and T
out
→∞.
There is, however, a problem to associate the particle states in the presence of interactions
with the ones we have derived from the noninteracting theory. The problem is that particles
can have selfinteractions long before and after the diﬀerent particles have scattered oﬀ each
other. We have to reconsider our notion of particle states, as in experiments we are unable
to switch oﬀ these selfinteractions. For simplicity we assume that the oneparticle states
are stable, as in the simple scalar theory we have been considering. This implies from
conservation of probability that
d
3
k 
out
< p
k >
in

2
= 1 , (9.5)
independently of p. In general, conservation of probability implies that the Smatrix is
unitary. Formally, unitarity of a Smatrix is guaranteed as soon as the Hamiltonian is
hermitian. Because of the necessity to regulate the quantum theory, e.g. by introducing a
cutoﬀ, this is generally no longer true and one has to show that unitarity is restored when
the cutoﬀ is removed. If this is not possible, the theory is illdeﬁned, or at best does not
make sense above the energies where unitarity is violated.
For the free theory, unitarity is of course satisﬁed. In this case, the only diagram con
tributing to G
J
is the one with a single propagator connecting two sources, which is also
called the connected twopoint function G
(2)
c
(J)
G
J
= G
(2)
c
(J) = −
i
2
dtds
¸
p
dp
0
2π
e
−ip
0
(t−s)
˜
J( p, t)
˜
J(− p, s)
p
2
−m
2
+ iε
. (9.6)
This implies that (T ≡ T
out
−T
in
)
out
< p
k >
in
= ˆ a( p)ˆ a
†
(
k) exp(G
(2)
c
(J))
J=0
= 2ip
0
( p)δ
k, p
dp
0
2π
e
ip
0
T
p
2
−m
2
+ iε
= e
−ip
0
( p)T
δ
k, p
,
(9.7)
where the p
0
integration is performed by deforming the contour to the upper halfplane,
giving a contribution from the pole at p
0
= −p
0
( p) only. It is trivial to see that this is the
same as what can be obtained within the Hamiltonian formulation.
In the presence of interactions, this result is no longer true, since the connected twopoint
function will deviate from the one in the free theory. In this case we can write (from now on
38
the symmetry factors will be absorbed in the expression associated to a diagram)
G
(2)
c
(J) = × × +
`
× ×
+
`
`
× ×
+· · ·
≡
`
× ×
, (9.8)
where we have written the connected twopoint function in terms of the oneparticle irre
ducible (1PI) twopoint function iΣ(p) (Σ is the socalled selfenergy)
iΣ(p) ≡
`
. (9.9)
In general, a 1PIgraph is a connected graph that remains connected when one arbitrary
propagator is being cut (except when cutting away a tadpole of the form
¯
, which we
will not allow. However, these tadpoles describe single particles popping in or out of the
vacuum, usually required to be absent. They can be removed by shifts in the ﬁelds.) The
external lines of these diagrams will carry no propagator. The diagrams of eq. (9.8) can be
converted to the result
1
2
d
4
p 
˜
J(p)
2
−i
p
2
−m
2
+ iε
+
−i
p
2
−m
2
+ iε
2
iΣ(p) +
−i
p
2
−m
2
+ iε
3
(iΣ(p))
2
+· · ·
=
1
2
d
4
p 
˜
J(p)
2
−i
p
2
−m
2
+ iε
∞
¸
n=0
Σ(p)
p
2
−m
2
+ iε
n
¸
= −
i
2
d
4
p

˜
J(p)
2
p
2
−m
2
−Σ(p) + iε
≡ G
(2)
c
(J) . (9.10)
Normally the selfenergy will not vanish at p
2
= m
2
, such that the selfinteractions shift the
pole in the twopoint function to another value, ˜ m
2
, i.e.
p
2
−m
2
−Σ(p) = 0 for p
2
0
= p
2
+ ˜ m
2
. (9.11)
Consequently, the mass of the oneparticle states is shifted (or renormalised). As we cannot
switch oﬀ the interactions in nature, the true or observable mass is ˜ m and not m, the latter is
also called the bare mass. The residue at the poles (i.e. p
2
= ˜ m
2
, called the massshell) will
in general also change from ±πi/p
0
( p) to ±πiZ/p
0
( p). On the massshell (i.e.
˜
J(p) vanishes
rapidly as a function of p
2
− ˜ m
2
) one therefore has
G
(2)
c
(J) = −
i
2
d
4
p
Z
˜
J(p)
2
p
2
− ˜ m
2
+ iε
. (9.12)
As long as the oneparticle states are stable, eq. (9.5) needs to remain valid, which can only
be achieved (see eq. (9.7)) by rescaling the wave functionals with a factor
√
Z (this is called
wavefunction renormalisation). It implies that eq. (9.4) needs to be modiﬁed to
out
< p
1
, p
2
, · · · , p
ℓ

k
1
,
k
2
, · · · ,
k
n
>
in
=
ℓ
¸
i=1
ˆ a
−
( p
i
)
n
¸
j=1
ˆ a
+
(
k
j
) exp(G
J
)
J=0
, (9.13)
where
ˆ a
+
( p) ≡ i
2
√
p
2
+ ˜ m
2
Z
1
2
δ
δ
˜
J( p, t = T
in
)
, ˆ a
−
( p) ≡ i
2
√
p
2
+ ˜ m
2
Z
1
2
δ
δ
˜
J(− p, t = T
out
)
.
(9.14)
39
As for the free theory, each of these operators ˆ a
±
( p) will replace one external line (propagator
plus source) by an appropriate wavefunction factor and puts these external lines on the
massshell. We will call the connected npoint function with amputated external lines the
amputated connected npoint function G
(amp)
c (p
1
, p
2
, · · · , p
n
) (in general not oneparticle
irreducible), i.e.
G
(n)
c
(J) ≡
n
¸
i=1
d
4
p
i
−i
˜
J(p
i
)
p
2
i
−m
2
−Σ(p
i
) + iε
¸
G
(amp)
c
(p
1
, p
2
, · · · , p
n
) . (9.15)
Diagrammatically this looks as follows
.
.
.
.
.
.
.
.
.
.
.
.
>
>
>
>
>>
> >
× ×
× ×
`
`
`
`
(9.16)
To get the Smatrix we have to compute (see eq. (9.13))
¸
ℓ
¸
i=1
ˆ a
−
( p
i
)
n
¸
j=1
ˆ a
+
(
k
j
) e
G
J
¸
J=0
= e
G
J=0
ℓ
¸
i=1
ˆ a
−
( p
i
)
n
¸
j=1
ˆ a
+
(
k
j
)
¸
¸
Σrqr=ℓ+n
¸
r
1
q
r
!
G
(r)
c
(J)
¸
qr
¸
.
(9.17)
The sum is over all possible partitions of ℓ + n. Let us ﬁrst consider the most important
term, corresponding to connected graphs, where q
1
= ℓ + n
ℓ
¸
i=1
ˆ a
−
( p
i
)
n
¸
j=1
ˆ a
+
(
k
j
) G
(ℓ+n)
c
(J)=
n
¸
j=1
−i
πZ/k
(j)
0
ℓ
¸
i=1
−i
πZ/p
(i)
0
G
(amp)
c
({−p
i
}, {k
j
})e
−iΣp
(i)
0
T
≡
−2πiV δ
4
(
¸
ℓ
i=1
p
i
−
¸
n
j=1
k
j
) M
ℓ+n
({−p
i
}, {k
j
})
¸
ℓ
i=1
2p
(i)
0
V
¸
n
j=1
2k
(j)
0
V
e
−iΣp
(i)
0
T
. (9.18)
Here M
ℓ
(a Lorentz scalar as we will see later) is the socalled reduced matrix element with
ℓ external lines, all on the massshell. Note that we have extracted the trivial energy factors
(remember that
¸
p
(i)
0
T
out
−
¸
k
(j)
0
T
in
=
¸
p
(i)
0
(T
out
− T
in
) ≡
¸
p
(i)
0
T), such that the limits
T
in
→ −∞ and T
out
→ ∞ can be taken. Each ˆ a( p)
±
will act on one of the factors between
curly brackets in eq. (9.15). Concentrating on one such a factor we have
d
4
p
−i
˜
J(p)G
(amp)
c (p, p
2
, · · · , p
n
)
p
2
−m
2
−Σ(p) + iε
=
d
4
p
√
2π
dt
−i
˜
J( p, t)e
ip
0
t
G
(amp)
c (p, p
2
, · · · , p
n
)
p
2
−m
2
−Σ(p) + iε
.
(9.19)
such that, using eq. (9.14)
ˆ a
−
( p)
d
4
p
−i
˜
J(p)G
(amp)
c (p, p
2
, · · · , p
n
)
p
2
−m
2
−Σ(p) + iε
=
p
0
( p)
πZ
dp
0
e
ip
0
Tout
G
(amp)
c ((p
0
, − p), · · · , p
n
)
p
2
−m
2
−Σ(p) + iε
= −i
√
2πV Z
G
(amp)
c (−p, p
2
, · · · , p
n
)
2p
0
( p)V
e
−ip
0
( p)Tout
. (9.20)
Since T
out
→ ∞, we can extend the p
0
contour integration to the upper halfplane, under
mild regularity conditions for G
(amp)
c (p, p
2
, · · · , p
n
) as Imp
0
→−∞(that can easily be shown
40
to be satisﬁed at any ﬁnite order in perturbation theory). Thus, the integral over p
0
only
gets a contribution from the pole at p
0
= −p
0
( p) = −
√
p
2
+ ˜ m
2
, with residue −Z/2p
0
( p)
(cmp. eq. (9.12)). For the creation operators one similarly ﬁnds
ˆ a
+
( p)
d
4
p
−i
˜
J(p)G
(amp)
c (p, p
2
, · · · , p
n
)
p
2
−m
2
−Σ(p) + iε
=
p
0
( p)
πZ
dp
0
e
ip
0
T
in
G
(amp)
c ((p
0
, p), · · · , p
n
)
p
2
−m
2
−Σ(p) + iε
= −i
√
2πV Z
G
(amp)
c (p, p
2
, · · · , p
n
)
2p
0
( p)V
e
ip
0
( p)T
in
. (9.21)
Likewise, as T
in
→−∞, we can now extend the p
0
contour integration to the lower halfplane
(under the same regularity conditions for the amputated npoint functions as for eq. (9.20)
to be valid), such that we pick up the contribution of the pole at p
0
= p
0
( p) with residue
Z/2p
0
( p). Combining these results proves the ﬁrst identity in eq. (9.18), the second is merely
a deﬁnition.
Note that the derivation is not valid for the case ℓ + n = 2, where the generalisation of
eq. (9.7) to the interacting case implies
out
< p
k >
in
= ˆ a
−
( p)ˆ a
+
(
k) exp(G
J
)
J=0
= δ
3
(
k −p)e
−i[E
0
+p
0
( p)]T
. (9.22)
Here we used eq. (9.12) and the fact that exp(G
J=0
) = exp(−iE
0
T), which is often also
normalised to 1, but till now we had only required this to be the case at zero couplings.
The reason this case is special is because eq. (9.15) requires us to deﬁne for the amputated
twopoint function
G
(amp)
c
(p
1
, p
2
) ≡ iδ
4
(p
1
+ p
2
)(p
2
1
−m
2
−Σ(p
1
) + iε) , (9.23)
which vanishes on the massshell. In using eq. (9.20) and eq. (9.21) it was implicitly assumed
that the amputated Green’s function has no zero that will cancel the pole.
The Feynman rules in momentum space for computing the reduced matrix elements will
obviously have to be modiﬁed for the external lines to a factor −i
√
2πV Z and an overall
factor i/(2πV ) (as always, in an inﬁnite volume one replaces V by (2π)
3
). If we associate a
momentum delta function to a vertex and a momentum integration to a propagator (as was
done up to now), the delta function for overall energy and momentum conservation should
not be written explicitly in eq. (9.18), since it is contained in the reduced matrix element.
Instead, if we choose to integrate over the independent loop momenta, implementing energy
and momentum conservation at each vertex (as will be done from now on, see the table
below), the deﬁnition of eq. (9.18) is the appropriate one. The overall factors of i, 2π and
V can be determined with the help of the two identities
L = P + 1 −
¸
n
V
n
, E + 2P =
¸
n
nV
n
. (9.24)
The proof for the ﬁrst identity was discussed below eq. (8.19). For the second identity we
put a dot on each end of a propagator (
) and one dot on each external line (
),
giving a total of 2P +E dots. The same dots can also be associated to each line of a vertex
(
> >
), giving
¸
nV
n
dots, thus proving the second identity (see also problem 15). To keep
the derivation general, we evaluate the overall factor in a ﬁnite volume
i
−P
i
−E+1
i
Σ(n−1)Vn
(2πV )
−
1
2
Σ(n−2)Vn
(2πV )
1
2
E−1
=
i
2πV
L
. (9.25)
41
This implies that we can shift all numerical factors from the propagators, vertices and ex
ternal lines to a factor i/(2πV ) (or i/(2π)
4
in an inﬁnite volume) for each loop, giving the
Feynman rules listed in the table for an inﬁnite volume. Note that the extraction of the factor
−2πiV in the deﬁnition of M is merely a convention (such that in lowest order M
n
= g
n
).
momentum space Itzykson and Zuber table 3
>
>
k
1
`
k
2
k
3
> >.
≡ g
ℓ
and
¸
k
i
= 0
>
>
k
1
`
k
2
k
3
> >.
≡ −i(2π)
4
g
ℓ
δ
4
(
¸
i
k
i
) vertex
k
≡
1
k
2
−m
2
+iε
k
≡
d
4
k
(2π)
4
i
k
2
−m
2
+iε
propagator
≡
√
Z
≡
√
Z external line
i
d
4
k
(2π)
4
1 loop factor
In the literature many diﬀerent conventions are being used. As an example, the table com
pares our Feynman rules with those of Itzykson and Zuber. Their convention for M
n
is
likewise to make it coincide to lowest order with the npoint vertex. However, as the latter
does already contain a factor −i(2π)
4
(in a ﬁnite volume −2πiV ), that factor should be
absent in relating the reduced matrix element to the amputated npoint function. Com
bining the extra factors of i and 2π in the Feynman rules of Itzykson and Zuber gives
i(2π)
−4
i
P−ΣVn
(2π)
4(ΣVn−P)
= [i/(2π)
4
]
L
, guaranteeing equivalence of the two sets of Feyn
man rules.
Concerning the symmetry factor associated to a particular diagram we note the following.
As we have generally ﬁxed the external momenta, interchanging external lines is no longer
allowed. But from eq. (9.15) we see that the symmetry factor n!, to be taken into account
for G
(n)
c
(J), will be compensated by the n derivatives on n sources. Hence, in computing
the reduced matrix elements the symmetry factors are determined without allowing for
permutations on the external lines.
To conclude this section we return to eq. (9.17), and discuss the contributions that will
be associated to the diagrams that are not connected. Each factor of q
r
! is compensated
for by the diﬀerentiations on
G
(r)
c
(J)
¸
qr
. For the corresponding connected components the
rules are identical to the ones speciﬁed above. In particular each connected component will
carry its own factor −i(2π)
4
δ
4
(
¸
p
i
) for the conservation of energy and momentum (p
i
is
now assumed to run over a subset of both the incoming and the negative of the outgoing
fourmomenta). In a physical picture the disconnected parts correspond to situations where
only a subset of the incoming particles will interact with each other (the ones connected
by a particular diagram). Quite often, the experimental situation is such that the energy
momentum conservation will only be compatible with the fully connected part. We just
have to avoid the incoming momenta to coincide with any of the outcoming momenta. In a
collider, this means one excludes particles that escape in the direction of the beams, where
indeed it is not possible to put a detector. As an illustration we will give the situation for
n = ℓ = 2 and all momenta nonzero (to avoid tadpole diagrams,
¯
) to second order in
42
the threepoint coupling g
3
, putting all other couplings to zero
out
< p
1
, p
2

k
1
,
k
2
>
in
= exp
−i[E
0
+ p
(1)
0
+ p
(2)
0
]T
× (9.26)
¸
δ
3
( p
1
−
k
1
)δ
3
( p
2
−
k
2
)
+ δ
3
( p
1
−
k
2
)δ
3
( p
2
−
k
1
)
.
.
.
.
. 



−i
(2π)
4
δ
4
(p
1
+ p
2
−k
1
−k
2
)
2p
(1)
0
(2π)
3
2p
(2)
0
(2π)
3
2k
(1)
0
(2π)
3
2k
(2)
0
(2π)
3
+
+O(g
3
3
)
¸
.
The ﬁrst two diagrams, which have to be treated with special care (see eqs. (9.22) and
(9.23)), represent the situation without scattering. By deﬁnition they have no higher order
corrections.
10 Cross sections
In many experimental situations we are interested in the scattering of two particles with
momenta k
1
and k
2
to a state with n particles with momenta p
1
, p
2
, · · · , p
n
. We denote
by
d
3
k
1
˜
Ψ
1
(
k
1
)
k
1
> and
d
3
k
2
˜
Ψ
2
(
k
2
)
k
2
> the wave functionals of the incoming particles.
This is to describe the more realistic case of a wave packet. The amplitude for scattering to
take place is hence given by
A = −i(2π)
4
d
3
k
1
d
3
k
2
δ
4
(
¸
n
i=1
p
i
−k
1
−k
2
)M
2+n
({−p
i
}, {k
j
})
¸
n
i=1
2p
(i)
0
( p
i
)(2π)
3
¸
2
j=1
2k
(j)
0
(
k
j
)(2π)
3
˜
Ψ
1
(
k
1
)
˜
Ψ
2
(
k
2
)e
−i[E
0
+Σp
(i)
0
]T
.
(10.1)
If we deﬁne the wave function in coordinate space as usual
Ψ
j
(x) =
d
3
k
2k
0
(
k)(2π)
3
e
−ikx
˜
Ψ
j
(
k) , (10.2)
we can compute the overlap of the two wave functions
d
4
x Ψ
1
(x)Ψ
2
(x)e
ipx
= (2π)
4
d
3
k
1
2k
0
(
k
1
)(2π)
3
d
3
k
2
2k
0
(
k
2
)(2π)
3
δ
4
(k
1
+k
2
−p)
˜
Ψ
1
(
k
1
)
˜
Ψ
2
(
k
2
) .
(10.3)
We assume that over the range of momenta in the wave packets, the reduced matrix elements
are constant (which can be achieved with arbitrary precision for arbitrarily narrow wave
packets in momentum space). This allows us to write for the scattering probability of two
particles into n particles, with momenta in between p
i
and p
i
+ dp
i
,
dW = M({−p
i
}, {
¯
k
j
})
2
f(p)
n
¸
i=1
d
3
p
i
2p
0
( p
i
)(2π)
3
,
f(p) ≡
d
4
xd
4
y Ψ
1
(x)Ψ
2
(x)Ψ
∗
1
(y)Ψ
∗
2
(y)e
ip(x−y)
, (10.4)
where p =
¸
n
i=1
p
i
= k
1
+k
2
. The momenta
¯
k
i
in the reduced matrix element are the central
values of the wave packet in momentum space for the two incoming particle beams. Under
the same assumption that the momentum spread in the beams is very small, the function
f(p) will be highly peaked around p =
¯
k
1
+
¯
k
2
, such that
f(p) ≈ δ
4
(p −
¯
k
1
−
¯
k
2
)
d
4
p f(p) = (2π)
4
δ
4
(p −
¯
k
1
−
¯
k
2
)
d
4
xΨ
1
(x)
2
Ψ
2
(x)
2
. (10.5)
43
The quantities Ψ
j
(x)
2
are of course related to the probability densities of the two particles
in their respective beams,
ρ
j
(x) ≡ i(Ψ
∗
j
(x)∂
t
Ψ
j
(x) −Ψ
j
(x)∂
t
Ψ
∗
j
(x)) ≈ 2
¯
k
(j)
0
Ψ
j
(x)
2
, (10.6)
again using the fact that the wave packet is highly peaked in momentum space. Putting
these results together we ﬁnd
dW = (2π)
4
δ
4
(
¸
p
i
−
¯
k
1
−
¯
k
2
)M({−p
i
}, {
¯
k
j
})
2
n
¸
i=1
d
3
p
i
2p
0
( p
i
)(2π)
3
d
4
x
ρ
1
(x)ρ
2
(x)
4
¯
k
(1)
0
¯
k
(2)
0
. (10.7)
Since ρ
j
(x) will depend on the experimental situation, we should normalise with respect to
the total number of possible interactions in the experimental setup, also called the integrated
luminosity L.
dt L(t) ≡
d
4
x ρ
1
(x)ρ
2
(x)v
1
−v
2
 . (10.8)
Here
d
3
x ρ
1
(x)ρ
2
(x) is the number of possible interactions per unit volume at a given
time and v
1
−v
2
 is the relative velocity of the two beams. We have assumed that either,
one of the velocities is zero (ﬁxed target), or that the two velocities are parallel (colliding
beams). Hence, L(t) =
d
3
x ρ
1
(x, t)ρ
2
(x, t)v
1
− v
2
 is a ﬂux, typically of the order of
10
28
−10
33
cm
−2
s
−1
. To consider the general case we note that we can also write
dt L(t) =
(
¯
k
1
·
¯
k
2
)
2
−m
2
1
m
2
2
d
4
x
ρ
1
(x)ρ
2
(x)
¯
k
(1)
0
¯
k
(2)
0
. (10.9)
After all, for a ﬁxed target situation
¯
k
2
= (m
2
,
0), such that
(
¯
k
1
·
¯
k
2
)
2
−m
2
1
m
2
2
¯
k
(1)
0
¯
k
(2)
0
=
(
¯
k
(1)
0
)
2
−m
2
1
¯
k
(1)
0
= v
1
 , (10.10)
whereas for colliding beams of particles and antiparticles with mass m, where
¯
k
1
= (E,
k) =
E(1, v) and
¯
k
2
= (E, −
k) = E(1, −v), one ﬁnds
(
¯
k
1
·
¯
k
2
)
2
−m
2
1
m
2
2
¯
k
(1)
0
¯
k
(2)
0
=
(E
2
+
k
2
)
2
−m
4
E
2
= 2
k/E = 2v , (10.11)
such that both expressions reduce to v
1
−v
2
. We leave it as an exercise to prove the result
for the general case of parallel beams.
We can therefore deﬁne a machine independent diﬀerential cross section dσ by normalising
the scattering probability by the total luminosity,
dσ = (2π)
4
δ
4
(
n
¸
i=1
p
i
−
¯
k
1
−
¯
k
2
)
M({−p
i
}, {
¯
k
j
})
2
4
(
¯
k
1
·
¯
k
2
)
2
−m
2
1
m
2
2
S
n
¸
i=1
d
3
p
i
2p
0
( p
i
)(2π)
3
. (10.12)
The parameter S is the inverse of the permutation factor for identical particles in the ﬁnal
state, as a detector will not be able to distinguish them. This will avoid double counting
when performing the phase space integrals. If there are n
r
identical particles of sort r in the
ﬁnal state, S =
¸
r
1/n
r
! (in the present case S = 1/n!).
Typical electromagnetic cross sections, as we will compute later, are of the order of
nanobarns (1 nb= 10
−33
cm
2
). With a luminosity of 10
33
cm
−2
s
−1
, approximately one collision
event per second will take place. In the weak interaction the cross sections are typically 5
to 6 orders of magnitude smaller, such that not more than one event per day will take place
in that case.
44
11 Decay rates
The deﬁnition of the decay rate (also called decay width) of an unstable particle is best
deﬁned by considering its selfenergy,
Σ(p) =
1
Z
`
, (11.1)
where the diagram for the 1PI twopoint function is now to be evaluated using the Feynman
rules in table 3 (pg. 42). The relation with the selfenergy follows from the fact that, apart
from the overall factor i/(2π)
4
, one has for each of the two external lines an extra factor
−i(2π)
2
√
Z as compared to the amputated 1PI 2point function, in total one therefore has
−iZ times the amputated 2point function. The latter indeed equals iΣ(p), see eq. (9.9).
We will now consider a simple example of a scalar ﬁeld theory with two types of ﬁelds, a
ﬁeld ϕ(x) associated with a light particle (mass m) and a ﬁeld σ(x) associated with a heavy
particle (mass M > 2m), which can decay in the lighter particles if we allow for a coupling
between one σ and two ϕ ﬁelds,
V (σ, ϕ) =
1
2
gσϕ
2
,
>
>
ϕ
ϕ
σ
≡ g , (11.2)
For the σ twopoint function in lowest order we ﬁnd
`
p p
=
`
p p
p−k
k
+
`
p p
+· · · . (11.3)
If σ is a stable particle (i.e. M < 2m), the loop in the ﬁrst diagram corresponds to virtual
ϕ particles moving between the vertices, since always k
2
= m
2
and (k −p)
2
= m
2
. However,
as soon as M > 2m, the loop integral will contain contributions where the ϕ particles can
be on the massshell and behave as a real particles, e.g. k
2
= m
2
and (k − p)
2
= m
2
in the
ﬁrst diagram. The real ϕ particles can escape to inﬁnity, thereby describing the decay of the
σ particle. Its number will reduce as a function of time.
Indeed we will see that only if M > 2m, the selfenergy will be able to develop a nonzero
imaginary part, γ ≡ Im(−Z
σ
Σ
σ
(p))
p
2
=M
2
= 0. On the massshell the σ propagator is in that
case modiﬁed for t > 0 to
d
4
p
(2π)
4
Z
σ
e
−ipx
p
2
−M
2
+ iγ + iε
= −
i
2
d
3
p
(2π)
3
Z
σ
e
i p·x
√
p
2
+ M
2
−iγ
e
−it
√
p
2
+M
2
−iγ
. (11.4)
The poles p
0
= ±
√
p
2
+ M
2
−iγ are now complex and for γ ≪ M
2
one has in a good
approximation
e
−it
√
p
2
+M
2
−iγ
= e
−it
√
p
2
+M
2
e
−
1
2
γt/
√
p
2
+M
2
. (11.5)
The amplitude of the wave function for the σ particle consequently decays with a decay rate
of Γ(p) = γ/
√
p
2
+ M
2
, the lifetime of this particle is hence τ(p) = 1/Γ(p).
We will evaluate the imaginary part of selfenergy for the σ 2point function in eq. (11.3)
ﬁrst to lowest order in g
Im(−Σ
σ
(p)) = Im
−
ig
2
2
d
4
k
(2π)
4
1
(k
2
−m
2
+ iε)((k −p)
2
−m
2
+ iε)
= Im
−
ig
2
2
d
4
x G
2
(x)e
ipx
, (11.6)
45
where we used the deﬁnition of the Green’s function, eq. (4.7). With the help of eq. (5.12)
we can write
G(x) = −i
d
3
k
(2π)
3
e
i
k·x
e
−ik
0
(
k)t
2k
0
(
k)
, (11.7)
which yields (after changing x to −x at the right place)
Im(−Σ
σ
(p)) = −
1
4
g
2
d
4
x (G(x)
2
+ G
∗
(−x)
2
)e
ipx
= (11.8)
1
4
g
2
(2π)
4
d
3
k
1
2k
(1)
0
(2π)
3
d
3
k
2
2k
(2)
0
(2π)
3
δ
3
(
k
1
+
k
2
−p)
δ(k
(1)
0
+ k
(2)
0
−p
0
) + δ(k
(1)
0
+ k
(2)
0
+ p
0
)
,
where we have implicitly deﬁned k
(j)
0
≡
k
2
j
+ m
2
. As we wish to study the σ 2point
function at t > 0 near the massshell, we can put p
0
≡
√
p
2
+ m
2
as well. In particular,
restricting ourselves to p
0
> 0, gives
Γ(p) ≡
Im(−Z
σ
Σ
σ
(p))
p
0
=
g
2
(2π)
4
4p
0
d
3
k
1
2k
(1)
0
(2π)
3
d
3
k
2
2k
(2)
0
(2π)
3
δ
4
(k
1
+ k
2
−p) , (11.9)
which is the result to lowest nontrivial order in g (to this order we can take Z
σ
= 1).
To any order we can, however, decompose the part of the σ selfenergy, with two ϕ
particles as an intermediate state, in its 1PI components as follows
`
p p
=
`
`
p p
`
`
(11.10)
where the full ϕ 2point function is given by 1/(p
2
−m
2
−Σ
ϕ
(p)+iε), which on the massshell
reduces to Z
ϕ
/(p
2
− ˜ m
2
+iε), the only part that actually contributes to Im(−Z
σ
Σ
σ
(p)). The
1PI ϕϕσ 3point function can easily be seen to be equal to M
ϕϕσ
({−k
i
}, p)/(Z
ϕ
√
Z
σ
) (or
its complex conjugate if in and outgoing lines are interchanged), since in lowest order it
should coincide with the ϕϕσ 3point vertex. In this way we easily ﬁnd the partial decay
rate dΓ(p) to be
dΓ(p) = (2π)
4
δ
4
(Σ
i
k
i
−p)
M
ϕϕσ
({−k
i
}, p)
2
2p
0
S
¸
i
d
3
k
i
2k
(i)
0
(2π)
3
, (11.11)
which, as it should be, is always positive. The symmetry factor S is the same as for the cross
section in eq. (10.12). The total decay rate is found by integrating over the phase space of
the outgoing particles Γ(p) ≡
dΓ(p). The large resemblance with the formula for the cross
section is no coincidence, as in both cases we have to calculate the probability for something
to happen (respectively a decay or a scattering). In its present form the formula for Γ(p) is
also valid for the decay of a particle in n other particles. The derivation is almost identical,
e.g. in eq. (11.8) one now encounters G
n
(x) instead of G
2
(x) and g now stands of course
for the coupling constant of n ϕ ﬁelds to the σ ﬁeld. It is not necessary for this coupling to
occur in the Lagrangian; at higher orders one can generate it from the lower couplings that
do occur in the Lagrangian.
46
12 The Dirac equation
To obtain a Lorentz invariant Schr¨odinger equation, we considered the square root of the
KleinGordon equation. This had the disadvantage that the Hamiltonian H =
√
p
2
+ m
2
contains an inﬁnite number of powers of p
2
/m
2
, the parameter in which the square root
should be expanded. It would have been better to treat space and time on a more equal
footing in the Schr¨odinger equation. This is what Dirac took as his starting point. As the
Schr¨odinger equation is linear in p
0
= i∂/∂t, one is looking for a Hamiltonian that is linear
in the momenta p
j
= i∂/∂x
j
(= −p
j
).
i
∂Ψ
∂t
= HΨ = −iα
k
∂Ψ
∂x
k
+ βmΨ . (12.1)
The question Dirac posed himself was to ﬁnd the simplest choice for α
k
and β, such that the
square of the Schr¨odinger equation gives the KleinGordon equation
p
2
0
= (−p
k
α
k
+ βm)
2
= p
2
+ m
2
. (12.2)
Dirac noted that only in case we allow α
k
and β to be noncommuting objects (i.e. matrices),
one can satisfy these equations. The above equation is equivalent to
β
2
= 1 ,
1
2
(α
j
α
k
+ α
k
α
j
) = 1δ
jk
and α
j
β + βα
j
= 0 . (12.3)
Historically, Dirac ﬁrst considered m = 0, but the massless case (m = 0) is somewhat
simpler, as it allows one to use β = 0 and α
k
= σ
k
for a solution of eq. (12.3). Here σ
k
are
the Paulimatrices, familiar from describing spin onehalf particles.
σ
1
=
0 1
1 0
, σ
2
=
0 −i
i 0
, σ
3
=
1 0
0 −1
. (12.4)
It is clear that two will be the smallest matrix dimension for which one can solve the equation
1
2
(α
j
α
k
+ α
k
α
j
) ≡
1
2
{α
j
, α
k
} = 1δ
jk
. It is not hard to prove that in a two dimensional
representation all solutions to this equation are given by
α
j
= ±Uσ
j
U
−1
, (12.5)
where U is an arbitrary nonsingular complex 2 × 2 matrix. We should, however, require
that H (and hence α
j
) is hermitian. This narrows U down to a unitary matrix, since
α
j
α
†
j
≡ Uσ
j
U
−1
U
†
−1
σ
†
j
U
†
= U([U
†
Uσ
j
]
−1
σ
j
U
†
U)U
−1
= 1 , (12.6)
such that U
†
Uσ
j
= σ
j
U
†
U for each j. The only 2 ×2 matrix that commutes with all Pauli
matrices is a multiple of the identity, which proves that U is unitary (up to an irrelevant
overall complex factor, which does not aﬀect α
j
).
Since the Hamiltonian is now a 2 ×2 matrix, the wave function Ψ(x) becomes a complex
two dimensional vector, also called a spinor, which describes particles with spin ¯ h/2
p
0
Ψ(x) = ∓ p · σΨ(x) . (12.7)
We have to demonstrate that the Dirac equation is covariant under Lorentz transformations.
We ﬁrst put the boosts to zero, because we already know from quantum mechanics how a
spinor transforms under rotations
p →p
′
= exp(ω ·
L) p , Ψ(x) →Ψ
′
(x
′
) = exp(
i
2
ω · σ)Ψ(x) . (12.8)
47
Here L
i
are real 3 ×3 matrices that generate the rotations in IR
3
L
i
jk
= ε
ijk
, (12.9)
such that [L
i
, L
j
] ≡ L
i
L
j
−L
j
L
i
= −ε
ijk
L
k
. These reﬂect the commutation relations of the
generators iσ
j
/2. We will later, in the context of nonAbelian gauge theories, show that this
describes the fact that SU(2) (the group of unitary transformations acting on the spinors)
is a representation of SO(3) (the group of rotations in IR
3
). To show the covariance of the
Dirac equation under rotations, i.e.
p
0
Ψ(x) = ∓ p · σΨ(x) →p
0
Ψ
′
(x
′
) = ∓ p
′
· σΨ
′
(x
′
) , (12.10)
we work out the Dirac equation in the rotated frame. Using eq. (12.8) we get
p
0
Ψ(x) = ∓ p
′
· exp(−
i
2
ω · σ)σ exp(
i
2
ω · σ)Ψ(x) , (12.11)
which should reduce to eq. (12.7). To prove this, we use the following general result for
matrices X and Y
e
X
Y e
−X
= exp(adX)(Y ) , adX(Y ) ≡ [X, Y ] , (12.12)
which is derived from the fact that f
1
(t) = e
tX
Y e
−tX
and f
2
(t) = exp(ad(tX))Y satisfy the
same diﬀerential equation, df
i
(t)/dt = [X, f
i
(t)]. Since also f
1
(0) = f
2
(0) it follows that
f
1
(1) = f
2
(1), being the above equation. Applying this result to Y = σ
k
and X = −iω · σ/2,
using the fact that
ad(−
i
2
ω · σ)σ
k
= [−
i
2
ω · σ, σ
k
] = (ω ·
L)
kj
σ
j
, (12.13)
the r.h.s. of eq.(12.11) becomes ∓ p
′
· σ
′
Ψ(x) = ∓ p · σΨ(x), where σ
′
= exp(ω ·
L)σ.
The interpretation of this Schr¨odinger equation caused Dirac quite some trouble, as its
eigenvalues are ± p , and it is not bounded from below. In the scalar theory we could avoid
this by just considering the positive root of the KleinGordon equation. Only when we
required localisation of the wave function inside the lightcone, we were forced to consider
negative energy states. In the present case, restricting to one of the eigenstates would break
the rotational invariance of the theory. For the massive case Dirac ﬁrst incorrectly thought
that the positive energy states describe the electron and the negative energy states the
proton. At that time antiparticles were unknown. Antiparticles were predicted by Dirac
because the only way he could make the theory consistent was to invoke the Pauli principle
and to ﬁll all the negative energy states. A hole in this sea of negative energy states, the
socalled Dirac sea, then corresponds to a state of positive energy. These holes describe
the antiparticle with the same mass as the particle. Obviously particle number will no
longer be conserved and also the Dirac equation will require “second quantisation” and the
introduction of a ﬁeld, which will be discussed later.
For the massive Dirac equation we need to ﬁnd a matrix β that anticommutes with all
α
i
. For 2 × 2 matrices this is impossible, since the Pauli matrices form a complete set of
anticommuting matrices. The smallest size turns out to be a 4 × 4 matrix. The following
representation is usually chosen
α
i
=
⊘ σ
i
σ
i
⊘
, β =
1
2
⊘
⊘ −1
2
, (12.14)
48
for which the nonrelativistic limit has a simple form. For the massless case it is often more
convenient to use the socalled Weyl representation
˜ α
i
=
σ
i
⊘
⊘ −σ
i
,
˜
β =
⊘ −1
2
−1
2
⊘
. (12.15)
We leave it as an exercise for the student to show that these two representations are related
by a 4 ×4 unitary transformation U, i.e. ˜ γ
µ
= Uγ
µ
U
−1
.
To study the covariance of the Dirac equation
p
0
Ψ(x) = (−α
i
p
i
+ βm)Ψ(x) (12.16)
under Lorentz transformations (note that now Ψ(x) has four complex components), it will
be proﬁtable to introduce a “fourvector” γ
µ
of 4 ×4 matrices
γ
µ
≡ (γ
0
, γ
i
) = (β, βα
i
) , (12.17)
such that the Dirac equation becomes
(−iγ
µ
∂
µ
+ m)Ψ(x) = (−γ
µ
p
µ
+ m)Ψ(x) ≡ (−p / + m)Ψ(x) = 0 . (12.18)
The Dirac gamma matrices satisfy anticommuting relations
{γ
µ
, γ
ν
} ≡ γ
µ
γ
ν
+ γ
ν
γ
µ
= 2g
µν
. (12.19)
As for the covariance under rotations, this equation is covariant under Lorentz boosts if there
exists a nonsingular complex 4 ×4 matrix S, such that Ψ(x) →Ψ
′
(x
′
) = SΨ(x) and
S
−1
γ
µ
S = K
µ
ν
γ
ν
, (12.20)
where K
µ
ν
is the Lorentz transformation acting on the momenta as p
′
µ
= K
µ
ν
p
ν
and on the
coordinates as x
µ′
= K
µ
ν
x
ν
. Like for the rotations, K can be written as an exponent
K = exp(ω) . (12.21)
Here ω
µ
ν
is a 4×4 matrix, which is antisymmetric when one of its indices is raised or lowered
by the metric, ω
µν
= g
µλ
ω
λ
ν
= −ω
νµ
.
We will now prove that
S = S(ω) = exp(−
i
4
ω
µν
σ
µν
) , σ
µν
≡
i
2
[γ
µ
, γ
ν
] (12.22)
satisﬁes eq. (12.20). Using the antisymmetry of ω
µν
and eq. (12.19) we ﬁnd
[
i
4
ω
µν
σ
µν
, γ
λ
] = −
1
4
ω
µν
[γ
µ
γ
ν
, γ
λ
]
= −
1
4
ω
µν
(γ
µ
γ
ν
γ
λ
−γ
λ
γ
µ
γ
ν
)
=
1
4
ω
µν
(γ
λ
γ
µ
γ
ν
+ γ
µ
γ
λ
γ
ν
−2γ
µ
g
λν
)
=
1
2
ω
µν
(g
λµ
γ
ν
−γ
µ
g
λν
) = ω
λ
µ
γ
µ
. (12.23)
Applying eq. (12.12) gives the proof for eq. (12.20). One says that S(ω) is a representation
of K(ω). Note that in general S(ω) is not a unitary transformation. This is because the
49
boosts form a noncompact part of the Lorentz group. There is, however, a relation between
S
†
and S
−1
,
γ
0
S
†
γ
0
= S
−1
, (12.24)
which is most easily proven in the Weyl representation, since ˜ γ
0
=
˜
β commutes with ˜ σ
ij
but anticommutes with ˜ σ
k0
(as always roman indices run from 1 to 3 and greek indices run
from 0 to 3), whereas ˜ σ
ij
is hermitian and ˜ σ
k0
is antihermitian, as follows from the explicit
expressions obtained from eqs. (12.15), (12.17) and (12.22)
˜ σ
ij
= ˜ σ
ij
=
i
2
[˜ γ
i
, ˜ γ
j
] = ε
ijk
σ
k
⊘
⊘ σ
k
, ˜ σ
0k
= ˜ σ
k0
=
i
2
[˜ γ
k
, ˜ γ
0
] = −i
σ
k
⊘
⊘ −σ
k
.
(12.25)
Note that eqs. (12.20) and (12.22) are independent of the representation in which we give
the gamma matrices, as any two such representations have to be related by a unitary trans
formation. In the Weyl representation
˜
S(ω) is block diagonal, like the Dirac equation for
m = 0 (as ˜ α
i
is block diagonal). The upper block corresponds to eq. (12.7) with the plus
sign and the lower block corresponds to the minus sign. We can verify eq. (12.8) by using
the fact that the Lorentz transformations contain the rotations through the identiﬁcation
ω
k
= −
1
2
ε
ijk
ω
ij
. With ω
0k
= 0, one ﬁnds
˜
S(ω) = 1
2
⊗exp(iω · σ/2), i.e. it acts on each 2 ×2
block by the same unitary transformation.
The boost parameters are described by ω
0i
. For a boost in the x direction we have that
χ = ω
01
is related to the boost velocity by v
1
= −tanh(χ). For K we ﬁnd in this case
K =
¸
¸
¸
¸
cosh(χ) sinh(χ) 0 0
sinh(χ) cosh(χ) 0 0
0 0 1 0
0 0 0 1
¸
. (12.26)
In the Weyl representation,
˜
S splits again in two blocks, but one is the inverse of the other
(and neither is unitary). To be precise,
˜
S restricted to the upperleft 2 × 2 block equals
exp(
1
2
ω
0k
σ
k
), whereas for the lowerright 2 ×2 block we ﬁnd exp(−
1
2
ω
0k
σ
k
).
As S is not unitary Ψ
†
(x)Ψ(x) is no longer invariant under Lorentz transformations.
But we claim it is nevertheless a probability density, namely the time component j
0
(x) of a
conserved current
j
µ
(x) = Ψ
†
(x)γ
0
γ
µ
Ψ(x) . (12.27)
We leave it as an exercise to show that the Dirac equation implies that the current is
conserved, ∂
µ
j
µ
(x) = 0. The combination Ψ
†
(x)γ
0
will occur so often, that it has acquired
its own symbol
Ψ(x) ≡ Ψ
†
(x)γ
0
. (12.28)
It transforms under a Lorentz transformation as
Ψ(x) →SΨ(x) and Ψ(x) →Ψ(x)S
−1
. (12.29)
We can use this to build the required Lorentz scalars, vectors and tensors
scalar: Ψ
′
(x
′
)Ψ
′
(x
′
) = Ψ(x)S
−1
SΨ(x) = Ψ(x)Ψ(x)
vector: Ψ
′
(x
′
)γ
µ
Ψ
′
(x
′
) = Ψ(x)S
−1
γ
µ
SΨ(x) = K
µ
ν
Ψ(x)γ
ν
Ψ(x)
tensor: Ψ
′
(x
′
)σ
µν
Ψ
′
(x
′
) = Ψ(x)S
−1
σ
µν
SΨ(x) = K
µ
λ
K
ν
κ
Ψ(x)σ
λκ
Ψ(x)
50
The Lagrangian is a Lorentz scalar, which we chose such that its equations of motion repro
duce the Dirac equation. As Ψ(x) is complex, it can be considered independent of Ψ(x) and
the following Lagrangian
S
Dirac
=
d
4
x L
Dirac
=
d
4
x Ψ(x)(iγ
µ
∂
µ
−m)Ψ(x) . (12.30)
gives the EulerLagrange equations
δ
−
S
δ
−
Ψ(x)
= (iγ
µ
∂
µ
−m)Ψ(x) = 0 ,
δ
−
S
δ
−
Ψ(x)
= Ψ(x)(−iγ
µ
←
∂
µ
−m) = 0 . (12.31)
The second equation is the complex conjugate of the ﬁrst, Ψ
†
(x)(i(γ
µ
)
†
←
∂
µ
+m) = 0, because
the gamma matrices satisfy
(γ
µ
)
†
= γ
0
γ
µ
γ
0
, (12.32)
which follows from the fact that (γ
0
)
†
= β
†
= β = γ
0
and (γ
i
)
†
= (βα
i
)
†
= α
i
β = −γ
i
, or
from the explicit representation of the gamma matrices
γ
0
=
1
2
⊘
⊘ −1
2
, γ
i
=
⊘ σ
i
−σ
i
⊘
. (12.33)
Hence
0 = Ψ
†
(x)(i(γ
µ
)
†
←
∂
µ
+m)γ
0
= Ψ(iγ
0
(γ
µ
)
†
γ
0
←
∂
µ
+m) = Ψ(iγ
µ
←
∂
µ
+m) . (12.34)
An important role will be played by a ﬁfth gamma matrix
γ
5
≡ iγ
0
γ
1
γ
2
γ
3
=
⊘ 1
2
1
2
⊘
, (12.35)
which anticommutes with all γ
µ
(see problem 21)
γ
5
γ
µ
= −γ
µ
γ
5
and (γ
5
)
2
= 1
4
. (12.36)
This implies that we can introduce projection operators
P
±
≡
1
2
(1 ±γ
5
) , (12.37)
which satisfy (P
±
)
2
= P
±
and P
±
P
∓
= 0. Their role is best described in the Weyl represen
tation, where
˜ γ
5
= i˜ γ
0
˜ γ
1
˜ γ
2
˜ γ
3
= −i˜ α
1
˜ α
2
˜ α
3
=
1
2
⊘
⊘ −1
2
, (12.38)
such that
˜
P
+
=
1
2
(1 + ˜ γ
5
) projects on the two upper and
˜
P
−
=
1
2
(1 − ˜ γ
5
) on the two lower
components of the fourspinors. In the massless case these two components are decoupled,
p
0
Ψ(x) = (− p ·
˜
α + m
˜
β)Ψ(x) =
− p · σ −m1
2
−m1
2
p · σ
Ψ . (12.39)
Hence, for m = 0 we have
Ψ(x) =
Ψ
+
(x)
Ψ
−
(x)
, p
0
Ψ
±
(x) = ∓ p · σΨ
±
(x) . (12.40)
51
which is identical to eq. (12.7). The eigenstates of the projection operators P
±
are called
helicity eigenstates. As long as m = 0, helicity is not conserved. But as we saw, for m = 0
the two helicity eigenstates decouple. One can deﬁne in that case consistently a particle
with a ﬁxed helicity, whose opposite helicity state does not occur (although its antiparticle
has opposite helicity). A very important example of such a particle is the neutrino, although
experiment has not yet been able to rule out a (tiny) mass for this particle (m
νe
< 10eV ).
See problem 22 for more details.
Apart from the invariance of the Dirac equation under Lorentz transformations and
translations (which are obvious symmetries of L
Dirac
) we also often want invariance under
parity (x →−x) and timereversal (t →−t). One easily checks that
PΨ(x) ≡ Ψ
′
(t, −x) = γ
0
Ψ(t, x) and TΨ(x) ≡ Ψ
′
(−t, x) = γ
5
γ
0
γ
2
Ψ
∗
(t, x) (12.41)
satisfy the Dirac equation, where P stands for parity and T for timereversal. This implies
that the Lorentz covariant combinations Ψ(x)γ
5
Ψ(x) and Ψ(x)γ
5
γ
µ
Ψ(x) are not invariant
under parity and timereversal. They are called pseudoscalars and pseudovectors. These
combinations play an important role in the weak interactions, where parity is not a symmetry.
A third discrete symmetry, charge conjugation C, will be discussed in section 17.
13 Plane wave solutions of the Dirac equation
As usual, the Dirac equation can be solved by Fourier decomposition in plane waves,
Ψ(x) =
d
3
k
2k
0
(
k)(2π)
3
˜
Ψ(
k)e
−ikx
. (13.1)
where
˜
Ψ(
k) are complex fourvectors that satisfy
(k/ −m)
˜
Ψ(
k) ≡ (k
µ
γ
µ
−m)
˜
Ψ(
k) = 0 , k
2
0
=
k
2
+ m
2
. (13.2)
The Lorentz invariance implies
˜
Ψ
′
(
k
′
) = S(ω)
˜
Ψ(
k) , k
′
µ
= K(ω)
µ
ν
k
ν
. (13.3)
As any
k can be obtained from a boost to
k =
0, all solutions of the Dirac equation can be
obtained from the ones at rest with
k =
0 (see problem 19)
(γ
0
k
0
−m)
˜
Ψ(
0) =
(k
0
−m)1
2
⊘
⊘ −(k
0
+ m)1
2
˜
Ψ(
0) . (13.4)
We see that for k
0
> 0 (k
0
= m), there are two independent solutions both of the form
˜
Ψ
+
(
0) =
Ψ
A
0
, (13.5)
where Ψ
A
is a spin onehalf twospinor of which
1
0
is the spinup and
0
1
is the spin
down state. The identiﬁcation of the spin degrees of freedom follows from the behaviour of
52
˜
Ψ under rotations (which leave
k =
0). We leave it as an exercise to verify that also in the
Dirac representation of the gamma matrices, like in the Weyl representation (see eq. (12.25)),
σ
ij
= ε
ijk
σ
k
⊘
⊘ σ
k
, (13.6)
such that Ψ
A
is easily seen to transform under a rotation as in eq. (12.8) (cmp. the discussion
below eq. (12.25)). There are also two solutions for k
0
< 0 (k
0
= −m) of the form
˜
Ψ
−
(
0) =
0
Ψ
B
, (13.7)
where likewise Ψ
B
is a spin onehalf twospinor of which
1
0
is the spinup and
0
1
is the
spindown state, which transform under rotations as Ψ
A
.
For any frame, i.e. for any value of
k, we will deﬁne the four independent solutions of
the Dirac equation as
k
0
=
k
2
+ m
2
: u
(α)
(
k) =
k/ + m
m+k
0

u
(α)
0
, u
(1)
0
=
¸
¸
¸
¸
1
0
0
0
¸
, u
(2)
0
=
¸
¸
¸
¸
0
1
0
0
¸
,
k
0
= −
k
2
+ m
2
: v
(α)
(−
k) =
k/ + m
m+k
0

v
(α)
0
, v
(1)
0
=
¸
¸
¸
¸
0
0
1
0
¸
, v
(2)
0
=
¸
¸
¸
¸
0
0
0
1
¸
. (13.8)
These solutions naturally split in positive energy (u
(α)
(
k) with α = 1, 2) and negative energy
(v
(α)
(−
k) with α = 1, 2) solutions; for
k =
0 and m = 0 easily seen to be proportional
to the solutions
˜
Ψ
±
(
0), to be precise u
(α)
(
0) = u
(α)
0
/
√
2m and v
(α)
(
0) = v
(α)
0
/
√
2m. The
normalisation we have chosen allows us to treat massless fermions at the same footing. In
that case we can, however, not transform to the restframe. This normalisation also implies
that
u
(α)
(
k)u
(β)
(
k) = −v
(α)
(−
k)v
(β)
(−
k) = 2mδ
αβ
. (13.9)
For example
u
(α)
(
k)u
(β)
(
k) =
u
(α)
0
γ
0
(k/ + m)
†
γ
0
(k/ + m)u
(β)
0
(m+ k
0
)
=
u
(α)
0
(k/ + m)
2
u
(β)
0
(m+ k
0
)
=
u
(α)
0
(k
2
+ m
2
+ 2k/m)u
(β)
0
(m + k
0
)
=
u
(α)
0
(k
2
+ m
2
+ 2k
0
m)u
(β)
0
(m + k
0
)
= 2mδ
αβ
, (13.10)
where we used k
2
= m
2
and u
(α)
0
γ
µ
u
(β)
0
= δ
αβ
δ
µ0
(see eq. (12.33)). The computation for
v
(α)
(
k) is left as an exercise. The fact that we ﬁnd a result that is independent of
k is
consistent with our claim that these spinors can also be obtained by applying the appropriate
boost to
k =
0, since
˜
Ψ(
k)
˜
Ψ(
k) is a Lorentz scalar.
That these spinors indeed satisfy eq. (13.2) follows from the fact that
(k/ −m)(k/ + m) = k
2
−m
2
= 0 . (13.11)
53
Note that v
(α)
(
k) can also be viewed as a positive energy solution for the complex conjugate
of the Dirac equation
k
0
=
k
2
+ m
2
: (k/ + m)v
(α)
(
k) = 0 . (13.12)
It will play the role of the wave function for the antiparticles (the holes) in the Dirac ﬁeld
theory. To see that our plane waves have the correct amplitude, we use the fact that the
probability density can be deﬁned in terms of the zerocomponent ρ(x) of the conserved
fourvector j
µ
(x) = Ψ(x)γ
µ
Ψ(x) (see eq. (12.27)), or
ρ(x) = Ψ
†
(x)Ψ(x) . (13.13)
Indeed,
˜
Ψ
†
(
k)
˜
Ψ(
k) transforms as a density, i.e. as the energy k
0
, and we ﬁnd
u
(α)
(
k)
†
u
(β)
(
k) = v
(α)
(
k)
†
v
(β)
(
k) = 2k
0
δ
αβ
, u
(α)
(
k)
†
v
(β)
(
k) = 0 . (13.14)
This can be veriﬁed by a direct computation, e.g. (k
0
> 0)
v
(α)
(
k)
†
v
(β)
(
k) =
v
(α)
0
†
(m−k/)
†
(m−k/)v
(β)
0
m + k
0
=
v
(α)
0
†
(k
2
0
+
k
2
+ m
2
−2mk
0
γ
0
)v
(β)
0
m + k
0
=
=
v
(α)
0
†
(k
2
0
+
k
2
+ m
2
+ 2mk
0
)v
(β)
0
m + k
0
= 2k
0
δ
αβ
, (13.15)
using v
(α)
0
†
γ
µ
v
(β)
0
= −δ
µ0
δ
αβ
(see eq. (12.33)). We leave the other identities as an exercise.
This implies that the plane wave solutions (k
0
= (
k
2
+ m
2
)
1
2
)
u
(α)
(
k)
2k
0
(2π)
3
e
−ikx
,
v
(α)
(
k)
2k
0
(2π)
3
e
ikx
, α = 1, 2 , (13.16)
are normalised to one (of course, in a ﬁnite volume we replace (2π)
3
by V and
d
3
k by
¸
k
).
As for the scalar ﬁeld we can introduce a Dirac ﬁeld
Ψ(x) =
d
3
k
2k
0
(
k)(2π)
3
2
¸
α=1
b
α
(
k)u
(α)
(
k)e
−ikx
+ d
†
α
(
k)v
(α)
(
k)e
ikx
. (13.17)
Since Ψ(x) is complex, there is no relation between d
α
and b
α
. In the quantum ﬁeld theory
these will play the role of the annihilation operators for the (anti)particles.
Finally we note that, for m = 0, we can deﬁne projection operators (which are 4 × 4
matrices)
Λ
+
(
k) =
2
¸
α=1
u
(α)
(
k) ⊗u
(α)
(
k)
2m
=
k/ + m
2m
, Λ
−
(
k) = −
2
¸
α=1
v
(α)
(
k) ⊗v
(α)
(
k)
2m
=
−k/ + m
2m
,
(13.18)
which can be veriﬁed from the explicit form of u and v, or by ﬁrst computing them in the
restframe where Λ
±
(
0) =
1
2
(1 ±γ
0
). In that case we ﬁnd
Λ
±
(
k) =
(±k/ + m)(1 ±γ
0
)(±k/ + m)
4m(m + k
0
)
=
±k/ + m
2m
, (13.19)
whose proof requires some gamma matrix gymnastics. Independent of the frame Λ
±
satisfy
Λ
2
±
(
k) = Λ
±
(
k) , Λ
∓
(
k)Λ
±
(
k) = 0 . (13.20)
Note that Tr(Λ
±
(
k)) = 2, a result that can also be derived from eq. (13.9).
54
14 The Dirac Hamiltonian
In the Dirac equation we encountered an additional diﬃculty, namely that the negative en
ergy solutions even arise at the level of the classical theory. In ﬁeld theory the negative
energy solutions had an interpretation in terms of antiparticles, and the ﬁeld theory Hamil
tonian was still positive, and most importantly, bounded from below (see section 2 and
problem 5). The ﬁeld theory Hamiltonian for the Dirac ﬁeld no longer has this property.
The Hamiltonian can again be derived through a Legendre transform of the Lagrangian
S =
d
4
x L =
d
4
x Ψ(x)(iγ
µ
∂
µ
−m)Ψ(x) . (14.1)
The canonical momentum is hence
π
a
(x) =
δS
δ
˙
Ψ
a
(x)
= (Ψ(x)iγ
0
)
a
= iΨ
∗
a
(x) . (14.2)
such that
H =
d
3
x (π
a
(x)
˙
Ψ
a
(x) −L) =
d
3
x Ψ(x)(−iγ
j
∂
j
+ m)Ψ(x)
=
d
3
x Ψ
†
(x)(−iα
j
∂
j
+ mβ)Ψ(x)
=
d
3
k k
0
(
k)
¸
α
b
†
α
(
k)b
α
(
k) −d
α
(
k)d
†
α
(
k)
. (14.3)
Note the resemblance with eq. (12.1) for the middle term. We used eq. (13.17) for the
expansion of the Dirac ﬁeld in plane waves. From this result it is clear that the Hamiltonian
is not bounded from below, and this would make the vacuum unstable, as the negative
energy states, described by the d
α
(
k), can lower the energy by an arbitrary amount. It is
wellknown how Dirac repaired this problem. He postulated that all negative energy states
are occupied, and that the states satisfy the Pauli principle, i.e. two particles can not occupy
the same quantum state (it is only in that case that we can make sense of what is meant with
ﬁlling all negative energy states). This implies that one should use anticommuting relations
for the creation and annihilation operators
{b
α
(
k), b
β
( p)} = 0 , {b
α
(
k), b
†
β
( p)} = δ
αβ
δ
3
(
k −p) ,
{d
α
(
k), d
β
( p)} = 0 , {d
α
(
k), d
†
β
( p)} = δ
αβ
δ
3
(
k − p) ,
{d
α
(
k), b
β
( p)} = 0 , {d
α
(
k), b
†
β
( p)} = 0 . (14.4)
Indeed, if we deﬁne a two particle state as 
k, p >≡ b
†
(
k)b
†
( p)0 > (suppressing the spinor
indices), the anticommutation relations imply that 
k, p >= − p,
k >.
A hole in the Dirac sea is by deﬁnition the state that is obtained by annihilating a
negative energy state in the Dirac sea. As annihilation lowers the total energy by the energy
of the annihilated state, which in this case is negative, the net energy is raised. The wave
function for the negative energy state is given by exp(ikx)v
(α)
(
k)/
√
k
0
and has momentum−
k
(k
0
≡ (
k
2
+m
2
)
1
2
), see eq. (13.8). The reason to associate its Fourier coeﬃcient in eq. (13.17)
with a creation operator d
†
α
(
k), is that conservation of energy and momentum implies that it
creates an antiparticle as a hole in the Dirac sea, with momentum
k and helicity
1
2
for α = 2,
whereas for α = 1 the helicity is −
1
2
(hence the helicity and momentum are opposite to the
55
negative energy state it annihilates). The wave function of an antiparticle with momentum
k is hence given by exp(−ikx)v
(α)
(
k)
†
/
√
k
0
. If we now use the anticommutation relations of
creation and annihilation operators (as a consequence of the Pauli exclusion principle), we
see that with the present interpretation the Dirac Hamiltonian is bounded from below
H =
d
3
k k
0
(
k)
¸
α
b
†
α
(
k)b
α
(
k) + d
†
α
(
k)d
α
(
k) −1
. (14.5)
As in the case for scalar ﬁeld theory we can normalise the energy of the vacuum state to zero
by adding a (inﬁnite) constant. Note that this constant has its sign opposite to the scalar
case (and is in magnitude four times as large). In socalled supersymmetric ﬁeld theories this
is no longer an accident as the Dirac ﬁelds will be related to the scalar ﬁelds by a symmetry,
which is however outside the scope of these lectures.
Important is also to note that the anticommuting relations are crucial to guarantee
locality of the Dirac ﬁeld. In this case Dirac ﬁelds speciﬁed in diﬀerent regions of spacetime
that are spacelike separated should anticommute. And indeed, in problem 24 you are asked
to prove that
{Ψ
a
(x), Ψ
†
b
(x
′
)} = 0 for (x −x
′
)
2
< 0 . (14.6)
Also, as in the scalar theory, we can couple a source to the free Dirac ﬁeld (we again
introduce ¯ ε as the expansion parameter for taking the interactions due to the source into
account in Hamiltonian perturbation theory)
L(x) = Ψ(x)(iγ
µ
∂
µ
−m)Ψ(x) − ¯ ε
¯
J(x)Ψ(x) − ¯ εΨ(x)J(x) ,
H(x) = Ψ(x)(−iγ
k
∂
k
+ m)Ψ(x) + ¯ ε
¯
J(x)Ψ(x) + ¯ εΨ(x)J(x) , (14.7)
where as before the Hamiltonian H is the spatial integral over the Hamiltonian density H(x),
H =
d
3
x H(x). Note that the sources J(x) and
¯
J(x) are independent, as for a complex
scalar ﬁeld, see problem 17. In problem 24 you are asked to prove that in Hamiltonian
perturbation theory one obtains
< 0Texp(−i
T
0
H(t)dt) 0 > e
iE
0
T
= 1 −i¯ ε
2
d
4
xd
4
y
¯
J(x)G
F
(x −y)J(y) +O(¯ ε
3
)
= 1 −i
× × >
¯ εJ
¯ ε
¯
J
+O(¯ ε
3
) . (14.8)
where
G
F
(x −y) =
d
4
p
(2π)
4
e
−ip(x−y)
p / −m + iε
=
d
4
p
(2π)
4
(p / + m)e
−ip(x−y)
p
2
−m
2
+ iε
. (14.9)
Hence, the Green’s function for fermions is in Fourier space given by G
F
(p) = (p /−m+iε)
−1
,
which is the inverse of the quadratic part of the Lagrangian, as for the scalar ﬁelds. In
problem 24 it will be evident that, nevertheless, the anticommuting properties of the Dirac
ﬁeld play a crucial role (cmp. section 5). It becomes, however, plausible there is also for the
fermions a path integral formulation, as splitting of a square (cmp. eqs. (7.11) and (7.12)) is
independent of the details of the path integrals. This will be the subject of the next section.
To conclude, we note that the coupling to an electromagnetic ﬁeld A
µ
(x) should be
achieved through the current j
µ
(x), deﬁned in eq. (12.27)
d
4
x j
µ
(x)A
µ
(x) =
d
4
x Ψ(x)γ
µ
A
µ
(x)Ψ(x) . (14.10)
56
Since this current is conserved, which can also be seen as a consequence of the Noether
theorem applied to the invariance of eq. (14.1) (the free Dirac Lagrangian) under global
phase transformations, the coupling is gauge invariant. Using the minimal coupling deﬁned
as in eq. (3.35),
D
µ
Ψ(x) ≡ (∂
µ
−ieA
µ
(x))Ψ(x) , (14.11)
one can ﬁx the normalisation of the electromagnetic current to be J
µ
(x) = −ej
µ
(x), since
(cmp. eq. (3.31))
Ψ(x)(iγ
µ
D
µ
−m)Ψ(x) = Ψ(x)(iγ
µ
∂
µ
−m)Ψ(x) + ej
µ
A
µ
(x) . (14.12)
The current j
µ
(x) is the charge density current, whose time component ρ(x) at the quantum
level is no longer positive deﬁnite. Using the anticommuting properties of eq. (14.4) one
ﬁnds
d
3
x ˆ ρ(x) =
d
3
x Ψ
†
(x)Ψ(x)
=
d
3
k
¸
α
b
†
α
(
k)b
α
(
k) + d
α
(
k)d
†
α
(
k)
=
d
3
k
¸
α
b
†
α
(
k)b
α
(
k) −d
†
α
(
k)d
α
(
k) + 1
= Q
0
/e +
d
3
k
¸
α
b
†
α
(
k)b
α
(
k) −d
†
α
(
k)d
α
(
k)
. (14.13)
The vacuum value of this operator is indicated by the (generally inﬁnite) constant Q
0
/e,
which can be normalised to zero. After all we want the state with all negative energy
states occupied to have zero charge. With the above normalisation of the electric current,
J
µ
(x) = −ej
µ
(x), we see that the b modes can be identiﬁed with the electrons with charge
−e and the d modes with their antiparticles, the positrons, with opposite electric charge
+e. To summarise, b
†
α
(
k) corresponds with the creation operator of a spinup (α = 1) or a
spindown (α = 2) electron of momentum
k, whereas d
†
α
(
k) corresponds with the creation
operator of a spinup (α = 2) or a spindown (α = 1) positron of momentum
k.
15 Path integrals for fermions
For scalar ﬁelds, which describe bosons, we used real or complex numbers (the eigenvalues
of the operators), in order to perform the path integral. For fermionic ﬁelds it is essential to
build the anticommuting properties into the path integral.
To this end we introduce a socalled Grassmann algebra, which exists of Grassmann
variables θ
i
that mutually anticommute
{θ
i
, θ
j
} ≡ θ
i
θ
j
+ θ
j
θ
i
= 0 . (15.1)
In particular, a Grassmann variable squares to zero
θ
2
= 0 (15.2)
A Grassmann variable can be multiplied by a complex number, with which it commutes. A
function of a single Grassmann variable has a ﬁnite Taylor series
f(θ) = a
0
+ a
1
θ , (15.3)
57
and spans a two dimensional (real or) complex vector space. This is exactly what we need
to describe a spin onehalf particle. Let us introduce the following notation
0 >=
1
0
, 1 >=
0
1
. (15.4)
With respect to the Hamiltonian
H
0
=
1
2
−W 0
0 W
(15.5)
0 > is the vacuum state (i.e. the state with lowest energy) and we interpret 1 > as the
oneparticle state (with energy W above the vacuum). An arbitrary spinor can be written
as a linear combination of these two states
Ψ >= a
0
0 > +a
1
1 >= a
0
0 > +a
1
b
†
0 > , (15.6)
where b is the fermionic annihilation operator, which in the spinor representation is given
by a 2 ×2 matrix
b =
0 1
0 0
, b
†
=
0 0
1 0
. (15.7)
We note that b
2
= (b
†
)
2
= 0, a property it has in common with a Grassmann variable. We
will now look for properties of θ such that
Ψ(θ) ≡< θΨ >= a
0
+ a
1
θ (15.8)
is a representation of the state Ψ >, similar to Ψ(x) =< xΨ > for the case of a single
bosonic (i.e. commuting, as opposed to anticommuting) degree of freedom. In the latter
case, the normalisation
dx Ψ
∗
(x)Ψ(x) = 1 (15.9)
is an important property we would like to impose here too (keeping in mind that for path
integrals we need to insert completeness relations). As < 0 = (1, 0) and < 1 = (0, 1), we
have
< Ψ =< 0a
∗
0
+ < 1a
∗
1
=< 1b
†
a
∗
0
+ < 1a
∗
1
. (15.10)
As in eq. (15.8) (i.e. b
†
→θ), we anticipate
Ψ
∗
(θ) ≡< Ψθ >= a
∗
0
θ + a
∗
1
. (15.11)
We wish to deﬁne integration over Grassmann variables, such that the normalisation of the
wave function is as usual
< ΨΨ >= a
0

2
+a
1

2
=
dθ Ψ
∗
(θ)Ψ(θ) . (15.12)
Since the norm should be a number, and as
dθ 1 is itself a Grassmann variable, the latter
should vanish. For the same reason
dθ θ (which is itself a commuting object, as is any
even product of Grassmann variables) can be seen as a number. Demanding the socalled
Grassmann integration to be linear in the integrand, and using θ
2
= 0, all possible ingredients
have been discussed. Indeed
dθ θ = 1 ,
dθ 1 = 0 (15.13)
58
is easily seen to give the desired result. Note that dθ is considered as an independent
Grassmann variable (which is important to realise when multiple Grassmann integrations
are involved).
We can now study the action of an operator (like a Hamiltonian H) on a state, which in
the spinor representation is given by 2 ×2 matrices.
MΨ >= Ψ
′
> or
M
11
M
12
M
21
M
22
a
0
a
1
=
a
′
0
a
′
1
. (15.14)
Translated to Grassmann variables, this gives
Ψ
′
(θ) = a
′
0
+ a
′
1
θ = M
11
a
0
+ M
12
a
1
+ (M
21
a
0
+ M
22
a
1
)θ ≡
dθ
′
M(θ, θ
′
)Ψ(θ
′
) , (15.15)
provided we deﬁne
M(θ, θ
′
) ≡ M
11
θ
′
+ M
12
+ M
21
θ
′
θ −M
22
θ = M
11
θ
′
+ M
12
−M
21
θθ
′
−M
22
θ . (15.16)
Indeed,
dθ
′
M(θ, θ
′
)Ψ(θ
′
) =
dθ
′
(M
11
θ
′
+ M
12
−M
21
θθ
′
−M
22
θ)(a
0
+ a
1
θ
′
)
=
dθ
′
{a
0
M
12
−a
0
M
22
θ + (a
1
M
12
+ a
0
M
11
)θ
′
−(a
0
M
21
+ a
1
M
22
)θθ
′
}
=
dθ
′
θ
′
{(a
1
M
12
+ a
0
M
11
) + (a
0
M
21
+ a
1
M
22
)θ}
= M
12
a
1
+ M
11
a
0
+ (M
21
a
0
+ M
22
a
1
)θ . (15.17)
The 2 ×2 identity matrix is hence represented by (note the sign)
1
2
(θ, θ
′
) = θ
′
−θ , (15.18)
which can be used to write the inﬁnitesimal evolution operator
exp(−iH∆t) = 1
2
−iH∆t +O(∆t
2
) , (15.19)
where H is a (possibly timedependent) 2 ×2 matrix. In the Grassmann representation this
reads
1
2
(θ, θ
′
) −iH(θ, θ
′
)∆t = θ
′
−θ −i∆t(H
11
θ
′
+ H
12
−H
21
θθ
′
−H
22
θ)
=
d
˜
θ exp(
˜
θ[1
2
(θ, θ
′
) −iH(θ, θ
′
)∆t]) . (15.20)
The last identity is exact, and a consequence of the fact that the Taylor series of any function
of a Grassmann variable truncates
exp(
˜
θx) = 1 +
˜
θx ,
d
˜
θ exp(
˜
θx) = x . (15.21)
This is valid both for x a Grassmann variable (in which case the ordering of x with respect
to
˜
θ is important, with the opposite ordering the result is −x), or for x a complex number.
Another useful property of Grassmann integration is that (y is a number)
dθ exp(θx + y) = xexp(y) . (15.22)
59
To prove this, we use that
exp(θx+y) =
∞
¸
n=0
1
n!
(θx+y)
n
=
∞
¸
n=0
1
n!
y
n
+θx
∞
¸
n=1
1
(n −1)!
y
n−1
= (1+θx) exp(y) . (15.23)
In general it is not true that the exponential function retains the property exp(x + y) =
exp(x) exp(y), for x and y arbitrary elements of the Grassmann algebra. It behaves as if x
and y are matrices, as it should since the Grassmann representation originates from a 2 ×2
matrix representation. More precisely
exp(θx) exp(θ
′
y) = (1 +θx)(1 +θ
′
y) = 1 + (θx + θ
′
y) +
1
2
(θx + θ
′
y)
2
+
1
2
[θx, θ
′
y]
= exp (θx + θ
′
y +
1
2
[θx, θ
′
y]) . (15.24)
This means that the CampbellBakerHausdorﬀ formula (eq. (6.44)) can be extended to this
case. It truncates after the single commutator term as neither θx nor θy can appear more
than once.
Let us apply this to the evolution operator, which in the Grassmann representation is
given by (see eqs. (15.19) and (15.20))
< θ
i+1
U(t
i+1
, t
i
)θ
i
>≡ U
i
(θ
i+1
, θ
i
) =
d
˜
θ
i
exp
˜
θ
i
[θ
i
−θ
i+1
−iH(t
i
; θ
i+1
, θ
i
)∆t]
+O(∆t
2
) ,
(15.25)
such that
dθ
i
< θ
i+1
U(t
i+1
, t
i
)θ
i
>< θ
i
Ψ >=
dθ
i
d
˜
θ
i
exp
˜
θ
i
[θ
i
−θ
i+1
−iH(t
i
; θ
i+1
, θ
i
)∆t]
Ψ(θ
i
) +O(∆t
2
) . (15.26)
which can be iterated, ﬁrst by one step, to
dθ
i
< θ
i+2
U(t
i+2
, t
i
)θ
i
>< θ
i
Ψ >≡
dθ
i+1
U
i+1
(θ
i+2
, θ
i+1
)
dθ
i
U
i
(θ
i+1
, θ
i
)Ψ(θ
i
)
=
dθ
i+1
d
˜
θ
i+1
exp
˜
θ
i+1
[θ
i+1
−θ
i+2
−iH(t
i+1
; θ
i+2
, θ
i+1
)∆t]
×
dθ
i
d
˜
θ
i
exp
˜
θ
i
[θ
i
−θ
i+1
−iH(t
i
; θ
i+1
, θ
i
)∆t]
Ψ(θ
i
) +O(∆t
2
) . (15.27)
Note that we have to be careful where we put the diﬀerentials, as they are Grassmann
variables themselves. If H is diagonal, as will often be the case for the application we have
in mind, exp
˜
θ[1
2
(θ, θ
′
) −iH(θ, θ
′
)∆t]
will be a commuting object (socalled Grassmann
even) and it does not matter if we put one of the diﬀerentials on one or the other side of
the exponential. The combination dθ
i
d
˜
θ
i
is likewise Grassmann even, and the pair can be
shifted to any place in the expression for the path integral. Hence, provided of course H is
diagonal, any change in the ordering can at most given an additional minus sign. We now
apply eq. (15.24) to the above product of exponentials,
exp(
˜
θ
i+1
[θ
i+1
−θ
i+2
−iH(t
i+1
; θ
i+2
, θ
i+1
)∆t]) exp
˜
θ
i
[θ
i
−θ
i+1
−iH(t
i
; θ
i+1
, θ
i
)∆t]
(15.28)
=exp
¸
i+1
¸
j=i
˜
θ
j
{θ
j
−θ
j+1
−iH(t
j
; θ
j+1
, θ
j
)∆t} −
1
2
∆t
2
[
˜
θ
i+1
H(t
i+1
; θ
i+2
, θ
i+1
),
˜
θ
i
H(t
i
; θ
i+1
, θ
i
)]
¸
60
and evaluate the commutator that appears in the exponent. With the explicit expression for
H (see eq. (15.20)) we ﬁnd
dθ
i+1
[
˜
θ
i+1
H(t
i+1
; θ
i+2
, θ
i+1
),
˜
θ
i
H(t
i
; θ
i+1
, θ
i
)] =
2
˜
θ
i+1
˜
θ
i
θ
i+2
H
21
(t
i+1
)H
12
(t
i
) −θ
i
H
12
(t
i+1
)H
21
(t
i
)
. (15.29)
For H diagonal the commutator term vanishes as was to be expected. In that case, the
CampbellBakerHausdorﬀ formula truncates to the trivial term both in the matrix and in
the Grassmann representations. This does not mean that there are no discretisation errors
in the fermionic path integral when H is diagonal, as can be seen from eq. (15.19). To be
precise, assuming for simplicity that H = −
1
2
Wσ
3
, one has
exp(−iH∆t) = 1
2
−i
H −
i
2
W tan(
1
4
W∆t)1
2
¸
2 sin(
1
2
W∆t)
W
, (15.30)
which shows that ∆t is eﬀectively modiﬁed to 2 sin(
1
2
W∆t)/W (it is interesting to contrast
this with the result we found for the harmonic oscillator in section 6 and problem 10),
whereas H is shifted by a multiple of the identity that vanishes linearly in ∆t.
The generalisation of eqs. (15.27) and (15.28) to N steps is now obvious and for H
diagonal one easily proves that the limit N →∞ can be taken:
dθ < θ
′
U(T, 0)θ >< θΨ >= lim
N→∞
dθ
0
U
N
(θ
N
= θ
′
, θ
0
= θ)Ψ(θ
0
)
= lim
N→∞
N−1
¸
j=0
dθ
j
d
˜
θ
j
exp
¸
N−1
¸
j=0
˜
θ
j
[θ
j
−θ
j+1
−iH(t
j
; θ
j+1
, θ
j
)∆t]
¸
Ψ(θ
0
) , (15.31)
where as usual one has ∆t = T/N. In complete analogy with eq. (6.10), reinstating the
dependence on Planck’s constant, we can write
< θ
′
Texp(−i
H(t)dt/¯ h)θ >
= lim
N→∞
d
˜
θ
0
N−1
¸
j=1
dθ
j
d
˜
θ
j
exp
i∆t
¯ h
N−1
¸
j=0
i
˜
θ
j
(θ
j+1
−θ
j
)
∆t
−
˜
θ
j
H(t
j
; θ
j+1
, θ
j
)
¸
¸
= lim
N→∞
d
˜
θ
0
N−1
¸
j=1
dθ
j
d
˜
θ
j
exp
i∆t
¯ h
N−1
¸
j=0
i
˜
θ
j
(θ
j+1
−θ
j
)
∆t
+
˜
θ
j
W(t
j
)
(θ
j+1
+ θ
j
)
2
¸
¸
≡
DΨ(t)DΨ(t) exp
¸
i
T
0
dt Ψ
†
(t)(i∂
t
+ W(t))Ψ(t)
¸
. (15.32)
Here we have replaced
˜
θ with Ψ
†
(which in this case agrees with Ψ, since in one time and no
space dimensions γ
0
≡ 1) and θ with Ψ. We have indicated the general case where W can
depend on time, but in absence of this time dependence, W is the energy of the oneparticle
state, created from the vacuum. Since we identiﬁed θ also with b
†
, the creation operator for
the oneparticle state, we see that WΨ
†
Ψ = Wbb
†
= −H +
1
2
W, what is to be expected from
a relation between the Lagrangian and Hamiltonian (the term
1
2
W is of course irrelevant) .
As for scalar ﬁeld theories, we will be mainly interested in vacuum expectation values
of the evolution operator. In the presence of a source term this will allow us to derive all
required matrix elements. For the present case we easily ﬁnd the vacuum wave function to
be
< θ0 >= 1 , < 0θ >= θ , (15.33)
61
such that
< 0U(T)0 >=
dθ
′
< 0θ
′
>
dθ < θ
′
U(T)θ >< θ0 >=
dθdθ
′
θ
′
< θ
′
U(T)θ > .
(15.34)
The order of the diﬀerentials is important here. However, as in the case of scalar ﬁeld
theories, we do not require the precise form of the vacuum wave function(al) for performing
perturbation theory.
Up to now we have described a spin onehalf particle pinneddown at a ﬁxed position.
It is obvious how this can be generalised to include the quantum mechanical description of
a moving spin onehalf particle in a onedimensional potential V (x). If also W depends on
the particle position (W(x)), the Hamiltonian becomes
H =
ˆ p
2
2m
+ V (ˆ x)
1
2
−
1
2
W(ˆ x)σ
3
. (15.35)
If we write < x, θΨ >≡ Ψ(x, θ) ≡ a
0
(x) + θa
1
(x), the path integral is easily found to be
< x
′
, θ
′
 exp(−iHT/¯ h)x, θ >= lim
N→∞
dp
0
d
˜
θ
0
2π¯ h
N−1
¸
j=1
dp
j
dx
j
2π¯ h
dθ
j
d
˜
θ
j
× (15.36)
exp
i∆t
¯ h
N−1
¸
j=0
p
j
(x
j+1
−x
j
)
∆t
−
p
2
j
2m
−V (x
j
) +
i
˜
θ
j
(θ
j+1
−θ
j
)
∆t
+
˜
θ
j
W(x
j
)
(θ
j+1
+ θ
j
)
2
¸
¸
≡
Dx(t)DΨ(t)DΨ(t) exp
¸
i
¯ h
T
0
dt
1
2
m˙ x
2
(t) −V (x(t)) + Ψ
†
(t)[i∂
t
+ W(x(t))]Ψ(t)
¸
.
A careful derivation of this formula and a step by step comparison with the matrix represen
tation in the spin degrees of freedom can be found in sections 1 to 3 of the paper “Fermionic
coordinates and supersymmetry in quantum mechanics”, Nuclear Physics B196 (1982) 509,
by P. Salomonson and J.W. van Holten. For further details see the lectures by L. Faddeev
in “Methods in ﬁeld theory”, Les Houches 1975, ed. R. Balian and J. ZinnJustin.
It is now also straightforward to derive the path integral for the Dirac Hamiltonian of
the previous section. Using as a basis the plane waves constructed there, the Hamiltonian
becomes a decoupled sum (in a ﬁnite volume) for each
k of four fermions, described by
b
†
1
(
k), b
†
2
(
k), d
†
1
(
k) and d
†
2
(
k). Each of which can be described by its own θ. Associating the
annihilation operators with their respective
˜
θ, and performing the Fourier transformation
back to coordinate space, it is left as an exercise to show that the path integral for fermions
is given by
DΨ(x)DΨ(x) exp
¸
i
¯ h
d
4
x
Ψ(x)(iγ
µ
∂
µ
−m)Ψ(x) −
¯
J(x)Ψ(x) −Ψ(x)J(x)
. (15.37)
Since the ﬁelds Ψ(x) and Ψ(x) are Grassmann variables, also the sources J(x) and
¯
J(x) are
Grassmann variables. Their order in the above equation is therefore important, when used
in further calculations. As promised, it is as simple as for scalar ﬁeld theories to calculate the
dependence of this path integral on the sources. Since Grassmann variables form a complex
linear space we can perform all calculations as in the scalar case, provided we keep track of
the order of the Grassmann odd variables. In particular we can make the replacement
Ψ(x) → Ψ(x) +
d
4
x
′
G
F
(x −x
′
)J(x
′
) ,
Ψ(x) → Ψ(x) +
d
4
x
′
¯
J(x
′
)G
F
(x
′
−x) , (15.38)
62
where G
F
(x) is the Green’s function deﬁned in eq. (14.9). The integration measure, as for
commuting variables, is invariant under a shift by a constant Grassmann variable, such that
we obtain (as for the scalar case we normalise the path integral to 1 for vanishing sources)
< 0U
J
¯
J
(T)0 > = < 0U
J=
¯
J=0
(T)0 > exp
−i
d
4
xd
4
y
¯
J(x)G
F
(x −y)J(y)
≡ < 0U
J=
¯
J=0
(T)0 > Z
2
(J,
¯
J) . (15.39)
Again, this result holds to arbitrary order in the sources, and agrees with what can be derived
to second order within Hamiltonian perturbation theory (see problem 24).
Interactions are taken into account by adding higher order terms to the Lagrangian, where
the order of the fermion ﬁelds is important. For example, the Lagrangian for a fermionic
and a scalar ﬁeld is given by
L = L
2
−V (Ψ, Ψ, σ) −J(x)σ(x) −
¯
J(x)Ψ(x) −Ψ(x)J(x) ,
L
2
= Ψ(x)(iγ
µ
∂
µ
−m)Ψ(x) +
1
2
∂
µ
σ(x)∂
µ
σ(x) −
1
2
M
2
σ
2
(x) . (15.40)
We ﬁnd as in eq. (8.6)
Z(J,
¯
J, J, g
n
) = exp
−i
d
4
x V (
−iδ
δJ(x)
,
iδ
δ
¯
J(x)
,
iδ
δJ(x)
)
Z
2
(J,
¯
J, J) ,
Z
2
(J,
¯
J, J) = exp
¸
−i
d
4
xd
4
y
1
2
J(x)G(x −y)J(y) +
¯
J(x)G
F
(x −y)J(y)
. (15.41)
Note the minus sign for the derivative with respect to J(x), which is because the source
stands behind the ﬁeld component Ψ(x). Derivatives of Grassmann variables are simply
deﬁned as one would intuitively expect
d
dθ
1 = 0 ,
d
dθ
θ = 1 ,
d
dθ
θ
′
= 0 , (15.42)
together with a generalised Leibnitz rule for functions f and g that are either even or odd
Grassmann variables, a property denoted by the sign or grading s(= ±1),
d
dθ
(fg) = s
f
f
d
dθ
g +
d
dθ
f
g . (15.43)
By declaring the derivative to be a linear function on the Grassmann algebra, it can be
uniquely extended to this algebra from the above set of rules. Note that these rules imply
that the Grassmann integral over a total Grassmann derivative vanishes. Comparing with
eq. (15.21) we note that apparently Grassmann integration and diﬀerentiation is one and
the same thing. Both project on the coeﬃcient in front of the Grassmann variable. The
vanishing of the integral over a total derivative and of the derivative of an integral is in that
perspective trivial. More importantly, to make sense of eq. (15.41), one easily shows the
following identity to hold
d
dθ
exp(θx) = x . (15.44)
An example of an interaction between the fermions and a scalar ﬁeld σ is given by the
socalled Yukawa interaction
V (Ψ(x), Ψ(x), σ(x)) = gΨ(x)Ψ(x)σ(x) . (15.45)
63
We can also consider the interaction of the fermions with the electromagnetic ﬁeld, whose
quantisation will be undertaken in the next section. For this we can take the minimal
coupling in eq. (14.12) (see also eq. (14.10)), such that
V (Ψ(x), Ψ(x), A
µ
(x)) = −eΨ(x)γ
µ
Ψ(x)A
µ
(x) , (15.46)
which will play a dominating role in describing Quantum Electrodynamics (QED).
As before, log Z(
¯
J, J, J) is the sum over connected diagrams. Diagrams that involve
fermions necessarily have as many lines ending in a source J as in a source
¯
J. This is
because the Lagrangian is Grassmann even, a requirement that can be related to the Lorentz
invariance. It does not in general require the Lagrangian to be bilinear in Ψ and Ψ. In section
19 and problem 30 we will discuss the fourFermi interaction, Ψ(x)γ
µ
Ψ(x)Ψ(x)γ
µ
Ψ(x), which
is clearly Lorentz invariant and Grassmann even. However, for many of the theories we
discuss, the Lagrangian is bilinear in the fermionic ﬁelds, because higher order terms will
generally not be renormalisable (except in one space and one time dimension). If no higher
order fermionic interactions occur, a fermionic line either forms a loop or it goes from a
source J to a source
¯
J. As changing the order of fermionic ﬁelds and sources gives a sign
change, this has consequences for the Feynman diagrams too. However, it is cumbersome
to determine the overall sign of a diagram. Fortunately, all we need is the relative sign of
the various diagrams that contribute to the Green’s function with a ﬁxed number of sources,
since the overall sign drops out in our computations of cross sections and decay rates. If one
diagram can be obtained from the other by crossing two fermion lines, this gives a relative
minus sign, as for
× ×
× ×
−→ −
.
.
.
.
.
× ×
× ×
.
.
..





 ﬁg. 6
It also implies that each loop formed by a fermion line carries a minus sign. Intuitively this
follows, as is indicated in the following ﬁgure by the dashed box, from the identity above
× ×
,
.
.
.
.
.
.
.
.
.






−→ −
× ×
,
ﬁg. 7
More accurately a fermion loop that connects vertices x
k
for k = 1 to n is associated to
n
¸
k=1
Ψ(x
k
)Ψ(x
k
) →
n
¸
k=1
−iδ
δJ(x
k
)
iδ
δ
¯
J(x
k
)
→
−
¸
{k}
Tr
iG
F
(x
k(1)
−x
k(2)
)iG
F
(x
k(2)
−x
k(3)
) · · · iG
F
(x
k(n)
−x
k(1)
)
, (15.47)
where {k} stands for the various orders in which the vertices are connected. For each vertex
we have only indicated the fermionic part Ψ(x
k
)Ψ(x
k
). For the examples of eqs. (15.45) and
(15.46), the scalar or vector ﬁeld contributions are not indicated, as they are not relevant
for the fermion loop. The trace is with respect to the spinor indices which are not displayed
explicitly to keep the notation simple. We used that
Ψ(x
k(j)
)Ψ(x
k(j+1)
) →
iδ
δ
¯
J(x
k(j)
)
−iδ
δJ(x
k(j+1)
)
→iG
F
(x
k(j)
−x
k(j+1)
) , (15.48)
64
where an extra minus sign arises since in log Z
2
(J,
¯
J) (see eq. (15.39))
¯
J comes ﬁrst, and
has to be anticommuted with δ/δJ, before this derivative can be taken. The overall minus
sign comes from the term that closes the loop
¯
Ψ(x
k(1)
)AΨ(x
k(n)
) = −AΨ(x
k(n)
)
¯
Ψ(x
k(1)
) , (15.49)
where A is Grassmann even. We contrast this, as an example, with a scalar loop for the ﬁeld
ϕ that arises in the theory discussed in section 11, which is described by the Lagrangian
L =
1
2
(∂
µ
ϕ)
2
−
1
2
m
2
ϕ
2
+
1
2
(∂
µ
σ)
2
−
1
2
M
2
σ
2
−
1
2
ϕ
2
σ −Jσ −jϕ. One ﬁnds (k(n + 1) ≡ k(1))
n
¸
k=1
ϕ(x
k
)ϕ(x
k
) →
n
¸
k=1
iδ
δJ(x
k
)
iδ
δJ(x
k
)
→
¸
{k}
n
¸
j=1
iG(x
k(j)
−x
k(j+1)
) , (15.50)
which completes the demonstration of the extra minus sign for fermion loops. Note that the
factors of i, associated to the derivatives with respect to the sources, are absorbed in the
vertices for the Feynman rules in the table on page 35, which is why the propagator in that
table equals −i times the Green’s function. This is also the Feynman rule for the fermion
propagator. But the extra minus sign in the derivative with respect to the fermionic source
J (see eq. (15.41)) is not absorbed in the vertex in order to guarantee that vertex factors are
assigned as in the scalar theory. That minus sign is, however, absorbed in eq. (15.48) due to
the anticommuting nature of the fermionic variables, which was in the ﬁrst place the reason
for the extra minus sign in eq. (15.41) to appear. Only the overall minus sign required in
closing a fermion loop remains as an extra factor.
Before we convert these Feynman rules to the ones involved in computing the scattering
matrix, cross sections and decay rates (see pg. 42 and eq. (9.25)), we need to determine
the wavefunction factors to be inserted for the external lines that correspond to the in and
outgoing fermion lines. For this we express the creation and annihilation operators in terms
of the fermionic ﬁelds (at t = 0), such that their insertion in the operator formulation can
be converted in the path integral, as in the scalar case, to derivatives with respect to the
sources. Using eq. (13.17) and the orthogonality relations of eq. (13.14) one ﬁnds (cmp.
eq. (7.26))
b
α
(
k) =
u
(α)
(
k)
†
2k
0
(
k)
d
3
x
(2π)
3
e
−i
k·x
ˆ
Ψ(x) , d
†
α
(
k) =
v
(α)
(
k)
†
2k
0
(
k)
d
3
x
(2π)
3
e
i
k·x
ˆ
Ψ(x) (15.51)
and through conjugation we get
b
†
α
(
k) =
d
3
x
(2π)
3
e
i
k·x
ˆ
Ψ
†
(x)
u
(α)
(
k)
2k
0
(
k)
, d
α
(
k) =
d
3
x
(2π)
3
e
−i
k·x
ˆ
Ψ
†
(x)
v
(α)
(
k)
2k
0
(
k)
.
(15.52)
In the Hamiltonian formulation the scattering matrix is given by
out
< ( p
1
, α
1
), ( p
2
, α
2
), · · · , ( p
ℓ
, α
ℓ
)(
k
1
, β
1
), (
k
2
, β
2
), · · · , (
k
n
, β
n
) >
in
=
< 0c
α
1
( p
1
)c
α
2
( p
2
) · · · c
α
ℓ
( p
ℓ
)U(T
out
, T
in
)c
†
β
1
(
k
1
)c
†
β
2
(
k
2
) · · · c
†
βn
(
k
n
)0 >, (15.53)
where in a shorthand notation we separate particles from antiparicles by the helicity index
c
1
(
k) ≡ b
1
(
k) , c
2
(
k) ≡ b
2
(
k) , c
3
(
k) ≡ d
1
(
k) , c
4
(
k) ≡ d
2
(
k) . (15.54)
65
Like in eq. (9.3), the insertion of a ﬁeld operator at the appropriate time is in the path
integral represented by a derivative with respect to the source
ˆ
b
†
α
(
k) =
γ
0
u
(α)
(
k)
a
2k
0
(
k)(2π)
3
d
3
x e
i
k·x
−iδ
δJ
a
(x, t = T
in
)
=
γ
0
u
(α)
(
k)
a
2k
0
(
k)
−iδ
δ
˜
J
a
(
k, t = T
in
)
,
ˆ
d
†
α
(
k) =
v
(α)
a
(
k)
∗
2k
0
(
k)(2π)
3
d
3
x e
i
k·x
iδ
δ
¯
J
a
(x, t = T
in
)
=
v
(α)
a
(
k)
∗
2k
0
(
k)
iδ
δ
˜
¯
J
a
(
k, t = T
in
)
,
ˆ
b
α
(
k) =
u
(α)
a
(
k)
∗
2k
0
(
k)(2π)
3
d
3
x e
−i
k·x
iδ
δ
¯
J
a
(x, t = T
out
)
=
u
(α)
a
(
k)
∗
2k
0
(
k)
iδ
δ
˜
¯
J
a
(−
k, t = T
out
)
,
ˆ
d
α
(
k) =
γ
0
v
(α)
(
k)
a
2k
0
(
k)(2π)
3
d
3
x e
−i
k·x
−iδ
δJ
a
(x, t = T
out
)
=
γ
0
v
(α)
(
k)
a
2k
0
(
k)
−iδ
δ
˜
J
a
(−
k, t = T
out
)
,
(15.55)
such that
out
< ( p
1
, α
1
), ( p
2
, α
2
), · · · , ( p
ℓ
, α
ℓ
)(
k
1
, β
1
), (
k
2
, β
2
), · · · , (
k
n
, β
n
) >
in
=
ℓ
¸
i=1
ˆ c
α
i
( p
i
)
n
¸
j=1
ˆ c
†
β
j
(
k
j
) exp(G
J,
¯
J
)
J=
¯
J=0
. (15.56)
The ˆ c
α
(
k) are of course deﬁned in terms of
ˆ
b
α
(
k) and
ˆ
d
α
(
k) as in eq. (15.54). We continue
as in section 9 by ﬁrst ﬁxing the wavefunction and mass renormalisations in terms of the
connected twopoint function (we leave it as an exercise to show that in the absence of
interactions
out
< ( p, α)(
k, β) >
in
= e
−ip
0
( p)T
δ
k, p
δ
α,β
).
G
(2)
c
(J,
¯
J) =
× × +
`
× ×
+
`
`
× ×
+· · ·
≡
`
× ×
, (15.57)
where the selfenergy is now a 4 × 4 matrix given by (−i×) the amputated 1PI 2point
function
iΣ
ab
(p) ≡
`
a
b
. (15.58)
(The 1PI diagram equals Σ
ab
(p), when evaluated with the Feynman rules for the reduced
matrix elements of table 4 (pg. 69) by dropping the wavefunction factors.) The convention
for these sort of diagrams is that momentum ﬂows in the direction of the arrow, which points
to the ﬁrst spinor index (here a). With these deﬁnitions the twopoint function becomes
G
(2)
c
(J,
¯
J) = −i
d
4
p
˜
¯
J
a
(−p)
¸
1
p / −m−Σ(p) + iε
¸
ab
˜
J
b
(p) , (15.59)
with between square brackets the inverse of the 4×4 matrix (p
µ
γ
µ
−m−Σ(p)+iε)
ab
, which is
the full propagator in the momentum representation for the conventions on pg. 42. As long
as we don’t break the Lorentz invariance, the full propagator near the poles is of the form of
the free propagator with a wavefunction renormalisation factor Z
F
and a renormalised mass
˜ m, such that on the massshell one has (cmp. eq. (9.12))
G
(2)
c
(J,
¯
J) = −iZ
F
d
4
p
˜
¯
J(−p)
¸
1
p / − ˜ m+ iε
¸
˜
J(p) . (15.60)
66
Performing the wavefunction renormalisation, we have to modify eqs. (15.55) and (15.56)
accordingly (cmp. eqs. (913) and (9.14)).
ˆ
b
α
+
(
k) =
γ
0
u
(α)
(
k)
a
2Z
F
k
2
+ ˜ m
2
−iδ
δ
˜
J
a
(
k, t = T
in
)
,
ˆ
b
α
−
(
k) =
u
(α)
a
(
k)
∗
2Z
F
k
2
+ ˜ m
2
iδ
δ
˜
¯
J
a
(−
k, t = T
out
)
,
ˆ
d
α
+
(
k) =
v
(α)
a
(
k)
∗
2Z
F
k
2
+ ˜ m
2
iδ
δ
˜
¯
J
a
(
k, t = T
in
)
,
ˆ
d
α
−
(
k) =
γ
0
v
(α)
(
k)
a
2Z
F
k
2
+ ˜ m
2
−iδ
δ
˜
J
a
(−
k, t = T
out
)
,
(15.61)
and, with the ˆ c
α
±
(
k) deﬁned as in eq. (15.54),
out
< ( p
1
, α
1
), ( p
2
, α
2
), · · · , ( p
ℓ
, α
ℓ
)(
k
1
, β
1
), (
k
2
, β
2
), · · · , (
k
n
, β
n
) >
in
=
ℓ
¸
i=1
ˆ c
α
i
−
( p
i
)
n
¸
j=1
ˆ c
β
j
+
(
k
j
) exp(G
J,
¯
J
)
J=
¯
J=0
. (15.62)
To compute the wavefunction factors for the external lines we express the npoint function
in terms of amputated npoint functions, as in eq. (9.15), with the diﬀerence that there has
to be an even number of external lines, since the number of J and
¯
J sources has to be equal
(we ignore for the moment any other bosonic ﬁelds that might be present, including those
in the external lines will be obvious). The amputated npoint function will now carry the
spinor index of each of the external lines and one has
G
(2n)
c
(J,
¯
J) ≡
n
¸
j=1
d
4
p
j
d
4
k
j
G
(amp)
c (p
1
, p
2
, · · · , p
n
; k
1
, k
2
, · · · , k
n
)
b
1
,b
2
,···,bn
a
1
,a
2
,···,an
× (15.63)
n
¸
j=1
˜
¯
J
c
j
(p
j
)
¸
−i
−p /
j
−m−Σ(−p
j
) + iε
¸
c
j
b
j
¸
−i
k/
j
−m−Σ(k
j
) + iε
¸
a
j
d
j
˜
J
d
j
(k
j
)
.
Note that we have assumed one particular ordering for the sources. Relative signs of the
diagrams are determined by the rules that were described above (it is not diﬃcult to convince
oneself that with respect to the fermion lines, any diagram can be generated from a given
one by permuting fermion lines). As in eqs. (9.20) and (9.21) we can compute the action of
ˆ c
α
±
(
k), from which the wavefunction factors will be obtained. Like in eq. (9.19), computing
the action of
ˆ
b
α
−
( p) and
ˆ
d
α
+
( p) we can restrict our attention to
A(· · ·)
···
···
≡
d
4
p
˜
¯
J
a
(p)
¸
−i
−p / −m−Σ(−p) + iε
¸
ab
G
(amp)
c (p, · · ·)
b···
···
=
d
4
p
√
2π
dt
˜
¯
J
a
( p, t)
¸
−ie
ip
0
t
−p / −m−Σ(−p) + iε
¸
ab
G
(amp)
c (p, · · ·)
b···
···
. (15.64)
If we deﬁne as usual p
0
( p) =
√
p
2
+ ˜ m
2
and for convenience of notation change p to −p in
eq. (15.64), we ﬁnd
ˆ
b
α
−
( p)A(· · ·)
···
···
=
u
(α)
a
( p)
∗
4πp
0
( p)Z
F
dp
0
¸
e
−ip
0
Tout
p / −m−Σ(p) + iε
¸
ab
G
(amp)
c (−p, · · ·)
b···
···
=
−i
(2π)
4
Z
F
u
(α)
a
( p)
∗
(p / + ˜ m)
ab
2p
0
( p)
G
(amp)
c (−p, · · ·)
b···
···
2p
0
( p)(2π)
3
e
−ip
0
( p)Tout
, (15.65)
67
which is obtained by deforming the p
0
integration contour to the lower halfplane (since
T
out
→∞), and computing the contribution from the pole at p
0
= p
0
( p), taking into account
eq. (15.60). Its residue is proportional to the matrix Z
F
(p / + ˜ m)/2p
0
( p), which is most easily
found using 1/(p / − ˜ m + iε) = (p / + ˜ m)/(p
2
− ˜ m
2
+ iε). We can now use the fact that the
spinor u
(α)
( p) satisﬁes the equations of motion (see eq. (13.2)), such that
u
(α)
a
( p)
∗
(p / + ˜ m)
ab
2p
0
( p)
=
(p / + ˜ m)
†
2p
0
( p)
u
(α)
( p)
∗
b
=
(γ
0
p /γ
0
+ ˜ m)
2p
0
( p)
u
(α)
( p)
∗
b
= u
(α)
( p)
b
. (15.66)
Consequently, eq.(15.65) becomes
ˆ
b
α
−
( p)A(· · ·)
···
···
= −iu
(α)
( p)
b
(2π)
4
Z
F
G
(amp)
c (−p, · · ·)
b···
···
2p
0
( p)(2π)
3
e
−ip
0
( p)Tout
. (15.67)
The wavefunction factor for an outgoing electron is therefore given by u
(α)
( p)
b
√
Z
F
. The
convention is that the momentum ﬂows out of the diagram, along the arrow (see the table
below), this is why the amputated npoint function has −p as its argument, like for the
scalar case, where we deﬁned for the amputated npoint function all momenta to ﬂow into
the diagram. This means that in the reduced matrix element M the outgoing electron
momenta occur precisely as indicated in eq. (9.18).
By similar arguments we obtain from eq. (15.64)
ˆ
d
α
+
( p)A(· · ·)
···
···
=
v
(α)
a
( p)
∗
4πp
0
( p)Z
F
dp
0
¸
e
ip
0
T
in
−p / −m−Σ(−p) + iε
¸
ab
G
(amp)
c (p, · · ·)
b···
···
=
−i
(2π)
4
Z
F
v
(α)
a
( p)
∗
(−p / + ˜ m)
ab
2p
0
( p)
G
(amp)
c (p, · · ·)
b···
···
2p
0
( p)(2π)
3
e
ip
0
( p)T
in
=
iv
(α)
( p)
b
(2π)
4
Z
F
G
(amp)
c (p, · · ·)
b···
···
2p
0
( p)(2π)
3
e
ip
0
( p)T
in
, (15.68)
such that the wavefunction factor for an incoming antiparticle (positron) is −v
(α)
( p)
b
√
Z
F
.
In this case the momentum ﬂows against the arrow of the fermion line, but does ﬂow into
the diagram as is required in the convention of the reduced matrix element.
To compute the action of
ˆ
b
α
+
( p) and
ˆ
d
α
−
( p), we restrict our attention to
B(· · ·)
···
···
≡
d
4
p G
(amp)
c (p, · · ·)
···
b···
¸
−i
p / −m−Σ(p) + iε
¸
ba
˜
J
a
(p) =
d
4
p
√
2π
dt G
(amp)
c (p, · · ·)
···
b···
¸
−ie
ip
0
t
p / −m−Σ(p) + iε
¸
ba
˜
J
a
( p, t) . (15.69)
One ﬁnds
ˆ
b
α
+
( p)B(· · ·)
···
···
=
dp
0
G
(amp)
c (p, · · ·)
···
b···
¸
e
ip
0
T
in
p / −m−Σ(p) + iε
¸
ba
¸
γ
0
u
(α)
( p)
a
4πp
0
( p)Z
F
=
−i
(2π)
4
Z
F
G
(amp)
c (p, · · ·)
···
b···
2p
0
( p)(2π)
3
(p / + ˜ m)
ba
2p
0
( p)
γ
0
u
(α)
( p)
a
e
ip
0
( p)T
in
=
−i
(2π)
4
Z
F
G
(amp)
c (p, · · ·)
···
b···
2p
0
( p)(2π)
3
u
(α)
b
( p)e
ip
0
( p)T
in
, (15.70)
68
and (again for convenience changing p to −p)
ˆ
d
α
−
( p)B(· · ·)
···
···
=
dp
0
G
(amp)
c (−p, · · ·)
···
b···
¸
e
−ip
0
Tout
−p / −m−Σ(−p) + iε
¸
ba
¸
γ
0
v
(α)
( p)
a
4πp
0
( p)Z
F
=
−i
(2π)
4
Z
F
G
(amp)
c (−p, · · ·)
···
b···
2p
0
( p)(2π)
3
(−p / + ˜ m)
ba
2p
0
( p)
γ
0
v
(α)
( p)
a
e
−ip
0
( p)T
in
=
i
(2π)
4
Z
F
G
(amp)
c (−p, · · ·)
···
b···
2p
0
( p)(2π)
3
v
(α)
b
( p)e
−ip
0
( p)Tout
. (15.71)
In both cases there is an extra minus sign from pulling δ/δJ through
¯
J in eq. (15.63). The
wavefunction factor for an incoming electron is hence u
(α)
b
( p)
√
Z
F
with the momentum ﬂow
ing along the fermionic arrow, whereas the wavefunction factor of an outcoming antiparticle
(or positron) is given by −v
(α)
b
( p)
√
Z
F
, where the momentum ﬂows opposite to the fermionic
arrow. The minus signs in front of some of the wavefunction factors are irrelevant (they can
be absorbed in the overall sign ambiguity).
In the table below we summarise the Feynman rules that correspond to the fermionic
pieces in computing the reduced matrix elements. We have chosen the convention that
the incoming momenta ﬂow in, and the outgoing momenta ﬂow out of the diagram. This
guarantees that eqs. (9.18), (10.12) and (11.11) (resp. the scattering matrix, cross section and
decay rate), remain valid in the presence of fermions. Consequently, all fermion momenta
in the table ﬂow from left to right. For conventions where momenta always ﬂow in the
direction of the fermion arrow the four momentum for wavefunction factors associated to
in and outgoing antiparticles (positrons) should be reversed. Signs from fermion loops and
exchanges of external lines will not be implicit in diagrams, as only relative signs are known.
table 4
>
>
k
3
`
k
2
a
k
1
b
≡ gδ
ab
and k
1
= k
2
+ k
3
Yukawa vertex
.
.
>
>
k
3
µ
`
k
2
a
k
1
b
≡ −eγ
µ
ab
and k
1
= k
2
+ k
3
photon vertex
b
a
k
≡
1
k/−m+iε
ab
fermion propagator
a
α, k
≡
√
Z
F
u
(α)
a
(
k) , k
0
=
k
2
+ ˜ m
2
incoming electron
,
a
α, k
≡
√
Z
F
v
(α)
(
k)
a
, k
0
=
k
2
+ ˜ m
2
incoming positron
a
α, k
≡
√
Z
F
u
(α)
(
k)
a
, k
0
=
k
2
+ ˜ m
2
outgoing electron
,
a
α, k
≡
√
Z
F
v
(α)
a
(
k) , k
0
=
k
2
+ ˜ m
2
outgoing positron
−1 ×i
d
4
k
(2π)
4
for each fermion loop
−1 interchange of fermion lines
69
16 Feynman rules for vector ﬁelds
As before, the quantisation for vector ﬁelds starts by expanding the ﬁeld in plane waves and
identifying the Fourier coeﬃcients with creation and annihilation operators.
A
µ
(x) =
d
3
k
2k
0
(
k)(2π)
3
¸
λ
a
λ
(
k)ε
(λ)
µ
(
k)e
−ikx
+ a
†
λ
(
k)ε
(λ)
µ
(
k)
∗
e
ikx
, (16.1)
where ε
(λ)
µ
(
k)e
−ikx
are independent plane wave solutions of the equations of motion. The
index λ enumerates the various solutions for ﬁxed momentum. They will be identiﬁed with
the spin components or helicity eigenstates of the vector.
We will ﬁrst discuss the simpler case of a massive vector ﬁeld, expected to describe a
massive spin one particle. In problem 12 we already saw that its Lagrangian is given by
L
A
= −
1
4
F
µν
F
µν
+
1
2
m
2
A
µ
A
µ
−A
µ
(x)J
µ
(x) , (16.2)
and that the free equations of motion (i.e. J
µ
= 0) are equivalent to
∂
µ
A
µ
(x) = 0 and (∂
µ
∂
µ
+ m
2
)A
ν
(x) = 0 . (16.3)
As usual, this implies the onshell condition k
2
0
=
k
2
+ m
2
, but also
k
µ
ε
(λ)
µ
(
k) = 0 . (16.4)
It removes one of the four degrees of freedom of a fourvector. Three independent components
remain, exactly what would be required for a particle with spin one. We may for example
choose
ε
(λ)
µ
(
0) = δ
λ
µ
(λ = 1, 2, 3) . (16.5)
in the restframe of the particle, which is extended to an arbitrary frame by applying a
Lorentz boost. They satisfy the Lorentz invariant normalisation
g
µν
ε
(λ)
µ
(
k)
∗
ε
(λ
′
)
ν
(
k) = −δ
λλ
′
. (16.6)
The minus sign is just a consequence of the fact that in our conventions g
ij
= −δ
ij
. The
spin wave functions also satisfy a completeness relation which is given by
Λ
µν
(
k) ≡
¸
λ
ε
(λ)
µ
(
k)ε
(λ)
ν
(
k)
∗
= −
g
µν
−
k
µ
k
ν
m
2
, (16.7)
most easily proven in the restframe. Since the spin wave functions are by construction
Lorentz vectors, the above expression forms a Lorentz tensor and its onshell extension to
an arbitrary frame is therefore unique. It is easily seen to project arbitrary vectors w
µ
on
to vectors that satisfy k
µ
w
µ
= 0. The propagator for the massive spin one ﬁeld was already
computed in problem 12
.
.
.
.
k
µ ν
=
−
g
µν
−
kµkν
m
2
k
2
−m
2
+ iε
=
Λ
µν
(
k)
k
2
−m
2
+ iε
. (16.8)
Especially for the massless spin one ﬁeld (i.e. the photon ﬁeld) to be discussed below, it
is advantageous to decompose the spin with respect to the direction of the particle’s motion,
70
which are called helicity eigenstates. We have helicities 0 and ±1, described by the spin
wave functions
ε
(0)
µ
(k) =
¸

k
m
,
−k
0
m
k
k
¸
, ε
(±)
µ
(k) =
1
2
√
2
¯ ε
(1)
µ
(k) ±i¯ ε
(2)
µ
(k)
, (16.9)
where k
µ
, ε
(0)
µ
(k), ¯ ε
(1)
µ
(k) and ¯ ε
(2)
µ
(k) form a complete set of real orthogonal fourvectors.
These new spin wave functions satisfy the same properties as in eqs. (16.6) and (16.7) and
are also deﬁned oﬀshell, where they satisfy
Λ
µν
(k) ≡
¸
λ
ε
(λ)
µ
(k)ε
(λ)
ν
(k)
∗
= −
g
µν
−
k
µ
k
ν
k
2
, (16.10)
Sums over λ will of course run over the set {0, +, −} in this case. We leave it as an exercise
to verify that rotations over an angle α around the axis pointing in the direction of
k leaves
ε
(0)
µ
(k) invariant and transforms ε
(±)
µ
(k) to e
±iα
ε
(±)
µ
(k).
The Hamiltonian for the massive spin one particles is given by
H =
d
3
k
1
2
k
0
(
k)
¸
λ
a
†
λ
(
k)a
λ
(
k) + a
λ
(
k)a
†
λ
(
k)
. (16.11)
which can be expressed in terms of three scalar ﬁelds ϕ
λ
, λ = 1, 2, 3, as
H =
d
3
k
1
2
˜ π
λ
(
k)
2
+
1
2
(
k
2
+ m
2
) ˜ ϕ
λ
(
k)
2
=
d
3
x
1
2
π
2
λ
(x) +
1
2
(∂
i
ϕ
λ
(x))
2
+
1
2
m
2
ϕ
2
λ
(x)
.
(16.12)
where we deﬁned, as in eqs.(2.6) and (2.7),
˜ ϕ
λ
(
k) =
a
λ
(
k) + a
†
λ
(−
k)
/
2k
0
(
k) , ˜ π
λ
(
k) =
i
2
2k
0
(
k)
a
†
λ
(
k) −a
λ
(−
k)
. (16.13)
The corresponding Lagrangian would be given by
L
ϕ
=
¸
λ
1
2
(∂
µ
ϕ
λ
(x))
2
−
1
2
m
2
ϕ
2
λ
(x) −ϕ
λ
(x)J
(λ)
(x)
, (16.14)
where we have added a source for each scalar ﬁeld. We introduce also a ﬁeld σ for the
component of the vector ﬁeld along k
µ
. Writing
˜
A
µ
(k) ≡
1
m
˜ σ(k)k
µ
+
¸
λ
˜ ϕ
λ
(k)ε
(λ)
µ
(k) ,
˜
J
µ
(k) ≡
1
m
˜
j(k)k
µ
+
¸
λ
˜
J
(λ)
(k)ε
(λ)
µ
(k) , (16.15)
a simple calculation shows that L
A
≡ L
ϕ
−
1
2
(∂
µ
σ(x))
2
−σ(x)j(x). We see that σ decouples
from the other components and behaves like a scalar particle with the wrong sign for the
kinetic term. This would lead to serious inconsistencies, which are circumvented if we take
∂
µ
J
µ
(x) = 0, such that we can put σ ≡ 0.
It is important to realise that it is the Lorentz invariance that requires us to describe a
spin one particle by a fourvector. From the point of view of the scalar degrees of freedom,
ϕ
λ
, this invariance seems to be lost when we introduce interactions in the Lagrangian of
eq. (16.14). Nevertheless, if we treat λ as a threevector index (in some internal space) and
demand the interactions to be O(3) invariant with respect to this index (i.e. invariance
under rigid rotations and reﬂections in the internal space), then we claim that the resulting
71
interactions do respect the Lorentz invariance. The reason is simple, because the O(3)
invariance requires that the λ index is always pairwise contracted. Eqs. (16.10) and (16.15)
guarantee that such a pair, written in terms of the vector ﬁeld A
µ
(x), is a Lorentz scalar as far
as it concerns the dependence on A, which is suﬃcient if the Lagrangian is Lorentz invariant
when treating λ as a dummy label. To eliminate the σ ﬁeld we have to enforce ∂
µ
A
µ
(x) = 0,
not only onshell, but also oﬀshell. This can be achieved by adding to eq. (16.2) a term
−λ(x)∂
µ
A
µ
(x) (cmp. problem 8). Using
DA
µ
(x)Dλ(x) exp
¸
−i
d
4
x λ(x)∂
µ
A
µ
(x)
=
D
˜
A
µ
(k)
¸
k
δ(k
µ
˜
A
µ
(k)) , (16.16)
we see that the socalled Lagrange multiplier ﬁeld λ(x) plays the role of removing the un
wanted degree of freedom. Since we modiﬁed the theory oﬀshell, the propagator in eq. (16.8)
has to be changed also, by replacing Λ(
k) by its oﬀshell value Λ(k) (eq. (16.10)). If none of
the vertices or sources couple to the σ ﬁeld, we might just as well replace it by −g
µν
. There
are a number of other ways to eliminate the σ degree of freedom, see e.g. section 323 in
Itzykson and Zuber. We will come back to massive vector particles in section 19.
For the computation of the scattering matrix we express the annihilation and creation
operators in terms of the vector ﬁelds at t = 0, using the relations
a
λ
(
k) = −
2k
0
(
k)ε
(λ)
µ
(
k)
∗
d
3
x
(2π)
3
ˆ
A
µ
(x)e
−i
k·x
,
a
†
λ
(
k) = −
2k
0
(
k)ε
(λ)
µ
(
k)
d
3
x
(2π)
3
ˆ
A
µ
(x)e
i
k·x
. (16.17)
which in the path integral turn into
ˆ a
†
λ
(
k) =
2k
0
(
k)ε
(λ)
µ
(
k)
−iδ
δ
˜
J
µ
(
k, t = T
in
)
, ˆ a
λ
(
k) =
2k
0
(
k)ε
(λ)
µ
(
k)
∗
−iδ
δ
˜
J
µ
(−
k, t = T
out
)
,
(16.18)
which is identical to eq. (9.3), when reexpressed in terms of J
(λ)
. Like for scalar and fermion
ﬁelds there will be a mass and wavefunction (denoted by Z
A
) renormalisation, determined
through the selfenergy of the vector ﬁeld, which is now a Lorentz tensor of rank two. It
is proportional to Λ
µν
(k) (to guarantee that the scalar ﬁeld σ introduced above decouples
from the other ﬁelds; alternatively it can be seen as a consequence of the O(3) invariance
with respect to the index λ). We can consequently deﬁne
Σ
µν
(p) ≡ Λ
µν
(p)Σ
A
(p) . (16.19)
The npoint Green’s functions can now be written in terms of amputated Green’s func
tions that carry fourvector indices for each external spin one line,
G
(n)
c
(J
µ
) ≡
n
¸
j=1
d
4
p
j
˜
J
µ
j
(p
j
)
−iΛ
µ
j
ν
j
(p
j
)
p
2
j
−m
2
−Σ
A
(p
j
) + iε
G
amp
c
(p
1
, p
2
, · · · , p
n
)
ν
1
,ν
2
,···,νn
.
(16.20)
Using the fact that onshell Λ
µ
ν
(
k)ε
(λ)
ν
(
k) = −ε
(λ)
µ
(
k), we ﬁnd for the incoming spin one line
a wavefunction factor
√
Z
A
ε
(λ)
µ
(
k) and for the outgoing line a factor
√
Z
A
ε
(λ)
µ
(
k)
∗
.
For a massless spin one ﬁeld (the photon) we would expect the helicity zero component
of the vector ﬁeld to be absent. First we have to redeﬁne, however, what we would mean
72
with the zero helicity component, because eq. (16.9) is singular in the limit of zero mass.
We take as our deﬁnition
ε
(0)
µ
(k) = n
µ
(k) ≡
1
2
√
2
¸
1,
k

k
¸
, ε
(±)
µ
(k) = (0, s
±
(
k)) , (16.21)
with
k · s
±
(
k) = 0. These still form with k
µ
four independent fourvectors and ε
(±)
µ
(
k) are
still transverse polarisations. However, it is no longer true that k
µ
ε
(0)
µ
(
k) will vanish onshell
(i.e. at k
2
= 0). This is easily seen to imply that onshell a
0
(
k) = 0. In other words, onshell
there is no longitudinal component for the photon. The extra degree of freedom is removed
by the gauge invariance of the massless vector ﬁeld, as was discussed at the end of section
4 and in problem 9. To perform the quantisation of the theory, one can go about as in the
massive case. Due to the presence of the fourvector n(k) it will be much more cumbersome
to demonstrate the Lorentz invariance. In section 20 it will be shown how in principle in
any gauge the path integral can be deﬁned and that the result is independent of the chosen
gauge. One could then choose a gauge that allows us to show the equivalence between the
path integral and the canonical quantisation. However, it is the great advantage of the path
integral formulation that calculations can be performed in a gauge in which the Lorentz
invariance is manifest. The gauge most suitable for that purpose is of course the Lorentz
gauge ∂
µ
A
µ
(x) = 0, see eqs. (4.21) and (4.22). The propagator is read oﬀ from eq. (4.26)
.
.
.
.
k
µ ν
=
−
g
µν
−(1 −
1
α
)
kµkν
k
2
+iε
k
2
+ iε
≡
Λ
(α)
µν
(k)
k
2
+ iε
, (16.22)
where α is an arbitrary parameter, on which physical observables like cross sections and decay
rates should not depend. It is in general not true anymore that the selfenergy is proportional
to Λ
(α)
µν
(k), but the gauge invariance does guarantee that Σ
µν
(k)
˜
j
ν
(k) is independent of α for
any conserved current, i.e. k
µ
˜
j
µ
(k) = 0. It can be shown that this in general implies
Σ
µν
(k) = Λ
(α
′
)
µν
(k)Σ(k) , (16.23)
for some, possibly inﬁnite, α
′
. Apart from a wavefunction renormalisation (Z
A
), the gauge
ﬁxing parameter α will in principle have to be renormalised too. One still has for any value of
α that Λ
(α)
(k)
µ
ν
ε
(±)
ν
(
k) = −ε
(±)
µ
(
k). The wavefunction factors for external photon lines are
therefore identical to the ones for the massive case, except that now only two helicity states
can appear. It is an important consequence of gauge invariance that unphysical degrees of
freedom decouple in the physical amplitudes. It also implies that the selfenergy vanishes
on shell (see problem 39), such that it will not give rise to a renormalisation of the mass.
The photon remains massless. That the gauge invariance must be crucial here is clear, as a
massive photon would have one extra degree of freedom. We will come back to this point in
section 19. In the table below we summarise the Feynman rules.
table 5
.
.
.
.
k
µ ν
≡
−
gµν−(1−
1
α
)
kµkν
k
2
+iε
k
2
+iε
photon propagator (Lorentz gauge)
.
.
.
µ
λ, k
≡
√
Z
A
ε
(λ)
µ
(
k) incoming photon
.
.
.
µ
λ, k
≡
√
Z
A
ε
(λ)
µ
(
k)
∗
outgoing photon
73
17 Quantum Electrodynamics  QED
QED is the ﬁeld theory that describes the interaction between the photon and the charged
fermions. In the Lorentz gauge (see eqs. (4.21) and (4.22)) the Lagrangian is given by
L = −
1
4
F
µν
(x)F
µν
(x) −
1
2
α(∂
µ
A
µ
(x))
2
+
¸
f
Ψ
f
(x)(iγ
µ
D
µ
−m
f
)Ψ
f
(x) . (17.1)
Here f is de socalled ﬂavour index, which distinguishes the various types of fermions (elec
trons, protons, etc.). The covariant derivative D
µ
is given as before (see eqs. (3.35) and
(14.11)) by
D
µ
Ψ
f
(x) = (∂
µ
+ iq
f
A
µ
(x))Ψ
f
(x) . (17.2)
For electrons we have q
f
= −e and for protons q
f
= e. For α = 0 the Lagrangian is invariant
under gauge transformations
A
µ
(x) →A
µ
(x) + ∂
µ
Λ(x) , Ψ
f
(x) →exp(−iq
f
Λ(x))Ψ
f
(x) . (17.3)
The Feynman rules are collected in the table below.
table 6
.
.
.
.
k
µ ν
≡
−
gµν−(1−
1
α
)
kµkν
k
2
+iε
k
2
+iε
photon propagator (Lorentz gauge)
.
.
>
>
k
3
µ
`
k
2
a
k
1
b
≡ qγ
µ
ab
and k
1
= k
2
+ k
3
photon vertex (fermion charge is q)
b
a
k
≡
1
k/−m+iε
ab
fermion propagator
Before calculating cross sections we wish to discuss in more detail the helicity of the
fermions and its relation to charge conjugation C. The latter relates, say electrons to
positrons, or in general particles to antiparticles, which is an important symmetry of the
theory. It, as well as parity (P) and time reversal (T) symmetry can be separately broken,
but the combination CPT is to be unbroken to allow for a local, relativistic invariant ﬁeld
theory. The spin components of the solutions in eq. (13.8) were based on a decomposition
along the zaxis in the restframe. Helicity, as for the photon, is deﬁned by decomposing
the spin in the direction of motion,
k. It is hence deﬁned in terms of the eigenvalues of the
operator
ˆ
k ·
J ≡
k
i
4
k
ε
ijk
σ
jk
=
1
2
ˆ
k · σ ⊘
⊘
1
2
ˆ
k · σ
. (17.4)
(
J is the spin part of the angular momentum operator, the equivalent of
1
2
σ for a two
component spinor.) This holds both in the Dirac and Weyl representations. It is easy to
verify that [
ˆ
k ·
J, k/ ] = 0, e.g. by making use of the fact that in the Dirac representation
k/ =
k
0
1
2
k · σ
−
k · σ −k
0
1
2
. (17.5)
74
This implies that we can choose u
(α)
0
and v
(α)
0
to be eigenstates of the helicity operator
ˆ
k ·
J
(consequently they become functions of
ˆ
k)
ˆ
k ·
J
˜
Ψ(
k) = ±
1
2
˜
Ψ(
k) . (17.6)
Instead of the label α we can use ± to indicate the helicity and we have
u
±
(
k) =
(k/ + m)
m+k
0

u
±
0
(
ˆ
k) ,
ˆ
k ·
Ju
±
0
(
ˆ
k) = ±
1
2
u
±
0
(
ˆ
k) ,
v
±
(
k) =
(−k/ + m)
m+k
0

v
±
0
(
ˆ
k) ,
ˆ
k ·
Jv
±
0
(
ˆ
k) = ∓
1
2
v
±
0
(
ˆ
k) . (17.7)
Note the ﬂip of helicity for the positron wave functions. For
k = (0, 0, k) these eigenstates
coincide with the decomposition in eq. (13.8). It is clear that we can deﬁne
u
±
0
(
ˆ
k) =
¸
¸
ϕ
±
(
ˆ
k)
0
0
¸
,
ˆ
k · σ ϕ
±
(
ˆ
k) = ±ϕ
±
(
ˆ
k) ,
v
±
0
(
ˆ
k) =
¸
¸
0
0
χ
±
(
ˆ
k)
¸
,
ˆ
k · σ χ
±
(
ˆ
k) = ∓χ
±
(
ˆ
k) , (17.8)
with ϕ
±
and χ
±
each an orthonormal set of twocomponent spinors. They can be related to
each other by
χ
±
(
ˆ
k) ≡ −iσ
2
ϕ
∗
±
(
ˆ
k) . (17.9)
Indeed, when we use that
σ
2
σ
i
σ
2
= −σ
∗
i
, i = 1, 2, 3 , (17.10)
which expresses the fact that SU(2) is socalled pseudo real, we ﬁnd
ˆ
k·σ
−iσ
2
ϕ
∗
±
(
ˆ
k)
=
−i
ˆ
k · σ
∗
σ
2
ϕ
±
(
ˆ
k)
∗
=
iσ
2
ˆ
k · σ ϕ
±
(
ˆ
k)
∗
= ∓
−iσ
2
ϕ
∗
±
(
ˆ
k)
. (17.11)
As eq. (17.9) relates the components of the electron wave function to those of the positron
wave function, it is the basis of the charge conjugation symmetry, which relates the solutions
of the Dirac equation to solutions of the complex conjugate Dirac equation (see eq. (12.31)),
which indeed interchanges positive and negative energy solutions, i.e. particles and antipar
ticles. To formulate this symmetry in the fourcomponent spinor language one introduces
the charge conjugation matrix (in the Dirac representation)
C ≡ −iγ
0
γ
2
=
⊘ −iσ
2
−iσ
2
⊘
, (17.12)
which satisﬁes
C
−1
= C
†
= −C , Cγ
µ
C
−1
= −γ
t
µ
. (17.13)
This can be proven from the explicit form of the Dirac matrices. The equivalent of eq. (17.10)
is given by
γ
2
γ
µ
γ
2
= −γ
∗
µ
. (17.14)
75
It is now easy to verify that
v
±
(
k) = C¯ u
t
±
(
k) , u
±
(
k) = C¯ v
t
±
(
k) . (17.15)
We just need to prove one of these identities, because charge conjugation is an involution,
i.e. applying it twice gives the identity
C
CΨ
t
t
= Cγ
t
0
C
∗
γ
†
0
Ψ = Ψ . (17.16)
We ﬁnd
C¯ u
t
±
(
k) = Cγ
t
0
u
∗
±
(
k) = iγ
2
u
∗
±
(
k) = iγ
2
(k/
∗
+ m)
m +k
0

¸
¸
ϕ
∗
±
(
ˆ
k)
0
0
¸
= i
(−k/ + m)
m+k
0

γ
2
¸
¸
ϕ
∗
±
(
ˆ
k)
0
0
¸
=
(−k/ + m)
m +k
0

⊘ iσ
2
−iσ
2
⊘
¸
¸
ϕ
∗
±
(
ˆ
k)
0
0
¸
=
(−k/ + m)
m+k
0

¸
¸
0
0
χ
±
(
ˆ
k)
¸
= v
±
(
k) . (17.17)
Under charge conjugation the charge that appears in the covariant derivative in eq. (17.2)
should change sign too. To show this we multiply the complex conjugate of the Dirac equation
with iγ
2
iγ
2
[(−iγ
µ
(∂
µ
−ieA
µ
) + m) Ψ]
∗
=
iγ
2
γ
∗
µ
γ
2
(∂
µ
+ ieA
µ
) + m
(iγ
2
Ψ
∗
)
= (−iγ
µ
(∂
µ
+ ieA
µ
) + m) (CΨ
t
) . (17.18)
That charge conjugation is really a symmetry of the quantum theory, is most convincingly
demonstrated by the fact that the Dirac Lagrangian is invariant under charge conjugation.
Using CΨ
t
= Ψ
t
γ
∗
0
C
†
γ
0
= −Ψ
t
C
−1
, the anticommuting properties of the fermi ﬁelds and
partial integration, we ﬁnd
d
4
x CΨ
t
(iγ
µ
(∂
µ
+ ieA
µ
) −m) CΨ
t
=
d
4
x −Ψ
t
C
−1
(iγ
µ
(∂
µ
+ ieA
µ
) −m) CΨ
t
=
d
4
x −Ψ
t
−iγ
t
µ
(∂
µ
+ ieA
µ
) −m
Ψ
t
=
d
4
x Ψ(iγ
µ
(∂
µ
−ieA
µ
) −m) Ψ . (17.19)
In particular we see that the electromagnetic current generated by the fermi ﬁelds transforms
as required for the interchange of particles and antiparticles, under which the charge changes
sign
j
µ
= Ψγ
µ
Ψ
C
→−Ψγ
µ
Ψ . (17.20)
An important consequence of the charge conjugation symmetry is Furry’s theorem, which
states that a fermionic loop with an odd number of vertices will not contribute to the
amplitude. Consider a fermion loop as in ﬁgure (8a) below
(a)
`
º
.
p
k
n
k
1
k
2
.
.
.
.
.
. (b)
`
,
p
k
n
k
1
k
2
.
.
.
.
.
. ﬁg. 8
for which the Feynman rules lead to the expression (note that spinor index contractions run
against the arrow of the fermion line and
¸
i
k
i
= 0)
q
n
Tr
1
p / −m + iε
γ
µ
1
1
p / + k/
1
−m + iε
γ
µ
2
1
p / + k/
1
+ k/
2
−m+ iε
γ
µ
3
· · · γ
µn
. (17.21)
76
Using the fact that for any matrix A we have Tr(A) = Tr(A
t
) = Tr(CA
t
C
−1
), we convert
eq. (17.21) to the expression
(−q)
n
Tr
γ
µn
· · · γ
µ
3
1
−p / −k/
1
−k/
2
−m + iε
γ
µ
2
1
−p / −k/
1
−m + iε
γ
µ
1
1
−p / −m+ iε
,
(17.22)
which is exactly (−1)
n
times the result of the Feynman diagram that is obtained by inverting
the orientation of the fermion line (i.e. the vertices are connected in the reversed order) as
indicated in ﬁgure (b). As both diagrams will occur, their contributions will cancel whenever
n is odd. It conﬁrms the intuition that particles and antiparticles contribute equally, except
for their opposite charge factors (±q)
n
.
We will now calculate the cross section for electronelectron scattering (the socalled
Møller cross section). In lowest nontrivial order there are only two diagrams that contribute,
as indicated in the ﬁgure.
.
.
.
>
>
>
>
>>
k
1
, t
1
k
2
, t
2
p
1
, s
1
p
2
, s
2
−
.
.
.
>
>
k
1
, t
1
k
2
, t
2
p
1
, s
1
p
2
, s
2
ﬁg. 9
The labels t
i
and s
i
indicate the helicities of the incoming and outgoing electrons. The
scattering matrix (ignoring the time dependent phase factor) for this process is given by
out
< ( p
1
, s
1
), ( p
2
, s
2
)(
k
1
, t
1
), (
k
2
, t
2
) >
in
=
−i(2π)
4
δ
4
(p
1
+ p
2
−k
1
−k
2
)M({(−p
1
, s
1
), (−p
2
, s
2
)}, {(k
1
, t
1
), (k
2
, t
2
)})
2p
(1)
0
( p
1
)(2π)
3
2p
(2)
0
( p
2
)(2π)
3
2k
(1)
0
(
k
1
)(2π)
3
2k
(2)
0
(
k
2
)(2π)
3
=
iδ
4
(p
1
+ p
2
−k
1
−k
2
)
(4π)
2
p
(1)
0
p
(2)
0
k
(1)
0
k
(2)
0
¯ u
s
1
(p
1
)eγ
µ
u
t
1
(k
1
)g
µν
¯ u
s
2
(p
2
)eγ
ν
u
t
2
(k
2
)
(k
1
−p
1
)
2
+ iε
−
¯ u
s
1
(p
1
)eγ
µ
u
t
2
(k
2
)g
µν
¯ u
s
2
(p
2
)eγ
ν
u
t
1
(k
1
)
(k
1
−p
2
)
2
+ iε
¸
. (17.23)
The relative minussign is of course a consequence of the socalled FermiDirac statistics,
which implements the Pauli principle. We got rid of the gauge dependent part of the photon
propagator (see table 6) by using the fact that the currents generated by ¯ u
s
(p)γ
µ
u
t
(k) are
conserved, such that
¯ u
s
(p)γ
µ
u
t
(k)(p
µ
−k
µ
) = ¯ u
s
(p)[(p / −m) −(k/ −m)]u
t
(k) = 0 , (17.24)
because onshell (k/ − m)u(k) = 0 (and hence also ¯ u(k)(k/ − m) = 0). Indirectly, through
current conservation, this is related to gauge invariance. It guarantees that the longitudinal
component of the photon does not contribute to the scattering matrix, which is thus seen
not to depend on the gauge ﬁxing parameter α.
The diﬀerential cross section for unpolarised electronelectron scattering is given by (see
eq. (10.12), from now on we will drop the distinction between
¯
k
i
and k
i
)
dσ =
¸
s
1
,s
2
d
3
p
1
2p
0
( p
1
)(2π)
3
d
3
p
2
2p
0
( p
2
)(2π)
3
(2π)
4
δ
4
(p
1
+ p
2
−k
1
−k
2
)
4
(k
1
· k
2
)
2
−m
4
×
1
4
¸
t
1
,t
2
M({(−p
1
, s
1
), (−p
2
, s
2
)}, {(k
1
, t
1
), (k
2
, t
2
)})
2
, (17.25)
77
where
1
4
¸
t
1
,t
2
stands for averaging over the polarisations of the incoming electrons. For
the total cross section we should multiply with a factor
1
2
to avoid double counting the
identical outgoing electrons, or restrict the scattering angle θ to the interval θ ∈ [0, π/2],
when integrating over the outgoing momenta. The latter convention will be followed here.
In the center of mass system the scattered particles move back to back in a direction which
is only determined modulo π, which is why θ ∈ [0, π/2], with θ measured from the incoming
particle direction (also deﬁned modulo π).
To calculate M
2
we use
¸
s,t
¯ u
s
(p)γ
µ
u
t
(k) (¯ u
s
(p)γ
ν
u
t
(k))
∗
=
¸
s,t
¯ u
s
(p)γ
µ
u
t
(k) ¯ u
t
(k)γ
0
γ
†
ν
γ
0
u
s
(p) =
Tr
γ
µ
¸
t
u
t
(k) ⊗ ¯ u
t
(k)γ
ν
¸
s
u
s
(p) ⊗ ¯ u
s
(p)
= Tr
γ
µ
(k/ + m)γ
ν
(p / + m)
, (17.26)
which can be represented graphically as follows
¸
s,t
.
.
.
.
>
>
>>
k, t k, t
p, s
ν µ
M
∗
M
=
¸
t
.
.
.
.
>
>
>>
k, t k, t
p
ν µ
=
¸
t
.
.
.
.
»
´
k, t
p
ν µ
=
.
.
.
.
,
k
p
ν µ
. ﬁg. 10
Hence, we add a Feynman rule for the socalled cut fermion propagator
b
a
k
= sign(k
0
)(k/ + m)
ab
. (17.27)
For antiparticles k
0
< 0 (see eq. (13.18) for the extra minus sign). These results can be
generalised to other ﬁelds too, by noting that our conventions have been such that the
propagators can be written as
¸
β
φ
β
(
k) ⊗
¯
φ
β
(
k)
∗
k
2
−m
2
+ iε
, (17.28)
where φ
β
(
k) are the wave functions for the incoming lines and
¯
φ
β
(
k)
∗
for the outgoing lines,
with β labelling the internal degrees of freedom (cmp. eq. (16.8)).
We can now use this result to compute M
2
¸
s
1
,s
2
,t
1
,t
2
M
2
=
.
.
.
.
.
,
,
k
2
p
2
p
1
k
1
−
.
.
.
.
.
,
,
k
2
p
2
p
1
k
1
+ (p
1
←→p
2
)
= e
4
Tr[γ
µ
(k/
1
+ m)γ
ν
(p /
1
+ m)] Tr[γ
µ
(k/
2
+ m)γ
ν
(p /
2
+ m)]
((k
1
−p
1
)
2
+ iε) ((k
2
−p
2
)
2
+ iε)
−
Tr[γ
µ
(k/
1
+ m)γ
ν
(p /
2
+ m)γ
µ
(k/
2
+ m)γ
ν
(p /
1
+ m)]
((k
1
−p
1
)
2
+ iε) ((k
1
−p
2
)
2
+ iε)
¸
+(p
1
←→p
2
) . (17.29)
To compute the traces over the gamma matrices we use the following identities (problem 21)
Tr(γ
µ
γ
ν
) = 4g
µν
, Tr(γ
µ
γ
ν
γ
α
γ
β
) = 4(g
µν
g
αβ
+ g
µβ
g
αν
−g
µα
g
νβ
) ,
¸
µ
γ
µ
γ
α
γ
β
γ
µ
= 4g
αβ
,
¸
µ
γ
µ
γ
ν
γ
α
γ
β
γ
µ
= −2γ
β
γ
α
γ
ν
, (17.30)
78
and the fact that the trace over an odd number of gamma matrices vanishes. This implies
Tr
γ
µ
(k/+m)γ
ν
(p /+m)
= Tr
γ
µ
k/γ
ν
p /
+4m
2
g
µν
= 4(m
2
−k·p)g
µν
+4k
µ
p
ν
+4k
ν
p
µ
, (17.31)
and
¸
µ
γ
µ
(k/ + m)γ
ν
(p / + m)γ
µ
= −2p /γ
ν
k/ −2m
2
γ
ν
+ 4mk
ν
+ 4mp
ν
. (17.32)
Together with momentum conservation (p
1
+ p
2
= k
1
+ k
2
) and the onshell conditions
(p
2
1
= p
2
2
= k
2
1
= k
2
2
= m
2
), which imply identities like p
1
· p
2
= k
1
· k
2
, we ﬁnd
¸
s
1
,s
2
,t
1
,t
2
M
2
= e
4
16
g
µν
(m
2
−k
1
· p
1
) + k
µ
1
p
ν
1
+ k
ν
1
p
µ
1
g
µν
(m
2
−k
2
· p
2
) + k
2
µ
p
2
ν
+ k
2
ν
p
2
µ
(k
1
−p
1
)
2
+ iε
(k
2
−p
2
)
2
+ iε
−κ
Tr
¸
2p /
2
γ
ν
k/
1
+ 2m
2
γ
ν
−4mk
ν
1
−4mp
ν
2
(k/
2
+ m)γ
ν
(p /
1
+ m)
(k
1
−p
1
)
2
+ iε
(k
1
−p
2
)
2
+ iε
¸
+ (p
1
↔p
2
)
= 32e
4
(k
1
· k
2
)
2
+ (k
1
· p
2
)
2
+ 2m
2
(k
1
· p
2
−k
1
· k
2
)
(k
1
−p
1
)
2
+ iε
2
−κ
(k
1
· k
2
)
2
−2m
2
k
1
· k
2
(k
1
−p
1
)
2
+ iε
(k
1
−p
2
)
2
+ iε
¸
+ (p
1
↔p
2
) . (17.33)
The parameter κ determines the relative sign of the “crossed” diagrams in eq. (17.29),
which arise from multiplying the direct electronelectron scattering diagram with the complex
conjugate of the one where the outgoing fermion lines were crossed. For FermiDirac statistics
κ ≡ −1. By keeping track of the dependence on κ one sees how scattering experiments can
be used to verify the anticommuting nature of the electrons.
We ﬁnally perform some kinematics and express the diﬀerential cross section in terms of
the scattering angle θ. We deﬁne in the center of mass frame
k
0
1
= k
0
2
= p
0
1
= p
0
2
≡ E , p
1
= − p
2
≡ p ,
k
1
= −
k
2
≡
k with  p = 
k . (17.34)
Deﬁning θ to be the angle between
k and p, i.e. p ·
k =
k
2
cos θ, we have
k
1
· k
2
= E
2
+
k
2
= 2E
2
−m
2
, k
1
· p
1
= E
2
−
k
2
cos θ = E
2
(1 −cos θ) + m
2
cos θ ,
(p
1
−k
1
)
2
= −4
k
2
sin
2
(
1
2
θ) , k
1
· p
2
= E
2
+
k
2
cos θ = E
2
(1 + cos θ) −m
2
cos θ ,
(p
2
−k
1
)
2
= −4
k
2
cos
2
(
1
2
θ) , (k
2
· k
1
)
2
−m
4
= 4E
2
(E
2
−m
2
) = 4
k
2
E
2
. (17.35)
Finally we use that identity (Ω is the solid angle, dΩ = sin θdθdφ)
d
3
p
1
d
3
p
2
δ
4
(p
1
+p
2
−k
1
−k
2
) =
p
2
dpdΩ δ
2
p
2
+ m
2
−2
k
2
+ m
2
=
dΩ
1
2
E
k .
(17.36)
Collecting all terms we ﬁnd for eq. (17.25) the result
dσ
dΩ
=
1
(2E(2π)
3
)
2
·
(2π)
4
1
2
E
k
4
4
k
2
E
2
·
1
4
¸
s
1
,s
2
,t
1
,t
2
M
2
=
1
2
10
π
2
E
2
¸
s
1
,s
2
,t
1
,t
2
M
2
, (17.37)
79
with
¸
s
1
,s
2
,t
1
,t
2
M
2
given by
32e
4
(2E
2
−m
2
)
2
+[E
2
(1+cos θ)−m
2
cos θ]
2
+2m
2
[E
2
(1+cos θ)−m
2
cos θ+m
2
−2E
2
]
16(
k
2
)
2
sin
4
(
1
2
θ)
+
(2E
2
−m
2
)
2
+[E
2
(1−cos θ) + m
2
cos θ]
2
+ 2m
2
[E
2
(1−cos θ)+m
2
cos θ+m
2
−2E
2
]
16(
k
2
)
2
cos
4
(
1
2
θ)
−κ
2(2E
2
−m
2
)(2E
2
−3m
2
)
16(
k
2
)
2
sin
2
(
1
2
θ) cos
2
(
1
2
θ)
¸
=
32e
4
(
k
2
)
2
[(2E
2
−m
2
)
2
+E
4
+(E
2
−m
2
)
2
cos
2
θ+2m
2
(m
2
−E
2
)](cos
4
(
1
2
θ)+sin
4
(
1
2
θ))
sin
4
θ
+
cos θ[2E
2
(E
2
−m
2
)+2m
2
(E
2
−m
2
)](cos
4
(
1
2
θ)−sin
4
(
1
2
θ))
sin
4
θ
−κ
2(2E
2
−m
2
)(2E
2
−3m
2
)
4 sin
2
θ
¸
=
16e
4
(
k
2
)
2
(E
2
−m
2
)
2
+
4(2E
2
−m
2
)
2
sin
4
θ
−
3(2E
2
−m
2
)
2
−m
4
+κ(2E
2
−m
2
)(2E
2
−3m
2
)
sin
2
θ
¸
, (17.38)
yielding (α
e
=
e
2
4π
=
e
2
4π¯hc
≈
1
137
is the ﬁnestructure constant and κ = −1)
dσ
dΩ
=
α
2
e
(2E
2
−m
2
)
2
4E
2
(E
2
−m
2
)
2
4
sin
4
θ
−
3
sin
2
θ
+
(E
2
−m
2
)
2
(2E
2
−m
2
)
2
1 +
4
sin
2
θ
¸
. (17.39)
This cross section is invariant under θ → π − θ, such that we cannot tell the two outgoing
electrons apart, as it should be. For electronelectron scattering we have to put κ = −1, but
we see from the dependence on κ in eq. (17.38) that one can easily distinguish experimentally
if electrons behave according to the FermiDirac statistics.
In problem 29 electronpositron scattering is studied with in the ﬁnal state an electron
and a positron (Bhabha scattering) or a muon and an antimuon. Both for e
−
e
−
−→e
−
e
−
and
e
−
e
+
−→e
−
e
+
one cannot take θ too close to zero (or π for e
−
e
−
). Apart from the fact that
the detector would be in the way of the beam, it is fundamentally impossible to distinguish
the scattered particles at θ = 0 (and θ = π for e
−
e
−
) from those in the beam. The divergence
of the diﬀerential cross section was therefore to be expected. For e
−
e
+
−→ µ
−
µ
+
this
divergence is absent and one can deﬁne the total cross section by integrating over all angles.
For E ≫m
e
and E ≫m
µ
one ﬁnds (see problem 29) σ =
1
3
πα
2
e
¯ h
2
c
2
/E
2
= 21.7nb/E
2
(GeV).
We now discuss electronphoton scattering, also known as Compton scattering. The
resulting cross section is called the KleinNishina formula. There are again two diagrams
that contribute in lowest nontrivial order to the scattering matrix.
.
.
. .
.
.
k, t
k
′
, t
′
p, s
p
′
, s
′
+
.
.
.
.
.
.
´
k, t
k
′
, t
′
p, s
p
′
, s
′
ﬁg. 11
The cross section is now given by
dσ =
¸
s
′
,t
′
d
3
p
′
2p
′
0
( p
′
)(2π)
3
d
3
k
′
2k
′
0
(
k
′
)(2π)
3
(2π)
4
δ
4
(p + p
′
−k −k
′
)
4p · k
×
1
4
¸
s,t
M({(−p
′
, s
′
), (−k
′
, t
′
)}, {(p, s), (k, t)})
2
, (17.40)
80
where, as for electronelectron scattering, we will discuss unpolarised cross sections. This
requires averaging over the polarisations of the incoming particles (at the end we will mention
the dependence on the photon polarisations). Note that the photon has also two helicity
eigenstates, together with the electron
¸
s,t
contains four terms. The reduced matrix element
for the two diagrams is given by
M=
¯ u
s
′ (p
′
)eγ
µ
ε
(t
′
)
µ
(k
′
)
∗
(p / + k/ + m)eγ
ν
ε
(t)
ν
(k)u
s
(p)
(p + k)
2
−m
2
+ iε
+ ((k, t) ←→(−k
′
, t
′
)) . (17.41)
We leave it as an exercise to verify that the cut photon propagator, for the choice of polari
sations discussed in eq. (16.21), is given by (k
2
≡ 0)
.
.
.
.
k
µ ν
≡
¸
t=±
ε
(t)
µ
(k)ε
(t)
ν
(k)
∗
= −
g
µν
−
k
µ
n
ν
(k) + k
ν
n
µ
(k)
k · n(k)
. (17.42)
Like for electronelectron scattering, we can compute M
2
graphically by
¸
s
1
,s
2
,t
1
,t
2
M
2
=
.
.
.
.
.
.
`
`
·
,
p
′
+k
′
p
p
′
p+k
+
.
.
.
.
.
.
.
.
`
`
·
,
p
′
−k
p
p
′
←k
+ (k
′
←→−k)
= e
4
Tr[γ
µ
(p / + k/ + m)γ
ν
(p / + m)γ
ν
′
(p / + k/ + m)γ
µ
′
(p /
′
+ m)]
((p + k)
2
−m
2
+ iε)
2
+
Tr[γ
µ
(p / + k/ + m)γ
ν
(p / + m)γ
µ
′
(p / −k/
′
+ m)γ
ν
′
(p /
′
+ m)]
((p + k)
2
−m
2
+ iε) ((p −k
′
)
2
−m
2
+ iε)
¸
×
¸
t
′
=±
ε
(t
′
)
µ
(k
′
)ε
(t
′
)
µ
′ (k
′
)
∗
¸
t=±
ε
(t)
ν
′ (k)ε
(t)
ν
(k)
∗
+ (k
′
←→−k) . (17.43)
The gauge invariance, i.e. conservation of the fermionic current, is again instrumental in
decoupling the longitudinal component of the photon ﬁeld. In this case the argument is
somewhat more subtle. Consider for example the term from the cut photon propagators
that contains k
ν
. It gives rise to the combination (using that p
2
= m
2
)
(p / + k/ + m)γ
ν
k
ν
(p / + m) = (p / + k/ + m){(p / + k/ −m) −(p / −m)}(p / + m)
= ((p + k)
2
−m
2
)(p / + m) . (17.44)
This means that one of the photon vertices is removed. There remain two diagrams, each
with one fermion loop and with an odd number of vertices. Furry’s theorem tells us that
these two diagrams add to zero. We may therefore just as well replace the cut photon
propagator by −g
µν
. Using this we ﬁnd
¸
s
1
,s
2
,t
1
,t
2
M
2
= e
4
Tr[γ
µ
(p / + k/ + m)γ
ν
(p / + m)γ
ν
(p / + k/ + m)γ
µ
(p /
′
+ m)]
((p + k)
2
−m
2
+ iε)
2
+
Tr[γ
µ
(p / + k/ + m)γ
ν
(p / + m)γ
µ
(p / −k/
′
+ m)γ
ν
(p /
′
+ m)]
((p + k)
2
−m
2
+ iε) ((p −k
′
)
2
−m
2
+ iε)
¸
+ (k
′
↔−k) . (17.45)
Taking the incoming electron at rest (p = (m,
0)), following similar steps as for electron
electron scattering, one will arrive at the result
dσ
dΩ
=
α
2
e
2m
2
k
′
0
k
0
2
k
0
k
′
0
+
k
′
0
k
0
−sin
2
θ
, (17.46)
81
where θ is the angle of the scattered photon with the direction of the incident photon. From
energy and momentum conservation one ﬁnds that
k
′
0
=
k
0
1 + (k
0
/m)(1 −cos θ)
. (17.47)
For a detailed derivation we refer to section 521 of Itzykson and Zuber and to section 86
of Berestetskii, e.a. (see section 1 for the reference).
In Itzykson and Zuber, as for most other textbooks, the result is derived by choosing the
photon polarisation such that ε(k) · p = 0 (keeping ε(k) · k = 0). With this choice it is even
possible to determine the polarised cross section (the polarisation of the electron is assumed
not to be observed)
dσ
dΩ
pol
=
α
2
e
4m
2
k
′
0
k
0
2
k
0
k
′
0
+
k
′
0
k
0
+ 4(ε
′
· ε)
2
−2
, (17.48)
where ε and ε
′
are the polarisations of resp. the incident and scattered photon. When
k
′
0
≪m one obtains the wellknown Thomson formula
dσ
dΩ
pol
=
α
2
e
m
2
(ε
′
· ε)
2
. (17.49)
The unpolarised cross section in this limit is obtained by summing over the scattered and
averaging over the incident polarisations
dσ
dΩ
=
α
2
e
2m
2
(1 + cos
2
θ) and σ =
8πα
2
e
3m
2
. (17.50)
18 NonAbelian gauge theories
Quantum Electrodynamics is an example of a U(1) gauge theory. U(1) is the group of the
unimodular complex numbers and determines the transformation of the charged ﬁelds
Ψ(x) →exp(−iqΛ(x))Ψ(x) ≡ g(x)Ψ(x) . (18.1)
It forms a group, which means that for any two elements g, h ∈U(1), the product is also in
U(1). Furthermore, any element has an inverse g
−1
, which satisﬁes gg
−1
= g
−1
g = 1. The
unit 1 satisﬁes g1 = 1g = g, for any g ∈U(1). U(1) is called an Abelian group because its
product is commutative. For every g, h ∈U(1), gh = hg.
It is now tempting to generalise this to other, in general noncommutative groups, which
are called nonAbelian groups. It was the way how Yang and Mills discovered SU(2) gauge
theories in 1954. Like for U(1) gauge theories they made the SU(2) transformation into a
local one, where at every point the ﬁeld can be transformed independently. (It should be
noted that they were originally after describing the isospin symmetry, that relates protons
to neutrons, which form a socalled isospin doublet.)
The simplest nonAbelian gauge group, for which no longer gh = hg, is SU(2). This
group is wellknow from the description of spin onehalf particles. It has a two dimensional
(spinor) representation, which can also be seen as a representation of the rotation group
SO(3). As a local gauge theory it does no longer act on the spinor indices, but on indices
related to some internal space, giving rise to socalled internal symmetries. The way the
82
gauge group G acts on the ﬁelds Ψ is described by a representation of the group G. A
representation deﬁnes a mapping ρ from G to the space of linear mappings Map(V ), of the
linear vector space V into itself.
ρ : G →Map(V ), ρ(g) : V →V . (18.2)
Mostly, V will be either IR
n
or
/
C
n
, in which case ρ(g) is resp. a real or a complex n × n
matrix. For ρ to be a representation, it has to preserve the group structure of G
ρ(g)ρ(h) = ρ(gh) , ρ(1) = id
V
. (18.3)
We will generally restrict the gauge symmetries to Liegroups for which one can write any
group element as an exponential of a Liealgebra element
g ≡ exp(X) , X ∈ L
G
. (18.4)
This Liealgebra has a noncommutative, antisymmetric bilinear product (required to satisfy
the Jabobi identity, as deﬁned in eq. (18.12))
(X, Y ) ∈ L
G
×L
G
→[X, Y ] ∈ L
G
. (18.5)
The CampbellBakerHausdorﬀ formula expresses that the logarithm of exp(X) exp(Y )
is an element of the Lie algebra, i.e. the product of two exponentials is again an exponential.
F(X, Y ) ≡ log (exp(X) exp(Y )) = X+Y +
1
2
[X, Y ]+
1
12
[X, [X, Y ]]+
1
12
[Y, [Y, X]]+· · · ∈ L
G
.
(18.6)
This formula will be of great help in ﬁnding a simple criterion for ρ to be a representation,
satisfying eq. (18.3). Apart from the group structure of Map(V ), it also has a Liealgebra
structure (the commutator of two n×n matrices is again a n×n matrix). The representations
of the group can be easily restricted to the Liealgebra.
ρ : L
G
→Map(V ) , (18.7)
in a way that preserves the Liealgebra structure
ρ([X, Y ]) = [ρ(X), ρ(Y )] = ρ(X)ρ(Y ) −ρ(Y )ρ(X) (18.8)
It is more or less by construction that we require
ρ(exp(X)) = exp(ρ(X)) , (18.9)
where on the lefthand side ρ is the group representation and on the righthand side it is
the Liealgebra representation. Without causing too much confusion we can use the same
symbol for the two objects. As a Liealgebra forms a linear vector space we can deﬁne a
basis on L
G
.
Z =
n
¸
a=1
z
a
T
a
∈ L
G
, z
a
∈ IR (or
/
C) , T
a
∈ L
G
. (18.10)
In here n is the dimension of the Liealgebra (and the Liegroup if, as we will assume
throughout, the exponential is locally an invertible mapping). The commutator, also called
Lieproduct, is completely determined by the structure constants f
abc
[T
a
, T
b
] =
¸
c
f
abc
T
c
. (18.11)
83
Using the Jacobi identity
[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0 , (18.12)
applied to X = T
a
, Y = T
b
and Z = T
c
, we ﬁnd (from now on sums over repeated group
indices are implicit)
f
bcd
f
ade
+ f
cad
f
bde
+ f
abd
f
cde
= 0 . (18.13)
This precisely coincides with the commutation relations of the socalled adjoint representa
tion
(T
a
ad
)
bc
≡ ρ
ad
(T
a
)
bc
= f
acb
. (18.14)
Indeed, one easily veriﬁes that
[ρ
ad
(T
a
), ρ
ad
(T
b
)] = f
abc
ρ
ad
(T
c
) . (18.15)
In general, since a representation preserves the commutation relations, it also preserves
the structure constants in terms of ρ(T
a
) ≡ T
a
ρ
, which forms a basis for the linear represen
tation space which is contained in V . They are called the generators of the representation.
With the help of eq. (18.6) we easily verify that ρ is a representation if and only if
[T
a
ρ
, T
b
ρ
] = f
abc
T
c
ρ
. (18.16)
This is because under the action of ρ one simply replaces T
a
by T
a
ρ
ρ
exp(x
a
T
a
)
= ρ(exp(X)) = exp(ρ(X)) = exp
x
a
T
a
ρ
. (18.17)
Similarly, the CampbellBakerHausdorﬀ formula, when expressed with respect to the Lie
algebra basis {T
a
}
exp
x
a
T
a
exp
y
b
T
b
= exp
{x
a
+ y
a
+
1
2
x
b
y
c
f
bca
+
1
12
(x
d
x
b
y
c
+ y
d
y
b
x
c
)f
bce
f
dea
+· · ·}T
a
.
(18.18)
directly determines the multiplication of the representation of group elements by replacing T
a
by T
a
ρ
, provided eq. (18.16) is satisﬁed. Note that the structure constants are antisymmetric
with respect to the ﬁrst two indices. They are also invariant under cyclic permutations of
the indices. This follows from the cyclic property of the trace
f
abd
Tr(T
d
ρ
T
c
ρ
) = Tr([T
a
ρ
, T
b
ρ
]T
c
ρ
) = Tr(T
a
ρ
[T
b
ρ
, T
c
ρ
]) = f
bcd
Tr(T
a
ρ
T
d
ρ
) , (18.19)
and from the fact that for compact groups the generators can be normalised such that
Tr(T
a
fnd
T
b
fnd
) = −
1
2
δ
ab
, (18.20)
where T
a
fnd
are the generators of the socalled fundamental or deﬁning representation of the
group G. This matrix representation is usually identiﬁed with the group (or algebra) itself,
which till now was seen more as an abstract entity. The simplest example is SU(2), the set
of complex unitary 2 × 2 matrices with unit determinant. Its fundamental representation
coincides with the spinor or spin onehalf representation. The structure constants and the
generators of the fundamental and adjoint representations were considered in sect. 12 (see
eq. (12.9))
T
a
fnd
= −
i
2
σ
a
, f
abc
= ε
abc
, ρ
ad
(T
a
) = −L
a
. (18.21)
84
Because the CampbellBakerHausdorﬀ formula plays such a crucial role in the theory
and in the practical implementation of group representations, we will now provide a more
abstract derivation of eq. (18.6) to all orders. The proof simply states how in the Taylor
expansion products of Liealgebra elements are regrouped in multiple commutators. A crucial
ingredient for deriving the CampbellBakerHausdorﬀ is the socalled derivation D, that maps
a product of Liealgebra elements into a multiple commutator.
DX = X , DX
i
1
X
i
2
· · · X
is
≡ [X
i
1
, [X
i
2
, · · · [X
i
s−1
, X
is
] · · ·]] , s > 1 . (18.22)
We also deﬁne for these products the adjoint map, ad, introduced in eq. (12.12)
adX
i
1
X
i
2
· · · X
is
≡ adX
i
1
adX
i
2
· · · adX
is
, (18.23)
which is easily seen to satisfy
ad([X, Y ]) = [adX, adY ] . (18.24)
It is more or less by deﬁnition that for any two products u and v of Liealgebra elements
D(uv) = aduDv . (18.25)
For two Liealgebra elements X and Y , it can easily be shown that
D[X, Y ] = D(XY ) −D(Y X) = adXDY −adY DX = [X, DY ] + [DX, Y ] (18.26)
and this allows us to prove by induction that a monomial Q (a polynomial of which all terms
are of the same order) of degree m in terms of Liealgebra elements X
i
, i = 1, 2, · · · , s is
an element of the Liealgebra (i.e. can be written as multiple commutators, called a Lie
monomial) if and only if DQ = mQ. If this equation is satisﬁed it is clear from the deﬁnition
of a derivation that Q is a Liemonomial. So it is suﬃcient to prove that the equation is
satisﬁed for Q a Liemonomial. In that case Q is a sum of terms, each of which can be
written as ad(X
i
1
)Q
(1)
with Q
(1)
a Liemonomial of degree m−1. Using eq. (18.26) therefore
yields Dad(X
i
1
)Q
(1)
= ad(X
i
1
)DQ
(1)
+ad(X
i
1
)Q
(1)
. Induction in m gives the required result.
Now it is trivial to regroup the terms in the Taylor expansion of eq. (18.6) in multiple
commutators. From the fact that any group element can be written as the exponent of a
Liealgebra element, we know that F(X, Y ) ∈ L
G
(at the worst one needs to restrict X and
Y to suﬃciently small neighbourhoods of the origin in L
G
). Consequently, in the Taylor
expansion of F(X, Y ) the collection of all terms at ﬁxed order m, denoted by F
m
(X, Y ), is
a monomial in X and Y and F
m
(X, Y ) is an element of the Liealgebra, such that
F(X, Y ) ≡
¸
m
F
m
(X, Y ) , F
m
(X, Y ) =
1
m
DF
m
(X, Y ) . (18.27)
It is not diﬃcult to work out the Taylor expansion for F(X, Y )
F(X, Y ) ≡ log (exp(X) exp(Y )) = log
¸
1 +
¸
i+j>0
X
i
Y
j
i!j!
¸
=
¸
k
(−1)
k−1
k
¸
¸
i+j>0
X
i
Y
j
i!j!
¸
k
,
(18.28)
from which we easily obtain the explicit expression for the CampbellBakerHausdorﬀ formula
in terms of multiple commutators,
F(X, Y ) =
¸
m
¸
{k,
¸
k
j=1
p
j
+q
j
=m, p
j
+q
j
>0}
(−1)
k−1
km
D(X
p
1
Y
q
1
X
p
2
Y
q
2
· · · X
p
k
Y
q
k
)
p
1
!q
1
!p
2
!q
2
! · · · p
k
!q
k
!
. (18.29)
85
We leave it to the industrious student to verify that
F
1
(X, Y ) = X + Y, F
2
(X, Y ) =
1
2
[X, Y ], F
3
(X, Y ) =
1
12
(adX)
2
Y + (adY )
2
X
¸
,
F
4
(X, Y ) = −
1
24
adXadY adX(Y ),
F
5
(X, Y ) = −
1
720
(adX)
4
Y + (adY )
4
X
¸
+
1
360
adX(adY )
3
X + adY (adX)
3
Y
¸
−
1
120
adXadY (adX)
2
Y + adY adX(adY )
3
X
¸
. (18.30)
After this intermezzo we return to the issue of constructing nonAbelian gauge theories.
The simplest way is by generalising ﬁrst the covariant derivative. U(1) gauge transformations
act on a complex ﬁeld as in eq. (18.1), and the covariant derivative is designed such that
D
µ
Ψ(x) →g(x)D
µ
Ψ(x) . (18.31)
Since the gauge ﬁeld transforms as in eq. (17.3), this is easily seen to imply that the covariant
derivative is deﬁned as in eq. (17.2) (these formula are of course also valid for complex scalar
ﬁelds, compare eq. (3.36)). For a nonAbelian gauge theory we consider ﬁrst a ﬁeld Ψ that
transforms as an irreducible representation (i.e. there is no nontrivial linear subspace that
is left invariant under the action of all gauge group elements)
Ψ →
g
Ψ ≡ ρ(g)Ψ . (18.32)
In the following, as in the literature, we shall no longer make a distinction between g and
ρ(g). It will always be clear from the context what is intended. The vector potential should
now be an element of the Liealgebra L
G
(more precisely a representation thereof)
A
µ
= A
a
µ
T
a
. (18.33)
For U(1), which is onedimensional, we need to deﬁne T
1
≡ i. The Liealgebra of the group
consisting of the unimodular complex numbers is the set of imaginary numbers, L
U(1)
= iIR.
Note that as an exception this generator is normalised diﬀerent from eq. (18.20), so as not
to introduce unconventional normalisations elsewhere. The real valued vector potential A
µ
will now be denoted by A
1
µ
and we see that under a gauge transformation
A
µ
→
g
A
µ
= gA
µ
g
−1
−q
−1
(∂
µ
g)g
−1
= gA
µ
g
−1
+ q
−1
g∂
µ
(g
−1
) . (18.34)
This is the form that generalises directly to the nonAbelian gauge groups with the covariant
derivative deﬁned by
D
µ
Ψ = (∂
µ
+ qA
µ
) Ψ , (18.35)
where A
µ
≡ A
a
µ
T
a
ρ
is a matrix acting on the ﬁelds Ψ. We leave it as an exercise to verify
that under a gauge transformation, eq. (18.31) remains valid for the nonAbelian case.
It is now a trivial matter to construct a Lagrangian that is invariant under local gauge
transformation. Assuming the representation is unitary, for a scalar ﬁeld Ψ one has
L
Ψ
= (D
µ
Ψ)
†
D
µ
Ψ−m
2
Ψ
†
Ψ , (18.36)
whereas if Ψ is a Dirac ﬁeld, carrying both spinor (representation of the Lorentz group) and
group indices, one has
L
Ψ
= Ψ(iγ
µ
D
µ
−m) Ψ , Ψ = Ψ
†
γ
0
, (18.37)
86
where Ψ
†
is the hermitian conjugate both with respect to the spinor and the group (repre
sentation) indices.
The part of the Lagrangian that describes the self interactions of the vector ﬁeld A
µ
has to be invariant under local gauge transformations too. In that respect U(1) or Abelian
gauge theories are special, since the homogeneous part of the transformation of the vector
potential is trivial, gA
µ
g
−1
= A
µ
. For nonAbelian gauge transformations this is no longer
true. For U(1) one easily veriﬁes that
D
µ
D
ν
Ψ−D
ν
D
µ
Ψ ≡ [D
µ
, D
ν
]Ψ = iqF
1
µν
Ψ , (18.38)
where F
1
µν
= ∂
µ
A
1
ν
−∂
ν
A
1
µ
is the electromagnetic ﬁeld strength, compare eq. (3.27). Because
the covariant derivative transforms in a simple way under gauge transformations, this formula
can be directly generalised to nonAbelian gauge theories
F
µν
≡ q
−1
[D
µ
, D
ν
]
g
→gF
µν
g
−1
. (18.39)
For U(1), where g is a number, this means that the ﬁeld strength is gauge invariant, as was
noted before. For nonAbelian gauge theories the ﬁeld strength itself is not gauge invariant.
Nevertheless, it is simple to construct a gauge invariant action for the gauge ﬁeld
L
A
=
1
2
Tr (F
µν
F
µν
) ≡ −
1
4
F
a
µν
F
µν
a
, (18.40)
where F
a
µν
are the components of the ﬁeld strength with respect to the Liealgebra basis,
F
µν
≡ F
a
µν
T
a
=
∂
µ
A
a
ν
−∂
ν
A
a
µ
+ qf
abc
A
b
µ
A
c
ν
T
a
. (18.41)
We see from L
A
and L
Ψ
that q plays the role of an expansion parameter. For q = 0 we
have n = dim(G) noninteracting photon ﬁelds. They couple with strength q to the scalar
or Dirac ﬁelds. For nonAbelian gauge theories, in addition the vector ﬁeld couples to itself.
These selfinteractions guarantee that there is invariance under the gauge group G, which
is much bigger than U(1)
n
, which is the symmetry that seems implied by the q = 0 limit.
The nonAbelian gauge invariance ﬁxes the “charges” of the ﬁelds with respect to each of
these U(1) gauge factors. Without the nonAbelian gauge symmetry there would have been
n independent “charges”.
The Lagrangian L
A
is the one that was discovered in 1954 by C.N. Yang and R.L. Mills.
The EulerLagrange equations for the Lagrangian L
A
are called the YangMills equations.
One easily shows that
∂
µ
F
µν
a
+ qf
abc
A
b
µ
F
µν
c
= 0 or [D
µ
, F
µν
] ≡ ∂
µ
F
µν
+ q[A
µ
, F
µν
] = 0 . (18.42)
For the coupling to fermions we read oﬀ from eq. (18.37) what the current for the YangMills
ﬁeld is
L
Ψ
= Ψ(iγ
µ
D
µ
−m)Ψ = Ψ(iγ
µ
∂
µ
−m)Ψ +iqA
a
µ
Ψγ
µ
T
a
Ψ . (18.43)
The current is therefore given by
J
a
µ
≡ −iqΨγ
µ
T
a
Ψ . (18.44)
The coupled YangMills equations read
∂
µ
F
µν
a
+ qf
abc
A
b
µ
F
µν
c
= J
ν
a
. (18.45)
In problem 31 it will be shown that the current is not gauge invariant, unlike for Abelian
gauge symmetries. Closely related is the fact that it is no longer true that the current is
conserved, i.e. ∂
µ
J
a
µ
= 0. Instead, it will be shown in problem 31 that ∂
µ
J
a
µ
+qf
abc
A
µ
b
J
c
µ
= 0.
87
19 The Higgs mechanism
We have seen in problem 30 that the fourFermi interaction in good approximation can be
written in terms of the exchange of a heavy vector particle. In lowest order we have resp.
the diagrams in ﬁg. 12a and ﬁg. 12b
»
´
(a)
»
´
.
.
.
.
M
(b)
ﬁg. 12
The ﬁrst diagram comes from a four fermion interaction term that can be written in terms
of the product of two currents J
µ
J
µ
, where J
µ
= Ψγ
µ
Ψ. Here each fermion line typically
carries its own ﬂavour index, which was suppressed for simplicity. Fig. 12b can be seen to
eﬀectively correspond to
−
˜
J
µ
(−k)
g
µν
−k
µ
k
ν
/M
2
k
2
−M
2
+ iε
˜
J
ν
(k) (19.1)
At values of the exchanged momentum k
2
≪ M
2
, one will not see a diﬀerence between
these two processes, provided the coupling constant for the fourFermi interactions (ﬁg. 12a)
is chosen suitably (see problem 30). This is because for small k
2
, the propagator can be
replaced by g
µν
/M
2
, which indeed converts eq. (19.1) to J
µ
J
µ
/M
2
. It shows that the four
Fermi coupling constant is proportional to M
−2
, such that its weakness is explained by the
heavy mass of the vector particle that mediates the interactions. Examples of fourFermi
interactions occur in the theory of βdecay, for example the decay of a neutron into a proton,
an electron and an antineutrino. In that case the current also contains a γ
5
(problem 40).
It turns out that the fourFermi theory cannot be renormalised. Its quantum corrections
give rise to an inﬁnite number of divergent terms, that can not be reabsorbed in a redeﬁnition
of a Lagrangian with a ﬁnite number of interactions. With the interaction resolved at higher
energies by the exchange of a massive vector particle, the situation is considerably better.
But it becomes crucial for the currents in question to be conserved, such that the k
µ
k
ν
part in the propagator has no eﬀect. It would give rise to violations of unitarity in the
scattering matrix at high energies (the σ ﬁeld deﬁned in eq. (16.15) has the wrong sign for
its kinetic part). To enforce current conservation, we typically use gauge invariance. But
gauge invariance would protect the vector particle from having a mass. The big puzzle
therefore was how to describe a massive vector particle that is nevertheless associated to the
vector potential of a gauge ﬁeld.
The answer can be found in the theory of superconductivity, which prevents magnetic
ﬁeld lines to penetrate in a superconducting sample. If there is, however, penetration in the
form of a quantised ﬂux tube, the magnetic ﬁeld decays exponentially outside the ﬂux tube.
This would indicate a mass term for the electromagnetic ﬁeld within the superconductor.
The LandauGinzburg theory that gives an eﬀective description of this phenomenon (the
microscopic description being given by the BCS theory of Cooper pairs) precisely coincides
with scalar quantum electrodynamics.
L = −
1
4
F
µν
F
µν
+ (D
µ
ϕ)
∗
D
µ
ϕ −κϕ
∗
ϕ −
1
4
λ (ϕ
∗
ϕ)
2
. (19.2)
In the LandauGinzburg theory ϕ describes the Cooper pairs. It is also called the order
parameter of the BCS theory. In usual scalar quantum electrodynamics we would put κ =
m
2
, where m is the mass of the charged scalar ﬁeld. But in the LandauGinzburg theory of
88
superconductivity, it happens to be the case that κ is negative. In that case the potential
V (ϕ) for the scalar ﬁeld has the shape of a Mexican hat.
ﬁg. 13 ﬁg. 13
V
Re ϕ
Imϕ
The minimum of the potential is no longer at ϕ = 0, but at ϕ
∗
ϕ = −2κ/λ, and is
independent of the phase of ϕ. To ﬁnd the physical excitations of this theory we have to
expand around the minimum. With a global phase rotation we can chose the point to expand
around to be real,
ϕ
0
=
−2κ/λ . (19.3)
But this immediately implies that the terms quadratic in the gauge ﬁeld give rise to a mass
term for the photon ﬁeld
D
µ
ϕ
0

2
= e
2
ϕ
2
0
A
µ
A
µ
≡
1
2
M
2
A
µ
A
µ
, M = 2e
−κ/λ . (19.4)
Furthermore, from the degeneracy of the minimum of the potential it follows that the ﬂuc
tuation along that minimum (the phase in ϕ = ϕ
0
exp(iχ)) has no mass (this is related to
the famous Goldstone theorem, which states that if choosing a minimum of the potential
would break the symmetry, called spontaneous symmetry breaking, there is always a mass
less particle). However, this phase χ is precisely related to the gauge invariance, and can
be rotated away by a gauge transformation. On the one hand χ corresponds to a massless
excitation, on the other hand it is the unphysical longitudinal component of the gauge ﬁeld.
But the photon became massive, and has to develop an additional physical polarisation,
which is precisely the longitudinal component. In a prosaic way one states that the massless
excitation (called a wouldbe Goldstone boson) was “eaten” by the longitudinal component
of the photon, which in the process got a mass (“got fat”).
This means we have four massive degrees of freedom, three for the massive vector particle
and one for the absolute value of the complex scalar ﬁeld (its mass is determined by the
quadratic part of the potential in the radial direction at ϕ = ϕ
0
). This is exactly the same
number as for ordinary scalar electrodynamics where κ > 0, because in that case the massless
photon has only two degrees of freedom, whereas the complex scalar ﬁeld represents two
massive real scalar ﬁelds. It looks, however, that there is a discontinuity in the description
of these degrees of freedom, when approaching κ = 0. But the interpretation of the phase
of the complex ﬁeld as a longitudinal component of the vector ﬁeld is simply a matter of
choosing a particular gauge. To count the number of degrees of freedom we implicitly made
two diﬀerent gauge choices
κ > 0 : ∂
µ
A
µ
= 0, Lorentz gauge ,
κ < 0 : Im ϕ = 0, Unitary gauge . (19.5)
89
There is a gauge, called the ’t Hooft gauge, that interpolates between these two gauges
F ≡ ∂
µ
A
µ
−2ieξϕ
0
Im ϕ = 0, ’t Hooft gauge . (19.6)
Rather than adding to the Lagrangian the gauge ﬁxing term L
gf
= −
1
2
α(∂
µ
A
µ
)
2
, one adds
L
gf
= −
1
2
αF
2
. At ξ = 0 this corresponds to the Lorentz gauge, at ξ = ∞ to the unitary
gauge. For the choice ’t Hooft made (ξ = 1/α) the terms that mix (ϕ − ϕ
0
) and A
µ
at
quadratic order disappear and one easily reads of the masses. Gauge ﬁxing will be discussed
in the next section, where it will be shown how extra unphysical degrees of freedom appear in
the path integral, so as to cancel the unphysical components of the gauge and scalar ﬁelds.
The scalar ﬁeld, whose interactions give the gauge ﬁeld a mass, is called the Higgs ﬁeld.
Problems 34 and 35 discuss the Higgs mechanism in detail for the GeorgiGlashow model,
which is a nonAbelian gauge theory with gauge group SO(3), coupled to an SO(3)vector
of scalar ﬁelds ϕ
a
.
20 Gauge ﬁxing and ghosts
The quantisation of gauge theories in the path integral formalism requires more discussion,
since the gauge condition (like the Lorentz gauge ∂
µ
A
µ
= 0) seems to remove only one degree
of freedom of the two that are eliminated in the Hamiltonian formulation (see section 16).
From a simple example it is easily demonstrated what the eﬀect of gauge ﬁxing on a (path)
integral is. For this we take f(x) to be a function on IR
3
, which is invariant under rotations
around the origin, such that it is a function f(r) of the radius r = x only. The symmetry
group is hence SO(3) and we can attempt to compute the integral
d
3
x f(x) by introducing
a “gauge” ﬁxing condition like x
2
= x
3
= 0. But it is clear that
d
3
x f(x) =
d
3
x δ(x
2
)δ(x
3
)f(x) =
dx
1
f(x
1
) . (20.1)
We know very well that we need a Jacobian factor for the radial integral
d
3
x f(x) = 4π
∞
0
r
2
dr f(r) . (20.2)
This Jacobian, arising in the change of variable to the invariant radial coordinates and
the angular coordinates, can be properly incorporated following the method introduced by
Faddeev and Popov. The starting point is a straightforward generalisation of the identity
dx f
′
(x)δ(f(x)) = 1, assuming the equation f(x) = 0 to have precisely one solution (in a
sense the righthand side of the equation counts the number of zeros). It reads
1 =
Dg  det (M(
g
A)) δ (F(
g
A)) , (20.3)
where F(A) ∈ L
G
is the gauge ﬁxing function (with the gauge condition F(A) = 0, e.g.
F(A) = ∂
µ
A
µ
= 0). The gauge transformation g of the gauge ﬁeld A is indicated by
g
A, see
eq. (18.34). Furthermore, M(A) : L
G
→L
G
plays the role of the Jacobian,
M(
g
A) =
∂F(
g
A)
∂g
≡
∂F(
e
X
g
A)
∂X
at X = 0 . (20.4)
Equivalently, with respect to the Liealgebra basis, where F(A) ≡ F
a
(A)T
a
, one has
M
ab
(A) =
dF
a
(
exp(tT
b
)
A)
dt
at t = 0 . (20.5)
90
To relate this to the previous equation, one makes use of the fact that
h
(
g
A) =
(hg)
A , (20.6)
which states that two successive gauge transformations, g and h, give the same result as a
single gauge transformation with hg.
As an example we consider the Lorentz gauge, with F(A) = ∂
µ
A
µ
, for which
F(
exp(X)
A) −F(A) = −q
−1
∂
µ
D
µ
ad
(A)(X) +O(X
2
) , (20.7)
where D
µ
ad
(A) is the covariant derivative in the adjoint representation
D
µ
ad
(A)(X) ≡ ∂
µ
X + q[A
µ
, X] . (20.8)
With respect to the Liealgebra basis this gives
qM
ab
(A) = −δ
ab
∂
µ
∂
µ
+ qf
abc
(∂
µ
A
c
µ
+ A
c
µ
∂
µ
) . (20.9)
For an Abelian gauge theory the structure constants f
abc
vanish and M(A) becomes in
dependent of the gauge ﬁeld. This means that det(M(A)) can be absorbed in an overall
normalisation of the pathintegral. For nonAbelian gauge theories this is no longer possible.
Before describing how the A dependence of det(M(A)) is incorporated, it is important to
note that we assumed the gauge condition F(
g
A) = 0 to have precisely one solution, which
can be arranged with the help of eq. (20.6) to occur at g = 1, in which case A is said to
satisfy the gauge condition. This is in general not correct, as was discovered by Gribov.
Even in our simple problem on IR
3
, the gauge condition x
2
= x
3
= 0 does not uniquely
specify the gauge, because we can go from (r, 0, 0) to (−r, 0, 0) through a rotation over 180
degrees. We have to introduce a further restriction to get the identity
d
3
x f(x) = 4π
d
3
x x
2
1
δ(x
2
)δ(x
3
)θ(x
1
)f(x) , (20.10)
where θ(x) = 0 for x < 0 and θ(x) = 1 for x ≥ 0. In perturbation theory only the gauge
ﬁelds near the origin in ﬁeld space are relevant, and gauge conditions are chosen so as to
avoid this problem in a small neighbourhood of the origin. The Lorentz gauge is such a
gauge condition, and the gauge ﬁxing or Gribov ambiguity is not an issue for computing
quantities in perturbation theory in q.
We still need to deﬁne what we mean with Dg in eq. (20.3). It stands for the integration
measure
¸
x
dg(x), with dg(x) for every x deﬁned as the socalled Haarmeasure on the group.
It is best described in the example of SU(2), which as a space is isomorphic with S
3
. When
S
3
is embedded in IR
4
as a unit sphere, n
2
µ
= 1, it is not too diﬃcult to see that g = n
4
+iσ
k
n
k
gives an element of SU(2), whereas exp(iχs
k
σ
k
) = cos(χ) +i sin(χ)s
k
σ
k
, with s
2
k
= 1, shows
that any element of SU(2) can be written in terms of n
µ
. The Haarmeasure coincides with
the standard integration measure on S
3
,
d
4
n δ(n
2
µ
− 1). The Haarmeasure is in general
invariant under the change of variables g →hg and g →gh, for h some ﬁxed group element.
We can insert eq. (20.3) in the path integral to obtain
Z =
DA
µ
Dg det (M(
g
A)) δ (F(
g
A)) exp (iS(A)) . (20.11)
We now use that the action S(A) is invariant under gauge transformations. We leave it as
an exercise to verify that likewise DA
µ
is invariant under the change of variables A →
g
A,
which trivially implies that
Z =
DA
µ
Dg det (M(A)) δ (F(A)) exp (iS(A)) . (20.12)
91
The dependence of the integrand on g has disappeared, and the integration over g gives an
overall (inﬁnite) normalisation factor, which is irrelevant. We next note that Z has to be
independent of the gauge ﬁxing function F, in particular F(A) − Y is just as good for the
gauge ﬁxing (provided of course we show that F(
g
A) = Y has a solution). This modiﬁcation
does not aﬀect the socalled FaddeevPopov operator M(A) and we ﬁnd
Z =
DA
µ
det (M(A)) δ (F(A) −Y ) exp (iS(A)) , (20.13)
independent of Y . Suitably normalising DY we can deﬁne
DY exp
−
αi
2
d
4
x Y
2
a
(x)
= 1 , (20.14)
which combined with the previous equation gives
Z =
DY DA
µ
det (M(A)) δ(F(A) −Y ) exp
i
d
4
x L(A) −
α
2
Y
2
a
(x)
=
DA
µ
det (M(A)) exp
i
d
4
x L(A) −
α
2
F
2
a
(A)
. (20.15)
For U(1) gauge theories with F(A) = ∂
µ
A
µ
this precisely reproduces the action of eq. (4.22)
in the Lorentz gauge, and in that case det(M(A)) is a constant.
For nonAbelian gauge theories we are left with the task of computing det(M(A)) for
each A, which is no longer constant. But here the path integral over Grassmann variables
comes to the rescue. In problem 25 we have seen that
D¯ ηDη exp
i
d
4
x ¯ η
a
(x)qM
ab
(A)η
b
(x)
= det(M(A)) , (20.16)
up to an overall normalisation. This implies that the path integral can also be written in
the Lorentz gauge as
Z =
DA
µ
D¯ ηDη exp
i
d
4
x L(A)−
α
2
(∂
µ
A
µ
a
(x))
2
+ ∂
µ
¯ η
a
(x)[∂
µ
η
a
(x) + qf
abc
A
b
µ
(x)η
c
(x)]
.
(20.17)
Since ¯ η and η are auxiliary ﬁelds, they should never appear as external lines. They are
therefore called ghosts. Ghosts can only appear in loops and because of the fermionic nature
of the ghost variables, every such loop gives a minus sign. The Feynman rules for the Lorentz
gauge are given in the following table.
table 7
no external ghost lines
k
a
b
≡ (qM)
−1
(A=0)
ab
=
δ
ab
k
2
+iε
ghost propagator (Lorentz gauge)
.
.
c, k
3
µ
`
a, k
1
b, k
2
≡ −iqf
abc
k
µ
1
ghost vertex (Lorentz gauge)
−1 ×i
d
4
k
(2π)
4
loop factor
92
Because one can easily derive that for a complex scalar ﬁeld (up to an overall constant)
Dϕ
∗
Dϕ exp
i
d
4
x ϕ
∗
a
(x)qM
ab
(A)ϕ
b
(x)
=
1
det(M(A))
, (20.18)
we can view a ghost as the elimination of a complex degree of freedom. It is in this way that
in the path integral the two unphysical degrees of freedom of a Lorentz vector are eliminated.
For QED both the ghost and the unphysical degrees of freedom have no interactions, and can
not appear as external lines either, which is why in QED the introduction of ghosts was never
necessary for a consistent description of the theory. For nonAbelian gauge theories, because
of the interaction of the ghost with the gauge ﬁeld, ghosts can no longer be ignored. To
have the ghosts eliminate the unphysical degrees of freedom, one should have the “masses”
(poles) of the ghosts to coincide with the “masses” of the unphysical degrees of freedom.
Furthermore the couplings of the ghost and unphysical ﬁelds to the physical ﬁelds should be
related. This is veriﬁed explicitly for the GeorgiGlashow model in problem 35. In general it
is guaranteed by the existence of an extra symmetry, discovered by Becchi, Rouet and Stora,
called the BRS symmetry s, which for example acts on the gauge ﬁeld as follows
sA
µ
= D
µ
ad
η . (20.19)
This is precisely an inﬁnitesimal gauge transformation. For more details see Itzykson and
Zuber, section 12.41.
21 The Standard Model
The standard model describes the electromagnetic and weak interactions, uniﬁed in the so
called electroweak theory of GlashowWeinbergSalam with the gauge group U(1)×SU(2)
and the strong interactions, known as Quantum Chromodynamics (QCD) with gauge group
SU(3). Theory and experiment, where tested, agree very well up to about 100 GeV, the
energies reached by presentday accelerators. Now the top quark has been found, at a mass
of 174 GeV, only the Higgs particle remains to be detected. Its mass should be smaller
than 1000 GeV (i.e. 1 TeV=Terra electronvolt) according to presentday theoretical insight.
Gravitation has been left out so far. Its natural scale in energy where quantum eﬀects would
become important is the Planck energy, E
pl
=
¯ hc
5
/G ≈ 10
19
GeV. It is very well possible
that a number of the fundamental parameters in the standard model will be determined,
either directly or indirectly, by gravitational interactions. The standard model should then
be considered as an eﬀective ﬁeld theory. The theory for which the standard model describes
its eﬀective lowenergy behaviour is called a uniﬁed theory. An intermediate stage, which
does not yet include gravity is a socalled Grand Uniﬁed Theory (GUT). The simplest ver
sion uniﬁes the electroweak and strong interactions using a gauge group SU(5) (which has
U(1)×SU(2)×SU(3) as a subgroup), thereby reducing the number of free parameters consid
erably. These GUT’s predict proton decay, albeit at the tremendously low rate of one decay
in every 10
30−31
years. Nevertheless, a swimming pool of (10 m)
3
contains enough protons to
verify that the proton decay is slower than can be comfortably accommodated by GUT’s.
Candidates that unify the standard model with gravity in the form of string theories and
supergravity have been unable to provide predictions that either rule them out experimen
tally or provide evidence in favour of these theories. Much is therefore still to be discovered,
in particular because theoretical insight of the last ten years has shown that a Higgs ﬁeld is
most likely not fundamental, although it is not yet ruled out that it will show its structure
93
only at Planck energies. If that is the case the mass of the Higgs should, however, not be
much bigger than 100 GeV.
The standard model consists of gauge ﬁelds B
µ
(for U(1)), W
a
µ
(for SU(2)) and A
a
µ
(for
SU(3), where a runs from 1 to 8, to be discussed later). These gauge ﬁelds have interactions
with a Higgs ﬁeld φ ∈
/
C
2
, which transforms under SU(2) as a spin onehalf representation
(i.e. the fundamental representation) with a coupling constant g. Under U(1) this Higgs
ﬁeld transforms with a coupling constant −
1
2
g
′
, whereas it is neutral under SU(3). These
couplings are represented in the covariant derivative
D
µ
φ = ∂
µ
φ −
i
2
g
′
B
µ
φ −
ig
2
3
¸
a=1
W
a
µ
σ
a
φ . (21.1)
The potential for the Higgs ﬁeld causes spontaneous breaking of part of the symmetries
V (φ) =
λ
4
(φ
†
φ −F
2
)
2
= κφ
†
φ +
λ
4
(φ
†
φ)
2
+ const. , (21.2)
where κ ≡ −
1
2
λF
2
. In this case the minimum of the potential, also called the vacuum, is
degenerate on a three dimensional sphere, speciﬁed by φ
†
φ = F
2
(φ ∈
/
C
2
∼ IR
4
), which
would give rise to three massless scalar particles according to the Goldstone theorem, but
all three will be “eaten” by longitudinal components of the gauge ﬁelds to which the Higgs
ﬁeld couples. We note that there are four gauge ﬁeld components, B
µ
and W
a
µ
for a = 1, 2
and 3. Indeed one combination among these four will not have something to “eat” and will
therefore stay massless. It plays the role of the photon ﬁeld as we got to know it in QED.
To see this write
φ =
0
F
+
ϕ
1
ϕ
2
≡ φ
0
+
ϕ
1
ϕ
2
, (21.3)
such that
(D
µ
φ
0
)
†
D
µ
φ
0
=
g
2
F
2
4
[(W
1
µ
)
2
+ (W
2
µ
)
2
] +
F
2
4
[g
′
B
µ
−gW
3
µ
]
2
. (21.4)
Apparently, the vector ﬁelds W
1,2
µ
will have a mass M
W
=
1
2
√
2gF, whereas the linear
combination
Z
µ
=
g
′
B
µ
−gW
3
µ
g
2
+ g
′
2
= sin θ
W
B
µ
−cos θ
W
W
3
µ
, tan θ
W
=
g
′
g
, (21.5)
receives a mass m
Z
=
1
2
√
2F(g
2
+g
′
2
)
1
2
= M
W
/ cos θ
W
. The linear combination perpendicular
to Z
µ
,
A
em
µ
= cos θ
W
B
µ
+ sin θ
W
W
3
µ
, (21.6)
remains massless. This gauge ﬁeld deﬁnes a U(1) subgroup of SU(2)×U(1) that leaves
φ
0
invariant. This U(1) subgroup is a combination of the U(1) subgroup of SU(2) gener
ated by exp(iχσ
3
) and the phase rotations exp(iχ) associated with the explicit U(1) group
with B
µ
as its gauge ﬁeld. It is trivial to verify that the product of these group elements,
exp(iχ) exp(iχσ
3
) indeed leaves φ
0
invariant. The gauge symmetry associated to this so
called diagonal U(1) subgroup therefore remains unbroken and corresponds to electromag
netism.
The Higgs ﬁeld has three massless components Re ϕ
1
, Imϕ
1
and Imϕ
2
, all eaten by the
vector particles W and Z, and one massive component η ≡ Re ϕ
2
with a mass
m
η
=
√
−2κ =
√
λF . (21.7)
94
It is this component that is called the Higgs ﬁeld. It does not couple to A
em
µ
, because like
ϕ
0
, also η is not aﬀected by the transformation exp(iχ) exp(iχσ
3
). Alternatively, this can be
seen from the covariant derivative
D
µ
ϕ
1
ϕ
2
=
∂
µ
−
ig
2
(σ
1
W
1
µ
+ σ
2
W
2
µ
) +
ig
2 cos θ
W
(cos
2
θ
W
σ
3
−sin
2
θ
W
)Z
µ
−
ig
2
sin θ
W
(σ
3
+ 1)A
em
µ
¸
ϕ
1
ϕ
2
. (21.8)
Using the fact that
σ
3
+ 1 =
2 0
0 0
(21.9)
it is clear that ϕ
2
has no electric charge, whereas ϕ
1
has a charge q = −g sin θ
W
. As
these are wouldbe Goldstone bosons, “eaten” by the vector ﬁelds, it will turn out that the
combinations W
±
µ
=
1
2
√
2(W
1
µ
∓iW
2
µ
) are charged with an electric charge of ±e, where
e = −g sin θ
W
. (21.10)
As a consequence, the two coupling constants g and g
′
are determined by the electric charge e
and the socalled weak mixing angle θ
W
, also called the Weinberg angle. From experiment it
follows that sin
2
θ
W
≈ 0.23. The Z vector ﬁeld will remain neutral under the electromagnetic
interactions. To verify the charge assignment to the vector ﬁelds, we have to ﬁnd the coupling
of the various ﬁelds to A
em
µ
. For this it is suﬃcient to consider the following part of the
Lagrangian
L
W,B
= −
1
4
∂
µ
W
a
ν
−∂
ν
W
a
µ
+ gε
abc
W
b
µ
W
c
ν
2
−
1
4
(∂
µ
B
ν
−∂
ν
B
µ
)
2
. (21.11)
with the obvious shorthand notations like (F
a
µν
)
2
= F
a
µν
F
µν
a
. After some algebra the above
equation can be rewritten as
L
W,B
= −
1
4
F
em
µν
+ ie(W
+
µ
W
−
ν
−W
+
ν
W
−
µ
)
2
−
1
4
∂
µ
Z
ν
−∂
ν
Z
µ
+ ig cos θ
W
(W
+
µ
W
−
ν
−W
+
ν
W
−
µ
)
2
−
1
2
D
em
µ
W
−
ν
−D
em
ν
W
−
µ
−ig cos θ
W
(Z
µ
W
−
ν
−Z
ν
W
−
µ
)
2
, (21.12)
where we have deﬁned
D
em
µ
= ∂
µ
−ieA
em
µ
, F
em
µν
= ∂
µ
A
em
ν
−∂
ν
A
em
µ
. (21.13)
We immediately readoﬀ that our charge assignments for Z and W
±
have been correct. Note
that the vector ﬁeld W
±
µ
has an extra magnetic moment, because of its coupling to F
em
µν
L
magn.mom.
= −ieF
em
µν
W
+
µ
W
−
ν
, (21.14)
which is a direct consequence of the spin of the vector ﬁeld (the magnetic moment for the
Dirac ﬁeld is discussed in problem 32).
We now introduce the fermions in the standard model. They are arranged according to
families. The ﬁrst family with the smallest masses consists of the electron, the neutrino,
the up and the down quark. Essential in the standard model is that invariance under parity
is broken explicitly by the weak interactions (as has been observed in the beta decay of
95
Cobalt60, see problem 40). This is achieved by coupling the left and righthanded helicity
eigenstates of the fermions diﬀerently to the gauge ﬁelds. It should be stressed that the
standard model does not explain why parity is violated; it was put in “by hand”. For each
fermion we deﬁne
Ψ
L
=
1
2
(1 −γ
5
)Ψ , Ψ
R
=
1
2
(1 +γ
5
)Ψ . (21.15)
The Dirac Lagrangian in terms of these helicity eigenstates can be written as
L
Ψ
= Ψ(iγ
µ
∂
µ
−m)Ψ = Ψ
R
(iγ
µ
∂
µ
)Ψ
R
+ Ψ
L
(iγ
µ
∂
µ
)Ψ
L
−m(Ψ
R
Ψ
L
+ Ψ
L
Ψ
R
) , (21.16)
such that diﬀerent transformation rules for Ψ
R,L
enforce m = 0, i.e. the absence of an
explicit mass term. The beauty of the Higgs mechanism is that it also provides a mass for
the fermions. This is achieved by coupling the scalar ﬁeld φ to the fermions, using a Yukawa
coupling
L
Ψ,φ
= −y(Ψ
R
φ
†
Ψ
L
+ Ψ
L
φΨ
R
) , (21.17)
where y is the Yukawa coupling constant. It also immediately ﬁxes the representation to
which Ψ
R,L
should belong. Since the Lagrangian has to be invariant with respect to the
gauge symmetries, and since the scalar ﬁeld is in the fundamental representation of SU(2),
we require that Ψ
L
is also in the fundamental representation, i.e. it is a doublet. On the other
hand Ψ
R
is taken to be invariant under SU(2) (also called the singlet representation). The
couplings of the fermions to the gauge ﬁeld B
µ
have to be chosen such that the Lagrangian
is neutral. This coupling is parametrised by the socalled hypercharge Y , in units of −
1
2
g
′
.
Y
H
≡ Y (φ) = 1 , Y
R
= Y
L
−1 , Y
R,L
≡ Y (Ψ
R,L
) . (21.18)
The mass of the fermions is now readoﬀ from eq. (21.17) by replacing φ with its socalled
vacuum expectation value φ
0
L
Ψ,φ
= −yF(Ψ
R
Ψ
L
2
+ Ψ
L
2
Ψ
R
) , (21.19)
where the index on Ψ
L
indicates the socalled isospin index, the spinor index of the two
dimensional fundamental representation for the internal SU(2) symmetry group. We also
see that Ψ
L
1
remains massless and this is exactly the neutrino. The electron is identiﬁed
with the pair (Ψ
R
, Ψ
L
2
) and has a mass m
e
= yF. We want the neutrino to have no electric
charge and this ﬁxes the hypercharge of Ψ
L
. It is most easily determined from the covariant
derivative, acting on the lefthanded fermion, deﬁned as in eqs. (21.1) and (21.8), since both
are in the same representation (electron and neutrino are also neutral with respect to the
strong interactions. The situation for the quarks will be discussed below).
D
µ
Ψ
L
=
∂
µ
−
ig
2
(σ
1
W
1
µ
+ σ
2
W
2
µ
) +
ig
2 cos θ
W
(cos
2
θ
W
σ
3
−Y
L
sin
2
θ
W
)Z
µ
−
ig sin θ
W
2
(σ
3
+ Y
L
)A
em
µ
¸
Ψ
L
. (21.20)
To make Ψ
L
1
decouple from the electromagnetic ﬁeld we require
Y
L
= −1 , Y
R
= −2 . (21.21)
This also allows us to ﬁnd the electric charge of Ψ
L
2
to be g sin θ
W
= −e, which as it should
be, coincides with the electron charge. The righthanded component should of course have
96
the same electric charge. In that case the covariant derivative is given by
D
µ
Ψ
R
= (∂
µ
−
ig
′
2
Y
R
B
µ
)Ψ
R
= (∂
µ
−
ig
′
2
Y
R
[sin θ
W
Z
µ
+ cos θ
W
A
em
µ
])Ψ
R
= (∂
µ
−ieA
em
µ
−ie tan θ
W
Z
µ
)Ψ
R
, (21.22)
with the expected coupling to the electromagnetic ﬁeld. Note that we can summarise our
assignments of the electric charge by introducing the charge operator in terms of the hyper
charge and the socalled isospin operator I
3
Q
em
= (
1
2
Y + I
3
)e . (21.23)
On a doublet (Ψ
L
and ϕ) one has I
3
=
1
2
σ
3
, whereas I
3
= 0 on a singlet (Ψ
R
).
We now discuss quarks. Also there the weak interactions act diﬀerently on the left and
righthanded components. The lefthanded up and down quarks are combined in a doublet
representation for SU(2). If we denote the quark ﬁelds by q, we assign q
L
1
to the lefthanded
component of the upquark (also denoted by u
L
) and q
L
2
to the lefthanded component of
the downquark (d
L
). This doublet gets a hypercharge Y (q
L
) =
1
3
, from which we read oﬀ
the electric charges
Q
em
q
L
= Q
em
u
L
d
L
=
2e
3
u
L
−e
3
d
L
. (21.24)
The righthanded components of both the up and down quarks are singlets under SU(2) and
their hypercharges are chosen to insure that they have the same electric charge as for their
lefthanded partners
Y (u
R
) =
4
3
and Y (d
R
) = −
2
3
. (21.25)
The quarks transform nontrivially under SU(3), the gauge group of the strong interac
tions. They form complex vectors in the three dimensional deﬁning or fundamental repre
sentation of SU(3). The generators for SU(3) are given by
T
1
= −
i
2
¸
¸
0 1 0
1 0 0
0 0 0
¸
, T
2
= −
i
2
¸
¸
0 −i 0
i 0 0
0 0 0
¸
, T
3
= −
i
2
¸
¸
1 0 0
0 −1 0
0 0 0
¸
,
T
4
= −
i
2
¸
¸
0 0 1
0 0 0
1 0 0
¸
, T
5
= −
i
2
¸
¸
0 0 −i
0 0 0
i 0 0
¸
, T
6
= −
i
2
¸
¸
0 0 0
0 0 1
0 1 0
¸
,
T
7
= −
i
2
¸
¸
0 0 0
0 0 −i
0 i 0
¸
, T
8
= −
i
2
√
3
¸
¸
1 0 0
0 1 0
0 0 −2
¸
, (21.26)
normalised in accordance with eq. (18.20). (In terms of the socalled GellMann matrices
one has T
a
= −iλ
a
/2.) We leave it as an exercise to determine the structure constants. Note
that the Liealgebra for the group SU(N) is given by traceless and antihermitian (X
†
= −X)
complex N×N matrices. The dimension of this Liealgebra is easily seen to be N
2
−1. Note
that det(exp(X)) = exp(Tr(X)), such that exp(X) has determinant one. Also, exp(X)
−1
=
exp(−X) = exp(X
†
) = exp(X)
†
guarantees that exp(X) is a unitary matrix.
The fractional electric charge of the quarks is not observable (otherwise we would have
had a diﬀerent unit for electric charge). The reason is that quarks are conjectured to always
form bound states that are neutral under SU(3). This can be achieved by either taking
97
three quarks in an antisymmetric combination to form a SU(3) singlet, or by combining a
quark and an antiquark. In the ﬁrst case one has a baryon, of which the proton (uud) and
the neutron (udd) are examples. The quarkantiquark bound state is called a meson, of
which the pions are examples (e.g. π
+
= u
¯
d and π
−
= ¯ ud). The bar over the symbol of
a particle of course denotes the antiparticle. Rather prosaically one associates to the three
SU(3) components of the quark ﬁeld the property colour. Choosing the three basic colours
red, blue and green makes a bound state of three quarks in an antisymmetric wave function,
where hence all colours are diﬀerent, into a colourless composite. Similarly, combing a quark
and an antiquark gives a colourless combination. It is not too diﬃcult to show that a bound
state of quarks and antiquarks is a singlet under SU(3) if and only if the net colour is white.
It is now also easily veriﬁed that with the particular fractional electric charges assigned to
the quarks, a colourless combination always has an electric charge that is an integer multiple
of the electron charge. For this note that both quarks have modulo e an electric charge
equal to −
1
3
e, whereas both antiquarks have modulo e a charge of
1
3
e. Three quarks bound
together therefore have zero charge modulo e. The same holds for a quarkantiquark bound
state.
That the strong interactions really are strong, follows from the fact that a quark and
antiquark can not be separated without creating a quarkantiquark pair from the vacuum,
to make sure that the separated components remain neutral under SU(3). This is achieved
by combining the quark (antiquark) of the pair created with the antiquark (quark) we try
to separate. The mechanism that prevents free quarks to appear is called conﬁnement,
which still lacks a solid theoretical understanding. Because the coupling constant is strong,
a perturbative expansion is no longer applicable. That nevertheless the theory of the strong
interactions is believed to be the correct theory to describe the forces amongst the quarks
(and therefore indirectly the nuclear forces) follows from the remarkable property that at
high energies the eﬀective coupling constant is small; at inﬁnite energy even zero. This is
called asymptotic freedom and will only brieﬂy be discussed in the next section. For a more
detailed discussion we refer to Itzykson and Zuber. In the table below we list the gauge
particles of the standard model.
Gauge particles table 8
name charge spin mass force
γ photon 0 1 0 electromagnetism
A gluon 0 1 0 strong force
W
±
Wparticle ±e 1 80 GeV weak force
Z Zparticle 0 1 91 GeV weak force
The strong interactions do not break parity invariance, i.e. the eight gluons A
a
µ
couple to
the lefthanded and righthanded components of the quark ﬁelds in the same way. However,
the socalled CabibboKobayashiMaskawa (CKM) mixing with two other families of quarks
(the strange and charm quark on the one hand and the bottom and top quark on the other
hand) gives in a very subtle way rise to violation of CP, that is the combination of charge
conjugation and parity (equivalent to time reversal T, since CPT is conserved). The electron
and neutrino, called leptons, in the ﬁrst family are replaced by the muon and its neutrino
for the second family and by the tau and associated neutrino for the third family. The
98
experiments described in the introduction (see problem 37) have shown that there are not
more than three of these families with massless neutrinos. In the standard model there is
room to add a righthanded partner for the neutrino ﬁeld, which couples to none of the gauge
ﬁelds. With a suitably chosen Yukawa coupling the neutrino can be given an arbitrarily small
mass. It is experimentally very hard to measure the mass of the neutrino; only upper bounds
have been established. The table below lists the properties of all the fermions observed in the
standard model (the top quark was only discovered in 1994 at Fermilab). For much more on
the standard model see in particular the book by J.C. Taylor mentioned in the introduction.
Fermion families table 9
name charge spin mass
d down quark e/3 1/2 10 MeV
u up quark 2e/3 1/2 5 MeV
e electron e 1/2 511 keV
ν
e
neutrino 0 1/2 0(<10 eV)
s strange quark e/3 1/2 250 MeV
c charm quark 2e/3 1/2 1,5 GeV
µ muon e 1/2 106 MeV
ν
µ
muonneutrino 0 1/2 0(<0,5 MeV)
b bottom quark e/3 1/2 4,8 GeV
t top quark 2e/3 1/2 174 GeV
τ tau e 1/2 1,8 GeV
ν
τ
tauneutrino 0 1/2 0(<164 MeV)
22 Loop corrections and renormalisation
Up to now, we have only considered the lowest order calculations of cross sections, for which
it is suﬃcient to consider treelevel diagrams that do not contain any loops. Loop integrals
typically give rise to inﬁnities, which can be regularised by considering for example a cutoﬀ in
momentum space, as was discussed in sect. 7. Another possibility of regularising the theory is
by discretising spacetime, amounting to a lattice formulation, see eq. (7.5). In both of these
cases there exists a maximal energy (equivalent to a minimal distance). The parameters,
like the coupling constants, masses and ﬁeld renormalisation constants will depend on this
cutoﬀ parameter, generically denoted by an energy Λ or a distance a = 1/Λ. How to give a
physical deﬁnition of the mass in terms of the full propagator and why ﬁeld renormalisation
is necessary was discussed in sect. 9. For the renormalisation of the coupling constant it is
best to deﬁne the physical coupling constant in terms of a particular scattering process, as
that is what can be measured in experiment. Alternatively, as these are strongly related,
the physical couplings can be deﬁned in terms of an amputated 1PI npoint function, with
prescribed momenta assigned to the external lines, all proportional to an energy scale called
µ ≪Λ. As an example consider the selfinteracting scalar ﬁeld, with a fourpoint coupling λ
99
(see for example eq. (21.2)). We deﬁne the physical fourpoint coupling constant in terms of
the 1PI fourpoint function with the momenta on the amputated lines set to some particular
value proportional to µ (the precise choice is not important for the present discussion). It
is clear that this gives a function λ
reg
(λ, µ, Λ). The dependence on other coupling constants
and the mass parameters is left implicit.
The theory is considered renormalisable if we can remove the cutoﬀ by adjusting λ
(also called the bare coupling constant) in such a way that at a ﬁxed value µ = µ
0
the
renormalised coupling λ
R
stays ﬁnite and takes on a prescribed (i.e. measured) value. It is
then obvious that the renormalised coupling constant λ
R
(µ) ≡ λ
reg
(λ(Λ), µ, Λ) is a function
of µ, coinciding at µ
0
with the prescribed value λ
R
. Since the physical coupling constant is
computed in terms of the full 1PI fourpoint function, the dependence on the energy scale is
caused by quantum corrections. Since the vacuum in ﬁeld theories is not really empty, as was
discussed in the context of the Casimir eﬀect in sect. 2, the computation is not much diﬀerent
from calculating eﬀective interactions in a polarised medium. In this case the polarisation is
due to the virtual particles that describe the quantum ﬂuctuations (zeropoint ﬂuctuations)
of the vacuum, and is therefore also called the vacuum polarisation. The energy dependent
couplings are called running couplings. It should be emphasised that the running of the
couplings is a manifestation of an anomaly (called the conformal anomaly), which is the
breaking of a symmetry of the Lagrangian by the quantum corrections. In the absence of
a mass, the scalar ﬁeld theory with a ϕ
4
interaction is at the classical level invariant under
scale transformations, ϕ(x) → κϕ(x/κ), where κ is the scale parameter. It is obvious that
the regularised couplings are not invariant under such a rescaling, because of the presence of
a cutoﬀ. What is not obvious is that, for the simple ﬁeld theories we have been considering
in four dimensions, the scale independence can not be recovered by removing the cutoﬀ (i.e.
taking Λ →∞).
By adjusting the bare coupling constants of the theory as a function of the cutoﬀ Λ,
to ensure that all regularised couplings stay ﬁnite when the cutoﬀ is moved to inﬁnity, the
calculations can be arranged such that nowhere explicit inﬁnities occur. When we say that
the contributions of the loops diverge, we mean that without adjusting the bare coupling
constant, their contributions are inﬁnite in the limit Λ →∞. A theory is called renormalis
able if only a ﬁnite number of bare coupling constants needs to be adjusted to have all 1PI
npoint functions ﬁnite. This can be shown to be equivalent to all 1PI npoint functions to
be completely determined as a function of a ﬁnite number of renormalised couplings, called
relevant couplings. It is only in such an instance that quantum ﬁeld theory has predictive
power. Renormalisability is therefore a necessary requirement for the theory to be insensitive
to what happens at very high energies with a maximal amount of predictive power. The
standard model falls in this class of theories.
Theoretical studies of the last ﬁve years or so have shown that the selfcoupling of the
Higgs ﬁeld will most likely vanish if we really take Λ → ∞, albeit in a logarithmic way.
Losely speaking the running of this coupling is such that the renormalised coupling increases
with increasing energy. The only way it can be avoided that the renormalised coupling will
become inﬁnite at some ﬁnite energy, is to either take the renormalised coupling equal to
zero or to keep the cutoﬀ ﬁnite. It depends on the parameters of the model, in particular
the Higgs mass, how large the cutoﬀ should be. If the Higgs mass is relatively light, this can
be at the Planck scale and has little consequence for the theory. If, however, the Higgs mass
turns out to be in the order of 1 TeV, the cutoﬀ has to be roughly smaller than 10 TeV.
As we can measure the selfcoupling of the Higgs ﬁeld and related quantities to be non
trivial (which is of course crucial for the spontaneous symmetry breaking and giving a mass
100
to the W and Z particles), the scalar sector of the standard model depends in a rather subtle
way on what happens at higher energies. This sensitivity to high energies is, however, much
and much weaker than in nonrenormalisable theories like for the fourfermi interactions.
It is outside the scope of these lectures to describe the computations necessary to make
the above more precise. In the following we give a sample calculation that will provide the
technical ingredients to perform such calculations and to illustrate some of these issues in a
simple setting. Also, problems 2(!), 38 and 39 illustrate further ingredients that are pertinent
to renormalising ﬁeld theories.
Let us end this discussion by noting that a running coupling can of course either increase
or decrease at increasing energy. The Higgs selfcoupling and the electromagnetic coupling
constant e are examples of couplings that increase at high energies. For the electric charge e,
this increase is very tiny and the cutoﬀ can be chosen much bigger than the Planck energy.
The analogy with a polarised medium is that the virtual particles in the ﬁeld of a charged
particle will screen its charge at large distances. When we probe the charged particle at
ever smaller distances the eﬀective charge becomes less screened and increases. Due to the
selfinteractions of a nonAbelian gauge ﬁeld, its charges show the eﬀect of antiscreening.
Here the eﬀective charge becomes bigger at larger distances. For the strong interactions this
is one way to understand conﬁnement. The energy of a single quark within a spherical shell
would increase without bound with increasing radius. A free quark would carry an inﬁnite
energy. To the contrary, at decreasing separations, the eﬀective charge becomes weaker and
weaker and the quarks start to behave as free particles. This is the asymptotic freedom
mentioned in the previous section.
We will now consider to oneloop order the selfenergy for the scalar ﬁeld ϕ with a mass
m, coupled to two ﬂavours of fermions with masses m
1
and m
2
, coupled through Yukawa
couplings described by the Lagrangian
L =
1
2
(∂
µ
ϕ)
2
−
1
2
m
2
ϕ
2
−
1
6
λϕ
3
+
¸
i
Ψ
(i)
(iγ
µ
∂
µ
−m
i
)Ψ
(i)
−gϕ
Ψ
(1)
Ψ
(2)
+ Ψ
(2)
Ψ
(1)
. (22.1)
The selfenergy for the scalar ﬁeld in oneloop order splits in two contributions Σ
1
and Σ
2
from a fermion and a scalar loop (in this order Z ≡ 1).
Σ
ϕ
(q) ≡ Σ
1
+ Σ
2
= −
`
,
q q
m
1
, k
m
2
, q+k
+
`
q q
k
q+k
. (22.2)
The numerical expressions for Σ
1
and Σ
2
are given by
Σ
1
= −
ig
2
(2π)
4
d
4
k
Tr ((k/ + m
1
)(k/ + q / + m
2
))
(k
2
−m
2
1
+ iε)((k + q)
2
−m
2
2
+ iε)
,
Σ
2
=
iλ
2
2(2π)
4
d
4
k
1
(k
2
−m
2
+ iε)((k + q)
2
−m
2
+ iε)
. (22.3)
Using eq. (11.1), requires us to employ the Feynman rules of table 3 to obtain these expres
sions. (Alternatively the Feynman rules of table 2 can be used, provided the selfenergy is
deﬁned through eq. (9.9).) These integrals are obviously divergent. Introducing a momen
tum cutoﬀ Λ we ﬁnd Σ
1
∼ Λ
2
and Σ
2
∼ log Λ to lowest nontrivial order in 1/Λ. One says
that Σ
1
is quadratically and Σ
2
logarithmically divergent. To simplify the integrands we
discuss the Feynman trick
1
ab
=
1
0
dx
1
(ax + b(1 −x))
2
, (22.4)
101
which we can apply to the computation of Σ
1
by substituting a = (k + q)
2
− m
2
2
+ iε and
b = k
2
− m
2
1
+ iε. For Σ
2
we have the same assignment, with in addition m
1
= m
2
= m.
It is useful to also have the generalisation of the Feynman trick for an arbitrary product of
scalar propagators.
k
¸
i=1
a
n
i
i
−1
=
Γ(
¸
i
n
i
)
¸
k
i=1
Γ(n
i
)
k
¸
i=1
1
0
x
n
i
−1
i
dx
i
δ
k
¸
i=1
x
i
−1
k
¸
i=1
a
i
x
i
−
¸
i
n
i
, (22.5)
which is proven by induction. In here Γ(z) is the gamma function, which satisﬁes Γ(z +1) =
zΓ(z), Γ(n + 1) = n! and Γ(
1
2
) =
√
π (see problem 2b). Consequently we ﬁnd
Σ
1
= −
4ig
2
(2π)
4
d
4
k
1
0
dx
k
2
+ k · q + m
1
m
2
(k + (1 −x)q)
2
+ x(1 −x)q
2
−xm
2
1
−(1 −x)m
2
2
+ iε
2
,
Σ
2
=
iλ
2
2(2π)
4
d
4
k
1
0
dx
1
(k + (1 −x)q)
2
+ x(1 −x)q
2
−m
2
+ iε
2
. (22.6)
We will show how to regularise these two integrals in two diﬀerent ways. First we use
dimensional regularisation introduced by ’t Hooft and Veltman (see problem 2) and then
discuss PauliVillars regularisation. In dimensional regularisation the loop integrations are
replaced by integrals in n, instead of 4, dimensions. The momentum integrations are always
of the form
I
n,α,β
(M) ≡
d
n
k
(k
2
)
α
(k
2
−M
2
+ iε)
β
. (22.7)
We can evaluate this integral by performing the socalled Wick rotation, where we replace
the integral over Rek
0
by an integration over Imk
0
. The integral over the two quarter circles
indicated in the ﬁgure, will vanish as the radius tends to inﬁnity. As there are no poles inside
the contour of integration, we ﬁnd
·
`
`
·
Rek
0
Imk
0
ﬁg. 14
I
n,α,β
(M) = i(−1)
α−β
d
n
k
(k
2
)
α
(k
2
+ M
2
−iε)
β
, k
2
= k
2
0
+
k
2
. (22.8)
We note that the integrand is a purely radial integral and as the surface area of a n di
mensional sphere is analytically known (S
n
= 2π
n/2
/Γ(n/2), e.g. S
2
= 2π, S
3
= 4π,
S
4
= 2π
2
,· · ·), we obtain
I
n,α,β
(M) = i(−1)
α−β
2π
n/2
Γ(n/2)
r
n−1
dr
r
2α
(r
2
+ M
2
−iε)
β
=
i(−1)
α−β
π
n/2
Γ(α + n/2)Γ(β −α −n/2)
(M
2
−iε)
β−α−n/2
Γ(β)Γ(n/2)
. (22.9)
We used the integral representation of the beta function
B(n, k) ≡
Γ(n)Γ(k)
Γ(n + k)
=
∞
0
dz
z
n−1
(z + 1)
k+n
. (22.10)
102
Shifting the integration variable k →k −(1 −x)q we ﬁnd
Σ
1
= −
4ig
2
(2π)
4
d
n
k
1
0
dx
k
2
−(1 −2x)k · q + m
1
m
2
+ x(x −1)q
2
(k
2
−
ˆ
M
2
x,q
+ iε)
2
,
Σ
2
=
iλ
2
2(2π)
4
d
n
k
1
0
dx
1
(k
2
−M
2
x,q
+ iε)
2
, (22.11)
where
ˆ
M
2
x,q
= xm
2
1
+ (1 −x)m
2
2
−x(1 −x)q
2
and M
2
x,q
= m
2
−x(1 −x)q
2
, (22.12)
which allows us to express Σ
i
in terms of the integrals I
n,α,β
(m)
Σ
1
= −
4ig
2
(2π)
4
1
0
dx
I
n,1,2
(
ˆ
M
x,q
) + (m
1
m
2
−x(1 −x)q
2
)I
n,0,2
(
ˆ
M
x,q
)
¸
,
Σ
2
=
iλ
2
2(2π)
4
1
0
dx I
n,0,2
(M
x,q
) . (22.13)
Substituting the expressions for I
n,α,β
(M) from eq. (22.9), we ﬁnd
Σ
1
=
4g
2
π
n/2
(2π)
4
1
0
dx
Γ(2 −n/2)
(m
1
m
2
−x(1 −x)q
2
)
(
ˆ
M
2
x,q
−iε)
2−n/2
Γ(2)
−
Γ(1 +n/2)Γ(1 −n/2)
(
ˆ
M
2
x,q
−iε)
1−n/2
Γ(2)Γ(n/2)
,
Σ
2
= −
λ
2
π
n/2
2(2π)
4
1
0
dx
Γ(2 −n/2)
(M
2
x,q
−iε)
2−n/2
Γ(2)
. (22.14)
This can be further simpliﬁed using
Γ(1 +n/2)Γ(1 −n/2)
Γ(2 −n/2)Γ(n/2)
=
n
2 −n
, (22.15)
such that
Σ
1
=
4g
2
π
n/2
Γ(2 −n/2)
(2π)
4
1
0
dx
(m
1
m
2
−x(1 −x)q
2
)
(
ˆ
M
2
x,q
−iε)
2−n/2
−
n/(2 −n)
(
ˆ
M
2
x,q
−iε)
1−n/2
,
Σ
2
= −
λ
2
π
n/2
Γ(2 −n/2)
2(2π)
4
1
0
dx
1
(M
2
x,q
−iε)
2−n/2
. (22.16)
The divergent part is now fully contained in Γ(2 −n/2), because
Γ(2 −n/2) =
Γ(1 +
1
2
(4 −n))
1
2
(4 −n)
=
2
4 −n
−γ +O(4 −n) , (22.17)
where γ = 0.57721 · · · is the Euler constant. We expand Σ
i
around n = 4
Σ
1
=
g
2
(2π)
2
2
4 −n
−γ
π
(n−4)/2
1
0
dx
¸
m
1
m
2
−x(1 −x)q
2
+ (2 +
1
2
(4 −n))
ˆ
M
2
x,q
(
ˆ
M
2
x,q
−iε)
(4−n)/2
¸
=
g
2
(2π)
2
2
4 −n
−γ
1
0
dx
m
1
m
2
−3x(1 −x)q
2
+ 2xm
2
1
+ 2(1 −x)m
2
2
×
1 −
1
2
(4 −n) log(π[
ˆ
M
2
x,q
−iε])
+
1
2
(4−n)
ˆ
M
2
x,q
¸
≡
Σ
(−1)
1
4−n
+Σ
(0)
1
+O(4 −n) ,
Σ
2
=
λ
2
8(2π)
2
1
0
dx
log(π[M
2
x,q
−iε]) + γ
−
2
4−n
¸
≡
Σ
(−1)
2
4−n
+Σ
(0)
2
+O(4 −n) .(22.18)
103
Note that q
2
= q
2
0
−q
2
and that the coupling constant λ for the scalar threepoint coupling
has the dimension of mass. We have split the result for Σ
i
in a pole term with residue Σ
(−1)
i
and a ﬁnite part Σ
(0)
i
for n →4.
Σ
(0)
1
=
g
2
4π
2
1
0
dx
ˆ
M
2
x,q
−
log(π[
ˆ
M
2
x,q
−iε]) + γ
×
m
1
m
2
+ 3x(x −1)q
2
+ 2xm
2
1
+ 2(1 −x)m
2
2
,
Σ
(0)
2
=
λ
2
8(2π)
2
1
0
dx
log(π[M
2
x,q
−iε]) + γ
,
Σ
(−1)
1
=
g
2
2π
2
(m
1
m
2
+ m
2
1
+ m
2
2
−
1
2
q
2
) , Σ
(−1)
2
= −
λ
2
(4π)
2
. (22.19)
We now note that the pole terms are of the same form as the treelevel expressions
obtained from the following extra term in the Lagrangian
∆L =
1
2
a(∂
µ
ϕ)
2
−
1
2
bϕ
2
. (22.20)
This means that we can choose a and b so as to precisely cancel the pole terms. To lowest
order we therefore have
Σ
ϕ
(L + ∆L) = Σ
ϕ
(L) + b −aq
2
= Σ
(0)
1
+ Σ
(0)
2
+O(n −4) , (22.21)
from which we can solve for a and b in terms of Σ
(−1)
i
a = −
g
2
(2π)
2
1
4 −n
, b =
λ
2
(4π)
2
−
g
2
2π
2
(m
1
m
2
+ m
2
1
+ m
2
2
)
1
4 −n
. (22.22)
Note that as long as we stay away from n = 4 everything is well deﬁned, including a
and b. The limit n → 4 is to be taken after we have expressed everything in terms of the
renormalised coupling constants and masses. We have taken here a slightly diﬀerent approach
for renormalising the theory. Rather than computing at n = 4 physical processes to ﬁx the
renormalised couplings, we have started with renormalised couplings and determined how
they have to depend on the bare couplings so as to cancel any inﬁnities that might arise as
n →4. It is clear that these two procedures are equivalent. For the physical interpretation
the ﬁrst procedure (due to Wilson) is more transparent, in a loop expansion the second
procedure is more natural. To ﬁnd the bare mass and the ﬁeld renormalisation (for the bare
λ coupling we should have considered the 1PI threepoint function with three ϕ external
lines) we write to oneloop order
L
B
= L + ∆L ≡
1
2
(∂
µ
ϕ
B
)
2
−
1
2
m
2
B
ϕ
2
B
, m
2
B
≡
m
2
+ b
1 +a
, ϕ
B
≡
Z
ϕ
ϕ =
√
1 +aϕ .
(22.23)
Often it can be determined by powercounting (of momenta) which diagrams give rise to
divergencies for n →4 or Λ →∞. The inﬁnities correspond to local counter terms (i.e with
a ﬁnite number of spacetime derivatives) in the Lagrangian. For the theory in eq. (22.1),
power counting easily shows that the ϕ fourpoint function is logarithmically divergent at
oneloop order, see the Feynman diagram below.
`
,
`
·
m
1
m
1
m
2
m
2
ﬁg. 15
104
We therefore need to introduce an independent parameter for the ϕ fourpoint coupling, so
as to adjust its bare coupling to depend in the proper way on the cutoﬀ, to ensure that
we can remove it. It can be shown that after adding to the Lagrangian in eq. (22.1) the
term −λ
4
ϕ
4
/4!, the theory becomes renormalisable to all orders in the loop expansion. The
relevant parameters are m, m
1
, m
2
, g, λ and λ
4
.
As we have seen in section 11, Σ
ϕ
(q) should have a nonvanishing imaginary part if the
scalar particle is unstable. It is clear that the scalar particle itself can not decay in two scalar
particles, but when m
1
+ m
2
< m it could decay in two fermions. Indeed, it is not diﬃcult
to show that on the mass shell (q
2
= m
2
) Σ
(0)
2
is real
Σ
(0)
2
(q
2
= m
2
) =
λ
2
8(2π)
2
1
0
dx log[(x −
1
2
)
2
+
3
4
] + γ + log(m
2
π) ∈ IR . (22.24)
We will show that ImΣ
(0)
1
(q
2
= m
2
) = 0 if and only if q
2
> (m
1
+m
2
)
2
, called the threshold
for decay. The only way Σ
(0)
1
can develop an imaginary part is when the argument of the
logarithm in eq. (22.19) becomes negative. The threshold is therefore determined by
min{
ˆ
M
2
x,q
x ∈ [0, 1]} = min{xm
2
1
+ (1 −x)m
2
2
−x(1 −x)q
2
x ∈ [0, 1]} < 0 . (22.25)
Let us ﬁrst consider the simplest case of equal fermion masses, m
1
= m
2
. In that case
min{
ˆ
M
2
x,q
x ∈ [0, 1]} = m
2
1
−
1
4
q
2
(m
1
= m
2
) , (22.26)
and the threshold is determined by q
2
> 4m
2
1
= (m
1
+m
2
)
2
. For the general case of unequal
fermion masses, the minimum is obtained for x =
1
2
(1 + (m
2
2
−m
2
1
)/q
2
). After some algebra
we ﬁnd
min{
ˆ
M
2
x,q
x ∈ IR} =
q
2
4
1 −
(m
1
−m
2
)
2
q
2
(m
1
+ m
2
)
2
q
2
−1
, (22.27)
which is indeed negative for q
2
> (m
1
+m
2
)
2
. The value of x where this minimum is attained
does not lie in the interval [0, 1] if (m
2
1
− m
2
2
)/q
2
 > 1, which can be used to rule out the
other region, q
2
< (m
1
− m
2
)
2
, where eq. (22.27) is negative. This therefore proves that
the kinematically determined threshold coincides with the threshold for ImΣ(q) = 0, as was
assumed in section 11.
A major advantage of dimensional regularisation is that it preserves the Lorentz and
gauge invariances. Furthermore, it is a local regulator. The lattice regularisation also can
be arranged to preserve the gauge invariance, but locality and Lorentz invariance are only
valid at distances much bigger than the lattice spacing a. A momentum cutoﬀ breaks both
the Lorentz and gauge invariance. PauliVillars regularisation is aimed at having a regulator
that preserves the Lorentz invariance. We will describe it for the Lagrangian of eq. (22.1),
using again the computation of Σ
ϕ
(q) to oneloop order as an illustration. For each of the
original ﬁelds ϕ and Ψ
(i)
one adds extra (ghost) ﬁelds, with either the same (e
ℓ
> 0) or
opposite (e
ℓ
< 0) statistics. This means that a loop of these ghost ﬁelds gets an additional
factor e
ℓ
. Furthermore, the mass of these ghost ﬁelds is shifted over M
ℓ
with respect to the
original (“parent”) ﬁeld (alternatively, if the original ﬁeld is a boson, one can shift m
2
over
M
2
ℓ
, see problem 39. With the present prescription we can treat bosons and fermions on the
same footing). By deﬁning e
0
= 1 and M
0
= 0, the index ℓ = 0 describes the original ﬁelds
of the model. We deﬁne furthermore M
ℓ
≡ b
ℓ
Λ with Λ ≫ m, m
1
, m
2
. To regularise Σ
ϕ
by
PauliVillars’ method, we choose
e
ℓ
= (1, −1, 2, −2) and b
ℓ
= (0, 4, 3, 1) , (22.28)
105
in other words the scalar and two fermion ﬁelds each have three ghost ﬁelds associated to
them but with nonstandard weights for the loops. We could stick to standard weights, such
that these ghost ﬁelds can be described in terms of either grassmann or bosonic variables by
taking e
ℓ
 ﬁelds, having either the same (e
ℓ
> 0) or reversed (e
ℓ
< 0) statistics with respect
to the original (“parent”) ﬁeld. It is straightforward to give the selfenergy including the
contribution of the ghost ﬁelds
Σ
PV
ϕ
(q) =
3
¸
ℓ=0
e
ℓ
Σ
1
(m
1
+ b
ℓ
Λ, m
2
+ b
ℓ
Λ; q) + Σ
2
(m+ b
ℓ
Λ; q)
¸
, (22.29)
in an obvious notation. The weights are chosen such that the momentum integrals can all
be performed. Nevertheless, the masses of the ghost particles, all proportional to Λ, now
play the role of a momentum cutoﬀ, as the Lagrangian will at that energy scale no longer
describe a physical theory. It is still convenient to evaluate the integrals in n, rather than
in 4 dimensions. We will see that eq. (22.28) guarantees that the terms proportional to
(4 −n)
−1
exactly cancel. Indeed, using eq. (22.19)
3
¸
ℓ=0
e
ℓ
Σ
(−1)
1
(m
1
+ b
ℓ
Λ, m
2
+ b
ℓ
Λ; q) + Σ
(−1)
2
(m+ b
ℓ
Λ; q)
= (22.30)
3
¸
ℓ=0
e
ℓ
g
2
2π
2
(m
1
+ b
ℓ
Λ)(m
2
+ b
ℓ
Λ) + (m
1
+ b
ℓ
Λ)
2
+ (m
2
+ b
ℓ
Λ)
2
−
g
2
q
2
(2π)
2
−
λ
2
(4π)
2
=
g
2
(m
1
m
2
+m
2
1
+m
2
2
−
1
2
q
2
)−
1
8
λ
2
2π
2
¸
ℓ
e
ℓ
+
3(m
1
+m
2
)g
2
2π
2
Λ
¸
ℓ
e
ℓ
b
ℓ
+
3g
2
2π
2
Λ
2
¸
ℓ
e
ℓ
b
2
ℓ
= 0 .
The ﬁnite result that remains (replacing Σ
(−1)
in the equation above by Σ
(0)
) is nevertheless
still dependent on Λ. To keep the following computation transparent we take m
1
= m
2
Σ
PV
ϕ
(q)
=
g
2
(2π)
2
1
0
dx
¸
ℓ
e
ℓ
1−3γ−3 log
π[(m
1
+b
ℓ
Λ)
2
−x(1−x)q
2
]
¸
(m
1
+b
ℓ
Λ)
2
−x(1−x)q
2
+
λ
2
8(2π)
2
1
0
dx
¸
ℓ
e
ℓ
log
π[(m+b
ℓ
Λ)
2
−x(1−x)q
2
]
+γ
¸
= −
3g
2
(2π)
2
1
0
dx
¸
ℓ
e
ℓ
log
(m
1
+ b
ℓ
Λ)
2
−x(1 −x)q
2
(m
1
+ b
ℓ
Λ)
2
−x(1 −x)q
2
+
λ
2
8(2π)
2
1
0
dx
¸
ℓ
e
ℓ
log
(m+ b
ℓ
Λ)
2
−x(1 −x)q
2
=
λ
2
8(2π)
2
1
0
dx log
m
2
−x(1−x)q
2
−
3g
2
(2π)
2
1
0
dx log
m
2
1
−x(1−x)q
2
m
2
1
−x(1−x)q
2
+a
1
Λ
2
+ a
2
Λ +a
3
log Λ +a
4
q
2
log Λ +a
5
+ a
6
q
2
+O(1/Λ) . (22.31)
The precise values of the coeﬃcients a
i
are not very important, but can be calculated ex
plicitly with some eﬀort. All Λ dependent terms can be absorbed in a redeﬁnition of Z
ϕ
and
the mass of the scalar ﬁeld, such that
Σ
PV
ϕ
(q) = −
3g
2
(2π)
2
1
0
dx log
m
2
1
−x(1 −x)q
2
m
2
1
−x(1 −x)q
2
+
λ
2
8(2π)
2
log
m
2
−x(1 −x)q
2
+ a
5
+ a
6
q
2
. (22.32)
106
Note that in dimensional regularisation (DR) we found the result Σ
(0)
1
+ Σ
(0)
2
, or
Σ
DR
ϕ
(q) = −
3g
2
(2π)
2
1
0
dx
log
π
m
2
1
−x(1 −x)q
2
+ γ −
1
3
¸
m
2
1
−x(1 −x)q
2
+
λ
2
8(2π)
2
1
0
dx
log
π
m
2
−x(1 −x)q
2
+ γ
¸
. (22.33)
However, the diﬀerence Σ
PV
ϕ
(q) −Σ
DR
ϕ
(q) = a
5
−b
1
+ (a
6
−b
2
)q
2
, where
b
1
=
λ
2
(γ + log π)
8(2π)
2
−
g
2
m
2
1
(3γ + 3 log π −1)
(2π)
2
, b
2
=
g
2
m
2
1
(3γ + 3 log π −1)
6(2π)
2
, (22.34)
can be absorbed in a ﬁnite redeﬁnition of the mass and of Z
ϕ
. If we deﬁne the renormalised
coupling in terms of some physical scattering process, such an ambiguity of course cannot
arise. In that case there is a unique relation between the bare and renormalised parameters.
This relation, however, depends on the regularisation used.
We will now discuss, without a detailed derivation, the renormalisation of gauge theories
to oneloop order in dimensional regularisation. The bare Lagrangian is given by
L = −
1
4
(∂
µ
A
ν
B
−∂
ν
A
µ
B
)
2
−
1
2
α
B
(∂
µ
A
µ
B
)
2
+ Ψ
B
(iγ
µ
∂
µ
−m
B
)Ψ
B
+e
B
A
µ
B
Ψ
B
γ
µ
Ψ
B
. (22.35)
In n dimensions we still want the action to be dimensionless (¯ h = 1), which implies that
L/µ
n
is dimensionless. From this we derive the dimensions of the ﬁelds and the parameters
in n dimensions,
[A
µ
B
] = µ
1
2
n−1
, [α
B
] = 1 , [Ψ
B
] = [Ψ
B
] = µ
1
2
n−
1
2
, [m
B
] = µ, [e
B
] = µ
2−
1
2
n
. (22.36)
If we deﬁne (as is customary) ε ≡ 4 − n, one ﬁnds (for details see Itzykson and Zuber, e.g.
sections 71 and 84. They use slightly diﬀerent notations.)
A
µ
B
≡ µ
−
1
2
ε
Z
A
A
µ
=µ
−
1
2
ε
1 −
e
2
6π
2
·
1
ε
+· · ·
1
2
A
µ
, m
B
≡Z
m
m=
1 −
3e
2
8π
2
·
1
ε
+· · ·
m ,
Ψ
B
≡ µ
−
1
2
ε
Z
Ψ
Ψ =µ
−
1
2
ε
1 −
e
2
8π
2
·
1
ε
+· · ·
1
2
Ψ , α
B
≡ Z
α
α =
1 +
e
2
6π
2
·
1
ε
+· · ·
α ,
e
B
≡ µ
1
2
ε
Z
e
e = µ
1
2
ε
1 +
e
2
12π
2
·
1
ε
+· · ·
e . (22.37)
We note that to oneloop order A
µ
≡ e
B
A
µ
B
and α
B
/e
2
B
are ﬁnite for n → 4. This is not
an accident, but the consequence of a socalled Ward identity, which as a consequence of
the gauge symmetry (through the BRS invariance mentioned at the end of sect. 20) relates
diﬀerent Z factors,
Z
2
e
= Z
α
= 1/Z
A
. (22.38)
It is therefore sometimes much more convenient to use
L = −
1
4e
2
B
(∂
µ
A
ν
−∂
ν
A
µ
)
2
−
α
B
2e
2
B
(∂
µ
A
µ
)
2
+ Ψ
B
(iγ
µ
D
µ
−m
B
)Ψ
B
, D
µ
≡ ∂
µ
−iA
µ
.
(22.39)
To all orders in the loop expansion the ﬁeld A
µ
and the gauge ﬁxing parameter α/e
2
remain
free of renormalisations. The same holds for nonAbelian gauge theories. By absorbing the
107
charge q (called coupling constant g ≡ q henceforth) in the gauge ﬁeld, the Lagrangian can
be expressed as
L =
1
2g
2
B
Tr(F
2
µν
) +
α
B
g
2
B
Tr(∂
µ
A
µ
)
2
+ Ψ
B
(iγ
µ
D
µ
−m
B
)Ψ
B
, (22.40)
where the gauge ﬁeld and the gauge ﬁxing parameter receive no renormalisations, in other
words they are already the renormalised ﬁeld and gauge ﬁxing parameter. The ﬁeld strength
F
µν
and covariant derivative D
µ
are now given by (compare eqs. (18.35) and (18.41))
D
µ
= ∂
µ
+ A
µ
, F
µν
= ∂
µ
A
ν
−∂
ν
A
µ
+ [A
µ
, A
ν
] . (22.41)
If the gauge group is SU(N) and there are n
f
ﬂavours of fermions the renormalisation of the
coupling constant is given by (see Itzykson and Zuber sect. 12.34)
g
B
= µ
1
2
ε
Z
g
g = µ
1
2
ε
1 −
(11N −2n
f
)g
2
48π
2
·
1
ε
+· · ·
g . (22.42)
As long as the number of fermion ﬂavours is small enough, we see that the one loop corrections
to the bare coupling constant diﬀers in sign from the equivalent expression for the Abelian
case. It is the selfinteractions of the nonAbelian gauge ﬁelds that are responsible for
the asymptotic freedom of its running coupling constant. The running of the coupling is
expressed in terms of the socalled betafunction
β(g) ≡ µ
∂g(g
B
(ε), µ, ε)
∂µ
, (22.43)
where the derivative is taken at ﬁxed ε and g
B
(g ≡ g
R
). For nonAbelian gauge theories
one ﬁnds (µ
0
is an integration constant)
β(g) = −
(11N −2n
f
)g
3
48π
2
+O(g
5
) ,
1
g
2
(µ)
=
(11N −2n
f
)
24π
2
log(µ/µ
0
) +O(g
2
(µ)) ,
(22.44)
whereas for QED (coupled to n
f
ﬂavours of fermions)
β(e) =
n
f
e
3
12π
2
+O(e
5
) ,
1
e
2
(µ)
= −
n
f
6π
2
log(µ/µ
0
) +O(e
2
(µ)) . (22.45)
For other regularisations the computation of the running coupling constant is similar, except
that ε is replaced roughly by 1/ log(Λ). To the order displayed, the betafunctions do not
depend on the regularisation scheme.
It is perhaps appropriate to end these lecture notes with as classic an experimental test
of renormalisation eﬀects in ﬁeld theory as the one for the Casimir energy in section 2. It
concerns the Lamb shift, measured in 1947, which is the very small energy splitting of the
2S1
2
and 2P1
2
orbitals in hydrogen atoms, receiving a contribution from vacuum polarisation
eﬀects (for a discussion of the other contributions see section 732 of Itzykson and Zuber).
In problem 39 it will be shown that to oneloop order the photon vacuum polarisation is
given by (compare eqs. (16.22) and (16.23). In the Landau gauge, α →∞, we can drop the
Λ
(α)
factors)
Σ
µν
(q) = −Λ
(α)
µβ
(q)(q
2
g
βγ
−q
β
q
γ
)ω(q
2
)Λ
(α)
γν
(q) . (22.46)
108
From the results of problem 39, where ω is computed with PauliVillars regularisation, it
can be deduced that (m is the electron mass)
ω(q
2
) = a log(Λ/m) + b + cq
2
for q
2
→0, Λ →∞ . (22.47)
The precise values of the coeﬃcients a and b are not very important, as the combination
a log(Λ/m) + b can be absorbed in the ﬁeld renormalisation (this means that a can be
read oﬀ from Z
A
given above). In section 711 of Itzykson and Zuber it is shown that
c = e
2
/(60π
2
m
2
). In the static limit, as is relevant for the hydrogen atom, q
2
= −q
2
and the
photon exchange can be accurately described by the Coulomb potential
−
e
2
4πr
= −e
2
d
3
q
e
i q·r
q
2
, (22.48)
which due to the vacuum polarisation is replaced by
−e
2
d
3
q
e
i q·r
q
2
(1 +ω(−q
2
))
= −e
2
d
3
q
1
q
2
+ c +· · ·
e
i q·r
= −
e
2
4πr
−
e
4
60m
2
π
2
δ
3
(r) .
(22.49)
The extra deltafunction interaction, that arises from the vacuum ﬂuctuations, only aﬀects
the wave functions that do not vanish in the origin. Consequently, only the energy of the S
orbitals will be shifted by this correction
∆E(nS1
2
) = −
4mα
5
e
15πn
3
, α
e
=
e
2
4π
, (22.50)
where n is the radial quantum number and α
e
is the ﬁnestructure constant.
109
Table of contents
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.
Introduction Motivation Quantisation of ﬁelds EulerLagrange equations Treelevel diagrams Hamiltonian perturbation theory Path integrals in quantum mechanics Path integrals in ﬁeld theory Perturbative expansion for ﬁeld theory The scattering matrix Cross sections Decay rates The Dirac equation Plane wave solutions of the Dirac equation The Dirac Hamiltonian Path integrals for fermions Feynman rules for vector ﬁelds Quantum Electrodynamics  QED NonAbelian gauge theories The Higgs mechanism Gauge ﬁxing and ghosts The Standard Model Loop corrections and renormalisation Index to Problems
1 2 4 8 13 17 20 27 33 37 43 45 47 52 55 57 70 74 82 88 90 93 99 110
0
0
Introduction
Field theory is most successful in describing the process of scattering of particles in the context of the Standard Model, and in particular in the Electromagnetic and Weak Interactions. The Large Electron Positron (LEP) collider operated 1989 till 2000. In a ring of 27 km in diameter, electrons and positrons were accelerated in opposite directions to energies of approximately 45 GeV. This energy is equivalent to half the mass (expressed as energy through E = mc2 ) of the neutral Z o vector boson mass, which mediates part of the weak interactions. The Z o particle can thus be created in electronpositron annihilation, at the regions where the electron and positron beams intersect. As a Z o can be formed out of an electron and its antiparticle, the positron, it can also decay into these particles. Likewise it can decay in a muonantimuon pair and other combinations (like hadrons). The crosssection for the formation of Z o particles shows a resonance peak around the energy where the Z o particle can be formed. The width of this peak is a measure for the probability of the decay of this particle. By the time you have worked yourself through this course, you should be able to understand how to calculate this crosssection, which in a good approximation is given by σ=
2 4παe E 2 /27 2 2 (E 2 − MZ o )2 + MZ o Γ2 o Z
2
,
(0.1)
e expressed in units where h = c = 1. αe = 4π ∼ 1/137.037, is the ﬁnestructure constant, ¯ E is twice the beam energy, MZ o the mass and ΓZ o the decay rate (or width) of the Z o vector boson. The latter gets a contribution from all particles in which the Z o can decay, in particular from the decay in a neutrino and antineutrino of the three known types (electron, muon and tau neutrinos). Any other unknown neutrino type (assuming their mass to be smaller than half the Z o mass) would contribute likewise. Neutrinos are very hard to detect directly, as they have no charge and only interact through the weak interactions (and gravity) with other matter. With the data obtained from the LEP collider (the ﬁgure is from the ALEPH collaboration) one has been able to establish that there are no unknown types of light neutrinos, i.e. Nν = 3, which has important consequences (also for cosmology).
35 30 25 20 15 10 5 0 1.05 1 0.95 88 89 90 91 92 93 94 95
ﬁg. 1
The main aim of this ﬁeld theory course is to give the student a working knowledge and understanding of the theory of particles and ﬁelds, with a description of the Standard Model towards the end. We feel that an essential ingredient of any ﬁeld theory course has 1
which are as important as the theoretical aspects (a point Veltman often emphasised forcefully) I can recommend “Field Theory in Particle Physics” by B. For more emphasis on the phenomenological aspects of ﬁeld theory.and perhaps one should spare the student the long struggle to arrive at a consistent formulation. Martinus Veltman and Gerard ’t Hooft. Zuber (McGrawHill. Veltman (Cambridge University Press. they were edited by Jeroen Snippe. New York. In Leiden I spent roughly 25% longer in front of the classroom (3 lectures of 45 minutes each for 14 weeks).R. Taylor (Cambridge Univ. de Wit and J. 1971). t) . These lecture notes reﬂect the ﬁeld theory courses I taught in the fall of 1992 at Utrecht. Consider the Schr¨dinger equation o ct d d x ∂Ψ(x. is very valuable. I also recommend “Diagrammatica: The path to Feynman diagrams”. which allowed me to spend more time and detail on certain aspects. Information can only propagate inside the future light cone. for its unique style.M. New York. 1994. 40 in total. by M. to understand why a ﬁeld needs to be introduced. 1976). t) = 0 for all (x. Nevertheless. Although this is not an attempt to do justice to history . In their present form. Smith (NorthHolland.but the traditional approach of introducing the concept is not very inspiring and most often lacks physical motivation.C. Of the many books on ﬁeld theory that exist by now.P.B. which form a crucial ingredient of these lectures. by J. useful towards the end of this course. t) d' E ﬁg. E. This is why the problems form an integral part of this course.0) ∂t NΨ Relativistic invariance should require that Ψ(x. with a lot of hindsight. because it oﬀers material substantially beyond the content of these notes. (1. 1996.to be to teach the student how Feynman rules are derived from ﬁrst principles. Berestetskii. Amsterdam. 1978) is a must.Press. 1986). it is equally essential that the student learns how to use these rules. I owe much to my teachers in this ﬁeld. and 1993. 0) = 0} of the wave function at t = 0. I will follow to a large extent their conventions. The set of problems. 1990) it is inevitable that there is some overlap. 1994). With the path integral approach this is feasible. for an introduction to the Standard Model. were initially compiled by KarelJan Schoutens with some additions by myself. Feynman and A. The discussion on unitarity is very informative and it has an appendix comparing diﬀerent conventions. the book “Quantum mechanics and path integrals” by R. which most likely has not completely crystalised yet either . In the following discussion I was inspired by “Relativistic Quantum Theory” from V. Peierls (1930). 1 Motivation Field theory is the ultimate consequence of the attempts to reconcile the principles of relativistic invariance with those of quantum mechanics. It is not too diﬃcult. Oxford. the book “Gauge theories of weak interactions”.E. Hibbs (McGrawHill. An important consequence of relativistic invariance is that no information should propagate at a speed greater than that of light. Itzykson and J. t) outside the light cone of the support NΨ = {xΨ(x. Pitaevskii (Pergamon Press. 2 . As Julius Wess put it during his course as a Lorentz professor at our Institute “you won’t become a good pianist by listening to good concerts”. I recommend the student to consider using “Quantum Field Theory” by C. Finally. The argument goes back to L. For path integrals.D. As I taught in Utrecht from ’t Hooft’s lecture notes “Inleiding in de gequantiseerde veldentheorie” (Utrecht. 1980) in addition to these lecture notes. 2 i¯ h = HΨ(x. Landau and R. 1998 and 2000 at Leiden.P.B. Lifshitz and L.
We will now verify by direct computation that localising the wave function within the light cone will indeed require negative energy states. it seems. t)Ψ(x.2) However. (1. which coincides with the limit set ¯ ¯ by the De Broglie wavelength. and once antiparticles are introduced.4) We shall show that this. this can be easily remedied. of course. Nevertheless. the initial condition uniquely ﬁxes the wave o 3 .e. We consider ﬁrst the positive square root of the KleinGordon equation and solve the Schr¨dinger equation for the initial condition o Ψ(x. As the Schr¨dinger equation is ﬁrst order in time. 0) = δ3 (x). that is position can in principle not be measured with arbitrary accuracy. For ordinary quantum mechanics. t) ∂x2 (1. Only a free particle. On the other hand.Naturally. But negative energy states will be interpreted as antiparticles. we will learn something important from that computation.1) this is clearly not the case. formulated in terms of a potential H= p2 + V (x) . if we would like to localise the particle more accurate than within its De Broglie wavelength. for this equation the usual deﬁnition of probability density is not conserved ∂t d3 xΨ∗ (x. we ﬁnd that ∆x > h/p ≥ hc/E. but in that case. Only for a free particle. t) = ∂t −¯ 2 c2 h ∂2 + m2 c4 Ψ(x. Using the relation E 2 = p2 c2 + m2 c4 the most obvious attempt for a relativistically invariant wave equation would be the KleinGordon equation −¯ 2 h ∂ 2 Ψ(x. which is unacceptable. But ﬁrst we will provide a simple heuristic argument based on the uncertainty relation. where momentum is conserved. which can annihilate with particles. ∂t2 ∂x2 (1. as a plane wave. As the momentum is bounded by the (positive) energy (p ≤ E/c) and as the maximal change in the momentum is of the order of p itself. If we take this serious.3) As this is a consequence of the fact that the equation is second order in time. however. i. ¯ From the uncertainty principle ∆x∆p > h/2 and the bound on the speed involved in any measurement of the position. a ﬁrst requirement should be that the Schr¨dinger equation itself is relativiso tically invariant. seems to be compatible with relativistic invariance. particle number is no longer conserved and we likewise loose the notion of position of a particle. the position is completely undetermined. namely that negative energies seem unavoidable when trying to localise wave functions within the light cone of NΨ . t) = −¯ 2 c2 h + m2 c4 Ψ(x. it follows that precision of a measurement of the momentum ¯ is limited by the available time ∆t∆p > h/c. such a measurement would be possible. From this we can solve any initial condition by convolution. t) = 0 . consistent with the plane wave description of such a free particle (the light cone of NΨ would in that case indeed give us no constraint). 2m (1. by taking the “square root” of the KleinGordon equation i¯ h ∂Ψ(x. t) ∂ 2 Ψ(x. violates the principle of causality. More instructive is to look at how accurately we can determine the position of a particle. the notion of a wave function looses its meaning. t) . the wave function propagates outside of its light cone. it seems to require an uncertainty in momentum that can only be achieved by allowing for negative energy states.
localisation is only possible if we allow for negative energy solutions. Drell (McGraw o Hill. New York.7) Outside of the light cone. we can try to introduce such a free particle as a quantum observable.function for all later times and there will be a unique answer to the question whether the wave function vanishes outside the light cone (i. h h 2 z 2 = m2 c2 (r 2 − c2 t2 )/¯ .e. Ψ∗ is a solution of the negative square root of the KleinGordon equation. mcr/¯ = z cosh(v) .e. t) = −i ∂ 2 4π 2 rc ∂r∂t −i ∂ 2 = 8π 2 rc ∂r∂t −i ∂ 2 = 2π 2 rc ∂r∂t du cos(z sinh(u) cosh(v))e−iz sinh(v) cosh(u) du e−iz sinh(u+v) + e−iz sinh(u−v) du cos(z sinh(u)) ≡ −i ∂ 2 Ko (z) . i. It decays exponentially. for t > x).6) such that (the last identity simply being the deﬁnition of the modiﬁed Bessel function Ko ) Ψ(x. spatial dimensions. t) = e−i(ko t−x·k)/¯ . we could take the real part of Ψ as the o wave function. z is real (r 2 > c2 t2 ) and Ψ(x. using the fact that in Fourier space the solution is trivial. t) is purely imaginary. t) = e e (2π¯ )3 h p2 dp sin(θ)dθ ipr cos(θ)/¯ −it√p2 c2 +m2 c4 /¯ h h e e = 2 h3 (2π) ¯ √ 1 2 2 2 4 h = pdp sin(pr/¯ )e−it p c +m c /¯ h 2 2π 2 r¯ h ∞ cos(pr/¯ ) −it√p2 c2 +m2 c4 /¯ h −i ∂ 2 h . This observable is hence described by a plane wave h ϕk (x. appendix C of “Relativistic Quantum Fields” by J. (2. 1965)) the following explicit expression Ψ(x. Bj¨rken and S.g. (1. Problem 1 asks you to investigate this in the simpler case of one. but does not vanish! Inside the light cone we ﬁnd by analytic continuation (see e.D. t) = √ √ 1 ∂2 iYo (mc c2 t2 − r 2 /¯ ) − sign(t)Jo (mc c2 t2 − r 2 /¯ ) . h h 4πrc ∂r∂t r 2 < c2 t2 . 2π 2 rc ∂r∂t (1. Apparently. t) thus requires just some skills in performing Fourier integrals. Ψ(x.5) dp √ 2 2 e = 2 r ∂r∂t 0 2π p c + m2 c4 We now introduce p = mc sinh(u) . (1. d3 p ip·x/¯ −it√p 2 c2 +m2 c4 /¯ h h Ψ(x. h ∞ −∞ ∞ −∞ ∞ 0 (1. but not its positive square root. It satisﬁes the KleinGordon equation. but free particles do not seem to be in contradiction with relativistic invariance.D. and corresponds to a negative energy solution.1) 4 . t) = 0 for x > ct and want to stay as close as possible to the solutions of the Schr¨dinger equation. For the latter we simply give the result here. instead of three. 2 Quantisation of ﬁelds As position is no longer a quantum observable.8) If we want to insist on locality. Computing Ψ(x. mc2 t/¯ = z sinh(v) .
t) satisfy the harmonic ˜ equation 2 ˜ 2 ∂ ϕ(k. π (k) ≡ ˜ h ∂ ¯ i ∂ ϕ(k) ˜ . Using the well known h h h h fact that eiHt/¯ a(k)e−iHt/¯ = e−iω(k)t a(k) and eiHt/¯ a† (k)e−iHt/¯ = eiω(k)t a† (k). which is therefore described by a ﬁeld ϕ(x. 0)e−iHt/¯ . These play the role the ˜ coordinates used to play in ordinary quantum mechanics and will require quantisation.6) k 1 H(k) = 2 ˜ (k)2 + 1 ω(k)2 ϕ(k)2 . with frequency ω(k) ≡ k0 (k)/¯ . As they satisfy a simple harmonic equation in time. h i¯ h ∂Ψ(ϕ) = H Ψ(ϕ) = ∂t H(k) Ψ(ϕ) .3) It satisﬁes the KleinGordon equation if the Fourier components ϕ(k. it is natural to quantise them as harmonic oscillators. t) 2 −¯ h = (c2 k 2 + m2 c4 )ϕ(k.8) 5 . which gives ϕ(x. t) = eiHt/¯ ϕ(x. t) ∂ 2 ϕ(x. (2.4) 2 ∂t Its solutions split in positive and negative frequency components h h ϕ(k. By superposition of these plane waves we can make a superposition of free particles. t) = L− 2 k 3 h ¯ 2ko (k) h h a† (k)e−i(k·x−ko t)/¯ + a(k)ei(k·x−ko t)/¯ . we can introduce annihilation and creation operators a(k) = a† (k) = 1 2¯ ω(k) h 1 2¯ ω(k) h ω(k)ϕ(k) + i˜ (k) ˜ π ω(k)ϕ∗ (k) − i˜ ∗ (k) ˜ π . t) = ϕ+ (k)e−iko t/¯ + ϕ− (k)eiko t/¯ ˜ ˜ ˜ (2. or rather the wave functional Ψ(ϕ). describes the distribution over the various free particle states. (2. ˜ ˜ (2.7) and express the ﬁeld operator (the equivalent of the coordinates) in terms of these creation and annihilation operators. t) = (2π¯ )− 2 h 3 h d3 k ϕ(k. (2. The Hamiltonian is then simply the sum of the harmonic oscillator Hamiltonian for each k.5) The wave function. the integral over the momenta is replaced by a sum as the momenta are in that case discrete.which satisﬁes the KleinGordon equation −¯ 2 h ∂ 2 ϕ(x. t) . which is a consequence of [a(k). n ∈ Z 3 . . t) . H] = −¯ ω(k)a† (k). The basic dynamical variables are ϕ(k). t) = −¯ 2 c2 h + m2 c4 ϕ(x. ∂t2 ∂x2 (2. H] = hω(k)a(k) and [a† (k). Like for h Z the harmonic oscillator. t)eik·x/¯ ˜ . we ﬁnd ¯ h ϕ(x.2) where k0 = c2 k 2 + m2 c4 is the energy of the free particle. t) ≡ ko (k)ϕ(k. π ˜ 2 In a ﬁnite volume with periodic boundary conditions. (2. To give the ﬁeld operator its time dependence we have to invoke h h the Heisenberg picture. k = 2πn¯ /L.
which is compatible with ϕ∗ (k) = ϕ(−k). however. as each 6 . for (x − x′ )2 > (t − t′ )2 c2 . in which the Lorentz invariance is intrinsic. when combining quantum mechanics with relativistic invariance. empty space is still full of zeropoint ﬂuctuations. that its energy is linear in the occupation number. only visible if we probe that empty space in one way or another. ϕ(x′ . namely that of an antiparticle with opposite charge. we would ﬁrst like to address a simple physical consequence of the introduction and subsequent quantisation of ﬁelds. which in ﬁeld theory is now interpreted as the number of free particles of momentum k. however. It is this interpretation that is known as second quantisation. Note that in the Heisenberg h h picture positive energy modes behave in time as eiEt/¯ . however. It is. t′ )] = 0. hence describing a separate degree of freedom.In an inﬁnite volume we replace L− 2 k by (2π¯ )− 2 d3 k. Also. Field theory thus seems to be nothing but the quantum mechanics of an inﬁnite number of degrees of freedom. For a real scalar ﬁeld. t). Interactions between the particles are simply introduced by modifying the KleinGordon equation to have nonlinear terms. as long as these two points are not causally connected. (2. ˜ ˜ ˜ ˜ required to describe a real ﬁeld (complex ﬁelds will be discussed in problem 5). We will resolve this by using the path integral approach.9) with nk the occupation number. a deﬁnition that makes sense as the energy of such a state is nk ko (k) above the state with zero occupation number (the “vacuum”). Before preparing for path integrals by discussing the action principle. which consequently can be interpreted as the annihilation of the removed particle with an antiparticle (described by the annihilation operator).10) It states that the action of an operator on the wave functional at a given spacetime point is independent of the action of the operator at an other spacetime point. which requires the choice of a time coordinate it remains to be established. which are. (2. Put diﬀerently. It states that empty space (all occupation numbers equal to zero) has nevertheless a nontrivial structure. The annihilation operator in this language therefore removes a particle (lowering the energy by the appropriate amount). We were forced to introduce the notion of ﬁelds and the interpretation involving antiparticles. This is implied by the following identity. its physical interpretation that crucially diﬀers from that of ordinary quantum mechanics. But for the complex ﬁeld of problem 5. It is the property of the harmonic oscillator. which makes the ﬁeld theory interpretation in terms of particles possible. formally. The Hilbert space is now given by the product of the Hilbert spaces of each k separately {nk } >= nk >= a† (k)nk k 3 3 k nk ! 0k > . but which can also be shown to be equivalent to the Hamiltonian formulation. We should therefore verify that indeed it does not give rise to propagation of information with a speed larger than the speed of light. the Fourier component with negative energy is independent of the one with positive energy. Apparently we can identify (up to a factor) ϕ− (k) with a† (−k) and ϕ+ (k) with a(k). Due to the description of the time evolution with a Hamiltonian. a particle is its own antiparticle and this description is perhaps somewhat unfamiliar. which for the free scalar ﬁeld will be veriﬁed in problem 6 [ϕ(x. after which in general the diﬀerent Fourier components no longer decouple. that these equations are covariant under Lorentz transformations. in the same way that the ground state of a Hydrogen atom is nontrivial.
Acad. but the essential ingredient is that Fourier components of the ﬁeld are aﬀected by the presence of the conducting plates. If furthermore we use periodic boundary conditions in the two other perpendicular directions over a distance L. are given by sin(πkx1 /x) with k a positive integer. One can mimic this by artiﬁcially cutting oﬀ the integral and sum at high momenta. (as long as gravity is left out of our considerations) one is only sensitive to differences in energy. ser. (2. perpendicular to the conducting walls. A famous and elegant method for probing the vacuum was introduced by Casimir (Proc. We would not expect the physical result to depend on the details of how we do this. Kon. but as Casimir observed.. by assuming that the ﬁeld has to vanish at the plates. par. The dependence of the vacuum energy on the probe can be used to discover the nontrivial structure of the vacuum. (2. ¨¨ ¨¨ ¨¨ ¨¨ T ¨ ¨ B ¨ ¨¨ ¨¨ L ¨¨ ¨ ¨¨ ¨ % ¨ L ' ﬁg. as otherwise we could use this experiment in an ingenious way to learn how nature behaves at arbitrarily high energies. whereas the quantisation of the momenta in the other two directions is as usual. 324). which gives a force. 3 x E ¨¨ ¨¨ c ¨ ¨ ¨ ¨ ¨¨ ¨¨ ¨ ¨ One can now formally take the inﬁnite volume limit c d ∞ F (x) = lim FL (x) = − 2 2 L→∞ 8π h dx k=1 ¯ d2 p p2 π¯ k h + x 2 . If we probe the vacuum. Ned. in practise no conducting plate can shield a ﬁeld perfectly and especially for high frequency the boundary conditions should be modiﬁed. We can also discuss this in the context of the simple scalar ﬁeld we have introduced before.11) However.13) The integral and the sum are clearly divergent. B51 (1948) 793).zeropoint energy is nonzero. then one easily veriﬁes that the force per unit area on the conducting plates is given by ∞ 1 d FL (x) = −dEo (x)/dx = − 2 2L dx n∈Z 2 k=1 Z 2π¯ cn h L 2 π¯ ck h + x 2 . its energy can only be put to zero for one particular value of the probe. who considered using two conducting plates in empty space. The energy of the vacuum is a function of the distance between the two plates. Strictly speaking we should discuss this in the situation of the quantised electromagnetic ﬁeld (see Itzykson and Zuber. For simplicity we will also take the mass of the scalar particles to vanish.12) where due to the vanishing boundary conditions the Fourier modes in the x1 direction. Wet. Indeed Casimir’s careful analysis showed that 7 . the energy of the vacuum in ﬁeld theory seems to be inﬁnite E0 = k k 2 c2 + m2 c4 = · · ·? (2.
∂µ ϕ. .1) ⊖ can be derived by variational calculus from an action principle S = d4 x L(ϕ.15) 480x4 Please note that we have disregarded the space outside the conducting plates. Therefore. Since the result is insensitive to the method of regularisation (only an overall constant contribution to Eo (x) depends on it.the result is independent of the cutoﬀ function chosen. the eﬀect of the zeropoint ﬂuctuations in the vacuum leads to a (very small) attractive force. The method of calculation is known as dimensional regularisation. we can choose a convenient way to perform the calculation. (2. It can be analytically extended to odd k=1 negative arguments. which was ten years later experimentally measured by Sparnaay (Physica. as we argued before).14) in which ζ(i) ≡ ∞ k −i is the Riemann ζ function. x. It is an important example of what we will later recognise as renormalisability of ﬁeld theory. y. x) = 2 (∂µ ϕ)2 − V (ϕ) . Details of this will be provided in problem 2. to be discussed at the end of the last section. (3. where in terms of Bernoulli coeﬃcients ζ(1 − 2i) = −B2i /(2i).3) d4 x −δϕ(∂µ ∂ µ ϕ + ∂V (ϕ) ) + ∂ϕ 8 dµ σ(δϕ∂ µ ϕ) = 0 . and vanishes when L → ∞. F (x) = − 3 EulerLagrange equations 1 −1 −1 The KleinGordon equation in Lorentz covariant form (x ≡ (ct. but that is not observable. g µν = ⊖ −1 . z) ≡ (x0 . Another famous example of the inﬂuence of zeropoint ﬂuctuations is the Lamb shift in atomic spectra (hyperﬁne splittings).2) We assume the ﬁeld to be given at the boundary of the domain M of integration (typically assuming the ﬁeld vanishes at inﬁnity) and demand the action to be stationary with respect to any variation ϕ(x) → ϕ(x) + δϕ(x) of the ﬁeld. x) . ∂µ ϕ. 24 (1958) 751). 1 L(ϕ. (2. one easily ﬁnds that the region outside the plates contributes with F (L − x) to the force. We will ﬁnd that F (x) = lim − π¯ k h n→2 x hc Γ(−(n + 1)/2) −(n+2) ¯ x = lim (n + 1)ζ(−n − 1)π 3n/2 n→2 8π Γ(−1/2) d n p p2 + c d ∞ 8π 2h2 dx k=1 ¯ 2 . where one works in an arbitrary dimension (n = 2) and then analytically extends the result to n = 2. (∂µ ϕ)2 ≡ ∂µ ϕ∂ µ ϕ = g µν ∂µ ϕ∂ν ϕ . (3. Imposing also periodic boundary conditions in that direction. δS(ϕ) ≡ S(ϕ + δϕ) − S(ϕ) = = M M d4 x ∂ µ ϕ∂µ δϕ − ∂M ∂V (ϕ) δϕ ∂ϕ (3. and we simply ﬁnd that 2 2 2 π 2 hc ¯ . x)) g µν ∂µ ∂ν ϕ(x) + m2 ϕ(x) = 0 . 1 V (ϕ) = 2 m2 ϕ2 . Also Γ(− 1 ) = − 3 Γ(− 3 ) is ﬁnite.
(3. Later we will see that one can not add arbitrary powers of the ﬁeld to this potential. we can derive from a Lagrangian with ϕ(x) and ϕ(x) ≡ ˙ ∂ϕ(x)/∂t as its independent variables. which is then split according to the explicit δ dependence of the action on the ﬁeld and its derivatives (usually an action will not contain higher than ﬁrst order spacetime derivatives). by generalising the dependence of the “potential” V (ϕ) to include higher order terms. Examples of these anomalies will be discussed later for the breaking of scale invariance and gauge invariance). ∂ϕ (3. except in two dimensions. if we substitute for ϕ(x) eq.6). (3. of the Lagrangian density L. It is now simple to add interactions to the KleinGordon equation. except at ∂M. which makes it much easier to guarantee Lorentz covariance. We can also write the EulerLagrange equations for arbitrary action S(ϕ) in terms of functional derivatives δS δS δS = − ∂µ =0 . the Hamiltonian through a Legendre transformation to the canonical pair of variables π(x) (the “momentum”) and ϕ(x) (the “coordinate”) π(x) = δS .10) which perhaps explains why V is called the potential. δ ϕ(x) ˙ H= H(x)d3 x = (π(x)ϕ(x) − L(x)) d3 x .8) For the KleinGordon ﬁeld we simply ﬁnd 1 1 1 H = 2 π(x)2 + 2 (∂i ϕ(x))2 + 2 m2 ϕ2 (x) . required to make sense of the path integral destroys the Lorentz invariance. The big advantage of using an action principle is that S is a Lorentz scalar. The variation δϕ is arbitratry. As in classical ﬁeld theory.8). (2. and this implies the EulerLagrange equation ∂µ ∂ µ ϕ + ∂V (ϕ) =0 . δπ(x) π(x) = − ˙ δH δH + ∂i δϕ(x) δ∂i ϕ(x) . ˙ (3.where dµ σ is the integration measure on the boundary ∂M.5) where − stands for the total functional derivative. like 1 V (ϕ) = 2 m2 ϕ2 + g 4 ϕ 4! .6) which is known as a scalar ϕ4 ﬁeld theory.7) The classical Hamilton equations of motion are given by ϕ(x) = ˙ δH . Please note that a functional derivative has the property δϕ(x)/δϕ(y) = δ4 (x−y). as one sees often. Lorentz covariance is much more easy to establish within this framework (there are instances where the regularisation.9) and in problem 5 one will see that this Hamiltonian coincides with eq. (3. where we assume δϕ vanishes. (3. − ϕ(x) δ δϕ(x) δ∂µ ϕ(x) − (3. For an interacting scalar ﬁeld one ﬁnds 1 1 H = 2 π(x)2 + 2 (∂i ϕ(x))2 + V (ϕ(x)) .4) which coincides with the KleinGordon equation. like in string theory. (2. As the action will be the starting point of the path integral formulation of ﬁeld theory. which is why in the above equation we take functional derivatives of the action S and not. 9 .
whose time component is the charge density.It is well known that the Hamiltonian equations imply that H itself is conserved with time. where Λ is a parameter (such as a shift of the coordinates or a phase rotation of a complex ﬁeld).17) In problem 3 these quantities will be deﬁned for a charged scalar ﬁeld. (3.13) In the same fashion one proves conservation of momentum in case the Lagrangian does not explicitly depend on space (∂L/∂xi = 0) 0= dL = dxi d3 x ∂µ ∂i ϕ(x) δS δ∂µ ϕ(x) = d dt d3 x π(x)∂i ϕ(x) . (3.11) Conservation of energy is one of the most important laws of nature and it is instructive to derive it more directly from the fact that L does not depend explicitly on time. which vanishes if we assume that the ﬁeld is time independent (or vanishes) at the boundary of the spatial integration domain. ∂µ T µν (x) = 0 . The above equation becomes now dL d = dt dt which can also be expressed as d dt d3 x (ϕ(x)π(x) − L) ≡ ˙ dH =0 . They can be derived as the space integral of the time component of a conserved current or tensor ∂µ J µ (x) = 0 . Assuming the current to vanish at spatial inﬁnity one easily ﬁnds d dt d3 x Jo (x) = d3 x ∂i Ji (x) = 0 (3. (3. then the following current is conserved J µ (x) = ∂ϕΛ (x) δS δ(∂µ ϕ(x)) ∂Λ 10 . dt (3. such that dL = dt d3 x ∂t ϕ(x) δS δS + ∂t (∂µ ϕ(x)) δϕ(x) δ∂µ ϕ(x) = d3 x ∂µ ∂t ϕ(x) δS δ∂µ ϕ(x) .12) The last term contains a total derivative. (3.18) The underlying principle is described by the Noether theorem. provided the Lagrangian (or Hamiltonian) has no explicit timedependence dH = dt d3 x π(x) ˙ δH δH + ϕ(x) ˙ δπ(x) δϕ(x) =0 .16) Both conservation of momentum and energy are examples of conservation laws that are consequence of symmetries (translation and time invariance). where Jµ (x) can be identiﬁed with the current.19) . which implies that if the Lagrangian L is invariant under ϕ → ϕΛ . (3. We deﬁne the Lagrangian L as an integral of the Lagrange density L over space. ˙ (3. (3.15) The conserved momentum is hence given by Pi = d3 x π(x)∂i ϕ(x) . L ≡ d3 x L.14) d3 x ϕ(x) ˙ δS d = δ ϕ(x) ˙ dt d3 x ϕ(x)π(x) . Indeed the total charge is conserved.
(3. due to the transformation property ∂ xµ ∂ xν αβ ¯ ¯ g (x) . For εµ not constant. (3. (3. in ¯ particular − det g(x) = 1.22) ∂xα ∂xβ For global translation invariance. δg = −g δgαβ g and δ − det g = 2 g − det gδgµν . (3. Tµν = ∂µ ϕ∂ν ϕ − gµν L.17)).25) which implies conservation of the energymomentum tensor (Too = √ From the fact that H).15) can be easily generalised to show that the energymomentum tensor. Hence. shows that δS 0= = ∂µ J µ (x) . As long as indices are contracted with the metric tensor g. Particularly with the latter in mind we demand therefore that the action S (and not just L) is invariant under ϕ(x) → ϕΛ(x) (x). which is the local version of translation invariance.24) ¯ x x x x x Now observe that g µν (x) is constant. is conserved (eq.1). but the associated Jacobian can be easily absorbed √ by − det g. (3.23) For the original coordinates x of Minkowski spacetime the metric is given as in eq. √ µν µα βν 1 µν δgµν = −∂µ εν − ∂ν εµ . For a scalar ﬁeld this leads to the following invariant action ¯ xµ = xµ + εµ (x) . This actually leads to the same conserved currents in case L is also invariant. For this we have to make the action invariant under such local coordinate redeﬁnitions. using general coordinate invariance. (3. g µν (¯) = ¯ x S= d4 x − det g ( 1 g µν ∂µ ϕ∂ν ϕ − V (ϕ)) 2 (3. like the gauge invariance related to local changes of phase and the general coordinate invariance in general relativity.20) We here considered the invariance under a global symmetry. which is simply the action itself. The term linear in ε therefore has to vanish. but note that it only involved the variation of the metric under the general coordinate transformation. εµ is constant and eqs. integrated over x.The proof is simple and uses the EulerLagrange equations to substitute ∂µ {δS/δ(∂µ ϕ(x))} for δS/δϕ(x) 0 = δS ∂ϕΛ (x) δS ∂(∂µ ϕΛ (x)) dL(ϕΛ ) = + dΛ δϕ(x) ∂Λ δ(∂µ ϕ(x)) ∂Λ δS ∂ϕΛ (x) δS ∂(∂µ ϕΛ (x)) = ∂µ + = ∂µ J µ (x) δ(∂µ ϕ(x)) ∂Λ δ(∂µ ϕ(x)) ∂Λ (3.21) δΛ(x) As an important example we will discuss how this construction leads to conservation of the energymomentum tensor. The same computation as above. as x now plays the role of a dummy ¯ ¯ ¯ integration variable. (3. and by expanding g to ﬁrst order in εµ (x) we ﬁnd S= 1 d4 x (1 − ∂λ ελ (x)) [ 2 g µν ∂µ ϕ(¯)∂ν ϕ(¯) − V (ϕ(¯))] + ∂α εµ (x)∂µ ϕ(¯)∂ α ϕ(¯) .14) and (3. but important in nature are also the local symmetries. we note that the integration measure d4 x is not a scalar under general coordinate transformations. still using the EulerLagrange equations. with Λ an arbitrary function of spacetime. 0= d4 x εµ (x)∂α (g µα L(x) − ∂ µ ϕ(x)∂ α ϕ(x)) ≡ − d4 x εµ (x)∂α T αµ (x) .26) T µν (x) ≡ −2 δgµν (x) 11 . we derive the identity δS . (3. such that the term independent of ε is a function of x. L will be invariant under general coordinate transformations. where the determinant is applied to the 4 × 4 matrix gµν .
see problem 23) like the one for the electromagnetic ﬁeld. but for any other bosonic ﬁeld theory (fermions form an exception. or the sources.27) This already implies one of the Maxwell equations (through the socalled Jacobi or integrability conditions) ∂µ Fνλ + ∂ν Fλµ + ∂λ Fµν = 0 . rotB − ∂0 E = J . that the equations of motion imply that the current is conserved. (3. Its action for a free particle is given by S0 = d4 x ∂µ ϕ∗ (x)∂ µ ϕ(x) − m2 ϕ∗ (x)ϕ(x) . with E i (x) = −F 0i (x) its electric and B i (x) = − 1 εijk F jk (x) its magnetic components. Our action is invariant under this symmetry if and only if the current is conserved. (3.26) always gives a symmetric energymomentum tensor and from the derivation it is clear that the result holds not only for a simple scalar ﬁeld. J) are given by 1 (3. which we discuss now.34) is conserved.33) It is invariant under a phase rotation ϕ(x) → exp(ieΛ)ϕ(x) and from Noether’s theorem we deduce that Jµ (x) ≡ ie(ϕ(x)∂µ ϕ∗ (x) − ϕ∗ (x)∂µ ϕ(x)) (3. (3. 12 . ¯ These Maxwell equations follow from the following action Sem (J) = d4 x (− 1 Fµν (x)F µν (x) − Aµ (x)J µ (x)) 4 . An example of a conserved current can be deﬁned for a complex scalar ﬁeld.31) We note. ∂0 B + rotE = 0 . In terms of the vector potential Aµ (x) one has 2 Fµν (x) = ∂µ Aν (x) − ∂ν Aµ (x) . The ﬁeld is given by the tensor Fµν (x). see problem 3.28) Written as εµνλσ ∂ν Fλσ = 0 they are easily seen (resp.32) the theory does not change. J µ = (cρ. (3.30) ∂µ F µν = J ν or divE = ρ . (3. Eq. (3. as is well known. for µ = 0 and µ = i) to give divB = 0 . it is important that any Lorentz vector (like the derivative ∂µ ϕ) or tensor appears in the Lagrangian with its indices down. We can extend this global phase symmetry to a local symmetry if we couple the scalar ﬁeld minimally to the vector potential S = d4 x − 1 Fµν (x)F µν (x) + (Dµ ϕ)∗ (x)D µ ϕ(x) − m2 ϕ∗ (x)ϕ(x) 4 Dµ ϕ(x) ≡ ∂µ ϕ(x) − ieAµ (x)ϕ(x) . With Noether’s theorem this makes us suspect that this is caused by a symmetry and indeed it is known that under the gauge transformation Aµ (x) → Aµ (x) + ∂µ Λ(x) (3. This gauge symmetry will play a crucial role in the quantisation of the electromagnetic ﬁeld. c We have chosen HeavisideLorentz units and in the future we will also often choose units such that h = c = 1.29) The dynamical equations determining the ﬁelds in terms of the currents. (3.35) . the result is to be evaluated for Minkowski space. Furthermore.In taking the derivative with respect to the metric.
(3. Note that we can write this action also as S = Sem (J) + S0 + d4 x e2 Aµ (x)Aµ (x)ϕ(x)2 . The result is given by Sem (J = 0) = − 1 4 d4 x g µλ g νσ Fµν Fλσ − det g . We leave it as an exercise to show how the action of the electromagnetic ﬁeld can be generalised to be invariant under general coordinate transformations and to derive from this the energymomentum tensor.7) (4. ϕ(k) = ˜ 1 (2π)2 d4 x eikx ϕ(x) (4. (3. We add to the Lagrangian density L a socalled source term.5) 1 (2π)2 d4 x eikx J(x) . (3.This guarantees the combined invariance under a local gauge transformation ϕ(x) → exp(ieΛ(x))ϕ(x) . (3.6) .37) with J as given in eq. (4. (4.4) where G is called the Green’s function. (4.36) which makes the covariant derivative Dµ ϕ(x) of the scalar ﬁeld transform as the scalar ﬁeld itself.3) If g = 0 it is easy to solve the equation (describing a free particle interacting with a given source) in Fourier space. which couples linearly to the ﬁeld ϕ (compare the driving force term for a harmonic oscillator) 1 (4. For sake of explicitness. We discuss the scalar case ﬁrst.2) The EulerLagrange equations are now given by 1 ∂µ ∂ µ ϕ(x) + m2 ϕ(x) + 2 gϕ2(x) + J(x) = 0 .38) 4 4 Treelevel diagrams In general. even for local phase rotations. Aµ (x) → Aµ (x) + ∂µ Λ(x) . in the presence of interactions.34). it is given by the following Fourier integral G(x − y) = e−ik(x−y) d4 k (2π)4 k 2 − m2 + iε 13 . the equations of motions can not be solved exactly and one has to resort to a perturbative expansion in a small parameter. as it is the solution of the equation (∂µ ∂ µ + m2 )G(x − y) = −δ4 (x − y) Explicitly. 3! (4. we will take the following expression for the potential 1 V (ϕ) = 2 m2 ϕ2 (x) + g 3 ϕ (x) . as it is as always the simplest. Introducing the Fourier coeﬃcients ˜ J(k) = it follows that ˜ (−k 2 + m2 )ϕ(k) + J(k) = 0 or ϕ(x) = ˜ d4 y G(x − y)J(y) . T µν = 1 g µν F λσ Fλσ − F µλ F ν λ .1) L = 2 (∂µ ϕ)2 − V (ϕ) − J(x)ϕ(x) .
except for the term ϕ 1 (n−1) (y)2 at n odd. ϕ(x) = ϕ0 (x) + gϕ1 (x) + g 2 ϕ2 (x) + · · · Obviously we have ϕ0 (x) = d4 y G(x − y)J(y) . (4. Now we have found the solution for the free ﬁeld coupled to a source. (4. speciﬁed by the term iε. 2 which occurs once. we can do perturbation in the strength of the coupling constant g.14) d4 y G(x − y) (2ϕ0 (y)ϕn−1(y) + 2ϕ1 (y)ϕn−2(y) + · · ·) .11) We can therefore interpret the righthand side as a source (up to a minus sign) and this allows us to solve ϕ1 (x) using the Green’s function ϕ1 (x) = 1 2 d4 y G(x − y)ϕ2 (y) = 0 1 2 G(x − y)d4y G(y − z)G(y − w)J(z)J(w)d4 zd4 w . whereas the negative energy part will be required to vanish for future inﬁnity. The eﬀect of the iε prescription is. A Green’s function is not uniquely speciﬁed by its 2nd order equations.8) .16) ˜ ˜ ˜ ˜ 14 . e. as we will see. These boundary conditions are.g. but also requires boundary conditions.15) Here we have written out the terms in the sum explicitly to indicate that all terms occur twice and are the product of two diﬀerent terms. (4. (4.9) (4. Because of the interpretation of the negative energy states as antiparticles.12) This looks particularly simple in Fourier space ϕ1 (k) = ˜ 1 1 2 k 2 − m2 + iε 2(2π) d4 p ˜ ˜ J(p)J(k − p) (p2 − m2 + iε)((k − p)2 − m2 + iε) n−1 . which travel “backwards” in time. Classically this would not make sense. 0 (4. i=0 (4. to shift the poles on the real axes to the 1 complex k0 plane at k0 = ±((k 2 + m2 ) 2 − iε).13) It is clear that this can be continued iteratively. In Fourier space one ﬁnds ϕn (k) = ˜ 1 1 2 k 2 − m2 + iε 2(2π) d4 p (2ϕ0 (p)ϕn−1 (k − p) + 2ϕ1 (p)ϕn−2 (k − p) + · · ·) .Please note our shorthand notation of k 2 for kµ k µ and k(x − y) for kµ (xµ − y µ ). ∂µ ∂ µ ϕ(x) + m2 ϕ(x) + J(x) = − 1 gϕ2 (x) 2 can be solved iteratively by substituting a series expansion for ϕ(x). (4. and we would require the solution to vanish outside the future light cone. ∂µ ∂ µ ϕn (x) + m2 ϕn (x) = − 1 2 which is solved by ϕn (x) = 1 2 ϕi (x)ϕn−1−i (x) . the quantum theory will require that the positive energy part vanishes for past inﬁnity. In section 5 we will see that this will imply the appropriate behaviour required by the quantum theory.10) whereas ϕ1 (x) will be determined by the equation 1 ∂µ ∂ µ ϕ1 (x) + m2 ϕ1 (x) = − 2 ϕ2 (x) . (4.
To understand why momentum is conserved at each vertex we use that in the coordinate formulation each vertex comes with an integration over its position.By induction it is now easy to prove that ϕ(x) = diagrams r ×+ 1 2 r × ¨ ¨ r r × r + 1 2 +1 2 = diagrams g #vertices N(diagram) × ¨ × ¨¨ ¨ ¨ r ¨ r rr r × r × dxiv <i. if V (ϕ) = gn ϕn /n!. Furthermore. each source with momentum ˜ k ﬂowing in the source by a factor J (k). This is the reason why these vertices are weighed by a factor 1/n! in the potential and hence by a factor 1/(n − 1)! in the equations of motion (to be precise. gn for an npoint function).e. a factor 1/(2π)2 (for an npoint function a factor (2π)4−2n ) and a momentum conserving delta function. by inserting the Fourier expansion of each of the terms that occur. is that each of these come with their own coupling constant (i. Each propagator. As each line entering the vertex carries a Green’s function that depends 15 . {ks } {iv } G(xi − xj ) (4. such that only three point vertices appear. whereas ks runs only over positions of the sources. being the inverse of the order of the permutation group (interchange of lines and vertices) that leaves the diagram invariant (which is also the number of ways the diagram can be constructed out of its building blocks). j > stands for the pairs of points in a diagram connected by a line (called propagator). coordinate space ≡ g d4 x y × x ≡ G(x − y) ≡ d4 xJ(x) ' momentum space T ¨r ¨r % j ¨r table 1 ki ) vertex propagator source k1 ¨r ¨x r k3 k2 ≡ ≡ g δ( (2π)2 4 i x k × k 1 d4 k k2 −m2 +iε ˜ ≡ J(k) It is straightforward to translate these Feynman rules to momentum space. and the factor (n − 1)! is part of the combinatorics involved in interchanging each of the n − 1 factors ϕ in the interaction term.17) Here the index iv runs over all vertices and sources (so it does not label the four spacetime components of a single point. λ for a fourpoint vertex).j> r × ¨ ¨¨ ¨ r× + r r r × + 1 8 r × ¨¨ × ¨ r r ¨ × rr × ··· J(xks ) . which carries a momentum k is replaced by a factor 1/(k 2 − m2 + iε) and d4 k. for each vertex at a point x we insert d4 x and a power of the coupling constant g. Finally each diagram comes with an overall factor 1/N(diagram). All that is required to generalise this to the arbitrary case with npoint vertices. the equation of motion gives 2 ∂µ ϕ(x) + J(x) = −gn ϕ(n−1) (x)/(n − 1)!. The Feynman rules to convert a diagram to the solution are apparently that each line (propagator) between points x and y contributes G(x − y) and each cross (source) at a point x contributes J(x). each vertex by a factor of the coupling constant (i. We have derived these rules for the case that λ = 0. The expression < i.e. frequently it will be assumed that it is clear from the context what is meant).
which can be ﬁxed to a particular value by a gauge transformation. As a last example in this section we will look again at the electromagnetic ﬁeld (whose particles are called photons).on that position (this being the only dependence). To impose the gauge ﬁxing. (4. ˜ because the current is conserved.20) The component of Aµ in the direction of kµ is for obvious reasons called the longitudinal component. we can add a term to the Lagrangian which enforces the gauge condition. and the gauge choice is prescribed by the gauge condition. we see that a vertex at the point x gives rise to d4 x α G(x − xα ) → d4 x α = (2π)4 α d4 kα e−ikα (x−xα ) 2 (2π)4 (kα − m2 + iε) d4 kα eikα xα δ4 ( 2 (2π)4 (kα − m2 + iε) kα ) . Without such a term the action is stationary under any longitudinal variation δAµ (x) = ∂µ Λ(x) of the vector ﬁeld. This is a direct consequence of the gauge invariance as the gauge transformation of eq. The equations of motion for this action now yield ∂µ F µν (x) + α∂ ν ∂µ Aµ (x) = J ν (x) . Needless to say that the ﬁnal answers have to be independent of the chosen conventions. α (4.24) . the choice of gauge has no eﬀect on the equations of motion. the source) does not couple to the unphysical longitudinal component of the photon ﬁeld. For any choice of the parameter α = 0 this is achieved by the action S= 1 d4 x − 4 Fµν (x)F µν (x) − 1 α(∂µ Aµ (x))2 − Aµ (x)J µ (x) 2 . The current (i.21) Because of the gauge invariance. 16 (4. or k µ Jµ (k) = 0. (4. An important example is the socalled Lorentz gauge ˜ ∂µ Aµ (x) = 0 or kµ Aµ (k) = 0 .19) ν Unfortunately the matrix −k 2 δµ + kµ k ν has no inverse as k µ is an eigenvector with zero eigenvalue. Fixing the longitudinal component of the electromagnetic ﬁeld (also called photon ﬁeld) is called gauge ﬁxing.18) Conventions in the literature can diﬀer on how the factors of i (which will appear in the quantum theory) and 2π are distributed over the vertices and propagators. In Fourier space the equations of motion are given by ν ˜ ˜ (−k 2 δµ + kµ k ν )Aµ (k) = J ν (k) .23) which implies the Lorentzgauge (assuming vanishing boundary conditions for the vector potential at inﬁnity).e. It stresses again the importance of gauge invariance and its associated conservation of currents. and the added term should be such that stationarity in that direction imposes the gauge condition.32) in Fourier language reads ˜ ˜ ˜ Aµ (k) → Aµ (k) + ikµ Λ(k) . (4. (4. (3. the variation δAµ (x) = ∂µ Λ(x) in the longitudinal direction leads to the equation −α d4 x ∂µ ∂ µ Λ(x)∂ν Aν (x) = 0 . (4.22) Indeed.
but now with a unique speciﬁcation of the required boundary conditions following from the time ordering in the quantum evolution equations. (4. t = 0) = . In the quantum theory this is not entirely trivial. ¯ (5. in deriving scattering amplitudes and cross sections. Also the Green’s function will reappear. H1 (t) = V d3 x J(x. a(k)eik·x + a† (k)e−ik·x 2V ko (k) k π(x.25) µ δν − k kν (1−α ) ν k 2 +iε ˜ J (k) . For the Lagrangian 1 L = 2 ∂µ ϕ∂ µ ϕ − 1 m2 ϕ2 − εJϕ . One needs to ﬁx the gauge to perform perturbation theory and then one has to prove that the result does not depend on the choice of gauge ﬁxing. like for the derivation of the classical perturbation theory of the previous section. Fortunately. however. ˜µ 5 Hamiltonian perturbation theory the Hamiltonian is given by We consider the Hamiltonian for a free scalar particle coupled to a source. ¯ 2 (5. there are regularisations that preserve the gauge invariance. Like in the scalar case it can be used to perform a perturbative expansion for the classical equations of motion. (5. k = 2πn/L. In the presence of fermions the situation can. This is the hard part in gauge theories. as the regularisation can break the gauge invariance explicitly. such that the momenta are discrete.26) A (k) = − k 2 + iε This is consequently the propagator of the electromagnetic ﬁeld (in the Lorentz gauge).4) . as long as α = 0.or in Fourier space which is invertible. Some diﬀerent choices of gauge ﬁxing will be explored in problems 8 and 9. We will see that the source can be used to create particles from the vacuum in the quantum theory. t) a(−k) + a† (k) .1) 2 1 1 H = 2 π 2 + 2 (∂i ϕ)2 + 1 m2 ϕ2 + εJϕ . Note that the photon propagator simpliﬁes dramatically if we choose α = 1. 1 ϕ(x. also simply called the photon propagator. The result is given by µ −1 ν ˜ ˜ (−k 2 δµ + (1 − α)kµ k ν )Aµ (k) = J ν (k) . (4. be considerably more tricky. L]3 with periodic boundary conditions. We have o H0 = k 1 k0 (k)(a† (k)a(k) + 2 ) . (5. We will quantise the theory in a ﬁnite volume V = ¯ [0. t)ϕ(x) = k 1 2k0(k) 17 ˜ J (k. and we work out the perturbation ¯ theory in the Schr¨dinger representation.3) k The Hamiltonian is now given by H(t) = H0 + εH1 (t). and it forms an important ingredient.2) where ε is a small expansion parameter. t = 0) = −i ko (k) a(k)eik·x − a† (k)e−ik·x 2V . like dimensional regularisation which we already encountered in section 2 (in discussing the Casimir eﬀect). but all ﬁnal results should be independent of the choice of α and even of the choice of gauge ﬁxing all together.
e. which is essential as H1 (t) does not commute with H1 (t′ ) for diﬀerent t and t′ . t1 )J(k. or 1 J(x. (5. t2 ) < 0 a(−p) + a† (p) ei(H0 −E0 )(t2 −t1 ) a(−k) + a† (k) 0 > 4k0 (p)k0 (k) k.8) 0 ˆ dt1 eiH0 t1 H1 (t1 )e−iH0 t1 Ψn−1 (t1 ) >= · · · Please note the time ordering. < 0a† 0 >=< 0a0 >= 0. (5. which has the property that a(k)0 >= 0 for all momenta. then it follows that d Ψ(t) >= −iH(t)Ψ(t) > . by transforming to Ψ(t) > we are using the interaction picture.6) εn Ψn (t) > .7) d ˆ ˆ Ψn (t) > = −ieiH0 t H1 (t)e−iH0 t Ψn−1 (t) > dt ˆ Actually. t). dt which can be evaluated by perturbing in ε. ¯ ˆ Ψ(t) > ≡ e−iH0 t Ψ(t) > . t) is the Fourier coeﬃcient of J(x.5) Let us start at t = 0 with the vacuum state 0 >. ˆ Ψ(t) >= ∞ n=0 (5. (5. which will often be assumed to vanish) ˆ < 0Ψ(t) > = e−iE0 t < 0Ψ(t) >= e−iE0 t {1 − i¯ ε − ε2 ¯ t 0 t 0 dt1 < 0H1 (t1 )0 > dt1 t1 0 dt2 < 0H1 (t1 )ei(H0 −E0 )(t2 −t1 ) H1 (t2 )0 > +O(¯3 )} . (5. (5. . To evaluate the remaining expectation value in the above equation we substitute H1 in terms of the creation and annihilation operators (see eq. t2 ) −ik0 (k)(t1 −t2 ) e .10) = 2k0 (k) k 18 . which is the usual way of performing Hamiltonian perturbation theory known from ordinary quantum mechanics.4)) < 0H1 (t1 )ei(H0 −E0 )(t2 −t1 ) H1 (t2 )0 >= ˜ ˜ J(p. i. as the vacuum expectation values of ¯ the creation and annihilation operators vanish. We can for example compute the probability that at time t Ψ(t) > is still in the ground state (whose energy we denote by E0 .p ˜ ˜ J(−k. ¯ ˆ . t1 )J(k. t)eik·x k . ˆ dt1 eiH0 t1 H1 (t1 )e−iH0 t1 Ψ1 (t1 ) > t1 0 dt1 eiH0 t1 H1 (t1 ) t dt2 eiH0 (t2 −t1 ) H1 (t2 )e−iH0 t2 0 > . These equations can be solved iteratively as follows ˆ Ψ1 (t) > = −i ˆ Ψ2 (t) > = −i = − ˆ Ψn (t) > = −i t 0 t 0 t 0 dt1 eiH0 t1 H1 (t1 )e−iH0 t1 0 > .˜ here J(k. t) = √ V ˜ J(k.9) ε It is simple to see that the term linear in ε will vanish. (5.
t1 )J(k. (5.11) as i lim < 0Ψ(t) > eiE0 t = 1 − ε2 ¯ t→∞ 2 where 1 ˜ J(ω. −k)J(ω. Instead. Equivalently we can start at t = −∞ and integrate the quantum equation of motion up to t = ∞. t)eiωt = √ 2πV I R×V d3 xdt J(x)eikx . for t < 0 the contour needs to be deformed to the lower halfplane and the pole at ω = ω+ ≡ k0 (k) − iε contributes with the residue 2πieik0 (k)t /(−2k0 (k)) (note that the contour now runs clockwise. t2 ) Especially the last identity is useful to relate this to the Green’s function we introduced in the previous section.Combining these results we ﬁnd < 0Ψ(t) > eiE0 t = 1 − ε2 ¯ ¯ = 1 − 1 ε2 2 t k 0 dt1 t t1 0 t 0 dt2 dt2 ˜ ˜ J (−k. k) + O(¯3 ) . 9' 6 t>0 E r ω− r 8' E r ω+ 7 ω− r t<0 ﬁg. It is important to note that we have chosen J(x) = 0 for t < 0. t1 )J(k. The socalled iε prescription. with a residue 2πie−ik0 (k)t /(−2k0 (k)) (see the ﬁgure).15) where G(x − y) is exactly the Green’s function we introduced in the previous section. ε (5. We have to require that J(x) vanishes suﬃciently rapidly at inﬁnity.12) This can be shown as follows. (5. we can deform the contour of integration to the upper halfplane (where eiωt decays exponentially) and only the pole at ω = ω− ≡ −k0 (k)+iε contributes.14) The last expression should be replaced by (2π)−2 IR4 d4 x J(x)eikx in case the volume is inﬁnite. In an inﬁnite volume we therefore ﬁnd for what is known as the vacuum to vacuum amplitude of the scattering matrix i ¯ lim < 0Ψ(t) > eiE0 t = 1 − ε2 t→∞ 2 i = 1 − ε2 ¯ 2 d4 k ˜ ˜ J(−k)J (k) + O(¯3 ) ε k 2 − m2 + iε d4 xd4 y J(x)G(x − y)J(y) + O(¯3 ) . k) = √ 2π ∞ k −∞ dω ω 2 − k 2 − m2 + iε ˜ ˜ J(−ω. which is equivalent with specifying the boundary conditions. When t > 0. ε (5. giving an extra minus sign). has 19 . Using contour deformation in the complex ω plane we ﬁnd ∞ −∞ dω eiωt ω2 − 2 k0 (k) + iε =− 2πi 2k0 (k) e−ik0 (k)t . t2 ) 2k0 (k) 2k0 (k) e−ik0 (k)(t1 −t2 ) + O(¯3 ) ε e−ik0 (k)t1 −t2  + O(¯3 ).11) ε k 0 dt1 ˜ ˜ J(−k.13) 1 ˜ dt J(k. 4 ω+ This means that we can rewrite eq. (5. (5.
(6. 6 Path integrals in quantum mechanics p2 ˆ + V (ˆ) . ε εJ ∗ ¯ (5.4) i ∂x An important ingredient in deriving the path integral expression will be the completeness relations ˆ = dx x >< x and ˆ = dp p >< p .6) 20 . (6. For a complex scalar ﬁeld.3) h ∂ ¯ < xp >= p < xp > . ε εJ ¯ (5.5) For arbitrary N we can use this to write h h < x′ e−iHT /¯ x > = < x′  e−iHT /N ¯ N Indeed. h In the coordinate representation its matrix is given by h < x′ e−iHT /¯ x > . This is why in this case the propagator has a direction. to distinguish them from numbervalued coordinate and momentum. We wish to study the timeevolution operator exp(−iHT /¯ ). (6. i.2) where x > is the position eigenfunction. whose wave function in the coordinate space is given by ˆ h eipx/¯ < xp >= √ 2π¯ h . (6. ϕ and ϕ∗ are independent and we need to introduce two sources by adding to the Lagrangian −ϕJ ∗ − ϕ∗ J (see problem 17). Note that we can use the diagrams introduced in the previous section to express this result (taking E0 = 0 from now on) as < 0Ψ(t) >= 1 − i 2 × εJ ¯ × + O(¯3 ) . We will also need the momentum eigenfunction p >.17) without a factor of one half because the sources J and J ∗ are independent and cannot be interchanged. x 2m h ∂ ¯ i ∂x For simplicity we will start with a onedimensional Hamiltonian H= p= ˆ .therefore been derived from the timeordering in the Hamiltonian evolution of the system and is thus prescribed by the requirement of causality. one veriﬁes that x >= ··· h < x′ e−iHT /N ¯ xN −1 > dxN −1 ¯ h < xN −1 e−iHT /N h xN −2 > dxN −2 < xN −2  · · · · · · e−iHT /N ¯ x2 > dx2 h h < x2 e−iHT /N ¯ x1 > dx1 < x1 e−iHT /N ¯ x > . pp >= pp >. It is not too diﬃcult to show that in this case < 0Ψ(t) >= 1 − i × εJ ¯ > × + O(¯3 ) .1) where we have indicated a hat on top of operators. (6.e. 1 1 (6.16) where the factor of a half is a consequence of the symmetry under interchanging the two sources.
7) for two operators A and B.We will now use the socalled Trotter formula e−i(A+B)/N = e−iA/N e−iB/N (1 + O(N −2 )) (6.8) By inserting the completeness relation for the momentum we can eliminate the operators h < xi+1 e−iHT /N ¯ xi > = h dpi < xi+1 pi >< pi e−iHT /N ¯ xi > p h x h dpi < xi+1 pi >< pi e−iˆ T /2mN ¯ e−iV (ˆ)T /N ¯ xi > h h dpi < xi+1 pi >< pi e−ipi T /2mN ¯ e−iV (xi )T /N ¯ xi > 2 2 ≈ = = dpi h eipi (xi+1 −xi )/¯ −i( p2 +V (xi ))T /N ¯ i h e 2m 2π¯ h . The integrals in the path integral are strongly oscillating and can only be deﬁned by analytic continuation.13) This is the deﬁnition of the path integral.1) this can be used to write for N → ∞ h p2 h x h e−iHT /N ¯ = e−iˆ T /2mN ¯ e−iV (ˆ)T /N ¯ . Writing ∆t = T /N. For ∆t = T /N ≡ −iT /N = −i∆τ the Gaussian integral over the momenta is easily evaluated ∞ −∞ dpi exp ipi (xi+1 − xi ) p2 ∆τ = − i h ¯ 2m¯ h 1 N 2 2πm¯ /∆τ exp − 1 m h 2 (xi+1 − xi )2 ∆τ h ¯ . which will be introduced later). (6.11) which leads to < x e ′ −iHT /¯ h x >= lim N →∞ mN 2πT h ¯ N −1 i=1 dxi exp − ∆τ h ¯ N −1 i=0 m(xi+1 − xi )2 + V (xi ) 2∆τ 2 (6. (6. (6. but formally it will often be written as < x e ′ −iHT /¯ h x >= lim mN 2πiT h ¯ 1 N 2 N −1 dxi exp i < x e ′ −iHT /¯ h x >= x(T )=x′ x(0)=x Dx(t) exp[iS/¯ ] . xN = x′ and x0 = x we ﬁnd N −1 N −1 N −1 < x e ′ −iHT /¯ h x > = dp0 2π¯ h N →∞ N −1 i=1 lim ··· dxi i=1 j=0 dpj i=0 N −1 i=0 h < xi+1 pi >< pi e−iH∆t/¯ xi > = lim N →∞ dxi dpi i∆t exp 2π¯ h h ¯ p2 pi (xi+1 − xi ) − i − V (xi ) ∆t 2m .10) It is important to observe that there is one more p integration than the number of x integrations.9) This can be done for each matrix element occurring in eq. (6. (6. and the error term is actually of the form [A. As parameter for this analytic continuation one chooses the time t. h 21 S= T 0 dt 1 2 mx2 (t) − V (x(t)) ˙ . (6.12) or after substituting T = iT we ﬁnd ∆t N −1 m(xi+1 − xi )2 − V (xi ) N →∞ h i=0 ¯ 2∆t2 i=1 (6. B]/N 2 .6). (One can also use the CampbellBakerHausdorﬀ formula. This can be seen by expanding the exponents. With the Hamiltonian of eq.14) . (6.
Glimm and A. what we will compute is the quantum partition function for the harmonic oscillator. the integral is only deﬁned through the analytic continuation in time.. We will now do an exact computation to give us some conﬁdence in the formalism. 1987). 2 mN ZN ≡ 2πT h ¯ 1 N 2 ∞ N −1 i=0 −∞ ··· ∆τ dxi exp − h ¯ N −1 1 2 m i=0 xi+1 − xi ∆τ 2 + 1 mω 2 x2 i 2 . with T = −iT . New York. In the continuous formulation we therefore have ¯ h Tr(e−T H/¯ ) = x(T )=x(0) Dx(τ ) exp[−SE /¯ ] h (6. but formal manipulations will be much easier to perform in this continuous formulation.16) It is the quantum thermal partition function (the Boltzmann distribution) with a temperature of h/kT .20) 22 . Springer. x ¡e ¡ e h ¡ e ' E ¤f £ £ ¤ f £ ¤ f £ f ¤ f £ ¤ d d£ ¤ i i i ¡f ¡ f h h i ∆t h h i h ¤ h i i ¡ i ¡ i¡ ¡d ¡ d¡ ¡ ¢ ¢ f ¢ f f ¢ d d¢ t ﬁg. has an important physical interpretation h dx < xe−HT /¯ x >= Tr(e−βH )β=T /¯ h . 5 For more details on this see “Quantum Physics: A functional integral point of view”. Furthermore.since the discretised version of the action with xj ≡ x(t = j∆t) is precisely N −1 Sdiscrete = ∆t i=0 1 2 m (xi+1 − xi )2 − V (xi ) ∆t2 . (6.x′ (T ) ≡ x(T )=x′ x(0)=x Dx(τ ) exp(− T 0 1 2 mx(τ )2 dτ /¯ ) . if we integrate over xN = x0 this analytically continued path integral. To be speciﬁc. Jaﬀe (2nd ed. by J.17) in which SE is the socalled Euclidean action SE = T 0 dτ 1 2 m(dx(τ )/dτ )2 + V (x(τ )) . (6.18) It is only in this Euclidean case that one can deﬁne the path integral in a mathematically rigorous fashion on the class of piecewise continuous functions in terms of the socalled Wiener measure dWx. (6.15) It is important to note that the continuous expression is just a notation for the discrete version of the path integral. ˙ h (6. (6.19) meaning that this measure is independent of the way the path is discretised. However. where V (x) = 1 mω 2 x2 .
(6. (6. from the deﬁnition of ZN .23) It is easy to verify that the Jacobian for the change of variables yi → bℓ is 1.21) −∞ ··· N −1 dy 1 1 √ i exp − (yi+1 − yi)2 + 2 ω 2 yi ˜ 2 2 2π i=0 . We start with the identity log(λ/µ) = − lim such that log[ZN (˜ )/ZN (˜ 0 )] = 1 lim ω ω 2 aց0 0 ∞ aց0 0 ∞ ds sa−1 (e−sλ − e−sµ ) . The path involved is thus periodic in time. (6.Note that we have now N integrations.22) We can diagonalise the quadratic term by using Fourier transformation 1 yk = √ N N −1 bℓ e2πikℓ/N . because we also integrate over x0 = x(0) = x(T ) = xN to implement the trace.25) We can convert the product to a sum using a Laplace transform. ω ) − Q(s. (6.26) (6. The integral can now easily be evaluated N −1 ℓ=0 ZN = 4 sin2 (πℓ/N) + ω 2 ˜ 1 −2 . (6.30) 1 2π dθ eiN kθ e2s cos(θ) ≡ IN k (2s) .27) We read oﬀ. ZN = ∞ N −1 ℓ=0 −∞ ··· N −1 db 1 √ ℓ exp − 2 4 sin2 (πℓ/N) + ω 2 bℓ 2 ˜ 2π ℓ=0 . ℓ=0 b∗ = bN −ℓ .29) This is a periodic function with period 1 (fs (x + 1) = fs (x)) and its discrete Fourier coeﬃcients can be computed exactly ˜ fs (k) = = 1 0 dx e2πikx fs (x) = 2π 0 1 N N 0 dx e2πikx e2s cos(2πx/N ) (6. 23 . and one obtains a result that must look familiar from the classical small oscillations problem for a ﬁnite number of weights connected by strings. ω = ω∆τ ˜ . We now rescale yi = xi to obtain the simple result ZN = ∞ N −1 i=0 m ∆τ h ¯ 1 2 . we mean by dbℓ = dRebℓ dImbℓ .28) fs (x) = e2s cos(2π(ℓ+x)/N ) ℓ=0 .24) Note that if bℓ is complex. (6. ds s−1+a (Q(s. (6. (6. that Q is a sum of exponentials N −1 Q(s. a general feature of the expression for the quantum partition function in terms of a path integral. ω0 )) ˜ ˜ . ω ) ≡ ˜ where ℓ=0 exp −s 4 sin2 (πℓ/N) + ω 2 ˜ 1 N N −1 ω ≡ Ne−s(˜ 2 +2) fs (0) . ℓ b∗ = b0 0 .
ω ω We therefore ﬁnd (using β = T /¯ = N∆τ /¯ ) h h ∂ F (˜ ) = hω ω ¯ ∂ω ˜ k∈Z Z = ∞ 0 ω ds e−(2+˜ 2 )s (6.27). The advantage of these manipulations is that the Laplace transform of this Besselfunction is know (see the same reference) √ ∞ [λ + λ2 − 1]−ν −λs √ .Please note that we have exchanged the sum over ℓ with extending the integration of x to the interval [0. “Handbook of mathematical functions”. (6. by M. where the free energy F is deﬁned as ZN (˜ ) = exp[−βF (˜ )] . 1978). (6.35) e−N Ωk = k∈Z Z 1 − 1 − e−N Ω 1 2 =− 1 ∂ e−ΩN/2 log N ∂Ω 1 − e−N Ω . New York. see eqs.31) ds e Iν (s) = 0 λ2 − 1 and as we can express fs (x) as a sum over these Besselfunctions fs (x) = k∈Z Z e−2πikx IN k (2s) . but to make the result more transparent we introduce the scaled eﬀective frequency Ω ωT /N = ω∆τ = ω ≡ 2 sinh( 1 Ω) ˜ 2 and using the identity 1 ω 2 + 1 = cosh(Ω).33) IN k (2s) 1 2 hω ¯ 1 2 ( 2 ω 2 + 1)2 − 1 k∈Z ˜ Z ω2 + 1 + ˜ −N k 1 ( 2 ω 2 + 1)2 − 1 ˜ . (6.37) Amazingly. as it can also be written as 1 ∂ ∆τ ∂ F (Ω) = − log h ∂Ω ¯ N ∂Ω ∞ n=0 e−(n+ 2 )ΩN 1 . (6.34) The geometric series is of course easily summed. The last identity is one of the deﬁnitions of the modiﬁed Besselfunction. see e.32) this allows us to evaluate eq. It is trivial to check now that the limit N → ∞ is well deﬁned and gives the required result. since N →∞ lim ΩN/T = ω . Stegun (Dover.21) and (6.35). (6. one can evaluate this exact expression in relatively simple terms (which will verify the above 24 . but with a frequency that is modiﬁed by the discretisation. even at ﬁnite N the euclidean path integral agrees with the quantum partition function of a harmonic oscillator. (6.g.38) In general the exact ﬁnite N path integral is no longer of a simple form. (6. one easily ﬁnds that ˜ 2 ∆τ ∂ F (Ω) = h ∂Ω ¯ 1 2 (6. Abramowitz and I. Nevertheless. N]. For technical reasons it is easier to compute the variation of the free energy with the frequency. (6.36) The last identity can be seen as the free energy of the harmonic oscillator with the frequency ΩN/T = Ω/∆τ .
[B. A]]+ · · · 12 12 . such that UH1 U † = H2 .B) . This can be corrected in two ways e −H1 (N )∆τ /¯ h h e−H2 (N )∆τ /¯ ∆τ p2 ˆ ∆τ V (ˆ) x ∆τ p2 ˆ ≡ exp − exp − exp − .41) leading to two equivalent expressions for the ﬁnite N path integral ZN (x′ . working out the products of the exponential can be done to all orders and one ﬁnds (see problem 10 for details) ∆τ Hi (N) = p2 ˆ + 1 Mi Ω2 x2 ˆ 2Mi 2 25 . B]]+ [B. For the harmonic oscillator. ZN (x′ . see also problem 10). 18. x. These objects will be discussed in sect. (6. [A. which shows that both choices are indeed physically equivalent. T ) = Tr(e−H1 (N )T /¯ ) = Tr(e−H2 (N )T /¯ ) .42) .40) But this Hamiltonian is not hermitian as one easily checks from the above expression. (6. F (A.43) It is actually not too diﬃcult to show that there exists a unitary transformation U. T ) = = dp0 i=1 ′ dp0 N −1 2π¯ i=1 h N −1 dxi dpi ∆τ exp 2π¯ h h ¯ N −1 N −1 i=0 ipi (xi+1 − xi ) p2 − i − V (xi ) ∆τ 2m ∆τ p2 ˆ ∆τ V (ˆ) x exp − xj > 2m¯ h h ¯ x > .result along a diﬀerent route. (6. but in the mathematics literature more elegant constructions are known. T ) = < x′  exp − ∆τ p2 ˆ ∆τ p2 ˆ exp (−H1 (N)T /¯ ) exp h x > 4m¯ h 4m¯ h ∆τ V (ˆ) x ∆τ V (ˆ) x exp (−H2 (N)T /¯ ) exp − h x > (6. since under conjugation the order of the exponents containing the kinetic and potential terms is reversed. x. T ) = < x′  exp 2¯ h 2¯ h In particular the partition function is given by ZN = h h dx ZN (x. by using the socalled CampbellBakerHausdorﬀ formula eA eB = eF (A. B) = A+ B + 1 [A. operators A and B. x. (6. So from now on we will take the potential arbitrary and in a sense we follow the derivation of the path integral in the reverse order. in general noncommuting.45) . x.44) which is a series in multiple commutators of the. In principle we can now compute Hi (N) for ﬁnite N as an expansion in 1/N. B]+ 2 1 1 [A. ≡ exp − 2¯ h 2m¯ h 2¯ h (6. ZN (x′ . (6. 4m¯ h h ¯ 4m¯ h ∆τ p2 ˆ ∆τ V (ˆ) x ∆τ V (ˆ) x exp − exp − .39) dxi dpi j=0 < xj+1 pj >< pj  exp − ∆τ V (ˆ) x ∆τ p2 ˆ exp − = < x  exp − 2m¯ h h ¯ N This means that we can deﬁne an eﬀective Hamiltonian by e −H(N )T /¯ h ∆τ V (ˆ) x ∆τ p2 ˆ exp − ≡ exp − 2m¯ h h ¯ N . based on properties of Lie groups and Lie algebras. It can be derived by expanding the exponentials.
35) and the eﬀective masses Mi deﬁned by M1 = 2m tanh( 1 Ω) 2 . one ﬁnds that the action changes to SΛ = S + e{Λ(x(T )) − Λ(x(0))}.52) So. at least for some examples. however.49) which is equivalent to the stationary phase approximation for the momentum integration (6. Using the path integral this means that < x ′  exp(−iHΛ T /¯ )x >= exp(ieΛ(x ′ )/¯ ) < x ′  exp(−iHT /¯ )x > exp(−ieΛ(x)/¯ ) . x)) = x − δp ∂p m (6. We leave it as an exercise to verify that the action. or ¯ ˆ p p ˆ ˆ A(x) · ˆ = ˆ · A(x) + i¯ ∂i Ai (x) .46) One can now explicitly verify that (cmp.37)) Tr(exp(−H1 (N)T /¯ )) = Tr(exp(−H2 (N)T /¯ )) = exp(−F (Ω)T /¯ ) .54) 26 . h h h (6. the matrix element < pi  exp(−iH∆t/¯ )xi > will depend on the speciﬁc h ordering for the position and momentum operators in H. ˙ (p · x − H(p. Diﬀerent orderings diﬀer by terms linear in h. is given by S= T 0 dt 1 2 ˙ ˙ mx 2 − V (x) + ex · A(x) .with Ω deﬁned as in eq. h (6.50) An interesting other example of the path integral is the case of the interaction of a charged particle with a magnetic ﬁeld. (xi+1 − xi )/∆t) h i=0 ¯ . 2m 2m 2 δ ˙ ˙ ∂H = x − p = 0 . h h h h (6. 2m (6. as it shows that the Gaussian integration over the momenta eﬀectuates the Legendre transform ˙ ˙ p2 (p − mx)2 mx 2 ˙ ip · x − i − iV (x) = −i +i − iV (x) . (6. by chosing the socalled Coulomb gauge ∂i Ai (x) = 0. h < x ′ e−iHT /¯ x > ≡ Dx(t) exp d n p0 (2π¯ )n h mN 2πiT h ¯ N −1 i=1 1 nN 2 i h ¯ T 0 ˙ dt L(x(t). eq. ˆ + V (x) . xi ) n (2π¯ ) h h i=0 ¯ ∆t N −1 = lim N →∞ dn xi exp i i=1 ∆t N −1 L(xi . (6.51) Now. the limit of increasingly ﬁner discretisation is in principle well deﬁned. ∆τ Ω M2 = m sinh(Ω) ∆τ Ω . In that case one has for the Hamiltonian ˆ (ˆ − eA(x))2 p ˆ H(ˆ x) = p. (6.53) Under a gauge transformation Ai (x) → Ai (x) + ∂i Λ(x). obtained from the Legendre transform.48) We have purposely also given the expression that involves the path integral as an integral over phase space. x(t)) = lim N →∞ (xi+1 − xi ) dn xi dn pi ∆t N −1 pi · exp i − H(pi. the problem of the operator ordering disappears.47) Now we have seen that. (6. (6. we can think of generalisation to an arbitrary number of dimensions (n) (for ﬁeld theory even to an inﬁnite number of dimensions).
also for the coordinate integrals we can use the stationary phase approximation (exact for a harmonic oscillator). As long as the Hamiltonian is quadratic in the momenta. and as nonquadratic path integrals can rarely be computed explicitly. in terms of which one can eﬃciently perform the perturbative computations. (7. and considers the rest as a perturbation. this proves that the path integral derived from eq. splits the action in a classical piece. It gives a way of deﬁning an expansion in h. such that S(x) = S(xcl ) + 1 2 q(0) = q(T ) = 0 .55) is precisely solved by the classical solutions. where in accordance to the correspondence principle. xcl (t). In that case xcl is the classical solution of the quadratic part only. (6. L]3 the Hamiltonian is given in the Fourier representation by (see eq.57) There is no term linear in q i (t). Mij (t) ≡ ∂ 2 V (x)  ∂xi ∂xj x=xcl(t) .56) − δ S(xcl ) dt′ dt q i (t) − i − j ′ q j (t′ ) + O(q 3 ) . Indeed the stationary phase condition − δS − xi (t) δ = δS d δS − =0 i (t) δx dt δ xi (t) ˙ (6. 2 (6. where the stationary phase approximation is 3 exact.58) 1 For the harmonic potential. which is related to the WKB approximation in quantum mechanics. this will be the way in which we will derive the Feynman rules for the quantum theory.1) 27 . as this term is proportional to the equations of motion. However. Introducing q(t). (6. (2. For the simple Lagrangian L = 2 mx 2 − V (x) one has δ S(xcl ) d2 = −δ(t − t′ ) mδij 2 + Mij (t) − q i (t)δ q j (t′ ) − δ dt −2 . or ˙ 1 equivalently to the stationary phase condition. As this can always be solved exactly.6)) H= k=2πn/L 1 2 1 ˜ ˜ (k)2 + 2 (k 2 + m2 )ϕ(k)2 + V (ϕ) + ϕ(k)J(−k. i. 7 Path integrals in ﬁeld theory For a scalar ﬁeld in a ﬁnite volume V = [0. the stationary phase approximation for the momentum integral is exact.ˆ h ˆ h Since it is easily shown that HΛ = exp(ieΛ(x)/¯ ) H exp(−ieΛ(x)/¯ ).53) has the correct properties under gauge transformations. there are no O(q ) corrections. that is independent of the boundary conditions and the classical path S(x) = S(xcl ) + T 0 ˙ 1 dt ( 1 mq 2 − 2 mω 2 q 2 ) . V = 2 mω 2 x2 . despite the fact that the derivation was performed by ﬁrst going to the Coulomb gauge. We expand around these solutions by writing x(t) = xcl (t) + q(t) . and a quantum piece described by a harmonic oscillator action for the ﬂuctuations around the classical path. δ q (t)δ q (t ) (6. that depends on the boundary conditions. with xcl (0) = x and xcl (T ) = x′ . 2 (6. the lowest order term reproduces ¯ the classical time evolution.e. t) π ˜ ˜ ˜ . however.59) In practical situations one splits from the action the part quadratic in the coordinates and velocities.
This achieves a substantial simpliﬁcation over Hamiltonian perturbation theory.e. The class of renormalisable ﬁeld theory is relatively small. In either case. such that the potential V (ϕ) only contains the interaction terms. The introduction of a cutoﬀ is called a regularisation. by adding the potential term to the Lagrange density. 0) and ϕ(x.2) ˜ − 1 (k 2 + m2 )ϕj (k)2 − ϕj (k)J(−k. t)2 − ϕ(k. ˜ ˜ ˜ k One of course identiﬁes ϕj (k) = ϕ(k. (7.3) The last expression is manifestly Lorentz invariant apart from the dependence on the boundary conditions on the ﬁelds (which should disappear once we take L and T to inﬁnity). and we ﬁnd in yet another shorthand notation for the measure the following expression for the path integral (implicitly assuming that the boundary values ϕ(x. t))2 − 1 m2 ϕ2 (x. t) exp i ˜ T 0 1 2 1 ˜ ˙ ϕ(k. t)J(−k. for V (ϕ) = 0 the Hamiltonian simply describes an inﬁnite set of decoupled harmonic oscillators. which can be truncated to a ﬁnite set by introducing a socalled momentum cutoﬀ k ≤ Λ. T ) are ﬁxed.4) In principle a path integral should be independent of the discretisation used in order to deﬁne it. such that the Fourier modes are given by j sin(πnj xj /L) (with nj > 0).As ϕ(x) is real we have ϕ∗ (k) = ϕ(−k). For the euclidean path integral. i. t)J(x. t))2 − 2 (∂i ϕ(x. It is now also trivial to reintroduce the interactions. we can write T 0 dt k 1 2 1 ˜ ˙ ϕ(k.T ] . t)2 − 2 (k 2 + m2 )ϕ(k. often by varying the parameters in a suitable way with the cutoﬀ. Or even simpler is to use Dirichlet boundary conditions. prescribed functions) Z= Dϕ(x) exp i d4 x 1 2 1 ∂µ ϕ(x)∂ µ ϕ(x) − 2 m2 ϕ2 (x) − V (ϕ) − ϕ(x)J(x) V ×[0. taking the limit Λ → ∞). t = j∆t) and performing the Fourier transformation ˜ ˜ once more. It is customary to write the term quadratic in the ˜ ˜ ﬁelds (the mass term) explicitly. one particular way that is used quite often is 28 .T ] d4 x 1 ∂µ ϕ(x)∂ µ ϕ(x) − 2 m2 ϕ2 (x) − ϕ(x)J(x) . or in the absence of interactions N −1 N −1 Z = N →∞ lim (2πi∆t)−N/2 k j=1 k dϕj (k) exp i∆t ˜ j=0 k ϕj+1(k) − ϕj (k)2 ˜ ˜ 2 2∆t (7. t) − ϕ(x. If we like we could split the Fourier modes in their real and imaginary components (the cos(x · k) and sin(x · k) modes). This will allow us to perform perturbation theory in a Lorentz covariant way (things are somewhat subtle as any ﬁnite choice of the momentum cutoﬀ does break the Lorentz invariance. t)J(−k. (7. even in the presence of interactions. ϕ(x) = 0 at the boundaries of the volume. t) dt . and there are some theories where this is not restored when removing the cutoﬀ. In this case we know how to write the path integral. t) ˜ ˜ ˜ = = T 0 dt V d3 x 1 2 1 2 1 (∂t ϕ(x. i. j∆t) ˜ ˜ 2 ≡ D ϕ(k. with real coeﬃcients. t)2 − 2 (k 2 + m2 )ϕ(k. t)2 − ϕ(k. For a ﬁnite momentum cutoﬀ the path integral is nothing but a simple generalisation of the one we deﬁned for quantum mechanics in n dimensions.e. t) 2 V ×[0. The ﬁeld theory is called renormalisable if the limit Λ → ∞ can be deﬁned in a suitable way.
dt its solution can be written as (note the absence of 1/n!) i U(t) = Texp −i t 0 ∞ (7. denoted by Ψ0 ({ϕ(k)}) =< {ϕ(k)}0 >. t2 > t1 .the lattice discretisation.11) 29 . ˜ ˜ ˜ ˜ ˜ ˜ ′ ′ (7. 2 2 The vacuum 0 > is the state where all k oscillators are in their ground state. L]3 ). (7. In the presence of a source. t2 )U(t2 . U(0) = 1 . It turns out that we do not need an explicit expression for this vacuum wave functional. t1 ) . dϕj exp −a4 (2πa) 2 a2 µ j j (7. For the latter we will compute.9) In section 5 we calculated the matrix element < 0U(T )0 > to second order in the source (from now on we put ε = 1). V = a3 M 3 and T = aN. We have ˜ ˜ < 0U(T )0 >= p t2 t1 H(t)dt . with J(x) = 0 for t < 0 and t > T (and for x ∈ [0. the same quantity as was calculated in section 5. whereas the momentum cutoﬀ is suitable for performing perturbation theory around the quadratic approximation of the action. This leads to an integral of the form (ϕj+eµ − ϕj )2 1 2 2 1 + 2 m ϕj + V (ϕj ) + Jj ϕj . using the path integral. t3 > t2 > t1 . This means that the ﬁeld now lives on a lattice and its argument takes the values ja where j ∈ Z 4 and a is the socalled lattice spacing.k0 ˜ (k 2 − m2 )ϕ(k)2 − ϕ(k)J(−k) ˜ ˜ . By suitably restricting the components of j.6) H(t)dt ≡ (−i)n n=0 t 0 dt1 t1 0 ··· tn−1 0 dtn H(t1 ) · · · H(tn ) . (7. The Lagrangian relevant for the path integral evaluation is ¯ / given by L = 1 ∂µ ϕ∂ µ ϕ− 1 m2 ϕ2 −Jϕ. (7. The lattice formulation is very suitable for numerically evaluating the path integral.10) where {ϕ(p)} plays the role of x and {ϕ (p)} the role of x . The path integral expression for ˜ ˜ the evolution operator therefore becomes < {ϕ′ (p)}U(T ){ϕ(p)} >= ˜ ˜ = D ϕ(k) exp i ˜ Dϕ(x) exp i 1 2 T 0 V L(ϕ(x))d4 x k.8) dϕ(p)dϕ′ (p) < 0{ϕ′(p)} >< {ϕ′ (p)}U(T ){ϕ(p)} >< {ϕ(p)}0 > . t1 ) = U(t3 . In a sense. the Hamiltonian depends on time and the evolution operator has to be written in a way that takes the time ordering into account. one keeps space and time ﬁnite. t1 ) ≡ Texp −i which satisﬁes the property that U(t3 . (7. the momentum cutoﬀ Λ is similar to the spacetime cutoﬀ 1/a. and ϕj is identiﬁed with ϕ(aj). where instead of a momentum cutoﬀ one makes not only time but also space discrete.5) where eµ is a unit vector in the µ direction. which in the end should Z be taken to zero. with appropriate boundary conditions on the ﬁelds. using Hamiltonian perturbation theory.7) For convenience we introduce the notation U(t2 . As the time evolution operator U(t) satisﬁes the Schr¨dinger equation o −N M 3 /2 d U(t) = H(t)U(t) .
k0 × exp − i 2 k ˜ J(k)2 k 2 − m2 + iε . so that causality is also properly implemented in the path integral approach. playing such an important role in the Hamiltonian formulation. This prescription also allows us to make the analytic continuation to imaginary time. thereby converting the path integral measure D ϕ(k.12) ˜ We can now shift the integration of ϕ(k) over J(k)/(k 2 − m2 + iε) and introduce the Green’s ˜ function in coordinate space (eq. To recover the result obtained in the Hamiltonian approach. it will be helpful to understand in more detail how the source can be used to create and annihilate particles. (7. k) and k 2 = k0 − k 2 . see eq.9).10) we now obtain the remarkable result that < 0U(T )0 >=< 0e−iH(J=0)T 0 > exp − i 2 d4 x d4 y J(x)G(x − y)J(y) (7. Finally we note that the oscillatory integrals occurring in the path integral can be dampened by replacing m2 by m2 − iε as this leads to replacement exp(i L(ϕ)d4x) → exp(i L(ϕ)d4x − ε ϕ2 d4 x).(6.7)) to get < {ϕ′ (p)}U(T ){ϕ(p)} >= ˜ ˜ = exp − i 2 d4 x d4 y J(x)G(x − y)J(y) i 2 = exp − d4 x d4 y J(x)G(x − y)J(y) < {ϕ′ (p)}e−iH(J=0)T {ϕ(p)} > ˜ ˜ D ϕ(k) exp i ˜ 1 2 k. Obviously our notations are such 2 that k ≡ (k0 . But before we will derive the Feynman rules from the perturbative expansion. and coincides with the prescription derived for the propagator in the Hamiltonian formulation. 30 .13) In the last step it is crucial to note that the source is taken to vanish for t ≤ 0 and for t ≥ T . as this will be our tool to write down the matrix elements of the evolution operator (the socalled scattering matrix. see eq. since the quadratic part of the action is the starting point of the perturbative expansion.14) to all orders in the source J.23). (4. hence ˜ we also ﬁnd a unit Jacobian for this change of variables. or for short Smatrix) with respect to the basis speciﬁed by particle number and momentum (the socalled Fock space). it will be useful to emphasise that the time ordering. this demonstrates the dramatic simpliﬁcations that can arise from using the path integral method for calculating quantum amplitudes. Even at the level of a noninteracting scalar ﬁeld theory.k0 (k 2 − m2 + iε)ϕ(k)2 ˜ . as otherwise the shift we performed in the ﬁeld would have changed the boundary values. This is no accident and is in general the way the (lowest order) propagator is directly read oﬀ from the Lagangian. One particularly feature that is noteworthy in the path integral calculation. Furthermore. is that the part of the Lagrangian that is quadratic in the ﬁelds represents the inverse propagator. exactly as was done in one dimension. If we take the limit of space and time to inﬁnity (L → ∞ and T → ∞). the sums over k can be converted into integrals. (2.We have here also performed the Fourier transformation with respect to time. (7. Using eq. (7. we simply split oﬀ a square < {ϕ′ (p)}U(T ){ϕ(p)} >= ˜ ˜ D ϕ(k) exp i ˜ 2 ˜ J(k) 2 2 1 ˜ (k − m + iε) ϕ(k) − 2 2 2 + iε k −m k. t) to the multiple integral over the (discrete tempo˜ ral) Fourier components D ϕ(k). is automatically implemented by the path integral.
˜ Each of the matrices for the three evolution operators can be written as a path integral. t1 ) in the path integral ˆ formulation simply means that one glues the paths in U(t3 .21) 31 . Without the insertion of the ﬁeld operators ϕ(x). t1 )0 >= ˆ ˆ < 0e−iH(T −t2 ) ϕ(x ′ )e−iH(t2 −t1 ) ϕ(x)e−iHt1 0 > ˆ ˆ −iHT 0 > < 0e .20) L d4 x) } < {ϕ(p)}0 > . ˜ where we implicitly assumed that the boundary conditions for the ﬁeld ϕ in the path integral is in momentum space given by ϕ(p. With the ﬁeld operator sandwiched between the two U matrices ˜ one simply includes its eigenvalue in the integrand over the paths. The way the ˜ ˜ ˜ ˜ time ordering in the path integral is manifest is now obvious. The matrix product involves an integral over the ﬁnal ﬁeld component of the matrix to the right. ˆ ˆ (7. One ﬁrst writes the product of the operators as a product of matrices in a suitable representation (e. t = 0) = ϕ(p) and ϕ(p. In the presence of a time dependent source one can write < 0U(T )ϕ(x ′ . this would describe the fact that U(t3 .16) t2 > t1 . t1 )0 > .8)) is given by ϕ(x. It is now trivial. t2 )U(t2 .17) . the ﬁeld representation {ϕ(p)} >). t1 )ϕ(x)U(t1 )0 > ˆ ˆ ˆ ˆ In this case the ﬁeld operators are of course given by ϕ(x. t) = U(t)† ϕ(x)U(t) .18) where in general U(t) depends on the source J. t2 )ϕ(x ′ )U(t2 . but a bit tedious. (7. The ﬁnal result can be ˜ written as dϕ(p)dϕ′(p) < 0{ϕ′ (p)} > { Dϕ(x) exp(i ˜ ˜ ˜ ϕ(x. t2 )ϕ(x. t1 )0 >= − ˆ ˆ δJ(x ′ . excluding the integral over the initial and ﬁnal ﬁeld components. (7. eq. t1 ) together by integrating over ϕ(k. t1 ) = U(t3 . The normalisation by < 0e−iHT 0 > is hence a rather trivial factor. which is also the initial ﬁeld component for the matrix to the left. t1 ) exp(i 0 T t2 t1 p L d4 x)ϕ(x ′ . t) = eiHt ϕ(x)e−iHt ˆ ˆ such that < 0ϕ(x ′ . t2 ). ˆ ˆ where the ﬁeld operator (cmp. (7. t1 ) ′ . We can even write a similar expression in the presence of the source. t2 )ϕ(x. t2 )δJ(x. t2 ) and U(t2 . to convert this matrix element to a path integral. (7. t2 )ϕ(x.In the Hamiltonian formulation we consider < 0ϕ(x ′ . (7. (2. t2 ) exp(i t2 t1 L d4 x) × (7. t1 )0 >=< 0U(T. since the ﬁeld operator (or its Fourier components) is diagonal on the states {ϕ(p)} >.19) . In perturbation theory this would not be needed.15) where we have made use of the fact that the vacuum 0 > is assumed to be an eigenstate of the Hamiltonian H (without a source term).g. t2 )ϕ(x. but it is useful from a general point of view to consider this situation too. which is particularly simple to derive in the path integral formulation δ2 < 0U(T )0 > < 0U(T )ϕ(x . t = T ) = ϕ′ (p). Note that the ﬁeld expectation values can also be written in terms of derivatives with respect to the sources.
at J = 0).23) with the boundary conditions as listed below eq. < 0ϕ(x ′ .8). ˜ (7. From eq. we ﬁnd for this matrix element in lowest nontrivial order < pU(T )0 > = −i = −i = −i T 0 dt < 0a(p)ei(t−T )H0 H1 (t)e−itH0 0 > T 0 ˜ dt < 0a(p)ei(p0 (p)+E0 )(t−T ) J(k. such that < 0ϕ(x)U(T )0 > = < 0U(T )ϕ(x. t)eip0 (p)t = −i J(p) . we will calculate both in the Hamiltonian and in the path integral formulations the matrix element < pU(T )0 > .27) 2p0 (p) V d3 x ϕ(x)e−ip·x ˆ . (7. (7.20). (7. Using the result of eqs.7) and (5. t1 )0 >J =0 = − < 0U(T )0 >−1 ˆ ˆ where in the path integral formulation one has < 0U(T )0 >= p δ 2 < 0U(T )0 > δJ(x ′ .25) p0 (p) √ To write down the path integral result.3) we ﬁnd a(p) + a† (−p) = such that (using < 0a† (−p) = 0) < pU(T )0 >= 2p0 (p) V d3 x e−ip·x < 0ϕ(x)U(T )0 > ˆ .24) Hence p > is the oneparticle state with momentum p. To study the role the source plays in creating and annihilating particles.(7. using that in a ﬁnite volume the Green’s function is given by dk0 e−ik(x−y) 1 . (7. t) T 0 (a(−k) + a† (k)) 2k0 (k) k e−iE0 t 0 > e−i(E0 +p0 (p))T 2p0 (p) πe−i(E0 +p0 (p))T ˜ ˜ dt J(p. t1 ) .22) dϕ(p)dϕ′ (p) < 0{ϕ′(p)} > ˜ ˜ ˜ Dϕ(x) exp(i T 0 L d4 x) < {ϕ(p)}0 >. J=0 (7. < 0UJ=0 (T ) =< 0e−iE0 T (with E0 the vacuum energy) and we can bring the trivial phase factor e−iE0 T to the other side by normalising with < 0UJ=0 (T )0 >. t2 )ϕ(x. T ) 2 We evaluate this to linear order in the source J.Since we assumed that 0 > is an eigenstate of H (i.19) and an obvious generalisation of eq. (5. T )0 >= i ˆ ˆ δ < 0U(T )0 > δJ(x.21). T ) i δ d4 xd4 y J(x)G(x − y)J(y) . (7. we ﬁrst express the annihilation operator in terms of the ﬁeld.14). (7. (5. as in eq. (7.26) We now use eqs.28) exp − = ie−iE0 T δJ(x.17). (7. p >≡ a† (p)0 > .e. (7.29) G(x − y) = V 2π k 2 − m2 + iε k 32 . (7. which is equivalent to the result Ψ(t) >= U(t)0 >. t2 )δJ(x.
30) becomes < 0U(T )p > = e−iE0 T = e−iE0 T 2p0 (p) V p0 (p) π d3 x eip·x dp0 d4 y G(x − y)J(y) + O(J 3 ) (7.32) ˜ J (−p.this renaming is just for ease of notation. 0) . p2 − m2 + iε in which case x0 = 0 is assumed. Again.33) 2p0 (p) V 2p0 (p) V d3 x eip·x < 0U(T )ϕ(x.30) ˜ J(p)e−ip0 T + O(J 3 ) . whereas e−ip0 T J(p) will vanish for Imp0 → −∞. Since J(p) 0 ˜ will vanish for Imp0 → ∞.(7. (7. the path integral trivially allows an extension to arbitrary order in the source. 33 . (7. √ −iE0 T i πe ˜ J(−p) exp − < 0U(T )p >= −i d4 xd4 y J(x)G(x − y)J(y) . we see that the amplitude for ˜ the annihilation of a oneparticle state is proportional to J(p) whereas it is proportional ˜ (exactly with the same factor) to J(−p) for the creation of a oneparticle state (in both cases p0 = p0 (p)). p0 ) + O(J 3 ) . analogously to eq. Now we must deform the contour for the p0 integration to the upper halfplane in a counterclockwise fashion such that the residue at the pole p0 = −p0 (p) contibutes. (7.such that < pU(T )0 > = e−iE0 T = e−iE0 T 2p0 (p) V p0 (p) π d3 x e−ip·x dp0 d4 y G(x − y)J(y) + O(J 3 ) (7. (7. 2 − m2 + iε p Note that in the last step we integrate over p0 as a dummy variable. ˜ x0 = T is assumed. For the p0 integration we need the analytic behaviour of J(p) for imaginary p0 in order to see if we are allowed to deform the integration contour√ such that ˜ = T dt eip0 t J(p. as indicated.30) the p0 integration can therefore be deformed to the lower halfplane in a clockwise fashion √ giving a minus sign and a residue from the pole at p0 = p0 (p) ≡ p 2 + m2 . J(p) ˜ ˜ only one of the poles in the integrand contribute. which yields the result √ −i(E0 +p0 (p))T i πe ˜ J(p) exp − < pU(T )0 >= −i d4 xd4 y J(x)G(x − y)J(y) . Also. (7. For later use.34) 2 p0 (p) Apart from the trivial diﬀerence of the factor exp(−ip0 T ). (7. In eq.25).31). we will also consider the matrix element < 0U(T )p > = = i The analogue of eq. which in the expression for the Green’s function above is called k0 .31) 2 p0 (p) To linear order in the source J this coincides with the result of eq. This gives. 0)0 > ˆ d3 x eip·x δ< 0U(T )0 > δJ(x. t)/ 2π.
2) ∂µ ϕ(x)∂ µ ϕ(x) − m2 ϕ2 (x) . (8.7) where the dependence on the coupling constants in GJ is implicit. which will now be studied in more detail. Z(J. (8.5) = j iδ δJ(x(j) ) = j iδ i exp − δJ(x(j) ) 2 d4 xd4 y J(x)G(x − y)J(y) to ﬁnd a somewhat formal. gn ) is synonymous with Z(J.4) = < 0U2 (T )Texp −i We can now use the fact that Dϕ(x) ϕ(x(j) ) exp i j d4 x (L2 (ϕ) − ϕ(x)J(x)) Dϕ(x) exp i d4 x (L2 (ϕ) − ϕ(x)J(x)) . (8. We now deﬁne GJ as GJ ≡ log Z(J. A diagram is connected if it can not be decomposed in the product of two diagrams that are not connected. Equivalently. gn ) ≡< 0U(T )0 >= Dϕ(x) exp(i d4 x L(ϕ)) (8. when evaluating the path integral for the quadratic approximation L(ϕ) = L2 (ϕ) − ϕ(x)J(x). gn ) = Dϕ(x) exp i d4 x {L2 (ϕ) − ϕ(x)J(x)} exp −i d4 x V (ϕ) 0 > ˆ . We will show that GJ can be seen as the sum of all connected diagrams. We can add the interaction as an operator. d4 x V (ϕ) (8. In general the Lagrange density for a scalar ﬁeld theory is given by L(ϕ) = L2 (ϕ) − V (ϕ) − J(x)ϕ(x) .8 Perturbative expansion in ﬁeld theory As we have seen in the previous section ( Dϕ(x) where relevant includes groundstate factors) Z(J.1) will play the role of a generating functional for calculating expectation values of products of ﬁeld operators. V (ϕ) = g4 g3 3 ϕ (x) + ϕ4 (x) + · · · 3! 4! . where L2 (ϕ) is quadratic in the ﬁelds. Note that at 34 . in absence of interactions.6) We have assumed the vacuum energy to be normalised to zero. such that Z(J = gn = 0) = 1. hence for a scalar ﬁeld L2 (ϕ) = 1 2 (8. gn )/Z(J = gn = 0). but in an expansion with respect to the coupling constants gn well deﬁned expression for the fully interacting path integral Z(J. gn ) .3) As mentioned before. it is customary to not include the mass term in the potential V . δJ(x) d4 x V ( d4 xd4 y J(x)G(x − y)J(y) (8. gn ) = exp −i ≡ exp −i i iδ ) exp − δJ(x) 2 iδ d4 x V ( ) Z2 (J) . Z(J. such that V describes the interactions.
the propagator will come with an extra factor −i. gn ). To compute the vacuum energy there is an overall factor i. one of the sources connected to a propagator 35 . Using ϕ(x) = 1 (2π)2 d4 k e−ikx ϕ(k) . coordinate space ≡ i y × x ℓ−1 momentum space T ¨r ¨r % j ¨r table 2 ki ) vertex propagator source k1 ¨r ¨x r gℓ d4 x k3 k2 ≡ iℓ−1 (2π)4−2ℓ gℓ δ4 ( ≡ −i d4 k k2 −m2 +iε i x ≡ −iG(x − y) ≡ d4 xJ(x) ' k × k ˜ ≡ J(k) We note that the propagator connected to a source comes down whenever a derivative in the source acts on Z2 (J). see eq.7). The factors of i exactly cancel each other. ˜ In the quantum theory we have to keep track of the factors i. The same factor of i applies for using the treelevel diagrams to solve the classical equations of motion. ˜ J(x) = 1 (2π)2 ˜ d4 k e−ikx J (k) . Compared to the Feynman rules of section 4. see the table below. which is to be integrated over. since E0 T = iGJ=0 = i log Z(J = 0. It is easy to see that these Feynman rules give identical results for these treelevel diagrams. (8. iGJ /T equals the energy of the ground state as a function of the coupling constants. The diﬀerent diagrams arise from the expansion of exp −i iδ ) = exp −i d4 x V ( δJ(x) gℓ d4 x ℓ=3 ℓ! iδ δJ(x) ℓ (8.J = 0. (8. (8.8) in powers of gℓ . A vertex will now √ carry a factor igℓ ((2π)2 /i)2−ℓ (in a ﬁnite volume this becomes igℓ ( 2πV /i)2−ℓ ). (8. As we saw in the derivation of the classical equations of motion. normalised so as to vanish at zero couplings. Each factor gℓ (iδ/δJ(x))ℓ will represent an ℓpoint vertex.9) we have for each vertex in the Fourier representation −igℓ iδ d4 x δJ(x) ℓ = −igℓ (2π)4 δ4 ( ℓ ℓ k(j) ) j=1 j=1 d4 k(j) iδ ˜ (2π)2 δ J(k(j) ) . with coordinate ℓ! x. the integral over x in the Fourier representation gives rise to conservation of momentum at the vertex.10) Note that factors of 2π are dropping out in the identities exp − i 2 d4 xd4 y J(x)G(x − y)J(y) = exp − i 2 d4 k ˜ ˜ J(k)J (−k) k 2 − m2 + iε . as compared to the Feynman rules introduced in section 4.11) d4 x ϕ(x)J(x) = ˜ d4 k ϕ(k)J(−k) . When this derivative acts on terms that have already come down from previous derivatives.
we replace the path integral by an ordinary integral. with the inverse of the order of the permutation group that leaves the topology of the diagram unchanged.13) (8. We can check them by reducing the path integral to zero dimensions. with X(iδ/δJ(x)) ≡ −i d4 x V (δ/δJ(x)) . [X. between two vertices r q q¨ .14) Due to the multiple commutators. the numerical factors in front of the diagrams indicate the combinatorial factors (for the ﬁrst diagram a factor 2 from interchanging the two vertices and a factor 3! 36 . and is speciﬁc to the quantum theory. and we can express GJ in a sum of multiple commutators using the CampbellBakerHausdorﬀ formula (see eq. The only thing that remains to be discussed is with which combinatorial factor each diagram should contribute. To prove that GJ only contains connected diagrams we write GJ = log {exp(X(iδ/δJ)) exp(Y (J))} 1 .is removed and this connects that propagator to the vertex associated to δ/δJ(x). g) = C 1 dϕ exp i{ 2 ϕMϕ − The constant C is simply to normalise Z(J = gn = 0) = 1. However. ¨ ¨ r possibility of closed loops did not occur in solving the classical equations of motion. Y ] + 1 1 [X.15) i∂ ∂J 6 i − JM −1 J 2 6 3 ··· 3 g2 = exp i 4 2 (3!)3 = exp 1 12 m + ∂ ∂J 1 8 JM −1 J m m O(g 3 ) + + O(g 3 ) = exp i . Y (J) ≡ − i 2 d4 xd4 yJ(x)G(x − y)J(y) . As an example consider Z(J. or ϕ(x) → ϕ and Dϕ(x) → dϕ. (8. 12 12 (8. 1 GJ = X + Y + 2 [X. Y ]] + [Y. This is.12) As J(x) and δ/δJ(x) form an algebra (similar to the algebra of x and p in quantum mechanics. ˆ ˆ however. or it connects two legs of the same vertex q j The . These combinatorial factors are clearly independent of the spacetime integrations and possible contractions of vector or other indices. also X and Y are elements from the algebra. X]] + · · · 1 . generalised to inﬁnite dimensions). if the components would commute they would not contribute to the commutators. [Y. 5g 2 + O(g 3 ) 24M 3 (8. the propagator either runs between a vertex and a source r q ×. As any derivative is connected to a vertex. In other words. once they have been moved to the right (this is why we consider the action on the identity). all components that do not commute are connected. Expanding the exponents we get in lowest nontrivial order g2 Z(J = 0) = 1 − 2(3!)3 = exp −i g 3! g 3 ϕ − ϕJ} 3! 3 i∂ i exp − JM −1 J ∂J 2 .44)). This is even true if we do not put the derivatives with respect to the source to zero. (8. as in section 4. (6.16) In the last term.
we will show that Z(J.from interchanging the three propagators. whereas we have shown in section 4 that these classical equations are solved by tree diagrams. (8. 9 The scattering matrix We would like to compute the amplitude for the transition of n incoming particles at t = Tin to ℓ outgoing particles at t = Tout in the limit where Tout → ∞ and Tin → −∞.19) Vn . gn ) = i Dϕ exp h ¯ ∞ (L(ϕ) − ϕJ) ≡ exp(GJ /¯ ) . · · · . We can expect such a relation. the number of independent momenta) in a diagram equals the number of loops plus the number of external lines. i. If we call L the number of loops of a diagram. ¯ (8.e. E external h lines (connected to a source) and P propagators has therefore an extra overall factor of h−E hP ¯ ¯ n=3 h−Vn ¯ . pℓ U(Tout .18) We can relate this to the number of loops by noting that the number of momentum integrations (i. · · · . as there are exactly E such external propagators. On the other hand. the propagator a factor h and the ¯ h ¯ coupling constants gn are replaced by gn /¯ . In problem 13 the exponentiation is checked for Z(J) to O(g 2 ) and O(J 5 ) (giving the simplest nontrivial check). Tin )k1. The diﬀerence with quantum mechanics is that the particle number is no longer conserved. as we have shown at the end of section 6 that the h → 0 ¯ limit is related to the classical equations of motion. eq. pℓ k1. 37 (9. out< p1 . k2 . The Feynman rules for this simple case are that each vertex gets a factor −g (in zero dimensions there are no factors 2π) and each propagator gets a factor −i/M. i. p2 .e. · · · . Hence (8.17) where GL. k2 . In the next ¯ sections we will often consider socalled amputated diagrams. L+ E −1. p2 . for the second diagram the latter factor is replaced by 4 as we can only interchange for each vertex the two legs that do not interconnect the two vertices). If we do not count these external propagators. kn > . (8.1) In terms of creation and annihilation operators this can be written as out< < 0a(p1 )a(p2 ) · · · a(pℓ )U(Tout . · · · . · · · . p1 . k2 .2) .e. h GJ = L=0 hL GL. L=1+P −E − n=3 which implies that the total number of h factors in a diagram is given by L − 1.J . We will now show how the number of loops in a diagram is related to the expansion in ¯ h. A diagram with Vn npoint vertices. To prove this we ﬁrst note that due to ¯ reinstating h the source term will get an extra factor 1/¯ . pℓ k1 . P − n=3 Vn . kn >in≡< p1 .19) has to be replaced by L = 1 + P − n=3 Vn . kn >in = (9. the number of momentum integrations is also the number of propagators minus the number of delta functions coming from the vertices. · · · . Tin )a† (k1 )a† (k2 ) · · · a† (kn )0 > . where the external propagators connected to a source are taken oﬀ from the expressions for the diagram. This means that a loop expansion is equivalent with an expansion in h. minus one for the overall conservation of energy and momentum. p2 .J is the sum of all connected diagrams with exactly L loops.
by introducing a cutoﬀ. 2π p2 − m2 + iε (9.From eqs. Because of the necessity to regulate the quantum theory. (9. t = Tin ) . If this is not possible. this result is no longer true.3) δ δ = i 2p0 (p) ˜ δJ(x.4) In principle. unitarity of a Smatrix is guaranteed as soon as the Hamiltonian is hermitian. as in experiments we are unable to switch oﬀ these selfinteractions. In this case we can write (from now on (2) ˆ a† out< pk >in = a(p)ˆ (k) exp(Gc (J))J =0 = 2ip0 (p)δk. this allows us to calculate the scattering. k2 . For simplicity we assume that the oneparticle states are stable. · · · . since the connected twopoint function will deviate from the one in the free theory. the only diagram contributing to GJ is the one with a single propagator connecting two sources. kn >in = a(pi ) ˆ i=1 j=1 a† (kj ) exp(GJ )J =0 ˆ . For the free theory. t = Tout ) This implies the following identity for the scattering matrix ℓ out< n p1 . t)J(−p. Formally. t = Tin ) δ J(p. It is trivial to see that this is the same as what can be obtained within the Hamiltonian formulation. t = Tout ) δ J(−p. p2 . this is generally no longer true and one has to show that unitarity is restored when the cutoﬀ is removed. (9. gn ) a† (p) = i ˆ a(p) = i ˆ 2p0 (p) V 2p0 (p) V d3 x eip·x d3 x e−ip·x δ δ = i 2p0 (p) ˜ δJ(x. s) e 2π p2 − m2 + iε .p .p 38 . In the presence of interactions.26). conservation of probability implies that the Smatrix is unitary. There is. e. however. In general. (7.5) independently of p. taking Tin → −∞ and Tout → ∞.28) and (7. giving a contribution from the pole at p0 = −p0 (p) only. In this case. unitarity is of course satisﬁed. or at best does not make sense above the energies where unitarity is violated.g.6) This implies that (T ≡ Tout − Tin ) dp0 eip0 T = e−ip0 (p)T δk.7) where the p0 integration is performed by deforming the contour to the upper halfplane. (7. the theory is illdeﬁned. · · · . which is also called the connected twopoint function G(2) (J) c GJ = G(2) (J) = − c i 2 dtds p ˜ ˜ dp0 −ip0 (t−s) J(p. pℓ k1 . a problem to associate the particle states in the presence of interactions with the ones we have derived from the noninteracting theory.32) we know how to implement these creation and annihilation operators on the generating functional Z(J. (9. The problem is that particles can have selfinteractions long before and after the diﬀerent particles have scattered oﬀ each other. . We have to reconsider our notion of particle states. (9. as in the simple scalar theory we have been considering. This implies from conservation of probability that d3 k  out< pk >in 2 = 1 .
(9. will not allow. m2 . usually required to be absent. &% (9.12) As long as the oneparticle states are stable. · · · .e. As we cannot switch oﬀ the interactions in nature.9) In general. (9.10) . kn >in = n a− (pi ) ˆ i=1 j=1 a+ (kj ) exp(GJ )J =0 ˆ . The diagrams of eq. i. · · · . c p − m2 − Σ(p) + iε −i iΣ(p) + 2 p − m2 + iε n 3 (iΣ(p))2 + · · · (9.11) Consequently. the latter is ˜ also called the bare mass. d d d d d &% d × '$ '$ ×+··· &% &% (9. such that the selfinteractions shift the pole in the twopoint function to another value. (9. J(p) vanishes rapidly as a function of p2 − m2 ) one therefore has ˜ G(2) (J) = − c i 2 d4 p ˜ ZJ(p)2 p2 − m2 + iε ˜ .14) 39 . ˜ 2 p2 − m2 − Σ(p) = 0 for p0 = p 2 + m2 ˜ ∞ −i Σ(p) p2 − m2 + iε n=0 p2 − m2 + iε ˜ J(p)2 d4 p 2 ≡ G(2) (J) . p2 = m2 . eq.the symmetry factors will be absorbed in the expression associated to a diagram) G(2) (J) c = × ≡ × '$ × + × ×+ &% '$ d d d d × .8) where we have written the connected twopoint function in terms of the oneparticle irreducible (1P I) twopoint function iΣ(p) (Σ is the socalled selfenergy) iΣ(p) ≡ '$ . k2 . these tadpoles describe single particles popping in or out of the vacuum.) The external lines of these diagrams will carry no propagator.e. the mass of the oneparticle states is shifted (or renormalised).5) needs to remain valid.13) where √ 2 p 2 + m2 ˜ a+ (p) ≡ i ˆ Z 1 2 δ . called the massshell) will ˜ ˜ in general also change from ±πi/p0 (p) to ±πiZ/p0 (p). a 1P Igraph is a connected graph that remains connected when one arbitrary q j which we propagator is being cut (except when cutting away a tadpole of the form .8) can be converted to the result 1 2 −i −i ˜ d4 p J(p)2 + 2 2 − m2 + iε p p − m2 + iε = 1 2 2 ˜ d4 p J(p)2 = − i 2 Normally the selfenergy will not vanish at p2 = m2 . (9. p2 . which can only √ be achieved (see eq. (9. ˜ δ J(−p. pℓ k1 .e. √ 2 p 2 + m2 ˜ a− (p) ≡ i ˆ Z 1 2 δ .4) needs to be modiﬁed to ℓ out< p1 . the true or observable mass is m and not m. They can be removed by shifts in the ﬁelds. (9. The residue at the poles (i. However. On the massshell (i. t = Tout ) (9.7)) by rescaling the wave functionals with a factor Z (this is called wavefunction renormalisation). It implies that eq. (9. ˜ t = Tin ) δ J(p.
using eq. p2 . i. all on the massshell. # . p2 . we can extend the p0 contour integration to the upper halfplane. · · · . Let us ﬁrst consider the most important term. "! .16) × × To get the Smatrix we have to compute (see eq.As for the free theory. 2 2 − Σ(p ) + iε pi − m i × # r r #¨ × ¨ d 96 d d d d d d d d d d dd r dd r d d d . p2 .e. · · · . (9. pn ) as Imp0 → −∞ (that can easily be shown mild regularity conditions for Gc 40 . Concentrating on one such a factor we have (amp) ˜ (p. We will call the connected npoint function with amputated external lines the (amp) (p1 . Note that we have extracted the trivial energy factors (i) (j) (i) (i) (remember that p0 Tout − k0 Tin = p0 (Tout − Tin ) ≡ p0 T ). r. · · · . (9. (9. . pn ) p0 (p) (p. under (amp) (p. Each a(p)± will act on one of the factors between ˆ curly brackets in eq. {kj }) −iΣp(i) T j=1 0 e (i) (j) ℓ 2p0 V n 2k0 V i=1 j=1 . pn ) −ip0 (p)Tout Gc e .14) a− (p) ˆ (amp) (amp) ˜ eip0 Tout Gc ((p0 . d .20) = −i 2πV Z 2p0 (p)V Since Tout → ∞. · · · . pn ) (in general not oneparticle amputated connected npoint function Gc irreducible). p2 . pn ) −iJ (p)Gc = d4 p 2 − m2 − Σ(p) + iε p dp √4 2π (amp) ˜ (p. pn ) −iJ (p. such that the limits Tin → −∞ and Tout → ∞ can be taken. p2 .15) Diagrammatically this looks as follows (9. "! ¨. . corresponding to connected graphs.17) The sum is over all possible partitions of ℓ + n. each of these operators a± (p) will replace one external line (propagator ˆ plus source) by an appropriate wavefunction factor and puts these external lines on the massshell. pn ) −iJ (p)Gc dp0 = d4 p 2 − m2 − Σ(p) + iε 2 − m2 − Σ(p) + iε p πZ p (amp) √ (−p. # . (9. . p2 . ¨d .15).19) such that. {kj })e−iΣp0 (i) (i) T ≡ −2πiV δ4 ( − n kj ) Mℓ+n ({−pi }. where q1 = ℓ + n ℓ n n a+ (kj ) eGJ ˆ J=0 ℓ n = eGJ =0 i=1 a− (pi ) ˆ j=1 a+ (kj ) ˆ Σrqr =ℓ+n r 1 G(r) (J) c qr ! qr . · · · . p2 . t)eip0 t Gc dt . −p).13)) ℓ n a− (pi ) ˆ i=1 j=1 (9. pn ) . (9.18) Here Mℓ (a Lorentz scalar as we will see later) is the socalled reduced matrix element with ℓ external lines. ¨ ¨ 87 r d d d d d dd dd d d dd dd r d d d d r ¨d "! ¨ "! (9. · · · . · · · . · · · . a− (pi ) ˆ i=1 a+ (kj ) ˆ j=1 G(ℓ+n) (J) = c ℓ i=1 pi j=1 −i (j) πZ/k0 ℓ i=1 ( −i πZ/p0 Gcamp) ({−pi }. . n G(n) (J) c ≡ d 4 pi i=1 ˜ −iJ (pi ) ( Gcamp) (p1 . 2 − m2 − Σ(p) + iε p (9.
(9. · · · . The reason this case is special is because eq. Instead. (9.18). with residue −Z/2p0 (p) (cmp. in an inﬁnite volume one replaces V by (2π)3 ). 1 (9. (9. Thus. E + 2P = n nVn . In using eq.24) The proof for the ﬁrst identity was discussed below eq.19). which is often also normalised to 1. (9. (9. we can now extend the p0 contour integration to the lower halfplane (under the same regularity conditions for the amputated npoint functions as for eq. (9. see the table below). such that we pick up the contribution of the pole at p0 = p0 (p) with residue Z/2p0 (p). (9.15) requires us to deﬁne for the amputated twopoint function ( Gcamp) (p1 .12) and the fact that exp(GJ=0 ) = exp(−iE0 T ).21) it was implicitly assumed that the amputated Green’s function has no zero that will cancel the pole. · · · . (9. giving a total of 2P + E dots. as Tin → −∞. where the generalisation of eq.to be satisﬁed at any ﬁnite order in perturbation theory). 2π and V can be determined with the help of the two identities L=P +1− Vn n . (9. To keep the derivation general.18) is the appropriate one. (9. (9.25) .23) which vanishes on the massshell. p2 .22) Here we used eq.21) = −i 2πV Z 2p0 (p)V Likewise. Combining these results proves the ﬁrst identity in eq. giving nVn dots. If we associate a momentum delta function to a vertex and a momentum integration to a propagator (as was done up to now). p2 . if we choose to integrate over the independent loop momenta. the delta function for overall energy and momentum conservation should not be written explicitly in eq. the deﬁnition of eq. The same dots can also be associated to each line of a vertex ( ¨r ). (8. · · · . The Feynman rules in momentum space for computing the reduced matrix elements will √ obviously have to be modiﬁed for the external lines to a factor −i 2πV Z and an overall factor i/(2πV ) (as always. For the second identity we put a dot on each end of a propagator ( ) and one dot on each external line ( ).12)). For the creation operators one similarly ﬁnds a+ (p) ˆ (amp) (amp) ˜ (p. we evaluate the overall factor in a ﬁnite volume i−P i−E+1 iΣ(n−1)Vn (2πV )− 2 Σ(n−2)Vn (2πV ) 2 E−1 = 41 1 1 i 2πV L . pn ) ((p0 .20) and eq. (9. (9. the second is merely a deﬁnition. Note that the derivation is not valid for the case ℓ + n = 2. pn ) ip0 (p)Tin Gc e . the integral over p0 only √ ˜ gets a contribution from the pole at p0 = −p0 (p) = − p 2 + m2 . thus proving the second identity (see also problem 15).18). implementing energy and momentum conservation at each vertex (as will be done from now on. since it is contained in the reduced matrix element. p). (9. pn ) −iJ (p)Gc eip0 Tin Gc p0 (p) d4 p dp0 = 2 − m2 − Σ(p) + iε 2 − m2 − Σ(p) + iε p πZ p (amp) √ (p.7) to the interacting case implies out< pk >in = a− (p)ˆ+ (k) exp(GJ )J =0 = δ3 (k − p)e−i[E0 +p0 (p)]T ˆ a .20) to be valid). p2 ) ≡ iδ4 (p1 + p2 )(p2 − m2 − Σ(p1 ) + iε) . eq. but till now we had only required this to be the case at zero couplings. The overall factors of i.
But from eq. Concerning the symmetry factor associated to a particular diagram we note the following. in computing c the reduced matrix elements the symmetry factors are determined without allowing for permutations on the external lines. will be compensated by the n derivatives on n sources. the experimental situation is such that the energymomentum conservation will only be compatible with the fully connected part. as the latter does already contain a factor −i(2π)4 (in a ﬁnite volume −2πiV ). As an example. In particular each connected component will carry its own factor −i(2π)4 δ4 ( pi ) for the conservation of energy and momentum (pi is now assumed to run over a subset of both the incoming and the negative of the outgoing fourmomenta). Quite often. For the corresponding connected components the rules are identical to the ones speciﬁed above. the table compares our Feynman rules with those of Itzykson and Zuber. However. In a collider. Hence. and discuss the contributions that will be associated to the diagrams that are not connected. As we have generally ﬁxed the external momenta. Combining the extra factors of i and 2π in the Feynman rules of Itzykson and Zuber gives i(2π)−4 iP −ΣVn (2π)4(ΣVn −P ) = [i/(2π)4 ]L . 42 .This implies that we can shift all numerical factors from the propagators. giving the Feynman rules listed in the table for an inﬁnite volume.15) we see that the symmetry factor n!. to be taken into account for G(n) (J). momentum space T ¨r ¨r % j ¨r Itzykson and Zuber ki = 0 T ¨r ¨r % j ¨r table 3 ki ) vertex propagator external line loop factor k1 k1 k3 k2 ≡ gℓ and ≡ ≡ i 1 k 2 −m2 +iε k3 k2 ≡ −i(2π)4 gℓ δ4 ( ≡ ≡ 1 i d4 k (2π)4 k 2 −m2 +iε i k k √ Z √ Z d4 k (2π)4 In the literature many diﬀerent conventions are being used. this means one excludes particles that escape in the direction of the beams. We just have to avoid the incoming momenta to coincide with any of the outcoming momenta. interchanging external lines is no longer allowed. that factor should be absent in relating the reduced matrix element to the amputated npoint function. To conclude this section we return to eq. Note that the extraction of the factor −2πiV in the deﬁnition of M is merely a convention (such that in lowest order Mn = gn ).17). As an illustration we will give the situation for q j to second order in ) n = ℓ = 2 and all momenta nonzero (to avoid tadpole diagrams. Their convention for Mn is likewise to make it coincide to lowest order with the npoint vertex. where indeed it is not possible to put a detector. vertices and external lines to a factor i/(2πV ) (or i/(2π)4 in an inﬁnite volume) for each loop. Each factor of qr ! is compensated qr (r) for by the diﬀerentiations on Gc (J) . In a physical picture the disconnected parts correspond to situations where only a subset of the incoming particles will interact with each other (the ones connected by a particular diagram). (9. (9. guaranteeing equivalence of the two sets of Feynman rules.
The momenta ki in the reduced matrix element are the central i=1 values of the wave packet in momentum space for the two incoming particle beams. (10.3) We assume that over the range of momenta in the wave packets.the threepoint coupling g3 .4) d4 xd4 y Ψ1 (x)Ψ2 (x)Ψ∗ (y)Ψ∗(y)eip(x−y) 1 2 ¯ where p = n pi = k1 + k2 . We denote ˜ ˜ by d3 k1 Ψ1 (k1 )k1 > and d3 k2 Ψ2 (k2 )k2 > the wave functionals of the incoming particles. {kj })2 f (p) f (p) ≡ d 3 pi 3 i=1 2p0 (pi )(2π) n . {kj }) ˜ ˜ Ψ1 (k1 )Ψ2 (k2 )e−i[E0 +Σp0 ]T . The ﬁrst two diagrams. This is to describe the more realistic case of a wave packet. The amplitude for scattering to take place is hence given by (i) pi − k1 − k2 )M2+n ({−pi }. 10 Cross sections In many experimental situations we are interested in the scattering of two particles with momenta k1 and k2 to a state with n particles with momenta p1 . By deﬁnition they have no higher order corrections. with momenta in between pi and pi + dpi . This allows us to write for the scattering probability of two particles into n particles. the reduced matrix elements are constant (which can be achieved with arbitrary precision for arbitrarily narrow wave packets in momentum space). (10.5) . p2 .2) we can compute the overlap of the two wave functions d4 x Ψ1 (x)Ψ2 (x)eipx = (2π)4 d3 k1 2k0 (k1 )(2π)3 d3 k2 2k0 (k2 )(2π)3 ˜ ˜ δ4 (k1 + k2 − p)Ψ1 (k1 )Ψ2 (k2 ) . Under the same assumption that the momentum spread in the beams is very small. which have to be treated with special care (see eqs. · · · . ¯ dW = M({−pi }.1) If we deﬁne the wave function in coordinate space as usual A = −i(2π)4 d3 k1 d3 k2 δ4 ( n i=1 Ψj (x) = d3 k 2k0 (k)(2π)3 ˜ e−ikx Ψj (k) . (i) (j) n 3 2 3 i=1 2p0 (pi )(2π) j=1 2k0 (kj )(2π) (10. represent the situation without scattering. k2 >in = exp −i[E0 + p0 + p0 ]T × δ3 (p1 − k1 )δ3 (p2 − k2 ) −i + δ3 (p1 − k2 )δ3 (p2 − k1 ) + $ $ $$$ (1) (2) (9. (9.23)).22) and (9. p2 k1. (10. .26) (1) (2) (1) (2) 2p0 (2π)3 2p0 (2π)3 2k0 (2π)3 2k0 (2π)3 (2π)4 δ4 (p1 + p2 − k1 − k2 ) 3 + O(g3 ) . putting all other couplings to zero out< p1 . (10. such that ¯ ¯ f (p) ≈ δ4 (p − k1 − k2 ) ¯ ¯ d4 p f (p) = (2π)4 δ4 (p − k1 − k2 ) 43 d4 xΨ1 (x)2 Ψ2 (x)2 . pn . the function ¯ ¯ f (p) will be highly peaked around p = k1 + k2 .
also called the integrated luminosity L.The quantities Ψj (x)2 are of course related to the probability densities of the two particles in their respective beams. (10. are of the order of nanobarns (1 nb= 10−33 cm2 ). as a detector will not be able to distinguish them. (1) ¯0 k (1) (10. or that the two velocities are parallel (colliding beams). such that ¯ ¯ (k1 · k2 )2 − m2 m2 1 2 = (1) ¯ (2) ¯ k0 k0 ¯ (k0 )2 − m2 1 = v1  . approximately one collision event per second will take place. S = r 1/nr ! (in the present case S = 1/n!). Typical electromagnetic cross sections. (10. {kj })2 d 3 pi 3 i=1 2p0 (pi )(2π) n d4 x ρ1 (x)ρ2 (x) .8) Here d3 x ρ1 (x)ρ2 (x) is the number of possible interactions per unit volume at a given time and v1 − v2  is the relative velocity of the two beams. n n ¯ d 3 pi M({−pi}.7) ¯(1) ¯(2) 4k 0 k 0 Since ρj (x) will depend on the experimental situation. Hence.12) dσ = (2π)4 δ4 ( pi − k1 − k2 ) 3 ¯ ¯ 4 (k1 · k2 )2 − m2 m2 i=1 2p0 (pi )(2π) i=1 1 2 The parameter S is the inverse of the permutation factor for identical particles in the ﬁnal state. (10. L(t) = d3 x ρ1 (x. typically of the order of 1028 − 1033 cm−2 s−1 .9) ¯ After all. 44 . E2 (10. We leave it as an exercise to prove the result for the general case of parallel beams. In the weak interaction the cross sections are typically 5 to 6 orders of magnitude smaller. With a luminosity of 1033 cm−2 s−1 . dt L(t) ≡ d4 x ρ1 (x)ρ2 (x)v1 − v2  . {kj })2 ¯ ¯ S . We have assumed that either. ∗ ¯(j) ρj (x) ≡ i(Ψ∗ (x)∂t Ψj (x) − Ψj (x)∂t Ψj (x)) ≈ 2k0 Ψj (x)2 j .10) ¯ whereas for colliding beams of particles and antiparticles with mass m. −k) = E(1. −v). t)ρ2 (x. t)v1 − v2  is a ﬂux. This will avoid double counting when performing the phase space integrals. for a ﬁxed target situation k2 = (m2 . If there are nr identical particles of sort r in the ﬁnal state. where k1 = (E. Putting these results together we ﬁnd dW = (2π) δ4 ( 4 ¯ ¯ ¯ pi − k1 − k2 )M({−pi }. we should normalise with respect to the total number of possible interactions in the experimental setup. one ﬁnds E(1.11) such that both expressions reduce to v1 − v2 . (10. such that not more than one event per day will take place in that case. To consider the general case we note that we can also write dt L(t) = ¯ ¯ (k1 · k2 )2 − m2 m2 1 2 d4 x ρ1 (x)ρ2 (x) ¯ (1) ¯(2) k0 k0 .6) again using the fact that the wave packet is highly peaked in momentum space. k) = ¯2 = (E. 0). one of the velocities is zero (ﬁxed target). as we will compute later. (10. We can therefore deﬁne a machine independent diﬀerential cross section dσ by normalising the scattering probability by the total luminosity. v) and k ¯ ¯ (k1 · k2 )2 − m2 m2 1 2 = (1) ¯ (2) ¯ k0 k0 (E 2 + k 2 )2 − m4 = 2k/E = 2v .
Indeed we will see that only if M > 2m. (11. k 2 = m2 and (k − p)2 = m2 in the ﬁrst diagram. The relation with the selfenergy follows from the fact that.e. the lifetime of this particle is hence τ (p) = 1/Γ(p). The real ϕ particles can escape to inﬁnity. (11. since always k 2 = m2 and (k − p)2 = m2 .3) ﬁrst to lowest order in g Im(−Σσ (p)) = Im − ig 2 2 ig 2 = Im − 2 d4 k 1 4 (k 2 − m2 + iε)((k − p)2 − m2 + iε) (2π) d4 x G2 (x)eipx 45 .9).6) .3) p &% p &% · · · . the loop integral will contain contributions where the ϕ particles can be on the massshell and behave as a real particles. r 2 σ rϕ ≡ g . see eq. apart from the overall factor i/(2π)4 .5) The amplitude of the wave function for the σ particle consequently decays with a decay rate √ of Γ(p) = γ/ p 2 + M 2 . as soon as M > 2m. M < 2m). (11. We will now consider a simple example of a scalar ﬁeld theory with two types of ﬁelds. &% The deﬁnition of the decay rate (also called decay width) of an unstable particle is best deﬁned by considering its selfenergy. a ﬁeld ϕ(x) associated with a light particle (mass m) and a ﬁeld σ(x) associated with a heavy particle (mass M > 2m). ¨ϕ ¨ (11. γ ≡ Im(−Zσ Σσ (p))p2 =M 2 = 0. The latter indeed equals iΣ(p). We will evaluate the imaginary part of selfenergy for the σ 2point function in eq.2) V (σ.4) =− (2π)4 p2 − M 2 + iγ + iε 2 (2π)3 p + M − iγ √ The poles p0 = ± p 2 + M 2 − iγ are now complex and for γ ≪ M 2 one has in a good approximation √ √ √ 1 2 2 2 2 2 2 e−it p +M −iγ = e−it p +M e− 2 γt/ p +M . the selfenergy will be able to develop a nonzero imaginary part. 1 Σ(p) = Z (11. the loop in the ﬁrst diagram corresponds to virtual ϕ particles moving between the vertices. For the σ twopoint function in lowest order we ﬁnd p−k '$ '$ p p = &% '$ (11. e. ϕ) = 1 gσϕ2 .g. one has for each of the two external lines an extra factor √ −i(2π)2 Z as compared to the amputated 1P I 2point function. (9.11 Decay rates '$ . which can decay in the lighter particles if we allow for a coupling between one σ and two ϕ ﬁelds. in total one therefore has −iZ times the amputated 2point function. Its number will reduce as a function of time. thereby describing the decay of the σ particle. On the massshell the σ propagator is in that case modiﬁed for t > 0 to √ d4 p i d3 p Zσ e−ipx Z eip·x −it p 2 +M 2 −iγ √ 2 σ 2 e .1) where the diagram for the 1P I twopoint function is now to be evaluated using the Feynman rules in table 3 (pg. 42). p + p + k If σ is a stable particle (i. (11. However.
9) which is the result to lowest nontrivial order in g (to this order we can take Zσ = 1).10) where the full ϕ 2point function is given by 1/(p2 −m2 −Σϕ (p)+iε). In this way we easily ﬁnd the partial decay rate dΓ(p) to be dΓ(p) = (2π)4 δ4 (Σi ki − p) Mϕϕσ ({−ki }. the only part that actually contributes to Im(−Zσ Σσ (p)). in its 1P I components as follows '$ p = &% # d d d d '$ '$ dd dd d d "! d d d d d d # p dd &% dd d &% d d dd d d d d d "! p p (11. at higher orders one can generate it from the lower couplings that do occur in the Lagrangian. (10.where we used the deﬁnition of the Green’s function. It is not necessary for this coupling to occur in the Lagrangian. To any order we can. where we have implicitly deﬁned k0 ≡ kj2 + m2 . restricting ourselves to p0 > 0.g. with two ϕ particles as an intermediate state. With the help of eq. decompose the part of the σ selfenergy. As we wish to study the σ 2point √ function at t > 0 near the massshell. In its present form the formula for Γ(p) is also valid for the decay of a particle in n other particles. The ˜ √ 1P I ϕϕσ 3point function can easily be seen to be equal to Mϕϕσ ({−ki }.7) G(x) = −i (2π)3 2k0 (k) which yields (after changing x to −x at the right place) 1 Im(−Σσ (p)) = − 4 g 2 1 4 d4 x (G(x)2 + G∗ (−x)2 )eipx = δ3 (k1 (2) 2k0 (2π)3 (j) (11. (11. however. p)/(Zϕ Zσ ) (or its complex conjugate if in. The large resemblance with the formula for the cross section is no coincidence. as it should be.7). In particular. The derivation is almost identical. (11. which on the massshell reduces to Zϕ /(p2 − m2 + iε). e.and outgoing lines are interchanged). we can put p0 ≡ p 2 + m2 as well. gives Γ(p) ≡ Im(−Zσ Σσ (p)) g 2 (2π)4 = p0 4p0 d3 k1 (1) 2k0 (2π)3 d3 k2 δ4 (k1 (2) 2k0 (2π)3 + k2 − p) . 46 . since in lowest order it should coincide with the ϕϕσ 3point vertex. (11. p)2 S 2p0 d3 ki (i) 2k0 (2π)3 . (5. The total decay rate is found by integrating over the phase space of the outgoing particles Γ(p) ≡ dΓ(p). (4. (11. The symmetry factor S is the same as for the cross section in eq.11) i which.12) we can write d3 k eik·x e−ik0 (k)t .8) (1) (2) (1) (2) g 2(2π)4 d3 k1 (1) 2k0 (2π)3 d3 k2 + k2 − p) δ(k0 + k0 − p0 ) + δ(k0 + k0 + p0 ) . is always positive. eq. in eq.12). as in both cases we have to calculate the probability for something to happen (respectively a decay or a scattering).8) one now encounters Gn (x) instead of G2 (x) and g now stands of course for the coupling constant of n ϕ ﬁelds to the σ ﬁeld.
This is what Dirac took as his starting point.7) We have to demonstrate that the Dirac equation is covariant under Lorentz transformations. σ3 = 1 0 0 −1 . Dirac ﬁrst considered m = 0. (12. which proves that U is unitary (up to an irrelevant overall complex factor. σ2 = 0 −i i 0 . (12. The above equation is equivalent to β2 = 1 . however. αk } = 1δjk . because we already know from quantum mechanics how a spinor transforms under rotations p → p ′ = exp(ω · L)p .e.4) It is clear that two will be the smallest matrix dimension for which one can solve the equation 1 (αj αk + αk αj ) ≡ 1 {αj . It would have been better to treat space and time on a more equal footing in the Schr¨dinger equation.6) such that U † Uσj = σj U † U for each j.1) The question Dirac posed himself was to ﬁnd the simplest choice for αk and β. We should. The only 2 × 2 matrix that commutes with all Paulimatrices is a multiple of the identity. (12.8) . (12. one is looking for a Hamiltonian that is linear o in the momenta pj = i∂/∂xj (= −pj ). which does not aﬀect αj ).2) Dirac noted that only in case we allow αk and β to be noncommuting objects (i. We ﬁrst put the boosts to zero. since † αj αj ≡ Uσj U −1 U † −1 † † σj U = U([U † Uσj ]−1 σj U † U)U −1 = 1 .3) Historically. such that the square of the Schr¨dinger equation gives the KleinGordon equation o p2 = (−pk αk + βm)2 = p 2 + m2 0 . require that H (and hence αj ) is hermitian.3). Here σk are the Paulimatrices. the parameter in which the square root should be expanded. one can satisfy these equations. ∂t ∂x (12. but the massless case (m = 0) is somewhat simpler. (12. It is not hard to prove that in a two dimensional 2 2 representation all solutions to this equation are given by αj = ±Uσj U −1 . familiar from describing spin onehalf particles. the wave function Ψ(x) becomes a complex two dimensional vector. as it allows one to use β = 0 and αk = σk for a solution of eq.5) where U is an arbitrary nonsingular complex 2 × 2 matrix. matrices). This narrows U down to a unitary matrix. (12. we considered the square root of the o √ KleinGordon equation.12 The Dirac equation To obtain a Lorentz invariant Schr¨dinger equation. (12. σ1 = 0 1 1 0 . i ∂Ψ ∂Ψ = HΨ = −iαk k + βmΨ . Since the Hamiltonian is now a 2 × 2 matrix. As the o Schr¨dinger equation is linear in p0 = i∂/∂t. which describes particles with spin h/2 ¯ p0 Ψ(x) = ∓p · σΨ(x) . also called a spinor. i Ψ(x) → Ψ′ (x′ ) = exp( ω · σ)Ψ(x) . 1 2 (αj αk + αk αj ) = 1δjk and αj β + βαj = 0 . 2 47 (12. This had the disadvantage that the Hamiltonian H = p 2 + m2 contains an inﬁnite number of powers of p 2 /m2 .
since the Pauli matrices form a complete set of anticommuting matrices. the socalled Dirac sea. Since also f1 (0) = f2 (0) it follows that f1 (1) = f2 (1).11) becomes ∓p ′ · σ ′ Ψ(x) = ∓p · σΨ(x). These reﬂect the commutation relations of the generators iσj /2. adX(Y ) ≡ [X.12) which is derived from the fact that f1 (t) = etX Y e−tX and f2 (t) = exp(ad(tX))Y satisfy the same diﬀerential equation. i.Here Li are real 3 × 3 matrices that generate the rotations in IR3 Li = εijk jk . To prove this. (12. The smallest size turns out to be a 4 × 4 matrix.10) which should reduce to eq. Obviously particle number will no longer be conserved and also the Dirac equation will require “second quantisation” and the introduction of a ﬁeld. show that this describes the fact that SU(2) (the group of unitary transformations acting on the spinors) is a representation of SO(3) (the group of rotations in IR3 ).11) (12. (12. Antiparticles were predicted by Dirac because the only way he could make the theory consistent was to invoke the Pauli principle and to ﬁll all the negative energy states. In the present case. being the above equation. A hole in this sea of negative energy states.13) the r. Y ] . (12. Using eq. Only when we required localisation of the wave function inside the lightcone. We will later.7). (12. β= 48 12 ⊘ ⊘ −12 .14) .e. Lj ] ≡ Li Lj − Lj Li = −εijk Lk . At that time antiparticles were unknown. as its o eigenvalues are ±p . and it is not bounded from below. we were forced to consider negative energy states. 2 2 (12. In the scalar theory we could avoid this by just considering the positive root of the KleinGordon equation. where σ ′ = exp(ω · L)σ. fi (t)]. in the context of nonAbelian gauge theories.8) we get i i p0 Ψ(x) = ∓p ′ · exp(− ω · σ)σ exp( ω · σ)Ψ(x) .(12.h.9) such that [Li . These holes describe the antiparticle with the same mass as the particle. For 2 × 2 matrices this is impossible. For the massive case Dirac ﬁrst incorrectly thought that the positive energy states describe the electron and the negative energy states the proton. of eq. σk ] = (ω · L)kj σj 2 2 . restricting to one of the eigenstates would break the rotational invariance of the theory. The interpretation of this Schr¨dinger equation caused Dirac quite some trouble. then corresponds to a state of positive energy. (12. we work out the Dirac equation in the rotated frame. For the massive Dirac equation we need to ﬁnd a matrix β that anticommutes with all αi . p0 Ψ(x) = ∓p · σΨ(x) → p0 Ψ′ (x′ ) = ∓p ′ · σΨ′ (x′ ) . The following representation is usually chosen αi = ⊘ σi σi ⊘ .s. (12. dfi (t)/dt = [X. To show the covariance of the Dirac equation under rotations. Applying this result to Y = σk and X = −iω · σ/2. we use the following general result for matrices X and Y eX Y e−X = exp(adX)(Y ) . which will be discussed later. using the fact that i i ad(− ω · σ)σk = [− ω · σ.
such that the Dirac equation becomes (−iγ µ ∂µ + m)Ψ(x) = (−γ µ pµ + m)Ψ(x) ≡ (−p + m)Ψ(x) = 0 . this equation is covariant under Lorentz boosts if there exists a nonsingular complex 4 × 4 matrix S.23) Applying eq.22) satisﬁes eq. γ λ ] = − 1 ωµν [γ µ γ ν . (12. (12. One says that S(ω) is a representation of K(ω). ˜ β= ⊘ −12 −12 ⊘ . (12. ωµν = gµλ ω λ ν = −ωνµ .20). 4 i σ µν ≡ [γ µ .17) As for the covariance under rotations. (12. (12.15) We leave it as an exercise for the student to show that these two representations are related by a 4 × 4 unitary transformation U.12) gives the proof for eq.16) under Lorentz transformations (note that now Ψ(x) has four complex components).19) (12. (12.19) we ﬁnd i [ ωµν σ µν . Using the antisymmetry of ωµν and eq. γ λ ] 4 4 = − 1 ωµν (γ µ γ ν γ λ − γ λ γ µ γ ν ) 4 1 = 4 ωµν (γ λ γ µ γ ν + γ µ γ λ γ ν − 2γ µ g λν ) 1 = 2 ωµν (g λµ γ ν − γ µ g λν ) = ω λ µ γ µ . Like for the rotations. (12. For the massless case it is often more convenient to use the socalled Weyl representation αi = ˜ σi ⊘ ⊘ −σi . ˜ To study the covariance of the Dirac equation p0 Ψ(x) = (−αi pi + βm)Ψ(x) (12.e.20) where K µ ν is the Lorentz transformation acting on the momenta as p′µ = Kµ ν pν and on the coordinates as xµ′ = K µ ν xν . γ µ = Uγ µ U −1 . γ ν ] 2 (12. / The Dirac gamma matrices satisfy anticommuting relations {γ µ . such that Ψ(x) → Ψ′ (x′ ) = SΨ(x) and S −1 γ µ S = K µ ν γ ν .20).for which the nonrelativistic limit has a simple form. K can be written as an exponent K = exp(ω) .18) (12. This is because the 49 . Note that in general S(ω) is not a unitary transformation. γ ν } ≡ γ µ γ ν + γ ν γ µ = 2g µν . it will be proﬁtable to introduce a “fourvector” γ µ of 4 × 4 matrices γ µ ≡ (γ 0 . βαi) .21) Here ω µ ν is a 4×4 matrix. (12. which is antisymmetric when one of its indices is raised or lowered by the metric. (12. i. γ i ) = (β. We will now prove that i S = S(ω) = exp(− ωµν σ µν ) .
(12.boosts form a noncompact part of the Lorentz group. With ω0k = 0. (12. There is. The boost parameters are described by ω0i . 2 As S is not unitary Ψ† (x)Ψ(x) is no longer invariant under Lorentz transformations. (12. We can verify eq.24) ˜ which is most easily proven in the Weyl representation.22) are independent of the representation in which we give the gamma matrices. ∂µ j µ (x) = 0. (12. To be precise. (12. one ﬁnds S(ω) = 12 ⊗ exp(iω · σ/2). (12. γ 0 ] = −i ˜ ˜ ⊘ 2 ⊘ −σk . as any two such representations have to be related by a unitary trans˜ formation.22) i σk σij = σ ij = [˜ i . but one is the inverse of the other ˜ (and neither is unitary).25) Note that eqs.7) with the plus ˜ sign and the lower block corresponds to the minus sign. like the Dirac equation for m = 0 (as αi is block diagonal). (12.26) ˜ In the Weyl representation. a relation between S † and S −1 . The combination Ψ† (x)γ 0 will occur so often. In the Weyl representation S(ω) is block diagonal. γ 0 S † γ 0 = S −1 . (12. whereas for the lowerright 2 × 2 block we ﬁnd exp(− 1 ω0k σk ). (12. The upper block corresponds to eq. that it has acquired its own symbol Ψ(x) ≡ Ψ† (x)γ 0 .28) It transforms under a Lorentz transformation as Ψ(x) → SΨ(x) and Ψ(x) → Ψ(x)S −1 . For a boost in the x direction we have that χ = ω01 is related to the boost velocity by v1 = − tanh(χ). S restricted to the upperleft 2 × 2 block equals 1 exp( 2 ω0k σk ).29) We can use this to build the required Lorentz scalars. however. namely the time component j 0 (x) of a conserved current j µ (x) = Ψ† (x)γ 0 γ µ Ψ(x) . γ j ] = εijk ˜ ˜ γ ˜ ⊘ 2 ⊘ σk .15). (12. (12.e. it acts on each 2 × 2 2 block by the same unitary transformation. But we claim it is nevertheless a probability density. i. vectors and tensors scalar: vector: tensor: ′ ′ Ψ (x′ )Ψ′ (x′ ) Ψ (x′ )γ µ Ψ′ (x′ ) Ψ (x′ )σ µν Ψ′ (x′ ) ′ = = Ψ(x)S −1 SΨ(x) Ψ(x)S −1 γ µ SΨ(x) = Ψ(x)Ψ(x) = K µ ν Ψ(x)γ ν Ψ(x) = K µ λ K ν κ Ψ(x)σ λκ Ψ(x) = Ψ(x)S −1 σ µν SΨ(x) 50 .17) and (12. i σk γ ˜ σ0k = σ k0 = [˜ k .20) and (12.27) We leave it as an exercise to show that the Dirac equation implies that the current is conserved. For K we ﬁnd in this case cosh(χ) sinh(χ) 0 0 sinh(χ) cosh(χ) 0 0 K= 0 0 1 0 0 0 0 1 . whereas σij is hermitian and σk0 is antihermitian.8) by using the fact that the Lorentz transformations contain the rotations through the identiﬁcation ˜ ωk = − 1 εijk ωij . as follows from the explicit ˜ ˜ expressions obtained from eqs. S splits again in two blocks. since γ 0 = β commutes with σij ˜ ˜ but anticommutes with σk0 (as always roman indices run from 1 to 3 and greek indices run ˜ from 0 to 3).
36) ⊘ 12 12 ⊘ .34) This implies that we can introduce projection operators 1 P± ≡ 2 (1 ± γ5 ) . (12. (12.32) which follows from the fact that (γ 0 )† = β † = β = γ 0 and (γ i )† = (βαi )† = αi β = −γ i . (12. γi = ⊘ −σi σi ⊘ . where 12 ⊘ γ5 = i˜0 γ1 γ2 γ3 = −i˜ 1 α2 α3 = ˜ γ ˜ ˜ ˜ α ˜ ˜ . (12.37) 1 ˜ ˜ ˜ ˜ such that P+ = 2 (1 + γ5 ) projects on the two upper and P− = 1 (1 − γ5 ) on the two lower 2 components of the fourspinors.33) (12.35) ← ← ← 12 ⊘ ⊘ −12 . (12.40) . or from the explicit representation of the gamma matrices γ0 = Hence 0 = Ψ† (x)(i(γ µ )† ∂µ +m)γ 0 = Ψ(iγ 0 (γ µ )† γ 0 ∂µ +m) = Ψ(iγ µ ∂µ +m) . Their role is best described in the Weyl representation. −p · σ −m12 −m12 Ψ . (12. which satisfy (P± )2 = P± and P± P∓ = 0.The Lagrangian is a Lorentz scalar. − Ψ(x) δ ← − (12.31) The second equation is the complex conjugate of the ﬁrst. for m = 0 we have Ψ(x) = Ψ+ (x) Ψ− (x) . − Ψ(x) δ − ← δS = Ψ(x)(−iγ µ ∂µ −m) = 0 . it can be considered independent of Ψ(x) and the following Lagrangian SDirac = d4 x LDirac = d4 x Ψ(x)(iγ µ ∂µ − m)Ψ(x) . (12. In the massless case these two components are decoupled. which we chose such that its equations of motion reproduce the Dirac equation. 51 (12. As Ψ(x) is complex. because the gamma matrices satisfy (γ µ )† = γ 0 γ µ γ 0 . An important role will be played by a ﬁfth gamma matrix γ5 ≡ iγ 0 γ 1 γ 2 γ 3 = which anticommutes with all γ µ (see problem 21) γ5 γ µ = −γ µ γ5 and (γ5 )2 = 14 .38) ⊘ −12 ˜ ˜ p0 Ψ(x) = (−p · α + mβ)Ψ(x) = Hence.30) gives the EulerLagrange equations δS = (iγ µ ∂µ − m)Ψ(x) = 0 . Ψ† (x)(i(γ µ )† ∂µ +m) = 0.39) p0 Ψ± (x) = ∓p · σΨ± (x) . p·σ (12.
See problem 22 for more details. x) = γ5 γ0 γ 2 Ψ∗ (t. where P stands for parity and T for timereversal.3) As any k can be obtained from a boost to k = 0. Ψ(x) = ˜ Ψ(k)e−ikx . They are called pseudoscalars and pseudovectors. although experiment has not yet been able to rule out a (tiny) mass for this particle (mνe < 10eV ). One can deﬁne in that case consistently a particle with a ﬁxed helicity. will be discussed in section 17. x) (12. This implies that the Lorentz covariant combinations Ψ(x)γ5 Ψ(x) and Ψ(x)γ5 γ µ Ψ(x) are not invariant under parity and timereversal.7). −(k0 + m)12 (13. charge conjugation C.5) 1 0 is the spinup and is the spin0 1 down state. for m = 0 the two helicity eigenstates decouple.1) ˜ where Ψ(k) are complex fourvectors that satisfy ˜ ˜ (k − m)Ψ(k) ≡ (kµ γ µ − m)Ψ(k) = 0 . As long as m = 0. (13. / The Lorentz invariance implies ˜ ˜ Ψ′ (k ′ ) = S(ω)Ψ(k) . x) and T Ψ(x) ≡ Ψ′ (−t. ′ kµ = K(ω)µ ν kν 2 k0 = k 2 + m2 . Apart from the invariance of the Dirac equation under Lorentz transformations and translations (which are obvious symmetries of LDirac) we also often want invariance under parity (x → −x) and timereversal (t → −t). (12.2) . These combinations play an important role in the weak interactions. One easily checks that P Ψ(x) ≡ Ψ′ (t. helicity is not conserved. But as we saw. whose opposite helicity state does not occur (although its antiparticle has opposite helicity). the Dirac equation can be solved by Fourier decomposition in plane waves. −x) = γ0 Ψ(t. where parity is not a symmetry. The identiﬁcation of the spin degrees of freedom follows from the behaviour of where ΨA is a spin onehalf twospinor of which 52 . 13 Plane wave solutions of the Dirac equation d3 k 2k0 (k)(2π)3 As usual. there are two independent solutions both of the form ˜ Ψ+ (0) = .which is identical to eq.4) We see that for k0 > 0 (k0 = m). The eigenstates of the projection operators P± are called helicity eigenstates. A very important example of such a particle is the neutrino. (13.41) satisfy the Dirac equation. (13. A third discrete symmetry. all solutions of the Dirac equation can be obtained from the ones at rest with k = 0 (see problem 19) ˜ (γ 0 k0 − m)Ψ(0) = (k0 − m)12 ⊘ ΨA 0 ⊘ ˜ Ψ(0) . (13.
to be precise u(α) (0) = u0 / 2m and v (α) (0) = v0 / 2m. This normalisation also implies that u(α) (k)u(β) (k) = −v (α) (−k)v (β) (−k) = 2mδαβ . (13. u0 = .11) (β) . The computation for v (α) (k) is left as an exercise. In that case we can. (13. σij = εijk σk ⊘ ⊘ σk . (13. (12. For any frame. (12. not transform to the restframe. We leave it as an exercise to verify that also in the Dirac representation of the gamma matrices.25)). (12. 0 0 0 0 0 0 0 0 (2) (1) .25)). however. v0 = . 2) solutions. The fact that we ﬁnd a result that is independent of k is consistent with our claim that these spinors can also be obtained by applying the appropriate ˜ ˜ boost to k = 0. which transform under rotations as ΨA .6) such that ΨA is easily seen to transform under a rotation as in eq. for k = 0 and m = 0 easily seen to be proportional (α) √ (α) √ ˜ to the solutions Ψ± (0).9) For example u(α) (k)u(β) (k) (α) u0 γ 0 (k + m)† γ 0 (k + m)u0 / / = (m + k0 ) (β) (α) (α) (β) u0 (k + m)2 u0 / = (m + k0 ) (β) (α) (β) = . That these spinors indeed satisfy eq. the discussion below eq.8) (cmp.2) follows from the fact that (k − m)(k + m) = k 2 − m2 = 0 . (12. There are also two solutions for k0 < 0 (k0 = −m) of the form ˜ Ψ− (0) = 0 ΨB . u0 = .e.7) is the where likewise ΨB is a spin onehalf twospinor of which spindown state. 2) and negative energy (v (α) (−k) with α = 1. 1 0 is the spinup and 0 1 (13. (13. i. like in the Weyl representation (see eq.10) u0 (k 2 + m2 + 2k m)u0 / (m + k0 ) (α) u (k 2 + m2 + 2k0 m)u0 = 0 (m + k0 ) = 2mδαβ where we used k 2 = m2 and u0 γ µ u0 = δαβ δµ0 (see eq.8) 0 1 1 0 k0 = k 2 + m2 : u(α) (k) = u0 (α) k0 = − k 2 + m2 : v (α) (−k) = v0 (α) These solutions naturally split in positive energy (u(α) (k) with α = 1. for any value of k. we will deﬁne the four independent solutions of the Dirac equation as k+m / m + k0  k+m / m + k0  0 1 1 0 (2) (1) . (13. v0 = .33)). The normalisation we have chosen allows us to treat massless fermions at the same footing.˜ Ψ under rotations (which leave k = 0). since Ψ(k)Ψ(k) is a Lorentz scalar. / / 53 (13.
for m = 0. (k0 > 0) v (α) (k) v † (β) (k) = v0 (m − k )† (m − k )v0 / / m + k0 (α) † (β) = v0 2 (k0 + k 2 + m2 − 2mk0 γ 0 )v0 m + k0 (β) = (13.15) = (α) † (β) v0 2 (k0 + k 2 + m2 + 2mk0 )v0 m + k0 (β) = 2k0 δαβ .13) . or by ﬁrst computing them in the restframe where Λ± (0) = 1 (1 ± γ 0 ).27)).16) d3 k by k ). we use the fact that the probability density can be deﬁned in terms of the zerocomponent ρ(x) of the conserved fourvector j µ (x) = Ψ(x)γ µ Ψ(x) (see eq. In that case we ﬁnd 2 Λ+ (k) = k+m / u(α) (k) ⊗ u(α) (k) = 2m 2m α=1 . Λ− (k) = − Λ± (k) = ±k + m / (±k + m)(1 ± γ 0 )(±k + m) / / = 4m(m + k0 ) 2m Λ∓ (k)Λ± (k) = 0 . . e. (12. To see that our plane waves have the correct amplitude. (13. 1 This implies that the plane wave solutions (k0 = (k 2 + m2 ) 2 ) u(α) (k) 2k0 (2π)3 e−ikx .e. (13. we can deﬁne projection operators (which are 4 × 4 matrices) v (α) (k) ⊗ v (α) (k) −k + m / = . as the energy k0 . using v0 γ µ v0 = −δµ0 δαβ (see eq. there is no relation between dα and bα . Finally we note that. ˜ ˜ Indeed.17) 2k0 (k)(2π)3 α=1 Since Ψ(x) is complex. Ψ† (k)Ψ(k) transforms as a density. 2m 2m α=1 (13. ± (13. 54 .14) This can be veriﬁed by a direct computation.Note that v (α) (k) can also be viewed as a positive energy solution for the complex conjugate of the Dirac equation k0 = k 2 + m2 : (k + m)v (α) (k) = 0 . (12. (α) † u(α) (k)† v (β) (k) = 0 . are normalised to one (of course. in a ﬁnite volume we replace (2π)3 by V and As for the scalar ﬁeld we can introduce a Dirac ﬁeld Ψ(x) = d3 k 2 † bα (k)u(α) (k)e−ikx + dα (k)v (α) (k)eikx . v (α) (k) 2k0 (2π)3 eikx . (13. or ρ(x) = Ψ† (x)Ψ(x) . In the quantum ﬁeld theory these will play the role of the annihilation operators for the (anti)particles. 2 . Independent of the frame Λ± satisfy Λ2 (k) = Λ± (k) .19) 2 2 whose proof requires some gamma matrix gymnastics. α = 1. a result that can also be derived from eq.g.33)). (13. (13. We leave the other identities as an exercise.12) It will play the role of the wave function for the antiparticles (the holes) in the Dirac ﬁeld theory. / (13.18) which can be veriﬁed from the explicit form of u and v. and we ﬁnd u(α) (k)† u(β) (k) = v (α) (k)† v (β) (k) = 2k0 δαβ (α) † (13.20) Note that Tr(Λ± (k)) = 2.9). i.
2) d3 x Ψ† (x)(−iαj ∂j + mβ)Ψ(x) d3 k k0 (k) α b† (k)bα (k) − dα (k)d† (k) α α . k >. can lower the energy by an arbitrary amount. {dα (k). It is wellknown how Dirac repaired this problem. and most importantly.17) for the expansion of the Dirac ﬁeld in plane waves. (13.14 The Dirac Hamiltonian In the Dirac equation we encountered an additional diﬃculty. if we deﬁne a two particle state as k.1) The canonical momentum is hence πa (x) = such that H = = = ˙ d3 x (πa (x)Ψa (x) − L) = d3 x Ψ(x)(−iγ j ∂j + m)Ψ(x) δS = (Ψ(x)iγ 0 )a = iΨ∗ (x) . He postulated that all negative energy states are occupied. As annihilation lowers the total energy by the energy of the annihilated state. which in this case is negative. is that conservation of energy and momentum implies that it α creates an antiparticle as a hole in the Dirac sea. as the negative energy states. and that the states satisfy the Pauli principle. We used eq. b† (p)} = δαβ δ3 (k − p) . with momentum k and helicity 1 for α = 2. This implies that one should use anticommuting relations for the creation and annihilation operators {bα (k). (13. the anticommutation relations imply that k. (14.3) Note the resemblance with eq. described by the dα (k). bβ (p)} = 0 . namely that the negative energy solutions even arise at the level of the classical theory. In ﬁeld theory the negative energy solutions had an interpretation in terms of antiparticles.8). p >= −p. The ﬁeld theory Hamiltonian for the Dirac ﬁeld no longer has this property. p >≡ b† (k)b† (p)0 > (suppressing the spinor indices). The wave function for the negative energy state is given by exp(ikx)v (α) (k)/ k0 and has momentum −k 1 (k0 ≡ (k 2 +m2 ) 2 ). (14. (12. and the ﬁeld theory Hamiltonian was still positive. The Hamiltonian can again be derived through a Legendre transform of the Lagrangian S= d4 x L = d4 x Ψ(x)(iγ µ ∂µ − m)Ψ(x) .e. β {dα (k).1) for the middle term. β (14. (13. bβ (p)} = 0 . The reason to associate its Fourier coeﬃcient in eq. i. {dα (k). dβ (p)} = 0 . and this would make the vacuum unstable. {bα (k). A hole in the Dirac sea is by deﬁnition the state that is obtained by annihilating a negative energy state in the Dirac sea.4) Indeed. b† (p)} = 0 . From this result it is clear that the Hamiltonian is not bounded from below. two particles can not occupy the same quantum state (it is only in that case that we can make sense of what is meant with ﬁlling all negative energy states). the net √ energy is raised.17) with a creation operator d† (k). 2 whereas for α = 1 the helicity is − 1 (hence the helicity and momentum are opposite to the 2 55 . see eq. d† (p)} = δαβ δ3 (k − p) . β {dα (k). bounded from below (see section 2 and problem 5). a ˙ a (x) δΨ (14.
which is however outside the scope of these lectures. as for a complex scalar ﬁeld.7) where as before the Hamiltonian H is the spatial integral over the Hamiltonian density H(x).12)) is independent of the details of the path integrals. Important is also to note that the anticommuting relations are crucial to guarantee locality of the Dirac ﬁeld. we see that with the present interpretation the Dirac Hamiltonian is bounded from below H= d3 k k0 (k) α † b† (k)bα (k) + dα (k)dα (k) − 1 α . (14.8) where GF (x − y) = Hence. see problem 17. ¯¯ ¯ (14. (14. (7. we can couple a source to the free Dirac ﬁeld (we again introduce ε as the expansion parameter for taking the interactions due to the source into ¯ account in Hamiltonian perturbation theory) ¯¯ ¯ L(x) = Ψ(x)(iγ µ ∂µ − m)Ψ(x) − εJ (x)Ψ(x) − εΨ(x)J (x) . we note that the coupling to an electromagnetic ﬁeld Aµ (x) should be achieved through the current j µ (x). section 5).negative energy state it annihilates). the Green’s function for fermions is in Fourier space given by GF (p) = (p − m + iε)−1 . Ψ† (x′ )} = 0 for (x − x′ )2 < 0 . k H(x) = Ψ(x)(−iγ ∂k + m)Ψ(x) + εJ (x)Ψ(x) + εΨ(x)J (x) . 56 (14. The wave function of an antiparticle with momentum √ k is hence given by exp(−ikx)v (α) (k)† / k0 . In socalled supersymmetric ﬁeld theories this is no longer an accident as the Dirac ﬁelds will be related to the scalar ﬁelds by a symmetry. deﬁned in eq. Note that the sources J (x) and J (x) are independent.9) . plausible there is also for the fermions a path integral formulation. In this case Dirac ﬁelds speciﬁed in diﬀerent regions of spacetime that are spacelike separated should anticommute. as for the scalar ﬁelds. as in the scalar theory. / which is the inverse of the quadratic part of the Lagrangian. Note that this constant has its sign opposite to the scalar case (and is in magnitude four times as large). In problem 24 you are asked to prove that in Hamiltonian perturbation theory one obtains < 0Texp(−i T 0 H(t)dt) 0 > eiE0 T = 1 − i¯2 ε = 1−i × εJ ¯ ¯ d4 xd4 y J (x)GF (x − y)J (y) + O(¯3 ) ε > × + O(¯3 ) . nevertheless.11) and (7. however.27) d4 x j µ (x)Aµ (x) = d4 x Ψ(x)γ µ Aµ (x)Ψ(x) . eqs. If we now use the anticommutation relations of creation and annihilation operators (as a consequence of the Pauli exclusion principle). To conclude.6) b Also. ε εJ ¯¯ . ¯ H = d3 x H(x). (12. (14. And indeed. It becomes.5) As in the case for scalar ﬁeld theory we can normalise the energy of the vacuum state to zero by adding a (inﬁnite) constant. the anticommuting properties of the Dirac ﬁeld play a crucial role (cmp. In problem 24 it will be evident that. as splitting of a square (cmp.10) d4 p e−ip(x−y) = (2π)4 / − m + iε p d4 p (p + m)e−ip(x−y) / 4 (2π) p2 − m2 + iε (14. in problem 24 you are asked to prove that {Ψa (x). This will be the subject of the next section.
(14. in order to perform the path integral. the positrons. which can be normalised to zero. which describe bosons. whose time component ρ(x) at the quantum level is no longer positive deﬁnite. whereas d† (k) corresponds with the creation α operator of a spinup (α = 2) or a spindown (α = 1) positron of momentum k.3) . A function of a single Grassmann variable has a ﬁnite Taylor series f (θ) = a0 + a1 θ 57 . Using the minimal coupling deﬁned as in eq. 15 Path integrals for fermions For scalar ﬁelds. the coupling is gauge invariant. θj } ≡ θi θj + θj θi = 0 . After all we want the state with all negative energy states occupied to have zero charge. with opposite electric charge +e. b† (k) corresponds with the creation operator of a spinup (α = 1) or a α spindown (α = 2) electron of momentum k. (14. (14.4) one ﬁnds d3 x ρ(x) = ˆ = = d3 x Ψ† (x)Ψ(x) d3 k α b† (k)bα (k) + dα (k)d† (k) α α b† (k)bα (k) − d† (k)dα (k) + 1 α α d3 k α d3 k α = Q0 /e + b† (k)bα (k) − d† (k)dα (k) α α . we used real or complex numbers (the eigenvalues of the operators). In particular. since (cmp. (14.12) The current j µ (x) is the charge density current.31)) Ψ(x)(iγ µ Dµ − m)Ψ(x) = Ψ(x)(iγ µ ∂µ − m)Ψ(x) + ej µ Aµ (x) . Using the anticommuting properties of eq. To this end we introduce a socalled Grassmann algebra. a Grassmann variable squares to zero θ2 = 0 (15.13) The vacuum value of this operator is indicated by the (generally inﬁnite) constant Q0 /e. (15. (3.1) A Grassmann variable can be multiplied by a complex number.11) one can ﬁx the normalisation of the electromagnetic current to be J µ (x) = −ej µ (x). which exists of Grassmann variables θi that mutually anticommute {θi . (3. J µ (x) = −ej µ (x).1) (the free Dirac Lagrangian) under global phase transformations. eq. With the above normalisation of the electric current. which can also be seen as a consequence of the Noether theorem applied to the invariance of eq.2) (15. Dµ Ψ(x) ≡ (∂µ − ieAµ (x))Ψ(x) . To summarise. with which it commutes.35). For fermionic ﬁelds it is essential to build the anticommuting properties into the path integral. we see that the b modes can be identiﬁed with the electrons with charge −e and the d modes with their antiparticles.Since this current is conserved. (14.
as opposed to anticommuting) degree of freedom. Indeed dθ θ = 1 . the normalisation dx Ψ∗ (x)Ψ(x) = 1 (15.8) is a representation of the state Ψ >. which in the spinor representation is given by a 2 × 2 matrix 0 1 0 0 b= .8) (i. we anticipate Ψ∗ (θ) ≡< Ψθ >= a∗ θ + a∗ 0 1 . commuting. For the same reason dθ θ (which is itself a commuting object. b† → θ). we have < Ψ =< 0a∗ + < 1a∗ =< 1b† a∗ + < 1a∗ . (15. 0) and < 1 = (0. as is any even product of Grassmann variables) can be seen as a number. and using θ2 = 0. (15.e.10) 0 1 0 1 As in eq. b† = .5) 0 > is the vacuum state (i. In the latter case.9) is an important property we would like to impose here too (keeping in mind that for path integrals we need to insert completeness relations).e. 1). As < 0 = (1.and spans a two dimensional (real or) complex vector space. We will now look for properties of θ such that Ψ(θ) ≡< θΨ >= a0 + a1 θ (15.4) −W 0 0 W (15.7) 0 0 1 0 We note that b2 = (b† )2 = 0.13) . Let us introduce the following notation 0 >= With respect to the Hamiltonian H0 = 1 2 1 0 .11) We wish to deﬁne integration over Grassmann variables.12) Since the norm should be a number. the latter should vanish. a property it has in common with a Grassmann variable. (15. such that the normalisation of the wave function is as usual < ΨΨ >= a0 2 + a1 2 = dθ Ψ∗ (θ)Ψ(θ) . all possible ingredients have been discussed.6) where b is the fermionic annihilation operator. (15. (15.e. Demanding the socalled Grassmann integration to be linear in the integrand. 58 dθ 1 = 0 (15. similar to Ψ(x) =< xΨ > for the case of a single bosonic (i. and as dθ 1 is itself a Grassmann variable. This is exactly what we need to describe a spin onehalf particle. (15. (15. 1 >= 0 1 . An arbitrary spinor can be written as a linear combination of these two states Ψ >= a0 0 > +a1 1 >= a0 0 > +a1 b† 0 > . the state with lowest energy) and we interpret 1 > as the oneparticle state (with energy W above the vacuum).
(15.22) . θ′ ) ≡ M11 θ′ + M12 + M21 θ′ θ − M22 θ = M11 θ′ + M12 − M21 θθ′ − M22 θ Indeed. which can be used to write the inﬁnitesimal evolution operator exp(−iH∆t) = 12 − iH∆t + O(∆t2 ) . and a consequence of the fact that the Taylor series of any function of a Grassmann variable truncates ˜ ˜ exp(θx) = 1 + θx . dθ′ M(θ.21) This is valid both for x a Grassmann variable (in which case the ordering of x with respect ˜ to θ is important. θ′ )Ψ(θ′ ) . (15. Another useful property of Grassmann integration is that (y is a number) dθ exp(θx + y) = x exp(y) . (15. MΨ >= Ψ′ > or M11 M21 M12 M22 a0 a1 = ′ a0 ′ a1 . which in the spinor representation is given by 2 × 2 matrices. θ′ )∆t = θ′ − θ − i∆t(H11 θ′ + H12 − H21 θθ′ − H22 θ) ˜ ˜ = dθ exp(θ[12 (θ. Note that dθ is considered as an independent Grassmann variable (which is important to realise when multiple Grassmann integrations are involved). ˜ ˜ dθ exp(θx) = x .18) where H is a (possibly timedependent) 2 × 2 matrix. with the opposite ordering the result is −x). or for x a complex number. θ′ )Ψ(θ′ ) = = = dθ′ (M11 θ′ + M12 − M21 θθ′ − M22 θ)(a0 + a1 θ′ ) . (15. θ′ ) − iH(θ. θ′ ) − iH(θ.20) The last identity is exact. We can now study the action of an operator (like a Hamiltonian H) on a state. (15. θ′ )∆t]) . θ′ ) = θ′ − θ . (15. 59 (15. this gives Ψ′ (θ) = a′0 + a′1 θ = M11 a0 + M12 a1 + (M21 a0 + M22 a1 )θ ≡ provided we deﬁne M(θ. In the Grassmann representation this reads 12 (θ.16) dθ′ M(θ.17) = M12 a1 + M11 a0 + (M21 a0 + M22 a1 )θ The 2 × 2 identity matrix is hence represented by (note the sign) 12 (θ.19) (15.is easily seen to give the desired result.14) Translated to Grassmann variables. (15.15) dθ′ {a0 M12 − a0 M22 θ + (a1 M12 + a0 M11 )θ′ − (a0 M21 + a1 M22 )θθ′ } dθ′ θ′ {(a1 M12 + a0 M11 ) + (a0 M21 + a1 M22 )θ} .
24) This means that the CampbellBakerHausdorﬀ formula (eq. θi )Ψ(θi ) (15. The combination dθi dθi is likewise Grassmann even. θi+1 . ﬁrst by one step. θi )∆t] + O(∆t2 ) .44)) can be extended to this case. (15. θi+1 . θ′ y]) . θi+2 . It behaves as if x and y are matrices. ti )θi >< θi Ψ >≡ = dθi+1 Ui+1 (θi+2 . provided of course H is diagonal. θi+2 . and the pair can be shifted to any place in the expression for the path integral.28) 1 ˜ ˜ ˜ θj {θj − θj+1 − iH(tj . Hence. More precisely 1 exp(θx) exp(θ′ y) = (1 + θx)(1 + θ′ y) = 1 + (θx + θ′ y) + 1 (θx + θ′ y)2 + 2 [θx. exp θ[12 (θ. If H is diagonal. θi+1 )∆t]) exp θi [θi − θi+1 − iH(ti . θi )∆t] Ψ(θi ) + O(∆t2 ) .24) to the above product of exponentials. for x and y arbitrary elements of the Grassmann algebra. any change in the ordering can at most given an additional minus sign. θi H(ti . (6. We now apply eq.19) and (15. θi+1 ). ti )θi >≡ Ui (θi+1 . θj+1 . (15. as they are Grassmann variables themselves. which in the Grassmann representation is given by (see eqs. θi )∆t] Ψ(θi ) + O(∆t2 ) . θi+1 . θi+1 )∆t] × ˜ ˜ dθi dθi exp θi [θi − θi+1 − iH(ti . θ′ ) − iH(θ. (15. θi ) = dθi exp θi [θi − θi+1 − iH(ti . ˜ ˜ exp(θi+1 [θi+1 − θi+2 − iH(ti+1 . (15.25) ˜ ˜ dθi+1 dθi+1 exp θi+1 [θi+1 − θi+2 − iH(ti+1 . such that dθi < θi+1 U(ti+1 . θ′ y] 2 1 = exp (θx + θ′ y + 2 [θx. as it should since the Grassmann representation originates from a 2 × 2 matrix representation. It truncates after the single commutator term as neither θx nor θy can appear more than once. Let us apply this to the evolution operator. θi )∆t] = exp i+1 j=i (15. θi+2 .23) n! n! (n − 1)! n=0 n=1 n=0 ∞ In general it is not true that the exponential function retains the property exp(x + y) = exp(x) exp(y). θj )∆t} − 2 ∆t2 [θi+1 H(ti+1 .27) Note that we have to be careful where we put the diﬀerentials. (15.20)) ˜ ˜ < θi+1 U(ti+1 . θi+1 . (15. to dθi < θi+2 U(ti+2 . θi+1 . θi )] 60 . as will often be the case for the application we have ˜ in mind. θ′ )∆t] will be a commuting object (socalled Grassmann even) and it does not matter if we put one of the diﬀerentials on one or the other side of ˜ the exponential. we use that exp(θx+ y) = ∞ ∞ 1 n 1 1 (θx+ y)n = y + θx y n−1 = (1 + θx) exp(y) . θi+1 ) dθi Ui (θi+1 .To prove this.26) which can be iterated. ti )θi >< θi Ψ >= ˜ ˜ dθi dθi exp θi [θi − θi+1 − iH(ti .
The generalisation of eqs.27) and (15.28) to N steps is now obvious and for H diagonal one easily proves that the limit N → ∞ can be taken: dθ < θ′ U(T. whereas H is shifted by a multiple of the identity that vanishes linearly in ∆t. To be precise. (15. the CampbellBakerHausdorﬀ formula truncates to the trivial term both in the matrix and in the Grassmann representations.20)) we ﬁnd ˜ ˜ dθi+1 [θi+1 H(ti+1 . θi )] = ˜ ˜ 2θi+1 θi θi+2 H21 (ti+1 )H12 (ti ) − θi H12 (ti+1 )H21 (ti ) . W is the energy of the oneparticle state.19).29) For H diagonal the commutator term vanishes as was to be expected. we will be mainly interested in vacuum expectation values of the evolution operator. created from the vacuum. θj+1 . 0)θ >< θΨ >= lim N −1 N →∞ dθ0 UN (θN = θ′ .30) 1 which shows that ∆t is eﬀectively modiﬁed to 2 sin( 2 W ∆t)/W (it is interesting to contrast this with the result we found for the harmonic oscillator in section 6 and problem 10).10). (15. what is to be expected from 2 1 a relation between the Lagrangian and Hamiltonian (the term 2 W is of course irrelevant) . θi H(ti. θj ) h j=0 ¯ ∆t = lim N →∞ ˜ (θj+1 + θj ) i∆t N −1 iθj (θj+1 − θj ) ˜ ˜ + θj W (tj ) dθj dθj exp h j=0 ¯ ∆t 2 T 0 ≡ DΨ(t)DΨ(t) exp i dt Ψ† (t)(i∂t + W (t))Ψ(t) . we see that W Ψ† Ψ = W bb† = −H + 1 W . In that case. For the present case we easily ﬁnd the vacuum wave function to be < θ0 >= 1 . (15. In complete analogy with eq. reinstating the dependence on Planck’s constant. Since we identiﬁed θ also with b† . (15. but in absence of this time dependence.32) ˜ Here we have replaced θ with Ψ† (which in this case agrees with Ψ. θi+1 ). we can write < θ′ Texp(−i = lim N →∞ ˜ dθj dθj exp N −1 j=0 H(t)dt/¯ )θ > h ˜ dθ0 ˜ dθ0 N −1 j=1 N −1 j=1 ˜ i∆t N −1 iθj (θj+1 − θj ) ˜ ˜ dθj dθj exp − θj H(tj . As for scalar ﬁeld theories. since in one time and no space dimensions γ0 ≡ 1) and θ with Ψ. θ0 = θ)Ψ(θ0 ) ˜ θj [θj − θj+1 − iH(tj . θi+1 . (15. (15.33) 61 . θj )∆t] Ψ(θ0 ) . We have indicated the general case where W can depend on time. With the explicit expression for H (see eq. one has 2 i 1 exp(−iH∆t) = 12 − i H − 2 W tan( 4 W ∆t)12 2 sin( 1 W ∆t) 2 W . θi+2 . (6. This does not mean that there are no discretisation errors in the fermionic path integral when H is diagonal. < 0θ >= θ .31) = lim N →∞ j=0 where as usual one has ∆t = T /N. In the presence of a source term this will allow us to derive all required matrix elements. the creation operator for the oneparticle state.and evaluate the commutator that appears in the exponent. (15. as can be seen from eq. (15. assuming for simplicity that H = − 1 W σ3 . θj+1 .
38) . It is obvious how this can be generalised to include the quantum mechanical description of a moving spin onehalf particle in a onedimensional potential V (x). For further details see the lectures by L. θΨ >≡ Ψ(x. As promised. θ >= lim h N →∞ dpj dxj ˜ dθj dθj × 2π¯ h (15. Their order in the above equation is therefore important. R. when used in further calculations. van Holten. and performing the Fourier transformation back to coordinate space. ed. Each of which can be described by its own θ. the path integral is easily found to be N −1 j=1 < x′ . However. ¯ d4 x′ J (x′ )GF (x′ − x) . Up to now we have described a spin onehalf particle pinneddown at a ﬁxed position. Using as a basis the plane waves constructed there.34) The order of the diﬀerentials is important here. d† (k) and d† (k). b† (k). (15. we do not require the precise form of the vacuum wave function(al) for performing perturbation theory. (15. it is left as an exercise to show that the path integral for fermions is given by DΨ(x)DΨ(x) exp i h ¯ ¯ d4 x Ψ(x)(iγ µ ∂µ − m)Ψ(x) −J (x)Ψ(x) − Ψ(x)J (x) . Les Houches 1975. A careful derivation of this formula and a step by step comparison with the matrix representation in the spin degrees of freedom can be found in sections 1 to 3 of the paper “Fermionic coordinates and supersymmetry in quantum mechanics”. it is as simple as for scalar ﬁeld theories to calculate the dependence of this path integral on the sources. provided we keep track of the order of the Grassmann odd variables. described by b† (k). ZinnJustin.37) ¯ Since the ﬁelds Ψ(x) and Ψ(x) are Grassmann variables.36) ˜ p2 iθj (θj+1 − θj ) ˜ (θj+1 + θj ) i∆t N −1 pj (xj+1 − xj ) − j − V (xj ) + + θj W (xj ) exp h j=0 ¯ ∆t 2m ∆t 2 ≡ Dx(t)DΨ(t)DΨ(t) exp i h ¯ T 0 dt 1 2 mx2 (t) − V (x(t)) + Ψ† (t)[i∂t + W (x(t))]Ψ(t) ˙ .such that < 0U(T )0 >= dθ′ < 0θ′ > (15. Balian and J. Faddeev in “Methods in ﬁeld theory”. also the sources J (x) and J (x) are Grassmann variables. Associating the 1 2 1 2 ˜ annihilation operators with their respective θ. Salomonson and J.35) dθ < θ′ U(T )θ >< θ0 >= dθdθ′ θ′ < θ′ U(T )θ > . If also W depends on the particle position (W (x)). It is now also straightforward to derive the path integral for the Dirac Hamiltonian of the previous section. the Hamiltonian becomes H= p2 ˆ x + V (ˆ) 12 − 1 W (ˆ)σ3 x 2 2m ˜ dp0 dθ0 2π¯ h . If we write < x. Nuclear Physics B196 (1982) 509.W. 62 (15. θ′  exp(−iHT /¯ )x. by P. θ) ≡ a0 (x) + θa1 (x). Since Grassmann variables form a complex linear space we can perform all calculations as in the scalar case. the Hamiltonian becomes a decoupled sum (in a ﬁnite volume) for each k of four fermions. In particular we can make the replacement Ψ(x) → Ψ(x) + Ψ(x) → Ψ(x) + d4 x′ GF (x − x′ )J (x′ ) . as in the case of scalar ﬁeld theories.
For example. ¯ . σ(x)) = gΨ(x)Ψ(x)σ(x) . to make sense of eq.43) By declaring the derivative to be a linear function on the Grassmann algebra. Derivatives of Grassmann variables are simply deﬁned as one would intuitively expect d 1=0 . (15. J) . J . σ) − J(x)σ(x) − J (x)Ψ(x) − Ψ(x)J (x) . the Lagrangian for a fermionic and a scalar ﬁeld is given by ¯ L = L2 − V (Ψ. and agrees with what can be derived to second order within Hamiltonian perturbation theory (see problem 24).44) dθ An example of an interaction between the fermions and a scalar ﬁeld σ is given by the socalled Yukawa interaction V (Ψ(x).9). ) Z2 (J . Both project on the coeﬃcient in front of the Grassmann variable. dθ d ′ θ =0 . where the order of the fermion ﬁelds is important.6) ¯ Z(J . J) = exp −i d4 x V ( d4 xd4 y 1 2 (15.40) ¯ J(x)G(x − y)J(y) + J (x)GF (x − y)J (y) iδ iδ −iδ ¯ . ¯ Again. The vanishing of the integral over a total derivative and of the derivative of an integral is in that perspective trivial. a property denoted by the sign or grading s(= ±1). (14. J ) . Interactions are taken into account by adding higher order terms to the Lagrangian. (15.41) Note the minus sign for the derivative with respect to J (x). Ψ. as for commuting variables. 1 1 L2 = Ψ(x)(iγ µ ∂µ − m)Ψ(x) + 2 ∂µ σ(x)∂ µ σ(x) − 2 M 2 σ 2 (x) . (15. We ﬁnd as in eq. J . (15. dθ (15. this result holds to arbitrary order in the sources.39) ¯ ≡ < 0UJ =J =0 (T )0 > Z2 (J . J. dθ d θ=1 . The integration measure. (15.21) we note that apparently Grassmann integration and diﬀerentiation is one and the same thing. J .41).where GF (x) is the Green’s function deﬁned in eq. (8. Ψ(x). one easily shows the following identity to hold d exp(θx) = x . δJ (x) δ J (x) δJ(x) . it can be uniquely extended to this algebra from the above set of rules. which is because the source stands behind the ﬁeld component Ψ(x). gn ) = exp −i ¯ Z2 (J . d d d (f g) = sf f g + f g dθ dθ dθ . is invariant under a shift by a constant Grassmann variable.45) . More importantly. Note that these rules imply that the Grassmann integral over a total Grassmann derivative vanishes. Comparing with eq.42) together with a generalised Leibnitz rule for functions f and g that are either even or odd Grassmann variables. such that we obtain (as for the scalar case we normalise the path integral to 1 for vanishing sources) < 0UJ J (T )0 > = < 0UJ =J =0 (T )0 > exp −i ¯ ¯ ¯ d4 xd4 y J (x)GF (x − y)J (y) (15. 63 (15.
it is cumbersome to determine the overall sign of a diagram. (15.10)). However. a fermionic line either forms a loop or it goes from a ¯ source J to a source J . The trace is with respect to the spinor indices which are not displayed explicitly to keep the notation simple. Ψ(x). For this we can take the minimal coupling in eq. Aµ (x)) = −eΨ(x)γ µ Ψ(x)Aµ (x) . For each vertex we have only indicated the fermionic part Ψ(xk )Ψ(xk ). This is because the Lagrangian is Grassmann even. because higher order terms will generally not be renormalisable (except in one space and one time dimension). Diagrams that involve ¯ fermions necessarily have as many lines ending in a source J as in a source J . J . for many of the theories we discuss. all we need is the relative sign of the various diagrams that contribute to the Green’s function with a ﬁxed number of sources. a requirement that can be related to the Lorentz invariance. (14. the scalar or vector ﬁeld contributions are not indicated. In section 19 and problem 30 we will discuss the fourFermi interaction. If one diagram can be obtained from the other by crossing two fermion lines. (15. It does not in general require the Lagrangian to be bilinear in Ψ and Ψ. J) is the sum over connected diagrams. this has consequences for the Feynman diagrams too. Intuitively this follows. Ψ(x)γ µ Ψ(x)Ψ(x)γµ Ψ(x).47) where {k} stands for the various orders in which the vertices are connected. which is clearly Lorentz invariant and Grassmann even. 7 × More accurately a fermion loop that connects vertices xk for k = 1 to n is associated to n k=1 n Ψ(xk )Ψ(xk ) → − {k} k=1 −iδ iδ ¯(xk ) δJ (xk ) δ J → Tr iGF (xk(1) − xk(2) )iGF (xk(2) − xk(3) ) · · · iGF (xk(n) − xk(1) ) . as for × × E E × × −→ $ $ z X − × $$ × $ × $ × ﬁg. ¯ As before. For the examples of eqs. as they are not relevant for the fermion loop. However. such that V (Ψ(x). (14. (15.12) (see also eq. log Z(J . as is indicated in the following ﬁgure by the dashed box.46).46) which will play a dominating role in describing Quantum Electrodynamics (QED).45) and (15. 6 It also implies that each loop formed by a fermion line carries a minus sign.48) . We used that Ψ(xk(j) )Ψ(xk(j+1) ) → −iδ iδ ¯(xk(j) ) δJ (xk(j+1) ) → iGF (xk(j) − xk(j+1) ) . this gives a relative minus sign. from the identity above $$ z X $$ $ × $ × ' 9 6 ' 1( E 0) E × −→ − ﬁg. δJ 64 (15.We can also consider the interaction of the fermions with the electromagnetic ﬁeld. As changing the order of fermionic ﬁelds and sources gives a sign change. whose quantisation will be undertaken in the next section. Fortunately. since the overall sign drops out in our computations of cross sections and decay rates. If no higher order fermionic interactions occur. the Lagrangian is bilinear in the fermionic ﬁelds.
¯ ¯ where an extra minus sign arises since in log Z2 (J . (7. 42 and eq.25)). (15. (15.39)) J comes ﬁrst.17) and the orthogonality relations of eq. which was in the ﬁrst place the reason for the extra minus sign in eq. (15. to derivatives with respect to the sources. (15.51) and through conjugation we get † bα (k) = d3 x (2π)3 ˆ eik·x Ψ† (x) u(α) (k) 2k0 (k) . Note that the factors of i. we need to determine the wavefunction factors to be inserted for the external lines that correspond to the in. J ) (see eq. d† (k) = α v (α) (k)† 2k0 (k) d3 x (2π)3 ˆ eik·x Ψ(x) (15. as in the scalar case. (13. · · · . (15. That minus sign is. (15. (13. before this derivative can be taken. eq. (kn . c2 (k) ≡ b2 (k) . c3 (k) ≡ d1 (k) . But the extra minus sign in the derivative with respect to the fermionic source J (see eq. and has to be anticommuted with δ/δJ . such that their insertion in the operator formulation can be converted in the path integral.41) to appear. · · · . are absorbed in the vertices for the Feynman rules in the table on page 35. cross sections and decay rates (see pg. which is described by the Lagrangian 1 L = 1 (∂µ ϕ)2 − 1 m2 ϕ2 + 1 (∂µ σ)2 − 1 M 2 σ 2 − 2 ϕ2 σ − Jσ − jϕ. (p2 . αℓ )(k1. The overall minus sign comes from the term that closes the loop ¯ ¯ Ψ(xk(1) )AΨ(xk(n) ) = −AΨ(xk(n) )Ψ(xk(1) ) . however. associated to the derivatives with respect to the sources. α2 ). β1 ).and outgoing fermion lines. α1 ).14) one ﬁnds (cmp. with a scalar loop for the ﬁeld ϕ that arises in the theory discussed in section 11. (9. which is why the propagator in that table equals −i times the Green’s function.48) due to the anticommuting nature of the fermionic variables. (k2. (15.26)) bα (k) = u(α) (k)† 2k0 (k) d3 x (2π)3 ˆ e−ik·x Ψ(x) .53) β β β where in a shorthand notation we separate particles from antiparicles by the helicity index c1 (k) ≡ b1 (k) . Tin )c† 1 (k1 )c† 2 (k2 ) · · · c† n (kn )0 > . Only the overall minus sign required in closing a fermion loop remains as an extra factor.54) .41)) is not absorbed in the vertex in order to guarantee that vertex factors are assigned as in the scalar theory. dα (k) = d3 x (2π)3 ˆ e−ik·x Ψ† (x) v (α) (k) 2k0 (k) . This is also the Feynman rule for the fermion propagator. One ﬁnds (k(n + 1) ≡ k(1)) 2 2 2 2 n k=1 n ϕ(xk )ϕ(xk ) → k=1 iδ iδ δJ(xk ) δJ(xk ) n → {k} j=1 iG(xk(j) − xk(j+1) ) . (15. (pℓ . Before we convert these Feynman rules to the ones involved in computing the scattering matrix.52) In the Hamiltonian formulation the scattering matrix is given by out< (p1 . (15. as an example. 65 c4 (k) ≡ d2 (k) . β2 ).49) where A is Grassmann even. We contrast this. Using eq. absorbed in eq. For this we express the creation and annihilation operators in terms of the fermionic ﬁelds (at t = 0).50) which completes the demonstration of the extra minus sign for fermion loops. βn ) >in = < 0cα1 (p1 )cα2 (p2 ) · · · cαℓ (pℓ )U(Tout .
(k2 . t = Tin ) iδ = d3 x e−ik·x ¯ δ Ja (x. (kn .p δα. 69) by dropping the wavefunction factors.J )J =J =0 ˆβ ¯ ¯ . ˜ (k. (p2 . With these deﬁnitions the twopoint function becomes ¯ G(2) (J . ˜a (k. b &% (The 1P I diagram equals Σab (p). the full propagator near the poles is of the form of the free propagator with a wavefunction renormalisation factor ZF and a renormalised mass m.60) . (15. α)(k. such that on the massshell one has (cmp. which points to the ﬁrst spinor index (here a).55) a 2k0 (k)(2π)3 −iδ = δJa (x. J ) c = ≡ '$ E × E × + × E× + '$ &% E d d dd × d E× . As long as we don’t break the Lorentz invariance. J ) = −iZF c ˜ ¯ d4 p J (−p) 66 1 ˜ J (p) . t = T ) ¯ 2k0 (k) δJ a out . β1 ). · · · . / − m + iε p ˜ (15. d d d d d &% '$ '$ E E E × ×+··· &% &% (15.58) a . (15. α2 ). β2 ). t = Tout ) d3 x e −ik·x (α) va (k)∗ u(α) (k)∗ a iδ . which is the full propagator in the momentum representation for the conventions on pg.56) ˆ The cα (k) are of course deﬁned in terms of ˆα (k) and dα (k) as in eq. the insertion of a ﬁeld operator at the appropriate time is in the path integral represented by a derivative with respect to the source ˆ† (k) = bα ˆ† dα (k) = ˆα (k) = b ˆ dα (k) = such that out< ℓ n γ 0 u(α) (k) a 2k0 (k)(2π)3 (α) va (k)∗ d3 x e ik·x −iδ = δJa (x.54). ˜ (−k. (pℓ . (9. αℓ )(k1 .59) ab with between square brackets the inverse of the 4×4 matrix (pµ γ µ −m−Σ(p)+iε)ab . · · · .Like in eq. β) >in = e−ip0 (p)T δk. ¯ G(2) (J . t = Tout ) γ 0 v (α) (k) −iδ (p1 .β ).) The convention for these sort of diagrams is that momentum ﬂows in the direction of the arrow. ˜ 2k0 (k) δ Ja (−k.57) where the selfenergy is now a 4 × 4 matrix given by (−i×) the amputated 1P I 2point function '$ E E iΣab (p) ≡ (15. (9. eq. t = T ) ¯ 2k0 (k) δJ a in iδ . t = Tin ) 2k0(k) δ J a 2k0 (k)(2π)3 u(α) (k)∗ a 2k0 (k)(2π)3 γ 0 v (α) (k) a iδ d3 x eik·x ¯ = δ Ja (x. t = Tout ) (15. βn ) >in = i=1 cαi (pi ) ˆ j=1 c† j (kj ) exp(GJ . 42. t = Tin ) γ 0 u(α) (k) −iδ . We continue ˆ b as in section 9 by ﬁrst ﬁxing the wavefunction and mass renormalisations in terms of the connected twopoint function (we leave it as an exercise to show that in the absence of interactions out< (p.12)) ˜ ¯ G(2) (J . J ) = −i c ˜ ¯ d4 p J a (−p) 1 / − m − Σ(p) + iε p ˜ Jb (p) . (15.3). when evaluated with the Feynman rules for the reduced matrix elements of table 4 (pg. α1 ).
k2. t) ··· −p − m − Σ(−p) + iε ab / √ If we deﬁne as usual p0 (p) = p 2 + m2 and for convenience of notation change p to −p in ˜ eq. t = T ) ¯ out 2 + m2 δ J a k ˜ a 2ZF ˜ ¯ δJ a (k. · · ·)b··· −ip0 (p)Tout ··· e 3 2p0 (p) 2p0 (p)(2π) 67 .55) and (15. t = Tout ) k 2 + m2 a ˜ −iδ . with the cα (k) deﬁned as in eq. computing ˆ± ˆ the action of ˆα (p) and dα (p) we can restrict our attention to b− + A(· · ·)··· ≡ ··· dp √4 2π ˜ ¯ d4 p J a (p) −i −p − m − Σ(−p) + iε / (amp) (p. ˆα (k) = b− u(α) (k)∗ a 2ZF iδ . since the number of J and J sources has to be equal (we ignore for the moment any other bosonic ﬁelds that might be present.61) and. t = Tin ) k 2 + m2 ˜ ˆ . any diagram can be generated from a given one by permuting fermion lines).b22. (15. with the diﬀerence that there has ¯ to be an even number of external lines. (p2 . (15. αℓ )(k1. t = Tin ) iδ −iδ .54). · · ·)b··· = Gc ··· . (15. · · · .···.19). we ﬁnd ˆα (p)A(· · ·)··· = b− ··· −i u(α) (p)∗ a 4πp0 (p)ZF / (α) ∗ (p F ua (p) dp0 e−ip0 Tout / − m − Σ(p) + iε p (amp) (−p.14)). k1 . · · ·)b··· = Gc ··· −i ˜ ¯ J cj (pj ) −pj − m − Σ(−pj ) + iε / j=1 cj bj aj dj ab −ieip0 t (amp) ˜ ¯ (p. β2 ). p2 . ˆ± out< (p1 . βn ) >in ℓ n = i=1 cαi (pi ) ˆ− j=1 c+j (kj ) exp(GJ . · · ·)b··· .64) Gc dt J a (p. (pℓ . As in eqs. (15. ˜ (−k.15). Like in eq. (9.···. kn )b1 . including those in the external lines will be obvious). · · · . · · · . (9. The amputated npoint function will now carry the spinor index of each of the external lines and one has ¯ G(2n) (J . α2 ).bnn × d 4 pj d 4 k j Gc a1 . (k2. pn . (9.20) and (9.J )J =J =0 ˆ ¯ ¯ β . (913) and (9.63) . (kn . Relative signs of the diagrams are determined by the rules that were described above (it is not diﬃcult to convince oneself that with respect to the fermion lines.a .62) To compute the wavefunction factors for the external lines we express the npoint function in terms of amputated npoint functions. α1 ). dα (k) = − γ 0 v (α) (k) 2ZF ˜ δ J (−k. · · · . we have to modify eqs.21) we can compute the action of cα (k). eqs. from which the wavefunction factors will be obtained.64). β1 ). (15. Note that we have assumed one particular ordering for the sources.a −i kj − m − Σ(kj ) + iε / ˜ Jdj (kj ) (15. (15.Performing the wavefunction renormalisation. J ) ≡ c n n j=1 (amp) (p1 .65) ab (2π)4 Z (amp) + m)ab Gc ˜ (−p. ˆα (k) = b+ γ 0 u(α) (k) a ˜ 2ZF k 2 + m2 ˆ dα (k) = + (α) va (k)∗ ˜ δ Ja (k. as in eq.56) accordingly (cmp. (15.
t) . (15. we restrict our attention to bα − ··· B(· · ·)··· ≡ (amp) ··· (p. · · ·)b··· dt Gc (amp) dp √4 2π One ﬁnds ˆα (p)B(· · ·)··· = b+ ··· −i −i −i / − m − Σ(p) + iε p ip0 t ba ˜ Ja (p) = ˜ Ja (p. along the arrow (see the table below).70) 68 . By similar arguments we obtain from eq. (15.67) √ The wavefunction factor for an outgoing electron is therefore given by u(α) (p)b ZF . (15. eq. like for the scalar case. · · ·)b··· a ba 4πp0 (p)ZF = (2π)4 Z F (amp) ··· (p. · · ·)b··· ip0 (p)Tin + m)ab Gc ˜ ··· e = 2p0 (p) 2p0 (p)(2π)3 F iv (α) (p)b (2π)4 Z (amp) (p.60). such that / (p u(α) (p)∗ a + m)ab ˜ = 2p0 (p) (p + m)† (α) / ˜ u (p) 2p0 (p) ∗ = b (γ 0/γ 0 + m) (α) p ˜ u (p) 2p0 (p) ∗ = u(α) (p)b .66) Consequently. · · ·)b··· −ip0 (p)Tout ··· ˆα (p)A(· · ·)··· = −iu(α) (p) (2π)4 ZF Gc e b− ··· b 3 2p0 (p)(2π) . (9.69) −ie / − m − Σ(p) + iε p eip0 Tin / − m − Σ(p) + iε p ba dp0 Gc ··· (p. γ 0 u(α) (p) (15. · · ·)b··· Gc 2p0 (p)(2π)3 ub (p)eip0 (p)Tin (α) (2π)4 Z F . · · ·)b··· Gc ··· 2p0 (p)(2π)3 eip0 (p)Tin . · · ·)b··· = Gc ··· (2π)4 Z / (α) ∗ (−p F va (p) (amp) (p.which is obtained by deforming the p0 integration contour to the lower halfplane (since Tout → ∞). taking into account eq. In this case the momentum ﬂows against the arrow of the fermion line. but does ﬂow into the diagram as is required in the convention of the reduced matrix element. The convention is that the momentum ﬂows out of the diagram. (15. This means that in the reduced matrix element M the outgoing electron momenta occur precisely as indicated in eq.64) ˆα d+ (p)A(· · ·)··· = ··· −i (α) va (p)∗ 4πp0 (p)ZF dp0 eip0 Tin −p − m − Σ(−p) + iε / ab (amp) (p. which is most easily / ˜ 2 2 found using 1/(p − m + iε) = (p + m)/(p − m + iε). (13. · · ·)b··· (p + m)ba 0 (α) Gc / ˜ γ u (p) eip0 (p)Tin = a 3 2p0 (p) 2p0 (p)(2π) (amp) ··· (p. (15. Its residue is proportional to the matrix ZF (p + m)/2p0 (p). b (15.(15. this is why the amputated npoint function has −p as its argument. and computing the contribution from the pole at p0 = p0 (p).18). · · ·)b··· d 4 p Gc (amp) ··· (p.68) √ such that the wavefunction factor for an incoming antiparticle (positron) is −v (α) (p)b ZF .2)). ˆ To compute the action of ˆ+ (p) and dα (p).65) becomes (amp) (−p. where we deﬁned for the amputated npoint function all momenta to ﬂow into the diagram. We can now use the fact that the / ˜ / ˜ ˜ spinor u(α) (p) satisﬁes the equations of motion (see eq.
whereas the wavefunction factor of an outcoming antiparticle √ (α) (or positron) is given by −vb (p) ZF . Signs from fermion loops and exchanges of external lines will not be implicit in diagrams.11) (resp. and the outgoing momenta ﬂow out of the diagram.18). kE ¨ a ' © α. (15. Consequently. In the table below we summarise the Feynman rules that correspond to the fermionic pieces in computing the reduced matrix elements. This guarantees that eqs. The √ (α) wavefunction factor for an incoming electron is hence ub (p) ZF with the momentum ﬂowing along the fermionic arrow. We have chosen the convention that the incoming momenta ﬂow in. where the momentum ﬂows opposite to the fermionic arrow. a ZF v (α) (k)a . The minus signs in front of some of the wavefunction factors are irrelevant (they can be absorbed in the overall sign ambiguity).and (again for convenience changing p to −p) ˆ dα (p)B(· · ·)··· = − ··· −i dp0 Gc (amp) ··· (−p. · · ·)b··· Gc 2p0 (p)(2π)3 vb (p)e−ip0 (p)Tout (α) i (2π)4 ZF .12) and (11.and outgoing antiparticles (positrons) should be reversed. (α) ZF va (k) . · · ·)b··· (−p + m)ba 0 (α) Gc γ v (p) e−ip0 (p)Tin = a 3 2p0 (p) 2p0 (p)(2π) (amp) ··· (−p. k E a' α. as only relative signs are known. · · ·)b··· e−ip0 Tout −p − m − Σ(−p) + iε / γ 0 v (α) (p) ba a 4πp0 (p)ZF = (2π)4 Z F (amp) ··· / ˜ (−p. ZF u(α) (k)a . cross section and decay rate). table 4 T ¨r bB r a ≡ gδab and k1 = k2 + k3 ¨j k2 k1 µ £ k3 ¢ £¡ E k3 Yukawa vertex µ ¢ T ¨¡ bB r a ≡ −eγab and k1 = k2 + k3 ¨j r k1 k2 photon vertex fermion propagator b k a a E α. (9. For conventions where momenta always ﬂow in the direction of the fermion arrow the four momentum for wavefunction factors associated to in. (10. the scattering matrix. k E ¨ a E © α. kE ≡ ≡ ≡ ≡ ≡ 1 k−m+iε ab / √ √ √ √ ZF u(α) (k) . remain valid in the presence of fermions. all fermion momenta in the table ﬂow from left to right. d4 k (2π)4 k0 = k0 = k0 = k0 = k 2 + m2 ˜ k 2 + m2 ˜ k 2 + m2 ˜ k 2 + m2 ˜ incoming electron incoming positron outgoing electron outgoing positron for each fermion loop interchange of fermion lines −1 × i −1 69 . (15.71) ¯ In both cases there is an extra minus sign from pulling δ/δJ through J in eq.63).
(16. (16. (16. the above expression forms a Lorentz tensor and its onshell extension to an arbitrary frame is therefore unique. but also (λ) k µ εµ (k) = 0 . In problem 12 we already saw that its Lagrangian is given by LA = − 1 Fµν F µν + 1 m2 Aµ Aµ − Aµ (x)J µ (x) . it is advantageous to decompose the spin with respect to the direction of the particle’s motion. 2.3) (16. this implies the onshell condition k0 = k 2 + m2 . The index λ enumerates the various solutions for ﬁxed momentum. Three independent components remain. (16. (16. Since the spin wave functions are by construction Lorentz vectors. It is easily seen to project arbitrary vectors wµ on to vectors that satisfy k µ wµ = 0. We may for example choose (λ) λ εµ (0) = δµ (λ = 1.5) in the restframe of the particle. They satisfy the Lorentz invariant normalisation (λ) g µν εµ (k)∗ ε(λ ) (k) = −δ λλ ν ′ ′ . exactly what would be required for a particle with spin one. expected to describe a massive spin one particle. The propagator for the massive spin one ﬁeld was already computed in problem 12 £ £ £ £ ¢¡¢¡¢¡¢¡ k µ ν = − gµν − k2 − kµ kν m2 m2 + iε = k2 Λµν (k) − m2 + iε . the photon ﬁeld) to be discussed below. J µ = 0) are equivalent to ∂µ Aµ (x) = 0 and (∂µ ∂ µ + m2 )Aν (x) = 0 . Aµ (x) = d3 k 2k0 (k)(2π)3 λ † (λ) (λ) aλ (k)εµ (k)e−ikx + aλ (k)εµ (k)∗ eikx .1) (λ) where εµ (k)e−ikx are independent plane wave solutions of the equations of motion. 2 As usual. 4 2 and that the free equations of motion (i.6) The minus sign is just a consequence of the fact that in our conventions gij = −δij .16 Feynman rules for vector ﬁelds As before. which is extended to an arbitrary frame by applying a Lorentz boost. We will ﬁrst discuss the simpler case of a massive vector ﬁeld. (16. the quantisation for vector ﬁelds starts by expanding the ﬁeld in plane waves and identifying the Fourier coeﬃcients with creation and annihilation operators.4) It removes one of the four degrees of freedom of a fourvector.2) (16. 3) . They will be identiﬁed with the spin components or helicity eigenstates of the vector.e.8) Especially for the massless spin one ﬁeld (i.7) most easily proven in the restframe. The spin wave functions also satisfy a completeness relation which is given by Λµν (k) ≡ (λ) (λ) εµ (k)εν (k)∗ = − gµν − λ kµ kν m2 .e. 70 .
which are circumvented if we take ∂µ J µ (x) = 0. (16. ε(±) (k) = µ 1 2 √ 2 ε(1) (k) ± i¯(2) (k) ¯µ εµ . (16. We have helicities 0 and ±1. ε(1) (k) and ε(2) (k) form a complete set of real orthogonal fourvectors. Writing 1 ˜ ˜ Aµ (k) ≡ σ (k)kµ + m (λ) ϕλ (k)εµ (k) . 2.(2.11) which can be expressed in terms of three scalar ﬁelds ϕλ . as in eqs.7). (16. described by the spin wave functions ε(0) (k) µ k −k0 k = . Nevertheless. This would lead to serious inconsistencies. λ = 1. −} in this case. ϕλ .14) where we have added a source for each scalar ﬁeld. ˜ λ 1 ˜ j(k)kµ + Jµ (k) ≡ ˜ m (λ) ˜ J (λ) (k)εµ (k) . ε(0) (k).13) The corresponding Lagrangian would be given by Lϕ = 1 2 λ (∂µ ϕλ (x))2 − 1 m2 ϕ2 (x) − ϕλ (x)J (λ) (x) λ 2 .which are called helicity eigenstates. ¯µ ¯µ µ These new spin wave functions satisfy the same properties as in eqs.6) and (16. m mk .10) Sums over λ will of course run over the set {0. such that we can put σ ≡ 0. where they satisfy Λµν (k) ≡ (λ) (λ) εµ (k)εν (k)∗ = − gµν − λ kµ kν k2 . (16. We introduce also a ﬁeld σ for the component of the vector ﬁeld along kµ . (16. It is important to realise that it is the Lorentz invariance that requires us to describe a spin one particle by a fourvector. if we treat λ as a threevector index (in some internal space) and demand the interactions to be O(3) invariant with respect to this index (i. µ µ The Hamiltonian for the massive spin one particles is given by H= 1 d3 k 2 k0 (k) a† (k)aλ (k) + aλ (k)a† (k) λ λ λ . (16. λ 2 (16. as H= d3 k 1 2 ˜λ (k)2 + 1 (k 2 + m2 )ϕλ (k)2 = π ˜ 2 d3 x 1 2 2 1 πλ (x) + 2 (∂i ϕλ (x))2 + 1 m2 ϕ2 (x) .6) and (2. From the point of view of the scalar degrees of freedom. then we claim that the resulting 71 . +. this invariance seems to be lost when we introduce interactions in the Lagrangian of eq.e.9) where kµ . ˜ πλ (k) = ˜ . 3. † ϕλ (k) = aλ (k) + aλ (−k) / 2k0 (k) . (16.7) and are also deﬁned oﬀshell. (16. We see that σ decouples from the other components and behaves like a scalar particle with the wrong sign for the kinetic term.14).12) i 2k0 (k) a† (k) − aλ (−k) λ 2 where we deﬁned. invariance under rigid rotations and reﬂections in the internal space). We leave it as an exercise to verify that rotations over an angle α around the axis pointing in the direction of k leaves (0) εµ (k) invariant and transforms ε(±) (k) to e±iα ε(±) (k).15) λ 1 a simple calculation shows that LA ≡ Lϕ − 2 (∂µ σ(x))2 − σ(x)j(x).
pn )ν1 . It is proportional to Λµν (k) (to guarantee that the scalar ﬁeld σ introduced above decouples from the other ﬁelds. problem 8). Eqs. the propagator in eq. For a massless spin one ﬁeld (the photon) we would expect the helicity zero component of the vector ﬁeld to be absent. however. when reexpressed in terms of J (λ) . If none of the vertices or sources couple to the σ ﬁeld. k (16. t = Tout ) (16. is a Lorentz scalar as far as it concerns the dependence on A. (16. see e.interactions do respect the Lorentz invariance. we might just as well replace it by −gµν . p2 .10)). t = Tin ) Σµν (p) ≡ Λµν (p)ΣA (p) . (16.19) a† (k) = ˆλ (λ) 2k0 (k)εµ (k) −iδ . what we would mean n 72 . written in terms of the vector ﬁeld Aµ (x).10) and (16.ν2 . section 323 in Itzykson and Zuber. G(n) (Jµ ) ≡ d4 pj J µj (pj ) 2 c pj − m2 − ΣA (pj ) + iε c j=1 (16. (16. Using DAµ (x)Dλ(x) exp −i d4 x λ(x)∂µ Aµ (x) = ˜ D Aµ (k) ˜ δ(kµ Aµ (k)) . using the relations (λ) aλ (k) = − 2k0 (k)εµ (k)∗ (λ) a† (k) = − 2k0 (k)εµ (k) λ d3 x (2π)3 d3 x (2π)3 ˆ Aµ (x)e−ik·x ˆ Aµ (x)eik·x . First we have to redeﬁne. There are a number of other ways to eliminate the σ degree of freedom. We can consequently deﬁne ˜ δ Jµ (k.···. . aλ (k) = ˆ (λ) 2k0 (k)εµ (k)∗ −iδ The npoint Green’s functions can now be written in terms of amputated Green’s functions that carry fourvector indices for each external spin one line. Like for scalar and fermion ﬁelds there will be a mass and wavefunction (denoted by ZA ) renormalisation. −iΛµj νj (pj ) ˜ Gamp (p1 .3).20) ν (λ) (λ) Using the fact that onshell Λµ (k)εν (k) = −εµ (k).16) we see that the socalled Lagrange multiplier ﬁeld λ(x) plays the role of removing the unwanted degree of freedom. This can be achieved by adding to eq. For the computation of the scattering matrix we express the annihilation and creation operators in terms of the vector ﬁelds at t = 0. (16.17) which in the path integral turn into . but also oﬀshell. determined through the selfenergy of the vector ﬁeld. The reason is simple.18) which is identical to eq. Since we modiﬁed the theory oﬀshell.15) guarantee that such a pair. which is now a Lorentz tensor of rank two. (9. ˜ δ Jµ (−k.8) has to be changed also. (16.g. alternatively it can be seen as a consequence of the O(3) invariance with respect to the index λ). because the O(3) invariance requires that the λ index is always pairwise contracted. by replacing Λ(k) by its oﬀshell value Λ(k) (eq. To eliminate the σ ﬁeld we have to enforce ∂µ Aµ (x) = 0. · · · . which is suﬃcient if the Lagrangian is Lorentz invariant when treating λ as a dummy label. not only onshell.νn .2) a term −λ(x)∂µ Aµ (x) (cmp. (16. We will come back to massive vector particles in section 19. we ﬁnd for the incoming spin one line √ √ (λ) (λ) a wavefunction factor ZA εµ (k) and for the outgoing line a factor ZA εµ (k)∗ .
In other words. it is the great advantage of the path integral formulation that calculations can be performed in a gauge in which the Lorentz invariance is manifest. except that now only two helicity states can appear. µ (16. because eq. The extra degree of freedom is removed by the gauge invariance of the massless vector ﬁeld. We take as our deﬁnition ε(0) (k) = nµ (k) ≡ µ 1 2 √ with k · s± (k) = 0. That the gauge invariance must be crucial here is clear. on which physical observables like cross sections and decay rates should not depend.e. such that it will not give rise to a renormalisation of the mass. It is in general not true anymore that the selfenergy is proportional to Λ(α) (k). To perform the quantisation of the theory. possibly inﬁnite. In section 20 it will be shown how in principle in any gauge the path integral can be deﬁned and that the result is independent of the chosen gauge. the gauge ﬁxing parameter α will in principle have to be renormalised too. However.26) £ £ £ £ ¢¡ ¢¡ ¢¡¢¡ k 2 1. at k = 0). k . It can be shown that this in general implies any conserved current. The propagator is read oﬀ from eq. (16. s± (k)) . k . but the gauge invariance does guarantee that Σµν (k)˜ν (k) is independent of α for j µν ˜µ (k) = 0. one can go about as in the massive case. (16. It is an important consequence of gauge invariance that unphysical degrees of freedom decouple in the physical amplitudes. It also implies that the selfenergy vanishes on shell (see problem 39). However. (4. µν ′ (16.with the zero helicity component. i. see eqs. Due to the presence of the fourvector n(k) it will be much more cumbersome to demonstrate the Lorentz invariance.9) is singular in the limit of zero mass. The photon remains massless. it is no longer true that k µ ε(0) (k) will vanish onshell µ 2 (i. kµ j Σµν (k) = Λ(α ) (k)Σ(k) . One still has for any value of ν α that Λ(α) (k)µ ε(±) (k) = −ε(±) (k). as was discussed at the end of section 4 and in problem 9. α′ . table 5 µ £ £ £ £ ¢¡ ¢¡ ¢¡¢¡ k µ £ £ £ £ ¢¡¢¡¢¡ λ. In the table below we summarise the Feynman rules.21) k µ ν = 1 k kν − gµν − (1 − α ) k2µ+iε k 2 + iε Λ(α) (k) µν ≡ 2 k + iε . ε(±) (k) = (0. The gauge most suitable for that purpose is of course the Lorentz gauge ∂µ Aµ (x) = 0.22).21) and (4.e. One could then choose a gauge that allows us to show the equivalence between the path integral and the canonical quantisation. k E ¨ £ µ £ £ ¢¡ ¢¡ £ © ¢¡E ν ≡ ≡ ≡ 1 − gµν −(1− α ) kµ kν k2 +iε k 2 +iε photon propagator (Lorentz gauge) incoming photon outgoing photon 73 √ √ (λ) ZA εµ (k) (λ) ZA εµ (k)∗ λ. These still form with kµ four independent fourvectors and ε(±) (k) are µ still transverse polarisations. This is easily seen to imply that onshell a0 (k) = 0. The wavefunction factors for external photon lines are ν µ therefore identical to the ones for the massive case. (4. We will come back to this point in section 19. as a massive photon would have one extra degree of freedom. onshell there is no longitudinal component for the photon.22) where α is an arbitrary parameter.23) for some. Apart from a wavefunction renormalisation (ZA ).
the equivalent of 1 σ for a two2 component spinor.3) The Feynman rules are collected in the table below.5) . (13. (17.).4) 1ˆ ⊘ k·σ 4k 2 (J is the spin part of the angular momentum operator. In the Lorentz gauge (see eqs. The covariant derivative Dµ is given as before (see eqs.8) were based on a decomposition along the zaxis in the restframe. (17. Ψf (x) → exp(−iqf Λ(x))Ψf (x) . k. It is hence deﬁned in terms of the eigenvalues of the operator 1ˆ ki k·σ ⊘ ˆ εijk σjk = 2 k·J ≡ .QED QED is the ﬁeld theory that describes the interaction between the photon and the charged fermions. relativistic invariant ﬁeld theory. by making use of the fact that in the Dirac representation / k= / k 0 12 −k · σ 74 k·σ −k0 12 . as for the photon. is deﬁned by decomposing the spin in the direction of motion. Helicity.2) For electrons we have qf = −e and for protons qf = e. protons. (17. say electrons to positrons. or in general particles to antiparticles.1) Here f is de socalled ﬂavour index. k ] = 0.21) and (4.22)) the Lagrangian is given by 1 L = − 4 Fµν (x)F µν (x) − 1 α(∂µ Aµ (x))2 + 2 f Ψf (x)(iγ µ Dµ − mf )Ψf (x) . (4. as well as parity (P ) and time reversal (T ) symmetry can be separately broken. The spin components of the solutions in eq. For α = 0 the Lagrangian is invariant under gauge transformations Aµ (x) → Aµ (x) + ∂µ Λ(x) .17 Quantum Electrodynamics . e. (17. (3. which is an important symmetry of the theory.g. which distinguishes the various types of fermions (electrons.35) and (14. etc. The latter relates. It. table 6 µ £ £ £ £ ¢¡ ¢¡ ¢¡¢¡ k k1 £¡ ¢ T ¨¡ bB r a ¨j r µ £ k3 ¢ E ν ≡ 1 − gµν −(1− α ) kµ kν k2 +iε k 2 +iε photon propagator (Lorentz gauge) photon vertex (fermion charge is q) fermion propagator k2 µ ≡ qγab and k1 = k2 + k3 b k a ≡ 1 k−m+iε ab / Before calculating cross sections we wish to discuss in more detail the helicity of the fermions and its relation to charge conjugation C. but the combination CPT is to be unbroken to allow for a local.) This holds both in the Dirac and Weyl representations.11)) by Dµ Ψf (x) = (∂µ + iqf Aµ (x))Ψf (x) . It is easy to ˆ verify that [k · J. (17.
(17. (17. Instead of the label α we can use ± to indicate the helicity and we have u± (k) = v ± (k) = (k + m) / (−k + m) / m + k0  m + k0  ˆ ˆ ˆ ˆ u± (k) . 3 . ˆ χ± (k) (17.e. 2 (17. (17. They can be related to each other by ˆ ˆ χ± (k) ≡ −iσ2 ϕ∗ (k) . which indeed interchanges positive and negative energy solutions. For k = (0.This implies that we can choose u0 and v0 ˆ (consequently they become functions of k) (α) (α) ˆ to be eigenstates of the helicity operator k · J (17. (12. which relates the solutions of the Dirac equation to solutions of the complex conjugate Dirac equation (see eq. k · σ χ± (k) = ∓χ± (k) . To formulate this symmetry in the fourcomponent spinor language one introduces the charge conjugation matrix (in the Dirac representation) C ≡ −iγ 0 γ 2 = which satisﬁes C −1 = C † = −C .10) is given by ∗ γ 2 γµ γ2 = −γµ . we ﬁnd ∗ ˆ ˆ ˆ ˆ k·σ −iσ2 ϕ± (k) = −ik · σ ∗ σ2 ϕ± (k) ∗ ˆ ˆ = iσ2 k · σ ϕ± (k) ∗ ˆ = ∓ −iσ2 ϕ∗ (k) ± . k · J v0 (k) = ∓ 1 v0 (k) .7) Note the ﬂip of helicity for the positron wave functions. when we use that ∗ σ2 σi σ2 = −σi . It is clear that we can deﬁne ˆ ϕ± (k) ˆ ˆ ˆ ˆ u± (k) = 0 . i = 1. The equivalent of eq. k · J u± (k) = ± 1 u± (k) . particles and antiparticles. i. 0 0 2 0 ± ˆ ± ˆ ± ˆ ˆ v0 (k) .12) . 0 0 0 ± ˆ ˆ ˆ ˆ v0 (k) = 0 .9) ± Indeed. (17.14) 75 . (17. 0.11) As eq. (17.8) with ϕ± and χ± each an orthonormal set of twocomponent spinors.8).9) relates the components of the electron wave function to those of the positron wave function. (17. k) these eigenstates coincide with the decomposition in eq.13) This can be proven from the explicit form of the Dirac matrices. (13. it is the basis of the charge conjugation symmetry.10) which expresses the fact that SU(2) is socalled pseudo real.6) 1 ˜ ˆ ˜ k · J Ψ(k) = ± 2 Ψ(k) . t Cγµ C −1 = −γµ ⊘ −iσ2 −iσ2 ⊘ . k · σ ϕ± (k) = ±ϕ± (k) . 2. (17.31)).
(17. (17. ¯± u± (k) = C v± (k) . (17.19) t In particular we see that the electromagnetic current generated by the fermi ﬁelds transforms as required for the interchange of particles and antiparticles. we ﬁnd d4 x CΨ (iγµ (∂ µ + ieAµ ) − m) CΨ t d4 x − Ψt −iγµ (∂ µ + ieAµ ) − m Ψ t t t = = d4 x − Ψt C −1 (iγµ (∂ µ + ieAµ ) − m) CΨ = d4 x Ψ (iγµ (∂ µ − ieAµ ) − m) Ψ . the anticommuting properties of the fermi ﬁelds and partial integration. ¯t (17. because charge conjugation is an involution. is most convincingly demonstrated by the fact that the Dirac Lagrangian is invariant under charge conjugation.It is now easy to verify that v± (k) = C ut (k) .15) We just need to prove one of these identities. Consider a fermion loop as in ﬁgure (8a) below ¢ '$ % ¡ £ £ £ £ ¢¡¢¡ ¢¡ ¢¡ (a) for which the Feynman rules lead to the expression (note that spinor index contractions run against the arrow of the fermion line and i ki = 0) q n Tr 1 1 1 γ µ1 γ µ2 γ µ3 · · · γ µn / − m + iε p / + k1 − m + iε p / / + k1 + k2 − m + iε p / / 76 . (17. i. (17. applying it twice gives the identity C CΨ We ﬁnd ∗ ˆ ˆ ϕ± (k) ϕ∗ (k) ± (−k + m) 2 / (k + m) / t γ 0 C ut (k) = Cγ0 u∗ (k) = iγ 2 u∗ (k) = iγ 2 ¯± 0 =i ± ± m + k0  m + k0  0 0 ∗ ˆ 0 ϕ (k) (−k + m) / (−k + m) / ⊘ iσ2 ± = 0 = v± (k) .18) That charge conjugation is really a symmetry of the quantum theory. t ∗ Using CΨ = Ψt γ0 C † γ0 = −Ψt C −1 . (17.20) An important consequence of the charge conjugation symmetry is Furry’s theorem. 8 .2) should change sign too.16) Under charge conjugation the charge that appears in the covariant derivative in eq. under which the charge changes sign C jµ = Ψγµ Ψ → −Ψγµ Ψ . (17. To show this we multiply the complex conjugate of the Dirac equation with iγ 2 iγ 2 [(−iγµ (∂ µ − ieAµ ) + m) Ψ]∗ = ∗ iγ 2 γµ γ2 (∂ µ + ieAµ ) + m (iγ 2 Ψ∗ ) t = (−iγµ (∂ µ + ieAµ ) + m) (CΨ ) . which states that a fermionic loop with an odd number of vertices will not contribute to the amplitude.21) p k1 &% kn E k2 £ ¢¡ £ (b) ¢¡ '$ B j £ £ £ £ ¢¡ ¢¡ ¢¡¢¡ p k1 &% kn ' k2 £ ¢¡ £ ﬁg.e.17) 0 = −iσ2 ⊘ m + k0 m + k0  χ± (k) ˆ 0 ∗ t t † t = Cγ0 C ∗ γ0 Ψ = Ψ .
s ¨ B 1 ﬁg. (−p2 . which is thus seen not to depend on the gauge ﬁxing parameter α. s1 ). We will now calculate the cross section for electronelectron scattering (the socalled Møller cross section). (k2 . we convert eq. such that us (p)γ µ ut (k)(pµ − kµ ) = us (p)[(p − m) − (k − m)]ut (k) = 0 . except for their opposite charge factors (±q)n .25) . t1 j £ p2 . It conﬁrms the intuition that particles and antiparticles contribute equally. The diﬀerential cross section for unpolarised electronelectron scattering is given by (see ¯ eq. (17.t2 M({(−p1 . t2 ) >in = δ4 (p1 + p2 − k1 − k2 )M({(−p1 . (k2 . 9 1 The labels ti and si indicate the helicities of the incoming and outgoing electrons. Indirectly. through / ¯ / current conservation. (−p2 . t1 ).22) which is exactly (−1)n times the result of the Feynman diagram that is obtained by inverting the orientation of the fermion line (i. t1 ). {(k1. t1 p1 . from now on we will drop the distinction between ki and ki) dσ = s1 . We got rid of the gauge dependent part of the photon propagator (see table 6) by using the fact that the currents generated by us (p)γ µ ut (k) are ¯ conserved. k2 .s2 d 3 p2 (2π)4 δ4 (p1 + p2 − k1 − k2 ) d 3 p1 × 2p0 (p1 )(2π)3 2p0 (p2 )(2π)3 4 (k1 · k2 )2 − m4 1 4 t1 . this is related to gauge invariance. It guarantees that the longitudinal component of the photon does not contribute to the scattering matrix. their contributions will cancel whenever n is odd. In lowest nontrivial order there are only two diagrams that contribute. (k2 . (17. s2 )}. (17.21) to the expression (−q)n Tr γ µn · · · γ µ3 1 1 1 γ µ2 γ µ1 −p − k1 − k2 − m + iε / / / −p − k1 − m + iε / / −p − m + iε / . t2 k1 . (p2 . t2 )})2 77 . s2 )}. as indicated in the ﬁgure. t2 )}) = −i(2π)4 (1) (2) (1) (2) 2p0 (p1 )(2π)32p0 (p2 )(2π)3 2k0 (k1 )(2π)3 2k0 (k2 )(2π)3 iδ4 (p1 + p2 − k1 − k2 ) us1 (p1 )eγ µ ut1 (k1 )gµν us2 (p2 )eγ ν ut2 (k2 ) ¯ ¯ 2 + iε (1) (2) (1) (2) (k1 − p1 ) (4π)2 p p k k 0 0 0 0 − us1 (p1 )eγ µ ut2 (k2 )gµν us2 (p2 )eγ ν ut1 (k1 ) ¯ ¯ 2 + iε (k1 − p2 ) . s2 )(k1 . The scattering matrix (ignoring the time dependent phase factor) for this process is given by out< (p1 . {(k1 . s2 − k2 .12).24) because onshell (k − m)u(k) = 0 (and hence also u(k)(k − m) = 0). As both diagrams will occur.e. the vertices are connected in the reversed order) as indicated in ﬁgure (b). s1 k1 . s1 ). s2 r r ¢¡ £d d ¢ £ ¡ ¢ d ¨¡ p . (10.23) The relative minussign is of course a consequence of the socalled FermiDirac statistics. t2 j £¨ rB ¨ r ¢¡ £ ¢¡ £ ¢ ¨¡ r ¨j B r p2 . ¯ ¯ / / (17. which implements the Pauli principle.Using the fact that for any matrix A we have Tr(A) = Tr(At ) = Tr(CAt C −1 ). s1 ). t1 ). (17.
(16. µ µ (17. eq.27) For antiparticles k0 < 0 (see eq.s2 .26) ¯ / / ' ¨ £ £ E ¢¡ ¢ © £¡ µ ¢ p £¡ν ¡ ¢ which can be represented graphically as follows k. π/2]. Tr(γµ γν γα γβ ) = 4(gµν gαβ + gµβ gαν − gµα gνβ ) .t2 k2 ' ¨ £ p2 ¢ £ ¡ £d ¢ ¡ ¢ d £ ¡ £¡ ¢ ¢ 1 d ¡ p © k' 1 + (p1 ←→ p2 ) = e4 Tr[γ µ (k1 + m)γ ν (p1 + m)] Tr[γµ (k2 + m)γν (p2 + m)] / / / / 2 + iε) ((k − p )2 + iε) ((k1 − p1 ) 2 2 µ ν Tr[γ (k1 + m)γ (p2 + m)γµ (k2 + m)γν (p1 + m)] / / / / − ((k1 − p1 )2 + iε) ((k1 − p2 )2 + iε) +(p1 ←→ p2 ) . t = t k. To calculate M2 we use us (p)γµ ut (k) (¯s (p)γν ut (k))∗ = ¯ u s. ﬁg.t1 . These results can be generalised to other ﬁelds too. we add a Feynman rule for the socalled cut fermion propagator k E / a = sign(k0 )(k + m)ab . For 4 the total cross section we should multiply with a factor 1 to avoid double counting the 2 identical outgoing electrons. π/2].t † us (p)γµ ut (k) ut (k)γ0 γν γ0 us (p) = ¯ ¯ Tr γµ t ut (k) ⊗ ut (k)γν ¯ s us (p) ⊗ us (p) = Tr γµ (k + m)γν (p + m) . We can now use this result to compute M2 M2 = k2 ' ¨ £ £ ¢ E ¢¡ © £¡ p2£ ¢ 1 ¢¡ − p £ ¨ ¡ E ¢ ¡ ' © k 1 s1 . The latter convention will be followed here. t p. t Hence. In the center of mass system the scattered particles move back to back in a direction which is only determined modulo π.t2 stands for averaging over the polarisations of the incoming electrons. γµ γν γα γβ γ µ = −2γβ γα γν . which is why θ ∈ [0. (17. (13. s j r s. by noting that our conventions have been such that the propagators can be written as ∗ ¯ β φβ (k) ⊗ φβ (k) .8)). (17.t s. (17. t© k = . b M M∗ k.28) k 2 − m2 + iε ¯ where φβ (k) are the wave functions for the incoming lines and φβ (k)∗ for the outgoing lines.t ¨ r E £¨ £E ¢¡ ¢B £ £ ¡ ¢ ν µ ¢¡ ¡ k. with θ measured from the incoming particle direction (also deﬁned modulo π). when integrating over the outgoing momenta.18) for the extra minus sign).30) 78 . with β labelling the internal degrees of freedom (cmp. 10 (17.29) To compute the traces over the gamma matrices we use the following identities (problem 21) Tr(γµ γν ) = 4gµν . γµ γα γβ γ µ = 4gαβ . or restrict the scattering angle θ to the interval θ ∈ [0.where 1 t1 . t j £ p £¨ rE B ¨ r ¢ = ¢¡ £¡ £ µ ¢ ¢ ν ¡ ¡ t d s £ d£ ¢E p ¢¡ £ ¡ £ ¢¡ν µ ¢¡ k.
(17. (17. Together with momentum conservation (p1 + p2 = k1 + k2 ) and the onshell conditions 2 2 (p2 = p2 = k1 = k2 = m2 ).s2 .34) Deﬁning θ to be the angle between k and p.t2 M2 . We ﬁnally perform some kinematics and express the diﬀerential cross section in terms of the scattering angle θ. (17. (17.and the fact that the trace over an odd number of gamma matrices vanishes.31) (17.t1 . i.s2 . 1 (p2 − k1 )2 = −4k 2 cos2 ( 2 θ) . (17. dΩ = sin θdθdφ) d3 p1 d3 p2 δ4 (p1 + p2 − k1 − k2 ) = p2 dpdΩ δ 2 p 2 + m2 − 2 k 2 + m2 = 1 dΩ 2 Ek . k1 = −k2 ≡ k with p = k .t1 . (k2 · k1 )2 − m4 = 4E 2 (E 2 − m2 ) = 4k 2 E 2 k1 · p1 = E 2 − k 2 cos θ = E 2 (1 − cos θ) + m2 cos θ . . p1 = −p2 ≡ p . (17.25) the result 1 (2π)4 1 Ek dσ 2 · = · dΩ (2E(2π)3 )2 4 4k 2 E 2 1 4 s1 . we ﬁnd 1 2 M = e 2 4 µ ν 16 g µν (m2 − k1 · p1 ) + k1 pν + k1 pµ 1 1 2 2 gµν (m2 − k2 · p2 ) + kµ p2 + kν p2 ν µ s1 .t2 M2 = 1 210 π 2 E 2 s1 .29). By keeping track of the dependence on κ one sees how scattering experiments can be used to verify the anticommuting nature of the electrons.33) The parameter κ determines the relative sign of the “crossed” diagrams in eq. For FermiDirac statistics κ ≡ −1.35) Finally we use that identity (Ω is the solid angle. which imply identities like p1 · p2 = k1 · k2 . This implies Tr γµ (k +m)γν (p +m) = Tr γµ k γν / +4m2 gµν = 4(m2 −k·p)gµν +4kµ pν +4kν pµ / / / p and µ .t1 .t2 (k1 − p1 )2 + iε (k2 − p2 )2 + iε + (p1 ↔ p2 ) −κ ν / / Tr 2p2 γ ν k1 + 2m2 γ ν − 4mk1 − 4mpν (k2 + m)γν (p1 + m) / / 2 (k1 − p1 )2 + iε (k1 − p1 )2 + iε −κ (k1 − p2 )2 + iε 2 = 32e4 (k1 · k2 )2 + (k1 · p2 )2 + 2m2 (k1 · p2 − k1 · k2 ) (k1 · k2 )2 − 2m2 k1 · k2 (k1 − p1 )2 + iε (k1 − p2 )2 + iε + (p1 ↔ p2 ) . We deﬁne in the center of mass frame 0 0 k 1 = k 2 = p0 = p0 ≡ E 1 2 . (17.e. which arise from multiplying the direct electronelectron scattering diagram with the complex conjugate of the one where the outgoing fermion lines were crossed. p · k = k 2 cos θ. (17.37) 79 . k1 · p2 = E 2 + k 2 cos θ = E 2 (1 + cos θ) − m2 cos θ .32) γµ (k + m)γν (p + m)γ µ = −2pγν k − 2m2 γν + 4mkν + 4mpν / / / / . we have k1 · k2 = E 2 + k 2 = 2E 2 − m2 .36) Collecting all terms we ﬁnd for eq. 1 (p1 − k1 )2 = −4k 2 sin2 ( 2 θ) .s2 .
s2 .t M({(−p′ . s′ k ′ . t′ d p′ . 2 1 For E ≫ me and E ≫ mµ one ﬁnds (see problem 29) σ = 3 παe h2 c2 /E 2 = 21. t′ + £ £ £ k. such that we cannot tell the two outgoing electrons apart. For electronelectron scattering we have to put κ = −1. t′ )}. s The cross section is now given by dσ = s′ .with s1 . (k. The resulting cross section is called the KleinNishina formula. (17.38) that one can easily distinguish experimentally if electrons behave according to the FermiDirac statistics. t £ £ £ E ¢¡¢¡¢¡ p. as it should be. k. t)})2 80 . but we see from the dependence on κ in eq.t1 .40) . (17. (17. s′ ﬁg. For e− e+ −→ µ− µ+ this divergence is absent and one can deﬁne the total cross section by integrating over all angles.t′ d d s £ £ £ E d ¢¡¢¡ ¢¡ k ′ . There are again two diagrams that contribute in lowest nontrivial order to the scattering matrix. In problem 29 electronpositron scattering is studied with in the ﬁnal state an electron and a positron (Bhabha scattering) or a muon and an antimuon. t ¢¡¢¡¢¡ E p. Both for e− e− −→ e− e− and e− e+ −→ e− e+ one cannot take θ too close to zero (or π for e− e− ). (17. ¯ We now discuss electronphoton scattering. also known as Compton scattering.7nb/E 2 (GeV).38) (E 2 −m2 )2 + − sin4 θ sin2 θ (k 2 )2 yielding (αe = e2 4π = 32e4 1 1 [(2E 2−m2 )2+E 4+(E 2− m2 )2 cos2 θ+2m2 (m2−E 2 )](cos4 ( 2 θ)+sin4 ( 2 θ)) + sin4 θ = e2 4π¯ c h ≈ 1 137 is the ﬁnestructure constant and κ = −1) . Apart from the fact that the detector would be in the way of the beam. s £ £ £ ¢¡¢¡¢¡ p′ . it is fundamentally impossible to distinguish the scattered particles at θ = 0 (and θ = π for e− e− ) from those in the beam. The divergence of the diﬀerential cross section was therefore to be expected.39) 4 α2 (2E 2 − m2 )2 4 3 (E 2 − m2 )2 dσ 1+ = e2 2 − + 2 2 )2 sin4 θ 2 − m2 )2 dΩ 4E (E − m sin θ (2E sin2 θ This cross section is invariant under θ → π − θ. (−k ′ . {(p. s).t2 M2 given by 32e4 + 1 16(k 2 )2 sin4 ( 2 θ) (2E 2−m2 )2+[E 2 (1−cos θ) + m2 cos θ]2+ 2m2 [E 2 (1−cos θ)+m2 cos θ+m2−2E 2 ] 16(k 2)2 cos4 ( 1 θ) 2 −κ 2(2E 2−m2 )(2E 2−3m2 ) (2E 2−m2 )2+[E 2 (1+cos θ)−m2 cos θ]2+2m2 [E 2 (1+cos θ)−m2 cos θ+m2−2E 2 ] 16(k 2 )2 sin2 ( 1 θ) cos2 ( 1 θ) 2 2 (k 2 )2 1 2(2E 2−m2 )(2E 2−3m2 ) cos θ[2E 2 (E 2−m2 )+2m2 (E 2−m2 )](cos4 ( 2 θ)−sin4 ( 1 θ)) 2 −κ sin4 θ 4 sin2 θ 16e4 4(2E 2 −m2 )2 3(2E 2 −m2 )2 −m4 +κ(2E 2 −m2 )(2E 2 −3m2 ) = . 11 d3 p ′ (2π)4 δ4 (p + p′ − k − k ′ ) d3 k ′ × ′ 2p′0 (p′ )(2π)3 2k0 (k ′ )(2π)3 4p · k 1 4 s. s′ ).
i.43) The gauge invariance. This requires averaging over the polarisations of the incoming particles (at the end we will mention the dependence on the photon polarisations). (17.t1 . t) ←→ (−k ′ .t1 . Note that the photon has also two helicity eigenstates. M= 2 − m2 + iε (p + k) ′ (17. The reduced matrix element for the two diagrams is given by (t / / us′ (p′ )eγ µ εµ ) (k ′ )∗ (p + k + m)eγ ν ε(t) (k)us (p) ¯ ν + ((k. each with one fermion loop and with an odd number of vertices.t2 M2 = # E p′ £ £ £ ¢¡ ¢¡ ¢¡ p+k £ £ £ p′ +k ′ T¢¡¢¡ c ¢¡ Tr[γ µ (p + k + m)γ ν (p + m)γ ν (p + k + m)γ µ (p′ + m)] / / / / / / =e 2 − m2 + iε)2 ((p + k) ′ ′ µ Tr[γ (p + k + m)γ ν (p + m)γ µ (p − k′ + m)γ ν (p′ + m)] / / / / / / + × ((p + k)2 − m2 + iε) ((p − k ′ )2 − m2 + iε) ′ ' "! p + + (k ′ ←→ −k) 4 (t εµ ) (k ′ )εµ′ (k ′ )∗ t′ =± t=± ′ (t′ ) εν ′ (k)ε(t) (k)∗ + (k ′ ←→ −k) . Consider for example the term from the cut photon propagators that contains kν . There remain two diagrams. Furry’s theorem tells us that these two diagrams add to zero.42) Like for electronelectron scattering. for the choice of polarisations discussed in eq.21).where.44) This means that one of the photon vertices is removed. is given by (k 2 ≡ 0) µ £ £ £ £ ¢¡ ¢¡ ¢¡ ¢¡ k ν ≡ t=± ε(t) (k)ε(t) (k)∗ = − gµν − µ ν ←k £ £ £ £ £ ¢¡¢¡ ¢¡ ¢¡ ¢¡ d p′ c ′ d £ d £ Td ¢¡¢¡¢¡p −k ' &p % ′ kµ nν (k) + kν nµ (k) k · n(k) . we will discuss unpolarised cross sections.t contains four terms. (17.41) We leave it as an exercise to verify that the cut photon propagator. conservation of the fermionic current.e. t′ )) . (16. one will arrive at the result α2 dσ = e2 dΩ 2m ′ k0 k0 2 ′ k0 k0 + − sin2 θ ′ k0 k0 Tr[γ µ (p + k + m)γ ν (p + m)γν (p + k + m)γµ (p′ + m)] / / / / / / + 2 − m2 + iε)2 ((p + k) .45) 2 − m2 + iε) ((p − k ′ )2 − m2 + iε) ((p + k) Taking the incoming electron at rest (p = (m. (17. as for electronelectron scattering.s2 . It gives rise to the combination (using that p2 = m2 ) (p + k + m)γ ν kν (p + m) = (p + k + m){(p + k − m) − (p − m)}(p + m) / / / / / / / / / 2 2 = ((p + k) − m )(p + m) . 0)).t2 M = e 2 4 Tr[γ µ (p + k + m)γ ν (p + m)γµ (p − k′ + m)γν (p′ + m)] / / / / / / + (k ′ ↔ −k) .46) 81 . We may therefore just as well replace the cut photon propagator by −gµν . following similar steps as for electronelectron scattering. together with the electron s. Using this we ﬁnd s1 . ν (t) (17.s2 . In this case the argument is somewhat more subtle. / (17. we can compute M2 graphically by s1 . is again instrumental in decoupling the longitudinal component of the photon ﬁeld.
For every g. which form a socalled isospin doublet. It is now tempting to generalise this to other.47) For a detailed derivation we refer to section 521 of Itzykson and Zuber and to section 86 of Berestetskii. This group is wellknow from the description of spin onehalf particles. Like for U(1) gauge theories they made the SU(2) transformation into a local one. (17.50) 18 NonAbelian gauge theories Quantum Electrodynamics is an example of a U(1) gauge theory.1) It forms a group. as for most other textbooks. (It should be noted that they were originally after describing the isospin symmetry. that relates protons to neutrons. for which no longer gh = hg. giving rise to socalled internal symmetries. Furthermore. (17. With this choice it is even possible to determine the polarised cross section (the polarisation of the electron is assumed not to be observed) dσ α2 = e2 dΩ pol 4m ′ k0 k0 2 ′ k0 k0 + + 4(ε′ · ε)2 − 2 ′ k0 k0 . h ∈ U(1). is SU(2). U(1) is the group of the unimodular complex numbers and determines the transformation of the charged ﬁelds Ψ(x) → exp(−iqΛ(x))Ψ(x) ≡ g(x)Ψ(x) . in general noncommutative groups. It was the way how Yang and Mills discovered SU(2) gauge theories in 1954. As a local gauge theory it does no longer act on the spinor indices. (17. (18. which are called nonAbelian groups. (see section 1 for the reference).48) where ε and ε′ are the polarisations of resp. The unit 1 satisﬁes g1 = 1g = g. the result is derived by choosing the photon polarisation such that ε(k) · p = 0 (keeping ε(k) · k = 0).where θ is the angle of the scattered photon with the direction of the incident photon. which satisﬁes gg −1 = g −1 g = 1. which means that for any two elements g. U(1) is called an Abelian group because its product is commutative.49) The unpolarised cross section in this limit is obtained by summing over the scattered and averaging over the incident polarisations 2 α2 8παe dσ = e 2 (1 + cos2 θ) and σ = dΩ 2m 3m2 . but on indices related to some internal space. It has a two dimensional (spinor) representation. (17.a. which can also be seen as a representation of the rotation group SO(3). gh = hg. the product is also in U(1). e. the incident and scattered photon. From energy and momentum conservation one ﬁnds that ′ k0 = k0 1 + (k0 /m)(1 − cos θ) . for any g ∈ U(1). any element has an inverse g −1 .) The simplest nonAbelian gauge group. In Itzykson and Zuber. where at every point the ﬁeld can be transformed independently. h ∈ U(1). The way the 82 . When ′ k0 ≪ m one obtains the wellknown Thomson formula dσ α2 = e (ε′ · ε)2 dΩ pol m2 .
For ρ to be a representation.e.3) We will generally restrict the gauge symmetries to Liegroups for which one can write any group element as an exponential of a Liealgebra element g ≡ exp(X) . Y ]]+ 12 [Y. ρ : G → Map(V ).gauge group G acts on the ﬁelds Ψ is described by a representation of the group G. the product of two exponentials is again an exponential. satisfying eq. Y ]) = [ρ(X).8) (18.6) This formula will be of great help in ﬁnding a simple criterion for ρ to be a representation. Y ) ≡ log (exp(X) exp(Y )) = X +Y + 2 [X. of the linear vector space V into itself. [X. a real or a complex n × n matrix. (18. it has to preserve the group structure of G ρ(g)ρ(h) = ρ(gh) . (18. (18. Apart from the group structure of Map(V ). it also has a Liealgebra structure (the commutator of two n×n matrices is again a n×n matrix).3). 1 1 1 F (X. As a Liealgebra forms a linear vector space we can deﬁne a basis on LG . the exponential is locally an invertible mapping). is completely determined by the structure constants fabc [T a . [Y. Y ) ∈ LG × LG → [X.5) The CampbellBakerHausdorﬀ formula expresses that the logarithm of exp(X) exp(Y ) is an element of the Lie algebra.7) (18. T b] = c fabc T c 83 . (18. Y ] ∈ LG . The commutator. Without causing too much confusion we can use the same symbol for the two objects. ρ(1) = idV . antisymmetric bilinear product (required to satisfy the Jabobi identity. X ∈ LG . The representations of the group can be easily restricted to the Liealgebra. n Z= a=1 za T a ∈ LG .11) . in a way that preserves the Liealgebra structure ρ([X. ρ(g) : V → V . in which case ρ(g) is resp. V will be either IRn or C n . (18. as deﬁned in eq. (18. i.9) where on the lefthand side ρ is the group representation and on the righthand side it is the Liealgebra representation. / za ∈ IR (or C) . (18.4) This Liealgebra has a noncommutative.10) In here n is the dimension of the Liealgebra (and the Liegroup if. ρ(Y )] = ρ(X)ρ(Y ) − ρ(Y )ρ(X) It is more or less by construction that we require ρ(exp(X)) = exp(ρ(X)) . (18. A representation deﬁnes a mapping ρ from G to the space of linear mappings Map(V ). (18. Y ]+ 12 [X.12)) (X. (18. as we will assume throughout. X]]+· · · ∈ LG . T a ∈ LG . also called Lieproduct.2) / Mostly. ρ : LG → Map(V ) .
19) and from the fact that for compact groups the generators can be normalised such that a b Tr(Tfnd Tfnd ) = − 1 δab 2 . (18. (18. one easily veriﬁes that [ρad (T a ). (18.12) applied to X = T a . Tρ ] = fabc Tρ . Tρ ]Tρ ) = Tr(Tρ [Tρ . we ﬁnd (from now on sums over repeated group indices are implicit) fbcd fade + fcad fbde + fabd fcde = 0 . since a representation preserves the commutation relations. (18. (18.16) a This is because under the action of ρ one simply replaces T a by Tρ a ρ exp(xa T a ) = ρ(exp(X)) = exp(ρ(X)) = exp xa Tρ . (18. They are called the generators of the representation. ρad (T b )] = fabc ρad (T c ) . (18. ρad (T a ) = −La .Using the Jacobi identity [X. (18. [Z. Note that the structure constants are antisymmetric with respect to the ﬁrst two indices. The simplest example is SU(2).20) a where Tfnd are the generators of the socalled fundamental or deﬁning representation of the group G. This matrix representation is usually identiﬁed with the group (or algebra) itself. (18. (12. Y ]] = 0 . Its fundamental representation coincides with the spinor or spin onehalf representation. the CampbellBakerHausdorﬀ formula. Tρ ]) = fbcd Tr(Tρ Tρ ) .6) we easily verify that ρ is a representation if and only if a b c [Tρ . They are also invariant under cyclic permutations of the indices. [Y.16) is satisﬁed.13) This precisely coincides with the commutation relations of the socalled adjoint representation a (Tad )bc ≡ ρad (T a )bc = facb .9)) i a (18. when expressed with respect to the Liealgebra basis {T a } (xd xb yc + yd yb xc )fbce fdea + · · ·}T a . This follows from the cyclic property of the trace 1 12 1 exp xa T a exp yb T b = exp {xa + ya + 2 xb yc fbca + d c a b c a b c a d fabd Tr(Tρ Tρ ) = Tr([Tρ . it also preserves a the structure constants in terms of ρ(T a ) ≡ Tρ . Z]] + [Y. provided eq. The structure constants and the generators of the fundamental and adjoint representations were considered in sect. which till now was seen more as an abstract entity. Y = T b and Z = T c .14) Indeed.18) directly determines the multiplication of the representation of group elements by replacing T a a by Tρ . [X. which forms a basis for the linear representation space which is contained in V .21) Tfnd = − σ a .15) In general. 12 (see eq. fabc = εabc . 2 84 .17) Similarly. X]] + [Z. (18. With the help of eq. (18. the set of complex unitary 2 × 2 matrices with unit determinant.
The proof simply states how in the Taylor expansion products of Liealgebra elements are regrouped in multiple commutators. So it is suﬃcient to prove that the equation is satisﬁed for Q a Liemonomial.e.26) therefore yields Dad(Xi1 )Q(1) = ad(Xi1 )DQ(1) + ad(Xi1 )Q(1) . Y ] (18. (18. DXi1 Xi2 · · · Xis ≡ [Xi1 . Y ) ≡ log (exp(X) exp(Y )) = log 1 + = i+j>0 i!j! F (X. Y ) is an element of the Liealgebra.28) from which we easily obtain the explicit expression for the CampbellBakerHausdorﬀ formula in terms of multiple commutators. (18.25) For two Liealgebra elements X and Y .29) . · · · . (12. adY ] . Y ]) = [adX. Y ] = D(XY ) − D(Y X) = adXDY − adY DX = [X. we will now provide a more abstract derivation of eq. Y ) . Y ) the collection of all terms at ﬁxed order m. X iY j F (X. Y ) (−1)k−1 X iY j .27) It is not diﬃcult to work out the Taylor expansion for F (X. (18. introduced in eq. m Fm (X. 2. Y ) = m {k. in the Taylor expansion of F (X. such that F (X.12) (18. pj +qj >0} (−1)k−1 D (X p1 Y q1 X p2 Y q2 · · · X pk Y qk ) km p1 !q1 !p2 !q2 ! · · · pk !qk ! 85 . (18. Y ) ≡ Fm (X. A crucial ingredient for deriving the CampbellBakerHausdorﬀ is the socalled derivation D.6) to all orders. s is an element of the Liealgebra (i. Now it is trivial to regroup the terms in the Taylor expansion of eq. Using eq. i = 1. If this equation is satisﬁed it is clear from the deﬁnition of a derivation that Q is a Liemonomial. Induction in m gives the required result. k j=1 k pj +qj =m. [Xi2 . (18. In that case Q is a sum of terms. denoted by Fm (X. Y ) = 1 DFm (X. From the fact that any group element can be written as the exponent of a Liealgebra element.6) in multiple commutators. that maps a product of Liealgebra elements into a multiple commutator. (18. called a Liemonomial) if and only if DQ = mQ. it can easily be shown that D[X. we know that F (X.26) and this allows us to prove by induction that a monomial Q (a polynomial of which all terms are of the same order) of degree m in terms of Liealgebra elements Xi . Consequently.23) It is more or less by deﬁnition that for any two products u and v of Liealgebra elements D(uv) = aduDv . DX = X . DY ] + [DX.22) We also deﬁne for these products the adjoint map. Xis ] · · ·]] . k i+j>0 i!j! k (18. each of which can be written as ad(Xi1 )Q(1) with Q(1) a Liemonomial of degree m − 1. m (18. s>1 .24) . can be written as multiple commutators. adXi1 Xi2 · · · Xis ≡ adXi1 adXi2 · · · adXis which is easily seen to satisfy ad([X. Y ) . ad.Because the CampbellBakerHausdorﬀ formula plays such a crucial role in the theory and in the practical implementation of group representations. · · · [Xis−1 . (18. is a monomial in X and Y and Fm (X. Y ) ∈ LG (at the worst one needs to restrict X and Y to suﬃciently small neighbourhoods of the origin in LG ). Y ).
We leave it to the industrious student to verify that F1 (X, Y ) = X + Y, F2 (X, Y ) = 1 [X, Y ], 2 F3 (X, Y ) =
1 360 1 12
(adX)2 Y + (adY )2 X ,
1 F5 (X, Y ) = − 720 (adX)4 Y + (adY )4 X +
1 F4 (X, Y ) = − 24 adXadY adX(Y ),
adX(adY )3 X + adY (adX)3 Y . (18.30)
1 − 120 adXadY (adX)2 Y + adY adX(adY )3 X
After this intermezzo we return to the issue of constructing nonAbelian gauge theories. The simplest way is by generalising ﬁrst the covariant derivative. U(1) gauge transformations act on a complex ﬁeld as in eq. (18.1), and the covariant derivative is designed such that Dµ Ψ(x) → g(x)Dµ Ψ(x) . (18.31)
Since the gauge ﬁeld transforms as in eq. (17.3), this is easily seen to imply that the covariant derivative is deﬁned as in eq. (17.2) (these formula are of course also valid for complex scalar ﬁelds, compare eq. (3.36)). For a nonAbelian gauge theory we consider ﬁrst a ﬁeld Ψ that transforms as an irreducible representation (i.e. there is no nontrivial linear subspace that is left invariant under the action of all gauge group elements) Ψ → g Ψ ≡ ρ(g)Ψ . (18.32)
In the following, as in the literature, we shall no longer make a distinction between g and ρ(g). It will always be clear from the context what is intended. The vector potential should now be an element of the Liealgebra LG (more precisely a representation thereof) Aµ = Aa T a µ . (18.33)
For U(1), which is onedimensional, we need to deﬁne T 1 ≡ i. The Liealgebra of the group consisting of the unimodular complex numbers is the set of imaginary numbers, LU(1) = iIR. Note that as an exception this generator is normalised diﬀerent from eq. (18.20), so as not to introduce unconventional normalisations elsewhere. The real valued vector potential Aµ will now be denoted by A1 and we see that under a gauge transformation µ Aµ → g Aµ = gAµ g −1 − q −1 (∂µ g)g −1 = gAµ g −1 + q −1 g∂µ (g −1 ) . (18.34)
This is the form that generalises directly to the nonAbelian gauge groups with the covariant derivative deﬁned by Dµ Ψ = (∂µ + qAµ ) Ψ , (18.35)
a where Aµ ≡ Aa Tρ is a matrix acting on the ﬁelds Ψ. We leave it as an exercise to verify µ that under a gauge transformation, eq. (18.31) remains valid for the nonAbelian case. It is now a trivial matter to construct a Lagrangian that is invariant under local gauge transformation. Assuming the representation is unitary, for a scalar ﬁeld Ψ one has
LΨ = (Dµ Ψ)† D µ Ψ − m2 Ψ† Ψ ,
(18.36)
whereas if Ψ is a Dirac ﬁeld, carrying both spinor (representation of the Lorentz group) and group indices, one has LΨ = Ψ (iγ µ Dµ − m) Ψ , 86 Ψ = Ψ† γ 0 , (18.37)
where Ψ† is the hermitian conjugate both with respect to the spinor and the group (representation) indices. The part of the Lagrangian that describes the self interactions of the vector ﬁeld Aµ has to be invariant under local gauge transformations too. In that respect U(1) or Abelian gauge theories are special, since the homogeneous part of the transformation of the vector potential is trivial, gAµ g −1 = Aµ . For nonAbelian gauge transformations this is no longer true. For U(1) one easily veriﬁes that
1 Dµ Dν Ψ − Dν Dµ Ψ ≡ [Dµ , Dν ]Ψ = iqFµν Ψ ,
(18.38)
1 where Fµν = ∂µ A1 − ∂ν A1 is the electromagnetic ﬁeld strength, compare eq. (3.27). Because ν µ the covariant derivative transforms in a simple way under gauge transformations, this formula can be directly generalised to nonAbelian gauge theories
Fµν ≡ q −1 [Dµ , Dν ] → gFµν g −1
g
.
(18.39)
For U(1), where g is a number, this means that the ﬁeld strength is gauge invariant, as was noted before. For nonAbelian gauge theories the ﬁeld strength itself is not gauge invariant. Nevertheless, it is simple to construct a gauge invariant action for the gauge ﬁeld
a µν LA = 1 Tr (Fµν F µν ) ≡ − 1 Fµν Fa 2 4
,
(18.40)
a where Fµν are the components of the ﬁeld strength with respect to the Liealgebra basis, a Fµν ≡ Fµν T a = ∂µ Aa − ∂ν Aa + qfabc Ab Ac T a ν µ µ ν
.
(18.41)
We see from LA and LΨ that q plays the role of an expansion parameter. For q = 0 we have n = dim(G) noninteracting photon ﬁelds. They couple with strength q to the scalar or Dirac ﬁelds. For nonAbelian gauge theories, in addition the vector ﬁeld couples to itself. These selfinteractions guarantee that there is invariance under the gauge group G, which is much bigger than U(1)n , which is the symmetry that seems implied by the q = 0 limit. The nonAbelian gauge invariance ﬁxes the “charges” of the ﬁelds with respect to each of these U(1) gauge factors. Without the nonAbelian gauge symmetry there would have been n independent “charges”. The Lagrangian LA is the one that was discovered in 1954 by C.N. Yang and R.L. Mills. The EulerLagrange equations for the Lagrangian LA are called the YangMills equations. One easily shows that
µν ∂µ Fa + qfabc Ab Fcµν = 0 or [Dµ , F µν ] ≡ ∂µ F µν + q[Aµ , F µν ] = 0 . µ
(18.42)
The current is therefore given by
For the coupling to fermions we read oﬀ from eq. (18.37) what the current for the YangMills ﬁeld is LΨ = Ψ(iγ µ Dµ − m)Ψ = Ψ(iγ µ ∂µ − m)Ψ + iqAa Ψγ µ T a Ψ . (18.43) µ
a Jµ ≡ −iqΨγµ T a Ψ .
(18.44)
The coupled YangMills equations read
µν ν ∂µ Fa + qfabc Ab Fcµν = Ja µ
.
(18.45)
In problem 31 it will be shown that the current is not gauge invariant, unlike for Abelian gauge symmetries. Closely related is the fact that it is no longer true that the current is a a c conserved, i.e. ∂ µ Jµ = 0. Instead, it will be shown in problem 31 that ∂ µ Jµ + qfabc Aµ Jµ = 0. b 87
19
The Higgs mechanism
We have seen in problem 30 that the fourFermi interaction in good approximation can be written in terms of the exchange of a heavy vector particle. In lowest order we have resp. the diagrams in ﬁg. 12a and ﬁg. 12b
d d d © s d
(a)
M £ £ £ £ d d ¢¡ ¢¡¢¡¢¡ d © s d (b)
ﬁg. 12
The ﬁrst diagram comes from a four fermion interaction term that can be written in terms of the product of two currents Jµ J µ , where Jµ = Ψγµ Ψ. Here each fermion line typically carries its own ﬂavour index, which was suppressed for simplicity. Fig. 12b can be seen to eﬀectively correspond to ˜ −J µ (−k) gµν − kµ kν /M 2 ˜ν J (k) k 2 − M 2 + iε (19.1)
At values of the exchanged momentum k 2 ≪ M 2 , one will not see a diﬀerence between these two processes, provided the coupling constant for the fourFermi interactions (ﬁg. 12a) is chosen suitably (see problem 30). This is because for small k 2 , the propagator can be replaced by gµν /M 2 , which indeed converts eq. (19.1) to J µ Jµ /M 2 . It shows that the fourFermi coupling constant is proportional to M −2 , such that its weakness is explained by the heavy mass of the vector particle that mediates the interactions. Examples of fourFermi interactions occur in the theory of βdecay, for example the decay of a neutron into a proton, an electron and an antineutrino. In that case the current also contains a γ 5 (problem 40). It turns out that the fourFermi theory cannot be renormalised. Its quantum corrections give rise to an inﬁnite number of divergent terms, that can not be reabsorbed in a redeﬁnition of a Lagrangian with a ﬁnite number of interactions. With the interaction resolved at higher energies by the exchange of a massive vector particle, the situation is considerably better. But it becomes crucial for the currents in question to be conserved, such that the kµ kν part in the propagator has no eﬀect. It would give rise to violations of unitarity in the scattering matrix at high energies (the σ ﬁeld deﬁned in eq. (16.15) has the wrong sign for its kinetic part). To enforce current conservation, we typically use gauge invariance. But gauge invariance would protect the vector particle from having a mass. The big puzzle therefore was how to describe a massive vector particle that is nevertheless associated to the vector potential of a gauge ﬁeld. The answer can be found in the theory of superconductivity, which prevents magnetic ﬁeld lines to penetrate in a superconducting sample. If there is, however, penetration in the form of a quantised ﬂux tube, the magnetic ﬁeld decays exponentially outside the ﬂux tube. This would indicate a mass term for the electromagnetic ﬁeld within the superconductor. The LandauGinzburg theory that gives an eﬀective description of this phenomenon (the microscopic description being given by the BCS theory of Cooper pairs) precisely coincides with scalar quantum electrodynamics.
1 L = − 1 Fµν F µν + (Dµ ϕ)∗ D µ ϕ − κϕ∗ ϕ − 4 λ (ϕ∗ ϕ)2 4
.
(19.2)
In the LandauGinzburg theory ϕ describes the Cooper pairs. It is also called the order parameter of the BCS theory. In usual scalar quantum electrodynamics we would put κ = m2 , where m is the mass of the charged scalar ﬁeld. But in the LandauGinzburg theory of 88
which in the process got a mass (“got fat”). which is precisely the longitudinal component. It looks. (19. But the photon became massive. three for the massive vector particle and one for the absolute value of the complex scalar ﬁeld (its mass is determined by the quadratic part of the potential in the radial direction at ϕ = ϕ0 ). when approaching κ = 0. With a global phase rotation we can chose the point to expand around to be real. In that case the potential V (ϕ) for the scalar ﬁeld has the shape of a Mexican hat. on the other hand it is the unphysical longitudinal component of the gauge ﬁeld. whereas the complex scalar ﬁeld represents two massive real scalar ﬁelds. However. 13 Re ϕ Im ϕ ﬁg. V ﬁg.5) . there is always a massless particle). 13 The minimum of the potential is no longer at ϕ = 0. In a prosaic way one states that the massless excitation (called a wouldbe Goldstone boson) was “eaten” by the longitudinal component of the photon. 89 Lorentz gauge . This is exactly the same number as for ordinary scalar electrodynamics where κ > 0. Im ϕ = 0. called spontaneous symmetry breaking. But the interpretation of the phase of the complex ﬁeld as a longitudinal component of the vector ﬁeld is simply a matter of choosing a particular gauge. because in that case the massless photon has only two degrees of freedom. (19. and has to develop an additional physical polarisation. (19.3) ϕ0 = −2κ/λ . and can be rotated away by a gauge transformation. To count the number of degrees of freedom we implicitly made two diﬀerent gauge choices κ>0: κ<0: ∂µ Aµ = 0.superconductivity. however. from the degeneracy of the minimum of the potential it follows that the ﬂuctuation along that minimum (the phase in ϕ = ϕ0 exp(iχ)) has no mass (this is related to the famous Goldstone theorem. On the one hand χ corresponds to a massless excitation. it happens to be the case that κ is negative. and is independent of the phase of ϕ. this phase χ is precisely related to the gauge invariance. 0 2 M = 2e −κ/λ . Unitary gauge . To ﬁnd the physical excitations of this theory we have to expand around the minimum. But this immediately implies that the terms quadratic in the gauge ﬁeld give rise to a mass term for the photon ﬁeld Dµ ϕ0 2 = e2 ϕ2 Aµ Aµ ≡ 1 M 2 Aµ Aµ . This means we have four massive degrees of freedom.4) Furthermore. but at ϕ∗ ϕ = −2κ/λ. which states that if choosing a minimum of the potential would break the symmetry. that there is a discontinuity in the description of these degrees of freedom.
3) where F (A) ∈ LG is the gauge ﬁxing function (with the gauge condition F (A) = 0. e. The scalar ﬁeld. whose interactions give the gauge ﬁeld a mass. Furthermore. coupled to an SO(3)vector of scalar ﬁelds ϕa . At ξ = 0 this corresponds to the Lorentz gauge. one adds Lgf = − 1 αF 2 . where F (A) ≡ Fa (A)T a . called the ’t Hooft gauge. ∂F (e g A) ∂F (g A) ≡ M( A) = ∂g ∂X g b X at X = 0 . (20. (18. 20 Gauge ﬁxing and ghosts The quantisation of gauge theories in the path integral formalism requires more discussion. since the gauge condition (like the Lorentz gauge ∂µ Aµ = 0) seems to remove only one degree of freedom of the two that are eliminated in the Hamiltonian formulation (see section 16). Gauge ﬁxing will be discussed in the next section.2) This Jacobian.6) 1 Rather than adding to the Lagrangian the gauge ﬁxing term Lgf = − 2 α(∂µ Aµ )2 . It reads 1= Dg  det (M(g A)) δ (F (g A)) . at ξ = ∞ to the unitary 2 gauge. F (A) = ∂µ Aµ = 0). with respect to the Liealgebra basis. M(A) : LG → LG plays the role of the Jacobian. ’t Hooft gauge . (19. (20. that interpolates between these two gauges F ≡ ∂µ Aµ − 2ieξϕ0 Im ϕ = 0. one has dFa (exp(tT ) A) Mab (A) = dt 90 at t = 0 . (20. Problems 34 and 35 discuss the Higgs mechanism in detail for the GeorgiGlashow model. From a simple example it is easily demonstrated what the eﬀect of gauge ﬁxing on a (path) integral is.There is a gauge. so as to cancel the unphysical components of the gauge and scalar ﬁelds. The gauge transformation g of the gauge ﬁeld A is indicated by g A. is called the Higgs ﬁeld. (20. For this we take f (x) to be a function on IR3 . such that it is a function f (r) of the radius r = x only.g. The starting point is a straightforward generalisation of the identity dx f ′ (x)δ(f (x)) = 1. For the choice ’t Hooft made (ξ = 1/α) the terms that mix (ϕ − ϕ0 ) and Aµ at quadratic order disappear and one easily reads of the masses. see eq.34). where it will be shown how extra unphysical degrees of freedom appear in the path integral. which is invariant under rotations around the origin.1) We know very well that we need a Jacobian factor for the radial integral d3 x f (x) = 4π ∞ 0 r 2 dr f (r) . assuming the equation f (x) = 0 to have precisely one solution (in a sense the righthand side of the equation counts the number of zeros). But it is clear that d3 x f (x) = d3 x δ(x2 )δ(x3 )f (x) = dx1 f (x1 ) . The symmetry group is hence SO(3) and we can attempt to compute the integral d3 x f (x) by introducing a “gauge” ﬁxing condition like x2 = x3 = 0. (20. arising in the change of variable to the invariant radial coordinates and the angular coordinates. which is a nonAbelian gauge theory with gauge group SO(3). can be properly incorporated following the method introduced by Faddeev and Popov.4) Equivalently.5) .
in which case A is said to satisfy the gauge condition. 0.8) With respect to the Liealgebra basis this gives qMab (A) = −δab ∂µ ∂ µ + qfabc (∂ µ Ac + Ac ∂ µ ) .12) . 0) through a rotation over 180 degrees. and the gauge ﬁxing or Gribov ambiguity is not an issue for computing quantities in perturbation theory in q. X] .6) to occur at g = 1.To relate this to the previous equation. g and h. and gauge conditions are chosen so as to avoid this problem in a small neighbourhood of the origin. it is not too diﬃcult to see that g = n4 +iσk nk µ 2 gives an element of SU(2).11) We now use that the action S(A) is invariant under gauge transformations. for h some ﬁxed group element. The Haarmeasure coincides with the standard integration measure on S 3 . for which µ where Dad (A) is the covariant derivative in the adjoint representation µ Dad (A)(X) ≡ ∂ µ X + q[Aµ .7) (20. The Haarmeasure is in general µ invariant under the change of variables g → hg and g → gh. µ F (exp(X) A) − F (A) = −q −1 ∂µ Dad (A)(X) + O(X 2 ) . (20. with dg(x) for every x deﬁned as the socalled Haarmeasure on the group. We have to introduce a further restriction to get the identity d3 x f (x) = 4π d3 x x2 δ(x2 )δ(x3 )θ(x1 )f (x) .3). Even in our simple problem on IR3 . which can be arranged with the help of eq. one makes use of the fact that h g ( A) = (hg) A .10) where θ(x) = 0 for x < 0 and θ(x) = 1 for x ≥ 0. 1 (20. We can insert eq. This means that det(M(A)) can be absorbed in an overall normalisation of the pathintegral. (20. For nonAbelian gauge theories this is no longer possible. shows that any element of SU(2) can be written in terms of nµ . n2 = 1. We leave it as an exercise to verify that likewise DAµ is invariant under the change of variables A → g A. 0. which trivially implies that Z= DAµ Dg det (M(A)) δ (F (A)) exp (iS(A)) 91 . (20. When S 3 is embedded in IR4 as a unit sphere. (20. 0) to (−r. because we can go from (r. µ µ (20. This is in general not correct. In perturbation theory only the gauge ﬁelds near the origin in ﬁeld space are relevant. As an example we consider the Lorentz gauge.6) which states that two successive gauge transformations. (20. the gauge condition x2 = x3 = 0 does not uniquely specify the gauge. We still need to deﬁne what we mean with Dg in eq.9) For an Abelian gauge theory the structure constants fabc vanish and M(A) becomes independent of the gauge ﬁeld. with F (A) = ∂µ Aµ . it is important to note that we assumed the gauge condition F (g A) = 0 to have precisely one solution. which as a space is isomorphic with S 3 . (20. d4 n δ(n2 − 1). whereas exp(iχsk σk ) = cos(χ) + i sin(χ)sk σk . with sk = 1. give the same result as a single gauge transformation with hg. The Lorentz gauge is such a gauge condition. (20. It is best described in the example of SU(2). as was discovered by Gribov. Before describing how the A dependence of det(M(A)) is incorporated.3) in the path integral to obtain Z= DAµ Dg det (M(g A)) δ (F (g A)) exp (iS(A)) . It stands for the integration measure x dg(x).
(20. and in that case det(M(A)) is a constant.16) up to an overall normalisation.14) which combined with the previous equation gives Z = = DY DAµ det (M(A)) δ(F (A) − Y ) exp i DAµ det (M(A)) exp i d4 x L(A) − d4 x L(A) − . k1 µ ≡ −iqfabc k1 ghost vertex (Lorentz gauge) −1 × i d4 k (2π)4 loop factor 92 . Ghosts can only appear in loops and because of the fermionic nature of the ghost variables. Suitably normalising DY we can deﬁne DY exp − d4 x Ya2 (x) = 1 . But here the path integral over Grassmann variables comes to the rescue. (4. They are ¯ therefore called ghosts.22) in the Lorentz gauge. For nonAbelian gauge theories we are left with the task of computing det(M(A)) for each A.The dependence of the integrand on g has disappeared. This modiﬁcation does not aﬀect the socalled FaddeevPopov operator M(A) and we ﬁnd Z= DAµ det (M(A)) δ (F (A) − Y ) exp (iS(A)) αi 2 . In problem 25 we have seen that D ηDη exp i ¯ d4 x η a (x)qMab (A)η b (x) = det(M(A)) . We next note that Z has to be independent of the gauge ﬁxing function F .17) Since η and η are auxiliary ﬁelds. they should never appear as external lines. in particular F (A) − Y is just as good for the gauge ﬁxing (provided of course we show that F (g A) = Y has a solution). ¯ a µ 2 (20. The Feynman rules for the Lorentz gauge are given in the following table. and the integration over g gives an overall (inﬁnite) normalisation factor. every such loop gives a minus sign.13) independent of Y . (20.15) α 2 F (A) 2 a For U(1) gauge theories with F (A) = ∂µ Aµ this precisely reproduces the action of eq. k2 £¡ T E ¢E ¡ µ £ c. which is irrelevant. k3 ¢¡ ≡ (qM )−1 (A=0)ab = δab k 2 +iε ghost propagator (Lorentz gauge) a. which is no longer constant. Z = DAµ D ηDη exp i d4 x L(A)− ¯ table 7 no external ghost lines E k b a b. This implies that the path integral can also be written in the Lorentz gauge as α (∂µ Aµ (x))2 + ∂ µ η a (x)[∂µ η a (x) + qfabc Ab (x)η c (x)] . ¯ (20. α 2 Y (x) 2 a (20.
Its natural scale in energy where quantum eﬀects would ¯ become important is the Planck energy. albeit at the tremendously low rate of one decay in every 1030−31 years. 1 TeV=Terra electronvolt) according to presentday theoretical insight. The theory for which the standard model describes its eﬀective lowenergy behaviour is called a uniﬁed theory. ghosts can no longer be ignored. 21 The Standard Model The standard model describes the electromagnetic and weak interactions. Gravitation has been left out so far. This is veriﬁed explicitly for the GeorgiGlashow model in problem 35. although it is not yet ruled out that it will show its structure 93 . Candidates that unify the standard model with gravity in the form of string theories and supergravity have been unable to provide predictions that either rule them out experimentally or provide evidence in favour of these theories. called the BRS symmetry s. uniﬁed in the socalled electroweak theory of GlashowWeinbergSalam with the gauge group U(1)×SU(2) and the strong interactions. In general it is guaranteed by the existence of an extra symmetry. which does not yet include gravity is a socalled Grand Uniﬁed Theory (GUT). one should have the “masses” (poles) of the ghosts to coincide with the “masses” of the unphysical degrees of freedom. For nonAbelian gauge theories. only the Higgs particle remains to be detected. It is very well possible that a number of the fundamental parameters in the standard model will be determined.Because one can easily derive that for a complex scalar ﬁeld (up to an overall constant) Dϕ∗ Dϕ exp i d4 x ϕ∗ (x)qM ab (A)ϕb (x) = a 1 det(M(A)) . Its mass should be smaller than 1000 GeV (i. known as Quantum Chromodynamics (QCD) with gauge group SU(3). For more details see Itzykson and Zuber. Now the top quark has been found. (20. because of the interaction of the ghost with the gauge ﬁeld. by gravitational interactions. Rouet and Stora.18) we can view a ghost as the elimination of a complex degree of freedom. An intermediate stage. where tested. section 12. These GUT’s predict proton decay. and can not appear as external lines either.41. (20. which is why in QED the introduction of ghosts was never necessary for a consistent description of the theory. It is in this way that in the path integral the two unphysical degrees of freedom of a Lorentz vector are eliminated. The standard model should then be considered as an eﬀective ﬁeld theory. For QED both the ghost and the unphysical degrees of freedom have no interactions. Epl = hc5 /G ≈ 1019 GeV. at a mass of 174 GeV. in particular because theoretical insight of the last ten years has shown that a Higgs ﬁeld is most likely not fundamental. either directly or indirectly. Furthermore the couplings of the ghost and unphysical ﬁelds to the physical ﬁelds should be related. thereby reducing the number of free parameters considerably. agree very well up to about 100 GeV. a swimming pool of (10 m)3 contains enough protons to verify that the proton decay is slower than can be comfortably accommodated by GUT’s. which for example acts on the gauge ﬁeld as follows µ sAµ = Dad η . discovered by Becchi.19) This is precisely an inﬁnitesimal gauge transformation.e. the energies reached by presentday accelerators. Nevertheless. Theory and experiment. The simplest version uniﬁes the electroweak and strong interactions using a gauge group SU(5) (which has U(1)×SU(2)×SU(3) as a subgroup). Much is therefore still to be discovered. To have the ghosts eliminate the unphysical degrees of freedom.
however. (21.only at Planck energies.5) g2 + g′2 √ 1 1 receives a mass mZ = 2 2F (g 2 +g ′2 ) 2 = MW / cos θW . and one massive component η ≡ Re ϕ2 with a mass √ √ (21.e. Under U(1) this Higgs ﬁeld transforms with a coupling constant − 1 g ′ .2 Apparently. Im ϕ1 and Imϕ2 . (21. This gauge ﬁeld deﬁnes a U(1) subgroup of SU(2)×U(1) that leaves φ0 invariant. Indeed one combination among these four will not have something to “eat” and will therefore stay massless. This U(1) subgroup is a combination of the U(1) subgroup of SU(2) generated by exp(iχσ3 ) and the phase rotations exp(iχ) associated with the explicit U(1) group with Bµ as its gauge ﬁeld. to be discussed later). a The standard model consists of gauge ﬁelds Bµ (for U(1)). where a runs from 1 to 8. a=1 (21. These gauge ﬁelds have interactions / with a Higgs ﬁeld φ ∈ C 2 .4) 4 4 √ 1. the vector ﬁelds Wµ will have a mass MW = 1 2gF . 2 and 3. . tan θW = remains massless. To see this write 0 ϕ1 ϕ1 φ= + ≡ φ0 + . which transforms under SU(2) as a spin onehalf representation (i.2) 1 where κ ≡ − 2 λF 2 . These 2 couplings are represented in the covariant derivative i ig Dµ φ = ∂µ φ − g ′ Bµ φ − 2 2 3 a Wµ σa φ .6) µ Zµ = . It plays the role of the photon ﬁeld as we got to know it in QED. (21. It is trivial to verify that the product of these group elements. is / degenerate on a three dimensional sphere. which would give rise to three massless scalar particles according to the Goldstone theorem. 94 3 g ′Bµ − gWµ 3 = sin θW Bµ − cos θW Wµ g′ g . The linear combination perpendicular to Zµ . If that is the case the mass of the Higgs should. but all three will be “eaten” by longitudinal components of the gauge ﬁelds to which the Higgs a ﬁeld couples. all eaten by the vector particles W and Z.3) F ϕ2 ϕ2 such that g2F 2 F2 ′ 1 2 3 (Dµ φ0 )† D µ φ0 = [(Wµ )2 + (Wµ )2 ] + [g Bµ − gWµ ]2 . not be much bigger than 100 GeV. Bµ and Wµ for a = 1. speciﬁed by φ† φ = F 2 (φ ∈ C 2 ∼ IR4 ). also called the vacuum. Wµ (for SU(2)) and Aa (for µ SU(3). exp(iχ) exp(iχσ3 ) indeed leaves φ0 invariant. (21. The gauge symmetry associated to this socalled diagonal U(1) subgroup therefore remains unbroken and corresponds to electromagnetism. In this case the minimum of the potential. We note that there are four gauge ﬁeld components. The Higgs ﬁeld has three massless components Re ϕ1 . whereas it is neutral under SU(3).1) The potential for the Higgs ﬁeld causes spontaneous breaking of part of the symmetries V (φ) = λ λ † (φ φ − F 2 )2 = κφ† φ + (φ† φ)2 + const. whereas the linear 2 combination .7) mη = −2κ = λF . 4 4 (21. the fundamental representation) with a coupling constant g. 3 Aem = cos θW Bµ + sin θW Wµ .
where e = −g sin θW . They are arranged according to families.B = − 4 ∂µ Wν − ∂ν Wµ + gεabc Wµ Wν 2 − 1 (∂µ Bν − ∂ν Bµ )2 4 . As these are wouldbe Goldstone bosons.11) a a µν with the obvious shorthand notations like (Fµν )2 = Fµν Fa .It is this component that is called the Higgs ﬁeld. the up and the down quark. em em em Fµν = ∂µ Aν − ∂ν Aµ . (21. because like µ ϕ0 . We now introduce the fermions in the standard model.23. The Z vector ﬁeld will remain neutral under the electromagnetic interactions.10) As a consequence.12) where we have deﬁned em Dµ = ∂µ − ieAem µ . From experiment it follows that sin2 θW ≈ 0. After some algebra the above equation can be rewritten as em + − + − LW. It does not couple to Aem .9) it is clear that ϕ2 has no electric charge. we have to ﬁnd the coupling of the various ﬁelds to Aem .B = − 1 Fµν + ie(Wµ Wν − Wν Wµ ) 4 2 2 em − em − − − 1 − 2 Dµ Wν − Dν Wµ − ig cos θW (Zµ Wν − Zν Wµ )2 + − + − 1 − 4 ∂µ Zν − ∂ν Zµ + ig cos θW (Wµ Wν − Wν Wµ ) . the neutrino.13) We immediately readoﬀ that our charge assignments for Z and W ± have been correct. µ ϕ2 2 2 0 0 0 (21. = −ieFµν Wµ Wν .8) Using the fact that σ3 + 1 = (21. Alternatively. (21. (21. Essential in the standard model is that invariance under parity is broken explicitly by the weak interactions (as has been observed in the beta decay of 95 . also η is not aﬀected by the transformation exp(iχ) exp(iχσ3 ). For this it is suﬃcient to consider the following part of the µ Lagrangian a a b c 1 LW. also called the Weinberg angle.14) which is a direct consequence of the spin of the vector ﬁeld (the magnetic moment for the Dirac ﬁeld is discussed in problem 32). (21. (21. the two coupling constants g and g ′ are determined by the electric charge e and the socalled weak mixing angle θW . this can be seen from the covariant derivative Dµ ϕ1 ϕ2 = ∂µ − ig ig 1 2 (σ1 Wµ + σ2 Wµ ) + (cos2 θW σ3 − sin2 θW )Zµ 2 2 cos θW ig ϕ1 − sin θW (σ3 + 1)Aem . it will turn out that the √ 1 2 ± 1 combinations Wµ = 2 2(Wµ ∓ iWµ ) are charged with an electric charge of ±e. “eaten” by the vector ﬁelds. whereas ϕ1 has a charge q = −g sin θW .mom. because of its coupling to Fµν em + − Lmagn. Note ± em that the vector ﬁeld Wµ has an extra magnetic moment. The ﬁrst family with the smallest masses consists of the electron. To verify the charge assignment to the vector ﬁelds.
It also immediately ﬁxes the representation to which ΨR. (21. (21. YR = −2 . YR = YL − 1 .17) by replacing φ with its socalled vacuum expectation value φ0 LΨ. This is achieved by coupling the scalar ﬁeld φ to the fermions.e. ΨL) and has a mass me = yF .19) where the index on ΨL indicates the socalled isospin index. It should be stressed that the standard model does not explain why parity is violated.φ = −yF (Ψ ΨL + Ψ2 ΨR ) . The electron is identiﬁed 1 with the pair (ΨR .φ = −y(Ψ φ† ΨL + Ψ φΨR ) . ΨR = 2 (1 + γ5 )Ψ . which as it should 2 be. The righthanded component should of course have 96 . 2 The Dirac Lagrangian in terms of these helicity eigenstates can be written as LΨ = Ψ(iγ µ ∂µ − m)Ψ = Ψ (iγ µ ∂µ )ΨR + Ψ (iγ µ ∂µ )ΨL − m(Ψ ΨL + Ψ ΨR ) . it is a doublet. (21. we require that ΨL is also in the fundamental representation.21) This also allows us to ﬁnd the electric charge of ΨL to be g sin θW = −e. It is most easily determined from the covariant derivative. For each fermion we deﬁne 1 (21.L should belong.1) and (21. We want the neutrino to have no electric 2 charge and this ﬁxes the hypercharge of ΨL .15) ΨL = 1 (1 − γ5 )Ψ . (21. Since the Lagrangian has to be invariant with respect to the gauge symmetries. 2 R L (21. On the other hand ΨR is taken to be invariant under SU(2) (also called the singlet representation). The situation for the quarks will be discussed below).18) The mass of the fermions is now readoﬀ from eq. see problem 40). (21. in units of − 1 g ′ . This is achieved by coupling the left.L enforce m = 0. the absence of an explicit mass term. i. We also see that ΨL remains massless and this is exactly the neutrino.L ) . it was put in “by hand”. acting on the lefthanded fermion.16) such that diﬀerent transformation rules for ΨR. The beauty of the Higgs mechanism is that it also provides a mass for the fermions.and righthanded helicity eigenstates of the fermions diﬀerently to the gauge ﬁelds. coincides with the electron charge. deﬁned as in eqs. using a Yukawa coupling R L LΨ. 2 YH ≡ Y (φ) = 1 . and since the scalar ﬁeld is in the fundamental representation of SU(2). (21. i.Cobalt60. since both are in the same representation (electron and neutrino are also neutral with respect to the strong interactions. the spinor index of the two dimensional fundamental representation for the internal SU(2) symmetry group. YR.e.L ≡ Y (ΨR.20) − µ 2 To make ΨL decouple from the electromagnetic ﬁeld we require 1 YL = −1 .17) where y is the Yukawa coupling constant. This coupling is parametrised by the socalled hypercharge Y . R L R L (21. The couplings of the fermions to the gauge ﬁeld Bµ have to be chosen such that the Lagrangian is neutral. Dµ Ψ L = ∂µ − ig ig 1 2 (cos2 θW σ3 − YL sin2 θW )Zµ (σ1 Wµ + σ2 Wµ ) + 2 2 cos θW ig sin θW (σ3 + YL)Aem ΨL .8).
20). from which we read oﬀ 3 the electric charges 2e L uL u 3 = −e L Qem q L = Qem . T = − 0 2 0 0 −i 0 i 6 0 .22) with the expected coupling to the electromagnetic ﬁeld. 0 0 0 0 0 1. 0 −2 0 0 −1 0 . (18. (In terms of the socalled GellMann matrices one has T a = −iλa /2. If we denote the quark ﬁelds by q.) We leave it as an exercise to determine the structure constants.the same electric charge.24) L d d 3 The righthanded components of both the up and down quarks are singlets under SU(2) and their hypercharges are chosen to insure that they have the same electric charge as for their lefthanded partners 4 2 Y (uR ) = and Y (dR ) = − . Note that the Liealgebra for the group SU(N) is given by traceless and antihermitian (X † = −X) complex N × N matrices. The fractional electric charge of the quarks is not observable (otherwise we would have had a diﬀerent unit for electric charge). The lefthanded up and down quarks are combined in a doublet L representation for SU(2). 2 (21. We now discuss quarks.25) 3 3 The quarks transform nontrivially under SU(3). In that case the covariant derivative is given by Dµ ΨR = (∂µ − ig ′ ig ′ YR Bµ )ΨR = (∂µ − YR [sin θW Zµ + cos θW Aem ])ΨR µ 2 2 = (∂µ − ieAem − ie tan θW Zµ )ΨR .23) 1 On a doublet (ΨL and ϕ) one has I3 = 2 σ3 . Also. whereas I3 = 0 on a singlet (ΨR ). such that exp(X) has determinant one. T = − i 0 2 0 0 0 1 0 0 i 5 0. µ (21. (21.26) normalised in accordance with eq. The generators for SU(3) are given by 0 i 1 T = − 1 2 0 0 i 4 T = − 0 2 1 0 i 7 T = − 0 2 0 1 0 0 0 0 0 0 0 i 0 0 −i i 2 0. Also there the weak interactions act diﬀerently on the left.and righthanded components. Note that det(exp(X)) = exp(Tr(X)). T = − 0 0 2 0 i 0 1 0 i 8 √ 0 −i . the gauge group of the strong interactions. T = − 2 3 0 0 0 1 i 3 0. (21. This doublet gets a hypercharge Y (q L ) = 1 . The dimension of this Liealgebra is easily seen to be N 2 − 1. T = − 0 2 0 0 0 0 1 0 . exp(X)−1 = exp(−X) = exp(X † ) = exp(X)† guarantees that exp(X) is a unitary matrix. we assign q1 to the lefthanded L component of the upquark (also denoted by uL ) and q2 to the lefthanded component of L the downquark (d ). The reason is that quarks are conjectured to always form bound states that are neutral under SU(3). They form complex vectors in the three dimensional deﬁning or fundamental representation of SU(3). Note that we can summarise our assignments of the electric charge by introducing the charge operator in terms of the hypercharge and the socalled isospin operator I3 Qem = ( 1 Y + I3 )e . 1 0 (21. This can be achieved by either taking 97 .
Rather prosaically one associates to the three SU(3) components of the quark ﬁeld the property colour. For a more detailed discussion we refer to Itzykson and Zuber. This is called asymptotic freedom and will only brieﬂy be discussed in the next section. In the ﬁrst case one has a baryon. The electron and neutrino. For this note that both quarks have modulo e an electric charge 1 equal to − 1 e. of which the proton (uud) and the neutron (udd) are examples. follows from the fact that a quark and antiquark can not be separated without creating a quarkantiquark pair from the vacuum. The same holds for a quarkantiquark bound state. In the table below we list the gauge particles of the standard model. The 98 . which still lacks a solid theoretical understanding. The bar over the symbol of ¯ a particle of course denotes the antiparticle. at inﬁnite energy even zero. Gauge particles name γ A W± Z photon gluon Wparticle Zparticle charge 0 0 ±e 0 spin 1 1 1 1 mass 0 0 80 GeV 91 GeV force electromagnetism strong force weak force weak force table 8 The strong interactions do not break parity invariance. This is achieved by combining the quark (antiquark) of the pair created with the antiquark (quark) we try to separate. to make sure that the separated components remain neutral under SU(3). the socalled CabibboKobayashiMaskawa (CKM) mixing with two other families of quarks (the strange and charm quark on the one hand and the bottom and top quark on the other hand) gives in a very subtle way rise to violation of CP . The quarkantiquark bound state is called a meson. whereas both antiquarks have modulo e a charge of 3 e. The mechanism that prevents free quarks to appear is called conﬁnement. into a colourless composite. That nevertheless the theory of the strong interactions is believed to be the correct theory to describe the forces amongst the quarks (and therefore indirectly the nuclear forces) follows from the remarkable property that at high energies the eﬀective coupling constant is small. the eight gluons Aa couple to µ the lefthanded and righthanded components of the quark ﬁelds in the same way. blue and green makes a bound state of three quarks in an antisymmetric wave function. It is not too diﬃcult to show that a bound state of quarks and antiquarks is a singlet under SU(3) if and only if the net colour is white. of ¯ which the pions are examples (e. Similarly. since CP T is conserved). or by combining a quark and an antiquark. Choosing the three basic colours red. However. π + = ud and π − = ud). that is the combination of charge conjugation and parity (equivalent to time reversal T .three quarks in an antisymmetric combination to form a SU(3) singlet. i.e. That the strong interactions really are strong. called leptons. a perturbative expansion is no longer applicable. a colourless combination always has an electric charge that is an integer multiple of the electron charge. It is now also easily veriﬁed that with the particular fractional electric charges assigned to the quarks. combing a quark and an antiquark gives a colourless combination.g. where hence all colours are diﬀerent. Three quarks bound 3 together therefore have zero charge modulo e. Because the coupling constant is strong. in the ﬁrst family are replaced by the muon and its neutrino for the second family and by the tau and associated neutrino for the third family.
with prescribed momenta assigned to the external lines. Loop integrals typically give rise to inﬁnities. as these are strongly related. With a suitably chosen Yukawa coupling the neutrino can be given an arbitrarily small mass.5). generically denoted by an energy Λ or a distance a = 1/Λ.C.5 MeV) 4. In both of these cases there exists a maximal energy (equivalent to a minimal distance). masses and ﬁeld renormalisation constants will depend on this cutoﬀ parameter. as that is what can be measured in experiment. see eq. As an example consider the selfinteracting scalar ﬁeld. for which it is suﬃcient to consider treelevel diagrams that do not contain any loops.5 GeV 106 MeV 0(<0. The table below lists the properties of all the fermions observed in the standard model (the top quark was only discovered in 1994 at Fermilab). as was discussed in sect. In the standard model there is room to add a righthanded partner for the neutrino ﬁeld. Alternatively. Fermion families name d u e νe s c µ νµ b t τ ντ down quark up quark electron neutrino strange quark charm quark muon muonneutrino bottom quark top quark tau tauneutrino table 9 charge e/3 2e/3 e 0 e/3 2e/3 e 0 e/3 2e/3 e 0 spin 1/2 1/2 1/2 1/2 1/2 1/2 1/2 1/2 1/2 1/2 1/2 1/2 mass 10 MeV 5 MeV 511 keV 0(<10 eV) 250 MeV 1. we have only considered the lowest order calculations of cross sections. 9.8 GeV 0(<164 MeV) 22 Loop corrections and renormalisation Up to now. like the coupling constants. amounting to a lattice formulation. 7. (7. How to give a physical deﬁnition of the mass in terms of the full propagator and why ﬁeld renormalisation is necessary was discussed in sect.8 GeV 174 GeV 1. For the renormalisation of the coupling constant it is best to deﬁne the physical coupling constant in terms of a particular scattering process. which can be regularised by considering for example a cutoﬀ in momentum space. with a fourpoint coupling λ 99 . The parameters.experiments described in the introduction (see problem 37) have shown that there are not more than three of these families with massless neutrinos. the physical couplings can be deﬁned in terms of an amputated 1P I npoint function. which couples to none of the gauge ﬁelds. all proportional to an energy scale called µ ≪ Λ. Another possibility of regularising the theory is by discretising spacetime. It is experimentally very hard to measure the mass of the neutrino. For much more on the standard model see in particular the book by J. only upper bounds have been established. Taylor mentioned in the introduction.
the cutoﬀ has to be roughly smaller than 10 TeV. the scale independence can not be recovered by removing the cutoﬀ (i. which is the breaking of a symmetry of the Lagrangian by the quantum corrections. In this case the polarisation is due to the virtual particles that describe the quantum ﬂuctuations (zeropoint ﬂuctuations) of the vacuum. µ. It depends on the parameters of the model. It should be emphasised that the running of the couplings is a manifestation of an anomaly (called the conformal anomaly).(see for example eq. The only way it can be avoided that the renormalised coupling will become inﬁnite at some ﬁnite energy. This can be shown to be equivalent to all 1P I npoint functions to be completely determined as a function of a ﬁnite number of renormalised couplings. It is only in such an instance that quantum ﬁeld theory has predictive power. and is therefore also called the vacuum polarisation. What is not obvious is that. their contributions are inﬁnite in the limit Λ → ∞. Losely speaking the running of this coupling is such that the renormalised coupling increases with increasing energy. In the absence of a mass. ϕ(x) → κϕ(x/κ). this can be at the Planck scale and has little consequence for the theory. Since the vacuum in ﬁeld theories is not really empty. called relevant couplings. to ensure that all regularised couplings stay ﬁnite when the cutoﬀ is moved to inﬁnity. in particular the Higgs mass. The theory is considered renormalisable if we can remove the cutoﬀ by adjusting λ (also called the bare coupling constant) in such a way that at a ﬁxed value µ = µ0 the renormalised coupling λR stays ﬁnite and takes on a prescribed (i.e. If. Renormalisability is therefore a necessary requirement for the theory to be insensitive to what happens at very high energies with a maximal amount of predictive power. is to either take the renormalised coupling equal to zero or to keep the cutoﬀ ﬁnite. It is then obvious that the renormalised coupling constant λR (µ) ≡ λreg (λ(Λ). as was discussed in the context of the Casimir eﬀect in sect. the dependence on the energy scale is caused by quantum corrections. the calculations can be arranged such that nowhere explicit inﬁnities occur. Λ).e. As we can measure the selfcoupling of the Higgs ﬁeld and related quantities to be nontrivial (which is of course crucial for the spontaneous symmetry breaking and giving a mass 100 . When we say that the contributions of the loops diverge. (21. how large the cutoﬀ should be. for the simple ﬁeld theories we have been considering in four dimensions. coinciding at µ0 with the prescribed value λR . By adjusting the bare coupling constants of the theory as a function of the cutoﬀ Λ. measured) value. The standard model falls in this class of theories. The dependence on other coupling constants and the mass parameters is left implicit. Theoretical studies of the last ﬁve years or so have shown that the selfcoupling of the Higgs ﬁeld will most likely vanish if we really take Λ → ∞. however. albeit in a logarithmic way. the computation is not much diﬀerent from calculating eﬀective interactions in a polarised medium. Since the physical coupling constant is computed in terms of the full 1P I fourpoint function.2)). A theory is called renormalisable if only a ﬁnite number of bare coupling constants needs to be adjusted to have all 1P I npoint functions ﬁnite. taking Λ → ∞). the Higgs mass turns out to be in the order of 1 TeV. µ. We deﬁne the physical fourpoint coupling constant in terms of the 1P I fourpoint function with the momenta on the amputated lines set to some particular value proportional to µ (the precise choice is not important for the present discussion). because of the presence of a cutoﬀ. It is obvious that the regularised couplings are not invariant under such a rescaling. the scalar ﬁeld theory with a ϕ4 interaction is at the classical level invariant under scale transformations. where κ is the scale parameter. 2. The energy dependent couplings are called running couplings. Λ) is a function of µ. If the Higgs mass is relatively light. It is clear that this gives a function λreg (λ. we mean that without adjusting the bare coupling constant.
Introducing a momentum cutoﬀ Λ we ﬁnd Σ1 ∼ Λ2 and Σ2 ∼ log Λ to lowest nontrivial order in 1/Λ. For the strong interactions this is one way to understand conﬁnement. When we probe the charged particle at ever smaller distances the eﬀective charge becomes less screened and increases.3) Using eq. at decreasing separations.1) The selfenergy for the scalar ﬁeld in oneloop order splits in two contributions Σ1 and Σ2 from a fermion and a scalar loop (in this order Z ≡ 1). To simplify the integrands we discuss the Feynman trick 1 = ab 1 0 dx 1 (ax + b(1 − x))2 101 . coupled through Yukawa couplings described by the Lagrangian 1 1 L = 2 (∂µ ϕ)2 − 1 m2 ϕ2 − 6 λϕ3 + 2 i Ψ(i) (iγ µ ∂µ − mi )Ψ(i) − gϕ Ψ(1) Ψ(2) + Ψ(2) Ψ(1) m1' . requires us to employ the Feynman rules of table 3 to obtain these expressions. In the following we give a sample calculation that will provide the technical ingredients to perform such calculations and to illustrate some of these issues in a simple setting. 38 and 39 illustrate further ingredients that are pertinent to renormalising ﬁeld theories. problems 2(!). The analogy with a polarised medium is that the virtual particles in the ﬁeld of a charged particle will screen its charge at large distances.4) . (22. Let us end this discussion by noting that a running coupling can of course either increase or decrease at increasing energy. A free quark would carry an inﬁnite energy. Σϕ (q) ≡ Σ1 + Σ2 = − q E &% q + q m2 . The energy of a single quark within a spherical shell would increase without bound with increasing radius.) These integrals are obviously divergent.2) q+k Tr ((k + m1 )(k + / + m2 )) / / q . To the contrary. the scalar sector of the standard model depends in a rather subtle way on what happens at higher energies. the eﬀective charge becomes weaker and weaker and the quarks start to behave as free particles. (9. For the electric charge e. Here the eﬀective charge becomes bigger at larger distances. One says that Σ1 is quadratically and Σ2 logarithmically divergent. its charges show the eﬀect of antiscreening. We will now consider to oneloop order the selfenergy for the scalar ﬁeld ϕ with a mass m. provided the selfenergy is deﬁned through eq. much and much weaker than in nonrenormalisable theories like for the fourfermi interactions. (22. Due to the selfinteractions of a nonAbelian gauge ﬁeld. Also. (11.1). coupled to two ﬂavours of fermions with masses m1 and m2 . This sensitivity to high energies is. (22. this increase is very tiny and the cutoﬀ can be chosen much bigger than the Planck energy. however.k '$ k '$ . The Higgs selfcoupling and the electromagnetic coupling constant e are examples of couplings that increase at high energies.9). This is the asymptotic freedom mentioned in the previous section. It is outside the scope of these lectures to describe the computations necessary to make the above more precise. (Alternatively the Feynman rules of table 2 can be used.to the W and Z particles). 2 + iε)((k + q)2 − m2 + iε) (k − m (22. 2 − m1 + iε)((k + q)2 − m2 + iε) 2 1 d4 k 2 . q+k The numerical expressions for Σ1 and Σ2 are given by Σ1 = − Σ2 ig 2 (2π)4 iλ2 = 2(2π)4 d4 k (k 2 &% q .
β (M) ≡ dn k 2 (k − M 2 + iε)β We can evaluate this integral by performing the socalled Wick rotation. As there are no poles inside the contour of integration. Γ(n + 1) = n! and Γ( 1 ) = π (see problem 2b). The integral over the two quarter circles indicated in the ﬁgure. instead of 4. = (M 2 − iε)β−α−n/2 Γ(β)Γ(n/2) ∞ 0 (22. which satisﬁes Γ(z + 1) = √ zΓ(z). S3 = 4π. dimensions. S4 = 2π 2 . (22.8) (k 2 + M 2 − iε)β We note that the integrand is a purely radial integral and as the surface area of a n dimensional sphere is analytically known (Sn = 2π n/2 /Γ(n/2). In dimensional regularisation the loop integrations are replaced by integrals in n. For Σ2 we have the same assignment.β (M) = i(−1)α−β dn k In. (22. In here Γ(z) is the gamma function.α.6) We will show how to regularise these two integrals in two diﬀerent ways. S2 = 2π. dx (k + (1 − x)q)2 + x(1 − x)q 2 − m2 + iε 2 . (22. k i=1 −1 ani i = Γ( ni ) k k i=1 Γ(ni ) i=1 i 1 0 n xi i −1 dxi k k − i ni δ i=1 xi − 1 ai xi i=1 .α. 1 It is useful to also have the generalisation of the Feynman trick for an arbitrary product of scalar propagators. k 2 = k0 + k 2 . 14 (k 2 )α 2 . will vanish as the radius tends to inﬁnity.which we can apply to the computation of Σ1 by substituting a = (k + q)2 − m2 + iε and 2 b = k 2 − m2 + iε. Consequently we ﬁnd 2 Σ1 = − Σ2 = 4ig 2 (2π)4 d4 k d4 k 1 0 1 0 dx iλ2 2(2π)4 2 (k + (1 − x)q)2 + x(1 − x)q 2 − xm2 − (1 − x)m2 + iε 1 k 2 + k · q + m1 m2 1 2 . First we use dimensional regularisation introduced by ’t Hooft and Veltman (see problem 2) and then discuss PauliVillars regularisation. where we replace the integral over Rek0 by an integration over Imk0 .5) which is proven by induction. (22.· · ·). (22.10) .α. The momentum integrations are always of the form (k 2 )α . we obtain In. k) ≡ Γ(n + k) 102 z n−1 dz (z + 1)k+n .g. we ﬁnd Imk0 6 8 T Eq c Eq c T Rek0 ﬁg.7) In.9) We used the integral representation of the beta function Γ(n)Γ(k) = B(n. with in addition m1 = m2 = m. e.β (M) = i(−1)α−β r 2α 2π n/2 r n−1 dr 2 Γ(n/2) (r + M 2 − iε)β i(−1)α−β π n/2 Γ(α + n/2)Γ(β − α − n/2) .
0. Substituting the expressions for In.0.9). The divergent part is now fully contained in Γ(2 − n/2).2 (Mx. (22.11) dx (k 2 2 and Mx.q 1 − iε)2−n/2 .q − iε)2−n/2 (Mx.q + iε)2 .2 (Mx.β (m) Σ1 = − Σ2 dx ˆ ˆ In.q (Mx.α. = 1 4−n (4 − n) 2 (22.(22. (22.q dx 2 ˆ x.q = m2 − x(1 − x)q 2 .q − iε)1−n/2 . (22. (22. 4−n (−1) (−1) Σ2 = λ2 8(2π)2 1 0 2 dx log(π[Mx.Shifting the integration variable k → k − (1 − x)q we ﬁnd Σ1 = − Σ2 = where ˆ2 Mx.2 (Mx.q ) .q ) .1. . because Γ(2 − n/2) = g2 = (2π)2 = g2 (2π)2 where γ = 0. we ﬁnd Σ1 = 4g π (2π)4 2 n/2 0 1 dx 1 Γ(2 − n/2) Σ2 = − λ2 π n/2 2(2π)4 0 dx 2 (Mx.16) 2 (Mx. (22.α.q + iε)2 1 2 − Mx. We expand Σi around n = 4 Σ1 2 − γ π (n−4)/2 4−n 2 −γ 4−n 1 0 1 0 1 Γ (1 + 2 (4 − n)) 2 − γ + O(4 − n) . (22.q − iε]) + 2 (4−n)Mx.q − iε]) + γ − 2 Σ (0) ≡ 2 +Σ2 +O(4 − n) .q − iε)(4−n)/2 (M 1 1 ˆ2 ˆ2 1 − 2 (4 − n) log(π[Mx. 2 (22.q ≡ Σ1 (0) +Σ1 +O(4 − n) .q .q Γ(2 − n/2) − iε)2−n/2 Γ(2) Γ(1 + n/2)Γ(1 − n/2) (m1 m2 − x(1 − x)q ) − .18) 4−n 4−n 103 .12) which allows us to express Σi in terms of the integrals In.14) This can be further simpliﬁed using n Γ(1 + n/2)Γ(1 − n/2) = Γ(2 − n/2)Γ(n/2) 2−n 1 0 .17) dx 2 m1 m2 − 3x(1 − x)q 2 + 2xm2 + 2(1 − x)m2 × 1 1 ˆ2 m1 m2 − x(1 − x)q 2 + (2 + 2 (4 − n))Mx.q = xm2 + (1 − x)m2 − x(1 − x)q 2 1 2 4ig 2 (2π)4 iλ2 = 2(2π)4 1 0 1 0 4ig 2 (2π)4 dn k dn k 1 0 1 0 dx iλ2 2(2π)4 k 2 − (1 − 2x)k · q + m1 m2 + x(x − 1)q 2 ˆ2 (k 2 − Mx.β (M) from eq. ˆ 2 − iε)2−n/2 Γ(2) ˆ 2 − iε)1−n/2 Γ(2)Γ(n/2) (Mx.13) dx In.15) such that Σ1 4g 2π n/2 Γ(2 − n/2) = (2π)4 λ2 π n/2 Γ(2 − n/2) 2(2π)4 dx 1 (m Σ2 = − 0 dx 2 n/(2 − n) 1 m2 − x(1 − x)q ) − ˆ2 ˆ2 (Mx.57721 · · · is the Euler constant.q ) + (m1 m2 − x(1 − x)q 2 )In.
We have split the result for Σi in a pole term with residue Σi (0) and a ﬁnite part Σi for n → 4.21) Note that as long as we stay away from n = 4 everything is well deﬁned. For the theory in eq.q − log(π[Mx.19) We now note that the pole terms are of the same form as the treelevel expressions obtained from the following extra term in the Lagrangian 1 ∆L = 2 a(∂µ ϕ)2 − 1 bϕ2 2 . 1 '$ d ' d m2 c m T 2 &% E m d d m2 + b ≡ 1+a .e with a ﬁnite number of spacetime derivatives) in the Lagrangian. Rather than computing at n = 4 physical processes to ﬁx the renormalised couplings. (22. Σ2 Σ1 (0) (−1) 1 λ2 2 dx log(π[Mx. To ﬁnd the bare mass and the ﬁeld renormalisation (for the bare λ coupling we should have considered the 1P I threepoint function with three ϕ external lines) we write to oneloop order LB = L + ∆L ≡ (∂µ ϕB ) − 1 2 g2 λ2 1 − 2 (m1 m2 + m2 + m2 ) 1 2 2 (4π) 2π 4−n .1). we have started with renormalised couplings and determined how they have to depend on the bare couplings so as to cancel any inﬁnities that might arise as n → 4.22) 2 1 2 m2 ϕ2 B B . 2 0 8(2π) λ2 g2 (−1) 1 (m1 m2 + m2 + m2 − 2 q 2 ) . We have taken here a slightly diﬀerent approach for renormalising the theory. b= (−1) (0) (0) (22. m ﬁg.23) Often it can be determined by powercounting (of momenta) which diagrams give rise to divergencies for n → 4 or Λ → ∞. m2 B (22. ϕB ≡ Zϕ ϕ = √ 1 + aϕ . (22.2 Note that q 2 = q0 − q 2 and that the coupling constant λ for the scalar threepoint coupling (−1) has the dimension of mass. see the Feynman diagram below. including a and b. in a loop expansion the second procedure is more natural. Σ1 (0) = g2 4π 2 1 0 ˆ2 ˆ2 dx Mx. (22. from which we can solve for a and b in terms of Σi a=− 1 g2 24−n (2π) . Σ2 = − = 1 2 2π 2 (4π)2 = . It is clear that these two procedures are equivalent. The limit n → 4 is to be taken after we have expressed everything in terms of the renormalised coupling constants and masses. (22.20) This means that we can choose a and b so as to precisely cancel the pole terms.q − iε]) + γ . The inﬁnities correspond to local counter terms (i. power counting easily shows that the ϕ fourpoint function is logarithmically divergent at oneloop order.q − iε]) + γ × m1 m2 + 3x(x − 1)q 2 + 2xm2 + 2(1 − x)m2 1 2 . For the physical interpretation the ﬁrst procedure (due to Wilson) is more transparent. 15 1 104 . To lowest order we therefore have Σϕ (L + ∆L) = Σϕ (L) + b − aq 2 = Σ1 + Σ2 + O(n − 4) .
3. PauliVillars regularisation is aimed at having a regulator that preserves the Lorentz invariance. In that case (m1 = m2 ) .27) which is indeed negative for q 2 > (m1 + m2 )2 . With the present prescription we can treat bosons and fermions on the same footing). where eq. g.24) We will show that ImΣ1 (q 2 = m2 ) = 0 if and only if q 2 > (m1 + m2 )2 .25) Let us ﬁrst consider the simplest case of equal fermion masses. see problem 39. 2. one can shift m2 over Mℓ2 . As we have seen in section 11. the mass of these ghost ﬁelds is shifted over Mℓ with respect to the original (“parent”) ﬁeld (alternatively. it is not diﬃcult (0) to show that on the mass shell (q 2 = m2 ) Σ2 is real Σ2 (q 2 = m2 ) = (0) λ2 8(2π)2 (0) 1 0 3 dx log[(x − 1 )2 + 4 ] + γ + log(m2 π) ∈ IR . 4. m1 = m2 .q x ∈ [0. (22. m2 . so as to adjust its bare coupling to depend in the proper way on the cutoﬀ. q 2 < (m1 − m2 )2 . By deﬁning e0 = 1 and M0 = 0. if the original ﬁeld is a boson. The only way Σ1 can develop an imaginary part is when the argument of the logarithm in eq. This therefore proves that the kinematically determined threshold coincides with the threshold for ImΣ(q) = 0. Furthermore. called the threshold (0) for decay.1) the term −λ4 ϕ4 /4!. (22. For the general case of unequal 1 2 fermion masses. 1) . To regularise Σϕ by PauliVillars’ method. It is clear that the scalar particle itself can not decay in two scalar particles. The lattice regularisation also can be arranged to preserve the gauge invariance. −1. A major advantage of dimensional regularisation is that it preserves the Lorentz and gauge invariances. the minimum is obtained for x = 1 (1 + (m2 − m1 )/q 2 ). the index ℓ = 0 describes the original ﬁelds of the model. 1] if (m2 − m2 )/q 2  > 1. to ensure that we can remove it. which can be used to rule out the 1 2 other region. λ and λ4 . (22. 105 (22. The threshold is therefore determined by ˆ2 min{Mx. (22. we choose eℓ = (1. −2) and bℓ = (0.19) becomes negative. using again the computation of Σϕ (q) to oneloop order as an illustration. A momentum cutoﬀ breaks both the Lorentz and gauge invariance.1). Σϕ (q) should have a nonvanishing imaginary part if the scalar particle is unstable. We will describe it for the Lagrangian of eq. The relevant parameters are m. For each of the original ﬁelds ϕ and Ψ(i) one adds extra (ghost) ﬁelds. m1 . Indeed.28) . m1 . (22. After some algebra 2 2 we ﬁnd ˆ2 min{Mx.27) is negative. The value of x where this minimum is attained does not lie in the interval [0. This means that a loop of these ghost ﬁelds gets an additional factor eℓ . but when m1 + m2 < m it could decay in two fermions. 1]} = min{xm2 + (1 − x)m2 − x(1 − x)q 2 x ∈ [0.26) and the threshold is determined by q 2 > 4m2 = (m1 + m2 )2 . It can be shown that after adding to the Lagrangian in eq. m2 .q x ∈ IR} = (m1 − m2 )2 q2 1− 4 q2 (m1 + m2 )2 −1 q2 . as was assumed in section 11. 1]} = m2 − 4 q 2 1 (22. Furthermore.We therefore need to introduce an independent parameter for the ϕ fourpoint coupling. 1 2 1 ˆ2 min{Mx. the theory becomes renormalisable to all orders in the loop expansion. it is a local regulator. with either the same (eℓ > 0) or opposite (eℓ < 0) statistics. 1]} < 0 . but locality and Lorentz invariance are only valid at distances much bigger than the lattice spacing a. (22.q x ∈ [0. 2 (22. We deﬁne furthermore Mℓ ≡ bℓ Λ with Λ ≫ m.
32) . m2 + bℓ Λ. All Λ dependent terms can be absorbed in a redeﬁnition of Zϕ and the mass of the scalar ﬁeld. such that ΣPV (q) = − ϕ 3g 2 (2π)2 1 0 dx log m2 − x(1 − x)q 2 1 2 m1 − x(1 − x)q 2 λ2 log m2 − x(1 − x)q 2 + a5 + a6 q 2 + 2 8(2π) 106 . (22. (22. (22. It is still convenient to evaluate the integrals in n. q) + Σ2 (m + bℓ Λ. the masses of the ghost particles. now play the role of a momentum cutoﬀ. (22. but can be calculated explicitly with some eﬀort. Indeed. all proportional to Λ. as the Lagrangian will at that energy scale no longer describe a physical theory. We will see that eq. m2 + bℓ Λ. To keep the following computation transparent we take m1 = m2 PV Σϕ (q) g2 = (2π)2 1 0 2 dx ℓ 1 0 1 0 eℓ 1−3γ −3 log π[(m1 +bℓ Λ)2 −x(1−x)q 2 ] ℓ (m1 +bℓ Λ)2 −x(1−x)q 2 λ + 8(2π)2 =− 3g (2π)2 λ 8(2π)2 2 2 2 dx eℓ log π[(m+bℓ Λ)2 −x(1−x)q 2 ] +γ eℓ log (m1 + bℓ Λ)2 − x(1 − x)q 2 ℓ dx ℓ 1 0 (m1 + bℓ Λ)2 − x(1 − x)q 2 + dx eℓ log (m + bℓ Λ)2 − x(1 − x)q 2 2 m1 −x(1−x)q 2 1 3g 2 1 λ = dx log m2 −x(1−x)q 2 − dx log m2 −x(1−x)q 2 1 8(2π)2 0 (2π)2 0 +a1 Λ2 + a2 Λ + a3 log Λ + a4 q 2 log Λ + a5 + a6 q 2 + O(1/Λ) . ℓ ℓ The ﬁnite result that remains (replacing Σ(−1) in the equation above by Σ(0) ) is nevertheless still dependent on Λ. such that these ghost ﬁelds can be described in terms of either grassmann or bosonic variables by taking eℓ  ﬁelds. The weights are chosen such that the momentum integrals can all be performed.19) 3 eℓ Σ1 ℓ=0 3 (−1) (m1 + bℓ Λ. rather than in 4 dimensions. q) + Σ2 (−1) (m + bℓ Λ. q) = (22.31) The precise values of the coeﬃcients ai are not very important.in other words the scalar and two fermion ﬁelds each have three ghost ﬁelds associated to them but with nonstandard weights for the loops. q) ℓ=0 .29) in an obvious notation. We could stick to standard weights. having either the same (eℓ > 0) or reversed (eℓ < 0) statistics with respect to the original (“parent”) ﬁeld.28) guarantees that the terms proportional to (4 − n)−1 exactly cancel. using eq. (22.30) = eℓ ℓ=0 g2q2 λ2 g2 (m1 + bℓ Λ)(m2 + bℓ Λ) + (m1 + bℓ Λ)2 + (m2 + bℓ Λ)2 − − 2π 2 (2π)2 (4π)2 eℓ + ℓ g 2 (m1 m2 +m2 +m2 − 1 q 2 )− 1 λ2 1 2 2 8 2π 2 3(m1 +m2 )g 2 Λ 2π 2 eℓ bℓ + ℓ 3g 2 2 Λ 2π 2 eℓ b2 = 0 . Nevertheless. It is straightforward to give the selfenergy including the contribution of the ghost ﬁelds ΣPV (q) = ϕ 3 eℓ Σ1 (m1 + bℓ Λ.
g.37) eB ≡ µ 2 ε Ze e = µ 2 ε 1 + 1 1 e2 · +··· e . the renormalisation of gauge theories to oneloop order in dimensional regularisation. (22. This is not B B an accident.39) To all orders in the loop expansion the ﬁeld A and the gauge ﬁxing parameter α/e remain free of renormalisations. 6π ε (22. By absorbing the 107 .36) 1 If we deﬁne (as is customary) ε ≡ 4 − n. e2 1 αB ≡ Zα α = 1 + 2 · + · · · α .Note that in dimensional regularisation (DR) we found the result Σ1 + Σ2 . The same holds for nonAbelian gauge theories. 8π 2 ε ΨB ≡ µ −1ε 2 −1ε 2 1 2 Ψ. which implies that h n L/µ is dimensionless. such an ambiguity of course cannot arise. (22. [eB ] = µ2− 2 n . e. 2 ε 12π We note that to oneloop order Aµ ≡ eB Aµ and αB /e2 are ﬁnite for n → 4. The bare Lagrangian is given by 1 1 L = − 4 (∂ µ Aν − ∂ ν Aµ )2 − 2 αB (∂µ Aµ )2 + ΨB (iγ µ ∂µ − mB )ΨB + eB Aµ ΨB γµ ΨB B B B B . [ΨB ] = [ΨB ] = µ 2 n− 2 . 1 [αB ] = 1 . depends on the regularisation used. one ﬁnds (for details see Itzykson and Zuber. 2 8(2π) (2π)2 b2 = g 2 m2 (3γ + 3 log π − 1) 1 . 2 Ze = Zα = 1/ZA . D µ ≡ ∂ µ − iAµ 2 . or DR Σϕ (q) = − 1 3g 2 1 dx log π m2 − x(1 − x)q 2 + γ − 3 1 (2π)2 0 1 λ2 dx log π m2 − x(1 − x)q 2 + γ + 8(2π)2 0 (0) (0) m2 − x(1 − x)q 2 1 . without a detailed derivation. 6(2π)2 (22. but the consequence of a socalled Ward identity. µ [AB ] = µ 2 n−1 .) Aµ B ≡ µ −1ε 2 ZA A = µ ZΨ Ψ = µ µ 1 −2ε e2 1 1− 2 · +··· 6π ε e2 1 1− 2 · +··· 8π ε 1 1 2 Aµ . however. mB ≡ Zm m = 1 − 3e2 1 · +··· m . (22. (22.38) It is therefore sometimes much more convenient to use L=− 1 αB (∂ µ Aν − ∂ ν Aµ )2 − 2 (∂µ Aµ )2 + ΨB (iγ µ Dµ − mB )ΨB 2 4eB 2eB µ . We will now discuss. In that case there is a unique relation between the bare and renormalised parameters. 1 1 [mB ] = µ .35) In n dimensions we still want the action to be dimensionless (¯ = 1). This relation. If we deﬁne the renormalised coupling in terms of some physical scattering process. where ϕ ϕ b1 = λ2 (γ + log π) g 2 m2 (3γ + 3 log π − 1) 1 − . From this we derive the dimensions of the ﬁelds and the parameters in n dimensions.33) However. (22. They use slightly diﬀerent notations. sections 71 and 84. the diﬀerence ΣPV (q) − ΣDR (q) = a5 − b1 + (a6 − b2 )q 2 . 20) relates diﬀerent Z factors.34) can be absorbed in a ﬁnite redeﬁnition of the mass and of Zϕ . which as a consequence of the gauge symmetry (through the BRS invariance mentioned at the end of sect.
(22. we see that the one loop corrections to the bare coupling constant diﬀers in sign from the equivalent expression for the Abelian case.40) where the gauge ﬁeld and the gauge ﬁxing parameter receive no renormalisations. 2 24π (22.44) whereas for QED (coupled to nf ﬂavours of fermions) β(g) = − g 2(µ) = β(e) = nf e3 + O(e5 ) . measured in 1947. The ﬁeld strength Fµν and covariant derivative Dµ are now given by (compare eqs. In problem 39 it will be shown that to oneloop order the photon vacuum polarisation is given by (compare eqs. (22.43) where the derivative is taken at ﬁxed ε and gB (g ≡ gR ).34) gB = µ 2 ε Zg g = µ 2 ε 1 − 1 1 (11N − 2nf )g 2 1 · +··· g 48π 2 ε .41)) Dµ = ∂µ + Aµ . It is perhaps appropriate to end these lecture notes with as classic an experimental test of renormalisation eﬀects in ﬁeld theory as the one for the Casimir energy in section 2. The running of the coupling is expressed in terms of the socalled betafunction β(g) ≡ µ ∂g(gB (ε). It is the selfinteractions of the nonAbelian gauge ﬁelds that are responsible for the asymptotic freedom of its running coupling constant. 2 48π 1 For other regularisations the computation of the running coupling constant is similar.46) γν 108 . In the Landau gauge. which is the very small energy splitting of the 2S 1 and 2P 1 orbitals in hydrogen atoms. receiving a contribution from vacuum polarisation 2 2 eﬀects (for a discussion of the other contributions see section 732 of Itzykson and Zuber). in other words they are already the renormalised ﬁeld and gauge ﬁxing parameter. Aν ] . For nonAbelian gauge theories one ﬁnds (µ0 is an integration constant) (11N − 2nf ) log(µ/µ0) + O(g 2 (µ)) . ε) ∂µ . (22. 6π 2 (22. (22. 12π 2 1 e2 (µ) =− nf log(µ/µ0 ) + O(e2 (µ)) .charge q (called coupling constant g ≡ q henceforth) in the gauge ﬁeld. except that ε is replaced roughly by 1/ log(Λ).42) As long as the number of fermion ﬂavours is small enough.23). To the order displayed.45) (11N − 2nf )g 3 + O(g 5 ) . the betafunctions do not depend on the regularisation scheme.41) If the gauge group is SU(N) and there are nf ﬂavours of fermions the renormalisation of the coupling constant is given by (see Itzykson and Zuber sect. we can drop the Λ(α) factors) (α) Σµν (q) = −Λµβ (q)(q 2 g βγ − q β q γ )ω(q 2)Λ(α) (q) . α → ∞. It concerns the Lamb shift. Fµν = ∂µ Aν − ∂ν Aµ + [Aµ . the Lagrangian can be expressed as L= αB 1 2 Tr(Fµν ) + 2 Tr(∂µ Aµ )2 + ΨB (iγ µ Dµ − mB )ΨB 2 2gB gB . µ. (22. (18.35) and (18.22) and (16. 12. (16.
Consequently. only aﬀects the wave functions that do not vanish in the origin.50) where n is the radial quantum number and αe is the ﬁnestructure constant. In section 711 of Itzykson and Zuber it is shown that c = e2 /(60π 2 m2 ). Λ → ∞ .49) The extra deltafunction interaction.From the results of problem 39. only the energy of the S orbitals will be shifted by this correction −e2 d3 q eiq·r = −e2 2 (1 + ω(−q 2 )) q d3 q ∆E(nS 1 ) = − 2 5 4mαe 15πn3 . q 2 = −q 2 and the photon exchange can be accurately described by the Coulomb potential − e2 = −e2 4πr d3 q eiq·r q2 . where ω is computed with PauliVillars regularisation.48) which due to the vacuum polarisation is replaced by 1 e2 e4 + c + · · · eiq·r = − − δ3 (r) . as is relevant for the hydrogen atom. αe = e2 4π . as the combination a log(Λ/m) + b can be absorbed in the ﬁeld renormalisation (this means that a can be read oﬀ from ZA given above). (22. (22. In the static limit. that arises from the vacuum ﬂuctuations. q2 4πr 60m2 π 2 (22.47) The precise values of the coeﬃcients a and b are not very important. 109 . it can be deduced that (m is the electron mass) ω(q 2 ) = a log(Λ/m) + b + cq 2 for q 2 → 0. (22.