Dict Dvi

A COURSE IN FIELD THEORY
Pierre van Baal

Instituut-Lorentz for Theoreti al Physi s,
University of Leiden, P.O.Box 9506,
2300 RA Leiden, The Netherlands.
Fall, 1998
Copyright P. van Baal

No part of the notes and problems should be
reprodu ed or made ele troni ally available
by others without permission of the author.
Table of ontents
0.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
Introdu tion
Motivation
Quantisation of elds
Euler-Lagrange equations
Tree-level diagrams
Hamiltonian perturbation theory
Path integrals in quantum me hani s
Path integrals in eld theory
Perturbative expansion for eld theory
The s attering matrix
Cross se tions
De ay rates
The Dira equation
Plane wave solutions of the Dira equation
The Dira Hamiltonian
Path integrals for fermions
Feynman rules for ve tor elds
Quantum Ele trodynami s - QED
Non-Abelian gauge theories
The Higgs me hanism
Gauge xing and ghosts
The Standard Model
Loop orre tions 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 Introdu tion
Field theory is most su essful in des ribing the pro ess of s attering of parti les in the ontext of the Standard Model, and in parti ular in the Ele tromagneti and Weak Intera tions.
From 1989, the Large Ele tron Positron (LEP) ollider is operational. In a ring of 27 km in
diameter, ele trons and positrons were a elerated in opposite dire tions to energies of approximately 45 GeV. This energy is equivalent to half the mass (expressed as energy through
E = m 2 ) of the neutral Z o ve tor boson mass, whi h mediates part of the weak intera tions.
The Z o parti le an thus be reated in ele tron-positron annihilation, at the regions where
the ele tron and positron beams interse t. As a Z o an be formed out of an ele tron and
its antiparti le, the positron, it an also de ay into these parti les. Likewise it an de ay
in a muon-antimuon pair and other ombinations (like hadrons). The ross-se tion for the
formation of Z o parti les shows a resonan e peak around the energy where the Z o parti le
an be formed. The width of this peak is a measure for the probability of the de ay of this
parti le. By the time you have worked yourself through this ourse, you should be able to
understand how to al ulate this ross-se tion, whi h in a good approximation is given by
4e2 E 2 =27
= 2
;
(0.1)
(E MZ2 o )2 + MZ2 o 2Z o
expressed in units where h = = 1. e = 4e 1=137:037, is the ne-stru ture onstant,
E is twi e the beam energy, MZ o the mass and Z o the de ay rate (or width) of the Z o
ve tor boson. The latter gets a ontribution from all parti les in whi h the Z o an de ay, in
parti ular from the de ay in a neutrino and antineutrino of the three known types (ele tron,
muon and tau neutrinos). Any other unknown neutrino type (assuming their mass to be
smaller than half the Z o mass) would ontribute likewise. Neutrinos are very hard to dete t
dire tly, as they have no harge and only intera t through the weak intera tions (and gravity)
with other matter. With the data obtained from the LEP ollider (the gure is from the
ALEPH ollaboration) one has been able to establish that there are no unknown types of
light neutrinos, i.e. N = 3, whi h has important onsequen es (also for osmology).
2
35
30
25
20
15
10
5
0
1.05
1
0.95
88
89
90
91
92
93
94
95
g. 1
The main aim of this eld theory ourse is to give the student a working knowledge
and understanding of the theory of parti les and elds, with a des ription of the Standard
Model towards the end. We feel that an essential ingredient of any eld theory ourse has
1
to be to tea h the student how Feynman rules are derived from rst prin iples. With the
path integral approa h 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 ourse.
As Julius Wess put it during his ourse as a Lorentz professor at our Institute \you won't
be ome a good pianist by listening to good on erts".
These le ture notes re e t the eld theory ourses I taught in the fall of 1992 at Utre ht
and in the fall of 1993, 1994 and 1996 at Leiden. I owe mu h to my tea hers in this eld,
Martinus Veltman and Gerard 't Hooft. As I taught in Utre ht from 't Hooft's le ture notes
\Inleiding in de gequantiseerde veldentheorie" (Utre ht, 1990) it is inevitable that there is
some overlap. In Leiden I spent roughly 25% longer in front of the lassroom (3 le tures of
45 minutes ea h for 14 weeks), whi h allowed me to spend more time and detail on ertain
aspe ts. The set of problems, 40 in total, were initially ompiled by Karel-Jan S houtens
with some additions by myself. In their present form, they were edited by Jeroen Snippe.
Of the many books on eld theory that exist by now, I re ommend the student to onsider
using \Quantum Field Theory" by C. Itzykson and J.-B. Zuber (M Graw-Hill, New York,
1980) in addition to these le ture notes, be ause it oers material substantially beyond the
ontent of these notes. I will follow to a large extent their onventions. I also re ommend
\Diagrammati a: The path to Feynman diagrams", by M. Veltman (Cambridge University
Press, 1994), for its unique style. The dis ussion on unitarity is very informative and it has
an appendix omparing dierent onventions. For more emphasis on the phenomenologi al
aspe ts of eld theory, whi h are as important as the theoreti al aspe ts (a point Veltman
often emphasised for efully) I an re ommend \Field Theory in Parti le Physi s" by B. de
Wit and J. Smith (North-Holland, Amsterdam, 1986). For path integrals, whi h form a
ru ial ingredient of these le tures, the book \Quantum me hani s and path integrals" by
R.P. Feynman and A.R. Hibbs (M Graw-Hill, New York, 1978) is a must. Finally, for an
introdu tion to the Standard Model, useful towards the end of this ourse, the book \Gauge
theories of weak intera tions", by J.C. Taylor (Cambridge Univ.Press, 1976), is very valuable.
1 Motivation
Field theory is the ultimate onsequen e of the attempts to re on ile the prin iples of relativisti invarian e with those of quantum me hani s. It is not too di ult, with a lot of
hindsight, to understand why a eld needs to be introdu ed. Although this is not an attempt
to do justi e to history - and perhaps one should spare the student the long struggle to arrive
at a onsistent formulation, whi h most likely has not ompletely rystalised yet either - but
the traditional approa h of introdu ing the on ept is not very inspiring and most often
la ks physi al motivation. In the following dis ussion I was inspired by \Relativisti Quantum Theory" from V.B. Berestetskii, E.M. Lifshitz and L.P. Pitaevskii (Pergamon Press,
Oxford, 1971). The argument goes ba k to L.D. Landau and R.E. Peierls (1930).
An important onsequen e of relativisti invarian e is that no information should propagate at a speed greater than that of light. Information an only propagate inside the future
light one. Consider the S hrodinger equation
t

x
(~x; t)

ih
= H (~x; t) :
(1:1)
g. 2
t
N
Relativisti invarian e should require that (~x; t) = 0 for all (~x; t) outside the light one of
the support N = f~xj (~x; 0) 6= 0g of the wave fun tion at t = 0.
2
Naturally, a rst requirement should be that the S hrodinger equation itself is relativisti ally invariant. For ordinary quantum me hani s, formulated in terms of a potential
p~2
+ V (~x) ;
(1.1)
H=
2m
this is learly not the ase. Using the relation E 2 = ~p2 2 + m2 4 the most obvious attempt
for a relativisti ally invariant wave equation would be the Klein-Gordon equation
2
2 (~x; t)
2 2 (~x; t) + m2 4 (~x; t) :
h 2
=
h

(1.2)
t2
~x2
However, for this equation the usual denition of probability density is not onserved
Z
t d3~x (~x; t) (~x; t) 6= 0 :
(1.3)
As this is a onsequen e of the fa t that the equation is se ond order in time, this an be
easily remedied, it seems, by taking the \square root" of the Klein-Gordon equation
v
u
(~x; t) u
ih
=t
t
!
2
2
2
2
4
h 2 + m (~x; t)
~x
(1.4)
We shall show that this, however, violates the prin iple of ausality, i.e. the wave fun tion
propagates outside of its light one, whi h is una eptable. Nevertheless, we will learn something important from that omputation, namely that negative energies seem unavoidable
when trying to lo alise wave fun tions within the light one of N . But rst we will provide
a simple heuristi argument based on the un ertainty relation.
From the un ertainty prin iple xp > h =2 and the bound on the speed involved in any
measurement of the position, it follows that pre ision of a measurement of the momentum
is limited by the available time tp > h = . Only for a free parti le, where momentum is
onserved, su h a measurement would be possible, but in that ase, of ourse, the position is
ompletely undetermined, onsistent with the plane wave des ription of su h a free parti le
(the light one of N would in that ase indeed give us no onstraint). More instru tive is
to look at how a urately we an determine the position of a parti le. As the momentum is
bounded by the (positive) energy (p E= ) and as the maximal hange in the momentum
is of the order of p itself, we nd that x > h =p h =E , whi h oin ides with the limit set
by the De Broglie wavelength.
If we take this serious, that is position an in prin iple not be measured with arbitrary
a ura y, the notion of a wave fun tion looses its meaning. On the other hand, if we would
like to lo alise the parti le more a urate than within its De Broglie wavelength, it seems
to require an un ertainty in momentum that an only be a hieved by allowing for negative
energy states. But negative energy states will be interpreted as antiparti les, and on e
antiparti les are introdu ed, whi h an annihilate with parti les, parti le number is no longer
onserved and we likewise loose the notion of position of a parti le. Only a free parti le, as
a plane wave, seems to be ompatible with relativisti invarian e.
We will now verify by dire t omputation that lo alising the wave fun tion within the light
one will indeed require negative energy states. We onsider rst the positive square root
of the Klein-Gordon equation and solve the S hrodinger equation for the initial ondition
(~x; 0) = 3 (~x). From this we an solve any initial ondition by onvolution. As the
S hrodinger equation is rst order in time, the initial ondition uniquely xes the wave
3
fun tion for all later times and there will be a unique answer to the question whether the
wave fun tion vanishes outside the light one (i.e. for t > j~xj). Problem 1 asks you to
investigate this in the simpler ase of one, instead of three, spatial dimensions. For the
latter we simply give the result here, using the fa t that in Fourier spa e the solution is
trivial. Computing (~x; t) thus requires just some skills in performing Fourier integrals.
Z
d3 ~p i~p~x=h itpp~ 2 2 +m2 4 =h
(~x; t) =
e
e
(2h )3
Z
p2 dp sin()d ipr os()=h itpp2 2 +m2 4 =h
=
e
e
(2 )2h 3
p
1 Z
it p2 2 +m2 4 =h
=
pdp
sin(
pr=
h

)
e
2 2 rh 2
p
i 2 Z 1
os(pr=h )
it p2 2 +m2 4 =h :
p
(1.5)
e
=
dp
2 2 r rt 0
p2 2 + m2 4
We now introdu e
p = m sinh(u) ; m r=h = z osh(v ) ; m 2 t=h = z sinh(v ) ;
z 2 = m2 2 (r2 2 t2 )=h 2 ;
(1.6)
su h that (the last identity simply being the denition of the modied Bessel fun tion Ko )
i 2 Z 1
(~x; t) =
du os(z sinh(u) osh(v ))e iz sinh(v) osh(u)
4 2 r rt 1

i 2 Z 1 iz sinh(u+v)
iz sinh(u v)
du
e
+
e
=
8 2 r rt 1
i 2
i 2 Z 1
du
os(
z
sinh(
u
))

K (z ) :
(1.7)
=
2 2 r rt 0
2 2 r rt o
Outside of the light one, z is real (r2 > 2 t2 ) and (~x; t) is purely imaginary. It de ays
exponentially, but does not vanish! Inside the light one we nd by analyti ontinuation (see
e.g. appendix C of \Relativisti Quantum Fields" by J.D. Bjorken and S.D. Drell (M Graw
Hill, New York, 1965)) the following expli it expression

p
p
1 2
(~x; t) =
iYo (m 2 t2 r2 =h ) sign(t)Jo (m 2 t2 r2 =h ) ; r2 < 2 t2 :
4r rt
(1.8)
If we want to insist on lo ality, i.e. (~x; t) = 0 for j~xj > t and want to stay as lose as
possible to the solutions of the S hrodinger equation, we ould take the real part of as the
wave fun tion. It satises the Klein-Gordon equation, but not its positive square root.
is a solution of the negative square root of the Klein-Gordon equation, and orresponds to
a negative energy solution. Apparently, lo alisation is only possible if we allow for negative
energy solutions.
2 Quantisation of elds
As position is no longer a quantum observable, but free parti les do not seem to be in
ontradi tion with relativisti invarian e, we an try to introdu e su h a free parti le as a
quantum observable. This observable is hen e des ribed by a plane wave
'~ (~x; t) = e i(ko t ~x~k)=h ;
(2.1)
k
whi h satises the Klein-Gordon equation

2
2 '(~x; t)
2 2 '(~x; t) + m2 4 '(~x; t) ;
h 2
=
h

t2
~x2
(2.2)
where k0 = 2~k2 + m2 4 is the energy of the free parti le. By superposition of these plane
waves we an make a superposition of free parti les, whi h is therefore des ribed by a eld
'(~x; t) = (2h )
3
2
d3~k '~(~k; t)ei~k~x=h :
(2.3)
It satises the Klein-Gordon equation if the Fourier omponents '~(~k; t) satisfy the harmoni
equation
2 '~(~k; t)
2
= ( 2~k2 + m2 4 )'~(~k; t) ko2 (~k)'~(~k; t) :
(2.4)
h
2
t
Its solutions split in positive and negative frequen y omponents
'~(~k; t) = '~+ (~k)e iko t=h + '~ (~k)eikot=h
(2.5)
The wave fun tion, or rather the wave fun tional ('), des ribes the distribution over the
various free parti le states. The basi dynami al variables are '~(~k). These play the role the
oordinates used to play in ordinary quantum me hani s and will require quantisation. As
they satisfy a simple harmoni equation in time, it is natural to quantise them as harmoni
os illators. The Hamiltonian is then simply the sum of the harmoni os illator Hamiltonian
for ea h ~k, with frequen y ! (~k) k0 (~k)=h .
ih
X
(')
= H (') = H(~k) (') ;
t
~k
H(~k) = j~ (~k)j2 + !(~k)2j'~(~k)j2 ; ~(~k) hi ~ :

'~(k)
(2.6)
1
2
1
2
In a nite volume with periodi boundary onditions, the integral over the momenta is
repla ed by a sum as the momenta are in that ase dis rete, ~k = 2~nh =L; ~n 2 ZZ3 . Like for
the harmoni os illator, we an introdu e reation and annihilation operators

1
! (~k)'~(~k) + i~ (~k) ;
2h! (~k)

1
! (~k)'~ (~k) i~ (~k) ;
ay (~k) = q
2h! (~k)
a(~k) = q
(2.7)
and express the eld operator (the equivalent of the oordinates) in terms of these reation
and annihilation operators. To give the eld operator its time dependen e we have to invoke
the Heisenberg pi ture, whi h gives '(~x; t) = eiHt=h '(~x; 0)e iHt=h . Using the well known
fa t that eiHt=h a(~k)e iHt=h = e i!(~k)t a(~k) and eiHt=h ay(~k)e iHt=h = ei!(~k)t ay (~k), whi h is a
onsequen e of [a(~k); H = h ! (~k)a(~k) and [ay (~k); H = h ! (~k)ay(~k), we nd
'(~x; t) = L
h y ~ i(~k~x
a (k)e
2ko (~k)
3 X
2
q
~k
ko t)=h
+ a(~k )ei(~k~x
ko t)=h
(2.8)
3 P
3 R
In an innite volume we repla e L 2 ~k by (2h ) 2 d3~k. Note that in the Heisenberg
pi ture positive energy modes behave in time as eiEt=h . Apparently we an identify (up to a
fa tor) '~ (~k) with ay ( ~k) and '~+ (~k) with a(~k), whi h is ompatible with '~ (~k) = '~( ~k ),
required to des ribe a real eld ( omplex elds will be dis ussed in problem 5).
The Hilbert spa e is now given by the produ t of the Hilbert spa es of ea h ~k separately
jfn~k g >= jn~k >=

Y
~k
~k
y(~k)n~k
aq
j0~ > ;
n~k ! k
(2.9)
with n~k the o upation number, whi h in eld theory is now interpreted as the number of
free parti les of momentum ~k, a denition that makes sense as the energy of su h a state
is n~k ko (~k) above the state with zero o upation number (the \va uum"). It is the property
of the harmoni os illator, that its energy is linear in the o upation number, whi h makes
the eld theory interpretation in terms of parti les possible. The annihilation operator in
this language therefore removes a parti le (lowering the energy by the appropriate amount),
whi h onsequently an be interpreted as the annihilation of the removed parti le with an
antiparti le (des ribed by the annihilation operator). For a real s alar eld, a parti le is its
own antiparti le and this des ription is perhaps somewhat unfamiliar. But for the omplex
eld of problem 5, the Fourier omponent with negative energy is independent of the one with
positive energy, hen e des ribing a separate degree of freedom, namely that of an antiparti le
with opposite harge.
Intera tions between the parti les are simply introdu ed by modifying the Klein-Gordon
equation to have non-linear terms, after whi h in general the dierent Fourier omponents
no longer de ouple. Field theory thus seems to be nothing but the quantum me hani s of an
innite number of degrees of freedom. It is, however, its physi al interpretation that ru ially
diers from that of ordinary quantum me hani s. It is this interpretation that is known as
se ond quantisation. We were for ed to introdu e the notion of elds and the interpretation
involving antiparti les, when ombining quantum me hani s with relativisti invarian e. 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, whi h for
the free s alar eld will be veried in problem 6
['(~x; t); '(~x0 ; t0 ) = 0; for (~x ~x0 )2 > (t t0 )2 2 :
(2.10)
It states that the a tion of an operator on the wave fun tional at a given spa e-time point
is independent of the a tion of the operator at an other spa e-time point, as long as these
two points are not ausally onne ted. Due to the des ription of the time evolution with
a Hamiltonian, whi h requires the hoi e of a time oordinate it remains to be established,
however, that these equations are ovariant under Lorentz transformations. We will resolve
this by using the path integral approa h, in whi h the Lorentz invarian e is intrinsi , but
whi h an also be shown to be equivalent to the Hamiltonian formulation.
Before preparing for path integrals by dis ussing the a tion prin iple, we would rst like
to address a simple physi al onsequen e of the introdu tion and subsequent quantisation
of elds. It states that empty spa e (all o upation numbers equal to zero) has nevertheless
a non-trivial stru ture, in the same way that the ground state of a Hydrogen atom is nontrivial. Put dierently, empty spa e is still full of zero-point u tuations, whi h are, however,
only visible if we probe that empty spa e in one way or another. Also, formally, as ea h
6
zero-point energy is non-zero, the energy of the va uum in eld theory seems to be innite
E0 =
~k
~k2 2 + m2 4 = ?
(2.11)
However, (as long as gravity is left out of our onsiderations) one is only sensitive to differen es in energy. If we probe the va uum, its energy an only be put to zero for one
parti ular value of the probe. The dependen e of the va uum energy on the probe an be
used to dis over the non-trivial stru ture of the va uum.
A famous and elegant method for probing the va uum was introdu ed by Casimir (Pro .
Kon. Ned. A ad. Wet., ser. B51 (1948) 793), who onsidered using two ondu ting plates in
empty spa e. The energy of the va uum is a fun tion of the distan e between the two plates,
whi h gives a for e. Stri tly speaking we should dis uss this in the situation of the quantised
ele tromagneti eld (see Itzykson and Zuber, par. 3-2-4), but the essential ingredient is
that Fourier omponents of the eld are ae ted by the presen e of the ondu ting plates.
We an also dis uss this in the ontext of the simple s alar eld we have introdu ed before,
by assuming that the eld has to vanish at the plates. For simpli ity we will also take the
mass of the s alar parti les to vanish. If furthermore we use periodi boundary onditions
in the two other perpendi ular dire tions over a distan e L, than one easily veries that the
for e per unit area on the ondu ting plates is given by
v
!
1 u
u 2 h
X
~n 2
t
1 d X
FL (x) = dEo (x)=dx =
2L2 dx ~n2ZZ2 k=1
!
h k 2
+
;
x
(2.12)
where due to the vanishing boundary onditions the Fourier modes in the x1 dire tion,
perpendi ular to the ondu ting walls, are given by sin(kx1 =x) with k a positive integer,
whereas the quantisation of the momenta in the other two dire tions is as usual.

6*

L

L
x ?

g. 3
One an now formally take the innite volume limit
v
u
u
t
!
h k 2

2
F (x) = Llim
F (x) =
:
d p~ p~ +
!1 L
x
8 2h 2 dx k=1 2
1 Z
d X
(2.13)
The integral and the sum are learly divergent, but as Casimir observed, in pra tise no
ondu ting plate an shield a eld perfe tly and espe ially for high frequen y the boundary
onditions should be modied. One an mimi this by arti ially utting o the integral and
sum at high momenta. We would not expe t the physi al result to depend on the details of
how we do this, as otherwise we ould use this experiment in an ingenious way to learn how
nature behaves at arbitrarily high energies. Indeed Casimir's areful analysis showed that
7
the result is independent of the uto fun tion hosen. It is an important example of what
we will later re ognise as renormalisability of eld theory. Sin e the result is insensitive to
the method of regularisation (only an overall onstant ontribution to Eo (x) depends on it,
but that is not observable, as we argued before), we an hoose a onvenient way to perform
the al ulation. Details of this will be provided in problem 2. The method of al ulation
is known as dimensional regularisation, where one works in an arbitrary dimension (n 6= 2)
and then analyti ally extends the result to n = 2. We will nd that
v
!
u
1 Z
u
d X
h k 2
t 2
F (x) = nlim
d p~ p~ +
!2 8 2h
x
2 dx k=1 n
h ( (n + 1)=2) (n+2)
x
:
(2.14)
= nlim
(n + 1) ( n 1) 3n=2
!2
8
( 1=2)
i
in whi h (i) 1
k=1 k is the Riemann fun tion. It an be analyti ally extended to odd
negative arguments, where in terms of Bernoulli oe ients (1 2i) = B2i =(2i). Also
( 12 ) = 32 ( 32 ) is nite, and we simply nd that
2 h
F (x) =
:
(2.15)
480x4
Please note that we have disregarded the spa e outside the ondu ting plates. Imposing
also periodi boundary onditions in that dire tion, one easily nds that the region outside
the plates ontributes with F (L x) to the for e, and vanishes when L ! 1. Therefore,
the ee t of the zero-point u tuations in the va uum leads to a (very small) attra tive
for e, whi h was ten years later experimentally measured by Sparnaay (Physi a, 24 (1958)
751). Another famous example of the in uen e of zero-point u tuations is the Lamb shift
in atomi spe tra (hyperne splittings), to be dis ussed at the end of the last se tion.
P
3 Euler-Lagrange equations
The Klein-Gordon equation in Lorentz ovariant form (x ( t; x; y; z ) (x0 ; ~x))
0
1
1
C
B
1
C
C
;
B
g '(x) + m2 '(x) = 0 ; g = B
A

1

1
an be derived by variational al ulus from an a tion prin iple
Z
S = d4 x L('; '; x) ; L('; '; x) = 1 ( ')2 V (') ;
(3.1)
( ')2 ' ' = g ' ' ; V (') = 21 m2 '2 :
(3.2)
We assume the eld to be given at the boundary of the domain M of integration (typi ally
assuming the eld vanishes at innity) and demand the a tion to be stationary with respe t
to any variation '(x) ! '(x) + '(x) of the eld,
S (')
S (' + ') S (') =

=
d4 x
'(
' +
d4 x
'
'
V (')
'
'
Z
V (')
) +
d (' ') = 0 ;
'
M
(3.3)
where d is the integration measure on the boundary M . The variation ' is arbitratry,
ex ept at M , where we assume ' vanishes, and this implies the Euler-Lagrange equation
' +
V (')
=0 ;
'
(3.4)
whi h oin ides with the Klein-Gordon equation. We an also write the Euler-Lagrange
equations for arbitrary a tion S (') in terms of fun tional derivatives
S
S
=
'(x) '(x)

S
=0 ;
'(x)
(3.5)
where stands for the total fun tional derivative, whi h is than split a ording to the expli it
dependen e of the a tion on the eld and its derivatives (usually an a tion will not ontain
higher than rst order spa e-time derivatives). Please note that a fun tional derivative has
the property '(x)='(y ) = 4 (x y ), whi h is why in the above equation we take fun tional
derivatives of the a tion S and not, as one sees often, of the Lagrangian density L.
The big advantage of using an a tion prin iple is that S is a Lorentz s alar, whi h
makes it mu h easier to guarantee Lorentz ovarian e. As the a tion will be the starting
point of the path integral formulation of eld theory, Lorentz ovarian e is mu h more easy
to establish within this framework (there are instan es where the regularisation, required
to make sense of the path integral destroys the Lorentz invarian e, like in string theory.
Examples of these anomalies will be dis ussed later for the breaking of s ale invarian e and
gauge invarian e). It is now simple to add intera tions to the Klein-Gordon equation, by
generalising the dependen e of the \potential" V (') to in lude higher order terms, like
1
V (') = 12 m2 '2 + '4 ;
4!
(3.6)
whi h is known as a s alar '4 eld theory. Later we will see that one an not add arbitrary
powers of the eld to this potential, ex ept in two dimensions.
As in lassi al eld theory, we an derive from a Lagrangian with '(x) and '_ (x)
'(x)=t as its independent variables, the Hamiltonian through a Legendre transformation
to the anoni al pair of variables (x) (the \momentum") and '(x) (the \ oordinate")
(x) =
Z
S
; H=
'_ (x)
H(x)d3~x =
( (x)'_ (x)
L(x)) d3~x :
(3.7)
The lassi al Hamilton equations of motion are given by
'_ (x) =
H
H
H
; _ (x) =
+ i
:
(x)
'(x)
i '(x)
For the Klein-Gordon eld we simply nd

H = 21 (x)2 + 21 (i '(x))2 + 21 m2 '2(x) ;
(3.8)
(3.9)
and in problem 5 one will see that this Hamiltonian oin ides with eq. (2.6), if we substitute
for '(x) eq. (2.8). For an intera ting s alar eld one nds
(3.10)
H = 1 (x)2 + 1 (i '(x))2 + V ('(x)) ;
2
whi h perhaps explains why V is alled the potential.

It is well known that the Hamiltonian equations imply that H itself is onserved with
time, provided the Lagrangian (or Hamiltonian) has no expli it time-dependen e
!
dH Z
H
H
=0 :
(3.11)
= d3~x _ (x)
+ '_ (x)
dt
(x)
'(x)
Conservation of energy is one of the most important laws of nature and it is instru tive to
derive it more dire tly from the fa t that L does not depend expli itly on time.
We dene
R
the Lagrangian L as an integral of the Lagrange density L over spa e, L d3~x L, su h
that
!
!
Z
S
S
S
dL Z
= d3~x t '(x)
:
= d3~x t '(x)
+ ( '(x))
dt
'(x) t
'(x)
'(x)
(3.12)
The last term ontains a total derivative, whi h vanishes if we assume that the eld is time
independent (or vanishes) at the boundary of the spatial integration domain. The above
equation be omes now
dL d Z
S
dZ
d ~x '_ (x)
d ~x '_ (x) (x) ;
(3.13)
=
=
dt dt 3
'_ (x) dt 3
whi h an also be expressed as
dH
dZ
d3~x ('_ (x) (x) L)
=0 :
(3.14)
dt
dt
In the same fashion one proves onservation of momentum in ase the Lagrangian does
not expli itly depend on spa e ( L=xi = 0)
!
dL Z
dZ
S
0=
=
d ~x (x)i '(x) :
(3.15)
= d3~x i '(x)
dxi
'(x)
dt 3
The onserved momentum is hen e given by
Pi =
d3~x (x)i '(x) :
(3.16)
Both onservation of momentum and energy are examples of onservation laws that are
onsequen e of symmetries (translation and time invarian e). They an be derived as the
spa e integral of the time omponent of a onserved urrent or tensor
J (x) = 0 ; T (x) = 0 :
(3.17)
In problem 3 these quantities will be dened for a harged s alar eld, where J (x) an be
identied with the urrent, whose time omponent is the harge density. Indeed the total
harge is onserved. Assuming the urrent to vanish at spatial innity one easily nds
Z
dZ
d ~x J (x) = d3~x i Ji (x) = 0
(3.18)
dt 3 o
The underlying prin iple is des ribed by the Noether theorem, whi h implies that if the
Lagrangian L is invariant under ' ! ' , where is a parameter (su h as a shift of the
oordinates or a phase rotation of a omplex eld), then the following urrent is onserved
S ' (x)
J ( x) =
:
(3.19)
( '(x))
10
The proof is simple and uses the Euler-Lagrange equations to substitute fS= ( '(x))g
for S='(x)
dL(')
S ' (x)
S ( ' (x))
0 =
=
+
d
'(x)
( '(x))

!
' (x)
S ( ' (x))
S
+
= J (x)
(3.20)
=
( '(x))

( '(x))

We here onsidered the invarian e under a global symmetry, but important in nature are
also the lo al symmetries, like the gauge invarian e related to lo al hanges of phase and the
general oordinate invarian e in general relativity. Parti ularly with the latter in mind we
demand therefore that the a tion S (and not just L) is invariant under '(x) ! '(x) (x), with
an arbitrary fun tion of spa e-time. This a tually leads to the same onserved urrents
in ase L is also invariant. The same omputation as above, still using the Euler-Lagrange
equations, shows that
S
0=
= J (x) :
(3.21)
(x)
As an important example we will dis uss how this onstru tion leads to onservation of the
energy-momentum tensor, using general oordinate invarian e, whi h is the lo al version
of translation invarian e. For this we have to make the a tion invariant under su h lo al
oordinate redenitions. As long as indi es are ontra ted with the metri tensor g , L will
be invariant under general oordinate transformations, due to the transformation property
x x
(3.22)
x = x + " (x) ; g (x) = g (x) :
x x
For global translation invarian e, " is onstant and eqs. (3.14) and (3.15) an be easily
generalised to show that the energy-momentum tensor, T = ' ' g L, is onserved
(eq. (3.17)). For " not onstant, we note that the integration measure d4 x is not a s alar
under
oordinate transformations, but the asso iated Ja obian an be easily absorbed
p general
by
det g, where the determinant is applied to the 4 4 matrix g . For a s alar eld this
leads to the following invariant a tion
S=
d4 x
det g ( 12 g ' ' V ('))
(3.23)
For the original

oordinates x of Minkowski spa e-time the metri is given as in eq. (3.1), in
q
parti ular
det g (x) = 1, and by expanding g to rst order in " (x) we nd
S=
d4 x (1 " (x)) [ 21 g '(x) '(x) V ('(x)) + " (x) '(x) '(x) : (3.24)
Now observe that g (x) is onstant, su h that the term independent of " is a fun tion of
x, integrated over x, whi h is simply the a tion itself, as x now plays the role of a dummy
integration variable. The term linear in " therefore has to vanish, but note that it only
involved the variation of the metri under the general oordinate transformation. Hen e,
0=
d4 x " (x) (g L(x) '(x) '(x))
d4 x " (x) T (x) ;
(3.25)
whi h implies onservation of the energy-momentum

tensor (Too = pH). From the fa t that
p

g = " " , g = g g g and det g = 12 g det gg , we derive
the identity
S
:
(3.26)
T (x) 2
g (x)
11
In taking the derivative with respe t to the metri , it is important that any Lorentz ve tor
(like the derivative ') or tensor appears in the Lagrangian with its indi es down. Furthermore, the result is to be evaluated for Minkowski spa e. Eq. (3.26) always gives a symmetri
energy-momentum tensor and from the derivation it is lear that the result holds not only
for a simple s alar eld, but for any other bosoni eld theory (fermions form an ex eption,
see problem 23) like the one for the ele tromagneti eld, whi h we dis uss now.
The eld is given by the tensor F (x), with Ei (x) = F0i (x) its ele tri and Bi (x) =
1"
2 ijk Fjk (x) its magneti omponents. In terms of the ve tor potential A (x) one has
F (x) = A (x) A (x) :
(3.27)
This already implies one of the Maxwell equations (through the so- alled Ja obi or integrability onditions)
F + F + F = 0 :
(3.28)

Written as " F = 0 they are easily seen (resp. for = 0 and = i) to give
divB~ = 0 ; 0 B~ + rotE~ = ~0 :
(3.29)
The dynami al equations determining the elds in terms of the urrents, or the sour es,
J = ( ; J~) are given by
1
F = J or divE~ = ; rotB~ 0 E~ = J~ :
(3.30)

We have hosen Heaviside-Lorentz units and in the future we will also often hoose units
su h that h = = 1.
These Maxwell equations follow from the following a tion
Sem (J ) =
d4 x (
1
4
F (x)F (x) A (x)J (x)) :
(3.31)
We note, as is well known, that the equations of motion imply that the urrent is onserved.
With Noether's theorem this makes us suspe t that this is aused by a symmetry and indeed
it is known that under the gauge transformation
A (x) ! A (x) + (x)
(3.32)
the theory does not hange. Our a tion is invariant under this symmetry if and only if the
urrent is onserved. This gauge symmetry will play a ru ial role in the quantisation of the
ele tromagneti eld.
An example of a onserved urrent an be dened for a omplex s alar eld. Its a tion
for a free parti le is given by
Z

S0 = d4 x ' (x) '(x) m2 ' (x)'(x) :
(3.33)
It is invariant under a phase rotation '(x) ! exp(ie)'(x) and from Noether's theorem we
dedu e that
J (x) ie('(x) ' (x) ' (x) '(x))
(3.34)
is onserved, see problem 3. We an extend this global phase symmetry to a lo al symmetry
if we ouple the s alar eld minimally to the ve tor potential
Z

S = d4 x 1 F (x)F (x) + (D ') (x)D '(x) m2 ' (x)'(x) ;
4
D '(x) '(x) ieA (x)'(x) :

12
(3.35)
This guarantees the ombined invarian e under a lo al gauge transformation
'(x) ! exp(ie(x))'(x) ; A (x) ! A (x) + (x) :
(3.36)
whi h makes the ovariant derivative D '(x) of the s alar eld transform as the s alar eld
itself, even for lo al phase rotations. Note that we an write this a tion also as
Z
S = Sem (J ) + S0 + d4 x e2 A (x)A (x)j'(x)j2 ;
(3.37)
with J as given in eq. (3.34). We leave it as an exer ise to show how the a tion of the ele tromagneti eld an be generalised to be invariant under general oordinate transformations
and to derive from this the energy-momentum tensor. The result is given by
Sem (J = 0) =
1
4
d4 x g g F F
det g ; T = 14 g F F
F F : (3.38)
4 Tree-level diagrams
In general, in the presen e of intera tions, the equations of motions an not be solved exa tly
and one has to resort to a perturbative expansion in a small parameter. We dis uss the s alar
ase rst, as it is as always the simplest. We add to the Lagrangian density L a so- alled
sour e term, whi h ouples linearly to the eld ' ( ompare the driving for e term for a
harmoni os illator)
(4.1)
L = 12 ( ')2 V (') J (x)'(x) :
For sake of expli itness, we will take the following expression for the potential
g
(4.2)
V (') = 21 m2 '2 (x) + '3 (x) :
3!
The Euler-Lagrange equations are now given by
'(x) + m2 '(x) + 1 g'2(x) + J (x) = 0 :
(4.3)
2
If g = 0 it is easy to solve the equation (des ribing a free parti le intera ting with a given
sour e) in Fourier spa e. Introdu ing the Fourier oe ients
J~(k) =
1 Z
1 Z
ikx
d x e J (x) ; '~(k) =
d x eikx '(x)
(2 )2 4
(2 )2 4
(4.4)
it follows that
( k2 + m2 )'~(k) + J~(k) = 0 or '(x) =
d4 y G(x y )J (y ) ;
where G is alled the Green's fun tion, as it is the solution of the equation
( + m2 )G(x y ) = 4 (x y )
(4.5)
(4.6)
Expli itly, it is given by the following Fourier integral
G(x y ) =
d4 k e ik(x y)
:
(2 )4 k2 m2 + i"
13
(4.7)
Please note our short-hand notation of k2 for k k and k(x y ) for k (x y ). A Green's
fun tion is not uniquely spe ied by its 2nd order equations, but also requires boundary onditions. These boundary onditions are, as we will see, spe ied by the term i". Be ause of
the interpretation of the negative energy states as antiparti les, whi h travel \ba kwards" in
time, the quantum theory will require that the positive energy part vanishes for past innity,
whereas the negative energy part will be required to vanish for future innity. Classi ally
this would not make sense, and we would require the solution to vanish outside the future
light one. The ee t of the i" pres ription
is, to shift the poles on the real axes to the
1
2
2
~
2
omplex k0 plane at k0 = ((k + m ) i"). In se tion 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 eld oupled to a sour e, we an do perturbation in the strength of the oupling onstant g .
'(x) + m2 '(x) + J (x) = 1 g'2 (x)
(4.8)
2
an be solved iteratively by substituting a series expansion for '(x),

'(x) = '0 (x) + g'1 (x) + g 2 '2 (x) + :
Obviously we have
(4.9)
'0 (x) = d4 y G(x y )J (y ) ;

whereas '1 (x) will be determined by the equation
'1 (x) + m2 '1 (x) = 21 '20 (x) :
(4.10)
(4.11)
We an therefore interpret the right-hand side as a sour e (up to a minus sign) and this
allows us to solve '1 (x) using the Green's fun tion
'1 (x) =
1
2
y )'20 (y ) =
d4 y G(x
1
2
G(x y )d4y G(y
This looks parti ularly simple in Fourier spa e

1
'~1 (k) =
2(2 )2 k2
Z
1
dp
m2 + i" 4 (p2
z )G(y
w)J (z )J (w)d4 zd4 w :

(4.12)
J~(p)J~(k p)
:
m2 + i")((k p)2 m2 + i")
It is lear that this an be ontinued iteratively, e.g.

nX1
'n (x) + m2 'n (x) = 12 'i (y )'n 1 i(y ) :
i=0
whi h is solved by
'n (x) =
1
2
d4 y G(x y ) (2'0 (y )'n 1(y ) + 2'1 (y )'n 2(y ) + ) :
(4.13)
(4.14)
(4.15)
Here we have written out the terms in the sum expli itly to indi ate that all terms o ur
twi e and are the produ t of two dierent terms, ex ept for the term ' 21 (n 1) (y )2 at n odd,
whi h o urs on e. In Fourier spa e one nds
Z
1
1
d p (2'~0 (p)'~n 1 (k p) + 2'~1 (p)'~n 2 (k p) + ) : (4.16)
'~n (k) =
2(2 )2 k2 m2 + i" 4
14
By indu tion it is now easy to prove that
'(x) =
diagrams
+
1
2

HHHH +

HHHHHH +

HH +
+ 12
1
2
1
8
HHH
H
Y
Y
g #verti es Z Y
=
dxiv
G(xi xj ) J (xks ) :
(4.17)
N
(diagram)
<i;j>
f
i
g
f
k
g
v
s
diagrams
Here the index iv runs over all verti es, sour es and the point x, (so it does not label the
four spa e-time omponents of a single point, frequently it will be assumed that it is lear
from the ontext what is meant), whereas ks runs only over positions of the sour es. The
expression < i; j > stands for the pairs of points in a diagram onne ted by a line ( alled
propagator).
The Feynman rules to onvert a diagram to the solution are apparently that ea h line
(propagator) between points x and y ontributes G(x y ) and ea h ross (sour e)
at a
R
point x ontributes J (x). Furthermore, for ea h vertex at a point x we insert d4 x and
a power of the oupling onstant g . Finally ea h diagram omes with an overall fa tor
1=N (diagram), being the inverse of the order of the permutation group (inter hange of
lines and verti es) that leaves the diagram invariant (whi h is also the number of ways the
diagram an be onstru ted out of its building blo ks). We have derived these rules for
the ase that = 0, su h that only three point verti es appear. All that is required to
generalise this to the arbitrary ase with n-point verti es, is that ea h of these ome with
their own oupling onstant (i.e. for a four-point vertex). This is the reason why these
verti es are weighed by a fa tor 1=n! in the potential and hen e by a fa tor 1=(n 1)! in
the equations of motion (to be pre ise, if V (') = gn'n =n!, the equation of motion gives
2 '(x) + J (x) = gn '(n 1) (x)=(n 1)!, and the fa tor (n 1)! is part of the ombinatori s
involved in inter hanging ea h of the n 1 fa tors ' in the intera tion term.
X
oordinate spa e
xHH
g d4x
R
y G(x y )
x
d4 xJ (x)
momentum spa e
k1
HjH6

k2
k3
k
k
(2)2 4 (
R
table 1
i ki )
vertex
d4 k k2 m1 2 +i" propagator
J~(k)
sour e
It is straightforward to translate these Feynman rules to momentum spa e, by inserting

the Fourier expansion of ea h of the terms that o ur. REa h propagator, whi h arries a
momentum k is repla ed by a fa tor 1=(k2 m2 + i") and d4 k, ea h sour e with momentum
k owing in the sour e by a fa tor J~(k), ea h vertex by a fa tor of the oupling onstant
(i.e. gn for an n-point fun tion), a fa tor 1=(2 )2 (for an n-point fun tion a fa tor (2 )4 2n )
and a momentum onserving delta fun tion. To understand why momentum is onserved at
ea h vertex we use that in the oordinate formulation ea h vertex omes with an integration
over its position. As ea h line entering the vertex arries a Green's fun tion that depends
15
on that position (this being the only dependen e), we see that a vertex at the point x gives
rise to
Z
Z
Z
Y
Y
d4 k e ik (x x )
d4 x G(x x ) ! d4 x
(2 )4 (k2 m2 + i")

Z
Y
X
d4 k eik x
= (2 )4
4 ( k ) :
(4.18)
4
2
2
(2 ) (k m + i")

Conventions in the literature an dier on how the fa tors of i (whi h will appear in the
quantum theory) and 2 are distributed over the verti es and propagators. Needless to say
that the nal answers have to be independent of the hosen onventions.
As a last example in this se tion we will look again at the ele tromagneti eld (whose
parti les are alled photons). In Fourier spa e the equations of motion are given by
( k2 + k k )A~ (k) = J~ (k) :
(4.19)
Unfortunately the matrix k2 + k k has no inverse as k is an eigenve tor with zero
eigenvalue. This is a dire t onsequen e of the gauge invarian e as the gauge transformation
of eq. (3.32) in Fourier language reads
~ k) :
A~ (k) ! A~ (k) + ik (
(4.20)
The omponent of A in the dire tion of k is for obvious reasons alled the longitudinal
omponent, whi h an be xed to a parti ular value by a gauge transformation. Fixing the
longitudinal omponent of the ele tromagneti eld (also alled photon eld) is alled gauge
xing, and the gauge hoi e is pres ribed by the gauge ondition. An important example is
the so- alled Lorentz gauge
A (x) = 0 or k A~ (k) = 0 :
(4.21)
Be ause of the gauge invarian e, the hoi e of gauge has no ee t on the equations of motion,
be ause the urrent is onserved, or k J~ (k) = 0. The urrent (i.e. the sour e) does not
ouple to the unphysi al longitudinal omponent of the photon eld. It stresses again the
importan e of gauge invarian e and its asso iated onservation of urrents.
To impose the gauge xing, we an add a term to the Lagrangian whi h enfor es the gauge
ondition. Without su h a term the a tion is stationary under any longitudinal variation
A (x) = (x) of the ve tor eld, and the added term should be su h that stationarity in
that dire tion imposes the gauge ondition. For any hoi e of the parameter 6= 0 this is
a hieved by the a tion
S=
d4 x
1
4
F (x)F (x)
1
2
( A (x))2
A (x)J (x)
(4.22)
Indeed, the variation A (x) = (x) in the longitudinal dire tion leads to the equation
Z
d4 x (x) A (x) = 0 ;
(4.23)
whi h implies the Lorentz-gauge (assuming vanishing boundary onditions for the ve tor
potential at innity).
The equations of motion for this a tion now yields
F (x) + A (x) = J (x) ;

16
(4.24)
or in Fourier spa e
( k2 + (1 )k k )A~ (k) = J~ (k) ;

(4.25)
whi h is invertible, as long as 6= 0. The result is given by
k k (1 1 )

k2 +i"

A~ (k) =
J~ (k) :
(4.26)
k2 + i"
This is onsequently the propagator of the ele tromagneti eld (in the Lorentz gauge), also
simply alled the photon propagator. Like in the s alar ase it an be used to perform a
perturbative expansion for the lassi al equations of motion.
Note that the photon propagator simplies dramati ally if we hoose = 1, but all nal
results should be independent of the hoi e of and even of the hoi e of gauge xing all
together. This is the hard part in gauge theories. One needs to x the gauge to perform
perturbation theory and than one has to prove that the result does not depend on the hoi e
of gauge xing. In the quantum theory this is not entirely trivial, as the regularisation an
break the gauge invarian e expli itly. Fortunately, there are regularisations that preserve the
gauge invarian e, like dimensional regularisation whi h we already en ountered in se tion 2
(in dis ussing the Casimir ee t). In the presen e of fermions the situation an, however, be
onsiderably more tri ky. Some dierent hoi es of gauge xing will be explored in problems
8 and 9.
5 Hamiltonian perturbation theory

We onsider the Hamiltonian for a free s alar parti le oupled to a sour e. We will see
that the sour e an be used to reate parti les from the va uum in the quantum theory,
and it forms an important ingredient, like for the derivation of the lassi al perturbation
theory of the previous se tion, in deriving s attering amplitudes and ross se tions. Also the
Green's fun tion will reappear, but now with a unique spe i ation of the required boundary
onditions following from the time ordering in the quantum evolution equations.
For the Lagrangian
L = 21 ' ' 21 m2 '2 "J' ;
(5.1)
the Hamiltonian is given by
H = 21 2 + 21 (i ')2 + 21 m2'2 + "J' ;
(5.2)
where " is a small expansion parameter. We will quantise the theory in a nit volume V =
[0; L3 with periodi boundary onditions, su h that the momenta are dis rete, ~k = 2~n=L.

X
1
q
a(~k)ei~k~x + ay (~k)e i~k~x ;
'(~x; t = 0) =
~k
2V ko (~k)
(~x; t = 0) =
v
u
Xu
t
~k
ko (~k) ~ i~k~x
a(k )e
2V

ay (~k)e i~k~x :
(5.3)
The Hamiltonian is now given by H (t) = H0 + "H1 (t), and we work out the perturbation
theory in the S hrodinger representation. We have
X
k0 (~k)(ay (~k)a(~k) + 21 ) ;
H0 =
~k
H1 (t) =
d3~x J (~x; t)'(~x) =

V
X
q
~k
17

1
J~(~k; t) a( ~k) + ay (~k) ;
2k0(~k)
(5.4)
here J~(~k; t) is the Fourier oe ient of J (~x; t), or

1
J (~x; t) = p
V
~k
J~(~k; t)ei~k~x :
(5.5)
Let us start at t = 0 with the va uum state j0 >, whi h has the property that a(~k)j0 >= 0
voor all momenta, than it follows that
d
j (t) >= iH (t)j (t) > ;
dt
whi h an be evaluated by perturbing in ".
j (t) > e
d ^
j (t) > =
dt n
iH0 t
^ t) > ; j (
^ t) >=
j (
ieiH0 t H1 (t)e
iH0 t
n=0
(5.6)
^ n ( t) > ;
"n j
j ^ n 1(t) > :
(5.7)
^ t) > we are using the intera tion pi ture, whi h is the

A tually, by transforming to j (
usual way of performing Hamiltonian perturbation theory known from ordinary quantum
me hani s. These equations an be solved iteratively as follows
j ^ 1(t) >
j ^ 2(t) >
0
Z
0
Z
dt1 eiH0 t1 H1 (t1 )e
iH0 t1
j0 > ;
dt1 eiH0 t1 H1 (t1 )e
iH0 t1
j ^ 1(t1 ) >
t1
dt eiH0 t1 H1 (t1 ) dt2 eiH0 (t2 t1 ) H1 (t2 )e iH0 t2 j0 > ;

0Z 1
0
t
i dt1 eiH0 t1 H1 (t1 )e iH0 t1 j ^ n 1 (t1 ) >= :
0
j ^ n(t) >
(5.8)
Please note the time ordering, whi h is essential as H1 (t) does not ommute with H1 (t0 ) for
dierent t and t0 . We an for example ompute the probability that at time t j (t) > is still
in the ground state (whose energy we denote by E0 , whi h will often be assumed to vanish)
< 0j (t) > = e
iE0 t
^ t) >= e
< 0j (
iE0 t
f1 i" 0 dt1 < 0jH1(t1 )j0 >
t1
dt1 dt2 < 0jH1 (t1 )ei(H0 E0 )(t2 t1 ) H1 (t2 )j0 > +O("3 )g : (5.9)
0
0
It is simple to see that the term linear in " will vanish, as the va uum expe tation values of
the reation and annihilation operators vanish, i.e. < 0jaj0 >=< 0jayj0 >= 0. To evaluate
the remaining expe tation value in the above equation we substitute H1 in terms of the
reation and annihilation operators (see eq. (5.4))
< 0jH1 (t1 )ei(H0 E0 )(t2 t1 ) H1 (t2 )j0 >=

~(p~; t1 )J~(~k; t2 )
X J
q
< 0j a( p~) + ay (p~) ei(H0 E0 )(t2 t1 ) a( ~k ) + ay (~k) j0 >
~k;~p
4k0 (p~)k0 (~k)
~( ~k; t1 )J~(~k; t2 ) ik0 (~k)(t1 t2 )
X J
e
:
(5.10)
=
2k0 (~k)
~k
"2
18
Combining these results we nd
J~( ~k; t1 )J~(~k; t2 )

e
2k0 (~k)
~k
Z
Z
2 X t dt1 t dt2 J~( ~k; t1 )J~(~k; t2 ) e
1"

2
0
2k0 (~k)
~k 0
X
< 0j (t) > eiE0 t = 1 "2
= 1
t1
dt
dt
0 1 0 2
ik0 (~k)(t1 t2 ) +
O("3)
ik0 (~k)jt1 t2 j +
O("3): (5.11)
Espe ially the last identity is useful to relate this to the Green's fun tion we introdu ed
in the previous se tion. Using ontour deformation in the omplex ! plane we nd
1
d! 2
1
!
2i ik0 (~k)jtj

ei!t
=
e
:
(5.12)
2
~
k0 (k) + i"
2k0 (~k)
This an be shown as follows. When t > 0, we an deform the ontour of integration to the
upper half-plane (where ei!t de ays exponentially) and only the pole at ! = ! k0 (~k)+ i"
ontributes, with a residue 2ie ik0 (~k)t =( 2k0 (~k)) (see the gure). Instead, for t < 0 the
ontour needs to be deformed to the lower half-plane and the pole at ! = !+ k0 (~k) i"
ontributes with the residue 2ieik0 (~k)t =( 2k0 (~k)) (note that the ontour now runs lo kwise,
giving an extra minus sign).
!r
-r
'
$
!+
t>0
g. 4
t
<
0
!r
-r
&
%
!+
Z
This means that we an rewrite eq. (5.11) as

lim < 0j (t) >
t!1
eiE0 t
=1
i 2 XZ 1
d!
"
2 ~k 1
J~( !; ~k)J~(!; ~k)

+ O("3 ) ;
! 2 ~k2 m2 + i"
(5.13)
where
1 Z
1 Z
i!t
~
~
~
~
J (!; k) = p
d3~xdt J (x)eikx :
(5.14)
dt J (k; t)e = p
2
2V IRV
R
The last expression should be repla ed by (2 ) 2 IR4 d4 x J (x)eikx in ase the volume is innite. It is important to note that we have hosen J (x) = 0 for t < 0. Equivalently we an
start at t = 1 and integrate the quantum equation of motion up to t = 1. We have to
require that J (x) vanishes su iently rapidly at innity.
In an innite volume we therefore nd for what is known as the va uum to va uum
amplitude of the s attering matrix
J~( k)J~(k)
i 2Z
" d4 k 2
+ O("3 )
2 Z
k m2 + i"
i 2
" d4 xd4 y J (x)G(x y )J (y ) + O("3 ) : (5.15)
= 1
2
where G(x y ) is exa tly the Green's fun tion we introdu ed in the previous se tion. The
so- alled i" pres ription, whi h is equivalent with spe ifying the boundary onditions, has
lim < 0j (t) > eiE0 t = 1
t!1
19
therefore been derived from the time-ordering in the Hamiltonian evolution of the system
and is thus pres ribed by the requirement of ausality. Note that we an use the diagrams
introdu ed in the previous se tion to express this result (taking E0 = 0 from now on) as
i
+ O("3) ;
(5.16)
2 "J
"J
where the fa tor of a half is a onsequen e of the symmetry under inter hanging the two
sour es.
For a omplex s alar eld, ' and ' are independent and we need to introdu e two
sour es by adding to the Lagrangian 'J ' J (see problem 17). It is not too di ult to
show that in this ase
< 0j (t) >= 1
< 0j (t) >= 1 i
+ O("3) ;
>
(5.17)
"J
"J
without a fa tor of one half be ause the sour es J and J are independent and annot be
inter hanged. This is why in this ase the propagator has a dire tion.
6 Path integrals in quantum me hani s

For simpli ity we will start with a one-dimensional Hamiltonian
p^2
h
H=
+ V (^x) ; p^ =
;
(6.1)
2m
i x
where we have indi ated a hat on top of operators, to distinguish them from number-valued
oordinate and momentum. We wish to study the time-evolution operator exp( iHT=h ).
In the oordinate representation its matrix is given by
< x0 je iHT=h jx > ;
(6.2)
where jx > is the position eigenfun tion. We will also need the momentum eigenfun tion
jp >, i.e. p^jp >= pjp >, whose wave fun tion in the oordinate spa e is given by
eipx=h
< xjp >= p
:
(6.3)
2h
Indeed, one veries that
h
< xjp >= p < xjp > :
(6.4)
i x
An important ingredient in deriving the path integral expression will be the ompleteness
relations
Z
Z
^1 = dx jx >< xj and ^1 = dp jp >< pj :
(6.5)
For arbitrary N we an use this to write
< x0 je
iHT=h
jx >

iHT=N h N
jx >= < x0 je iHT=N hjxN 1 > dxN

< xN 1 je iHT=N h jxN 2 > dxN 2 < xN 2 j e iHT=N h jx2 > dx2
< x2 je iHT=N h jx1 > dx1 < x1 je iHT=N h jx > :

= < x0 j e
20
(6.6)
We will now use the so- alled 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 an be seen by expanding the exponents, and the error
term is a tually of the form [A; B =N 2 . (One an also use the Campbell-Baker-Hausdor
formula, whi h will be introdu ed later). With the Hamiltonian of eq. (6.1) this an be used
to write for N ! 1
2
e iHT=N h = e ip^ T=2mN h e iV (^x)T=N h :
(6.8)
By inserting the ompleteness relation for the momentum we an eliminate the operators
Z
iHT=N

h
< xi+1 je
jxi > = dpi < xi+1 jpi >< pije iHT=N h jxi >

=
2
dpi < xi+1 jpi >< pi je ip^ T=2mN h e
2
dpi < xi+1 jpi >< pi je ipi T=2mN h e
iV (^x)T=N h
jxi >
iV (xi )T=N h jx >
i
eipi (xi+1 xi )=h i( 2pm2i +V (xi ))T=N h

e
:
(6.9)
2h
This an be done for ea h matrix element o urring in eq. (6.6). Writing t = T=N , xN = x0
and x0 = x we nd
Z
Z N 1
NY1
NY1
Y
0
iHT=

h

< x je
jx > = Nlim
dxi
dpj
< xi+1 jpi >< pi je iH t=h jxi >
!1
i=1
j =0
i=0
"
!#
Z
Z
N
1
N
1
dp0 Y dxi dpi
it X pi (xi+1 xi ) p2i
= Nlim
exp
V (xi ) : (6.10)
!1 2h
i=1
2h
h i=0
t
2m
It is important to observe that there is one more p integration than the number of x integrations.
The integrals in the path integral are strongly os illating and an only be dened by
analyti ontinuation. As parameter for this analyti ontinuation one hooses the time
t. For t = T=N iT =N = i the Gaussian integral over the momenta is easily
evaluated
"
#
#
"
Z 1
q
ipi (xi+1 xi ) p2i
(xi+1 xi )2
1
dpi exp
= 2mh = exp 2 m
: (6.11)
h
2mh
h
1
whi h leads to
!#
"
1

mN 2 N Z NY1
NX1 m(xi+1 xi )2
0
iHT=

h
< x je
jx >= Nlim
+ V (xi )
dxi exp
!1 2 T h

h i=0
2 2
i=1
(6.12)
or after substituting T = iT we nd
!#
"
1

t NX1 m(xi+1 xi )2
mN 2 N Z NY1
0
iHT=

h
V (xi )
dxi exp i
< x je
jx >= Nlim
!1 2iT h

h i=0
2t2
i=1
(6.13)
This is the denition of the path integral, but formally it will often be written as
Z x(T )=x0
Z T
o
n
0
iHT=
h

< x je
jx >= x(0)=x Dx(t) exp[iS=h ; S = 0 dt 12 mx_ 2 (t) V (x(t)) ; (6.14)
=
dpi
21
sin e the dis retised version of the a tion with xj x(t = j t) is pre isely
NX1
(xi+1 xi )2
1m
Sdis rete = t
2
t2
i=0
V (xi )
(6.15)
It is important to note that the ontinuous expression is just a notation for the dis rete
version of the path integral, but formal manipulations will be mu h easier to perform in
this ontinuous formulation. Furthermore, the integral is only dened through the analyti
ontinuation in time.
However, if we integrate over xN = x0 this analyti ally ontinued path integral, with
T = iT , has an important physi al interpretation
Z
dx < xje H T =h jx >= Tr(e
H )
=T =h
(6.16)
It is the quantum thermal partition fun tion (the Boltzmann distribution) with a temperature of h =kT . In the ontinuous formulation we therefore have
Tr(e T H=h ) =
x(T )=x(0)
Dx( )
exp[ SE =h
(6.17)
in whi h SE is the so- alled Eu lidean a tion
SE =
d
1
2
o
m(dx( )=d )2 + V (x( )) :
(6.18)
It is only in this Eu lidean ase that one an dene the path integral in a mathemati ally
rigorous fashion on the lass of pie ewise ontinuous fun tions in terms of the so- alled
Wiener measure
Z x(T )=x0
Z
Z T
1 mx
_ ( )2 d=h ) ;
(6.19)
dWx;x0 (T )
D
x( ) exp(
x(0)=x
0 2
meaning that this measure is independent of the way the path is dis retised.
B

AAA D
BB
B
D

D
- D
t
DD
D
E
E
EE
BB

EE BB

E
EE
t
g. 5
For more details on this see \Quantum Physi s: A fun tional integral point of view", by J.
Glimm and A. Jae (2nd ed., Springer, New York, 1987).
We will now do an exa t omputation to give us some onden e in the formalism. To be
spe i , what we will ompute is the quantum partition fun tion for the harmoni os illator,
where V (x) = 12 m! 2 x2 .
ZN
#
"
1
NX1 1 xi+1 xi 2 1 2 2
mN 2 N Z 1 Z NY1
m
dxi exp

+ 2 m! xi : (6.20)
2 T h
h i=0 2

1
i=0
22
Note that we have now N integrations, be ause we also integrate over x0 = x(0) = x(T ) = xN
to implement the tra e. The path involved is thus periodi in time, a general feature of the
expression for the quantum partition fun tion in terms of a path integral. We now res ale
yi = xi
m 12
; !~ = ! ;
h
(6.21)
to obtain the simple result

"
NX1
1 Z NY1 dyi
1
p
ZN =
exp
2 (yi+1
1
i=0 2
i=0
Z
yi)2 + 12 !~ 2 yi2
(6.22)
We an diagonalise the quadrati term by using Fourier transformation

1 NX1 2ik`=N
b` e
; b` = bN ` ; b0 = b0 :
(6.23)
yk = p
N `=0
It is easy to verify that the Ja obian for the hange of variables yi ! b` is 1, and one obtains
a result that must look familiar from the lassi al small os illations problem for a nite
number of weights onne ted by strings,
#
"
Z N 1
Z 1
NX1

Y db
`
2
2
2
1
p exp 2
ZN =
:
(6.24)
4 sin (`=N ) + !~ jb` j
1
`=0 2
`=0
Note that if b` is omplex, we mean by db` = dReb` dImb` . The integral an now easily be
evaluated
NY1
1
4 sin2 (`=N ) + !~ 2 2 :
ZN =
(6.25)
`=0
We an onvert the produ t to a sum using a Lapla e transform. We start with the identity
log(=) =
su h that
1
lim
ds sa 1 (e
a&0 0
Z
s
s )
(6.26)
1
log[ZN (~! )=ZN (~!0 ) = 21 alim
ds s 1+a (Q(s; !~ ) Q(s; !~ 0 )) :
(6.27)
&0 0
We read o, from the denition of ZN , that Q is a sum of exponentials
NX1
i
h

2
(6.28)
Q(s; !~ )
exp s 4 sin2 (`=N ) + !~ 2 Ne s(~! +2) fs(0) ;
`=0
where
1 NX1 2s os(2(`+x)=N )
fs (x) =
e
:
(6.29)
N `=0
This is a periodi fun tion with period 1 (fs(x + 1) = fs (x)) and its dis rete Fourier oe ients an be omputed exa tly
Z 1
1ZN
dx e2ikx e2s os(2x=N )
dx e2ikx fs (x) =
f~s (k) =
N 0
0 Z
1 2
=
d eiNk e2s os() INk (2s) :
(6.30)
2 0
Z
23
Please note that we have ex hanged the sum over ` with extending the integration of x to
the interval [0; N . The last identity is one of the denitions of the modied Besselfun tion,
see e.g. \Handbook of mathemati al fun tions", by M. Abramowitz and I. Stegun (Dover,
New York, 1978). The advantage of these manipulations is that the Lapla e transform of
this Besselfun tion is know (see the same referen e)
Z
ds e
s I
p
[ + 2 1 j j
p2
;
(s) =
1
(6.31)
and as we an express fs (x) as a sum over these Besselfun tions

X
fs(x) =
e 2ikx INk (2s) ;
k2ZZ
(6.32)
this allows us to evaluate eq. (6.27). For te hni al reasons it is easier to ompute the variation
of the free energy with the frequen y, where the free energy F is dened as
ZN (~!) = exp[ F (~!) :
(6.33)
We therefore nd (using = T =h = N =h )

Z 1
X

2
ds e (2+~! )s INk (2s)
F (~!) = h !
!~
k2ZZ 0

q
X
h !
2 + 1 + ( 1 !~ 2 + 1)2
1!
~
= q
2
2
2 ( 21 !~ 2 + 1)2 1 k2ZZ
N jk j
(6.34)
The geometri series is of ourse easily summed, but to make the result more transparent
we introdu e the s aled ee tive frequen y
! T =N = ! = !~ 2 sinh( 21
)
(6.35)
and using the identity 21 !~ 2 + 1 = osh(

), one easily nds that

F (
) =
h
1
2
k2ZZ
N
jkj
1
=
1 e N
1
2
!
e
N=2
1
:
log
=
N
1 e N
(6.36)
The last identity an be seen as the free energy of the harmoni os illator with the frequen y
N=T =
= , as it an also be written as
!
1
X
1 )
N

1
(
n
+
2
:
(6.37)
e
F (
) =
log
h
N
n=0
Amazingly, even at nite N the eu lidean path integral agrees with the quantum partition
fun tion of a harmoni os illator, but with a frequen y that is modied by the dis retisation,
see eqs. (6.21) and (6.35). It is trivial to he k now that the limit N ! 1 is well dened
and gives the required result, sin e
lim
N=T = ! :
N !1
(6.38)
In general the exa t nite N path integral is no longer of a simple form. Nevertheless,
one an evaluate this exa t expression in relatively simple terms (whi h will verify the above
24
result along a dierent 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
dp0 NY1 Z dxi dpi
NX1 ipi (xi+1 xi ) p2i
0
ZN (x ; x; T ) =
exp
V (xi )
2h i=1
2h
h i=0

2m
!
!
Z
NY1
NY1 Z
V (^x)
p^2
exp
jxj >
< xj +1 jpj >< pj j exp
dxi dpi
= dp0
2mh
h
j =0
i=1
(
!
!)
p^2
V (^x) N
0
= < x j exp
exp
jx > :
(6.39)
2mh
h
This means that we an dene an ee tive Hamiltonian by
!
!)
(
V (^x) N
p^2
H
(
N
)
T
=

h
e
exp 2mh exp
h
(6.40)
But this Hamiltonian is not hermitian as one easily he ks from the above expression, sin e
under onjugation the order of the exponents ontaining the kineti and potential terms is
reversed. This an be orre ted in two ways
!
2!
2!

V
(^
x
)

p
^

p
^
H
(
N
)

=

h
exp
;
e 1
exp 4mh exp
h
4mh
!
!
!
V (^x)
p^2
V (^x)
H

2 (N ) =h
e
exp
exp
exp
;
(6.41)
2h
2mh
2h
leading to two equivalent expressions for the nite N path integral
!
!
p^2
p^2
0
0
exp ( H1 (N )T =h ) exp
jx >
ZN (x ; x; T ) = < x j exp
4mh
4mh
!
!
V (^x)
V (^x)
0
0
ZN (x ; x; T ) = < x j exp
exp ( H2 (N )T =h ) exp
jx > (6.42)
:
2h
2h
In parti ular the partition fun tion is given by
ZN =
dx ZN (x; x; T ) = Tr(e
H1 (N )T =h )
= Tr(e
H2 (N )T =h )
(6.43)
It is a tually not too di ult to show that there exists a unitary transformation U , su h
that UH1 U y = H2 , whi h shows that both hoi es are indeed physi ally equivalent.
In prin iple we an now ompute Hi (N ) for nite N as an expansion in 1=N , by using
the so- alled Campbell-Baker-Hausdor formula
1
1
eA eB = eF (A;B) ; F (A; B ) = A + B + 12 [A; B + [A; [A; B + [B; [B; A+ ; (6.44)
12
12
whi h is a series in multiple ommutators of the, in general non- ommuting, operators A and
B . It an be derived by expanding the exponentials, but in the mathemati s literature more
elegant onstru tions are known, based on properties of Lie groups and Lie algebras. These
obje ts will be dis ussed in se t. 18. For the harmoni os illator, working out the produ ts
of the exponential an be done to all orders and one nds (see problem 10 for details)
p^2 1
+ M
2 x^2 ;
(6.45)
Hi (N ) =
2Mi 2 i
25
with
dened as in eq. (6.35) and the ee tive masses Mi dened by
2m tanh( 12
)
m sinh(
)
M1 =
; M2 =
:

One an now expli itly verify that ( mp. eq. (6.37))
(6.46)
Tr(exp( H1 (N )T =h )) = Tr(exp( H2 (N )T =h )) = exp( F (

)T =h ) :
(6.47)
Now we have seen that, at least for some examples, the limit of in reasingly ner dis retisation is in prin iple well dened, we an think of generalisation to an arbitrary number
of dimensions (n) (for eld theory even to an innite number of dimensions).
#
"
Z T
D~x(t) exp hi 0 dt L(~x(t); ~x_ (t))
"
!#
Z
(~xi+1 ~xi )
t NX1
dn p0 Z NY1 dn xi dn pi
p~
exp i
H (p~i; ~xi )
= Nlim
!1 (2h
)n i=1 (2h )n
h i=0 i
t
"
#
1

mN 2 nN Z NY1
t NX1
= Nlim
dn xi exp i
L(~xi ; (~xi+1 ~xi )=t) : (6.48)
!1 2iT h

h i=0
i=1
We have purposely also given the expression that involves the path integral as an integral
over phase spa e, as it shows that the Gaussian integration over the momenta ee tuates
the Legendre transform
(p~ m~x_ )2 m~x_ 2
~p2
iV (~x) = i
+i
iV (~x) ;
(6.49)
i~p ~x_ i
2m
2m
2
whi h is equivalent to the stationary phase approximation for the momentum integration
H _ p~
(p~ ~x_ H (p~; ~x)) = ~x_
= ~x
=0 :
(6.50)
~p
~p
m
An interesting other example of the path integral is the ase of the intera tion of a harged
parti le with a magneti eld. In that ase one has for the Hamiltonian
(p^~ eA~ (~x^))2
+ V (~x^) :
(6.51)
H (p^~; ~x^) =
2m
Now, however, the matrix element < ~pi j exp( iH t=h )j~xi > will depend on the spe i
ordering for the position and momentum operators in H . Dierent orderings dier by terms
linear in h , or
A~ (~x^ ) ^~p = ^~p A~ (~x^) + ih i Ai (~x^) :
(6.52)
So, by hosing the so- alled Coulomb gauge i Ai (~x) = 0, the problem of the operator ordering
disappears. We leave it as an exer ise to verify that the a tion, obtained from the Legendre
transform, is given by
< ~x 0 je iHT=hj~x >
S=
dt
1
2
m~x_ 2
V (~x) + e~x_ A~ (~x)
(6.53)
Under a gauge transformation Ai (~x) ! Ai (~x) + i (~x), one nds that the a tion hanges to
S = S + ef(~x(T )) (~x(0))g. Using the path integral this means that
< ~x 0 j exp( iH T=h )j~x >= exp(ie(~x 0 )=h ) < ~x 0 j exp( iHT=h )j~x > exp( ie(~x)=h ) :
(6.54)
26
Sin e it is easily shown that H = exp(ie(~x^)=h ) H exp( ie(~x^ )=h ), this proves that the
path integral derived from eq. (6.53) has the orre t properties under gauge transformations,
despite the fa t that the derivation was performed by rst going to the Coulomb gauge.
As long as the Hamiltonian is quadrati in the momenta, the stationary phase approximation for the momentum integral is exa t. However, also for the oordinate integrals we
an use the stationary phase approximation (exa t for a harmoni os illator), whi h is related
to the WKB approximation in quantum me hani s. It gives a way of dening an expansion
in h , where in a ordan e to the orresponden e prin iple, the lowest order term reprodu es
the lassi al time evolution. Indeed the stationary phase ondition
S
S
= i
i
x (t) x (t)
d S
=0
dt x_ i (t)
(6.55)
is pre isely solved by the lassi al solutions, ~x l (t), with ~x l (0) = ~x and ~x l (T ) = ~x0 . We
expand around these solutions by writing
~x(t) = ~x l (t) + ~q(t) ; ~q(0) = ~q(T ) = ~0 ;
(6.56)
su h that
2 S (~x l ) j 0
q (t ) + O (q 3 ) :
(6.57)
i
j
0
q (t) q (t )
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 ondition. For the simple Lagrangian L = 21 m~x_ 2 V (~x)
one has
!
2 S (~x l )
d2
2 V (~x)
0
=
(
t
t
)
m
+
M
(
t
)
;
M
(
t
)

j
:
(6.58)
ij
ij
ij
q i (t) q j (t0 )
dt2
xi xj ~x=~x l(t)
For the harmoni potential, V = 21 m! 2~x2 , where the stationary phase approximation is
exa t, i.e. there are no O(q 3 ) orre tions. Introdu ing ~q(t), however, splits the a tion in a
lassi al pie e, that depends on the boundary onditions, and a quantum pie e des ribed by a
harmoni os illator a tion for the u tuations around the lassi al path, that is independent
of the boundary onditions and the lassi al path
S (~x) = S (~x l ) + 21
dt0 dt q i (t)
S (~x) = S (~x l ) +
dt ( 21 m~q_ 2
m! 2~q 2 ) :
(6.59)
0
In pra ti al situations one splits from the a tion the part quadrati in the oordinates and
velo ities, and onsiders the rest as a perturbation. In that ase x l is the lassi al solution
of the quadrati part only. As this an always be solved exa tly, and as non-quadrati path
integrals an rarely be omputed expli itly, this will be the way in whi h we will derive the
Feynman rules for the quantum theory, in terms of whi h one an e iently perform the
perturbative omputations.
1
2
7 Path integrals in eld theory

For a s alar eld in a nite volume V = [0; L3 the Hamiltonian is given in the Fourier
representation by (see eq. (2.6))

X
2 + 1 (~k2 + m2 )j'~(~k)j2 + V ('~) + '~(~k)J~( ~k; t) :
1 j
~
H=
~
(
k
)
j
(7.1)
2
2
~k=2~n=L
27
As '(x) is real we have '~ (~k) = '~( ~k). It is ustomary to write the term quadrati in the
elds (the mass term) expli itly, su h that the potential V (') only ontains the intera tion
terms. If we like we ould split the Fourier modes in their real and imaginary omponents
(the os(~x ~k) and sin(~x ~k) modes). Or even simpler is to use Diri hlet boundary onditions,
i.e. '(x) = 0 at the boundaries of the volume, su h that the Fourier modes are given by
Q
j sin(nj xj =L) (with nj > 0), with real oe ients. In either ase, for V (') = 0 the
Hamiltonian simply des ribes an innite set of de oupled harmoni os illators, whi h an
be trun ated to a nite set by introdu ing a so- alled momentum uto j~kj . In this
ase we know how to write the path integral, even in the presen e of intera tions. The
introdu tion of a uto is alled a regularisation. The eld theory is alled renormalisable
if the limit ! 1 an be dened in a suitable way, often by varying the parameters in a
suitable way with the uto. The lass of renormalisable eld theory is relatively small. For
a nite momentum uto the path integral is nothing but a simple generalisation of the one
we dened for quantum me hani s in n dimensions, or in the absen e of intera tions
"
Z N 1
NX1 X j'
Y
Y Y
~j +1(~k) '~j (~k)j2
N=
2
~
(2
i

t
)
d
'
~
Z = Nlim
(7.2)
j (k ) exp it
2
!1 ~
2
t
j
=0
j
=1
~
~
k
k
k
2
2 ~j (~k)j2
1 (~
2 k + m )j'
D'~(~k; t)
exp i
Xn
~k
1
2
j'~_ (~k; t)j2
1
2
'~j (~k)J~( ~k; j t)
(~k2 + m2 )j'~(~k; t)j2
1
o
'~(~k; t)J~( ~k; t) dtA :
One of ourse identies '~j (~k) = '~(~k; t = j t) and performing the Fourier transformation
on e more, we an write
T
dt
Xn
~k
1
2
j'~_ (~k; t)j2
1
2
(~k 2 + m2 )j'~(~k; t)j2
o
(i '(~x; t))2 21 m2 '2 (~x; t) '(~x; t)J (~x; t)
0
V
Z
n
o
=
d4 x 12 '(x) '(x) 21 m2 '2 (x) '(x)J (x) :
(7.3)
V [0;T
The last expression is manifestly Lorentz invariant apart from the dependen e on the boundary onditions on the elds (whi h should disappear on e we take L and T to innity). This
will allow us to perform perturbation theory in a Lorentz ovariant way (things are somewhat subtle as any nite hoi e of the momentum uto does break the Lorentz invarian e,
and there are some theories where this is not restored when removing the uto, i.e. taking
the limit ! 1). This a hieves a substantial simpli ation over Hamiltonian perturbation
theory. It is now also trivial to reintrodu e the intera tions, by adding the potential term
to the Lagrange density, and we nd in yet another shorthand notation for the measure
the following expression for the path integral (impli itly assuming that the boundary values
'(~x; 0) and '(~x; T ) are xed, pres ribed fun tions)
Z=
dt
D'(x)
n
d3~x 21 (t '(~x; t))2
'~(~k; t)J~( ~k; t)
exp i
d4 x 12
1
2
'(x) '(x)
1 m2 '2 (x)
V (') '(x)J (x)
:
(7.4)
In prin iple a path integral should be independent of the dis retisation used in order
to dene it. For the eu lidean path integral, one parti ular way that is used quite often is
V [0;T
28
the latti e dis retisation, where instead of a momentum uto one makes not only time but
also spa e dis rete. This means that the eld now lives on a latti e and its argument takes
the values ja where j 2 ZZ4 and a is the so- alled latti e spa ing, whi h in the end should
be taken to zero. By suitably restri ting the omponents of j , with appropriate boundary
onditions on the elds, one keeps spa e and time nite, V = a3 M 3 and T = aN . This leads
to an integral of the form
0
(
)1
Z
2
X ('
X
Y
'
)
3
j
j +e
1
d'j exp a4
(2a) NM =2
+ 21 m2 '2j + V ('j ) + Jj 'j A :
2
2
a

j
j
(7.5)
where e is a unit ve tor in the dire tion, and 'j is identied with '(aj ). In a sense,
the momentum uto is similar to the spa e-time uto 1=a. The latti e formulation is
very suitable for numeri ally evaluating the path integral, whereas the momentum uto
is suitable for performing perturbation theory around the quadrati approximation of the
a tion. For the latter we will ompute, using the path integral, the same quantity as was
al ulated in se tion 5, using Hamiltonian perturbation theory.
In the presen e of a sour e, the Hamiltonian depends on time and the evolution operator
has to be written in a way that takes the time ordering into a ount. As the time evolution
operator U (t) satises the S hrodinger equation
d
i U (t) = H (t)U (t) ; U (0) = 1 ;
(7.6)
dt
its solution an be written as (note the absen e of 1=n!)
U (t) = Texp
H (t)dt
( i)n
0
0
n=0
For onvenien e we introdu e the notation
U (t2 ; t1 ) Texp
dt1
t2
t1
t1
tn
dtn H (t1 ) H (tn) : (7.7)
H (t)dt ; t2 > t1 ;
(7.8)
whi h satises the property that

U (t3 ; t2 )U (t2 ; t1 ) = U (t3 ; t1 ) ; t3 > t2 > t1 :
(7.9)
In se tion 5 we al ulated the matrix element < 0jU (T )j0 > to se ond order in the sour e
(from now on we put " = 1). The Lagrangian relevant for the path integral evaluation is
given by L = 21 ' ' 12 m2 '2 J', with J (x) = 0 for t < 0 and t > T (and for ~x 2= [0; L3 ).
The va uum j0 > is the state where all ~k os illators are in their ground state. It turns out
that we do not need an expli it expression for this va uum wave fun tional, denoted by
0 (f'~(~k)g) =< f'~(~k)gj0 >. We have
< 0jU (T )j0 >=
p~
d'~(p~)d'~0 (p~) < 0jf'~0(p~)g >< f'~0 (p~)gjU (T )jf'~(p~)g >< f'~(p~)gj0 > ;
(7.10)
where f'~(p~)g plays the role of x and f'~0 (p~)g the role of x0 . The path integral expression for
the evolution operator therefore be omes
< f'~0 (p~)gjU (T )jf'~(p~)g >=
D'(x)
exp i
D'~(k)
exp B
i
~k;k0
1
2
(k 2
29
L('(x))d4x
m2 )j'~(k)j2
1
o
'~(k)J~( k)
C
A
(7.11)
We have here also performed the Fourier transformation with respe t to time, thereby onverting the path integral measure D'~(~k; t) to the multiple integral over the (dis rete temporal) Fourier omponents D'~(k), exa tly as was done in one dimension, see eq.(6.23), hen e
we also nd a unit Ja obian for this hange of variables. Obviously our notations are su h
that k (k0 ; ~k) and k2 = k02 ~k 2 . If we take the limit of spa e and time to innity (L ! 1
and T ! 1), the sums over k an be onverted into integrals. Finally we note that the
os illatory integrals o urring in the pathR integral an be dampened
by repla ing
m2 by
R
R
m2 i" as this leads to repla ement exp(i L(')d4x) ! exp(i L(')d4x " '2 d4 x). This
pres ription also allows us to make the analyti ontinuation to imaginary time, and oin ides with the pres ription derived for the propagator in the Hamiltonian formulation, so
that ausality is also properly implemented in the path integral approa h. To re over the
result obtained in the Hamiltonian approa h, we simply split o a square
1
0
2

Z

~
X
J (k ) C
2 m2 + i") '~(k)
1
A
< f'~0 (p~)gjU (T )jf'~(p~)g >= D'~(k) exp B
i
2 (k
2

k m2 + i"
~k;k0
exp
!
i X jJ~(k)j2
:
2 k k2 m2 + i"
(7.12)
We an now shift the integration of '~(k) over J~(k)=(k2 m2 + i") and introdu e the Green's
fun tion in oordinate spa e (eq. (4.7)) to get
< f'~0 (p~)gjU (T )jf'~(p~)g >=

= exp
i
2
0
Z
D'~(k)
d4 x d4 y J (x)G(x y )J (y )
exp B
i
~k;k0
1
2
(k 2
m2 + i")j'~(k)j2 C
A
i
d4 x d4 y J (x)G(x y )J (y ) < f'~0 (p~)gje iH (J =0)T jf'~(p~)g > : (7.13)
2
In the last step it is ru ial to note that the sour e is taken to vanish for t 0 and for t T ,
as otherwise the shift we performed in the eld would have hanged the boundary values.
Using eq. (7.10) we now obtain the remarkable result that

iZ
< 0jU (T )j0 >=< 0je iH (J =0)T j0 > exp
d4 x d4 y J (x)G(x y )J (y )
(7.14)
2
to all orders in the sour e J . Even at the level of a non-intera ting s alar eld theory,
this demonstrates the dramati simpli ations that an arise from using the path integral
method for al ulating quantum amplitudes. One parti ularly feature that is noteworthy
in the path integral al ulation, is that the part of the Lagrangian that is quadrati in the
elds represents the inverse propagator. This is no a ident and is in general the way the
(lowest order) propagator is dire tly read o from the Lagangian, sin e the quadrati part of
the a tion 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 su h an important role in the Hamiltonian formulation, is automati ally
implemented by the path integral. Furthermore, it will be helpful to understand in more
detail how the sour e an be used to reate and annihilate parti les, as this will be our tool
to write down the matrix elements of the evolution operator (the so- alled s attering matrix,
or for short S -matrix) with respe t to the basis spe ied by parti le number and momentum
(the so- alled Fo k spa e), see eq. (2.9).
= exp
30
In the Hamiltonian formulation we onsider
< 0j'^(~x 0 ; t2 )'^(~x; t1 )j0 > ; t2 > t1 ;

where the eld operator ( mp. eq. (2.8)) is given by
'^(~x; t) = eiHt '^(~x)e
su h that
< 0j'^(~x 0 ; t2 )'^(~x; t1 )j0 >=
iHt
iH (T t2 ) '
^(~x 0 )e iH (t2 t1 ) '^(~x)e iHt1
< 0j e
< 0j e
(7.15)
iHT
j0 >
(7.16)
j0 > ;
(7.17)
where we have made use of the fa t that the va uum j0 > is assumed to be an eigenstate of
the Hamiltonian H (without a sour e term). The normalisation by < 0je iHT j0 > is hen e a
rather trivial fa tor. We an even write a similar expression in the presen e of the sour e. In
perturbation theory this would not be needed, but it is useful from a general point of view
to onsider this situation too.
In the presen e of a time dependent sour e one an write
< 0jU (T )'^(~x 0 ; t2 )'^(~x; t1 )j0 >=< 0jU (T; t2 )'^(~x 0 )U (t2 ; t1 )'^(~x)U (t1 )j0 >
(7.18)
In this ase the eld operators are of ourse given by
'^(~x; t) = U (t)y '^(~x)U (t) ;
(7.19)
where in general U (t) depends on the sour e J . It is now trivial, but a bit tedious, to onvert
this matrix element to a path integral. One rst writes the produ t of the operators as a
produ t of matri es in a suitable representation (e.g. the eld representation jf'~(p~)g >).
Ea h of the matri es for the three evolution operators an be written as a path integral,
ex luding the integral over the initial and nal eld omponents. The matrix produ t involves
an integral over the nal eld omponent of the matrix to the right, whi h is also the initial
eld omponent for the matrix to the left. Without the insertion of the eld operators
'^(x), this would des ribe the fa t that U (t3 ; t2 )U (t2 ; t1 ) = U (t3 ; t1 ) in the path integral
formulation simply means that one glues the paths in U (t3 ; t2 ) and U (t2 ; t1 ) together by
integrating over '~(~k; t2 ). With the eld operator sandwi hed between the two U matri es
one simply in ludes its eigenvalue in the integrand over the paths, sin e the eld operator
(or its Fourier omponents) is diagonal on the states jf'~(p~)g >. The nal result an be
written as
Z
p~
d'~(p~)d'~0(p~) < 0jf'~0 (p~)g > f
D'(x) exp(i
Z
t2
t1
Z t
2
0
L d4 x)'(~x ; t2 ) exp(i L d4x)
t1
L d4x) g < f'~(p~)gj0 > ; (7.20)

0
where we impli itly assumed that the boundary onditions for the eld ' in the path integral
is in momentum spa e given by '~(p~; t = 0) = '~(p~) and '~(p~; t = T ) = '~0 (p~). The way the
time ordering in the path integral is manifest is now obvious. Note that the eld expe tation
values an also be written in terms of derivatives with respe t to the sour es, whi h is
parti ularly simple to derive in the path integral formulation
2
0
< 0jU (T )'^(~x ; t2 )'^(~x; t1 )j0 >=
< 0j U ( T ) j 0 > :
(7.21)
J (~x 0 ; t2 )J (~x; t1 )
'(~x; t1 ) exp(i
31
Sin e we assumed that j0 > is an eigenstate of H (i.e. at J = 0), < 0jUJ =0 (T ) =< 0je iE0 T
(with E0 the va uum energy) and we an bring the trivial phase fa tor e iE0 T to the other
side by normalising with < 0jUJ =0 (T )j0 >, as in eq. (7.17).
"
2 < 0j U ( T ) j 0 > #
:
(7.22)
< 0j'^(~x 0 ; t2 )'^(~x; t1 )j0 >J =0 = < 0jU (T )j0 > 1
J (~x 0 ; t2 )J (~x; t1 ) J =0
where in the path integral formulation one has
< 0jU (T )j0 >=
p~
d'~(p~)d'~0 (p~) < 0jf'~0(p~)g >
D'(x) exp(i
L d4x) < f'~(p~)gj0 >;

(7.23)
with the boundary onditions as listed below eq. (7.20).

To study the role the sour e plays in reating and annihilating parti les, we will al ulate
both in the Hamiltonian and in the path integral formulations the matrix element
< p~jU (T )j0 > ;
jp~ > ay(p~)j0 > :
(7.24)
Hen e jp~ > is the one-parti le state with momentum ~p. Using the result of eqs. (5.7) and
(5.8), whi h is equivalent to the result j (t) >= U (t)j0 >, we nd for this matrix element
in lowest non-trivial order
< p~jU (T )j0 > =

=
=
dt < 0ja(p~)ei(t T )H0 H1 (t)e
itH0
j0 >
(a( ~k) + ay (~k)) iE0 t

dt < 0ja(p~)ei(p0 (p~)+E0 )(t T ) J~(~k; t) q
e
j0 >
~k 0
2k0 (~k)
pe i(E0 +p0(p~))T
i(E0 +p0 (p~))T Z T
e q
ip
~
)
t
0 (p
~
q
J~(p) : (7.25)
dt J (p~; t)e
= i
i
0
2p0 (p~)
p0 (p~)
To write down the path integral result, we rst express the annihilation operator in terms
of the eld. From eq. (5.3) we nd
s
a(p~) + ay ( p~) =
su h that (using < 0jay ( p~) = 0)
2p0 (p~) Z
d3~x '^(~x)e i~p~x ;
V
(7.26)
2p0 (p~) Z
d3~x e i~p~x < 0j'^(~x)U (T )j0 > :
(7.27)
V
We now use eqs. (7.14), (7.19) and an obvious generalisation of eq. (7.21), su h that
< ~pjU (T )j0 >=
< 0j U ( T ) j 0 >
J (~x; T )

iZ
iE
0T
= ie
d xd y J (x)G(x y )J (y ) :(7.28)
exp
J (~x; T )
2 4 4
We evaluate this to linear order in the sour e J , using that in a nite volume the Green's
fun tion is given by
1 X Z dk0 e ik(x y)
;
(7.29)
G(x y ) =
V ~k 2 k2 m2 + i"
< 0j'^(~x)U (T )j0 > = < 0jU (T )'^(~x; T )j0 >= i
32
su h that
s
Z
2p0 (p~) Z
d3~x e i~p~x d4 y G(x y )J (y ) + O(J 3 )
V
s
p0 (p~) Z
J~(p)e ip0 T
iE
0T
= e
dp0 2
+ O (J 3 ) :
(7.30)

p m2 + i"
< p~jU (T )j0 > = e
iE0 T
Note that in the last step we integrate over p0 as a dummy variable, whi h in the expression
for the Green's fun tion above is alled k0 - this renaming is just for ease of notation. Also,
x0 = T is assumed. For the p0 integration we need the analyti behaviour of J~(p) for
imaginary p0 in order to see if we are allowed to deform the integration
ontourpsu h that
R
only one of the poles in the integrand ontribute. Sin e J~(p) = 0T dt eip0 t J~(p~; t)= 2, J~(p)
will vanish for Imp0 ! 1, whereas e ip0 T J~(p) will vanish for Imp0 ! 1. In eq. (7.30)
the p0 integration an therefore be deformed to the lower half-plane
in a lo kwise fashion
p
2
giving a minus sign and a residue from the pole at p0 = p0 (p~) p~ + m2 , whi h yields the
result
pe i(E0 +p0(p~))T

iZ
q
< p~jU (T )j0 >= i
J~(p) exp
d4 xd4 y J (x)G(x y )J (y ) : (7.31)
2
p0 (p~)
To linear order in the sour e J this oin ides with the result of eq.(7.25). Again, the path
integral trivially allows an extension to arbitrary order in the sour e, as indi ated.
For later use, we will also onsider the matrix element
s
2p0 (p~) Z
d3~x ei~p~x < 0jU (T )'^(~x; 0)j0 >
V
s
2p0 (p~) Z
< 0jU (T )j0 >
= i
d3~x ei~p~x
:
V
J (~x; 0)
< 0jU (T )jp~ > =
(7.32)
The analogue of eq. (7.30) be omes

s
Z
2p0 (p~) Z
d3~x ei~p~x d4 y G(x y )J (y ) + O(J 3 )
V
s
p0 (p~) Z
J~( p~; p0 )
iE
0T
= e
+ O (J 3 ) ;
(7.33)
dp0 2

p m2 + i"
< 0jU (T )jp~ > = e
iE0 T
in whi h ase x0 = 0 is assumed. Now we must deform the ontour for the p0 integration
to the upper half-plane in a ounter- lo kwise fashion su h that the residue at the pole
p0 = p0 (p~) ontibutes. This gives, analogously to eq. (7.31),
< 0jU (T )jp~ >= i
pe
q
iE0 T
p0 (p~)
J~( p) exp

iZ
d xd y J (x)G(x y )J (y ) :
2 4 4
(7.34)
Apart from the trivial dieren e of the fa tor exp( ip0 T ), we see that the amplitude for
the annihilation of a one-parti le state is proportional to J~(p) whereas it is proportional
(exa tly with the same fa tor) to J~( p) for the reation of a one-parti le state (in both
ases p0 = p0 (p~)).
33
8 Perturbative expansion in eld theory

As we have seen in the previous se tion (
D'(x) where relevant in ludes groundstate fa tors)
Z (J; gn ) < 0jU (T )j0 >=
D'(x)
exp(i d4 x L('))
(8.1)
will play the role of a generating fun tional for al ulating expe tation values of produ ts of
eld operators, whi h will now be studied in more detail. In general the Lagrange density
for a s alar eld theory is given by
L(') = L2(') V (') J (x)'(x) ;
(8.2)
where L2 (') is quadrati in the elds, hen e for a s alar eld

L2(') = 21 '(x) '(x) m2 '2 (x) ; V (') = g3!3 '3 (x) + g4!4 '4 (x) + : (8.3)
As mentioned before, it is ustomary to not in lude the mass term in the potential V ,
su h that V des ribes the intera tions. We an add the intera tion as an operator, when
evaluating the path integral for the quadrati approximation L(') = L2 (') '(x)J (x),
Z
Z (J; gn) =
D'(x)
Z
exp i d4 x
= < 0jU2 (T )Texp
fL2(') '(x)J (x)g

i d4 x V ('^) j0 >
exp
i d 4 x V ( ')
(8.4)
We an now use the fa t that

Z
Z
D'(x) '(x(j) ) exp i d4x (L2(') '(x)J (x))

Y
j
j
Z

Z
i
D'(x) exp i d4x (L2(') '(x)J (x))
J (x(j ) )
!

iZ
i
exp
d xd y J (x)G(x y )J (y ) ;
J (x(j ) )
2 4 4
(8.5)
to nd a somewhat formal, but in an expansion with respe t to the oupling onstants gn

well dened expression for the fully intera ting path integral
Z (J; gn ) = exp
exp

i
iZ
i d4 x V (
d4 xd4 y J (x)G(x y )J (y )
) exp
J (x)
2
!
Z
i
) Z2 (J ) ;
(8.6)
i d4 x V (
J (x)
Z
We have assumed the va uum energy to be normalised to zero, in absen e of intera tions, su h
that Z (J = gn = 0) = 1. Equivalently, Z (J; gn) is synonymous with Z (J; gn )=Z (J = gn = 0).
We now dene GJ as
GJ log Z (J; gn) ;
(8.7)
where the dependen e on the oupling onstants in GJ is impli it. We will show that
GJ an be seen as the sum of all onne ted diagrams. A diagram is onne ted if it an
not be de omposed in the produ t of two diagrams that are not onne ted. Note that at
34
J = 0, iGJ =T equals the energy of the ground state as a fun tion of the oupling onstants,
normalised so as to vanish at zero ouplings.
The dierent diagrams arise from the expansion of
0
Z
X g
i
i
`
i d4 x V (
) = exp i d4 x
J (x)
`=3 `! J (x)
Z
exp
! 1
(8.8)
in powers of g`. Ea h fa tor g``! (i=J (x))` will represent an `-point vertex, with oordinate
x, whi h is to be integrated over. As we saw in the derivation of the lassi al equations
of motion, the integral over x in the Fourier representation gives rise to onservation of
momentum at the vertex. Using
' ( x) =
1 Z
dke
(2 )2 4
ikx '
~( k )
; J (x) =
1 Z
dke
(2 )2 4
ikx J~(k )
(8.9)
we have for ea h vertex in the Fourier representation
ig`
i
d4 x
J (x)
X
= ig` (2 )4 4 ( k(j ) )
j =1
d4 k(j ) i
2 ~
j =1 (2 ) J (k(j ) )
Y
(8.10)
Note that fa tors of 2 are dropping out in the identities

exp
!

iZ
iZ
J~(k)J~( k)
d xd y J (x)G(x y )J (y ) = exp
dk
;
2 4 4
2 4 k2 m2 + i"
Z
Z
d4 x '(x)J (x) = d4 k '~(k)J~( k) :
(8.11)
In the quantum theory we have to keep tra k of the fa tors i. Compared to the Feynman
rules of se tion 4, the propagator will ome with an extra fa tor
will now
p i. A2 vertex
`
2
2
`
arry a fa tor ig` ((2 ) =i) (in a nite volume this be omes ig` ( 2V =i) ), see the table
below. To ompute the va uum energy there is an overall fa tor i, sin e E0 T = iGJ =0 =
i log Z (J = 0; gn). The same fa tor of i applies for using the tree-level diagrams to solve
the lassi al equations of motion. It is easy to see that these Feynman rules give identi al
results for these tree-level diagrams, as ompared to the Feynman rules introdu ed in se tion
4. The fa tors of i exa tly an el ea h other.
oordinate spa e
xHH
i`
y
x
1 g` R d4 x
iG(x y )
R
d4 xJ (x)
momentum spa e
k1
HHj6

k2
k3
k
k
table 2
P
i` 1 (2 )4 2` g`4 ( i ki ) vertex
R
d4 k k2
J~(k)
i
m2 +i"
propagator
sour e
We note that the propagator onne ted to a sour e omes down whenever a derivative
in the sour e a ts on Z2 (J ), see eq. (8.7). When this derivative a ts on terms that have
already ome down from previous derivatives, one of the sour es onne ted to a propagator
35
is removed and this onne ts that propagator to the vertex asso iated to =J (x). As any
derivative is onne ted to a vertex, the propagator either runs between a vertex and a sour e
H q , between two verti es H
q qH , or it onne ts two legs of the same vertex q j. The
possibility of losed loops did not o ur in solving the lassi al equations of motion, and is
spe i to the quantum theory.
To prove that GJ only ontains onne ted diagrams we write
GJ = log fexp(X (i=J )) exp(Y (J ))g 1 ;
(8.12)
with
Z
iZ
X (i=J (x)) i d4 x V (=J (x)) ; Y (J )
d xd yJ (x)G(x y )J (y ) : (8.13)
2 4 4
As J (x) and =J (x) form an algebra (similar to the algebra of x^ and p^ in quantum me hani s,
however, generalised to innite dimensions), also X and Y are elements from the algebra, and
we an express GJ in a sum of multiple ommutators using the Campbell-Baker-Hausdor
formula (see eq. (6.44)).

1
1
1
GJ = X + Y + 2 [X; Y + [X; [X; Y + [Y; [Y; X + 1 :
(8.14)
12
12
Due to the multiple ommutators, all omponents that do not ommute are onne ted.
However, if the omponents would ommute they would not ontribute to the ommutators.
This is even true if we do not put the derivatives with respe t to the sour e to zero, on e
they have been moved to the right (this is why we onsider the a tion on the identity).
The only thing that remains to be dis ussed is with whi h ombinatorial fa tor ea h
diagram should ontribute. This is, as in se tion 4, with the inverse of the order of the
permutation group that leaves the topology of the diagram un hanged. These ombinatorial
fa tors are learly independent of the spa e-time integrations and possible ontra tions of
ve tor or other indi es. We an he k them by redu ing the path integral to zero dimensions,
or '(x) ! ' and D'(x) ! d'. In other words, we repla e the path integral by an ordinary
integral. As an example onsider

Z
g 3
1
Z (J; g ) = C d' exp if 2 'M'
' 'J g
3!
1
0
!3

i
i
g
A exp
JM 1 J :
(8.15)
= exp i
3! J
2
The onstant C is simply to normalise Z (J = gn = 0) = 1. Expanding the exponents we get

in lowest non-trivial order
!
3
g2
i 6 i
1
Z (J = 0) = 1
JM J
2(3!)3 J
2
0
1
!6
!
2

3
g

5g 2
1
3
3

A
= exp i 4 3
JM J + O(g ) = exp i
+ O (g )
2 (3!) J
24M 3

1 m 1 m m
3
+
+ O (g ) :
(8.16)
= exp
12
8
In the last term, the numeri al fa tors in front of the diagrams indi ate the ombinatorial
fa tors (for the rst diagram a fa tor 2 from inter hanging the two verti es and a fa tor 3!
36
from inter hanging the three propagators, for the se ond diagram the latter fa tor is repla ed
by 4 as we an only inter hange for ea h vertex the two legs that do not inter onne t the
two verti es). The Feynman rules for this simple ase are that ea h vertex gets a fa tor g
(in zero dimensions there are no fa tors 2 ) and ea h propagator gets a fa tor i=M . In
problem 13 the exponentiation is he ked for Z (J ) to O(g 2 ) and O(J 5 ) (giving the simplest
non-trivial he k).
We will now show how the number of loops in a diagram is related to the expansion in
h . We an expe t su h a relation, as we have shown at the end of se tion 6 that the h ! 0
limit is related to the lassi al equations of motion, whereas we have shown in se tion 4 that
these lassi al equations are solved by tree diagrams. If we all L the number of loops of a
diagram, we will show that

1
X
iZ
Z (J; gn ) = D' exp
(L(') 'J ) exp(GJ =h ) ; GJ = h L GL;J ; (8.17)
h
L=0
where GL;J is the sum of all onne ted diagrams with exa tly L loops. This means that a
loop expansion is equivalent with an expansion in h . To prove this we rst note that due to
reinstating h the sour e term will get an extra fa tor 1=h , the propagator a fa tor h and the
oupling onstants gn are repla ed by gn =h . A diagram with Vn n-point verti es, E external
lines ( onne ted to a sour e) and P propagators has therefore an extra overall fa tor of
Z
h E h P
n=3
h
Vn
(8.18)
We an relate this to the number of loops by noting that the number of momentum integrations (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 onservation 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 fun tions oming from the verti es,
P
i.e. P
n=3 Vn . Hen e
X
Vn ;
(8.19)
L=1+P E
n=3
whi h implies that the total number of h fa tors in a diagram is given by L 1. In the next
se tions we will often onsider so- alled amputated diagrams, where the external propagators
onne ted to a sour e are taken o from the expressions for the diagram. IfPwe do not ount
these external propagators, eq. (8.19) has to be repla ed by L = 1 + P
n=3 Vn , as there
are exa tly E su h external propagators.
9 The s attering matrix

We would like to ompute the amplitude for the transition of n in oming parti les at t = Tin
to ` outgoing parti les at t = Tout in the limit where Tout ! 1 and Tin ! 1. The
dieren e with quantum me hani s is that the parti le number is no longer onserved.
out< ~p1 ; ~p2 ; ; ~p`j~k1 ; ~k2 ; ; ~kn >in< p~1 ; ~p2 ; ; ~p`jU (Tout ; Tin )j~k1; ~k2 ; ; ~kn > : (9.1)
In terms of reation and annihilation operators this an be written as
out< ~p1 ; p~2 ; ; ~p` j~k1; ~k2 ; ; ~kn >in =
< 0ja(p~1 )a(p~2 ) a(p~` )U (Tout ; Tin )ay(~k1 )ay (~k2 ) ay (~kn )j0 > : (9.2)
37
From eqs. (7.26), (7.28) and (7.32) we know how to implement these reation and annihilation
operators on the generating fun tional Z (J; gn )
s
= i 2p0 (p~) ~
;
J (~x; t = Tin )
J (p~; t = Tin )
q
= i 2p0 (p~) ~
: (9.3)
d3~x e i~p~x
J (~x; t = Tout )
J ( p~; t = Tout )
2p0 (p~) Z
y
a^ (p~) = i
d3~x ei~p~x
V
2p0 (p~) Z
a^(p~) = i
V
s
This implies the following identity for the s attering matrix

out< ~p1 ; ~p2 ; ; ~p`j~k1 ; ~k2 ; ; ~kn >in =
i=1
a^(p~i )
j =1
a^y (~kj ) exp(GJ )jJ =0 :
(9.4)
In prin iple, this allows us to al ulate the s attering, taking Tin ! 1 and Tout ! 1.
There is, however, a problem to asso iate the parti le states in the presen e of intera tions
with the ones we have derived from the non-intera ting theory. The problem is that parti les
an have self-intera tions long before and after the dierent parti les have s attered o ea h
other. We have to re onsider our notion of parti le states, as in experiments we are unable
to swit h o these self-intera tions. For simpli ity we assume that the one-parti le states
are stable, as in the simple s alar theory we have been onsidering. This implies from
onservation of probability that
Z
d3~k j out< p~j~k >in j2 = 1 ;
(9.5)
independently of p~. In general, onservation of probability implies that the S -matrix is

unitary. Formally, unitarity of a S -matrix is guaranteed as soon as the Hamiltonian is
hermitian. Be ause of the ne essity to regulate the quantum theory, e.g. by introdu ing a
ut-o, this is generally no longer true and one has to show that unitarity is restored when
the ut-o is removed. If this is not possible, the theory is ill-dened, or at best does not
make sense above the energies where unitarity is violated.
For the free theory, unitarity is of ourse satised. In this ase, the only diagram ontributing to GJ is the one with a single propagator onne ting two sour es, whi h is also
alled the onne ted two-point fun tion G(2)
(J )
iZ
GJ = G(2)
(
J
)
=

2
This implies that (T
dtds
p~
dp0
e
2
~
p2
~ p~; s)
:
m2 + i"
~; t)J (
ip0 (t s) J (p
(9.6)
Tout Tin)
~)~k;~p
out< ~pj~k >in = a^(p~)ây (~k) exp(G(2)
(J ))jJ =0 = 2ip0 (p
dp0
eip0 T
=e
2 p2 m2 + i"
ip0 (p~)T
~k;~p
;
(9.7)
where the p0 integration is performed by deforming the ontour to the upper half-plane,
giving a ontribution from the pole at p0 = p0 (p~) only. It is trivial to see that this is the
same as what an be obtained within the Hamiltonian formulation.
In the presen e of intera tions, this result is no longer true, sin e the onne ted two-point
fun tion will deviate from the one in the free theory. In this ase we an write (from now on
38
the symmetry fa tors will be absorbed in the expression asso iated to a diagram)
G(2)
(J ) =

+
;

+
(9.8)
where we have written the onne ted two-point fun tion in terms of the one-parti le irredu ible (1P I ) two-point fun tion i(p) ( is the so- alled self-energy)

i(p)

(9.9)
In general, a 1P I -graph is a onne ted graph that remains onne ted when one arbitrary
propagator is being ut (ex ept when utting away a tadpole of the form q j, whi h we
will not allow. However, these tadpoles des ribe single parti les popping in or out of the
va uum, usually required to be absent. They an be removed by shifts in the elds.) The
external lines of these diagrams will arry no propagator. The diagrams of eq. (9.8) an be
onverted to the result
9
8
!2
!3
Z
<
i
i
i
2 + =
2
1
~
+
i
(
p
)
+
(
i
(
p
))
d
p
j
J
(
p
)
j
4
2
;
: p2
m2 + i"
p2 m2 + i"
p2 m2 + i"
!n )
(
Z
1
X
(p)
i
= 21 d4 p jJ~(p)j2 2
p m2 + i" n=0 p2 m2 + i"
jJ~(p)j2
iZ
d4 p 2
G(2)
(9.10)
=
(J ) :
2
p m2 (p) + i"
Normally the self-energy will not vanish at p2 = m2 , su h that the self-intera tions shift the
pole in the two-point fun tion to another value, m
~ 2 , i.e.
p2 m2 (p) = 0 for p20 = ~p 2 + m
~2 :
(9.11)
Consequently, the mass of the one-parti le states is shifted (or renormalised). As we annot
swit h o the intera tions in nature, the true or observable mass is m
~ and not m, the latter is
also alled the bare mass. The residue at the poles (i.e. p2 = m
~ 2 , alled the mass-shell) will
in general also hange from i=p0 (p~) to iZ=p0 (p~). On the mass-shell (i.e. J~(p) vanishes
rapidly as a fun tion of jp2 m
~ 2 j) one therefore has
iZ
Z jJ~(p)j2
G(2)
(
J
)
=
d
p
:
(9.12)

2 4 p2 m
~ 2 + i"
As long as the one-parti le states are stable, eq. (9.5) needs to remain valid,
p whi h an only
be a hieved (see eq. (9.7)) by res aling the wave fun tionals with a fa tor Z (this is alled
wavefun tion renormalisation). It implies that eq. (9.4) needs to be modied to
out< ~p1 ; ~p2 ; ; ~p`j~k1 ; ~k2 ; ; ~kn >in=
where
2 p~ 2 + m
~2
a^+ (p~) i
Z
!1
i=1
a^ (p~i )
j =1
a^+ (~kj ) exp(GJ )jJ =0 ;
(9.13)
!1
p
~2 2
2 p~ 2 + m
; a^ (p~) i
:
~
~
Z
J (p~; t = Tin )
J ( p~; t = Tout )
(9.14)
39
As for the free theory, ea h of these operators a^ (p~) will repla e one external line (propagator
plus sour e) by an appropriate wavefun tion fa tor and puts these external lines on the
mass-shell. We will all the onne ted n-point fun tion with amputated external lines the
amputated onne ted n-point fun tion G( amp) (p1 ; p2 ; ; pn) (in general not one-parti le
irredu ible), i.e.
)
n (Z
Y
iJ~(pi )
(
n
)
G( amp) (p1 ; p2 ; ; pn ) :
(9.15)
d4 pi 2
G (J )
2
pi m (pi ) + i"
i=1
Diagrammati ally this looks as follows
H

H...H
..
.

...
.
HH..

H

'$

(9.16)

&%

To get the S -matrix we have to ompute (see eq. (9.13))

0
1 n (qr ) or A

a^ (p~i ) a^+ (~kj ) eGJ A = eGJ =0 a^ (p~i ) a^+ (~kj )
:
G (J )
r
!
r
j
=1
i
=1
j =1
i=1

rq
=
`
+
n
r
J =0
(9.17)
The sum is over all possible partitions of ` + n. Let us rst onsider the most important
term, orresponding to onne ted graphs, where q1 = ` + n
Y
`
n
Y
(i)
(
j) Y
(
`
+
n
)
~
i Z=p(0i) G( amp) (f pi g; fkj g)e ip0 T
a^ (p~i ) a^+ (kj ) G (J )=
i Z=k0
i=1
i=1
j =1
j =1
P`
Pn
2iV 4 ( qi=1Qp` i (ij)=1 Qkjn) M`+(jn) (f pi g; fkj g) e ip0(i)T :
(9.18)
2
p
V
2
k
V
i=1 0
j =1 0
Here M` (a Lorentz s alar as we will see later) is the so- alled redu ed matrix element with
` external lines, all
on the mass-shell.
Note
that we have extra ted
the trivial energy fa tors
P
P (j )
P
P
(remember that p(0i) Tout
k0 Tin = p(0i) (Tout Tin ) p(0i) T ), su h that the limits
Tin ! 1 and Tout ! 1 an be taken. Ea h a^(p~) will a t on one of the fa tors between
urly bra kets in eq. (9.15). Con entrating on one su h a fa tor we have
Z
iJ~(p)G( amp) (p; p2 ; ; pn) Z d4 p Z
iJ~(p~; t)eip0 t G( amp) (p; p2 ; ; pn)
d4 p
dt
= p
:
p2 m2 (p) + i"
p2 m2 (p) + i"
2
(9.19)
su h that, using eq. (9.14)
s
Z
eip0 Tout G( amp) ((p0 ; p~); ; pn )
p0 (p~) Z
iJ~(p)G( amp) (p; p2 ; ; pn)
dp0
=
a^ (p~) d4 p
p2 m2 (p) + i"
Z
p2 m2 (p) + i"
) ( p; p ; ; p )
p
G( ampq
2
n
e ip0 (p~)Tout :
(9.20)
= i 2V Z
2p0 (p~)V
`
Sin e Tout ! 1, we an extend the p0 ontour integration to the upper half-plane, under
mild regularity onditions for G( amp) (p; p2 ; ; pn ) as Imp0 ! 1 (that an easily be shown
40
to be satised at any nite order in perturbation theory).

the integral over p0 only
pp~ 2 Thus,
gets a ontribution from the pole at p0 = p0 (p~) =
+m
~ 2 , with residue Z=2p0 (p~)
( mp. eq. (9.12)). For the reation operators one similarly nds
s
Z
iJ~(p)G( amp) (p; p2 ; ; pn)
eip0 Tin G( amp) ((p0 ; ~p); ; pn )
p0 (p~) Z
a^+ (p~) d4 p
dp0
=
p2 m2 (p) + i"
Z
p2 m2 (p) + i"
p
G( ampq) (p; p2 ; ; pn ) ip0 (p~)Tin
e
:
(9.21)
= i 2V Z
2p0 (p~)V
Likewise, as Tin ! 1, we an now extend the p0 ontour integration to the lower half-plane
(under the same regularity onditions for the amputated n-point fun tions as for eq. (9.20)
to be valid), su h that we pi k up the ontribution of the pole at p0 = p0 (p~) with residue
Z=2p0 (p~). Combining these results proves the rst identity in eq. (9.18), the se ond is merely
a denition.
Note that the derivation is not valid for the ase ` + n = 2, where the generalisation of
eq. (9.7) to the intera ting ase implies
(9.22)
out< ~pj~k >in = a^ (p~)â+ (~k) exp(GJ )j = 3 (~k ~p)e i[E0 +p0 (p~)T :
J =0
Here we used eq. (9.12) and the fa t that exp(GJ =0 ) = exp( iE0 T ), whi h is often also
normalised to 1, but till now we had only required this to be the ase at zero ouplings.
The reason this ase is spe ial is be ause eq. (9.15) requires us to dene for the amputated
two-point fun tion
G( amp) (p1 ; p2 ) i4 (p1 + p2 )(p21 m2 (p1 ) + i") ;
(9.23)
whi h vanishes on the mass-shell. In using eq. (9.20) and eq. (9.21) it was impli itly assumed
that the amputated Green's fun tion has no zero that will an el the pole.
The Feynman rules in momentum spa e for omputing the redu ed
p matrix elements will
obviously have to be modied for the external lines to a fa tor i 2V Z and an overall
fa tor i=(2V ) (as always, in an innite volume one repla es V by (2 )3 ). If we asso iate a
momentum delta fun tion to a vertex and a momentum integration to a propagator (as was
done up to now), the delta fun tion for overall energy and momentum onservation should
not be written expli itly in eq. (9.18), sin e it is ontained in the redu ed matrix element.
Instead, if we hoose to integrate over the independent loop momenta, implementing energy
and momentum onservation at ea h vertex (as will be done from now on, see the table
below), the denition of eq. (9.18) is the appropriate one. The overall fa tors of i, 2 and
V an be determined with the help of the two identities
L=P +1
V n ; E + 2P =
nVn :
(9.24)
The proof for the rst identity was dis ussed below eq. (8.19). For the se ond identity we
put a dot on ea h end of a propagator ( ` ` ) and one dot on ea h external line ( ` ` ),
giving` a total of P
2P + E dots. The same dots an also be asso iated to ea h line of a vertex
( `H` ), giving nVn dots, thus proving the se ond identity (see also problem 15). To keep
the derivation general, we evaluate the overall fa tor in a nite volume
P i E +1 i(n
1)Vn (2V ) 12 (n 2)Vn (2V ) 21 E 1 =
41
i L
:
2V
(9.25)
This implies that we an shift all numeri al fa tors from the propagators, verti es and external lines to a fa tor i=(2V ) (or i=(2 )4 in an innite volume) for ea h loop, giving the
Feynman rules listed in the table for an innite volume. Note that the extra tion of the fa tor
2iV in the denition of M is merely a onvention (su h that in lowest order Mn = gn ).
momentum spa e
k1
HHj6

k2
k3
g` and
1
k2 m2 +i"
Itzykson and Zuber
ki = 0
k1
HHj6

k2
k3

i (2d4 k)4
R

1
table 3
i(2 )4 g`4 (
d4 k
(2)4 k2
i
m2 +i"
i ki )
vertex
propagator
external line
loop fa tor
In the literature many dierent onventions are being used. As an example, the table ompares our Feynman rules with those of Itzykson and Zuber. Their onvention for Mn is
likewise to make it oin ide to lowest order with the n-point vertex. However, as the latter
does already ontain a fa tor i(2 )4 (in a nite volume 2iV ), that fa tor should be
absent in relating the redu ed matrix element to the amputated n-point fun tion. Combining the extra fa tors of i and 2 in the Feynman rules of Itzykson and Zuber gives
i(2 ) 4 iP Vn (2 )4(Vn P ) = [i=(2 )4 L , guaranteeing equivalen e of the two sets of Feynman rules.
Con erning the symmetry fa tor asso iated to a parti ular diagram we note the following.
As we have generally xed the external momenta, inter hanging external lines is no longer
allowed. But from eq. (9.15) we see that the symmetry fa tor n!, to be taken into a ount
for G( n) (J ), will be ompensated by the n derivatives on n sour es. Hen e, in omputing
the redu ed matrix elements the symmetry fa tors are determined without allowing for
permutations on the external lines.
To on lude this se tion we return to eq. (9.16), and dis uss the ontributions that will
be asso iated to the diagrams
that are
not onne ted. Ea h fa tor of r! is ompensated
n
or
(
q
)
r
for by the dierentiations on G (J ) . For the orresponding onne ted omponents the
rules are identi al to the ones spe ied
above. In parti ular ea h onne ted omponent will
P
arry its own fa tor i(2 )4 4 ( pi ) for the onservation of energy and momentum (pi is
now assumed to run over a subset of both the in oming and the negative of the outgoing
four-momenta). In a physi al pi ture the dis onne ted parts orrespond to situations where
only a subset of the in oming parti les will intera t with ea h other (the ones onne ted
by a parti ular diagram). Quite often, the experimental situation is su h that the energymomentum onservation will only be ompatible with the fully onne ted part. We just
have to avoid the in oming momenta to oin ide with any of the out oming momenta. In a
ollider, this means one ex ludes parti les that es ape in the dire tion of the beams, where
indeed it is not possible to put a dete tor. As an illustration we will give the situation for
n = ` = 2 and all momenta non-zero (to avoid tadpole diagrams, q j) to se ond order in
42
the three-point oupling g3 , putting all other ouplings to zero

(1) + p(2) T
~
~
<
p
~
;
p
~
j
k
;
k
>
=
exp
i
[
E
+
p
out 1 2 1 2 in
0 0
0
"
+ 3 (p~1 ~k2 )3 (p~2 ~k1 )

3 (p~1 ~k1 )3 (p~2 ~k2 )
(2 )4 4 (p1 + p2 k1 k2 )
i q (1)
+
(2)
(1)
(2)
3
3
3
3
2p0 (2 ) 2p0 (2 ) 2k0 (2 ) 2k0 (2 )
(9.26)
XXXX
PPPP + O(g33)
!
!#
The rst two diagrams, whi h have to be treated with spe ial are (see eqs. (9.22) and
(9.23)), represent the situation without s attering. By denition they have no higher order
orre tions.
10 Cross se tions
In many experimental situations we are interested in the s attering of two parti les with
momenta
k1 and k2 to a Rstate with n parti les with momenta p1 ; p2 ; ; pn. We denote
R
~
~
~ 2 (~k2 )j~k2 > the wave fun tionals of the in oming parti les.
by d3 k1 1 (~k1 )j~k1 > and d3~k2
This is to des ribe the more realisti ase of a wave pa ket. The amplitude for s attering to
take pla e is hen e given by
P
Z
q4 ( ni=1 pi k1 k2 )M2+n(f pi g; fkj g) ~ ~ ~ ~ i[E0 +p(0i) T
4
~
~
A = i(2) d3k1d3k2 Qn (i)
:
Q2
(j ) (~k )(2 )3 1 (k1 ) 2 (k2 )e
3
2
p
(
p
~
)(2

)
2
k
i
j
i=1 0
j =1 0
(10.1)
If we dene the wave fun tion in oordinate spa e as usual
Z
d ~k
~ j (~k) ;
j (x) = q 3
e ikx
(10.2)
3
~
2k0 (k)(2 )
we an ompute the overlap of the two wave fun tions
Z
Z
d ~k
d3~k2
q
d4 x 1 (x) 2 (x)eipx = (2 )4 q 3 1
4 (k1 + k2 p) ~ 1 (~k1 ) ~ 2 (~k2 ) :
2k0 (~k1 )(2 )3 2k0 (~k2 )(2 )3
(10.3)
We assume that over the range of momenta in the wave pa kets, the redu ed matrix elements
are onstant (whi h an be a hieved with arbitrary pre ision for arbitrarily narrow wave
pa kets in momentum spa e). This allows us to write for the s attering probability of two
parti les into n parti les, with momenta in between pi and pi + dpi ,
n
Y
d3 p~i
2

;
dW = jM(f pi g; fkj g)j f (p)
2p0 (p~i )(2 )3
i
=1
Z
f (p) d4 xd4 y 1 (x) 2 (x) 1 (y ) 2(y )eip(x y) ;
(10.4)
P
where p = ni=1 pi = k1 + k2 . The momenta ki in the redu ed matrix element are the entral
values of the wave pa ket in momentum spa e for the two in oming parti le beams. Under
the same assumption that the momentum spread in the beams is very small, the fun tion
f (p) will be highly peaked around p = k1 + k2 , su h that
Z
Z
f (p) 4 (p k1 k2 ) d4 p f (p) = (2 )4 4 (p k1 k2 ) d4 xj 1 (x)j2 j 2 (x)j2 : (10.5)
43
The quantities j j (x)j2 are of ourse related to the probability densities of the two parti les
in their respe tive beams,
j (x) i( j (x)t j (x) j (x)t j (x)) 2k0(j ) j j (x)j2 ;
(10.6)
again using the fa t that the wave pa ket is highly peaked in momentum spa e. Putting
these results together we nd
Z
n
X
Y
(x) (x)
d3 ~pi
4
2

d4 x 1 (1) 2(2) : (10.7)
dW = (2 ) 4 ( pi k1 k2 )jM(f pi g; fkj g)j
3
~i )(2 )
4k0 k0
i=1 2p0 (p
Sin e j (x) will depend on the experimental situation, we should normalise with respe t to
the total number of possible intera tions in the experimental setup, also alled the integrated
luminosity L.
Z
Z
dt L(t) d4 x 1 (x)2 (x)j~v1 ~v2 j :
(10.8)
R
Here d3~x 1 (x)2 (x) is the number of possible intera tions per unit volume at a given
time and j~v1 ~v2 j is the relative velo ity of the two beams. We have assumed that either,
one of the velo ities is zero
(xed target), or that the two velo ities are parallel ( olliding
R
beams). Hen e, L(t) = d3~x 1 (~x; t)2 (~x; t)j~v1 ~v2 j is a ux, typi ally of the order of
1028 1033 m 2 s 1 . To onsider the general ase we note that we an also write
Z
Z
q
(x) (x)
2
2
2

dt L(t) = (k1 k2 ) m1 m2 d4 x 1(1) 2(2)
:
(10.9)
k0 k0
After all, for a xed target situation k2 = (m2 ; ~0), su h that
q
q
(k1 k2 )2 m21 m22
(k0(1) )2 m21
=
= j~v1 j ;
(10.10)
k0(1) k0(2)
k0(1)
whereas for olliding beams of parti les and antiparti les with mass m, where k1 = (E; ~k) =
E (1; ~v) and k2 = (E; ~k) = E (1; ~v ), one nds
q
q
(k1 k2 )2 m21 m22
(E 2 + ~k 2 )2 m4
= 2j~kj=E = 2j~v j ;
(10.11)
=
E2
k0(1) k0(2)
su h that both expressions redu e to j~v1 ~v2 j. We leave it as an exer ise to prove the result
for the general ase of parallel beams.
We an therefore dene a ma hine independent dierential ross se tion d by normalising
the s attering probability by the total luminosity,
n
X
jM(f pig; fkj g)j2 S Yn
d3 ~pi
d = (2 )4 4 ( pi k1 k2 ) q
:
(10.12)
4 (k1 k2 )2 m21 m22 i=1 2p0 (p~i )(2 )3
i=1
The parameter S is the inverse of the permutation fa tor for identi al parti les in the nal
state, as a dete tor will not be able to distinguish them. This will avoid double ounting
when performingQthe phase spa e integrals. If there are nr identi al parti les of sort r in the
nal state, S = r 1=nr ! (in the present ase S = 1=n!).
Typi al ele tromagneti ross se tions, as we will ompute later, are of the order of
nanobarns (1 nb= 10 33 m2 ). With a luminosity of 1033 m 2 s 1 , approximately one ollision
event per se ond will take pla e. In the weak intera tion the ross se tions are typi ally 5
to 6 orders of magnitude smaller, su h that not more than one event per day will take pla e
in that ase.
44
11 De ay rates
The denition of the de ay rate (also alled de ay width) of an unstable parti le is best
dened by onsidering its self-energy,
1
(p) =
Z

(11.1)
where the diagram for the 1P I two-point fun tion is now to be evaluated using the Feynman
rules in table 3 (pg. 42). The relation with the self-energy follows from the fa t that, apart
form thepoverall fa tor i=(2 )4 , one has for ea h of the two external lines an extra fa tor
i(2 )2 Z as ompared to the amputated 1P I 2-point fun tion, in total one therefore has
iZ times the amputated 2-point fun tion. The latter indeed equals i(p), see eq. (9.9).
We will now onsider a simple example of a s alar eld theory with two types of elds, a
eld '(x) asso iated with a light parti le (mass m) and a eld (x) asso iated with a heavy
parti le (mass M > 2m), whi h an de ay in the lighter parti les if we allow for a oupling
between one and two ' elds,
' g ;

V (; ') = 12 g'2 ;
(11.2)
H
H'
For the two-point fun tion in lowest order we nd
p k

(11.3)
p + p
p + :
p
p = p

k
If is a stable parti le (i.e. M < 2m), the loop in the rst diagram orresponds to virtual
' parti les moving between the verti es, sin e always k2 6= m2 and (k p)2 6= m2 . However,
as soon as M > 2m, the loop integral will ontain ontributions where the ' parti les an
be on the mass-shell and behave as a real parti les, e.g. k2 = m2 and (k p)2 = m2 in the
rst diagram. The real ' parti les an es ape to innity, thereby des ribing the de ay of the
parti le. Its number will redu e as a fun tion of time.
Indeed we will see that only if M > 2m, the self-energy will be able to develop a non-zero
imaginary part, Im( Z (p))jp2 =M 2 6= 0. On the mass-shell the propagator is in that
ase modied for t > 0 to
pp~ 2+M 2 i
i Z d3 p~ p Z ei~p~x
Z e ipx
it
=
:
(11.4)
e
M 2 + i + i"
2 (2 )3 p~ 2 + M 2 i
p
The poles p0 = i~p 2 + M 2 i are now omplex and for M 2 one has in a good
approximation
p
p
p
1
e it p~ 2 +M 2 i = e it p~ 2+M 2 e 2 t= p~ 2 +M 2 :
(11.5)
The amplitude
of the wave fun tion for the parti le onsequently de ays with a de ay rate
p
of (p) = = p~ 2 + M 2 , the life-time of this parti le is hen e (p) = 1= (p).
We will evaluate the imaginary part of self-energy for the 2-point fun tion in eq. (11.3)
rst to lowest order in g
!
ig 2 Z d4 k
1
Im( (p)) = Im
2 (2 )4 (k2 m2 + i")((k p)2 m2 + i")
!
ig 2 Z
2
ipx
;
(11.6)
d4 x G (x)e
= Im
2
Z
d4 p
(2 )4 p2
45
where we used the denition of the Green's fun tion, eq. (4.7). With the help of eq. (5.12)
we an write
Z
d3~k ei~k~xe ik0 (~k)jtj
G(x) = i
;
(11.7)
(2 )3 2k0 (~k)
whi h yields (after hanging x to x at the right pla e)
Z
d4 x (G(x)2 + G ( x)2 )eipx =

(11.8)
Z

d3~k1
d3~k2
(1) + k(2) p ) + (k(1) + k(2) + p ) ;
1 g 2 (2 )4
~
~
(
k
+
k
p
~
)
(
k
3
1
2
0
0
0
0
0
0
4
2k0(1) (2 )3 2k0(2) (2 )3
q
where we have impli itly dened k0(j ) ~kj2 + m2 . As pwe wish to study the 2-point
fun tion at t > 0 near the mass-shell, we an put p0 p~ 2 + m2 as well. In parti ular,
restri ting ourselves to p0 > 0, gives
Im( Z (p)) g 2 (2 )4 Z
d3~k2
d3~k1
(p)
4 (k1 + k2 p) ;
(11.9)
=
p0
4p0
2k0(1) (2 )3 2k0(2) (2 )3
Im( (p)) =
1
4
g2
whi h is the result to lowest non-trivial order in g (to this order we an take Z = 1).
To any order we an, however, de ompose the part of the self-energy, with two '
parti les as an intermediate state, in its 1P I omponents as follows
PP

PP

p =

(11.10)

where the full ' 2-point fun tion is given by 1=(p2 m2 ' (p)+ i"), whi h on the mass-shell
redu es to Z' =(p2 m
~ 2 + i"), the only part that a tually ontributes to Im( Z (pp)). The
1P I '' 3-point fun tion an easily be seen to be equal to M'' (f ki g; p)=(Z' Z ) (or
its omplex onjugate if in- and outgoing lines are inter hanged), sin e in lowest order it
should oin ide with the '' 3-point vertex. In this way we easily nd the partial de ay
rate d (p) to be
d (p) = (2 )4 4 (i ki
p)
jM'' (f kig; p)j2 S

2p0
d3~ki
;
2k0(i) (2 )3
(11.11)
whi h, as it should be, is always positive. The symmetry fa tor S is the same as for the ross
se tion in eq. (10.12). The total
de ay rate is found by integrating over the phase spa e of
R
the outgoing parti les (p) d (p). The large resemblan e with the formula for the ross
se tion is no oin iden e, as in both ases we have to al ulate the probability for something
to happen (respe tively a de ay or a s attering). In its present form the formula for (p) is
also valid for the de ay of a parti le in n other parti les. The derivation is almost identi al,
e.g. in eq. (11.8) one now en ounters Gn (x) instead of G2 (x) and g now stands of ourse
for the oupling onstant of n ' elds to the eld. It is not ne essary for this oupling to
o ur in the Lagrangian; at higher orders one an generate it from the lower ouplings that
do o ur in the Lagrangian.
46
12 The Dira equation

To obtain a Lorentz invariant S hrodinger equation, we onsidered the square root
of the
p
2
Klein-Gordon equation. This had the disadvantage that the Hamiltonian H = p~ + m2
ontains an innite number of powers of ~p 2 =m2 , the parameter in whi h the square root
should be expanded. It would have been better to treat spa e and time on a more equal
footing in the S hrodinger equation. This is what Dira took as his starting point. As the
S hrodinger equation is linear in p0 = i=t, one is looking for a Hamiltonian that is linear
in the momenta pj = i=xj .

i = H = ik k + m :
(12.1)
t
x
The question Dira posed himself was to nd the simplest hoi e for k and , su h that the
square of the S hrodinger equation gives the Klein-Gordon equation
p20 = (p~ ~ + m)2 = p~ 2 + m2 :
(12.2)
Dira noted that only in ase we allow k and to be non- ommuting obje ts (i.e. matri es),
one an satisfy these equations. The above equation is equivalent to
(12.3)
2 = 1 ; 1 (j k + k j ) = 1jk and j + k = 0 :
2
Histori ally, Dira rst onsidered m 6= 0, but the massless ase (m = 0) is somewhat
simpler, as it allows one to use = 0 and k = k for a solution of eq. (12.3). Here k are
the Pauli-matri es, familiar from des ribing spin one-half parti les.

0
1
0
i
1
0
1 = 1 0
; 2 = i 0
; 3 = 0 1
:
(12.4)
It is lear that two will be the smallest matrix dimension for whi h one an solve the equation
1 ( + ) 1 f ; g = 1 . It is not hard to prove that in a two dimensional
j k
k j
j k
jk
2
2
representation all solutions to this equation are given by
j = Uj U 1 ;
(12.5)
where U is an arbitrary non-singular omplex 2 2 matrix. We should, however, require
that H (and hen e j ) is hermitian. This narrows U down to a unitary matrix, sin e
j y Uj U 1 U y 1 y U y = U ([U y Uj 1 j U y U )U 1 = 1 ;
(12.6)
j
su h that U y Uj = j U y U for ea h j . The only 2 2 matrix that ommutes with all Paulimatri es is a multiple of the identity, whi h proves that U is unitary (up to an irrelevant
overall omplex fa tor, whi h does not ae t j ).
Sin e the Hamiltonian is now a 2 2 matrix, the wave fun tion (x) be omes a omplex
two dimensional ve tor, also alled a spinor, whi h des ribes parti les with spin h =2
p0 (x) = p~ ~ (x) :
(12.7)
We have to demonstrate that the Dira equation is ovariant under Lorentz transformations.
We rst put the boosts to zero, be ause we already know from quantum me hani s how a
spinor transforms under rotations
i
(12.8)
~p ! ~p 0 = exp(~! L~ )p~ ; (x) ! 0 (x0 ) = exp( ~! ~ ) (x) :
2
47
Here Li are real 3 3 matri es that generate the rotations in IR3
Lijk = "ijk ;
(12.9)
su h that [Li ; Lj Li Lj Lj Li = "ijk Lk . These re e t the ommutation relations of the

generators ij =2. We will later, in the ontext of non-Abelian gauge theories, show that this
des ribes the fa t that SU(2) (the group of unitary transformations a ting on the spinors)
is a representation of SO(3) (the group of rotations in IR3 ). To show the ovarian e of the
Dira equation under rotations, i.e.
p0 (x) = p~ ~ (x) ! p0 0 (x0 ) = p~ 0 ~ 0 (x0 ) ;
(12.10)
we work out the Dira equation in the rotated frame. Using eq. (12.8) we get
i
i
~! ~ )~ exp( ~! ~ ) (x) ;
(12.11)
2
2
whi h should redu e to eq. (12.7). To prove this, we use the following general result for
matri es X and Y
p0 (x) = p~ 0 exp(
eX Y e
= exp(adX )(Y ) ; adX (Y ) [X; Y ;
(12.12)
whi h is derived from the fa t that f1 (t) = etX Y e tX and f2 (t) = exp(ad(tX ))Y satisfy the
same dierential equation, dfi (t)=dt = [X; fi (t). Sin e also f1 (0) = f2 (0) it follows that
f1 (1) = f2 (1), being the above equation. Applying this result to Y = k and X = i~! ~=2,
using the fa t that
i
i
~ )kj j ;
ad( ~! ~ )k = [ ~! ~ ; k = (~! L
(12.13)
2
2
~ )~ .
the r.h.s. of eq.(12.11) be omes p~ 0 ~ 0 (x) = p~ ~ (x), where ~ 0 = exp(~! L
The interpretation of this S hrodinger equation aused Dira quite some trouble, as its
eigenvalues are jp~ j, and it is not bounded from below. In the s alar theory we ould avoid
this by just onsidering the positive root of the Klein-Gordon equation. Only when we
required lo alisation of the wave fun tion inside the light- one, we were for ed to onsider
negative energy states. In the present ase, restri ting to one of the eigenstates would break
the rotational invarian e of the theory. For the massive ase Dira rst in orre tly thought
that the positive energy states des ribe the ele tron and the negative energy states the
proton. At that time antiparti les were unknown. Antiparti les were predi ted by Dira
be ause the only way he ould make the theory onsistent was to invoke the Pauli prin iple
and to ll all the negative energy states. A hole in this sea of negative energy states, the
so- alled Dira sea, then orresponds to a state of positive energy. These holes des ribe
the antiparti le with the same mass as the parti le. Obviously parti le number will no
longer be onserved and also the Dira equation will require \se ond quantisation" and the
introdu tion of a eld, whi h will be dis ussed later.
For the massive Dira equation we need to nd a matrix that anti ommutes with all
i . For 2 2 matri es this is impossible, sin e the Pauli matri es form a omplete set of
anti ommuting matri es. The smallest size turns out to be a 4 4 matrix. The following
representation is usually hosen
i =
i
i
; = 12

48
12
(12.14)
for whi h the non-relativisti limit has a simple form. For the massless ase it is often more
onvenient to use the so- alled Weyl representation

~i = i
; ~ =
i
12
12 :

(12.15)
We leave it as an exer ise for the student to show that these two representations are related
by a 4 4 unitary transformation U , i.e. ~ = U U 1 .
To study the ovarian e of the Dira equation
p0 (x) = (i pi + m) (x)
(12.16)
under Lorentz transformations (note that now (x) has four omplex omponents), it will
be protable to introdu e a \four-ve tor" of 4 4 matri es
( 0 ; i ) = (; i) ;
(12.17)
su h that the Dira equation be omes
( i + m) (x) = ( p + m) (x) (=p + m) (x) = 0 :
(12.18)
The Dira gamma matri es satisfy anti ommuting relations
f ; g + = 2g :
(12.19)
As for the ovarian e under rotations, this equation is ovariant under Lorentz boosts if there
exists a non-singular omplex 4 4 matrix S , su h that (x) ! 0 (x0 ) = S (x) and
S 1 S = K ;
(12.20)
where K is the Lorentz transformation a ting on the momenta as p0 = K p and on the
oordinates as x 0 = K x . Like for the rotations, K an be written as an exponent
K = exp(! ) :
(12.21)
Here ! is a 4 4 matrix, whi h is antisymmetri when one of its indi es is raised or lowered
by the metri , ! = g ! = ! .
We will now prove that
i
i
! ) ; [ ;
4
2
satises eq. (12.20). Using the antisymmetry of ! and eq. (12.19) we nd
S = S (! ) = exp(
i
[ ! ; =
4
=
=
=
1
4
(12.22)
! [ ;
! ( )

2 g )
1
4 ! ( +
1 ! (g
g ) = ! :
2
1
4
(12.23)
Applying eq. (12.12) gives the proof for eq. (12.20). One says that S (! ) is a representation
of K (! ). Note that in general S (! ) is not a unitary transformation. This is be ause the
49
boosts form a non- ompa t part of the Lorentz group. There is, however, a relation between
S y and S 1 ,
0S y 0 = S 1 ;
(12.24)
whi h is most easily proven in the Weyl representation, sin e ~ 0 = ~ ommutes with ~ij
but anti- ommutes with ~k0 (as always roman indi es run from 1 to 3 and greek indi es run
form 0 to 3), whereas ~ij is hermitian and ~k0 is anti-hermitian, as follows from the expli it
expressions obtained from eqs. (12.15), (12.17) and (12.22)
~ij =
~ ij

i i j
i k 0

k
k
0
= [~ ; ~ = "ijk
; ~0k = ~ = [~ ; ~ = i k
2
2
k
:
(12.25)
Note that eqs. (12.20) and (12.22) are independent of the representation in whi h we give
the gamma matri es, as any two su h representations have to be related by a unitary transformation. In the Weyl representation S~(! ) is blo k diagonal, like the Dira equation for
m = 0 (as ~i is blo k diagonal). The upper blo k orresponds to eq. (12.7) with the plus
sign and the lower blo k orresponds to the minus sign. We an verify eq. (12.8) by using
the fa t that the Lorentz transformations ontain the rotations through the identi ation
!k = 12 "ijk !ij . With !0k = 0, one nds S~(! ) = 12
exp(i~! ~ =2), i.e. it a ts on ea h 2 2
blo k by the same unitary transformation.
The boost parameters are des ribed by !0i . For a boost in the x dire tion we have that
= !01 is related to the boost velo ity by v1 = tanh(). For K we nd in this ase
0
osh() sinh()
B
sinh() osh()
K=B
B

0
0
0
0
0
0
1
0
k
0
0C
C
:
A
0C
1
(12.26)
In the Weyl representation, S~ splits again in two blo ks, but one is the inverse of the other
(and neither is unitary). To be pre ise, S~ restri ted to the upper-left 2 2 blo k equals
exp( 21 !0k k ), whereas for the lower-right 2 2 blo k we nd exp( 12 !0k k ).
As S is not unitary y (x) (x) is no longer invariant under Lorentz transformations.
But we laim it is nevertheless a probability density, namely the time omponent j 0 (x) of a
onserved urrent
j (x) = y (x) 0 (x) :
(12.27)
We leave it as an exer ise to show that the Dira equation implies that the urrent is
onserved, j (x) = 0. The ombination y(x) 0 will o ur so often, that it has a quired
its own symbol
(x) y(x) 0 :
(12.28)
It transforms under a Lorentz transformation as
(x) ! S (x) and (x) ! (x)S 1 :
(12.29)
We an use this to build the required Lorentz s alars, ve tors and tensors
0 (x0 ) 0 (x0 ) = (x)S 1 S (x) = (x) (x)
s alar:
ve tor: 0 (x0 ) 0 (x0 ) = (x)S 1 S (x) = K (x) (x)
tensor: 0 (x0 ) 0 (x0 ) = (x)S 1 S (x) = K K (x) (x)
50
The Lagrangian is a Lorentz s alar, whi h we hose su h that its equations of motion reprodu e the Dira equation. As (x) is omplex, it an be onsidered independent of (x) and
the following Lagrangian
SDira =
d4 x LDira =
d4 x (x)(i
m) (x) :
(12.30)
gives the Euler-Lagrange equations
S
= (i
(x)
S
= (x)( i m) = 0 :
(x)
m) (x) = 0 ;
(12.31)
The se ond equation is the omplex onjugate of the rst, y(x)(i( )y +m) = 0, be ause
the gamma matri es satisfy
( )y = 0 0 ;
(12.32)
whi h follows from the fa t that ( 0 )y = y = = 0 and ( i )y = (i )y = i = i , or
from the expli it representation of the gamma matri es

0 = 12
12
; i =
i
i
(12.33)
Hen e
0 = y(x)(i( )y +m) 0 = (i 0 ( )y 0 +m) = (i +m) :
(12.34)
An important role will be played by a fth gamma matrix
5 i 0 1 2 3 =
12
12 ;

(12.35)
whi h anti ommutes with all (see problem 21)
5 = 5 and ( 5 )2 = 14 :
(12.36)
This implies that we an introdu e proje tion operators
P 21 (1 5 ) ;
(12.37)
whi h satisfy (P )2 = P and P P = 0. Their role is best des ribed in the Weyl representation, where

1

2
~5 = i ~0 ~1 ~2 ~3 = i~1 ~2 ~3 = 1
;
(12.38)
2
su h that P~+ = 21 (1 + ~5 ) proje ts on the two upper and P~ = 21 (1 ~5 ) on the two lower
omponents of the four-spinors. In the massless ase these two omponents are de oupled,

p0 (x) = (p~ ~~ + m~) (x) = p~m1~
2
m12 :
p~ ~
(12.39)
p0 (x) = p~ ~ (x) :
(12.40)
Hen e, for m = 0 we have

(
x
)
+
(x) = (x)
;

51
whi h is identi al to eq. (12.7). The eigenstates of the proje tion operators P are alled
heli ity eigenstates. As long as m 6= 0, heli ity is not onserved. But as we saw, for m = 0
the two heli ity eigenstates de ouple. One an dene in that ase onsistently a parti le
with a xed heli ity, whose opposite heli ity state does not o ur (although its antiparti le
has opposite heli ity). A very important example of su h a parti le is the neutrino, although
experiment has not yet been able to rule out a (tiny) mass for this parti le (me < 10eV ).
See problem 22 for more details.
Apart from the invarian e of the Dira equation under Lorentz transformations and
translations (whi h are obvious symmetries of LDira ) we also often want invarian e under
parity (~x ! ~x) and time-reversal (t ! t). One easily he ks that
P (x) 0 (t; ~x) = 0 (t; ~x) and T (x) 0 ( t; ~x) = 5 0 2 (t; ~x)
(12.41)
satisfy the Dira equation, where P stands for parity and T for time-reversal. This implies
that the Lorentz ovariant ombinations (x) 5 (x) and (x) 5 (x) are not invariant
under parity and time-reversal. They are alled pseudos alars and pseudove tors. These
ombinations play an important role in the weak intera tions, where parity is not a symmetry.
A third dis rete symmetry, harge onjugation C , will be dis ussed in se tion 17.
13 Plane wave solutions of the Dira equation

As usual, the Dira equation an be solved by Fourier de omposition in plane waves,
(x) =
d3~k
~ ~k)e
(
3
2k0 (~k)(2 )
Z
q
ikx
(13.1)
~ ~k) are omplex four-ve tors that satisfy

where (
~ ~k) (k
(k= m) (
~ ~k) = 0 ; k02 = ~k 2 + m2 :
m) (
(13.2)
The Lorentz invarian e implies

~ 0 (~k 0 ) = S (! ) (
~ ~k) ; k0 = K (! ) k :

(13.3)
As any ~k an be obtained from a boost to ~k = ~0, all solutions of the Dira equation an be
obtained from the ones at rest with ~k = ~0 (see problem 19)
( 0 k0

~
~
m) (0) = (k0
m)12
~~
(k0 + m)12 (0) :
(13.4)
We see that for k0 > 0 (k0 = m), there are two independent solutions both of the form

A
~
~
+ (0) = 0
;
(13.5)

where A is a spin one-half two-spinor of whi h 1 is the spin-up and 0 is the spin0
1
down state. The identi ation of the spin degrees of freedom follows from the behaviour of
52
~ under rotations (whi h leave ~k = ~0). We leave it as an exer ise to verify that also in the

Dira representation of the gamma matri es, like in the Weyl representation (see eq. (12.25)),

ij = "ijk k
;
k
(13.6)
su h that A is easily seen to transform under a rotation as in eq. (12.8) ( mp. the dis ussion
below eq. (12.25)). There are also two solutions for k0 < 0 (k0 = m) of the form

0
~
~
(0) =
;
B
(13.7)

1
where likewise B is a spin one-half two-spinor of whi h
is the spin-up and 0 is the
0
1
spin-down state, whi h transform under rotations as A.
For any frame, i.e. for any value of ~k, we will dene the four independent solutions of
the Dira equation as

0
1
q
B C
B C
k
=
+
m
1C
(2) = B
B0C
k0 = ~k 2 + m2 : u() (~k) = q
B C ;
C ; u0
u(0) ; u(1)
0 =B
0A
0A
m + jk0 j
0
0
0 1
0 1
0
0
q
B C
B C
k
=
+
m
0C
0
(2) = B
C
B C : (13.8)
k0 = ~k 2 + m2 : v () ( ~k ) = q
v0() ; v0(1) = B
B C ; v0
0A
1A
m + jk0 j
1
0
These solutions naturally split in positive energy (u() (~k) with = 1; 2) and negative energy
(v () ( ~k ) with = 1; 2) solutions; for ~k = ~0 and mp6= 0 easily seen to be proportional
~ (~0), to be pre ise u() (~0) = u(0) = 2m and v () (~0) = v0() =p2m. The
to the solutions
normalisation we have hosen allows us to treat massless fermions at the same footing. In
that ase we an, however, not transform to the restframe. This normalisation also implies
that
u() (~k)u() (~k) = v () ( ~k)v () ( ~k ) = 2m :
(13.9)
For example
u(0) 0 (k= + m)y 0 (k= + m)u(0) u(0) (k= + m)2 u(0)

(

)
(

)
~
~
=
=
u (k)u (k) =
(m + k0 )
(m + k0 )
u(0) (k2 + m2 + 2k=m)u(0) u(0) (k2 + m2 + 2k0 m)u(0)
=
= 2m ;
(m + k0 )
(m + k0 )
(13.10)
where we used k2 = m2 and u(0) u(0) = 0 (see eq. (12.33)). The omputation for
v () (~k) is left as an exer ise. The fa t that we nd a result that is independent of ~k is
onsistent with our laim that these spinors an also be obtained by applying the appropriate
~ ~k) (
~ ~k) is a Lorentz s alar.
boost to ~k = ~0, sin e (
That these spinors indeed satisfy eq. (13.2) follows from the fa t that
(k= m)(k= + m) = k2 m2 = 0 :
(13.11)
53
Note that v () (~k) an also be viewed as a positive energy solution for the omplex onjugate
of the Dira equation
q
k0 = ~k 2 + m2 : (k= + m)v () (~k) = 0 :
(13.12)
It will play the role of the wave fun tion for the antiparti les (the holes) in the Dira eld
theory. To see that our plane waves have the orre t amplitude, we use the fa t that the
probability density an be dened in terms of the zero- omponent (x) of the onserved
four-ve tor j (x) = (x) (x) (see eq. (12.27)), or
(x) = y (x) (x) :
(13.13)
~ y (~k) (
~ ~k) transforms as a density, i.e. as the energy k0 , and we nd
Indeed,
u() (~k)yu() (~k) = v () (~k)yv () (~k) = 2k0 ; u() (~k)yv () (~k) = 0 :
(13.14)
This an be veried by a dire t omputation, e.g. (k0 > 0)
() y(m k=)y(m k=)v () v () y(k2 + ~k 2 + m2 2mk 0 )v ()
v
0
0 =
0
0 = 0
0
(

)
y
(

)
v (~k) v (~k) =
m + k0
m + k0
y
v0() (k02 + ~k 2 + m2 + 2mk0 )v0()
= 2k0 ;
(13.15)
=
m + k0
y
using v0() v0() = 0 (see eq. (12.33)). We leave the other identities as an exer ise.
1
This implies that the plane wave solutions (k0 = (~k 2 + m2 ) 2 )
v () (~k) ikx
e
; = 1; 2 ;
(13.16)
2k0 (2 )3
R
P
are normalised to one (of ourse, in a nite volume we repla e (2 )3 by V and d3~k by ~k ).
As for the s alar eld we an introdu e a Dira eld
Z
2

X
d3~k
(x) = q
(13.17)
b (~k)u() (~k)e ikx + dy (~k)v () (~k)eikx :
2k0 (~k)(2 )3 =1
Sin e (x) is omplex, there is no relation between d and b . In the quantum eld theory
these will play the role of the annihilation operators for the (anti-)parti les.
Finally we note that, for m 6= 0, we an dene proje tion operators (whi h are 4 4
matri es)
2 v () (~k)
v () (~k)
2 u() (~k)
u() (~k) k= + m
X
X
k= + m
=
; (~k) =
=
;
+ (~k) =
2m
2m
2m
2m
=1
=1
(13.18)
whi h an be veried from the expli it form of u and v , or by rst omputing them in the
restframe where (~0) = 12 (1 0 ). In that ase we nd
(k= + m)(1 0 )(k= + m) k= + m
=
;
(13.19)
(~k) =
4m(m + k0 )
2m
whose proof requires some gamma matrix gymnasti s. Independent of the frame satisfy
2 (~k) = (~k) ; (~k) (~k) = 0 :
(13.20)
Note that Tr((~k)) = 2, a result that an also be derived from eq. (13.9).
u() (~k)
e
2k0 (2 )3
ikx
54
14 The Dira Hamiltonian

In the Dira equation we en ountered an additional di ulty, namely that the negative energy solutions even arise at the level of the lassi al theory. In eld theory the negative
energy solutions had an interpretation in terms of antiparti les, and the eld theory Hamiltonian was still positive, and most importantly, bounded from below (see se tion 2 and
problem 5). The eld theory Hamiltonian for the Dira eld no longer has this property.
The Hamiltonian an again be derived through a Legendre transform of the Lagrangian
S=
d4 x L =
d4 x (x)(i
m) (x) :
(14.1)
The anoni al momentum is hen e
S
= ( (x)i 0 )a = i a (x) :
a (x) = _
a (x)
(14.2)
su h that
H =
=
=
d3~x (a (x) _ a (x)
L) = d3~x (x)( i j j + m) (x)
d3~x y (x)( ij j + m ) (x)

d3~k k0 (~k)
X
by (~k)b (~k) d (~k)dy (~k)
(14.3)
Note the resemblan e with eq. (12.1) for the middle term. We used eq. (13.17) for the
expansion of the Dira eld in plane waves. From this result it is lear that the Hamiltonian
is not bounded from below, and this would make the va uum unstable, as the negative
energy states, des ribed by the d (~k), an lower the energy by an arbitrary amount. It is
well-known how Dira repaired this problem. He postulated that all negative energy states
are o upied, and that the states satisfy the Pauli prin iple, i.e. two parti les an not o upy
the same quantum state (it is only in that ase that we an make sense of what is meant with
lling all negative energy states). This implies that one should use anti ommuting relations
for the reation and annihilation operators
fb (~k); b (p~)g = 0 ; fb (~k); by (p~)g = 3 (~k ~p) ;

fd(~k); d (p~)g = 0 ; fd(~k); dy (p~)g = 3 (~k p~) ;
fd(~k); b (p~)g = 0 ; fd(~k); by (p~)g = 0 :
(14.4)
Indeed, if we dene a two parti le state as j~k; p~ > by (~k)by (p~)j0 > (suppressing the spinor
indi es), the anti ommutation relations imply that j~k; ~p >= jp~; ~k >.
A hole in the Dira sea is by denition the state that is obtained by annihilating a
negative energy state in the Dira sea. As annihilation lowers the total energy by the energy
of the annihilated state, whi h in this ase is negative, the net p
energy is raised. The wave
fun tion for the negative energy state is given by exp(ikx)v () (~k)= k0 and has momentum ~k
1
(k0 (~k 2 + m2 ) 2 ), see eq. (13.8). The reason to asso iate its Fourier oe ient in eq. (13.17)
with a reation operator dy (~k), is that onservation of energy and momentum implies that it
reates an antiparti le as a hole in the Dira sea, with momentum ~k and heli ity 12 for = 2,
whereas for = 1 the heli ity is 21 (hen e the heli ity and momentum are opposite to the
55
negative energy state it annihilates). The wave fun tion of an antiparti le with momentum
~k is hen e given by exp( ikx)v () (~k)y=pk0 . If we now use the anti ommutation relations of
reation and annihilation operators (as a onsequen e of the Pauli ex lusion prin iple), we
see that with the present interpretation the Dira Hamiltonian is bounded from below
H=
d3~k k0 (~k)
X
by (~k)b (~k) + dy (~k)d (~k) 1
(14.5)
As in the ase for s alar eld theory we an normalise the energy of the va uum state to zero
by adding a (innite) onstant. Note that this onstant has its sign opposite to the s alar
ase (and is in magnitude four times as large). In so- alled supersymmetri eld theories this
is no longer an a ident as the Dira elds will be related to the s alar elds by a symmetry,
whi h is however outside the s ope of these le tures.
Important is also to note that the anti ommuting relations are ru ial to guarantee
lo ality of the Dira eld. In this ase Dira elds spe ied in dierent regions of spa e-time
that are spa e-like separated should anti ommute. And indeed, in problem 24 you are asked
to prove that
f a (x); yb (x0 )g = 0 for (x x0 )2 < 0 :
(14.6)
Also, as in the s alar theory, we an ouple a sour e to the free Dira eld (we again
introdu e " as the expansion parameter for taking the intera tions due to the sour e into
a ount in Hamiltonian perturbation theory)
L(x)
H(x)
= (x)(i m) (x) "J(x) (x) " (x)J (x) ;

= (x)( i k k + m) (x) + "J(x) (x) + " (x)J (x) ;
(14.7)
whereR as before the Hamiltonian H is the spatial integral over the Hamiltonian density H(x),
H = d3~x H(x). Note that the sour es J (x) and J(x) are independent, as for a omplex
s alar eld, see problem 17. In problem 24 you are asked to prove that in Hamiltonian
perturbation theory one obtains
< 0jTexp( i
H (t)dt) j0 >
eiE0 T
i"2
= 1
= 1 i
where
"J
d4 xd4 y J(x)GF (x y )J (y ) + O("3 )

>
+ O("3) :
"J
(14.8)
Z
d4 p (=p + m)e ip(x y)
d4 p e ip(x y)
=
:
(14.9)
(2 )4 =p m + i"
(2 )4 p2 m2 + i"
Hen e, the Green's fun tion for fermions is in Fourier spa e given by GF (p) = (=p m + i") 1 ,
whi h is the inverse of the quadrati part of the Lagrangian, as for the s alar elds. In
problem 24 it will be evident that, nevertheless, the anti ommuting properties of the Dira
eld play a ru ial role ( mp. se tion 5). It be omes, however, plausible there is also for the
fermions a path integral formulation, as splitting of a square ( mp. eqs. (7.11) and (7.12)) is
independent of the details of the path integrals. This will be the subje t of the next se tion.
To on lude, we note that the oupling to an ele tromagneti eld A (x) should be
a hieved through the urrent j (x), dened in eq. (12.27)
G F (x y ) =
d4 x
j (x)A (x)
d4 x (x) A (x) (x) :

56
(14.10)
Sin e this urrent is onserved, whi h an also be seen as a onsequen e of the Noether
theorem applied to the invarian e of eq. (14.1) (the free Dira Lagrangian) under global
phase transformations, the oupling is gauge invariant. Using the minimal oupling dened
as in eq. (3.35),
D (x) ( ieA (x)) (x) ;
(14.11)
one an x the normalisation of the ele tromagneti urrent to be J (x) = ej (x), sin e
( mp. eq. (3.31))
(x)(i D
m) (x) = (x)(i
m) (x) + ej A (x) :
(14.12)
The urrent j (x) is the harge density urrent, whose time omponent (x) at the quantum
level is no longer positive denite. Using the anti ommuting properties of eq. (14.4) one
nds
Z
d3~x ^(~x) =
=
=
d3~x y(x) (x)

X
by (~k)b (~k) + d (~k)dy (~k)
d3~k
d3~k
X
= Q0 =e +
by (~k)b (~k) dy (~k)d (~k) + 1
d3~k
X
by (~k)b (~k) dy (~k)d (~k)
(14.13)
The va uum value of this operator is indi ated by the (generally innite) onstant Q0 =e,
whi h an be normalised to zero. After all we want the state with all negative energy
states o upied to have zero harge. With the above normalisation of the ele tri urrent,
J (x) = ej (x), we see that the b modes an be identied with the ele trons with harge
e and the d modes with their antiparti les, the positrons, with opposite ele tri harge
+e. To summarise, by (~k) orresponds with the reation operator of a spin-up ( = 1) or a
spin-down ( = 2) ele tron of momentum ~k, whereas dy (~k) orresponds with the reation
operator of a spin-up ( = 2) or a spin-down ( = 1) positron of momentum ~k.
15 Path integrals for fermions

For s alar elds, whi h des ribe bosons, we used real or omplex numbers (the eigenvalues
of the operators), in order to perform the path integral. For fermioni elds it is essential to
build the anti ommuting properties into the path integral.
To this end we introdu e a so- alled Grassmann algebra, whi h exists of Grassmann
variables i that mutually anti ommute
fi ; j g i j + j i = 0 :
In parti ular, a Grassmann variable squares to zero
2 = 0
(15.1)
(15.2)
A Grassmann variable an be multiplied by a omplex number, with whi h it ommutes. A

fun tion of a single Grassmann variable has a nite Taylor series
f () = a0 + a1 ;
57
(15.3)
and spans a two dimensional (real or) omplex ve tor spa e. This is exa tly what we need
to des ribe a spin one-half parti le. Let us introdu e the following notation

1
j0 >= 0 ;

0
j1 >= 1 :
(15.4)
With respe t to the Hamiltonian
H0 =
W 0
0 W
1
2
(15.5)
j0 > is the va uum state (i.e.
the state with lowest energy) and we interpret j1 > as the

one-parti le state (with energy W above the va uum). An arbitrary spinor an be written
as a linear ombination of these two states
j >= a0 j0 > +a1 j1 >= a0 j0 > +a1byj0 > ;
(15.6)
where b is the fermioni annihilation operator, whi h in the spinor representation is given
by a 2 2 matrix

b = 00 10
; by = 01 00
:
(15.7)
We note that b2 = (by )2 = 0, a property it has in ommon with a Grassmann variable. We
will now look for properties of su h that
() < j >= a0 + a1
(15.8)
is a representation of the state j >, similar to (x) =< xj > for the ase of a single
bosoni (i.e. ommuting, as opposed to anti ommuting) degree of freedom. In the latter
ase, the normalisation
Z
dx (x) (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 ompleteness relations). As < 0j = (1; 0) and < 1j = (0; 1), we
have
< j =< 0ja0 + < 1ja1 =< 1jbya0 + < 1ja1 :
(15.10)
As in eq. (15.8) (i.e. by ! ), we anti ipate
() < j >= a0 + a1 :
(15.11)
We wish to dene integration over Grassmann variables, su h that the normalisation of the

wave fun tion is as usual
Z
< j >= ja0 j2 + ja1 j2 = d () () :
(15.12)
R
Sin e the norm should be a number, and

as d 1 is itself a Grassmann variable, the latter
R
should vanish. For the same reason d (whi h is itself a ommuting obje t, as is any
even produ t of Grassmann variables) an be seen as a number. Demanding the so- alled
Grassmann integration to be linear in the integrand, and using 2 = 0, all possible ingredients
have been dis ussed. Indeed
Z
d = 1 ;
58
d 1 = 0
(15.13)
is easily seen to give the desired result. Note that d is onsidered as an independent
Grassmann variable (whi h is important to realise when multiple Grassmann integrations
are involved).
We an now study the a tion of an operator (like a Hamiltonian H ) on a state, whi h in
the spinor representation is given by 2 2 matri es.

0
M
M
a
a0
11
12
0
or
=
:
M21 M22 a1
a01
Translated to Grassmann variables, this gives

M j >= j 0 >
0 () = a00 + a01 = M11 a0 + M12 a1 + (M21 a0 + M22 a1 )
(15.14)
d0 M (; 0 ) (0 ) ; (15.15)
provided we dene
M (; 0 ) M11 0 + M12 + M21 0
M22 = M11 0 + M12
M21 0
M22 :
(15.16)
Indeed,
Z
d0 M (; 0 ) (0 ) =

=
d0 fa0 M12
d0 (M11 0 + M12
M21 0
M22 )(a0 + a1 0 )
a0 M22 + (a1 M12 + a0 M11 )0
(a0 M21 + a1 M22 )0 g
= d0 0 f(a1 M12 + a0 M11 ) + (a0 M21 + a1 M22 )g

= M12 a1 + M11 a0 + (M21 a0 + M22 a1 ) :
(15.17)
The 2 2 identity matrix is hen e represented by (note the sign)

12 (; 0 ) = 0
;
whi h an be used to write the innitesimal evolution operator

exp( iH t) = 12 iH t + O(t2 ) ;
(15.18)
(15.19)
where H is a (possibly time-dependent) 2 2 matrix. In the Grassmann representation this

reads
12 (; 0 )
iH (; 0 )t = Z0 it(H11 0 + H12 H21 0 H22 )

=
d~ exp(~[12 (; 0 ) iH (; 0 )t) :
(15.20)
The last identity is exa t, and a onsequen e of the fa t that the Taylor series of any fun tion
of a Grassmann variable trun ates
~ ) = 1 + x
~ ;
exp(x
~ )=x :
d~ exp(x
(15.21)
This is valid both for x a Grassmann variable (in whi h ase the ordering of x with respe t
to ~ is important, with the opposite ordering the result is x), or for x a omplex number.
Another useful property of Grassmann integration is that (y is a number)
Z
d exp(x + y ) = x exp(y ) :
59
(15.22)
To prove this, we use that

exp(x + y ) =
1 1
n=0 n!
(x + y )n =
1 1
1
X
y n + x
n=0 n!
n=1 (n
1)!
y n 1 = (1+ x) exp(y ) : (15.23)
In general it is not true that the exponential fun tion 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 matri es, as it should sin e the Grassmann representation originates from a 2 2
matrix representation. More pre isely
exp(x) exp(0 y ) = (1 + x)(1 + 0 y ) = 1 + (x + 0 y ) + 1 (x + 0 y )2 + 1 [x; 0 y
= exp (x + 0 y + 21 [x; 0 y ) :
(15.24)
This means that the Campbell-Baker-Hausdor formula (eq. (6.44)) an be extended to this
ase. It trun ates after the single ommutator term as neither x nor y an appear more
than on e.
Let us apply this to the evolution operator, whi h in the Grassmann representation is
given by (see eqs. (15.19) and (15.20))
Z
< i+1 jU (ti+1 ; ti )ji > Ui (i+1 ; i ) = d~i exp ~i [i
su h that
Z
i+1
iH (ti ; i+1 ; i )t + O(t2 ) ;

(15.25)

di < i+1 jU (ti+1 ; ti )ji >< i j >=

Z
di d~i exp ~i [i
i+1

iH (ti ; i+1 ; i )t (i ) + O(t2 ) : (15.26)
whi h an be iterated, rst by one step, to

Z
di < i+2 jU (ti+2 ; ti )ji >< i j >

=
di+1 Ui+1 (i+2 ; i+1 ) di Ui (i+1 ; i ) (i )
di+1 d~i+1 exp ~i+1 [i+1 i+2 iH (ti+1 ; i+2 ; i+1 )t
Z

di d~i exp ~i [i i+1 iH (ti ; i+1 ; i )t (i ) + O(t2 ) :
(15.27)
Note that we have to be areful where we put the dierentials, as they are Grassmann
variables themselves.
If H is diagonal, as will often be the ase for the appli ation we have

in mind, exp ~[12 (; 0 ) iH (; 0 )t will be a ommuting obje t (so- alled Grassmann
even) and it does not matter if we put one of the dierentials on one or the other side of
the exponential. The ombination di d~i is likewise Grassmann even, and the pair an be
shifted to any pla e in the expression for the path integral. Hen e, provided of ourse H is
diagonal, any hange in the ordering an at most given an additional minus sign. We now
apply eq. (15.24) to the above produ t of exponentials,

exp(~i+1 [i+1 i+2 iH (ti+1 ; i+2 ; i+1 )t) exp ~i [i i+1 iH (ti ; i+1 ; i )t (15.28)
0
1
i+1
X
=exp ~j fj j +1 iH (tj ; j +1 ; j )tg 21 t2 [~i+1 H (ti+1 ; i+2 ; i+1 ); ~i H (ti ; i+1 ; i )A
j =i
60
and evaluate the ommutator that appears in the exponent. With the expli it expression for
H (see eq. (15.20)) we nd
Z
di+1 [~i+1 H (ti+1 ; i+2 ; i+1 ); ~i H (ti; i+1 ; i ) =

2~i+1 ~i i+2 H21 (ti+1 )H12 (ti ) i H12 (ti+1 )H21 (ti ) :
(15.29)
For H diagonal the ommutator term vanishes as was to be expe ted. In that ase, the
Campbell-Baker-Hausdor formula trun ates to the trivial term both in the matrix and in
the Grassmann representations. This does not mean that there are no dis retisation errors
in the fermioni path integral when H is diagonal, as an be seen from eq. (15.19). To be
pre ise, assuming for simpli ity that H = 12 W 3 , one has
n
o
2 sin( 21 W t)
exp( iH t) = 12 i H 2i W tan( 14 W t)12
;
(15.30)
W
whi h shows that t is ee tively modied to 2 sin( 12 W t)=W (it is interesting to ontrast
this with the result we found for the harmoni os illator in se tion 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 ! 1 an be taken:
Z
d < 0 jU (T; 0)j >< j >= Nlim

!1
0
= Nlim
!1
NY1 Z
j =0
dj d~j exp
NX1
j =0
d0 UN (N = 0 ; 0 = ) (0 )
~j [j
j +1
iH (tj ; j +1 ; j )tA (0 ) ; (15.31)
where as usual one has t = T=N . In omplete analogy with eq. (6.10), reinstating the
dependen e on Plan k's onstant, we an write
< 0 jTexp( i
H (t)dt=h )j >

2
!3
Z
NY1 Z
NX1 i~ (
i

t

)
j
j
+1
j
= Nlim
d~0
dj d~j exp 4
~j H (tj ; j +1 ; j ) 5
!1
h

t
j =1
j =0
2
!3
Z
Z
NY1
NX1 i~ (
(

+

)

)
i

t
j
+1
j
j
j
+1
j
5
= Nlim
d~0
dj d~j exp 4
+ ~j W (tj )
!1
h

t
2
j =1
j =0
Z
D (t)D (t)
"
exp i
dt y(t)(it + W (t)) (t)
(15.32)
Here we have identied ~ with y (whi h in this ase agrees with , sin e in one time and
no spa e dimensions 0 1) and with . We have indi ated the general ase where W an
depend on time, but in absen e of this time dependen e, W is the energy of the one-parti le
state, reated from the va uum. Sin e we identied also with by , the reation operator for
the one-parti le state, we see that W y = W bby = H + 21 W , what is to be expe ted from
a relation between the Lagrangian and Hamiltonian (the term 12 W is of ourse irrelevant) .
As for s alar eld theories, we will be mainly interested in va uum expe tation values
of the evolution operator. In the presen e of a sour e term this will allow us to derive all
required matrix elements. For the present ase we easily nd the va uum wave fun tion to
be
< j0 >= 1 ; < 0j >= ;
(15.33)
61
su h that
< 0jU (T )j0 >=
d0 < 0j0 > d < 0 jU (T )j >< j0 >=
dd0 0 < 0 jU (T )j > :

(15.34)
The order of the dierentials is important here. However, as in the ase of s alar eld
theories, we do not require the pre ise form of the va uum wave fun tion(al) for performing
perturbation theory.
Up to now we have des ribed a spin one-half parti le pinned-down at a xed position.
It is obvious how this an be generalised to in lude the quantum me hani al des ription of
a moving spin one-half parti le in a one-dimensional potential V (x). If also W depends on
the parti le position (W (x)), the Hamiltonian be omes
!
p^ 2
(15.35)
H=
+ V (^x) 12 21 W (^x)3 :
2m
If we write < x; j > (x; ) a0 (x) + a1 (x), the path integral is easily found to be
Z
dp0 d~0 NY1 Z dpj dxj
< x0 ; 0 j exp( iHT=h )jx; >= Nlim
d d~
(15.36)
!1
2h j =1
2h j j
2
!3
2
NX1 p (x
~j (j +1 j )
p
i

t
i

(

+

)
x
)
j
+1
j
j
j
+1
j
j
5
exp 4
V (xj ) +
+ ~j W (xj )
h j =0
t
2m
t
2
"
#
Z

i Z T 1 2
y
Dx(t)D (t)D (t) exp h 0 dt 2 mx_ (t) V (x(t)) + (t)[it + W (x(t)) (t) :
A areful derivation of this formula and a step by step omparison with the matrix representation in the spin degrees of freedom an be found in se tions 1 to 3 of the paper \Fermioni
oordinates and supersymmetry in quantum me hani s", Nu lear Physi s B196 (1982) 509,
by P. Salomonson and J.W. van Holten. For further details see the le tures by L. Faddeev
in \Methods in eld theory", Les Hou hes 1975, ed. R. Balian and J. Zinn-Justin.
It is now also straightforward to derive the path integral for the Dira Hamiltonian of
the previous se tion. Using as a basis the plane waves onstru ted there, the Hamiltonian
be omes a de oupled sum (in a nite volume) for ea h ~k of four fermions, des ribed by
by1 (~k), by2 (~k), dy1 (~k) and dy2 (~k). Ea h of whi h an be des ribed by its own . Asso iating the
annihilation operators with their respe tive ~, and performing the Fourier transformation
ba k to oordinate spa e, it is left as an exer ise to show that the path integral for fermions
is given by

Z
Z

D (x)D (x) exp hi d4x (x)(i m) (x) J(x) (x) (x)J (x) : (15.37)
Sin e the elds (x) and (x) are Grassmann variables, also the sour es J (x) and J(x) are
Grassmann variables. Their order in the above equation is therefore important, when used
in further al ulations. As promised, it is as simple as for s alar eld theories to al ulate the
dependen e of this path integral on the sour es. Sin e Grassmann variables form a omplex
linear spa e we an perform all al ulations as in the s alar ase, provided we keep tra k of
the order of the Grassmann odd variables. In parti ular we an make the repla ement
!
(x) !
(x)
(x) + d4 x0 GF (x x0 )J (x0 ) ;
Z
(x) + d4 x0 J(x0 )GF (x0 x) ;
62
(15.38)
where GF (x) is the Green's fun tion dened in eq. (14.9). The integration measure, as for
ommuting variables, is invariant under a shift by a onstant Grassmann variable, su h that
we obtain (as for the s alar ase we normalise the path integral to 1 for vanishing sour es)

< 0jUJ J (T )j0 > = < 0jUJ =J=0 (T )j0 > exp i d4 xd4 y J(x)GF (x y )J (y )
< 0jUJ =J=0(T )j0 > Z2(J ; J) :
(15.39)
Again, this result holds to arbitrary order in the sour es, and agrees with what an be derived
to se ond order within Hamiltonian perturbation theory (see problem 24).
Intera tions are taken into a ount by adding higher order terms to the Lagrangian, where
the order of the fermion elds is important. For example, the Lagrangian for a fermioni
and a s alar eld is given by
L
L2
= L2 V ( ; ; ) J (x) (x) J(x) (x)

= (x)(i m) (x) + 12 (x) (x)
1
2
(x)J (x) ;
M 2 2 (x) :
(15.40)
We nd as in eq. (8.6)
!
i
i
i
; ;
) Z2 (J ; J; J ) ;
Z (J ; J; J; gn) = exp i d4 x V (
J (x) J (x) J (x)

Z

Z2 (J ; J; J ) = exp i d4 xd4 y 12 J (x)G(x y )J (y ) + J(x)GF (x y )J (y ) : (15.41)
Z
Note the minus sign for the derivative with respe t to J (x), whi h is be ause the sour e
stands behind the eld omponent (x). Derivatives of Grassmann variables are simply
dened as one would intuitively expe t
d
d 0
d
1=0 ;
=1 ;
=0 ;
(15.42)
d
d
d
together with a generalised Leibnitz rule for fun tions f and g that are either even or odd
Grassmann variables, a property denoted by the sign or grading s(= 1),
!
d
d
d
(fg ) = sf f g +
f g :
d
d
d
(15.43)
By de laring the derivative to be a linear fun tion on the Grassmann algebra, it an 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 dierentiation is one and
the same thing. Both proje t on the oe ient 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
perspe tive trivial. More importantly, to make sense of eq. (15.41), one easily shows the
following identity to hold
d
exp(x) = x :
(15.44)
d
An example of an intera tion between the fermions and a s alar eld is given by the
so- alled Yukawa intera tion
V ( (x); (x); (x)) = g (x) (x) (x) :

63
(15.45)
We an also onsider the intera tion of the fermions with the ele tromagneti eld, whose
quantisation will be undertaken in the next se tion. For this we an take the minimal
oupling in eq. (14.12) (see also eq. (14.10)), su h that
V ( (x); (x); A (x)) = e (x) (x)A (x) ;
(15.46)
whi h will play a dominating role in des ribing Quantum Ele trodynami s (QED).
As before, log Z (J; J ; J ) is the sum over onne ted diagrams. Diagrams that involve
fermions ne essarily have as many lines ending in a sour e J as in a sour e J. This is
be ause the Lagrangian is Grassmann even, a requirement that an be related to the Lorentz
invarian e. It does not in general require the Lagrangian to be bilinear in and . In se tion
19 and problem 30 we will dis uss the four-Fermi intera tion, (x) (x) (x) (x), whi h
is learly Lorentz invariant and Grassmann even. However, for many of the theories we
dis uss, the Lagrangian is bilinear in the fermioni elds, be ause higher order terms will
generally not be renormalisable (ex ept in one spa e and one time dimension). If no higher
order fermioni intera tions o ur, a fermioni line either forms a loop or it goes from a
sour e J to a sour e J. As hanging the order of fermioni elds and sour es gives a sign
hange, this has onsequen es for the Feynman diagrams too. However, it is umbersome
to determine the overall sign of a diagram. Fortunately, all we need is the relative sign of
the various diagrams that ontribute to the Green's fun tion with a xed number of sour es,
sin e the overall sign drops out in our omputations of ross se tions and de ay rates. If one
diagram an be obtained from the other by rossing two fermion lines, this gives a relative
minus sign, as for

-
XzXX XX

:
g. 6
It also implies that ea h loop formed by a fermion line arries a minus sign. Intuitively this
follows, as is indi ated in the following gure by the dashed box, from the identity above

XzXX XX
:

' $
-
g. 7

More a urately a fermion loop that onne ts verti es xk for k = 1 to n is asso iated to
n
i
i
(xk ) (xk ) !
!

k=1 J (xk ) J (xk )
k=1
h
X
Tr iGF (xk(1) xk(2) )iGF (xk(2)
Y
fkg
xk(3) ) iGF (xk(n)
xk(1) ) ; (15.47)
where fkg stands for the various orders in whi h the verti es are onne ted. For ea h vertex
we have only indi ated the fermioni part (xk ) (xk ). For the examples of eqs. (15.45) and
(15.46), the s alar or ve tor eld ontributions are not indi ated, as they are not relevant
for the fermion loop. The tra e is with respe t to the spinor indi es whi h are not displayed
expli itly to keep the notation simple. We used that
i
i
(xk(j ) ) (xk(j +1) ) !
J (xk(j ) ) J (xk(j +1) )
64
! iGF (xk(j) xk(j+1) ) ;
(15.48)
where an extra minus sign arises sin e in log Z2 (J ; J) (see eq. (15.39)) J omes rst, and
has to be anti ommuted with = J , before this derivative an be taken. The overall minus
sign omes from the term that loses the loop
xk(1) )A (xk(n) ) = A (xk(n) ) (
xk(1) ) ;
(
(15.49)
where A is Grassmann even. We ontrast this, as an example, with a s alar loop for the eld
' that arises in the theory dis ussed in se tion 11, whi h is des ribed by the Lagrangian
L = 21 ( ')2 12 m2 '2 + 21 ( )2 12 M 2 2 21 '2 J j'. One nds (k(n + 1) k(1))
n
i
i
'(xk )'(xk ) !
k=1
k=1 J (xk ) J (xk )
Y
X Y
fkg j =1
iG(xk(j )
xk(j +1) ) ;
(15.50)
whi h ompletes the demonstration of the extra minus sign for fermion loops. Note that the
fa tors of i, asso iated to the derivatives with respe t to the sour es, are absorbed in the
verti es for the Feynman rules in the table on page 35, whi h is why the propagator in that
table equals i times the Green's fun tion. This is also the Feynman rule for the fermion
propagator. But the extra minus sign in the derivative with respe t to the fermioni sour e
J (see eq. (15.41)) is not absorbed in the vertex in order to guarantee that vertex fa tors are
assigned as in the s alar theory. That minus sign is, however, absorbed in eq. (15.48) due to
the anti ommuting nature of the fermioni variables, whi h was in the rst pla e the reason
for the extra minus sign in eq. (15.41) to appear. Only the overall minus sign required in
losing a fermion loop remains as an extra fa tor.
Before we onvert these Feynman rules to the ones involved in omputing the s attering
matrix, ross se tions and de ay rates (see pg. 42 and eq. (9.25)), we need to determine
the wavefun tion fa tors to be inserted for the external lines that orrespond to the in- and
outgoing fermion lines. For this we express the reation and annihilation operators in terms
of the fermioni elds (at t = 0), su h that their insertion in the operator formulation an
be onverted in the path integral, as in the s alar ase, to derivatives with respe t to the
sour es. Using eq. (13.17) and the orthogonality relations of eq. (13.14) one nds ( mp.
eq. (7.26))
u() (~k)y Z
b (~k) = q
2k0 (~k)
v () (~k)y Z
d3~x
^ ~x) ; dy (~k) = q
e i~k~x (
(2 )3
2k0 (~k)
d3~x i~k~x ^
e (~x) (15.51)
(2 )3
and through onjugation we get
v () (~k)
d3~x
e i~k~x ^ y(~x) q
:
(2 )3
2k0 (~k)
(15.52)
In the Hamiltonian formulation the s attering matrix is given by
by (~k) =

Z
() (~k)
u
d
~
x
~k~x ^ y
3
i
~
q
e (~x) q
; d (k) =
(2 )3
2k0 (~k)
out< (p~1 ; 1 ); (p~2 ; 2 ); ; (p~` ; ` )j(~k1; 1 ); (~k2; 2 ); ; (~kn; n ) >in =

< 0j 1 (p~1 ) 2 (p~2 ) ` (p~` )U (Tout ; Tin ) y1 (~k1 ) y2 (~k2 ) yn (~kn)j0 > ; (15.53)
where in a short-hand notation we separate parti les form antipari les by the heli ity index
1 (~k) b1 (~k) ; 2 (~k) b2 (~k) ; 3 (~k) d1 (~k) ; 4 (~k) d2 (~k) :

65
(15.54)
Like in eq. (9.3), the insertion of a eld operator at the appropriate time is in the path
integral represented by a derivative with respe t to the sour e
h
i
h
i
0 u() (~k) Z
0 u() (~k)

i
i
a
a
^by (~k) = q
;
d3~x ei~k~x
= q
~
~
Ja (~x; t = Tin )
2k0 (~k)(2 )3
2k0(~k ) Ja (k; t = Tin )
() (~k) Z
v () (~k)
i
i
v
a
y
^
~
= qa
d3~x ei~k~x
d (k) = q
;
~
Ja (~x; t = Tin )
2k0 (~k)(2 )3
2k0 (~k) Ja (~k; t = Tin )
() ~ Z
u() (~k)
i
i
^b (~k) = q ua (k)
;
= qa
d3~x e i~k~x
Ja (~x; t = Tout )
2k0 (~k)(2 )3
2k0 (~k) J~a ( ~k; t = Tout )
h
i
h
i
0 v () (~k)
0 v () (~k) a Z

i
i
a
d^ (~k) = q
= q
;
d3~x e i~k~x
Ja (~x; t = Tout )
2k0 (~k)(2 )3
2k0 (~k) J~a ( ~k; t = Tout )
(15.55)
su h that
out< (p~1 ; 1 ); (p~2 ; 2 ); ; (p~` ; `)j(~k1 ; 1 ); (~k2 ; 2 ); ; (~kn; n ) >in
=
i=1
î (p~i )
j =1
^yj (~kj ) exp(GJ ;J )jJ =J=0 :
(15.56)
The ^ (~k) are of ourse dened in terms of ^b (~k) and d^ (~k) as in eq. (15.54). We ontinue
as in se tion 9 by rst xing the wavefun tion and mass renormalisations in terms of the
onne ted two-point fun tion (we leave it as an exer ise to show that in the absen e of
intera tions out< (p~; )j(~k; ) >in = e ip0 (p~)T ~k;~p; ).

G(2)
(J ; J ) =
-
-

- + -
- +
- ;
-

where the self-energy is now a 4 4 matrix given by ( i) the amputated 1P I

fun tion
iab (p) b

-a

(15.57)
2-point
(15.58)
(The 1P I diagram equals ab (p), when evaluated with the Feynman rules for the redu ed
matrix elements of table 4 (pg. 69) by dropping the wavefun tion fa tors.) The onvention
for these sort of diagrams is that momentum ows in the dire tion of the arrow, whi h points
to the rst spinor index (here a). With these denitions the two-point fun tion be omes
#
"
Z
1
(2)
~
J~ (p) ;
(15.59)
G (J ; J) = i d4 p Ja ( p)
=p m (p) + i" ab b
with between square bra kets the inverse of the 4 4 matrix (p m (p)+ i")ab , whi h is
the full propagator in the momentum representation for the onventions on pg. 42. As long
as we don't break the Lorentz invarian e, the full propagator near the poles is of the form of
the free propagator with a wavefun tion renormalisation fa tor ZF and a renormalised mass
m
~ , su h that on the mass-shell one has ( mp. eq. (9.12))
"
#
Z
1
(2)
~
G (J ; J) = iZF d4 p J( p)
J~(p) :
(15.60)
=p m
~ + i"
66
Performing the wavefun tion renormalisation, we have to modify eqs. (15.55) and (15.56)
a ordingly ( mp. eqs. (913) and (9.14)).
h
i
0 u() (~k)
() ~

i
i
a
^b (~k) = r uaq(k)
^b+ (~k) = r q
;
;
~
~a (~k; t = Tin )
a ( ~k; t = Tout )
J
2
2
2
2
~
~
2ZF k + m
2ZF k + m
~
~
h
i
() (~k)
0 v () (~k) a
i
i
v
a

^
~
^
~
r
;
d
(
k
)
=
d+ (k) = r q
;
q
~a (~k; t = Tin )
~
~
J
(
k;
t
=
T
)
J
a
out
~2
~2
2Z ~k 2 + m
2Z ~k 2 + m
F
(15.61)
and, with the ^ (~k) dened as in eq. (15.54),

out< (p~1 ; 1 ); (p~2 ; 2 ); ; (p~` ; ` )j(~k1; 1 ); (~k2; 2 ); ; (~kn; n ) >in
=
i=1
î (p~i )
j =1
^+j (~kj ) exp(GJ ;J )jJ =J=0 :
(15.62)
To ompute the wavefun tion fa tors for the external lines we express the n-point fun tion
in terms of amputated n-point fun tions, as in eq. (9.15), with the dieren e that there has
to be an even number of external lines, sin e the number of J and J sour es has to be equal
(we ignore for the moment any other bosoni elds that might be present, in luding those
in the external lines will be obvious). The amputated n-point fun tion will now arry the
spinor index of ea h of the external lines and one has
Z
n
Y
(p1 ; p2 ; ; pn ; k1 ; k2; ; kn)ab11;b;a22;;;b;ann
(15.63)
d4 pj d4 kj G(amp)
G(2 n) (J ; J)

j =1
8
9
"
#
"
#
n <
=
Y
i
i
~ (pj )
~
J
:
J
(
k
)
j
:
=pj m ( pj ) + i" j bj k=j m (kj ) + i" aj dj dj j ;
j =1
Note that we have assumed one parti ular ordering for the sour es. Relative signs of the
diagrams are determined by the rules that were des ribed above (it is not di ult to onvin e
oneself that with respe t to the fermion lines, any diagram an be generated from a given
one by permuting fermion lines). As in eqs. (9.20) and (9.21) we an ompute the a tion of
^ (~k), from whi h the wavefun tion fa tors will be obtained. Like in eq. (9.19), omputing
the a tion of ^b (p~) and d^+(p~) we an restri t our attention to
"
#
Z
i

~
A( ) d4 p Ja (p)
(p; )b =
G(amp)

=p m ( p) + i" ab
#
"
Z
Z
ip0 t
d4 p
ie
~

p
(p; )b : (15.64)
G(amp)
dt J a (p~; t)

=
p
m
(
p
)
+
i"
2
ab
p
If we dene as usual p0 (p~) = p~ 2 + m
~ 2 and for onvenien e of notation hange p to p in
eq. (15.64), we nd
"
#
() Z
e ip0 Tout
^b (p~)A( ) = q ua (p~)
( p; )b =
dp0
G(amp)

=
p
m
(
p
)
+
i"
4p0 (p~)ZF
ab
q
(=p + m
~ )ab G(amp)
( p; )b ip0 (p~)Tout

(

)

4
q
e
;
(15.65)
i (2 ) ZF ua (p~)
2p0 (p~)
2p0 (p~)(2 )3
67
whi h is obtained by deforming the p0 integration ontour to the lower half-plane (sin e
Tout ! 1), and omputing the ontribution from the pole at p0 = p0 (p~), taking into a ount
eq. (15.60). Its residue is proportional to the matrix ZF (=p + m
~ )=2p0 (p~), whi h is most easily
2
2
found using 1=(=p m
~ + i") = (=p + m
~ ) =( p m
~ + i"). We an now use the fa t that the
spinor u() (p~) satises the equations of motion (see eq. (13.2)), su h that
!
!
(=p + m
~ )ab
( 0=p 0 + m
~ ) ()
(=p + m
~ )y ()
(

)

ua (p~)
=
u (p~) =
u (p~) = u() (p~)b : (15.66)
2p0 (p~)
2p0 (p~)
2
p
(
p
~
)
0
b
b
Consequently, eq.(15.65) be omes
(amp)
q
b
^b (p~)A( ) = iu() (p~)b (2 )4 ZF G q ( p; ) e ip0 (p~)Tout :
(15.67)
2p0 (p~)(2 )3
p
The wavefun tion fa tor for an outgoing ele tron is therefore given by u() (p~)b ZF . The
onvention is that the momentum ows out of the diagram, along the arrow (see the table
below), this is why the amputated n-point fun tion has p as its argument, like for the
s alar ase, where we dened for the amputated n-point fun tion all momenta to ow into
the diagram. This means that in the redu ed matrix element M the outgoing ele tron
momenta o ur pre isely as indi ated in eq. (9.18).
By similar arguments we obtain from eq. (15.64)
"
#
() (p~) Z
ip0 Tin
v
e
a

d^+ (p~)A( ) = q
(p; )b =
dp0
G(amp)

=
p
m
(
p
)
+
i"
4p0 (p~)ZF
ab
q
(p; )b ip0 (p~)Tin
( =p + m
~ )ab G(amp)

(

)

4
q
e
=
i (2 ) ZF va (p~)
2p0 (p~)
2p0 (p~)(2 )3
(amp)(p; )b
q
Gq
eip0 (p~)Tin ;
(

)
4
iv (p~)b (2 ) ZF
(15.68)
3
2p0 (p~)(2 )
p
su h that the wavefun tion fa tor for an in oming antiparti le (positron) is v () (p~)b ZF .
In this ase the momentum ows against the arrow of the fermion line, but does ow into
the diagram as is required in the onvention of the redu ed matrix element.
To ompute the a tion of ^b+ (p~) and d^ (p~), we restri t our attention to
#
"
Z
i
(amp)

J~ (p) =
(p; )b
B ( ) d4 p G
=p m (p) + i" ba a
"
#
Z
ip0 t
ie
d4 p Z
(amp)
p
(p; )b
dt G
J~ (p~; t) : (15.69)
=p m (p) + i" ba a
2
One nds
h
i
"
# )
(
0 u() (p~)
Z
ip
T
0
in

e
a =
^b+ (p~)B ( ) =
q
(p; )b
dp0 G(amp)

=p m (p) + i" ba 4p0 (p~)ZF
q
~ )ba h 0 () i ip0 (p~)Tin
(p; )b (=p + m
G(amp)
4
u (p~) a e
=
i (2 ) ZF q
2p0 (p~)
2p0 (p~)(2 )3
q
(p; )b () ip0 (p~)Tin
G(amp)
4
ub (p~)e
;
(15.70)
i (2 ) ZF q
2p0 (p~)(2 )3
68
and (again for onvenien e hanging p to p)
d^ (p~)B ( ) =
(
Z
( p; )b
dp0 G(amp)

"
ip0 Tout
e
=p m ( p) + i"
ba
i
0 v () (p~) a
q
=
4p0 (p~)ZF
( p; )b ( =p + m
G(amp)
~ )ba h 0 () i ip0 (p~)Tin
4
i (2 ) ZF q
v (p~) a e
=
2p0 (p~)
2p0 (p~)(2 )3
q
G(amp) ( p; )b ()
vb (p~)e ip0 (p~)Tout :
(15.71)
i (2 )4 ZF q
3
2p0 (p~)(2 )
In both ases there is an extra minus sign from pulling = Jpthrough J in eq. (15.63). The
wavefun tion fa tor for an in oming ele tron is hen e u(b) (p~) ZF with the momentum owing along the fermioni arrow, whereas
the wavefun tion fa tor of an out oming antiparti le
(
) p
(or positron) is given by vb (p~) ZF , where the momentum ows opposite to the fermioni
arrow. The minus signs in front of some of the wavefun tion fa tors are irrelevant (they an
be absorbed in the overall sign ambiguity).
In the table below we summarise the Feynman rules that orrespond to the fermioni
pie es in omputing the redu ed matrix elements. We have hosen the onvention that
the in oming momenta ow in, and the outgoing momenta ow out of the diagram. This
guarantees that eqs. (9.18), (10.12) and (11.11) (resp. the s attering matrix, ross se tion and
de ay rate), remain valid in the presen e of fermions. Consequently, all fermion momenta
in the table ow from left to right. For onventions where momenta always ow in the
dire tion of the fermion arrow the four momentum for wavefun tion fa tors asso iated to
in- and outgoing antiparti les (positrons) should be reversed. Signs from fermion loops and
ex hanges of external lines will not be impli it in diagrams, as only relative signs are known.
q
table 4
k3
6
b*
HjH a
k2
k1
k3

b*
HjH6a
k1
k2
gab and k1 = k2 + k3
Yukawa vertex
e ab and k1 = k2 + k3
k= m1+i" ab
a
b k
a- pZF u() (~k) ; k0 =
a
; k a pZF v () (~k) ; k0 =
a
; k a
- pZ u() (~k) ; k =
photon vertex
F
a
; ka- pZF va() (~k) ;
; k

fermion propagator
q
~k 2 + m
~ 2 in oming ele tron
~k 2 + m
~ 2 in oming positron
~k 2 + m
~ 2 out oing ele tron
~ 2 outgoing positron
k0 = ~k 2 + m
1 i (2d4 k)4
1
R
for ea h fermion loop

inter hange of fermion lines
69
16 Feynman rules for ve tor elds

As before, the quantisation for ve tor elds starts by expanding the eld in plane waves and
identifying the Fourier oe ients with reation and annihilation operators.
A (x) =
X
d3~k
a (~k)"() (~k)e
2k0 (~k)(2 )3
Z
q
ikx + ay (~k )"() (~k ) eikx

(16.1)
where "() (~k)e ikx are independent plane wave solutions of the equations of motion. The
index enumerates the various solutions for xed momentum. They will be identied with
the spin omponents or heli ity eigenstates of the ve tor.
We will rst dis uss the simpler ase of a massive ve tor eld, expe ted to des ribe a
massive spin one parti le. In problem 12 we already saw that its Lagrangian is given by
(16.2)
LA = 1 F F + 1 m2 AA A (x)J (x) ;
4
and that the free equations of motion (i.e. J = 0) are equivalent to

A (x) = 0 and ( + m2 )A (x) = 0 :
(16.3)
As usual, this implies the on-shell ondition k02 = ~k 2 + m2 , but also

k "() (~k) = 0 :
(16.4)
It removes one of the four degrees of freedom of a four-ve tor. Three independent omponents
remain, exa tly what would be required for a parti le with spin one. We may for example
hoose
"() (~0) = ( = 1; 2; 3) :
(16.5)
in the restframe of the parti le, whi h is extended to an arbitrary frame by applying a
Lorentz boost. They satisfy the Lorentz invariant normalisation
0
0
g "() (~k) "( ) (~k) = :
(16.6)

The minus sign is just a onsequen e of the fa t that in our onventions gij = ij . The
spin wave fun tions also satisfy a ompleteness relation whi h is given by
(~k)
"() (~k)"() (~k) =

g
k k
m2
(16.7)
most easily proven in the restframe. Sin e the spin wave fun tions are by onstru tion
Lorentz ve tors, the above expression forms a Lorentz tensor and its on-shell extension to
an arbitrary frame is therefore unique. It is easily seen to proje t arbitrary ve tors w on
to ve tors that satisfy k w = 0. The propagator for the massive spin one eld was already
omputed in problem 12

g km k2

(~k)

=
=
:

k2 m2 + i"
k2 m2 + i"
k

(16.8)
Espe ially for the massless spin one eld (i.e. the photon eld) to be dis ussed below, it
is advantageous to de ompose the spin with respe t to the dire tion of the parti le's motion,
70
whi h are alled heli ity eigenstates. We have heli ities 0 and 1, des ribed by the spin
wave fun tions
1
j~kj ; k0 ~kA ; "() (k) =

"(0)

(k ) =
~
m mjkj
1
2

(2) ;
2 "(1)
(k ) i" (k )
(16.9)
(1)
(2)
where k , "(0)
(k ), " (k ) and " (k ) form a omplete set of real orthogonal four-ve tors.
These new spin wave fun tions satisfy the same properties as in eqs. (16.6) and (16.7) and
are also dened o-shell, where they satisfy
(k)
"() (k)"() (k) =

g
k k
k2
(16.10)
Sums over will of ourse run over the set f0; +; g in this ase. We leave it as an exer ise
to verify that rotations over an angle around the axis pointing in the dire tion of ~k leaves
()
i ()
"(0)
(k ) invariant and transforms " (k ) to e " (k ).
The Hamiltonian for the massive spin one parti les is given by
H=

X
ay (~k)a (~k) + a (~k)ay (~k) :
d3~k 21 k0 (~k)
(16.11)
whi h an be expressed in terms of three s alar elds ' , = 1; 2; 3, as
H=
d3~k
1
2
j~(~k)j2 + (~k 2 + m2 )j'~(~k)j2

1
2
where we dened, as in eqs.(2.6) and (2.7),

d3~x 12 2 (~x) + 21 (i ' (~x))2 + 12 m2 '2 (~x) :
(16.12)

q

iq
'~ (~k) = a (~k) + ay ( ~k ) = 2k0 (~k) ; ~ (~k) = 2k0 (~k) ay (~k) a ( ~k) : (16.13)
2
The orresponding Lagrangian would be given by

X
L' = 1 ( '(x))2 1 m2 '2 (x)

' (x)J () (x)
(16.14)
where we have added a sour e for ea h s alar eld. We introdu e also a eld for the
omponent of the ve tor eld along k . Writing
X
X
1
1
A~ (k) ~ (k)k + '~ (k)"() (k) ; J~ (k) ~j (k)k + J~() (k)"() (k) ; (16.15)
m
m

a simple al ulation shows that LA L' 21 ( (x))2 (x)j (x). We see that de ouples
from the other omponents and behaves like a s alar parti le with the wrong sign for the
kineti term. This would lead to serious in onsisten ies, whi h are ir umvented if we take
J (x) = 0, su h that we an put 0.
It is important to realise that it is the Lorentz invarian e that requires us to des ribe a
spin one parti le by a four-ve tor. From the point of view of the s alar degrees of freedom,
' , this invarian e seems to be lost when we introdu e intera tions in the Lagrangian of
eq. (16.14). Nevertheless, if we treat as a three-ve tor index (in some internal spa e) and
demand the intera tions to be O(3) invariant with respe t to this index (i.e. invarian e
under rigid rotations and re e tions in the internal spa e), then we laim that the resulting
71
intera tions do respe t the Lorentz invarian e. The reason is simple, be ause the O(3)
invarian e requires that the index is always pairwise ontra ted. Eqs. (16.10) and (16.15)
guarantee that su h a pair, written in terms of the ve tor eld A (x), is a Lorentz s alar as far
as it on erns the dependen e on A, whi h is su ient if the Lagrangian is Lorentz invariant
when treating as a dummy label. To eliminate the eld we have to enfor e A (x) = 0,
not only on-shell, but also o-shell. This an be a hieved by adding to eq. (16.2) a term
(x) A (x) ( mp. problem 8). Using
Z
DA(x)D(x)
exp
i d4 x (x)
A (x)
DA~(k)
(k A~ (k)) ;
(16.16)
we see that the so- alled Lagrange multiplier eld (x) plays the role of removing the unwanted degree of freedom. Sin e we modied the theory o-shell, the propagator in eq. (16.8)
has to be hanged also, by repla ing (~k) by its o-shell value (k) (eq. (16.10)). If none of
the verti es or sour es ouple to the eld, we might just as well repla e it by g . There
are a number of other ways to eliminate the degree of freedom, see e.g. se tion 3-2-3 in
Itzykson and Zuber. We will ome ba k to massive ve tor parti les in se tion 19.
For the omputation of the s attering matrix we express the reation and annihilation
operators in terms of the ve tor elds at t = 0, using the relations
Z
q
d ~x
2k0 (~k)"() (~k) q 3 A^ (~x)e i~k~x ;
a (~k) =
(2 )3
Z
q
d ~x
(16.17)
2k0 (~k)"() (~k) q 3 A^ (~x)ei~k~x :
ay (~k) =
(2 )3
whi h in the path integral turn into
q
q
i
i
; a^ (~k) = 2k0 (~k)"() ( ~k) ~ ~
;
a^y (~k) = 2k0 (~k)"() (~k) ~ ~
J (k; t = Tin )
J ( k; t = Tout )
(16.18)
whi h is identi al to eq. (9.3), when re-expressed in terms of J () . Like for s alar and fermion
elds there will be a mass and wavefun tion (denoted by ZA ) renormalisation, determined
through the self-energy of the ve tor eld, whi h is now a Lorentz tensor of rank two. It
is proportional to (k) (to guarantee that the s alar eld introdu ed above de ouples
from the other elds; alternatively it an be seen as a onsequen e of the O(3) invarian e
with respe t to the index ). We an onsequently dene
(p) (p)A (p) :
(16.19)
The n-point Green's fun tions an now be written in terms of amputated Green's fun tions that arry four-ve tor indi es for ea h external spin one line,
8
Z <
Y
=
ij j (pj )
G( n) (J ) : d4 pj J~j (pj ) 2
Gamp
(p1 ; p2 ; ; pn )1 ;2 ;;n :

2
;
m

(
p
)
+
i"
p
A j
j
j =1
(16.20)

(

)
(

)
~
~
~
Using the fa t that on-shell (k)" (k) = " (k), we nd for the in oming spin one line
p
p
a wavefun tion fa tor ZA"() (~k) and for the outgoing line a fa tor ZA "() (~k) .
For a massless spin one eld (the photon) we would expe t the heli ity zero omponent
of the ve tor eld to be absent. First we have to redene, however, what we would mean
72
with the zero heli ity omponent, be ause eq. (16.9) is singular in the limit of zero mass.
We take as our denition
1
0
~
p
k
1
A

"(0)
; "() (k) = (0; ~s (~k)) ;
(16.21)
(k ) = n (k ) 2 2 1; ~
jkj
with ~k ~s(~k) = 0. These still form with k four independent four-ve tors and "() (~k) are still
~
transverse polarisations. However, it is not longer true that k "(0)
(k ) will vanish on-shell (i.e.
at k2 = 0). This is easily seen to imply that on-shell a(0) (~k) = 0. In other words, on-shell
there is no longitudinal omponent for the photon. The extra degree of freedom is removed
by the gauge invarian e of the massless ve tor eld, as was dis ussed at the end of se tion
4 and in problem 9. To perform the quantisation of the theory, one an go about as in the
massive ase. Due to the presen e of the four-ve tor n(k) it will be mu h more umbersome
to demonstrate the Lorentz invarian e. In se tion 20 it will be shown how in prin iple in
any gauge the path integral an be dened and that the result is independent of the hosen
gauge. One ould then hoose a gauge that allows us to show the equivalen e between the
path integral and the anoni al quantisation. However, it is the great advantage of the path
integral formulation that al ulations an be performed in a gauge in whi h the Lorentz
invarian e is manifest. The gauge most suitable for that purpose is of ourse the Lorentz
gauge A (x) = 0, see eqs. (4.21) and (4.22). The propagator is read o from eq. (4.26)
1 ) k2 k () (k)
(1
k +i"

;
(16.22)
2
2
=
k + i"
k + i"
where is an arbitrary parameter, on whi h physi al observables like ross se tions and de ay
rates should not depend. It is in general not true anymore that the self-energy is proportional
to () (k), but the gauge invarian e does guarantee that (k)~j (k) is independent of for
any onserved urrent, i.e. k~j (k) = 0. It an be shown that this in general implies
0
(k) = ( ) (k)(k) ;
(16.23)
for some, possibly innite, 0 . Apart from a wavefun tion renormalisation (ZA ), the gauge
xing parameter will in prin iple have to be renormalised too. One still has for any value of
that () (k) "() (~k) = "() (~k). The wavefun tion fa tors for external photon lines are
therefore identi al to the ones for the massive ase, ex ept that now only two heli ity states
an appear. It is an important onsequen e of gauge invarian e that unphysi al degrees of
freedom de ouple in the physi al amplitudes. It also implies that the self-energy vanishes
on shell (see problem 39), su h that it will not give rise to a renormalisation of the mass.
The photon remains massless. That the gauge invarian e must be ru ial here is lear, as a
massive photon would have one extra degree of freedom. We will ome ba k to this point in
se tion 19. In the table below we summarise the Feynman rules.
table 5

g (1 1 ) kk2+ki"
k

photon propagator (Lorentz gauge)
k2 +i"

g
k

; k -

; k
pZA"() (~k)
pZ "() (~k)
A
in oming photon
outgoing photon
73
17 Quantum Ele trodynami s - QED

QED is the eld theory that des ribes the intera tion between the photon and the harged
fermions. In the Lorentz gauge (see eqs. (4.21) and (4.22)) the Lagrangian is given by
X
L = 1 F (x)F (x) 1 (A (x))2 + f (x)(i D mf ) f (x) :
(17.1)
4
Here f is de so- alled avour index, whi h distinguishes the various types of fermions (ele trons, protons, et .). The ovariant derivative D is given as before (see eqs. (3.35) and
(14.11)) by
D f (x) = ( + iqf A (x)) f (x) :
(17.2)
For ele trons we have qf = e and for protons qf = e. For = 0 the Lagrangian is invariant
under gauge transformations
A (x) ! A (x) + (x) ; f (x) ! exp( iqf (x)) f (x) :
(17.3)
The Feynman rules are olle ted in the table below.

k3

b*
HjH6a
k2
k1
b k a
g
q ab

(1 1 ) kk2+ki"
k2 +i"
table 6
photon propagator (Lorentz gauge)
and k1 = k2 + k3 photon vertex (fermion harge is q )

i
fermion propagator
k= m+i" ab
Before al ulating ross se tions we wish to dis uss in more detail the heli ity of the
fermions and its relation to harge onjugation C . The latter relates, say ele trons to
positrons, or in general parti les to anti-parti les, whi h is an important symmetry of the
theory. It, as well as parity (P ) and time reversal (T ) symmetry an be separately broken,
but the ombination CPT is to be unbroken to allow for a lo al, relativisti invariant eld
theory. The spin omponents of the solutions in eq. (13.8) were based on a de omposition
along the z -axis in the restframe. Heli ity, as for the photon, is dened by de omposing
the spin in the dire tion of motion, ~k. It is hen e dened in terms of the eigenvalues of the
operator
!
1^
^k J~ ki "ijk jk = 2 k ~
(17.4)
12 k^ ~ :
4j~kj
(J~ is the spin part of the angular momentum operator, the equivalent of 21 ~ for a two omponent spinor.) This holds both in the Dira and Weyl representations. It is easy to
~ k= = 0, e.g. by making use of the fa t that in the Dira representation
verify that [k^ J;
k= =
k 0 12
~k ~
74
~k ~
k 0 12
(17.5)
This implies that we an hoose u(0) and v0() to be eigenstates of the heli ity operator k^ J~
( onsequently they be ome fun tions of k^)
~ ~k) :
~ ~k) = 12 (
k^ J~ (
(17.6)
Instead of the label we an use to indi ate the heli ity and we have
(k= + m) ^
~ 0 (k^) = 12 u0 (k^) ;
u0 (k) ; k^ Ju
m + jk0 j
( k= + m) ^
~ 0(k^) = 12 v0 (k^) :
v (~k) = q
v0 (k) ; k^ Jv
m + jk0 j
u (~k) =
(17.7)
Note the ip of heli ity for the positron wave fun tions. For ~k = (0; 0; k) these eigenstates
oin ide with the de omposition in eq. (13.8). It is lear that we an dene
1
' (k^)
^ ' (k^) = ' (k^) ;
0 C
u0 (k^) = B
A ; k ~

0
1
0
0
^ (k^) = (k^) ;
0 C
v0 (k^) = B
A ; k ~

(k^)
0
(17.8)
with ' and ea h an orthonormal set of two- omponent spinors. They an be related to
ea h other by
(k^) i2 ' (k^) :
(17.9)
Indeed, when we use that
2 i 2 = i ; i = 1; 2; 3 ;
(17.10)
whi h expresses the fa t that SU(2) is so- alled pseudo real, we nd
k^ ~
i2 ' (k^) =
ik^ ~ 2 ' (k^) = i2 k^ ~ ' (k^) =
i2 ' (k^)
: (17.11)
As eq. (17.9) relates the omponents of the ele tron wave fun tion to those of the positron
wave fun tion, it is the basis of the harge onjugation symmetry, whi h relates the solutions
of the Dira equation to solutions of the omplex onjugate Dira equation (see eq. (12.31)),
whi h indeed inter hanges positive and negative energy solutions, i.e. parti les and antiparti les. To formulate this symmetry in the four- omponent spinor language one introdu es
the harge onjugation matrix (in the Dira representation)
C i 0 2 =
i2
whi h satises
i2 ;

(17.12)
C 1 = C y = C ; C C 1 = t :
(17.13)
This an be proven from the expli it form of the Dira matri es. The equivalent of eq. (17.10)
is given by
2 2 = :
(17.14)
75
It is now easy to verify that
v (~k) = C ut(~k) ; u (~k) = C vt (~k) :
(17.15)
We just need to prove one of these identities, be ause harge onjugation is an involution,
i.e. applying it twi e gives the identity

C C t
= C 0t C 0y = :
We nd
(17.16)
1
0
(k^) 1
'
' (k^)
( k= + m) 2 B C
B
0 C
0 A
C ut(~k) = C 0t u(~k) = i 2 u (~k) = i 2 q
A = iq

m + jk0 j
m + jk0 j
0
0
1
1
0
0

^
( k= + m) B 0 C
(q k= + m) i2 B ' (k) C q
0 A = v (~k) : (17.17)
0
=
A =

i

2
m + jk0j
m + jk0 j (k^)
0
(k= + m)
Under harge onjugation the harge that appears in the ovariant derivative in eq. (17.2)
should hange sign too. To show this we multiply the omplex onjugate of the Dira equation
with i 2

i 2 [( i ( ieA ) + m) = i 2 2 ( + ieA ) + m ( i 2 )
= (

i (
+ ieA ) + m) (C t ) :
(17.18)
That harge onjugation is really a symmetry of the quantum theory, is most onvin ingly
demonstrated by the fa t that the Dira Lagrangian is invariant under harge onjugation.
Using C t = t 0 C y 0 = t C 1 , the anti- ommuting properties of the fermi elds and
partial integration, we nd
Z
Z
d4 x C t (i ( + ieA ) m) C t = d4 x t C 1 (i ( + ieA ) m) C t =
Z
d4 x
i t (
+ ieA )
m =
d4 x (i (
ieA ) m) :
(17.19)
In parti ular we see that the ele tromagneti urrent generated by the fermi elds transforms
as required for the inter hange of parti les and anti-parti les, under whi h the harge hanges
sign
C
:
(17.20)
j = !
An important onsequen e of the harge onjugation symmetry is Furry's theorem, whi h
states that a fermioni loop with an odd number of verti es will not ontribute to the
amplitude. Consider a fermion loop as in gure (8a) below
k2
k2

(a)
Y
-p
*j
-p

n
(b)
g. 8
k1
k
k1
k
for whi h the Feynman rules lead to the expression
(note that spinor index ontra tions run
P
against the arrow of the fermion line and i ki = 0)
q n Tr

n

1
1
1
1
2
=p m + i" =p + k=1 m + i" =p + k=1 + k=2
76
m + i"
3
n
(17.21)
Using the fa t that for any matrix A we have Tr(A) = Tr(At ) = Tr(CAt C 1 ), we onvert
eq. (17.21) to the expression
q )n Tr
n
3
1
=p k=1 k=2
2
1
1
=p m + i"
;
(17.22)
whi h is exa tly ( 1)n times the result of the Feynman diagram that is obtained by inverting
the orientation of the fermion line (i.e. the verti es are onne ted in the reversed order) as
indi ated in gure (b). As both diagrams will o ur, their ontributions will an el whenever
n is odd. It onrms the intuition that parti les and anti-parti les ontribute equally, ex ept
for their opposite harge fa tors (q )n .
We will now al ulate the ross se tion for ele tron-ele tron s attering (the so- alled
Mller ross se tion). In lowest non-trivial order there are only two diagrams that ontribute,
as indi ated in the gure.
k2 ; tj2 p2 ; s2
k2 ; tj2 p2 ; s2
H
H
(
m + i"
H*
=p k=1
m + i"
H
R
*

*HjH
k1 ; t1 p1 ; s1
k1 ; t1

p1 ; s1
g. 9
The labels ti and si indi ate the heli ities of the in oming and outgoing ele trons. The
s attering matrix (ignoring the time dependent phase fa tor) for this pro ess is given by
out< (p~1 ; s1 ); (p~2 ; s2 )j(~k1 ; t1 ); (~k2 ; t2 ) >in =
(p1 + p2 k1 k2 )M(f( p1 ; s1 ); ( p2 ; s2 )g; f(k1 ; t1 ); (k2 ; t2 )g)
=
i(2 )4 4 q
(2) (p~ )(2 )3 2k(1) (~k )(2 )3 2k(2) (~k )(2 )3
3
2p(1)
(
p
~
)(2

)
2
p
0 1
0 2
0 1
0 2

n
i4 (p1q+ p2 k1 k2 ) us1 (p1 )e ut1 (k1 )g us2 (p2 )e ut2 (k2 )
(2) (1) (2)
(k1 p1 )2 + i"
(4 )2 p(1)
0 p0 k0 k0
us1 (p1 )e ut2 (k2 )g us2 (p2 )e ut1 (k1 ) o
: (17.23)
(k1 p2 )2 + i"
The relative minus-sign is of ourse a onsequen e of the so- alled Fermi-Dira statisti s,
whi h implements the Pauli prin iple. We got rid of the gauge dependent part of the photon
propagator (see table 6) by using the fa t that the urrents generated by us (p) ut (k) are
onserved, su h that
us(p) ut (k)(p
k ) = us (p)[(=p m) (k= m)ut (k) = 0 ;
(17.24)
be ause on-shell (k= m)u(k) = 0 (and hen e also u(k)(k= m) = 0). Indire tly, through
urrent onservation, this is related to gauge invarian e. It guarantees that the longitudinal
omponent of the photon does not ontribute to the s attering matrix, whi h is thus seen
not to depend on the gauge xing parameter .
The dierential ross se tion for unpolarised ele tron-ele tron s attering is given by (see
eq. (10.12), from now on we will drop the distin tion between ki and ki)
X
d3 ~p1
d3 ~p2
(2 )4 q4 (p1 + p2 k1 k2 )
d =

~1 )(2 )3 2p0 (p~2 )(2 )3
s1 ;s2 2p0 (p
4 (k1 k2 )2 m4
X
1
jM(f( p1; s1); ( p2 ; s2)g; f(k1; t1); (k2; t2 )g)j2 ; (17.25)
4
t1 ;t2
77
where 14 t1 ;t2 stands for averaging over the polarisations of the in oming ele trons. For
the total ross se tion we should multiply with a fa tor 21 to avoid double ounting the
identi al outgoing ele trons, or restri t the s attering angle to the interval 2 [0; =2,
when integrating over the outgoing momenta. The latter onvention will be followed here.
In the enter of mass system the s attered parti les move ba k to ba k in a dire tion whi h
is only determined modulo , whi h is why 2 [0; =2, with measured from the in oming
parti le dire tion (also dened modulo ).
To al ulate jMj2 we use
X
s;t
us(p) ut (k) (us(p) ut (k)) =
Tr
ut (k)
ut (k)
s;t
us(p) ut (k) ut (k) 0 y 0 us(p) =

us (p)
us (p) = Tr (k= + m) (=p + m) ; (17.26)
whi h an be represented graphi ally as follows

k;jt p; s
k

HH- -
k; t X k;jHtH- p *k; t X k;-t I
*
X

=
=
- :

=

p

p
t
t
s;t

M M
g. 10
Hen e, we add a Feynman rule for the so- alled ut fermion propagator
-k a = sign(k0)(k= + m)ab :
(17.27)
b
For anti-parti les k0 < 0 (see eq. (13.18) for the extra minus sign). These results an be
generalised to other elds too, by noting that our onventions have been su h that the
propagators an be written as
P
~ ~
(k )
(k )
;
(17.28)
k2 m2 + i"
where (~k) are the wave fun tions for the in oming lines and (~k) for the outgoing lines,
with labelling the internal degrees of freedom ( mp. eq. (16.8)).
We an now use this result to ompute jMj2
k
k
2
2
p
!

2

-

X
R

p1 p2
jMj2 =
+ (p1 ! p2 )

p
s1 ;s2 ;t1 ;t2
1
k
1
1 k
Tr[ (k=1 + m) (=p1 + m) Tr[ (k=2 + m) (=p2 + m)

= e4
((k1 p1 )2 + i") ((k2 p2 )2 + i")
)
Tr[ (k=1 + m) (=p2 + m) (k=2 + m) (=p1 + m)
((k1 p1 )2 + i") ((k1 p2 )2 + i")
+(p1
! p2 ) :
(17.29)
To ompute the tra es over the gamma matri es we use the following identities (problem 21)
Tr( ) = 4g ; Tr( ) = 4(g g + g g g g ) ;
X
X
= 4g ;
= 2 ;
(17.30)

78
and the fa t that the tra e over an odd number of gamma matri es vanishes. This implies

Tr (k= + m) (=p+ m) = Tr k= =p +4m2 g = 4(m2 k p)g +4k p +4k p ; (17.31)
and
(k= + m) (=p + m) = 2=p k= 2m2 + 4mk + 4mp :
(17.32)
Together with momentum onservation (p1 + p2 = k1 + k2 ) and the on-shell onditions

(p21 = p22 = k12 = k22 = m2 ), whi h imply identities like p1 p2 = k1 k2 , we nd

s1 ;s2 ;t1 ;t2
jMj2 = e4
16
g (m2
k1 p1

Tr 2=p2 k=1 + 2m2

(k1
) + k p
1 1
(k1
+ k p

1 1
g (m2

p1 )2 + i" (k2
4mk1
k2 p2 ) + k2 p2 + k2 p2

p2 )2 + i"
4mp2 (k=2 + m) (=p1 + m)

2
2
p1 ) + i" (k1 p2 ) + i"

)
+ (p1 $ p2 )
(k k )2 + (k1 p2 )2 + 2m2 (k1 p2 k1 k2 )

= 32e4 1 2

2
2
(k1 p1 ) + i"
)
2 2m2 k1 k2
(
k

k
)
1
2

+ (p1 $ p2 ) : (17.33)

2
2
(k1 p1 ) + i" (k1 p2 ) + i"
(
The parameter determines the relative sign of the \ rossed" diagrams in eq. (17.29),
whi h arise from multiplying the dire t ele tron-ele tron s attering diagram with the omplex
onjugate of the one where the outgoing fermion lines were rossed. For Fermi-Dira statisti s
1. By keeping tra k of the dependen e on one sees how s attering experiments an
be used to verify the anti ommuting nature of the ele trons.
We nally perform some kinemati s and express the dierential ross se tion in terms of
the s attering angle . We dene in the enter of mass frame
k10 = k20 = p01 = p02 E ; p~1 = p~2 ~p ; ~k1 = ~k2 ~k with jp~j = j~kj : (17.34)
Dening to be the angle between ~k and p~, i.e. p~ ~k = ~k 2 os , we have
k1 k2 = E 2 + ~k 2 = 2E 2 m2 ; k1 p1 = E 2 ~k 2 os = E 2 (1 os ) + m2 os ;
(p1 k1 )2 = 4~k 2 sin2 ( 12 ) ; k1 p2 = E 2 + ~k 2 os = E 2 (1 + os ) m2 os ;
(p2 k1 )2 = 4~k 2 os2 ( 21 ) ; (k2 k1 )2 m4 = 4E 2 (E 2 m2 ) = 4~k 2 E 2 : (17.35)
Finally we use that identity (
is the solid angle, d
= sin dd)
Z
d3 p~1 d3 ~p2 4 (p1 + p2 k1 k2 ) =
q
p2 dpd
2 p~ 2 + m2
Colle ting all terms we nd for eq. (17.25) the result
2 ~k 2 + m2 =
d
12 E j~kj :
(17.36)
X
1
(2q
)4 12 E j~kj 1 X
1
d
2=

=

jMj
jMj2 ;
4
2
10
2
2
d
(2E (2 )3 ) 4 4~k 2 E 2 s1 ;s2 ;t1 ;t2
2 E s1 ;s2 ;t1 ;t2
79
(17.37)
with
s1 ;s2 ;t1 ;t2
jMj2 given by
(
(2E 2 m2 )2+[E 2 (1+ os ) m2 os 2+2m2 [E 2 (1+ os ) m2 os + m2 2E 2
4
32e
+
16(~k 2 )2 sin4 ( 12 )
(2E 2 m2 )2+[E 2 (1 os ) + m2 os 2+ 2m2 [E 2 (1 os )+ m2 os + m2 2E 2
16(~k 2)2 os4 ( 21 )
)
2(2E 2 m2 )(2E 2 3m2 )
~2 2 2 1
16(k ) sin ( 2 ) os2 ( 12 )
(
32e4 [(2E 2 m2 )2+ E 4+(E 2 m2 )2 os2 +2m2 (m2 E 2 )( os4 ( 21 )+sin4 ( 12 ))
= ~2 2
+
sin4
(k )
)
2(2E 2 m2 )(2E 2 3m2 )
os [2E 2 (E 2 m2 )+2m2 (E 2 m2 )( os4 ( 12 ) sin4 ( 21 ))

sin4
4 sin2
(
)
16e4
4(2E 2 m2 )2
4(E 2 m2 )2 3(2E 2 m2 )2
= 2 2 2 (E 2 m2 )2 +

;
(17.38)
(E m )
sin4
4 sin2
2
2
1 is the ne-stru ture onstant)
yielding (e = 4e = 4e h 137
!)
(
3
(E 2 m2 )2
4
4
d e2 (2E 2 m2 )2
+
+
:
(17.39)
=
1
d
4E 2 (E 2 m2 )2 sin4 sin2 (2E 2 m2 )2
sin2
This ross se tion is invariant under ! , su h that we annot tell the two outgoing
ele trons apart, as it should be. For ele tron-ele tron s attering we have to put = 1 and
we see that one an easily distinguish experimentally if ele trons behave a ording to the
Fermi-Dira statisti s.
In problem 29 ele tron-positron s attering is studied with in the nal state an ele tron
and a positron (Bhabha s attering) or a muon and an antimuon. Both for e e ! e e and
e e+ ! e e+ one annot take too lose to zero (or for e e ). Apart from the fa t that
the dete tor would be in the way of the beam, it is fundamentally impossible to distinguish
the s attered parti les at = 0 (and = for e e ) from those in the beam. The divergen e
of the dierential ross se tion was therefore to be expe ted. For e e+ ! + this
divergen e is absent and one an dene the total ross se tion by integrating over all angles.
For E me and E m one nds (see problem 29) = 31 e2h 2 2 =E 2 = 21:7nb=E 2 (GeV).
We now dis uss ele tron-photon s attering, also known as Compton s attering. The
resulting ross se tion is alled the Klein-Nishina formula. There are again two diagrams
that ontribute in lowest non-trivial order to the s attering matrix.
k; t
-

p; s
The ross se tion is now given by
d =
p0 ; s0
k0 ; t0
k; t p; s
- I
p0 ; s0

k0 ; t0
d3 ~p 0
(2 )4 4 (p + p0 k k0 )
d3~k0

0 ~0 )(2 )3 2k0 (~k0 )(2 )3
4
j
p

k
j
s0 ;t0 2p0 (p
0
X
1
jM(f( p0; s0); ( k0; t0)g; f(p; s); (k; t)g)j2 ;
4
g. 11
s;t
80
(17.40)
where, as for ele tron-ele tron s attering, we will dis uss unpolarised ross se tions. This
requires averaging over the polarisations of the in oming parti les (at the end we will mention
the dependen e on the photon polarisations).
Note that the photon has also two heli ity
P
eigenstates, together with the ele tron s;t ontains four terms. The redu ed matrix element
for the two diagrams is given by
us0 (p0 )e "(t0 ) (k0 ) (=p + k= + m)e "(t) (k)us(p)
M=
+ ((k; t) ! ( k0 ; t0 )) : (17.41)
(p + k)2 m2 + i"
We leave it as an exer ise to verify that the ut photon propagator, for the hoi e of polarisations dis ussed in eq. (16.21), is given by (k2 0)
!
k
X
k n (k) + k n (k)
(
t
)
(
t
)

g
:
(17.42)
t= " (k)" (k) =
k n(k )
Like for ele tron-ele tron
s attering, we an ompute jMj2 graphi ally by
0
X
s1 ;s2 ;t1 ;t2
jMj2 =
p+k 6 ?+

p

0 0
p k

p

k

p0 0
p

R ?
6
p
+ (k 0
! k)
Tr[ (=p + k= + m) (=p + m) 0 (=p + k= + m) 0 (=p0 + m)

4
=e
((p + k)2 m2 + i")2
)
Tr[ (=p + k= + m) (=p + m) 0 (=p k=0 + m) 0 (=p0 + m)
+

((p + k)2 m2 + i") ((p k0 )2 m2 + i")
X
X
0
0
"(t ) (k0 )"(t0 ) (k0 ) "(t0) (k)"(t) (k) + (k0 ! k) : (17.43)
t=
t0 =
The gauge invarian e, i.e. onservation of the fermioni urrent, is again instrumental in
de oupling the longitudinal omponent of the photon eld. In this ase the argument is
somewhat more subtle. Consider for example the term from the ut photon propagators
that ontains k . It gives rise to the ombination (using that p2 = m2 )
(
(=p + k= + m) k (=p + m) = (=p + k= + m)f(=p + k= m) (=p m)g(=p + m)

= ((p + k)2 m2 )(=p + m) :
(17.44)
This means that one of the photon verti es is removed. There remain two diagrams, ea h
with one fermion loop and with an odd number of verti es. Furry's theorem tells us that
these two diagrams add to zero. We may therefore just as well repla e the ut photon
propagator by g . Using this we nd
Tr[ (=p + k= + m) (=p + m) (=p + k= + m) (=p0 + m)

2
4
jMj = e
+
((p + k)2 m2 + i")2
s1 ;s2 ;t1 ;t2
)
Tr[ (=p + k= + m) (=p + m) (=p k=0 + m) (=p0 + m)
+ (k0 $ k) : (17.45)
((p + k)2 m2 + i") ((p k0 )2 m2 + i")
X
Taking the in oming ele tron at rest (p = (m; ~0)), following similar steps as for ele tronele tron s attering, one will arrive at the result
!
!
e2 k00 2 k0 k00
d
2
=
+
sin
;
(17.46)
d
2m2 k0
k00 k0
81
where is the angle of the s attered photon with the dire tion of the in ident photon. From
energy and momentum onservation one nds that
k0
k00 =
:
1 + (k0 =m)(1 os )
(17.47)
For a detailed derivation we refer to se tion 5-2-1 of Itzykson and Zuber and to se tion 86
of Berestetskii, e.a. (see se tion 1 for the referen e).
In Itzykson and Zuber, as for most other textbooks, the result is derived by hoosing the
photon polarisation su h that "(k) p = 0 (keeping "(k) k = 0). With this hoi e it is even
possible to determine the polarised ross se tion (the polarisation of the ele tron is assumed
not to be observed)
!
!
!
d
e2 k00 2 k0 k00
=
+ + 4("0 ")2 2 ;
(17.48)
d
pol 4m2 k0
k00 k0
where " and "0 are the polarisations of resp. the in ident and s attered photon. When
k00 m one obtains the well-known Thomson formula
!
e2 0 2
d
= (" ") :
(17.49)
d
pol m2
The unpolarised ross se tion in this limit is obtained by summing over the s attered and
averaging over the in ident polarisations
2
8e2
d
= e2 (1 + os2 ) and =
:
(17.50)
d
2m
3m2
18 Non-Abelian gauge theories

Quantum Ele trodynami s is an example of a U(1) gauge theory. U(1) is the group of the
unimodular omplex numbers and determines the transformation of the harged elds
(x) ! exp( iq (x)) (x) g (x) (x) :
(18.1)
It forms a group, whi h means that for any two elements g; h 2 U(1), the produ t is also in
U(1). Furthermore, any element has an inverse g 1 , whi h satises gg 1 = g 1 g = 1. The
unit 1 satises g 1 = 1g = g , for any g 2 U(1). U(1) is alled an Abelian group be ause its
produ t is ommutative. For every g; h 2 U(1), gh = hg .
It is now tempting to generalise this to other, in general non- ommutative groups, whi h
are alled non-Abelian groups. It was the way how Yang and Mills dis overed SU(2) gauge
theories in 1954. Like for U(1) gauge theories they made the SU(2) transformation into a
lo al one, where at every point the eld an be transformed independently. (It should be
noted that they were originally after des ribing the isospin symmetry, that relates protons
to neutrons, whi h form a so- alled isospin doublet.)
The simplest non-Abelian gauge group, for whi h no longer gh = hg , is SU(2). This
group is well-know from the des ription of spin one-half parti les. It has a two dimensional
(spinor) representation, whi h an also be seen as a representation of the rotation group
SO(3). As a lo al gauge theory it does no longer a t on the spinor indi es, but on indi es
related to some internal spa e, giving rise to so- alled internal symmetries. The way the
82
gauge group G a ts on the elds is des ribed by a representation of the group G. A

representation denes a mapping from G to the spa e of linear mappings Map(V ), of the
linear ve tor spa e V into itself.
: G ! Map(V ); (g ) : V
!V :
(18.2)
Mostly, V will be either IRn or C= n , in whi h ase (g ) is resp. a real or a omplex n n
matrix. For to be a representation, it has to preserve the group stru ture of G
(g )(h) = (gh) ; (1) = idV :
(18.3)
We will generally restri t the gauge symmetries to Lie-groups for whi h one an write any
group element as an exponential of a Lie-algebra element
g exp(X ) ; X 2 LG :
(18.4)
This Lie-algebra has an asso iative, but non- ommutative, bilinear produ t
(X; Y ) 2 LG LG ! [X; Y 2 LG :
(18.5)
The Campbell-Baker-Hausdor formula expresses that the logarithm of exp(X ) exp(Y )

is an element of the Lie algebra, i.e. the produ t of two exponentials is again an exponential.
F (X; Y ) log (exp(X ) exp(Y )) = X + Y + 12 [X; Y + 121 [X; [X; Y + 121 [Y; [Y; X + 2 LG :
(18.6)
This formula will be of great help in nding a simple riterion for to be a representation,
satisfying eq. (18.3). Apart from the group stru ture of Map(V ), it also has a Lie-algebra
stru ture (the ommutator of two n n matri es is again a n n matrix). The representations
of the group an be easily restri ted to the Lie-algebra.
: LG ! Map(V ) ;
(18.7)
in a way that preserves the Lie-algebra stru ture
([X; Y ) = [(X ); (Y ) = (X )(Y ) (Y )(X )
(18.8)
It is more or less by onstru tion that we require
(exp(X )) = exp((X )) ;
(18.9)
where on the left-hand side is the group representation and on the right-hand side it is
the Lie-algebra representation. Without ausing too mu h onfusion we an use the same
symbol for the two obje ts. As a Lie-algebra forms a linear ve tor spa e we an dene a
basis on LG .
n
X
(18.10)
Z = za T a 2 LG ; za 2 IR (or C= ) ; T a 2 LG :
a=1
In here n is the dimension of the Lie-algebra (and the Lie-group if, as we will assume
throughout, the exponential is lo ally an invertible mapping). The ommutator, also alled
Lie-produ t, is ompletely determined by the stru ture onstants fab
[T a ; T b =
83
fab T :
(18.11)
Using the Ja obi identity

[X; [Y; Z + [Y; [Z; X + [Z; [X; Y = 0 ;
(18.12)
applied to X = T a , Y = T b and Z = T , we nd (from now on sums over repeated group

indi es are impli it)
fb d fade + f ad fbde + fabd f de = 0 :
(18.13)
This pre isely oin ides with the ommutation relations of the so- alled adjoint representation
(Tada )b ad (T a )b = fab :
(18.14)
Indeed, one easily veries that
[ad (T a ); ad (T b ) = fab ad (T ) :
(18.15)
In general, sin e a representation preserves the ommutation relations, it also preserves

the stru ture onstants in terms of (T a ) Ta , whi h forms a basis for the linear representation spa e whi h is ontained in V . They are alled the generators of the representation.
With the help of eq. (18.6) we easily verify that is a representation if and only if
[Ta ; Tb = fab T :
(18.16)
This is be ause under the a tion of one simply repla es T a by Ta

exp(xa T a ) = (exp(X )) = exp((X )) = exp xa Ta
(18.17)
Similarly, the Campbell-Baker-Hausdor formula, when expressed with respe t to the Liealgebra basis fT a g

exp xa T a exp yb T b = exp fxa + ya + 21 xb y fab + 121 (xd xb y + yd yb x )fb efdea + gT a :
(18.18)
dire tly determines the multipli ation of the representation of group elements by repla ing T a
by Ta , provided eq. (18.16) is satised. Note that the stru ture onstants are antisymmetri
with respe t to the rst two indi es. They are also invariant under y li permutations of
the indi es. This follows from the y li property of the tra e
fabd Tr(Td T ) = Tr([Ta ; Tb T ) = Tr(Ta [Tb ; T ) = fb d Tr(Ta Td ) ;
(18.19)
and from the fa t that for ompa t groups the generators an be normalised su h that
a Tb ) =
Tr(Tfnd
fnd
1
2
ab ;
(18.20)
a are the generators of the so- alled fundamental or dening representation of the
where Tfnd
group G. This matrix representation is usually identied with the group (or algebra) itself,
whi h till now was seen more as an abstra t entity. The simplest example is SU(2), the set
of omplex unitary 2 2 matri es with unit determinant. Its fundamental representation
oin ides with the spinor or spin one-half representation. The stru ture onstants and the
generators of the fundamental and adjoint representations were onsidered in se t. 12 (see
eq. (12.9))
a = i a ; f = " ; (T a ) = La :
Tfnd
(18.21)
ab
ab
ad
2
84
Be ause the Campbell-Baker-Hausdor formula plays su h a ru ial role in the theory

and in the pra ti al implementation of group representations, we will now provide a more
abstra t derivation of eq. (18.6) to all orders. The proof simply states how in the Taylor
expansion produ ts of Lie-algebra elements are regrouped in multiple ommutators. A ru ial
ingredient for deriving the Campbell-Baker-Hausdor is the so- alled derivation D, that maps
a produ t of Lie-algebra elements into a multiple ommutator.
DX = X ; DXi Xi Xis [Xi ; [Xi ; [Xis ; Xis ; s > 1 :

1
(18.22)
We also dene for these produ ts the adjoint map, ad, introdu ed in eq. (12.12)
adXi1 Xi2 Xis adXi1 adXi2 adXis ;
(18.23)
whi h is easily seen to satisfy

ad([X; Y ) = [adX; adY :
(18.24)
It is more or less by denition that for any two produ ts u and v of Lie-algebra elements
D(uv) = aduDv :
(18.25)
For two Lie-algebra elements X and Y , it an easily be shown that
D[X; Y = D(XY ) D(Y X ) = adX DY
adY DX = [X; DY + [DX; Y
(18.26)
and this allows us to prove by indu tion that a monomial Q (a polynomial of whi h all terms
are of the same order) of degree m in terms of Lie-algebra elements Xi , i = 1; 2; ; s is
an element of the Lie-algebra (i.e. an be written as multiple ommutators, alled a Liemonomial) if and only if DQ = mQ. If this equation is satised it is lear from the denition
of a derivation that Q is a Lie-monomial. So it is su ient to prove that the equation is
satised for Q a Lie-monomial. In that ase Q is a sum of terms, ea h of whi h an be
written as ad(Xi1 )Q(1) with Q(1) a Lie-monomial of degree m 1. Using eq. (18.26) therefore
yields Dad(Xi1 )Q(1) = ad(Xi1 )DQ(1) +ad(Xi1 )Q(1) . Indu tion in m gives the required result.
Now it is trivial to regroup the terms in the Taylor expansion of eq. (18.6) in multiple
ommutators. From the fa t that any group element an be written as the exponent of a
Lie-algebra element, we know that F (X; Y ) 2 LG (at the worst one needs to restri t X and
Y to su iently small neighbourhoods of the origin in LG ). Consequently, ea h term in the
Taylor expansion of F (X; Y ) of order m (denoted by Fm (X; Y ), being a monomial in X and
Y ) has to be an element of the Lie-algebra and satises
X
1
F (X; Y ) Fm (X; Y ) ; Fm (X; Y ) = DFm (X; Y ) :
(18.27)
m
m
It is not di ult to work out the Taylor expansion for F (X; Y )
k
iY j
X ( 1)k 1
X X iY j
X
A =

A
F (X; Y ) log (exp(X ) exp(Y )) = log 1 +
;
k
i+j>0 i!j !
i+j>0 i!j !
k
(18.28)
from whi h we easily obtain the expli it expression for the Campbell-Baker-Hausdor formula
in terms of multiple ommutators,
X
X
( 1)k 1 D (X p1 Y q1 X p2 Y q2 X pk Y qk )
: (18.29)
F (X; Y ) =
p1 !q1 !p2 !q2 ! pk !qk !
m fk;Pk p +q =m; p +q >0g km
j j
j j
0
j =1
85
We leave it to the industrious student to verify that

n
o
F1 (X; Y ) = X + Y; F2 (X; Y ) = 21 [X; Y ; F3 (X; Y ) = 121 (adX )2 Y + (adY )2 X ;
F4 (X; Y ) = 241 adX adY adX (Y );
n
o
n
o
F5 (X; Y ) = 7201 (adX )4 Y + (adY )4 X + 3601 adX (adY )3 X + adY (adX )3 Y
n
2 Y + adY adX (adY )3 X o : (18.30)
1
ad
X
ad
Y
(ad
X
)
120
After this intermezzo we return to the issue of onstru ting non-Abelian gauge theories.
The simplest way is by generalising rst the ovariant derivative. U(1) gauge transformations
a t on a omplex eld as in eq. (18.1), and the ovariant derivative is designed su h that
D (x) ! g (x)D (x) :
(18.31)
Sin e the gauge eld transforms as in eq. (17.3), this is easily seen to imply that the ovariant
derivative is dened as in eq. (17.2) (these formula are of ourse also valid for omplex s alar
elds, ompare eq. (3.36)). For a non-Abelian gauge theory we onsider rst a eld that
transforms as an irredu ible representation (i.e. only the origin is left invariant under the
a tion of all gauge group elements)
! g (g ) :
(18.32)
In the following, as in the literature, we shall no longer make a distin tion between g and
(g ). It will always be lear from the ontext what is intended. The ve tor potential should
now be an element of the Lie-algebra LG (more pre isely a representation thereof)
A = Aa T a :
(18.33)
For U(1), whi h is one-dimensional, we need to dene T 1 i. The Lie-algebra of the group

onsisting of the unimodular omplex numbers is the set of imaginary numbers, LU(1) = iIR.
Note that as an ex eption this generator is normalised dierent from eq. (18.20), so as not
to introdu e un onventional normalisations elsewhere. The real valued ve tor potential A
will now be denoted by A1 and we see that under a gauge transformation
A ! g A = gA g 1
q 1 ( g )g 1 = gA g 1 + q 1 g (g 1 ) :
(18.34)
This is the form that generalises dire tly to the non-Abelian gauge groups with the ovariant
derivative dened by
D = ( + qA ) ;
(18.35)
where A Aa Ta is a matrix a ting on the elds . We leave it as an exer ise to verify
that under a gauge transformation, eq. (18.31) remains valid for the non-Abelian ase.
It is now a trivial matter to onstru t a Lagrangian that is invariant under lo al gauge
transformation. For a s alar eld one has
L = (D )y D m2 y ;
(18.36)
whereas if is a Dira eld, arrying both spinor (representation of the Lorentz group) and
group indi es, one has
(18.37)
L = (i D m) ; = y 0 ;
86
where y is the hermitian onjugate both with respe t to the spinor and the group (representation) indi es.
The part of the Lagrangian that des ribes the self intera tions of the ve tor eld A
has to be invariant under lo al gauge transformations too. In that respe t U(1) or Abelian
gauge theories are spe ial, sin e the homogeneous part of the transformation of the ve tor
potential is trivial, gA g 1 = A . For non-Abelian gauge transformations this is no longer
true. For U(1) one easily veries that
D D D D [D ; D = iqF 1 ;
(18.38)

1 = A1 A1 is the ele tromagneti eld strength, ompare eq. (3.27). Be ause

where F

the ovariant derivative transforms in a simple way under gauge transformations, this formula
an be dire tly generalised to non-Abelian gauge theories
g
F q 1 [D ; D !
gF g 1 :
(18.39)
For U(1), where g is a number, this means that the eld strength is gauge invariant, as was
noted before. For non-Abelian gauge theories the eld strength itself is not gauge invariant.
Nevertheless, it is simple to onstru t a gauge invariant a tion for the gauge eld
LA =
Tr (F F )
a F ;
F
(18.40)
a
a are the omponents of the eld strength with respe t to the Lie-algebra basis,
where F
1
2
1
4
a T a = Aa Aa + qf Ab A T a :
F F
(18.41)

ab
We see from LA and L that q plays the role of an expansion parameter. For q = 0 we
have n = dim(G) non-intera ting photon elds. They ouple with strength q to the s alar
or Dira elds. For non-Abelian gauge theories, in addition the ve tor eld ouples to itself.
These self-intera tions guarantee that there is invarian e under the gauge group G, whi h
is mu h bigger than U(1)n, whi h is the symmetry that seems implied by the q = 0 limit.
The non-Abelian gauge invarian e xes the \ harges" of the elds with respe t to ea h of
these U(1) gauge fa tors. Without the non-Abelian gauge symmetry there would have been
n independent \ harges".
The Lagrangian LA is the one that was dis overed in 1954 by C.N. Yang and R.L. Mills.
The Euler-Lagrange equations for the Lagrangian LA are alled the Yang-Mills equations.
One easily shows that
Fa + qfab Ab F = 0 or [D ; F F + q [A ; F = 0 :

(18.42)
For the oupling to fermions we read o from eq. (18.37) what the urrent for the Yang-Mills
eld is
L = (i D m) = (i m) + iqAa T a :
(18.43)
The urrent is therefore given by
(18.44)
Ja iq T a :
The oupled Yang-Mills equations read
Fa + qfab Ab F = Ja :
(18.45)
In problem 31 it will be shown that the urrent is not gauge invariant, unlike for Abelian
gauge symmetries. Closely related is the fa t that it is no longer true that the urrent is
onserved, i.e. Ja 6= 0. Instead, it will be shown in problem 31 that Ja + qfab Ab J = 0.
87
19 The Higgs me hanism

We have seen in problem 30 that the four-Fermi intera tion in good approximation an be
written in terms of the ex hange of a heavy ve tor parti le. In lowest order we have resp.
the diagrams in g. 12a and g. 12b
R

R
I

(a)

I

g. 12
(b)
The rst diagram omes from a four fermion intera tion term that an be written in terms
of the produ t of two urrents J J , where J = . Here ea h fermion line typi ally
arries its own avour index, whi h was suppressed for simpli ity. Fig. 12b an be seen to
ee tively orrespond to
!
g k k =M 2 ~

~
J (k )
(19.1)
J ( k) 2
k M 2 + i"
At values of the ex hanged momentum k2 M 2 , one will not see a dieren e between
these two pro esses, provided the oupling onstant for the four-Fermi intera tions (g. 12a)
is hosen suitably (see problem 30). This is be ause for small k2 , the propagator an be
repla ed by g =M 2 , whi h indeed onverts eq. (19.1) to J J =M 2 . It shows that the fourFermi oupling onstant is proportional to M 2 , su h that its weakness is explained by the
heavy mass of the ve tor parti le that mediates the intera tions. Examples of four-Fermi
intera tions o ur in the theory of -de ay, for example the de ay of a neutron into a proton,
an ele tron and an anti-neutrino. In that ase the urrent also ontains a 5 (problem 40).
It turns out that the four-Fermi theory annot be renormalised. Its quantum orre tions
give rise to an innite number of divergent terms, that an not be reabsorbed in a redenition
of a Lagrangian with a nite number of intera tions. With the intera tion resolved at higher
energies by the ex hange of a massive ve tor parti le, the situation is onsiderably better.
But it be omes ru ial for the urrents in question to be onserved, su h that the k k
part in the propagator has no ee t. It would give rise to violations of unitarity in the
s attering matrix at high energies (the eld dened in eq. (16.15) has the wrong sign for
its kineti part). To enfor e urrent onservation, we typi ally use gauge invarian e. But
gauge invarian e would prote t the ve tor parti le from having a mass. The big puzzle
therefore was how to des ribe a massive ve tor parti le that is nevertheless asso iated to the
ve tor potential of a gauge eld.
The answer an be found in the theory of super ondu tivity, whi h prevents magneti
eld lines to penetrate in a super ondu ting sample. If there is, however, penetration in the
form of a quantised ux tube, the magneti eld de ays exponentially outside the ux tube.
This would indi ate a mass term for the ele tromagneti eld within the super ondu tor.
The Landau-Ginzburg theory that gives an ee tive des ription of this phenomenon (the
mi ros opi des ription being given by the BCS theory of Cooper pairs) pre isely oin ides
with s alar quantum ele trodynami s.
L = 1 F F + (D ') D' '' 1 ('')2 :
(19.2)
4

In the Landau-Ginzburg theory ' des ribes the Cooper pairs. It is also alled the order
parameter of the BCS theory. In usual s alar quantum ele trodynami s we would put =
m2 , where m is the mass of the harged s alar eld. But in the Landau-Ginzburg theory of
88
super ondu tivity, it happens to be the ase that is negative. In that ase the potential
V (') for the s alar eld has the shape of a Mexi an hat.
V
g. 13
g. 13
Re '
Im '
The minimum of the potential is no longer at ' = 0, but at ' ' = 2=, and is
independent of the phase of '. To nd the physi al ex itations of this theory we have to
expand around the minimum. With a global phase rotation we an hose the point to expand
around to be real,
q
2= :
(19.3)
'0 =
But this immediately implies that the terms quadrati in the gauge eld give rise to a mass
term for the photon eld
q
jD'0 j2 = e2 '20A A M 2 A A ; M = 2e
1
2
= :
(19.4)
Furthermore, from the degenera y of the minimum of the potential it follows that the u tuation along that minimum (the phase in ' = '0 exp(i)) has no mass (this is related to
the famous Goldstone theorem, whi h states that if hoosing a minimum of the potential
would break the symmetry, alled spontaneous symmetry breaking, there is always a massless parti le). However, this phase is pre isely related to the gauge invarian e, and an
be rotated away by a gauge transformation. On the one hand orresponds to a massless
ex itation, on the other hand it is the unphysi al longitudinal omponent of the gauge eld.
But the photon be ame massive, and has to develop an additional physi al polarisation,
whi h is pre isely the longitudinal omponent. In a prosai way one states that the massless
ex itation ( alled a would-be Goldstone boson) was \eaten" by the longitudinal omponent
of the photon, whi h in the pro ess got a mass (\got fat").
This means we have four massive degrees of freedom, three for the massive ve tor parti le
and one for the absolute value of the omplex s alar eld (its mass is determined by the
quadrati part of the potential in the radial dire tion at ' = '0 ). This is exa tly the same
number as for ordinary s alar ele trodynami s where > 0, be ause in that ase the massless
photon has only two degrees of freedom, whereas the omplex s alar eld represents two
massive real s alar elds. It looks, however, that there is a dis ontinuity in the des ription
of these degrees of freedom, when approa hing = 0. But the interpretation of the phase
of the omplex eld as a longitudinal omponent of the ve tor eld is simply a matter of
hoosing a parti ular gauge. To ount the number of degrees of freedom we impli itly made
two dierent gauge hoi es
>0:
<0:
A = 0;
Im ' = 0;
Lorentz gauge ;
Unitary gauge :
89
(19.5)
There is a gauge, alled the 't Hooft gauge, that interpolates between these two gauges
F A
2ie'0 Im ' = 0;
't Hooft gauge :
(19.6)
Rather than adding to the Lagrangian the gauge xing term Lgf = 12 ( A )2 , one adds
Lgf = 12 F 2. At = 0 this orresponds to the Lorentz gauge, at = 1 to the unitary
gauge. For the hoi e 't Hooft made ( = 1=) the terms that mix (' '0 ) and A at
quadrati order disappear and one easily reads of the masses. Gauge xing will be dis ussed
in the next se tion, where it will be shown how extra unphysi al degrees of freedom appear in
the path integral, so as to an el the unphysi al omponents of the gauge and s alar elds.
The s alar eld, whose intera tions give the gauge eld a mass, is alled the Higgs eld.
Problems 34 and 35 dis uss the Higgs me hanism in detail for the Georgi-Glashow model,
whi h is a non-Abelian gauge theory with gauge group SO(3), oupled to an SO(3)-ve tor
of s alar elds 'a .
20 Gauge xing and ghosts

The quantisation of gauge theories in the path integral formalism requires more dis ussion,
sin e the gauge ondition (like the Lorentz gauge A = 0) seems to remove only one degree
of freedom of the two that are eliminated in the Hamiltonian formulation (see se tion 16).
From a simple example it is easily demonstrated what the ee t of gauge xing on a (path)
integral is. For this we take f (~x) to be a fun tion on IR3 , whi h is invariant under rotations
around the origin, su h that it is a fun tion f (r) of the radius r =R j~xj only. The symmetry
group is hen e SO(3) and we an attempt to ompute the integral d3~x f (~x) by introdu ing
a \gauge" xing ondition like x2 = x3 = 0. But it is lear that
Z
d3~x f (~x) 6=
d3~x (x2 ) (x3 )f (~x) =
dx1 f (x1 )
(20.1)
We know very well that we need a Ja obian fa tor for the radial integral
Z
Z 1
d3~x f (~x) = 4
r2 dr f (r) :
(20.2)
0
This Ja obian, arising in the hange of variable to the invariant radial oordinates and
the angular oordinates, an be properly in orporated following the method introdu ed by
Faddeev
and Popov. The starting point is a straightforward generalisation of the identity
R
dx jf 0 (x)j (f (x)) = 1, assuming the equation f (x) = 0 to have pre isely one solution (in a
sense the right-hand side of the equation ounts the number of zeros). It reads
1=
Dg j det (M (g A)) j (F (g A)) ;
(20.3)
where F (A) 2 LG is the gauge xing fun tion (with the gauge ondition F (A) = 0, e.g.
F (A) = A = 0). The gauge transformation g of the gauge eld A is indi ated by g A, see
eq. (18.34). Furthermore, M (A) : LG ! LG plays the role of the Ja obian,
F (g A) F (eXg A)
=
X
at X = 0 :
(20.4)
g
Equivalently, with respe t to the Lie-algebra basis, where F (A) Fa (A)T a , one has
dF (exp(tT b ) A)
at t = 0 :
(20.5)
Mab (A) = a
dt
M (g A)
90
To relate this to the previous equation, one makes use of the fa t that
h (g A) = (hg) A ;
(20.6)
whi h states that two su essive gauge transformations, g and h, give the same result as a
single gauge transformation with hg .
As an example we onsider the Lorentz gauge, with F (A) = A , for whi h
F (exp(X ) A) F (A) = q 1 Dad (A)(X ) + O(X 2) ;
(20.7)

where Dad
(A) is the ovariant derivative in the adjoint representation

Dad
(A)(X ) X + q [A ; X :
(20.8)
With respe t to the Lie-algebra basis this gives
qMab (A) = ab + qfab ( A + A ) :
(20.9)
For an Abelian gauge theory the stru ture onstants fab vanish and M (A) be omes independent of the gauge eld. This means that det(M (A)) an be absorbed in an overall
normalisation of the path-integral. For non-Abelian gauge theories this is no longer possible.
Before des ribing how the A dependen e of det(M (A)) is in orporated, it is important to
note that we assumed the gauge ondition F (g A) = 0 to have pre isely one solution, whi h
an be arranged with the help of eq. (20.6) to o ur at g = 1, in whi h ase A is said to
satisfy the gauge ondition. This is in general not orre t, as was dis overed by Gribov.
Even in our simple problem on IR3 , the gauge ondition x2 = x3 = 0 does not uniquely
spe ify the gauge, be ause we an go from (r; 0; 0) to ( r; 0; 0) through a rotation over 180
degrees. We have to introdu e a further restri tion to get the identity
Z
Z
d3~x f (~x) = 4 d3~x x21 (x2 ) (x3 )(x1 )f (~x) ;
(20.10)
where (x) = 0 for x < 0 and (x) = 1 for x 0. In perturbation theory only the gauge
elds near the origin in eld spa e are relevant, and gauge onditions are hosen so as to
avoid this problem in a small neighbourhood of the origin. The Lorentz gauge is su h a
gauge ondition, and the gauge xing or Gribov ambiguity is not an issue for omputing
quantities in perturbation theory in q .
We still
need to dene what we mean with Dg in eq. (20.3). It stands for the integration
Q
measure x dg (x), with dg (x) for every x dened as the so- alled Haar-measure on the group.
It is best des ribed in the example of SU(2), whi h as a spa e is isomorphi with S 3 . When
S 3 is embedded in IR4 as a unit sphere, n2 = 1, it is not too di ult to see that g = n4 + ik nk
gives an element of SU(2), whereas exp(isk k ) = os() + i sin()sk k , with s2k = 1, shows
that any element of SU(2) an be written Rin terms of n . The Haar-measure oin ides with
the standard integration measure on S 3 , d4 n (n2 1). The Haar-measure is in general
invariant under the hange of variables g ! hg and g ! gh, for h some xed group element.
We an insert eq. (20.3) in the path integral to obtain
Z=
D A D g
det (M (g A)) (F (g A)) exp (iS (A)) :
(20.11)
We now use that the a tion S (A) is invariant under gauge transformations. We leave it as
an exer ise to verify that likewise DA is invariant under the hange of variables A ! g A,
whi h trivially implies that
Z=
D A D g
det (M (A)) (F (A)) exp (iS (A)) :

91
(20.12)
The dependen e of the integrand on g has disappeared, and the integration over g gives an
overall (innite) normalisation fa tor, whi h is irrelevant. We next note that Z has to be
independent of the gauge xing fun tion F , in parti ular F (A) Y is just as good for the
gauge xing (provided of ourse we show that F (g A) = Y has a solution). This modi ation
does not ae t the so- alled Faddeev-Popov operator M (A) and we nd
Z
DA det (M (A)) (F (A) Y ) exp (iS (A)) ;

independent of Y . Suitably normalising DY we an dene
Z=
(20.13)

i Z
2
DY exp 2 d4x Ya (x) = 1 ;
whi h ombined with the previous equation gives

Z =
Z
DY DA det (M (A)) (F (A) Y ) exp i d4x L(A)
(20.14)
2
Y (x)
2 a
2
F (A) :
(20.15)
= DA det (M (A)) exp i d4 x L(A)
2 a
For U(1) gauge theories with F (A) = A this pre isely reprodu es the a tion of eq. (4.22)
in the Lorentz gauge, and in that ase det(M (A)) is a onstant.
For non-Abelian gauge theories we are left with the task of omputing det(M (A)) for
ea h A, whi h is no longer onstant. But here the path integral over Grassmann variables
omes to the res ue. In problem 25 we have seen that
Z
DD
Z
exp i d4 x
a (x)Mab (A) b (x)
= det(M (A)) ;
(20.16)
up to an overall normalisation. This implies that the path integral an also be written in
the Lorentz gauge as
Z

Z

2

a
a
b

Z = DA DD exp i d4 x L(A)
( A (x)) + (x)[ (x) + qfab A (x) (x) :
2 a
(20.17)
Sin e and are auxiliary elds, they should never appear as external lines. They are
therefore alled ghosts. Ghosts an only appear in loops and be ause of the fermioni nature
of the ghost variables, every su h loop gives a minus sign. The Feynman rules for the Lorentz
gauge are given in the following table.
no external ghost lines
-k
- -6
b
; k3
b; k2

a; k1
(A=0)ab = k2+abi"
table 7
ghost propagator (Lorentz gauge)
iqfab k1
ghost vertex (Lorentz gauge)
1 i (2d4 k)4
R
loop fa tor
92
Be ause one an easily derive that for a omplex s alar eld (up to an overall onstant)
Z

Z
D'D' exp i d4x 'a(x)M ab (A)'b(x) = det(M1 (A)) ;
(20.18)
we an view a ghost as the elimination of a omplex degree of freedom. It is in this way that
in the path integral the two unphysi al degrees of freedom of a Lorentz ve tor are eliminated.
For QED both the ghost and the unphysi al degrees of freedom have no intera tions, and an
not appear as external lines either, whi h is why in QED the introdu tion of ghosts was never
ne essary for a onsistent des ription of the theory. For non-Abelian gauge theories, be ause
of the intera tion of the ghost with the gauge eld, ghosts an no longer be ignored. To
have the ghosts eliminate the unphysi al degrees of freedom, one should have the \masses"
(poles) of the ghosts to oin ide with the \masses" of the unphysi al degrees of freedom.
Furthermore the ouplings of the ghost and unphysi al elds to the physi al elds should be
related. This is veried expli itly for the Georgi-Glashow model in problem 35. In general it
is guaranteed by the existen e of an extra symmetry, dis overed by Be hi, Rouet and Stora,
alled the BRS symmetry s, whi h for example a ts on the gauge eld as follows

sA = Dad
:
(20.19)
This is pre isely an innitesimal gauge transformation. For more details see Itzykson and
Zuber, se tion 12.4-1.
21 The Standard Model

The standard model des ribes the ele tromagneti and weak intera tions, unied in the so alled ele tro-weak theory of Glashow-Weinberg-Salam with the gauge group U(1)SU(2)
and the strong intera tions, known as Quantum Chromodynami s (QCD) with gauge group
SU(3). Theory and experiment, where tested, agree very well up to about 100 GeV, the
energies rea hed by presentday a elerators. Now the top quark has been found, at a mass
of 174 GeV, only the Higgs parti le remains to be dete ted. Its mass should be smaller
than 1000 GeV (i.e. 1 TeV=Terra ele tronvolt) a ording to presentday theoreti al insight.
Gravitation has been left out so far. Its naturalqs ale in energy where quantum ee ts would
be ome important is the Plan k energy, Epl = h 5 =G 1019 GeV. It is very well possible
that a number of the fundamental parameters in the standard model will be determined,
either dire tly or indire tly, by gravitational intera tions. The standard model should then
be onsidered as an ee tive eld theory. The theory for whi h the standard model des ribes
its ee tive low-energy behaviour is alled a unied theory. An intermediate stage, whi h
does not yet in lude gravity is a so- alled Grand Unied Theory (GUT). The simplest version unies the ele tro-weak and strong intera tions using a gauge group SU(5) (whi h has
U(1)SU(2)SU(3) as a subgroup), thereby redu ing the number of free parameters onsiderably. These GUT's predi t proton de ay, albeit at the tremendously low rate of one de ay
in every 1030 31 years. Nevertheless, a swimming pool of (10 m)3 ontains enough protons to
verify that the proton de ay is slower than an be omfortably a ommodated by GUT's.
Candidates that unify the standard model with gravity in the form of string theories and
supergravity have been unable to provide predi tions that either rule them out experimentally or provide eviden e in favour of these theories. Mu h is therefore still to be dis overed,
in parti ular be ause theoreti al insight of the last ten years has shown that a Higgs eld is
most likely not fundamental, although it is not yet ruled out that it will show its stru ture
93
only at Plan k energies. If that is the ase the mass of the Higgs should, however, not be
mu h bigger than 100 GeV.
The standard model onsists of gauge elds B (for U(1)), Wa (for SU(2)) and Aa (for
SU(3), where a runs from 1 to 8, to be dis ussed later). These gauge elds have intera tions
with a Higgs eld 2 C= 2 , whi h transforms under SU(2) as a spin one-half representation
(i.e. the fundamental representation) with a oupling onstant g . Under U(1) this Higgs
eld transforms with a oupling onstant 12 g 0 , whereas it is neutral under SU(3). These
ouplings are represented in the ovariant derivative
3
i 0
ig X
D =
W a :
(21.1)
g B
2
2 a=1 a
The potential for the Higgs eld auses spontaneous breaking of part of the symmetries

(21.2)
V () = (y F 2 )2 = y + (y )2 + onst. ;
4
4
where 12 F 2 . In this ase the minimum of the potential, also alled the va uum, is
degenerate on a three dimensional sphere, spe ied by y = F 2 ( 2 C= 2 IR4 ), whi h
would give rise to three massless s alar parti les a ording to the Goldstone theorem, but
all three will be \eaten" by longitudinal omponents of the gauge elds to whi h the Higgs
eld ouples. We note that there are four gauge eld omponents, B and Wa for a = 1; 2
and 3. Indeed one ombination among these four will not have something to \eat" and will
therefore stay massless. It plays the role of the photon eld as we got to know it in QED.
To see this write

1 + '1
= F0 + '
;
(21.3)
0
'2
'2
su h that
g2F 2 1 2
F2
[(W ) + W2 )2 + [g 0 B gW3 2 :
(21.4)
4
4
p
Apparently, the ve tor elds W1;2 will have a mass MW = 21 2gF , whereas the linear
ombination
g 0B gW3
g0
3
Z = q
;
(21.5)
=
sin

B
os

W
;
tan

=
W
W
W
g
g2 + g02
p
1
re eives a mass mZ = 12 2F (g 2 + g 02 ) 2 = MW = os W . The linear ombination perpendi ular
to Z ,
3
Aem
(21.6)
= os W B + sin W W ;
remains massless. This gauge eld denes a U(1) subgroup of SU(2)U(1) that leaves
0 invariant. This U(1) subgroup is a ombination of the U(1) subgroup of SU(2) generated by exp(i3 ) and the phase rotations exp(i) asso iated with the expli it U(1) group
with B as its gauge eld. It is trivial to verify that the produ t of these group elements,
exp(i) exp(i3 ) indeed leaves 0 invariant. The gauge symmetry asso iated to this so alled diagonal U(1) subgroup therefore remains unbroken and orresponds to ele tromagnetism.
The Higgs eld has three massless omponents Re '1 , Im '1 and Im'2 , all eaten by the
ve tor parti les W and Z , and one massive omponent Re '2 with a mass
(D 0 )yD 0 =
m =
2 = F :
94
(21.7)
It is this omponent that is alled the Higgs eld. It does not ouple to Aem
, be ause like
'0 , also is not ae ted by the transformation exp(i) exp(i3 ). Alternatively, this an be
seen from the ovariant derivative

ig
1 = n
1 + 2 W 2 ) + ig ( os2 W 3 sin2 W )Z
D '
(

W

1

'2
2
2 os W

o
ig
'
1
em
sin W (3 + 1)A
:
(21.8)
'2
2
Using the fa t that

2
0
3 + 1 = 0 0
(21.9)
it is lear that '2 has no ele tri harge, whereas '1 has a harge q = g sin W . As
these are would-be Goldstone
bosons, \eaten" by the ve tor elds, it will turn out that the
p
ombinations W = 21 2(W1 iW2 ) are harged with an ele tri harge of e, where

e = g sin W :
(21.10)
As a onsequen e, the two oupling onstants g and g 0 are determined by the ele tri harge e
and the so- alled weak mixing angle W , also alled the Weinberg angle. From experiment it
follows that sin2 W 0:23. The Z ve tor eld will remain neutral under the ele tromagneti
intera tions. To verify the harge assignment to the ve tor elds, we have to nd the oupling
of the various elds to Aem
. For this it is su ient to onsider the following part of the
Lagrangian

(21.11)
LW;B = 41 Wa Wa + g"ab WbW 2 41 ( B B )2 :
a )2 = F a F . After some algebra the above
with the obvious short-hand notations like (F
a
equation an be rewritten as

LW;B = 41 Fem + ie(W+W W+W ) 2

+W
+ W )2
1 Z

Z
+
ig
os

(
W
W

W

4
em
1 jD em W
D W
ig os W (Z W
Z W )j2 ;
(21.12)
2
where we have dened

Dem =

em = Aem
ieAem
; F

Aem
:

(21.13)
We immediately read-o that our harge assignments for Z and W have been orre t. Note
em
that the ve tor eld W has an extra magneti moment, be ause of its oupling to F
(21.14)
Lmagn.mom. = ieF em W +W ;

whi h is a dire t onsequen e of the spin of the ve tor eld (the magneti moment for the
Dira eld is dis ussed in problem 32).
We now introdu e the fermions in the standard model. They are arranged a ording to
families. The rst family with the smallest masses onsists of the ele tron, the neutrino,
the up and the down quark. Essential in the standard model is that invarian e under parity
is broken expli itly by the weak intera tions (as has been observed in the beta de ay of
95
Cobalt-60, see problem 40). This is a hieved by oupling the left- and right-handed heli ity
eigenstates of the fermions dierently to the gauge elds. It should be stressed that the
standard model does not explain why parity is violated; it was put in \by hand". For ea h
fermion we dene
(21.15)
L = 12 (1 + 5 ) ; R = 21 (1 5 ) :
The Dira Lagrangian in terms of these heli ity eigenstates an be written as
L = (i m) = R (i ) R + L (i ) L m( R L + L R ) ;
(21.16)
su h that dierent transformation rules for R;L enfor e m = 0, i.e. the absen e of an
expli it mass term. The beauty of the Higgs me hanism is that it also provides a mass for
the fermions. This is a hieved by oupling the s alar eld to the fermions, using a Yukawa
oupling
(21.17)
L ; = y( R y L + L R) ;
where y is the Yukawa oupling onstant. It also immediately xes the representation to
whi h R;L should belong. Sin e the Lagrangian has to be invariant with respe t to the
gauge symmetries, and sin e the s alar eld is in the fundamental representation of SU(2),
we require that L is also in the fundamental representation, i.e. it is a doublet. On the other
hand R is taken to be invariant under SU(2) (also alled the singlet representation). The
ouplings of the fermions to the gauge eld B have to be hosen su h that the Lagrangian
is neutral. This oupling is parametrised by the so- alled hyper harge Y , in units of 12 g 0 .
YH Y () = 1 ; YR = YL
1 ; YR;L Y ( R;L ) :
(21.18)
The mass of the fermions is now read-o from eq. (21.17) by repla ing with its so- alled
va uum expe tation value 0
L ; = yF ( R L2 + L2 R) ;
(21.19)
where the index on L indi ates the so- alled isospin index, the spinor index of the two
dimensional fundamental representation for the internal SU(2) symmetry group. We also
see that L1 remains massless and this is exa tly the neutrino. The ele tron is identied
with the pair ( R ; L2 ) and has a mass me = yF . We want the neutrino to have no ele tri
harge and this xes the hyper harge of L . It is most easily determined from the ovariant
derivative, a ting on the left-handed fermion, dened as in eqs. (21.1) and (21.8), sin e both
are in the same representation (ele tron and neutrino are also neutral with respe t to the
strong intera tions. The situation for the quarks will be dis ussed below).
D L =

ig
ig
( os2 W 3
(1 W1 + 2 W2 ) +
2
2 os W
o
ig sin W
L :
(3 + YL)Aem

2
YL sin2 W )Z
(21.20)
To make L1 de ouple from the ele tromagneti eld we require
YL = 1 ; YR = 2 :
(21.21)
This also allows us to nd the ele tri harge of L2 to be g sin W = e, whi h as it should
be, oin ides with the ele tron harge. The right-handed omponent should of ourse have
96
the same ele tri harge. In that ase the ovariant derivative is given by
D R = (
= (
ig 0
ig 0
YR B ) R = (
YR [sin W Z + os W Aem
) R

2
2
R ;
ieAem
ie
tan

Z
)
(21.22)
W

with the expe ted oupling to the ele tromagneti eld. Note that we an summarise our
assignments of the ele tri harge by introdu ing the harge operator in terms of the hyper harge and the so- alled isospin operator I3
Qem = ( 1 Y + I3 )e :
(21.23)
2
On a doublet ( L and ') one has I3 = 21 3 , whereas I3 = 0 on a singlet ( R ).

We now dis uss quarks. Also there the weak intera tions a t dierently on the left- and
right-handed omponents. The left-handed up and down quarks are ombined in a doublet
representation for SU(2). If we denote the quark elds by q , we assign q1L to the left-handed
omponent of the up-quark (also denoted by uL ) and q2L to the left-handed omponent of
the down-quark (dL ). This doublet gets a hyper harge Y (q L ) = 31 , from whi h we read o
the ele tri harges
2e L
L
u
em
L
em
3eu L
:
(21.24)
=
Q q =Q
L
d
3d
The right-handed omponents of both the up and down quarks are singlets under SU(2) and
their hyper harges are hosen to insure that they have the same ele tri harge as for their
left-handed partners
4
2
Y (uR ) =
and Y (dR ) =
:
(21.25)
3
3
The quarks transform non-trivially under SU(3), the gauge group of the strong intera tions. They form omplex ve tors in the three dimensional dening or fundamental representation of SU(3). The generators for SU(3) are given by the Gell-Mann matri es
0
i B0
1
T = 1
2 0
0
i 0
T4 = B
0
2 1
0
i B0
7
T = 0
2 0
1
0
0
0
0
0
0
0
i
0
i B0 i
C
2
0A; T = i 0
2 0 0
0
1
0
1
i 0 0
0C
0
T5 = B
A;
0
2
0
i 0
0
1
0
i B1
C
8
p
iA; T =
0
2 3 0
0
0
0C
T3 =
A;
0
1
i
0C
T6 =
A;
0
1
0 0
1 0 C
;
A
0 2
i B1 0 0C
1 0A;
0
2 0 0 0
1
0
0
0
0
iB
0 1C
A;
0
2 0 1 0
(21.26)
normalised in a ordan e with eq. (18.20). We leave it as an exer ise to determine the
stru ture onstants. Note that the Lie-algebra for the group SU(N) is given by tra eless
and antihermitian (X y = X ) omplex N N matri es. The dimension of this Lie-algebra
is easily seen to be N 2 1. Note that det(exp(X )) = exp(Tr(X )), su h that exp(X ) has
determinant one. Also, exp(X ) 1 = exp( X ) = exp(X y) = exp(X )y guarantees that exp(X )
is a unitary matrix.
The fra tional ele tri harge of the quarks is not observable (otherwise we would have
had a dierent unit for ele tri harge). The reason is that quarks are onje tured to always
form bound states that are neutral under SU(3). This an be a hieved by either taking
97
three quarks in an antisymmetri ombination to form a SU(3) singlet, or by ombining a

quark and an antiquark. In the rst ase one has a baryon, of whi h the proton (uud) and
the neutron (udd) are examples. The quark-antiquark bound state is alled a meson, of
whi h the pions are examples (e.g. + = ud and = ud). The bar over the symbol of
a parti le of ourse denotes the antiparti le. Rather prosai ally one asso iates to the three
SU(3) omponents of the quark eld the property olour. Choosing the three basi olours
red, blue and green makes a bound state of three quarks in an antisymmetri wave fun tion,
where hen e all olours are dierent, into a olourless omposite. Similarly, ombing a quark
and an antiquark gives a olourless ombination. It is not too di ult to show that a bound
state of quarks and antiquarks is a singlet under SU(3) if and only if the net olour is white.
It is now also easily veried that with the parti ular fra tional ele tri harges assigned to
the quarks, a olourless ombination always has an ele tri harge that is an integer multiple
of the ele tron harge. For this note that both quarks have modulo e an ele tri harge
equal to 31 e, whereas both antiquarks have modulo e a harge of 13 e. Three quarks bound
together therefore have zero harge modulo e. The same holds for a quark-antiquark bound
state.
That the strong intera tions really are strong, follows from the fa t that a quark and
antiquark an not be separated without reating a quark-antiquark pair from the va uum,
to make sure that the separated omponents remain neutral under SU(3). This is a hieved
by ombining the quark (antiquark) of the pair reated with the antiquark (quark) we try
to separate. The me hanism that prevents free quarks to appear is alled onnement,
whi h still la ks a solid theoreti al understanding. Be ause the oupling onstant is strong,
a perturbative expansion is no longer appli able. That nevertheless the theory of the strong
intera tions is believed to be the orre t theory to des ribe the for es amongst the quarks
(and therefore indire tly the nu lear for es) follows from the remarkable property that at
high energies the ee tive oupling onstant is small; at innite energy even zero. This is
alled asymptoti freedom and will only brie y be dis ussed in the next se tion. For a more
detailed dis ussion we refer to Itzykson and Zuber. In the table below we list the gauge
parti les of the standard model.
Gauge parti les

A
W
Z
name
harge spin mass
photon
gluon
W-parti le
Z-parti le
0
0
e
0
1
1
1
1
for e
table 8
0
ele tromagnetism
0
strong for e
80 GeV
weak for e
91 GeV
weak for e
The strong intera tions do not break parity invarian e, i.e. the eight gluons Aa ouple to
the left-handed and right-handed omponents of the quark elds in the same way. However,
the so- alled Cabibbo mixing with two other families of quarks (the strange and harm 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 ombination of harge onjugation and parity (equivalent
to time reversal T , sin e CP T is onserved). The ele tron and neutrino, alled leptons,
in the rst family are repla ed by the muon and its neutrino for the se ond family and
by the tau and asso iated neutrino for the third family. The experiments des ribed in the
98
introdu tion (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 right-handed partner
for the neutrino eld, whi h ouples to none of the gauge elds. With a suitably hosen
Yukawa oupling the neutrino an 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 dis overed in 1994 at Fermilab). For mu h more on the standard model
see in parti ular the book by J.C. Taylor mentioned in the introdu tion.
Fermion families
table 9
name
harge spin mass
d
u
e
e
down quark
up quark
ele tron
neutrino
-e/3
2e/3
-e
0
1/2
1/2
1/2
1/2
10 MeV
5 MeV
511 keV
0(<10 eV)
s

strange quark -e/3

harm quark
2e/3
muon
-e
muon-neutrino 0
1/2
1/2
1/2
1/2
250 MeV
1,5 GeV
106 MeV
0(<0,5 MeV)
b
t

bottom quark
top quark
tau
tau-neutrino
1/2
1/2
1/2
1/2
4,8 GeV
174 GeV
1,8 GeV
0(<164 MeV)
-e/3
2e/3
-e
0
22 Loop orre tions and renormalisation

Up to now, we have only onsidered the lowest order al ulations of ross se tions, for whi h
it is su ient to onsider tree-level diagrams that do not ontain any loops. Loop integrals
typi ally give rise to innities, whi h an be regularised by onsidering for example a uto in
momentum spa e, as was dis ussed in se t. 7. Another possibility of regularising the theory is
by dis retising spa e-time, amounting to a latti e formulation, see eq. (7.5). In both of these
ases there exists a maximal energy (equivalent to a minimal distan e). The parameters,
like the oupling onstants, masses and eld renormalisation onstants will depend on this
uto parameter, generi ally denoted by an energy or a distan e a = 1=. How to give a
physi al denition of the mass in terms of the full propagator and why eld renormalisation
is ne essary was dis ussed in se t. 9. For the renormalisation of the oupling onstant it is
best to dene the physi al oupling onstant in terms of a parti ular s attering pro ess, as
that is what an be measured in experiment. Alternatively, as these are strongly related,
the physi al ouplings an be dened in terms of an amputated 1P I n-point fun tion, with
pres ribed momenta assigned to the external lines, all proportional to an energy s ale alled
. As an example onsider the self-intera ting s alar eld, with a four-point oupling
99
(see for example eq. (21.2)). We dene the physi al four-point oupling onstant in terms of
the 1P I four-point fun tion with the momenta on the amputated lines set to some parti ular
value proportional to (the pre ise hoi e is not important for the present dis ussion). It
is lear that this gives a fun tion reg(; ; ). The dependen e on other oupling onstants
and the mass parameters is left impli it.
The theory is onsidered renormalisable if we an remove the ut-o by adjusting
(also alled the bare oupling onstant) in su h a way that at a xed value = 0 the
renormalised oupling R stays nite and takes on a pres ribed (i.e. measured) value. It is
then obvious that the renormalised oupling onstant R () reg((); ; ) is a fun tion
of , oin iding at 0 with the pres ribed value R . Sin e the physi al oupling onstant is
omputed in terms of the full 1P I four-point fun tion, the dependen e on the energy s ale is
aused by quantum orre tions. Sin e the va uum in eld theories is not really empty, as was
dis ussed in the ontext of the Casimir ee t in se t. 2, the omputation is not mu h dierent
from al ulating ee tive intera tions in a polarised medium. In this ase the polarisation is
due to the virtual parti les that des ribe the quantum u tuations (zero-point u tuations)
of the va uum, and is therefore also alled the va uum polarisation. The energy dependent
ouplings are alled running ouplings. It should be emphasised that the running of the
ouplings is a manifestation of an anomaly ( alled the onformal anomaly), whi h is the
breaking of a symmetry of the Lagrangian by the quantum orre tions. In the absen e of
a mass, the s alar eld theory with a '4 intera tion is at the lassi al level invariant under
s ale transformations, '(x) ! '(x=), where is the s ale parameter. It is obvious that
the regularised ouplings are not invariant under su h a res aling, be ause of the presen e of
a uto. What is not obvious is that, for the simple eld theories we have been onsidering
in four dimensions, the s ale independen e an not be re overed by removing the uto (i.e.
taking ! 1).
By adjusting the bare oupling onstants of the theory as a fun tion of the uto ,
to ensure that all regularised ouplings stay nite when the uto is moved to innity, the
al ulations an be arranged su h that nowhere expli it innities o ur. When we say that
the ontributions of the loops diverge, we mean that without adjusting the bare oupling
onstant, their ontributions are innite in the limit ! 1. A theory is alled renormalisable if only a nite number of bare oupling onstants needs to be adjusted to have all 1P I
n-point fun tions nite. This an be shown to be equivalent to all 1P I n-point fun tions to
be ompletely determined as a fun tion of a nite number of renormalised ouplings, alled
relevant ouplings. It is only in su h an instan e that quantum eld theory has predi tive
power. Renormalisability is therefore a ne essary requirement for the theory to be insensitive
to what happens at very high energies with a maximal amount of predi tive power. The
standard model falls in this lass of theories.
Theoreti al studies of the last ve years or so have shown that the self- oupling of the
Higgs eld will most likely vanish if we really take ! 1, albeit in a logarithmi way.
Losely speaking the running of this oupling is su h that the renormalised oupling in reases
with in reasing energy. The only way it an be avoided that the renormalised oupling will
be ome innite at some nite energy, is to either take the renormalised oupling equal to
zero or to keep the uto nite. It depends on the parameters of the model, in parti ular
the Higgs mass, how large the uto should be. If the Higgs mass is relatively light, this an
be at the Plan k s ale and has little onsequen e for the theory. If, however, the Higgs mass
turns out to be in the order of 1 TeV, the uto has to be roughly smaller than 10 TeV.
As we an measure the self- oupling of the Higgs eld and related quantities to be nontrivial (whi h is of ourse ru ial for the spontaneous symmetry breaking and giving a mass
100
to the W and Z parti les), the s alar se tor of the standard model depends in a rather subtle
way on what happens at higher energies. This sensitivity to high energies is, however, mu h
and mu h weaker than in non-renormalisable theories like for the four-fermi intera tions.
It is outside the s ope of these le tures to des ribe the omputations ne essary to make
the above more pre ise. In the following we give a sample al ulation that will provide the
te hni al ingredients to perform su h al ulations 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 eld theories.
Let us end this dis ussion by noting that a running oupling an of ourse either in rease
or de rease at in reasing energy. The Higgs self- oupling and the ele tromagneti oupling
onstant e are examples of ouplings that in rease at high energies. For the ele tri harge e,
this in rease is very tiny and the uto an be hosen mu h bigger than the Plan k energy.
The analogy with a polarised medium is that the virtual parti les in the eld of a harged
parti le will s reen its harge at large distan es. When we probe the harged parti le at
ever smaller distan es the ee tive harge be omes less s reened and in reases. Due to the
self-intera tions of a non-Abelian gauge eld, its harges show the ee t of anti-s reening.
Here the ee tive harge be omes bigger at larger distan es. For the strong intera tions this
is one way to understand onnement. The energy of a single quark within a spheri al shell
would in rease without bound with in reasing radius. A free quark would arry an innite
energy. To the ontrary, at de reasing separations, the ee tive harge be omes weaker and
weaker and the quarks start to behave as free parti les. This is the asymptoti freedom
mentioned in the previous se tion.
We will now onsider to one-loop order the self-energy for the s alar eld ' with a mass
m, oupled to two avours of fermions with masses m1 and m2 , oupled through Yukawa
ouplings des ribed by the Lagrangian

X
L = 1 ( ')2 1 m2 '2 1 '3 + (i) (i mi ) (i) g' (1) (2) + (2) (1) : (22.1)
2
The self-energy for the s alar eld in one-loop order splits in two ontributions 1 and 2
from a fermion and a s alar loop (in this order Z 1).
' (q ) 1 + 2 =

m ; q+k
m
1; k

q + q
q+k
The numeri al expressions for 1 and 2 are given by

Tr ((k= + m1 )(k= + =q + m2 ))
ig 2 Z
d4 k 2
;
1 =
4
(2 )
(k m21 + i")((k + q )2 m22 + i")
i2 Z
1
2 =
d4 k 2
:
4
2
2(2 )
(k m + i")((k + q )2 m2 + i")
(22.2)
(22.3)
Using eq. (11.1), requires us to employ the Feynman rules of table 3 to obtain these expressions. (Alternatively the Feynman rules of table 2 an be used, provided the self-energy is
dened through eq. (9.9).) These integrals are obviously divergent. Introdu ing a momentum uto we nd 1 2 and 2 log to lowest non-trivial order in 1=. One says
that 1 is quadrati ally and 2 logarithmi ally divergent. To simplify the integrands we
dis uss the Feynman tri k
1
1 Z1
= dx
;
(22.4)
ab
(ax + b(1 x))2
0
101
whi h we an apply to the omputation of 1 by substituting a = (k + q )2 m22 + i" and

b = k2 m21 + i". For 2 we have the same assignment, with in addition m1 = m2 = m.
It is useful to also have the generalisation of the Feynman tri k for an arbitrary produ t of
s alar propagators.
P
! 1
!
!
P
Z

k
k
k
k
i ni
Y
X
X
( i ni ) Y 1 ni 1
n
i
= Qk
; (22.5)
ai
ai xi
xi 1
xi dxi
i=1 (ni ) i=1 0
i=1
i=1
i=1
whi h is proven by indu tion. In here
(z ) is the gamma fun tion, whi h satises (z + 1) =
p
1
z (z ), (n + 1) = n! and ( 2 ) = (see problem 2b). Consequently we nd
Z 1
4ig 2 Z
k2 + k q + m1 m2
1 =
;
dx
d
k

2
(2 )4 4 0
(k + (1 x)q )2 + x(1 x)q 2 xm21 (1 x)m22 + i"
Z 1
i2 Z
1
:
(22.6)
d
k
2 =
dx
2
4
4
2(2 )
0
(k + (1 x)q )2 + x(1 x)q 2 m2 + i"
We will show how to regularise these two integrals in two dierent ways. First we use
dimensional regularisation introdu ed by 't Hooft and Veltman (see problem 2) and than
dis uss Pauli-Villars regularisation. In dimensional regularisation the loop integrations are
repla ed by integrals in n, instead of 4, dimensions. The momentum integrations are always
of the form
Z
(k2 )
In;; (M ) dn k 2
:
(22.7)
(k M 2 + i")
We an evaluate this integral by performing the so- alled Wi k rotation, where we repla e
the integral over Rek0 by an integration over Imk0 . The integral over the two quarter ir les
indi ated in the gure, will vanish as the radius tends to innity. As there are no poles inside
the ontour of integration, we nd
Imk0
$
?
- - 6
6 ? Rek0
q
g. 14
&
(k 2 )
; k2 = k02 + ~k 2 :
(22.8)
(k2 + M 2 i")
We note that the integrand is a purely radial integral and as the surfa e area of a n dimensional sphere is analyti ally known (Sn = 2 n=2 = (n=2), e.g. S2 = 2 , S3 = 4 ,
S4 = 2 2 , ), we obtain
2 n=2 Z n 1
r2

In;; (M ) = i( 1)
r dr 2
(n=2)
(r + M 2 i")
i( 1) n=2 ( + n=2) ( n=2)
:
(22.9)
=
(M 2 i") n=2 ( ) (n=2)
We used the integral representation of the beta fun tion
zn 1
(n) (k) Z 1
=
:
(22.10)
dz
( n 1; k )
(n + k )
(z + 1)k+n
0
In;; (M ) = i( 1)
dnk
102
Shifting the integration variable k ! k (1 x)q we nd

Z 1
k2 (1 2x)k q + m1 m2 + x(x 1)q 2
4ig 2 Z
d
k
dx
1 =
;
n
2 + i")2
(2 )4
0
(k2 M^ x;q
Z 1
i2 Z
1
2 =
dn k dx 2
(22.11)
4
2 + i")2 ;
2(2 )
(k Mx;q
0
where
2 = xm2 + (1 x)m2 x(1 x)q 2 and M 2 = m2 x(1 x)q 2 ;
M^ x;q
(22.12)
1
2
x;q
whi h allows us to express i in terms of the integrals In;; (m)
o
4ig 2 Z 1 n
2
^
^
dx
I
(
M
)
+
(
m
m
x
(1
x
)
q
)
I
(
M
)
;
1 =
n;1;2
x;q
1 2
n;0;2
x;q
(2 )4 0
i2 Z 1
2 =
dx In;0;2 (Mx;q ) :
(22.13)
2(2 )4 0
Substituting the expressions for In;; (M ) from eq. (22.9), we nd
8
9
Z 1
2
2
n=
2
<
(m1 m2 x(1 x)q )
(1 + n=2) (1 n=2) =
4g
(2
n=
2)
dx
1 =
:
2 i")2 n=2 (2) (M^ x;q
2 i")1 n=2 (2) (n=2) ; ;
(2 )4 0
(M^ x;q
(2 n=2)
2 n=2 Z 1
:
(22.14)
dx 2
2 =
4
2(2 ) 0
(Mx;q i")2 n=2 (2)
This an be further simplied using
n
(1 + n=2) (1 n=2)
=
;
(22.15)
(2 n=2) (n=2)
2 n
su h that
9
8
4g 2 n=2 (2 n=2) Z 1 < (m1 m2 x(1 x)q 2 )
n=(2 n) =
1 =
dx :
2 i")2 n=2
2 i")1 n=2 ; ;
(2 )4
0
(M^ x;q
(M^ x;q
1
2 n=2 (2 n=2) Z 1
:
(22.16)
dx 2
2 =
4
2(2 )
(Mx;q i")2 n=2
0
The divergent part is now fully ontained in (2 n=2), be ause
(1 + 12 (4 n))
2
+ O(4 n) ;
(22.17)
(2 n=2) =
=
1 (4
4 n
n)
2
where = 0:57721 is the Euler onstant. We expand i around n = 4
1
0

Z 1
2 + (2 + 1 (4 n))M^ 2
m
m
x
(1
x
)
q
g2 2
1
2
x;q A
2
1 =
(n 4)=2 dx
2
(4
n
)
=
2
^
(2 )2 4 n
0
(Mx;q i")
(

Z
2

1
g
2
2
2
2
=
dx
m
m
3
x
(1
x
)
q
+
2
xm
+
2(1
x
)
m

1 2
1
2
(2 )2 4 n
0
)
i
h
( 1)
2
2
+ O(4 n) ;
1 21 (4 n) log( [M^ x;q i") + 21 (4 n)M^ x;q 1 +(0)
4 n 1
)
(

2 Z 1
2
(2 1) (0)
2
2 =
dx
log(

[
M
i"
)
+

+ + O(4 n) :(22.18)
x;q
8(2 )2 0
4 n
4 n 2
103
Note that q 2 = q02 ~q 2 and that the oupling onstant for the s alar three-point oupling
has the dimension of mass. We have split the result for i in a pole term with residue (i 1)
and a nite part (0)
i for n ! 4.

Z
g2 1
(0)
2
2
^
^
dx Mx;q
log( [Mx;q i") +
1 =
4 2 0
!

2
2
2
m1 m2 + 3x(x 1)q + 2xm1 + 2(1 x)m2
;
2 Z 1
(0)
2 i") + ;
2 =
dx
log(

[
M
x;q
8(2 )2 0
g2
2
2 + m2 1 q 2 ) ; ( 1) =
(
m
m
+
m
:
(22.19)
(1 1) =
1
2
2
1
2 2
2 2
(4 )2
We now note that the pole terms are of the same form as the tree-level expressions
obtained form the following extra term in the Lagrangian
L = 21 a( ')2 12 b'2 :
(22.20)
This means that we an hoose a and b so as to pre isely an el the pole terms. To lowest
order we therefore have
(0)
' (L + L) = ' (L) + b aq 2 = (0)
(22.21)
1 + 2 + O(n 4) ;
from whi h we an solve for a and b in terms of ( 1)
i
!
g2 1
g2
2
1
2
2
a=
; b=
(m1 m2 + m1 + m2 )
:
(22.22)
2
2
2
(2 ) 4 n
(4 ) 2
4 n
Note that as long as we stay away from n = 4 everything is well dened, in luding a
and b. The limit n ! 4 is to be taken after we have expressed everything in terms of the
renormalised oupling onstants and masses. We have taken here a slightly dierent approa h
for renormalising the theory. Rather than omputing at n 6= 4 physi al pro esses to x the
renormalised ouplings, we have started with renormalised ouplings and determined how
they have to depend on the bare ouplings so as to an el any innities that might arise as
n ! 4. It is lear that these two pro edures are equivalent. For the physi al interpretation
the rst pro edure (due to Wislon) is more transparent, in a loop expansion the se ond
pro edure is more natural. To nd the bare mass and the eld renormalisation (for the bare
oupling we should have onsidered the 1P I three-point fun tion with three ' external
lines) we write to one-loop order
q
p
m2 + b
2
2
2
2
1
1
LB = L + L 2 ( 'B ) 2 mB 'B ; mB 1 + a ; 'B Z'' = 1 + a' :
(22.23)
Often it an be determined by power- ounting (of momenta) whi h diagrams give rise to
divergen ies for n ! 4 or ! 1. The innities orrespond to lo al ounter terms (i.e with
a nite number of spa e-time derivatives) in the Lagrangian. For the theory in eq. (22.1),
power ounting easily shows that the ' four-point fun tion is logarithmi ally divergent at
one-loop order, see the Feynman diagram below.
m
1

g. 15
m2 ? 6m2
-
m1

104
We therefore need to introdu e an independent parameter for the ' four-point oupling, so
as to adjust its bare oupling to depend in the proper way on the uto, to ensure that
we an remove it. It an be shown that after adding to the Lagrangian in eq. (22.1) the
term 4 '4 =4!, the theory be omes renormalisable to all orders in the loop expansion. The
relevant parameters are m, m1 , m2 , g , and 4 .
As we have seen in se tion 11, ' (q ) should have a non-vanishing imaginary part if the
s alar parti le is unstable. It is lear that the s alar parti le itself an not de ay in two s alar
parti les, but when m1 + m2 < m it ould de ay in two fermions. Indeed, it is not di ult
to show that on the mass shell (q 2 = m2 ) (0)
2 is real
2 Z1
2 = m2 ) =
(0)
(
q
dx log[(x 12 )2 + 43 log(m2 ) 2 IR :
(22.24)
2
8(2 )2 0
We will show that Im(0)
1 (q 2 = m2 ) 6= 0 if and only if q 2 > (m1 + m2 )2 , alled the threshold
for de ay. The only way (0)
1 an develop an imaginary part is when the argument of the
logarithm in eq. (22.19) be omes negative. The threshold is therefore determined by
2 jx 2 [0; 1g = minfxm2 + (1 x)m2 x(1 x)q 2 jx 2 [0; 1g < 0 :
(22.25)
minfM^ x;q
1
2
Let us rst onsider the simplest ase of equal fermion masses, m1 = m2 . In that ase
2 jx 2 [0; 1g = m2 1 q 2 (m1 = m2 ) ;
(22.26)
minfM^ x;q
1 4
and the threshold is determined by q 2 > 4m21 = (m1 + m2 )2 . For the general ase of unequal
fermion masses, the minimum is obtained for x = 21 (1 + (m22 m21 )=q 2 ). After some algebra
we nd
!
2 ! (m1 + m2 )2
2
(
m
m
)
q
1
2
2
1
1 ;
(22.27)
minfM^ x;q jx 2 IRg =
4
q2
q2
whi h is indeed negative for q 2 > (m1 + m2 )2 . The value of x where this minimum is attained
does not lie in the interval [0; 1 if j(m21 m22 )=q 2 j > 1, whi h an be used to rule out the
other region, q 2 < (m1 m2 )2 , where eq. (22.27) is negative. This therefore proves that
the kinemati ally determined threshold oin ides with the threshold for Im(q ) 6= 0, as was
assumed in se tion 11.
A major advantage of dimensional regularisation is that it preserves the Lorentz and
gauge invarian es. Furthermore, it is a lo al regulator. The latti e regularisation also an
be arranged to preserve the gauge invarian e, but lo ality and Lorentz invarian e are only
valid at distan es mu h bigger than the latti e spa ing a. A momentum uto breaks both
the Lorentz and gauge invarian e. Pauli-Villars regularisation is aimed at having a regulator
that preserves the Lorentz invarian e. We will des ribe it for the Lagrangian of eq. (22.1),
using again the omputation of ' (q ) to one-loop order as an illustration. For ea h of the
original elds ' and (i) one adds extra (ghost) elds, with either the same (e` > 0) or
opposite (e` < 0) statisti s. This means that a loop of these ghost elds gets an additional
fa tor e` . Furthermore, the mass of these ghost elds is shifted over M` with respe t to the
original (\parent") eld (alternatively, if the original eld is a boson, one an shift m2 over
M`2 , see problem 39. With the present pres ription we an treat bosons and fermions on the
same footing). By dening e0 = 1 and M0 = 0, the index ` = 0 des ribes the original elds
of the model. We dene furthermore M` b` with m; m1 ; m2 . To regularise ' by
Pauli-Villars' method, we hoose
e` = (1; 1; 2; 2) and b` = (0; 4; 3; 1) ;

105
(22.28)
in other words the s alar and two fermion elds ea h have three ghost elds asso iated to
them but with non-standard weights for the loops. We ould sti k to standard weights, su h
that these ghost elds an be des ribed in terms of either grassmann or bosoni variables by
taking je` j elds, having either the same (e` > 0) or reversed (e` < 0) statisti s with respe t
to the original (\parent") eld. It is straightforward to give the self-energy in luding the
ontribution of the ghost elds
3 n
o
X
PV
(
q
)
=
e

(
m
+
b

;
m
+
b
;
q
)
+

(
m
+
b
;
q
)
;
(22.29)
` 1 1
`
2 `
2
`
'
`=0
in an obvious notation. The weights are hosen su h that the momentum integrals an all
be performed. Nevertheless, the masses of the ghost parti les, all proportional to , now
play the role of a momentum uto, as the Lagrangian will at that energy s ale no longer
des ribe a physi al theory. It is still onvenient 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 exa tly an el. Indeed, using eq. (22.19)
3 ( 1)

X
e` 1 (m1 + b` ; m2 + b` ; q ) + (2 1) (m + b` ; q ) =
(22.30)
`=0
!
3

X
g2
g2q2
2
2
2
e`
(m1 + b` )(m2 + b` ) + (m1 + b` ) + (m2 + b` )
=
2 2
(2 )2 (4 )2
`=0
g 2 (m1 m2 + m21 + m22 12 q 2 ) 18 2 X
3g 2 2 X 2
3(m1 + m2 )g 2 X

e
b
+
e` b` = 0 :
e
+
`
`
`
2 2
2 2
2 2
`
`
`
The nite result that remains (repla ing ( 1) in the equation above by (0) ) is nevertheless
still dependent on . To keep the following omputation transparent we take m1 = m2
PV
' (q )
(
)

h
g2 Z 1 X
2
2
2 x(1 x)q 2 i
=
dx
e
1
3

3
log

[(
m
+
b
)
x
(1
x
)
q
(
m
+
b
)
`
1
`
1
`
(2 )2 0
`
(
)

2 Z 1 X
2
2
dx e` log [(m + b`) x(1 x)q +
+
8(2 )2 0
`
Z 1
2

X
3g
2 x(1 x)q 2 h(m1 + b` )2 x(1 x)q 2 i
dx
e
log
(
m
+
b
)
=
`
1
`
(2 )2 0
`
Z 1
2

X

2 x(1 x)q 2
+
dx
e
log
(
m
+
b
)
`
`
8(2 )2 0
`
2 Z1

2 Z 1
2 x(1 x)q 2 3g
2 x(1 x)q 2 hm2 x(1 x)q 2 i
=
dx
log
m
dx
log
m
1
1
8(2 )2 0
(2 )2 0
+a1 2 + a2 + a3 log + a4 q 2 log + a5 + a6 q 2 + O(1=) :
(22.31)
The pre ise values of the oe ients ai are not very important, but an be al ulated expli itly with some eort. All dependent terms an be absorbed in a redenition of Z' and
the mass of the s alar eld, su h that

3g 2 Z 1
2 x(1 x)q 2 hm2 x(1 x)q 2 i
dx
log
m
PV
(
q
)
=
1
1
'
(2 )2 0

2
2 x(1 x)q 2 + a5 + a6 q 2 :
(22.32)
log
m
+
8(2 )2
106
(0)
Note that in dimensional regularisation (DR) we found the result (0)
1 + 2 , or
(
)
i
h
h
3g 2 Z 1
2
2
2 x(1 x)q 2 i
DR
1
+

dx
log

m
x
(1
x
)
q
m
' (q ) =
1
1
3
(2 )2 0
)
(
h
2 Z 1
2 x(1 x)q 2 i + :
(22.33)
dx
log

m
+
8(2 )2 0
DR
2
However, the dieren e PV
' (q ) ' (q ) = a5 b1 + (a6 b2 )q , where
2 ( + log ) g 2 m21 (3 + 3 log 1)
g 2 m21 (3 + 3 log
b1 =
;
b
=
2
8(2 )2
(2 )2
6(2 )2
1)
(22.34)
an be absorbed in a nite redenition of the mass and of Z' . If we dene the renormalised
oupling in terms of some physi al s attering pro ess, su h an ambiguity of ourse annot
arise. In that ase there is a unique relation between the bare and renormalised parameters.
This relation, however, depends on the regularisation used.
We will now dis uss, without a detailed derivation, the renormalisation of gauge theories
to one-loop order in dimensional regularisation. The bare Lagrangian is given by
L = 1 ( A A )2 1 B ( A )2 + B (i mB ) B + eB A B B : (22.35)
B
In n dimensions we still want the a tion to be dimensionless (h = 1), whi h implies that
L=n is dimensionless. From this we derive the dimensions of the elds and the parameters
in n dimensions,
1
1
1
1
[A = 2 n 1 ; [B = 1 ; [ B = [ B = 2 n 2 ; [mB = ; [eB = 2 2 n : (22.36)
B
If we dene (as is ustomary) " 4 n, one nds (for details see Itzykson and Zuber, e.g.
se tions 7-1 and 8-4. They use slightly dierent notations.)
!1
!
2
3e2 1
e2 1

ZA
+ A ; mB Zm m = 1 82 " + m ;
1
6 2 "
!1
!
q
2 1
2 1
2
1"
1"
e
e
B 2 Z = 2 1
+ ; B Z = 1 + 6 2 " + ;
8 2 "
!
2 1
1"
1"
e
eB 2 Ze e = 2 1 +
+ e :
(22.37)
12 2 "
We note that to one-loop order A eB AB and B =e2B are nite for n ! 4. This is
not an a ident, but the onsequen e of a so- alled Ward identity, whi h as a onsequen e of
the gauge symmetry (through the BRS invarian e mentioned at the end of se t. 20) relates
dierent Z fa tors,
Ze2 = Z = 1=ZA :
(22.38)
It is therefore sometimes mu h more onvenient to use
L = 4e12 ( A A )2 2e2 ( A )2 + B (i D mB ) B ; D iA :
B
B
(22.39)

To all orders in the loop expansion the eld A and the gauge xing parameter remain
free of renormalisations. The same holds for non-Abelian gauge theories. By absorbing the
AB
1
2"
A =
1
2"
107
harge q ( alled oupling onstant g q hen eforth) in the gauge eld, the Lagrangian an
be expressed as
L = 2g12 Tr(F2 ) + g2 Tr( A )2 + B (i D mB ) B ;

B
(22.40)
where the gauge eld and the gauge xing parameter re eive no renormalisations, in other
words they are already the renormalised eld and gauge xing parameter. The eld strength
F and ovariant derivative D are now given by ( ompare eqs. (18.35) and (18.41))
D = + A ; F = A
A + [A ; A :
(22.41)
If the gauge group is SU(N) and there are nf avours of fermions the renormalisation of the
oupling onstant is given by (see Itzykson and Zuber se t. 12.3-4)
!
2 1
1"
1"
(11
N
2
n
)
g
f
gB = 2 Zg g = 2 1
" + g :
(22.42)
48 2
As long as the number of fermion avours is small enough, we see that the one loop orre tions
to the bare oupling onstant diers in sign from the equivalent expression for the Abelian
ase. It is the self-intera tions of the non-Abelian gauge elds that are responsible for
the asymptoti freedom of its running oupling onstant. The running of the oupling is
expressed in terms of the so- alled beta-fun tion
(g )
g (gB ("); ; ")

;

(22.43)
where the derivative is taken at xed " and gB (g gR ). For non-Abelian gauge theories
one nds (0 is an integration onstant)
(11N 2nf )g 3
1
(11N 2nf )
(g ) =
+ O (g 5 ) ; 2 =
log(=0) + O(g 2 ()) ;
2
2
48
g ( )
24
(22.44)
whereas for QED ( oupled to nf avours of fermions)
n e3
1
nf
(e) = f 2 + O(e5 ) ; 2 =
log(=0 ) + O(e2 ()) :
(22.45)
12
e ( )
6 2
For other regularisations the omputation of the running oupling onstant is similar, ex ept
that " is repla ed roughly by 1= log(). To the order displayed, the beta-fun tions do not
depend on the regularisation s heme.
It is perhaps appropriate to end these le ture notes with as lassi an experimental test
of renormalisation ee ts in eld theory as the one for the Casimir energy in se tion 2. It
on erns the Lamb shift, measured in 1947, whi h is the very small energy splitting of the
2S 21 and 2P 21 orbitals in hydrogen atoms, re eiving a ontribution from va uum polarisation
ee ts (for a dis ussion of the other ontributions see se tion 7-3-2 of Itzykson and Zuber).
In problem 39 it will be shown that to one-loop order the photon va uum polarisation is
given by ( ompare eqs. (16.22) and (16.23). In the Landau gauge, ! 1, we an drop the
() fa tors)
(q ) = () (q )(q 2 g q q )! (q 2)( ) (q ) :
(22.46)
108
From the results of problem 39, where ! is omputed with Pauli-Villars regularisation, it
an be dedu ed that (m is the ele tron mass)
! (q 2 ) = a log(=m) + b + q 2 for q 2 ! 0; ! 1 :
(22.47)
The pre ise values of the oe ients a and b are not very important, as the ombination
a log(=m)+ b an be absorbed in the eld renormalisation (this means that a an be read o
from ZA given above). In se tion 7-1-1 of Itzykson and Zuber it is shown that = e2 =(60m2 ).
In the stati limit, as is relevant for the hydrogen atom, q 2 = ~q 2 and the photon ex hange
an be a urately des ribed by the Coulomb potential
Z
ei~q~r
e2
= e2 d3 ~q 2 ;
(22.48)
4r
~q
whi h due to the va uum polarisation is repla ed by
e4
(~r) :
60m2 2 3
(22.49)
The extra delta-fun tion intera tion, that arises from the va uum u tuations, only ae ts
the wave fun tions that do not vanish in the origin. Consequently, only the energy of the S
orbitals will be shifted by this orre tion
4me5
e2
;

=
;
(22.50)
E (nS 12 ) =
e
15n3
4
where n is the radial quantum number and e is the ne-stru ture onstant.
e2
Z
ei~q~r
1
e2
2
i~q~r =
d3 ~q 2
d
~
q
=
e
+

+

e
3
~q (1 + ! ( ~q 2 ))
~q 2
4r
!
109

Dict Dvi

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Dict Dvi

Caricato da

Copyright:

Formati disponibili

A COURSE IN FIELD THEORY

Pierre van Baal

Copyright P. van Baal

t d3~x (~x; t) (~x; t) 6= 0 :

whi h satis es the Klein-Gordon equation

d3~k '~(~k; t)ei~k~x=h :

H(~k) = j~ (~k)j2 + !(~k)2j'~(~k)j2 ; ~(~k)  hi  ~ :

jfn~k g >= jn~k >=

One an now formally take the in nite volume limit

( ')2   '  ' = g   ' ' ; V (') = 21 m2 '2 :

 S (' + ') S (') =

The lassi al Hamilton equations of motion are given by

For the Klein-Gordon eld we simply nd

whi h perhaps explains why V is alled the potential.

d3~x  (x)i '(x) :

det g ( 12 g   ' ' V ('))

For the original

d4 x " (x) (g  L(x)   '(x) '(x)) 

d4 x " (x) T  (x) ;

whi h implies onservation of the energy-momentum

F (x) =  A (x)  A (x) :

F (x)F  (x) A (x)J  (x)) :

A (x) ! A (x) +  (x)

D '(x)   '(x) ieA (x)'(x) :

This guarantees the ombined invarian e under a lo al gauge transformation

'(x) ! exp(ie(x))'(x) ; A (x) ! A (x) +  (x) :

S = Sem (J ) + S0 + d4 x e2 A (x)A (x)j'(x)j2 ;

Expli itly, it is given by the following Fourier integral

an be solved iteratively by substituting a series expansion for '(x),

'0 (x) = d4 y G(x y )J (y ) ;

G(x y )d4y G(y

This looks parti ularly simple in Fourier spa e

w)J (z )J (w)d4 zd4 w :

It is lear that this an be ontinued iteratively, e.g.

d4 y G(x y ) (2'0 (y )'n 1(y ) + 2'1 (y )'n 2(y ) + ) :

By indu tion it is now easy to prove that

It is straightforward to translate these Feynman rules to momentum spa e, by inserting

F (x)F  (x)

 F  (x) +    A (x) = J  (x) ;

( k2  + (1 )k k )A~ (k) = J~ (k) ;

5 Hamiltonian perturbation theory

d3~x J (~x; t)'(~x) =

here J~(~k; t) is the Fourier oe ient of J (~x; t), or

^ t) > we are using the intera tion pi ture, whi h is the

dt1 eiH0 t1 H1 (t1 )e

dt1 eiH0 t1 H1 (t1 )e

dt eiH0 t1 H1 (t1 ) dt2 eiH0 (t2 t1 ) H1 (t2 )e iH0 t2 j0 > ;

< 0j (t) > = e

f1 i" 0 dt1 < 0jH1(t1 )j0 >

Combining these results we nd

J~( ~k; t1 )J~(~k; t2 )

2i ik0 (~k)jtj

This means that we an rewrite eq. (5.11) as

J~( !; ~k)J~(!; ~k)

< 0j (t) >= 1 i

6 Path integrals in quantum me hani s

jx >= < x0 je iHT=N hjxN 1 > dxN

We will now use the so- alled Trotter formula

eipi (xi+1 xi )=h i( 2pm2i +V (xi ))T=N h

dx < xje H T =h jx >= Tr(e

in whi h SE is the so- alled Eu lidean a tion

to obtain the simple result

We an diagonalise the quadrati term by using Fourier transformation

t d3~x (~x; t) (~x; t) 6= 0 :

whi h satises the Klein-Gordon equation

d3~k '~(~k; t)ei~k~x=h :

H(~k) = j~ (~k)j2 + !(~k)2j'~(~k)j2 ; ~(~k) hi ~ :

One an now formally take the innite volume limit

( ')2 ' ' = g ' ' ; V (') = 21 m2 '2 :

S (' + ') S (') =

d3~x (x)i '(x) :

det g ( 12 g ' ' V ('))

d4 x " (x) (g L(x) '(x) '(x))

d4 x " (x) T (x) ;

F (x) = A (x) A (x) :

F (x)F (x) A (x)J (x)) :

A (x) ! A (x) + (x)

D '(x) '(x) ieA (x)'(x) :

'(x) ! exp(ie(x))'(x) ; A (x) ! A (x) + (x) :

S = Sem (J ) + S0 + d4 x e2 A (x)A (x)j'(x)j2 ;

F (x)F (x)

F (x) + A (x) = J (x) ;

( k2 + (1 )k k )A~ (k) = J~ (k) ;

f1 i" 0 dt1 < 0jH1(t1 )j0 >

2i ik0 (~k)jtj

jx >= < x0 je iHT=N hjxN 1 > dxN

eipi (xi+1 xi )=h i( 2pm2i +V (xi ))T=N h

dx < xje H T =h jx >= Tr(e

We therefore nd (using = T =h = N =h )

Tr(exp( H1 (N )T =h )) = Tr(exp( H2 (N )T =h )) = exp( F (

< ~x 0 je iHT=hj~x >

whi h satises the property that

jp~ > ay(p~)j0 > :

Z (J; gn ) < 0jU (T )j0 >=

Note that fa tors of 2 are dropping out in the identities

to be satised at any nite order in perturbation theory).

1)Vn (2V ) 12 (n 2)Vn (2V ) 21 E 1 =