IC/95/315

INTERNAL REPORT Outline

(Limited Distribution)
The purpose of these lectures is to provide an elementary introduction to the subject,
of quantum field theory, leading up to the perturbative formulation of quantum electro-
International Atomic Energy Agency dynamics, but not including renormalization theory. The emphasis is on perturbation
i •;:-• v and theory. Fields are represented as collections of harmonic oscillators with weak couplings.
V ' - •. \ ns Educational Scientific and Cultural Organization Feynman rules, and their graphical representation, are treated in some detail. They are
-\
INTERNATIONAL CENTRE FOR THEORETICAL PHYSICS applied to the perturbative computation of Green's functions and S-matrix elements,
The course begins with a brief review of sotne very elementary ideas from classical and
quantum theory. These include a discussion of the quantized simple harmonic oscillator
and generalizations to sets of coupled oscillators, forced oscillators, etc. This leads to the
notion of asymptotic states and the 5-matrix for oscillator systems (Lecture 5).
The interaction picture and perturbation series are introduced in Lecture 6. This
leads to the definition of chronological ordering and causal Green's functions. A version
of Wick's theorem is arrived at. In the following three lectures free fields are quantized,
QED LECTURE NOTES Wick's theorem is generalized and the LSZ formulas for extracting asymptotic state matrix
elements from Green's functions are derived.
1994 Lectures 10 and 11 are concerned with symmetries, both spacetime and internal.
Noether's theorem is derived and the algebras of symmetry generators are illustrated.
The relationship between physical observables and 5-matrix elements is the subject of
Lecture 12 where reaction rates are discussed together with formulae for differential cross
sections.
Lectures 13-16 deal with perturbation series for Green's functions and 5-matrix ele-
ments. Wick's theorem is applied to formulate the Feynman rules, firstly for scalar fields
3. Strathdee
International Centre for Theoretical Physics, Trieste, Italy. with cubic or quartic self-interactions, then for Yukawa theory. The rules are developed
firstly in the configuration space version and then in the momentum space version. Be-
cause the combinatoric subtleties can be confusing, we illustrate the rules with many
examples.
The remaining lectures are devoted to electrodynamics. The gauge principle is intro-
duced, covariant derivatives are defined and the quantization of the electromagnetic field
is described. The quantization is carried out in a class of covariant gauges where the
important distinction between physical and unphysical states is made. The LSZ rules for
obtaining physical state matrix elements are derived. Finally, the formalism is illustrated
in Lecture 19 where the Klein-Nishina formula for electron-photon scattering is derived.
A brief introduction to the problem of ultraviolet divergences is given in Lecture 20.

MIRAMARE - TRIESTE
September 1995
 Ill

Table of Contents 1 Classical theory

1. Classical theory 3
2. Canonical quantization 9 Classical mechanics is based on the action principle. This principle ensures the consistency
3. Quantized oscillators 14 of equations of motion and it is used in analyzing general properties such as symmetry
4. Fermions and bosons 20 and conservation laws. Here we give a very brief review of classical theory, discussing
5. The forced oscillator 25 the relation between Lagrangians and Hamiltoiuans and the use of different kinds of
6. Perturbation theory 31 coordinate systems. We consider the theory of weak perturbations and the role of oscillator
7. Free fields 37 variables,
8. Green's functions and Wick's theorem 45 a) Lagrangians and Hamiltonians
9. Interacting fields 50
10. Symmetries in quantum field theory I (Poincare Group) 60 The action functional, S(<j),for a system with one degree of freedom, i.e. one coordinate
11, Symmetries in quantum field theory II (Internal) 67 variable q{t), is given by the time integral of a Lagrangian function:
12. Cross sections 72
13. Perturbation theory (I) 80 S{q) = ]it dtL(q(t).q(t)) (1)
14. Perturbation theory (II) - Feynman graphs 88
15, Fermions and Yukawa theory 98 where q = dq/dt, is the velocity. For example, the Lagrangian for a particle moving in a
16. Momentum space 106 potential is
17. Electrodynamics (I) 117 2
L= |q - V(q) (2)
18. Electrodynamics (II) 124
L9. Corn pt on scattering 131 where m is the mass and V is the potential.
20. Ultraviolet divergences 139 The trajectory, q(t), followed by the particle is the one that minimizes the value of
S(q) - subject to the boundary conditions, q{tt) = qt and <)(<;>) = tfe. To find this
trajectory we have to calculate the functional derivative, fiS/f>q(t). and set it equal to
zero. The functional S(q) can be thought of intuitively as a function of infinitely many
variables, {q{t)}, where t is the range tj < t < t 2 . The functional derivative is then
like a generalization of partial derivatives. Thus, if we have a function of n variables,
f(xi,... .xn), then its partial derivatives are defined by writing, for small bx3.

where the dots indicate quadratic and higher powers of the small quantities 6xi,..,,(
For the functional, S(q), we write

S(q + 6q) = S(q) dtbq{t) (3)

6q(t)
When S(q) has the simple form (1) we can write SS/Sq in terms of the partial derivatives
OL/Oq and OL/dq. Firstly,

6S(q) =

-I,
d L d L
j (
dt bq—- + 6q~r c

d V "<l "q
Since q = dq/dt it follows that bq — diq/dt. Hence we can use the identity,
r.dL d (c dL\ r d (dL^
 where we have used the chain rule for differentiation, the definition (6), and in the last
step, the Lagrangian equation of motion (5). Equations (8a) and (8b) are known as the
-L hamiltonian equations of motion.
For the example (2) we have p = mq so that
i)L\ (c dL
*-s'^'>+ +,.,
H(Q,P) = Pq-^q2 + V(q)
If we now require Sq(ti) = 6g(tj) = 0 then bS{q) has the form indicated in (3) with
d
— - ®L (2L (4)
Jq ~ ~dq ~ dt \~dq~ and the equations of motion read
l
Hence, the minimum of S(q), subject to the condition that q(ti) and q(t2) are fixed
numbers, is determined by the equation

V dy
(
dq dt\dq All of this generalizes immediately to system with N degrees of freedom where the
This is the Lagrangian equation of motion. For the example (2) we have Lagrangian is given as a function of JV coordinates <?i ,...,<?* and N velocities qt,... , qN.
dL 3V 3L We aje mainly interested in systems where iV —» oo, which are known as "field theories".
ami m
IT
oq = —T~
aq ^dq = 1 One of the simplest field theories is characterized by the Lagrangian
so that the equation of motion reads =I (9)
iJV
- — mq = 0
dq where <p = <i>(xt i) is a real-valued function of space, x, and time t. The integrand L is the
"Lagrangian density". To put L into the general form discussed above we must introduce
The Lagrangian is always given as a function of coordinates, q, and velocities, q. For
a complete set of functions, ua(x), on space, writing
many purposes it is more convenient, to use a formulation in which the velocities have been
eliminated in favour of "momenta". The canonical momentum is denned as a function of tj>(x, t) = ^2qa(t}un(x) (10)
q and q by
P= ~ (6) The set of coordinates qa(t), a = 1,2... carries the same information as the field <j>(x, t).
Usually it is easy to solve this equation for q as a function of q and p. The Hamiltonian If we substitute the expansion (10) into (9) and perform the integration then we obtain
function H(q.p) is then defined by the so-called "Legendre transformation", the Lagrangian L(qi,q?t.. .?i,<J2,...). The action principle then leads to equations of
motion of the type (5).
H(q.p)=pq~L{q,q) (7)
dL d_ /dL_
where q is expressed as a function of q and p on the right hand side, it follows that = 0. a = 1,2,... (11)
dqa dt \3qa
dH . dq i dL\
There are many different complete sets of functions. One of the most useful expansions
is the Fourier series.
= Q (8a)
dH dL\ dL
dq dq) dq where V is the volume of a large box. The coordinates, q% = q-k)' are complex numbers
dL given by

= -P (sb)
'More pr«x-ise]y, the equation bS/bq — 0 gives an "extrernum". It may be a minimum, a maximum,
t>r a saddle point depending on the circumstances. b) Weak perturbations
 It is always possible to choose coordinates in such a way that the Lagrangian takes Although the successive steps lead rapidly to unwieldy expressions we see that, at least
the real form. in principle, we could obtain the first few terms in an expansion of go in powers of <?„,,„•
The basic point here is that many dynamical systems, including field theories, can be
£ = E J ( £ - < £ £ ) - f ( 9 i , 92. •••) (12) treated perturbatively as systems of coupled anharmonit oscillators. At least this makes
where the first and second derivatives of V vanish at the origin, sense if the anharmonic couplings V',,,3-,. ... are sufficiently small. We shall nov develop
the classical theory any further since our main interest is with the quantized theories.
9i = 72 = • • • = 0

If the frequencies, LJO, are real numbers then the system is said to be "stable". We shall
always assume this. The equations of motion are
qa = -u2aqa + Ia{q), a = 1,2 (13)
where Ia = -dVjdqa. They resemble the equations for simple harmonic oscillators sub-
jected to an external force, Ia, Here, of course, the forces are not external because Ia
depends on all the coordinates qi,qi,. •.

where
d3v (14)
, etc.

However, if the coefficients Va^ ... are sufficiently small, it may be possible to regard the
coordinates themselves as small, and hence to solve the equations of motion approximately.
The perturbative analysis of such equations is the main subject of this course.
In the linear approximation we have /„ = 0 and the equations (13) are easy to solve

where An and f)u are real integration constants. Equivalently,

(fii , , , 1 _ .... I
(15)

where afl is a complex integration constant and the term "c.c" indicates the complex
conjugate. To implement the perturbative programme we can replace (13) by equivalent
integral equations.

9«(0 = 9o.».W + j " (16)

where </„,;«(') is a solution of the homogeneous equation,

ft*, = =%»»<:-**•'+c.c. (17)

that represents the early time (initial) boundary conditions. In principle one could solve
(16) by iteration. The first step would be to substitute for qp(t') in the integrand of (16)
the expression

9fl(t') = 93,m(t') + — / df'si

6
Exercises 2 Canonical quantization
In quantum theory coordinates and momenta are represented by operators subject to the
1.1) With L = L{q,q,q) derive the equation of motion. What are appropriate boundary commutation rules
conditions to make 6S = 0?
[<ja, h\ = 0, [qa,Pin] = ' « W , \Pa.pp] - 0
1.2) Given the classical Hamiltonian,
Other observables are functions of these ones, A = A(q,p). States of the system are
pi + V(q) represented by vectors, \iji), in an infinite dimensional complex vector space on which the
operators act.
find the Lagrangian, L(q,q). Assume that g'J is a non-singular, symmetric matrix
whose inverse is j/y. a) Hiibert space

The infinite dimensional complex vector space admits a complex-valued inner product
between pairs of vectors,
1.3) Verify that the integral equation (16) is equivalent to the differential equation (13) < W ) = {4>WY
under the assumption that Ia(<i(t)) goes to zero when ( —* —oo. The norm, {ip\ip}, is real and positive. It is always possible to choose a complete, or-
thonormal basis of vectors, |n), n = 1,2,... such that

Operators act on these vectors.

= Y,\n){n\A\m){rn\M>)

i.e. A is represented in this basis by the infinite dimensional matrix, (n|-4|m). The
hermitian conjugate, A+, is defined by

The eigenvalues of a hermitian operator are real and the eigenvectors axe orthogonal
= \n)An
(n'|n> = 6nn,
In the eigenvector basis A is represented by a diagonal matrix,
(n'\A\n) = An6nn,
(If the eigenvalues ale continuous then
(\'\A\\) = A{\)6{\ - V)
where S( } is the Dirac delta function). The "expectation value" of A in the state \il>) is
given by
with j(n|v)| 2 interpreted as a probability distribution. It follows that U(e)~l\q) is an eigenvector of q,
If two hermitian operators commute. [A. B] = AB - BA = 0, then their eigenvectors
arc common, qU(£)-l\q) = U{
A\n) = \n}An and B|n) = \n)Bn .
The matrices representing A and B are both diagonal in the eigenvector basis,

(n\A\m) = AnSnm, (n\B\m) = Bn6nm

for arbitrary f. Expanding around s = 0 gives the identity,

b) Functions of operators

The operator j(A) is denned to have the same eigenvectors as A, with eigenvalues where is the wave function. In general, if A = ^(^,p) then it is represented
given by f(An), by
f(A)\n) = \n)f{A») -—
i dq
For example,
The eigenstates of p satisfy
P\P) = \?)P
where ( is a number. It follows that
so that
-e<* = Ae'A = etAA {q\v\v) = -^-(Q\P) = (q[p)p
dt
which is solved by

<•) Coordinate representation

This implies the normalization of momentum eigenstates such that,
The coordinate <j usually has a continuous spectrum of eigenvalues, -oo < q < oo,
(p'\p) = ]dq{p'\q){q\p)
Q\Q) = \<i)q
= I' dqeMp-^
W\<I) = HQ'-Q)

To see how the momentum operator, p, is represented in this basis, define the unitary
operator, where we have used the completeness of the coordinate basis
£/(£)= e*"*
d) Time dependence
It is easy to verify that U(e)+ = U(s)~'. Consider the operator
The time dependence of physical states - in the Schrodinger picture - is governed by
Q[e) = U(e)qU(£)'1 the equation.
The derivative with respect to e is

where H = H(q,p) is the Hamiltonian. In the coordinate basis,

= tf(j«-jtf)<r'
I

Also Q((t) = IJ because £/(0) = 1. Hence, where t(q,t) = {q\il',t) is the wave function. If H is independent oft then

10 11
 The time dependence of matrix elements of A is given by Exercises

2.1) Prove that U(t) = e "" is unitary when ( is real and H is hermitian.

where A(t) is defined by 2.2) C a l c u l a t e t h e positive s q u a r e root m a t r i x , A i / S , w h e r e A is t h e h e r m i t i a n 2 x 2

matrix
a + cos 8 sin 6
A =
The use of time dependent operators and time independent states is known as the "Heisen- sinS a — cos 6
berg picture". The time dependence of operators in the Heisenberg picture is governed w i t h a > 1.
by the equation

Other pictures can be useful. Suppose H = HQ + V where Ha is simple and V is small.

Define the unitary operator fi(i) by

e '• = e '•'

and the time-dependent operators by

A(t) = e*"v°Ae-*"'°

Then the time dependence of matrix elements is given by

where
A(
and

If V is small then fj can be obtained as a power series in V. This is known as the

interaction picture.

12 13
( . . ..*. >_ Bt .

3 Quantized oscillators and also to write the Hamiltonian in the manifestly time-independent form
The simple harmonic oscillator is governed by the Lagrangian, H = -(aa*+<i+a) (8)
m .^ k ^ To find the eigenvalues of H we use the commutators
(1)
or the Hamiltonian, [H.a] = -huM

(2) [H, a+] = huja+ (9)

Notice that H is a positive operator, i.e.
where the mass, m, and spring constant, k, are positive. We shall study the quantum
properties of this system. ^ ^ 0
a) Energy spectrum +
where \ip\) = a \ip) and |i^>) = a\4>), for any \iji). This means that the eigenvalues of H
are all positive. Hence there must be a lowest value, Eo,
In the quantum theory [q,p\ — i,h and the Heisenberg picture equations of motion are
H\0) = |0}En
Now consider the state u|0), using the commutator (9),
p = -\j?,H] = -kq Ha\0) = ([H,a] + aH)\Q)
In
= (-hu + EB)a\Q)
Hence, on eliminating p, the equation of motion for q is
This seems to indicate that a|0) is also an eigenvector and that the corresponding eigen-
q= -w q (3) value is Eo - huj, which is less than Eo. This contradicts the assumption that ER is the
lowest value. Hence, we must have
where ^ = Jk/m is the "classical oscillator frequency". This is the only important a|0} = 0 (10)
parameter. Indeed, we can eliminate A; and m by rescaling the coordinate, The "ground state", |0) is annihilated by the operator, a. Using (7), (8) and (10), it is
fm easy to compute Eo- One finds
|0)-Lw (11)

so that All the other eigenstates are obtained by acting on |0) with the creation operator. a+
They are
2 1 / a
}, n=l,2,,., (12)
2
They satisfy2
where hQ = P. The general solution of the equation of motion is written in the form
H\n) = |n)(n+-) (13a)
(13b)
J
T o prove t h e s e results w e use t h e following identity. L«t A, B l n Bi,... b e any operators. Then,
where the operator, a, is an integration constant. We can express this operator in terms
of Q and Q.

\A,Bl}B2...Ba + S i [A, B2)B3... Bn + B,B-,AB3 • • • Bn

(6} \A,B,]B3...Bn + BnlA,Bi\B3...Bn + ...

+BiB2 .,. Bn-i[A, Bn] + B,Bt... BnA
Since {Q(t), Q(t)] = i, according to the canonical rule, it is easy to show
Hence, the very useful formula

1A,S, ...Bn\ = [A, i...BK-i[A, Bn] (14)

14 15
 (Q\\) =
The operators, a and o + , are known as annihilation and creation operator, respectively.
The eigenstate |n) is the state with n "quanta". Applying a + creates another quantum,
i.e.
a+\n) = \n (16)
whereas a annihilates one.
a|n> = 171 - \)y/7wii (15) c) Coherent states

b) Coordinate basis These are superpositions of n-quanta states, denned by

It is sometimes useful to employ the coordinate basis. What is the ground state wave
function, (Q|0)? It must satisfy the equation

0 = <Q|a|O)
where a = a\ + (Q2 is a complex number. The coherent state wave function, (Q|u), is
= (Q\(uQ+l-P)\Q) found by solving the differential equation
, d
^r(Qln) = <Q|a+|u)

which implies
<Q|0> = ce
The general solution is
wliere c is a constant, Normalization gives
(Q\a) = cW1 F[Q - a)
= I'dQ(0\Q){Q\0) where F is some function. The boundary condition at a = 0 gives
J -oo
F(Q) = e-^
= \c\2 jdQe-"?

Hence
Hence, (18)
\ TV

m - The coherent states form an "overcomplete" basis which is to say they are complete but.
not orthogonal. It is easy to prove
In particular, this formula implies (19a)
\H\arf\ = nfiuj(a+)"
which leads directly to (13a). On the other hand and
m +
(o|o (o )"|o) = (oifr-'Ma+m (19b)
IT J-x
m 1 +
= n2u(0|a - (a )" Coherent states have a simple time dependence following from the formula
= n(n -

If in > n we can repeat the operation n times to reach the matrix element (0|am""[0) = 0. If m < n we This gives
can repeat it m times to reach (0|(a+)"~m|0) = 0 (because (0|a+ =0). If m - n we obtain
(0|on(o+)"|0) =n!(2u)"(0|0)
Then, choosing <0|0> = 1, this leads to (13b),
(20)

16 17
In this basis the "evolution kernel" is given by Exercises
(21)
3.1) Prove the formulae (19a) and (19b).
In the coordinate basis this kernel is therefore
3.2) Using the formulae (5) and (7) calculate the commutator [Q{i),Q(t')] for 1 ^ f

where we have used the completeness condition (19b). On substituting the wave functions
(18) it is possible to perform the integrations over a and 0 to obtain

V" f (22)
^'' '^ ' \2irismut/
In the basis of energy eigenstates the result is very simple,

(23)

18 19
4 Fermions and bosons (These states are not normalized). It is clear from (4) that they are symmetric,

The formalism for a single quantized oscillator can be generalized immediately to a system
of JV independent oscillators. The associated quanta are known as "bosons". We discuss
the formalism ;ind then introduce another kind of oscillator whose quanta are known as
"fermions". b) Coupled oscillators
a) Bosons It often happens that the oscillator Lagrangian arises in a mixed form where it is
difficult to identify the independent oscillators. In such cases it is always possible to
The Lagrangian for N independent oscillators is
simplify the description by transforming to "normal coordinates". Suppose we have

(1) (9)

Canonical quantization gives the (equal time) commutators where mQ(j and kap are symmetric positive matrices. We make the transition to normal
coordinates in two steps. Firstly, compute the matrix (m1/2)O(3 according to the general
methods of Lecture 2. Define the new variables
1 ^ , ,,.,.
(2)
The Heisenberg picture equations of motion are solved by
so that (9) becomes
(3)

and the time independent creation and annihilation operators are found to satisfy the
commutation rules where m'1/1kTn^1^2 is a positive symmetric matrix. This matrix can be diagonalized by
an orthogonal transformation. That is, we can find a matrix R satisfying R~l = RT
n. a.,] = 0, [aa. a+] = 2w(t<W, K. 41 = ° (4) (transpose of R) such that
The Hamiltonian is R'lm'U2krn~mR = diag matrix
h
H
= \ (5)
Exactly as before the ground state is annihilated by the aa, (10)
aa|0)=0, a=l JV (6)
It follows that the eigenstates of H are given by The eigenvalues of m are utf,... oij., i.e. (10) reads
rn-^)a(lRa, = floX, 7 = 1, • • -, N
10} (7)

which correspond to the eigenvalues The columns of R are eigenvectors of m~1'2fcm~l''J. Define the normal coordinates
\ j Qa =
E(nu...,nN) = '£(na + -)hu>a (8)
l '
(11)
An equivalent way of writing these states is in the form of symmetric tensors
|Q,,Q2,...) = <<...|0) and write the Lagrangian finally as,

(12)

20 21
c) Fermions The other states are obtained by acting with i>+,
Consider the very simple Hamiltonian k,as,...) = 6J 6 + , . . . |0)
= ± ( 6 + ) - ( f c + ) " . . . (6+)»»|0) (21)
H = \b-b (13)
These states are antisymmetric in the labels cxy, ttj,...
where b ad b+ obey the rules |a,,Q2,--.) = -\cti.au-..)

b2 = 0, bb+ + b+b = 2UJ, (14) They exist only when the labels are all different from each other, which means that the
numbers of quanta of each type nt,... ,nN can equal only 0 or 1, There are only 2N
As before we find the oscillator type commutators independent states. The energy eigenvalues are
[H,b] = -hub E(nu...,nN) = (22)
(15)
The quanta are known as "fermions".
which suggests that b and b+ can be interpreted as annihilation and creation operators. The next step is to invent a Lagrangian and appropriate quantization rules for fermion
Also H is positive and so must have a lowest eigenvalue. The ground state is defined by systems, Write
6|0)=O (16) (23)
+ 2
However, since (ft ) = 0 there is only one excited state Viewed as a "classical" system this would give the equations of motion
(24)
(17)
The canonical momentum would be
There are no states orthogonal to both |0) and |1): the Hilbert space is 2-dimensional. _ 0L _ . ,|+
(25)
Hence b, b+ and H are represented by 2 x 2 matrices (Pauli)

b 0 V ^ \ so that (23) takes the canonical form

=
f) 0 ) N

0 0
1
N
0 0 \ hu, (26)

In order to have Heisenberg picture equations of motion in the form (24) we impose the
This can be generalized to A' pairs of operators like b, b+. The Hamiltonian (13) is anticommutation rules
generalized to
H = lh+Ja (18)
(27)
where anlicommutation rules are imposed.
The general solution of the equations of motion is
(19) 1 .
where the notation means
(28)
Notice that bf = 0, etc., so that the Hilbert space is again very limited. The ground state
satisfies where the time independent operators ba, b* satisfy (19) and the Hamiltonian (26) reduces
ba\0) = 0 , a = 1 , JV (20) to the form (18).

22 23
Exercises 5 The forced oscillator
As a preliminary step towards the treatment of interacting oscillators, we examine the
4.1) Diagonalize the Hamiltonian rase of a single oscillator acted on by an arbitrary external force. We shall from now on
employ units in which h = 1. This saves some writing.

a) Heisenberg picture
i.e. find a and a+ as linear combinations of the canonical q and p such that
The classical forced oscillator is described by the Lagrangian
+
H= -(a a

Give the ground state wave function.

where J{t) is an arbitrary function representing the external force. The classical equation
4.2) Diagonalize the fermion Hamiltonian of motion is
Q = -JQ + J (2)
Canonical quantization proceeds as before but the Hamiltonian is now dependent on time.
In the Schroeding picture.
where
(3)
Solve this problem by making the ansatz,
Transforming to the Heisenberg picture is therefore more complicated. Write

(4)
ri2 = J —(6+sin 0 + 62 cos 0)
V <JJ where U(t) is a unitary operator that satisfies the Schroedinger equations,
showing that S! and 0 can be found such that
idM*-) = H(t)U(t)
1
H = —()Ji 1)1 + 1/2 Vv + const {t)-[ = -U(t)~lH(t) (5)

where It follows that

dtQ(t) = -i

OtP(t) = l-

on using the canonical commutator [Q, P] = i. Eliminating P(t) shows that the Heisen-
berg picture equations agree with the classical equation (2).
Assume that J(t) vanishes outside a finite time interval. Then at sufficiently early or
late times the equations are homogeneous (J = 0) and the general solution must reduce
to the form studied previously. For early times write

Q(t) = Q>n(t)
(6a)

24 25
 • «<**!>*» • » •

while, on the other hand, for late times write is known as the 5-matrix. It is unitary since the in- and out-states comprise two distinct
and complete orthonormal bases
Q(t) = Qimt(t)
1 / out){m out\ri in)
(6b)
2uj
= (?i in\n in)
To find the relation between Qm and Qoul we must integrate the equation of motion (2).
The result is, according to (1.16),
We can write
- / - t')J(t') (7) (m tmi|n in) = ( m in\S\n in) (13)
+
The right hand side clearly reduces to Qtrl(t) for sufficiently early t. Also, for sufficiently where S is a unitary operator, S — S~\ Another way to express this is
late times the integral on the right hand side becomes mdependent of its upper limit, i.e. (n out | = (n in\S

Qoutit) = QinW + - r dt'BiBuj{t - t')J{t') (8) which leads to the operator formula
u) J -oo

Define the Fourier transform of J(t),

(14)
J{E) = fX dtJ(t)eiEt (9)
Our task is to solve (14) for 5. It will turn out that S is expressed in terms of
exponentials of the operators ain and o,"^. To show this consider firstly the operator
Hence (8) reads function,
Q,At) ==Q,n(t) - ~ ^ J{~u) + ^ - e - " J ( «
Q,,,At) a(a) = e""»u m e~» ,
where a is a complex number. Take the derivative
a
out — ain + iJ{ij)

C = at - i J(-w) (10)
Since J(t) is real we have J(-UJ) = J(UJ)*, so the equations (10) are consistent. Thus
we find that aout differs from ain by a numerical (not operator) term proportional to the (because the equal time commutation rule [Q,Q] = i, at sufficiently early times, gives
Fourier component, of the external force at the oscillator frequency. [ain,a,Y1] = 2w). Since o(0) = ain we find
This is not a trivial effect. It implies, for example, that the Hilbert space contains
two distinct ground states, |0m) and |0oui), appropriate to early times and late times, a(a) = am
respectively. They are defined by
This suggests that S should contain the factor eQO*» with a = ij(u)/2uj. In order to be
ain\Qin) = 0 and <w|0 out) = 0 (11) unitary it needs to include also the factor, e"™*""1, which commutes with ain. We therefore
try the ansatz
The quantity |{0out|0m)|2 measures the probability that the oscillator remains in its
5 = eaa™e-a'"">e' (15)
ground state after the external force has acted on it. Generally, the oscillator will make
transitions, and we need to compute the probabih'ties, \(mout \n in)\2. where 7 is a complex number chosen to make S~l = S+. Is this possible? Observe firstly,

b) The S-matrix 5-1 = e a-a, n e - Q at e -, (16a)

The matrix of transition amplitudes, 5 + = e - att -e Q "°'"e 7 ' (16b)

Smn = (m md\n in) (12) To establish the equality of these operators we need the Campbell, Baker, Hausdorif
formula

26 27
which is valid if the commutator, [A, fl], is a number (see Exercise (5.1)). Applying this we obtain by integrating over f,
to {16a} and (16b) gives
r
Hence, unitarity is established if -LJ|Q| 2 — -y = OJ|Q|3 + 7*, i.e. J-o

(18)
where Q is any real number. Thus we have the result,
which agrees with (10).
(19)
It is now a simple matter to evaluate the transition amplitudes
1 / + \
(m out\n in)0 — { |0 in)

For example,
<n out |0 in) = 4 = f2 - ^ (20)
Vn! V
The probability distribution of the excited oscillator states resulting from the action of
the external force is a Poisson distribution,
x"
\{n out\i) m)\2 = —e~J

where

Having analyzed the response of a single oscillator to an external force we can gen-
eralize directly to a system of independent, oscillators. The 5 operator will simply be a
product of commuting factors corresponding to the individual oscillators.

c) Asymptotic condition
In the derivation we assumed that the external force vanishes outside a finite time
interval so that the oscillator is free at sufficiently early and late times. This assumption
is stronger than necessary. For the integration (7) to make sense we need only assume
that J(t) —» 0 as t —» ±00 rapidly enough to ensure convergence of the Fourier integral
(9). We then have
Q(t). f Q,M
Q™(t) ,. t t-,-oc

or, more precisely,

lim \eiiJt{idt+u>)Q(t)\ = ain
t—*—OC I- J

lim le^iidt + u)Q(tj\ =<w (21)

FYom the identity,
dt

28 29
 - T ••' -

Exercise 6 Perturbation theory

5.L) Prove the CBH formula (17). To do this proceed as follows: In the last lecture we obtained the S-matrix for a forced oscillator by integrating the
Heisenberg picture equation of motion. Now we consider the same problem but in the
(a) Let ( be a real parameter and consider the operator interaction picture, This method is important because it can be applied to systems with
more complicated interactions where it is used to develop perturbation series.
IV'(i) = e-'V ( ' t + f l )
a) Interaction picture
Derive the differential equation
In terms of the time-independent Schroedinger picture dynamical variables, Q and P,
~ = B(t)W(t) the time-dependent Hamiltonian of the forced oscillator is

where B(() = e'iABetA

where
(b) Evaluate B(t) on the assumption that [A. B] is a number
(c) Solve the differential equation for W{t) and set t — 1.
V(t) = -J(t)Q (1)
The time evolution operator, U{t), used in transforming to the Heisenberg picture, obeys
the equation
id, 1/(0 = H(t)U(t) (2)
We look for a solution of the form
U(t) = e-""° (3)
The unitary operator, Q(t), obeys the equation

= €
= v,(tMt) (4)

where V/(t) is denned by

V,(t) = eil"°V(t)e-lt"a
= -J(t)Qt(t) (5)

The subscript, /, indicates "Interaction picture". Thus

idtP, = [P/,ffoj = - ^ Q /
These operators obey the free equations of motion, in contrast to the Heisenberg picture
operators,
QH[t) = U{t)-lQU[t)
(6)

etc., which obey the exact equations of motion,

31
 The external force determines the time dependence of SI. If we assume that J{t) etc. By definition the T-product is symmetric. We can write the factors in any order
vanishes outside a finite time interval then Q(t) becomes constant at sufficiently early and inside the T-product. In effect, the operators commute,
late times. Let us choose the boundary condition,
T({V,{t1),VI(t2)])=0
!!(-oc) = l (7)
We can now write (11) in the form
Then, according to (61), we have Qn(t) — Qi(t) at sufficiently early times, where we also
have Qn{t) = Qm{t)- Since Qin(t) and Qi{t) obey the free equations, this means
(13)
Qi(t) = QmM (8)
because the enlarged domain of integration contains n! copies of the original one. (There
at alt times. Applying the same logic at late times we find are n! distinct ways of ordering the n times, ([,..., ( n ). The series (13) is usually written
in the compact symbolic form
Qoui(t) = n(+oo)~lQin(t)il(+oo) (9)

Comparing this with (5.14) we see that, up to an undetermined phase factor, (13')

n(+oo) = 5 (10) Hence, in particular,

(14)
To find the ^-matrix we could solve the differential equation for tl(t), subject to the
boundary condition (7) and take the limit, t —> +oo. The equation (4) can be solved If the operator V, is in some sense "small" then we may keep only the first few terms of
exactly for the case of the forced oscillator (and we shall do it in the following) but it can the expansion (14) in order to make approximate calculations of 5-matrix elements.
be solved approximately in other more complicated and interesting systems. This is why
the interaction picture approach is useful. c) The forced oscillator (Causal Green's function)

b) Chronological ordering For the case of the forced oscillator, Vt = ~J(t)Qin(t), we can evaluate il(t) explicitly
and verify that f!(+oc) = S. We start with the ansatz,
It is easy to solve (4) by iteration. The answer is

= £(-*)"
£(-i f_ju j" du ... ft"'ldtnVl{t1)...V,(tn) (11)
J-oc- where Q{() is complex and 0(t) is real. The operator (15) is clearly unitary. The CBH
formula (5.17) gives
This is clear because idt acting on the nlh term of the series gives Vs(t) multiplied by the \l(t) = e .»e~ '"e~ '
l
" T m
(lo)
(n — I)"1 term. (It is understood that the 0th term equals 1, the unit operator.) Also, the
boundary condition fi(-oo) = 1 is satisfied if the integrals converge. which can be substituted into (4). Firstly
A very useful and important notational tool which simplifies the appearance of (11) is
the cktvnotogical product. The commutator [Vj(J), V/(t')] does not vanish in general, and
the order of factors in (11) is therefore important: the factors, V[{tj), are arranged from = aafn — a*a,n + nj(a'a — a'a) + i/3
left to right in order of decreasing time, £( > t? > .,. > tn. Let us define the chronological
products. This must equal the operator

-iVl{t) = tJM{a <,-«. + „+<,«*)

Hence oc and 0 must satisfy

T(V,(U)V,{t2)V,{h)) = i
Of = -—J(£)elcJ
2u>
(12)
0 = iw(a*a - a*a) (17)

32 33
These equations are easy to solve,
where E is an infinitesimal positive real number brought in to make the integral converge
at ( —* ±oo. The inverse transform is
a{t) = — / dt'.JW )<•<"*'
Z^j J-x-
1 G(t) =
/"' ' C ) + c.c. (18)
dE
= ,1™ r (25)
In the limit t —* oo we have 2ir E'2 —

i [•*• , The Feynman Green's function arises in a particular choice of boundary conditions for
the integrated form of the equation of motion,
2u,' J-oc

Q(t) = Q\?(t) + Q[J(t) + i r dt'G(t - t')J(t') (26)

fi(t) - --^- r rf*' / ' dt"J(t')e^l''i"'>J{n (19) J-oo
where
Substitute these limits into (16) and we see, on comparing with (5.19) that the same
S-matrix is recovered, except, that we have now determined a particular value for the
previously arbitrary phase, 0. The operator (14) is completely well defined, including its
phase. QLlW = ^<^!J" (27)
There is a particularly neat way to incorporate the phase in a formula which will prove
(The significance of equations (23),..., (27) is explored in the exercises.)
very useful in the following. Introduce the step function,
d) Normal ordering (Wick's theorem)
(20)
In the expressions (16) for Q(t) and (5.19) for S the creation operators, afn, are
and write (in the limit t —> oo) placed to the left of the annihilation operators, ain. This is known as "normal" ordering.
It is a convenient arrangement for calculating 5-matrix elements, and we can improve
the appearance of many formula by introducing a notational symbol that means simply,
"normal order these operators".
1 roc
roc roc
,c Thus, for example,
: = : n o := n T a (28)
Notice that, in effect
f : [o, o+] : = 0
4uJ J-
Operators "commute" inside the normal ordering symbol. Hence, in particular

here G is defined by

2ui

(22)

This function is known as the ''causal" or "Feynman" Green's function. It satisfies the ''' : (29)
differential equation
(d? + wi)G{t-t') = -i6{t-tl) (23) Finally, on collecting the formulae (14), (21) and (29) we obtain the fundamental result
which relates chronological products to normal products,
and its Fourier transform is given by
e> JdU(t)Qir.M\ _. e-±fdtdt'JWO(t-t')J(?)

G(E) = f dtG(t)e'
J-oo This result is known as Wick's theorem. It can be generalized directly to systems of more
(24) than one oscillator and plays a central role in perturbative calculations in quantuin field
- Ul2 + l£ theories.

34
35
Exercises 7 Free Fields
To begin the study of quantum field theory we consider the free scalar and Dirac spinor
6.1) Verify that Git) — e~'-J"l/2uj satisfies the differential equation {&f ' = -i6{t). fields. The Lagrangians are bilinear, the equations of motion are linear and it is fruitful
6.2) Compute the Fourier transform G(E). To make the time integral converge it is to discuss them in the language of oscillators, We therefore define plane wave expansions,
necessary to give u an imaginary part, iniLj < 0. Separate the integral over t into isolate the oscillator variables, and construct creation and annihilation operators for the
two parts (—oo,0) and (0. -Hoc) and verify (24). various quanta of these systems.

O.IJ) Another Green's function has the Fourier representation a) Lagrangians

dE The Lagrangians of interest in relativistic field theories are always local. This means
GI(t)=\imoj" they can be expressed as the integral over space of a local quantity, the Lagrange density,
2ff (E + xef -
which is itself a function of the field variables and their partial derivatives,
Evaluate this integral.
0.4) Verify the solution (26), i.e. show that it satisfies a) the equation of motion, b) the = Jd3xC(4>,d<t>) (1)
boundary conditions at t —> ±oo. (Use the result (5.10))
Two cases are of particular interest.
(i.5) P r o v e t h e formula, (0 in\T(Qm{t}QJt'))\Q in) = G{t - t').
1) Real scalar

(2)

where d = (9T, dy, (),) denotes the spatial gradient operator, and M is a positive parame-
ter, the rest mass. We employ units in which the velocity of light, c is unity.

2) Dirac spinor

L =
(3)

where d^ = (dt, d^, dy, dz). The field variables t> are in the form of a column with 4 complex
components, whereas v' is a row with 4 components and the 7" arc 4 x 4 matrices. There
are many equivalent representations of these matrices. They are required only to satisfy
the Dirac algebra,
{y,7"} = 2»r (4)
where tf is the metric tensor. Its non-vanishing components are

r," = -rfx = -vm = ~j)" = 1

The field -ip is known as a Dirac spinor, and !/>, its "adjoint", is obtained from ip by
hermitian conjugation,
ti = ^ + 7o (5)
Here and in the following we shall use the notation,
x° = t, x' = x, x'1 = y, x3 = z

36 37
and we shall use the summation convention.
where the integral extends over all t-space. However, ir, is often convenient to confine the
system to a large box of volume V. imposing periodic boundary conditions on the surface
of the box. In this version the integral (9) extends over the interior of the box and k
takes discrete values so that e~'-- has the appropriate periodicity. Then the inverse (10)
Indices are raised or lowered by means of the metric tensor, is replaced by

On substituting (10) into (2) and integrating over space we obtain

where
where i)ttv is the inverse, in the sense of matrices of, rf" ^(k)2 = k' + M2 (12)
which will be recognized as the Lagrangian for an infinite collection of independent oscil-
lators. Of course Q(k, t) is complex but this is a minor complication. Since <p(z, t) is real
The classical equations of motion derived from (1) are partial differential equations. it follows from (9) that
Thus Q(Lt)* = Q(-k.t)
6 / dtL = fd4xSC We could substitute

where Q^ is real and even, and Q? is real and odd. thereby obtaining L as a sum of real
oscillators.
dC The equations of motion for the oscillators in the Heisenberg picture arc
(13)
and are solved in the usual way,
00 '' \d[du<b (6)

Hence (14)
(a2 + M2)0 = o (7)
(i 0-m)il> = 0 (8a) where a(fc) and a{k}+ are annihilation and creation operators,
(8b)
2
where 9 = rfO^d,, = d'g — d'f — 0$ — 0%. These equations are covariant with respect to
Lorentz transformations but we shall ignore this feature for the present. [a,{k}+,a{k'y] = 0 (15)

b) Plane wave expansion for the scalar These rules follow from the canonical quantization of (11) which gives equal time com-
mutation relations like
The Fourier transform of <f>(x, t) is defined by the space integral

Q(k, t)= j d3xip{k, t)e~1-- (9)

(In the box version of the theory, Vbkt1 would replace (27r)3<5j(fc — k'} in these formulae.).
If there is no space boundary then the integral extends over all space and the wave vector The ground state is defined by
k is continuous. The inverse transform then reads a(i)lO) = 0 (16)
The eigenstates of energy are then obtained by acting with the creation operators. The
c.Oe 1 " (10) iV-quanta states are

38
39
corresponding to the energy and the equations of motion are
ki k.v) = (18)
where Ea is the ground state energy. What is it? Using the box version, the Hamiltonian = -i+(k,t)H{k) (25)
takes the form
To quantize this system we impose anticommutation rules like (4.27),
H
= h
(19)
! + , 4>(k',t)+} =0 (26)
+
on using the commutator, [a(k), a(k') ] = %j{k)V6w, Hence To solve the equations of motion (25), obtaining the Dirac analogue of the general solution
(4.28), it is necessary first to diagonalize the 4 x 4 matrix, ff(fc).
Since H(k) is hermitian, its eigenvalues must be real and its eigenvectors orthogonal.
To get an idea of the structure we adopt a particular representation of the Dirac matrices.
which is just the sum of the ground state energies of the individual oscillators in the In terms of 2 x 2 matrices,
system. This sum diverges. We must therefore choose a more reasonable Hamiltonian for
the quantized field. We choose 0 1 0 £ (27)
7o =
1 0 -2 0
H
= (20)
where
0 -i
-i
It is easy to verify that the matrices (27) satisfy (4). Then (23) gives
in the limit V —> CJC. From now on, Eo = 0 by definition.
H(k} =
-k-a. m (28)
c) Plane wave expansion for the apinor k • a

The Fourier transform oitli(x.,t) is defined by The eigenvalues E, of this matrix are given by the zeroes of the determinant, det(//(A) —
E). It is tedious but not difficult to show
i>(k.t) = f SxMx.t)e-*2-
Jv
det(H(k) - E2f
(21)
where w(ft) = \/k2 + m2 > 0. There are two eigenvectors with E = ui and two with
Substitute this expansion into (3) to obtain E = —w. To distinguish the two eigenvectors with the same energy we can use the
eigenvalues of some other operator that, commutes with H. For example, the "helicity"
L = I d xii'+ v iat + «7o7 • o — "i7ol viz., i) operator,
Jv - ' 1 / I r. \

(29)
(22) 0 k-s.
which has eigenvalues, ± i commutes with H(k). Thus, we have
where H(k) is a 4 x 4 hermittan matrix defined by
= ?o7 ' k + ™7o ( 23 ) H(k)u(k,\)
The Lagrangian (22) will be recognized as a sum of fermions of the type (4.23). The H(k)v(-ky\) = -w{k)v(-k,\) (30a)
Hamiltonian is = Xu(k,\)
1, (24)
V •
= Xv(-k,\) (30b)

40 41
where A = ± | . (We have used —k as the argument of the negative energy eigenvectors and define the ground state by
for a reason that will become clear shortly). The four eigenvectors are orthogonal and we
choose to normalize them such that = 0 and (36)
+ +
Excited states are obtained by acting with b(h, A) and d(k, A) on the ground state.
These operators create "particles" and '"antiparticles", respectively.
!.'(-£, A)+u(~fc A') = (31) d) Covariant expressions
The completeness of these eigenvectors is expressed by
By substituting (14) into (10) and making the change of integration variable, k —» —k,
£ (u{L A)u(fc, A) + + u(-fc, X)v(-k, A)+) = 2w(k)l (32) in the negative frequency part we obtain

where 1 is the 4-dimensional unit matrix.

With these eigenvectors we can now give the general solution of the equations of motion
(25), This expression is Lorentz covariant. It is usually abbreviated to the form
1
ML t) = (37)
+
+d(-k, X) v(-k,X)eMt)t\ (33)
where k® = u{k) and kx = k^x* = wt - k • x. The same can be done with the Dirac
where b and d are time-independent operators. (Since tp(x, t) is complex, there is no spinor.
hermiticity condition on MLt). Hence b and d are independent. This contrasts with
the bosonic formula (14) where Q(k,t}+ = Q(-k,t).) They satisfy the anticommutation
rules,
{b(k,X),b{k',X')} =0

+ +
J -~^ (38)
) ,b(k\X') }=0 (34)
+
and similarly for d(k, A). Also b atiticommutes with both d and d . All these rules follow in the limit V —» oo. The 4-component spinors u and u satisfy
from (26), (31) and (33). (^-m)u(A,A) =0, (fi + m)v{k,X) = 0 . (39)
On substituting (33) into the formula (24) for the Hamiltonian, using the eigenvalue
equations (30a) and the orthogonalities (31). we obtain at ( = 0,

This expression suffers from the same divergent ground state problem as the bosonic
expression (19). We therefore adopt the simpler form

= T? E E kKk, A)+6(i, A) + d(k, X)+d{k, A)) (35)

42 43
Exercises 8 Green's functions and Wick's theorem
7.1) Give the box version of (11) and apply canonical quantization to find the equal time Having described the quantization of free scalar and spinor fields, our next step is to
commutation rules for Q(k, t). From these derive the box version of the commutators consider their response to external forces. This work is a direct generalization to systems
(15). with infinitely many oscillators of the analysis given in Lectures 5 and 6.
7.2) Verify (39) using (23) and (30a). a) Fieids with sources
7.3) Use the Dirac equations (30a) and completeness condition (32) to prove the formulae In Lecture 5 we characterized the forced oscillator by adding to the Lagrangian of the
free oscillator the term J{t)Q. This is easily generalized to the field systems. To the
5 > ) = > i : + m and £>(/t,X)v(k, A) =fi- m Lagrangian (7.1) we make the addition,
:i
£ + Jd x(.J4>- (i)
where J(x) and T)(x) are arbitrary functions of x and t representing the external forces
acting on boson and fermion oscillators, respectively. In the field theory context these
modifications are known as "source" terms. They cause the equations of motion (7,7) and
(7.8) to become inhomogeneous
(d2 =J

= 7} (2)

These equations of motion can be treated by the Fourier method, writing

etc., so that
k,t) = J(k,t)
(3)
By integrating these equations one finds

(4)

where kg = yls1 + M 2 , and analogous formulae for b and d.

iJ
*, A) = dln(k, A) - i Id4xeikIfj(x)v(kt A) (5)

where fc0 = y^ + rn7. There is one qualitatively new feature in the formulae (8.5).
Being fermiouic quantities, the operators bm and bolit must satisfy anticommutation rules.
This means that the source field r?(x) cannot be an ordinary function. It is necessarily a
fermionic quantity and it too must satisfy anticommutation rules, viz.

(6)

44 45
Such quantities, which anticornmutc with themselves and all other fermionic quantities, where u = \flc2 + M2. Use this in (9) to obtain
are known as "antkommuting (-numbers".
On the basis of (8.4) and (8.5) it is possible to construct the 5-matrix and give the field
theory version of Wick's theorem (6.30). Bui first we consider the field theory analogues
of the Green's function. G(t - t').

b) Causal Green's functions

(10)
One way to define the causal Green's functions is through the integrated form of the
equations of motion (8.2) with appropriate boundary conditions, This shows that AF is a sum of independent oscillator Green's functions like (6.22). The
result (10) can be put into a manifestly covariant form by substituting
<P(x) = 4>£ x'A^x - x')J(x')
2LJ i^oo 2?r B-<J1
jd*x'SF(x - x')r,(x') (7)
to obtain
where the superscripts (+) and (-) indicate positive- and negative-frequency parts, re- --ik(x-x') (11)
spectively. The Green's functions, A F and Sp must satisfy the differential equations,
where £ is a positive infinitesimal and
(9 2 + M'J)AF(x - x') = -164(1 - x')
k7 = k k* = k2 -k2
{i P-m)S,.ix-x')=i8t(x~x') (8)
k{x - x') = ^ ( x " - 2-'*) = fcp(t - t') - k • ( r - x')
so that, (7) is consistent with (2) and It is clear from the Fourier representation (11) that A F satisfies the differential equation,
(d2 + M2)Af = —16. However, it is instructive to show this also from the definition (9).
(<92 4- M2)0m(x) = 0, etc.
This involves the commutation rules (at equal times)
In addition to the differential equations, A F and SF must satisfy appropriate boundary
conditions according to (7). For example, if J(x) vanishes outside a finite region of =0. (12a)
spacetime, then at sufficiently late times, (j>(x) = 4>out(x) and (7) implies (12b)
}
<p|: (x) + i JdVA F (x - x')J(x') (late times) which follow from canonical quantization. We also need the formulas

This equation must be the Fourier transform of (4). In particular it implies that Ap(x —1')
can have no negative frequency part when x" is in the future of i'**. This kind of argument da0(x'o - xa) = ^6(x 0 - i o )
is enough to determine SF and A F completely. However, there is a very simple and largely which can be verified by integrating x0 in the neighbourhood of x'o. Now, according t o
automatic way to implement these arguments. It is based on the formula (6.22) for the
(9)
oscillator Green's function. Write
9 0 A F (x - x') = <0|r(cW.(x)<Kx'))|0}
= (0|T(<Kx)<Hz'))|0) but the commutator vanishes by (12a). Next,
= 8{xa-x'<>){Q\<l>{x)<t>(x')\Q) + - x') = 6(xa - x o
+0[x'a - xo){0\<t>(x')<t>{x)\Q) (9) ii<{x - x'
where 8 is the chronological step function and |0) = |0in), <i>{x) = 4>in(x). Substitute the - x'
plane wavo expansion (7.37) so that by (12b) and the equation of motion. But the space derivatives can be taken outside
the T product. Hence the equation is verified. The same procedure gives the fermion
Green's function, Sp. Write

_Le-i*(i-i')
- / (2TT) 3 2W = 8(x0 - x'0)(Q\t(x)4>(x')\0) - 6(x'o - (13)

46 47
Note the minus sign in the definition of the T product. It implies, in effect Exercises

8.1) Prove the formulas (8.5). Do this by the following steps:

Now we have
a) Using the orthogonality (7.31) of u and t'. show

where we have used the canonical equal time rule lim (e-^v+(-t\U>(k,t)) =d(n{-t,
t—••'OQ \ /
(14) and similarly for &ou( and d+Bt
together with the definition ij) = i)+io- It follows immediately that b) Using properties (7.30a) of H(k), defined by (7.23), show

(i p- m)SF(x - x') = i8i{x - x') (15) t)) = eitMu(k, A)(j70fib - 7 • k - m)i>(k, t)

Using the identity t)) = e l u " s ( - t \){iyad0 - 7 • k - m)i>{L 0

(* P-m){i jS + m) = -(d1 + m2)
c) Integrate t, using the results a) to obtain
which can be proved using the algebra, {7W, 7"} = lif", we obtain
W £ , A ) - h=i(&, •*) = - ^ dte^uik, \)(if°dQ - j • k - m)4'(k, t)
SF(x-x') = (i P + m)AF(x - x')
x-x"\ 1
J5-
i

This is usually written out _> in _ , - J

1
2 g-itd-i ) (16) d) Use the equation of motion
- m+
8.2) Lorentz invariance implies that Ap(i) is a function of x2 = t2 — x*. Show that for
which clearly satisfies (15). spacelike x^, it can be represented by the Euclidean integral
c) Wick's theorem
/ fd"x
t 11 jfc(iI(i

The discussion of Lecture 6 can be repeated now for each Fourier component to com-
pute the operator Q(t) and take the limit t —> cc, obtaining S. The result is
where k2 - fcf + k% + kj. + kj and xt = ±it is real.
(17)

and similarly for the Dirac field,

T
= : e'J I'l'+'l") . f, J 1 1 i ,F x x -ox (18)

We shall not give any further discussion since nothing new is involved. We remark only
that these formulae can be used to generate relations between chronological and normal
products by expanding in powers of the sources and comparing coefficients. Such relations
are used in the perturbative computation of Green's functions and 5-matrix elements in
interacting field theories.

48 49
9 Interacting fields state of the Hydrogen atom among the in- and out-states but. if the interaction vanishes
for ( -» ±oo, the atom would fall apart. In fact this problem too can be solved although
Field theories of real physical interest are characterized by Lagrangians which include it is beyond the scope of this course.
terms of higher order than quadratic and which therefore give rise to non-linear equations It is usually assumed hat the ground state is unique and has zero energy, H\0) = 0.
of motion. In general, they do not include time dependent parameters or external forces:
This implies that the in- and out-vacua are represented by the same vector.
the interactions conserve energy. We discuss some of the main features of such theories
and. in particular the relationship between Green's functions and S-matrix elements, the
(6)
so-called '"reduction formulae".
The same is true of 1-particle states,
a) Energy conservation
\k in) = \k out) (7)
In the absence of time-dependent parameters or external forces the Hamiltonian is
independent of time and the observables have the characteristic Heisenberg picture de- Interactions are between particles - if only one is present, nothing happens.
pendence, In the following lectures we shall use the formula
Q(t) = eltHQe-"H (1)
Their matrix elements between eigenstates of energy are time-dependent only in phase, 5 =

(8)
We often meet operators in the form of integrals over t, e.g.
where Vj(t) = V(4>in,ipin} and verify that this operator has well defined, energy conserving
matrix elements in the limit e —> 0 and that the assumptions (G) and (7) are satisfied
F = j°° dtQ(t) (3) order by order in the perturbation expansion.
The matrix elements of such operators have a particular structure, b) Green's functions and 5-matrix elements

(a\F\8) = The only Green's functions considered up to this point are the causal functions A/,-
- E,i){a\Q\(3) and SF. These can be generalized to functions of many variables, which contain all the
n (4)
dynamical information in the system, including the S-matrix. In order to motivate this
i.e. they vanish unless Ec, = Ed: they conserve energy. The S'-matrix is of this type. But generalization we shall start by developing the "LSZ reduction formulae".
there is a complication. In Lecture 6 we obtain 5 as the limit t —* oc of an operator, Let qa(t) be a real bosonic variable in the Heisenberg picture. It obeys the equation
il{t), that satisfies the equation of motion (1.13)

WtQ(t) = V,(t)Q(t)
where V((jj,... ,< contains cubic or higher powers. If V vanishes for t —> ±oo then
where we have
u ,-HHo
V,{t) = c ""Ve
and the initial condition lim, M £l(t) = 1, is imposed. For this to make sense we should
have V/(t) —* 0 for t —• ±oo. If the time dependence of Vj comes entirely from Ho, then it
(10)
will not vanish in these limits. Its matrix elements between eigenstates of H$ will oscillate
like (2). This is a very non-trivial technical problem that we shall evade by making the
consider the integral
replacement
V -» Ve-' |(i (5)
r did, U"°\ldt +wn)T{qa[t)q0l(t1). . . q^(tn
V
(11)
where £ is a small positive quantity. It serves to switch off the interactions at t —* ±oo. J-OQ

This has two obvious consequences. Firstly, energy is no longer conserved and, secondly, where *i, . . . , ( „ are fixed. Being the integral of a total derivative it is easily evaluated to
stable bound states are lost. The first problem is minor, we can take e —» 0 at the end of give
the calculations and recover energy conservation. The second is physically serious. For fo ( « i ) . . . ) - T(qg, { t j .. .)aQ]171
example, in a theory of protons and electrons we should be able to include the ground

50 51
because, for sufficiently late ( the T-product puts qa(t) on the left while for sufficiently All S-matrix elements in this system are therefore expressed as Fourier transforms of
early t it must be put on the right. On the other hand, carrying out the differentiations the functions
in (10) gives
If the qa{t) are Fourier components of p(x, t), i.e.

q3i (t,). (12)

Similarly, from the formulae then

aoul(k)T{...) - T ( . . .)ain(k) = (15a)

we obtain and
- I dte-"*°'i[e? + w3a)T ((«o( (.. .)a in (fc) + - (., . =)
= oi ou ,T(.i ft (t 1 )...)-T( ( i A (t l )--'-Xi« (13) + M'2)T(4>{x,t)...) (15b)
The creation and annihilation operators for bosonic quanta satisfy the usual commutation
These formulae are examples of the "LSZ reduction formulae". They are used to express
rules.
matrix elements in the basis of asymptotic states, (a out\F\@ in), in terms of the ground
[aa,a0} = 0, [«Q,aJ] = W o / 3 , [<,a+] = 0
state expectation values, (OlTtf ^(ii)^(ar 2 ).. .)|0).
Consider an ^'-matrix element To illustrate, for vacuum and 1-particle states,

( f i i f l a . . - out|/3i./3a • • • in) = (0 out\aauaut • • • a£, i n • • • |0 in) (kout\Q) =

where, for simplicity, we suppose that the Q'S never equal any of the 0's. Then we can {k out | U. in) = (k in\k' in) + jdlxelkxi(32 + *' in)
write
Here we see that, because of the stability conditions (6) and (7), the integrals on the
{«, .. . out \&i ...m) — (a2 . . . out | a ai<ml | pi ... in)
right hand side must vanish. It will be seen, at a later stage when we discuss translation
= {a-2...out\ aa,ln invariance, that (Q|0(:r)|O) is independent of x while

(0\<t>(x)\k'in) =

and, in the same way, and using (12) Hence the stability of vacuum and 1-particle states reduces to the conditions,

{O|0(O}|O)=O and (0\4>{0)\k'in) < oo

(Q 2 ... out | <JQ1(tiPi • • •*") = jdtiCi^LH{<3l+uil2){ai... out \ T(qa,(U)qa,(t2))\&)
Reduction formulae can be easily developed also for the Dirac fermions
By repeating this process all the out-state operators can be reduced away. Then we can
use (13) to reduce out all the in-state operators until we are left with the ground state.
Thus,
- m)T{ip(x)...) (16a)

... f
= jd"xe!kxT((ip(x).. .)(i "p +m)v(k,X)-i (16b)
Note the sign of the frequency that distinguishes outgoing from incoming quanta. If we
relax the assumption that Q'S are different from /?'s then there will be more terms of the
The — sign applies if T(...) includes an even number of fermionic operators, the +
same kind but with factors of 6a^ -
sign for odd numbers.

52 53
 .*a> x m-

The vacuum expectation values, Since £/(£)' = U(t)~l and HH{tY = Hu{t), it follows that

idtU(t)-' = ~HH{t)U(t)-1
among which Sp and Af are special cases, are known as "Green's functions" = -V(t)-lHs(t) (22)

c) Perturbation series for Green's functions The operators il(t),i\in(t) and i1ml(t) all satisfy the same equation,
The Green's functions carry more dynamical information than do the S-matrix el- idtn,(t) = Q,{t)VH{t)
ements. They can be used to compute matrix elements of many physically interesting
operators such as the energy-momentum tensor, Tli,,(x), and the electromagnetic current
0 (23)
vector, J)i(x)- It is therefore important to set up a perturbative procedure for calculating
Green's functions. To do this we return to the unitary operator, Q(t), that relates the where Vw(t) = V{QH(t),t). They are distinguished from each other by the boundary
Heisenberg and interaction pictures. The discussion here will generalize the treatment conditions,
given in Lecture 0.
Firstly, remember that the perturbative method will be useful only if the Hamiltonian £2,(0) = 1
has the form
H = H°(Q,P) + V[Q,t) (17) = 1 (24)
where H° is simple and V is small, Typically, H° will be a bilinear function of the canonical
variables, Q and P corresponding to a set of independent simple harmonic oscillators. The We shall always assume that V(t) vanishes for t —> ±oo so that fl(i) —» constant and the
perturbation, V, will be a polynomial, pQ3 + XQ* + ..., with small coefficients which may conditions (24) make sense. Using the relations (18) we can write the equations (23) in
depend on time. the alternative form
In the following we shall work with the Heisenberg, Schroedinger, and several different
inter action pictures. These "pictures"' are related to each other by time-dependent unitary idSMt) =
operators. Operators in the various pictures are distinguished from each other by a idfiut) = v,n(t)nln{t)
subscript notation, Qn,Qs,Qi,QintQnut-
Qn(t) = - 1 QsU(t) where V,(t) = V{Qi(t),t), etc. From (23) or (25) we can find the equations of motion
obeyed by the various interaction picture operators. For example,

idtQin(t) = !<9((ni'i

and likewise far Pn{t), Ps, etc. The Hamiltonians Hnt Hs, # / , # ; „ and Hcut a r e defined
in the same way. For example.

HH(t) = HD(QHlPH)+V(QH,t)
Hin(t) = H°{Qin,Pin)
(19) (26)

etc.
where H°n = H°(Qm(t), PiT1(t)). This operator is actually independent of t because
The Heisenberg equations of motion always take the form

[QH{t),HH(t)\ (20)
Since Hfn is independent of t we can formally integrate the equation of motion (26) to
while the Schroedinger picture operators, Q$, are always independent of time. This write
implies that the unitary operator, U(t), must satisfy the equation, Qinit) = e!tH?" Qm{0)e-itH* (27)

= U(t)HH(t) Since Hfn is "simple", i.e. a collection of harmonic oscillators, it is not difficult to calculate
its eigenvalues and eigenvectors as in Lecture 3. In other words we can understand the
= Hs{t)U(t) (21)
formula (27) in all its details. The same is true for the other interaction pictures, Qi and
54
55
QolU. In all of these pictures, the dynamical variables are "free" in the sense that their Elimination of i1s(oo} here gives the relation
time development (27) is given by independent, uncoupled, unforced, harmonic oscillators.
All the information about interactions is contained in the unitary operators, i~l(t). 5= (33)
To compute Q(t) we can use perturbation series since V is small. Thus, as explained
in Lecture 6, the series solution of (25) for Qin is given by We shall not be using them in this course but the operators £lin(0) and f2out(0). known
as Moller operators, were often used in the older literature on scattering theory.
„ , ft fU
There is another way to write the series solution (28),
WO = E H ) " / dtJ f dt2...

f-oo
/ co
J-co
(
*;•••/'
J-x
*»r{V,n
(34)

where T is the antichronological ordering. To see this one must verify that (34) satisfies
(28) the differential equation (25) and the boundary condition £!jn{-oo) = 1. The first is
almost obvious,
This clearly satisfies the appropriate boundary condition, fi,n(—oo) = 1. Similarly, we
have for t > 0,
idtU,n(t) =
n,(t)=T(e-&dt'vdl')Y t > 0 (29)
On the other hand, for t < 0, we should write
= vin(t)nin{t)
) (30)
We have used the fact that t < t' in (34) so that Vin(t) must stand to the left of all
where T denotes the anti-ckronological ordering: earlier times to the left and later to the operators Vin(t') in the antichronological ordering. This may seem obscure but it is easy
right. Finally, we have, for all i, to justify by expanding the exponential to write

(31)

which satisfies (25) and the boundary condition £2om(oo) = 1. £ i" r du r dt2... r *„ vin(t,).
From (23) it follows that
and then differentiate with respect to t.
The boundary condition is somewhat less clear. If (34) is correct this condition would
imply
= o
The product. ilin(t){loui{t}~1 is independent of time. We define this product as the i/r. *'"»(•'A
S matrix, i.e.
s = nin(t)nml(tr
because taking the hermitian conjugate changes the sign of i : % —> -i, and reverses the
order of operators: T —> T. Hence, if (34) is true then S is unitary. On the other hand,
(32) (34) implies fJjn(+oo) = S and this in itself is a satisfactory boundary condition. Since
where in the last two equalities we have used the boundary conditions, fi^oo) = 1 and (25) is a first order differential equation we can obtain a unique solution by fixing its value
fl m (-oo) = 1, respectively. at t = +oc. Hence (34) is correct and S is unitary.
Similarly, the products Q.m{t}$li(t) ~l and Qmtitftlilt}'1 are also independent of time. Now, using (34) and (28) we can go back to the formulae (18) to write
Using the boundary condition Qi{Q) = 1 we obtain Qnit) = n m ( c )-'Q m (on m (t)

= n,(oo) = S-lT(Qm{t)S) (35)

56 57
•aw t

where, again, the formal manipulation with chronological exponentials can be justified Exercises
by expanding and examining them term by term. The steps used to obtain (35) can be
repeated for chronological products of two or more Hiesenberg operators to obtain the
general formula 9.1) Old-fashioned perturbation expansions for S'-mntrix elements can be obtained from
the series
T{QH(U)QH{h)...) = S-iT(Qln(h)Qm{t2) • • • S) (36)
(0\S\a) - Y.(-i
This is known as the Gell-Mann Low formula and it is one of the most important in quan- o
tum field theory. It plays a central role in many perturbative calculations. In particular,
by integrating ti, . . . , ( „ . Write
the vacuum expectation value of (36) gives the perturbation series for Green's functions,
V(t) = e-l\t\eitHaV(,-itH0
(Q\T{QH(t1)QH{t-i)...)\Q) =
= (0\S~lnQia(tl)Qin{t2)...S)\0) and assume that \a), (0\ are eigenstates of H§. Proceed as follows:
= {0\S-lT (37) a) Verify the formula
which is to be expanded in powers of Vm. Usually the vacuum is stable so that

Ea - Ho + K '
The operator V,n = V(Qin{t),t) is expressed in terms of the free operators Qin and this
makes it possible to calculate its matrix elements. Rules for computing the terms of in t h e limit E —* 0.
the series (37) as functional of A F and SF, using Wick's theorem (8.17), (8.18) will be b) I n t e g r a t e successively („, ( n _ i , . . . , (i t o o b t a i n
developed in Lectures 13-17.
Although we shall not pursue this further, we remark here that perturbation series for 1
{0\S\a) = {0\a} - 2m6(Eli - En)(0\{V • V +
Green's functions can be developed using either Qm or Q^t and T or f ordering. For Ea - Hn + if
example (35) is only one of four essentially equivalent representations for the Hiesenberg 1 1
field: +V- -V-.-V + ...}\a)
Ea~ Ho + ie' Ea - Ha + ie
Quit) =S c) Show that the series can be formally summed,
= T(QoM(t)S)S
{0\S\a) = {P\a) - V + feV}\n)
_ ff
= S T(Qml{t)S-') (38)
where H = Ho + V is the total Hamiltonian
In practice we use only the first of these. d) Define the state
— V\a)
' Ea-H-
Show that this state is given by the limit

|* n ) = t_liniofi(0,t)|£»)

and verify that it is an eigenstate of the total Hamiltonian,

This method works only if H and Ha have the same spectrum of eigenvalues, Ea
so that fi(0, —oo) is unitary.

58 59
10 Symmetries in quantum field theory I (Poincare Notice, in particular, that to every transformation there is a reverse transformation that
cancels it,
Group) (A,6)-'(A,b) = (l,0) (7)
where the inverse is given by
The equations of relativistic quantum field theory are covariant with respect to inhomo-
geneous Lorentz transformations (Poincare group) and often also with respect to other (8)
transformations associated with Lie groups, U(l), SU(2), etc. We review some of the main
1 12 V
features of these transformations and discuss how they are represented in. the theory. with {A" ) VA p = 8$. Prom (3) it is easy to see that

a} Poincare group (A"1)*, = W V ^

= A," (9)
The inhomogeneous Lorentz transformations are linear,

3? -» x" = A" „ x" + 6" (1) b) Scalars, vectors, tensors and spinors

where A and 6 are independent of x. The metric, defined by the quadratic form The main question is to determine how the field variables, (j>(x), ip(x), etc. respond
to changes in the coordinate description. A given point in spacetime is characterized by
^dx^dx" = (dx°f - {dx'f - (dx1)2 - {dx3)2 (2) the coordinates, i*1, in one system and by x* = AJJ + f in another system. A scalar
quantity is one whose numerical value at the given point is independent of the system. In
is invariant. This means that the matrices A must satisfy one system this value is represented by the function, 4>{x). In the other it is represented
by another function </>', such that
h" p = IJ V (3)

The "identity" transformation, i " = xu corresponds to A£ = 8£ and 6^ = 0. Infinites-

imal transformations are "close" to the identity,

A" „ = «J + E" „ (4)

M
where e£ and b are small quantities. Since (4) must satisfy the constraint (3) it follows for all x. Next, a vector quantity is represented by a set of functions which undergo a
that Efu, = ri^ye* „ satisfies linear rearrangement when the coordinate system is changed,
£ji* = -£i/)i (5)

The successive application of two transformations is equivalent to a single composite

transformation. This is the group property, For example, the gradient of a scalar, 4>ix{x) = ) , is a vector because
1
x"->x' ' = A" „ x" + 6"
x" — x"» = A'" „ x" + b'»

= A"
In general, tensors of rank N transform according to
where
(12)
Notice for example, that the product of two vectors is a tensor of rank 2. For such a
product we can also make a scalar by "contracting" the indices,
This composition can be expressed abstractly as a rule for multiplying transformations,
= AM "A A "<t>u(z)Xp(x) (tensor)
) = (A", 6")
= (A'A,A'6 + ft') (6) (scalar)

60 61

r
 Finally, the Dirac spinor is a set of 4 complex functions tpa(x) for which the transfor- etc. This is a generalization of the concept of Schroedinger and Heisenberg pictures -
mation rule is which dealt with the behaviour of matrix elements under time translations, i.e. a special
<•>:') = a,, J(A)^M (13) case of inhomogeneous Lorentz transformations.
where the matrices u{A) are constructed from Dirac matrices. For an infinitesimal trans- On comparing (18) with (10) and (13) we find
formation, A = 1 -f £, they are given by
U{A,b)-l<j>(x)U{A,b) = <p(A-l(x-b))

o(l +"£) = (14} ix-b)) (19)

The operators, U(A,b) constitute a "representation" of the group in the sense that
It is not important to know the explicit form of a(A) for arbitrary A, but two general their multiplication obeys the composition rule (6).
properties are important. Firstly, they represent the group composition rule, in the sense
of matrix multiplication,
U{A', b')U{A, b) = E/(A'A, A'b + b') (20)
a(A>(A) = a(A'A)
In particular, the identity transformation, (1,0) is represented by the unit operator,
and. secondly, they are pseudo-unitary,
f(l,0)=l (21)
a(A)+-yo<i(A) = 7a (15)
Hence, for infinitesimal transformations we can write
where a + is the hennitian conjugate of a. Hence, the transformation rule for the conjugate
spinor V — V'+7o is
(22)
&{x') = ^(i)a(A)" 1 (16)
which shows that 4>{x)ip{x) is a scalar. to first order in the small quantities, 6 and £. The so-called "infinitesimal generators",
Another property of the matrices a(\) is given by the formula and J^, are hermitian because U is unitary. The infinitesimal version of (19) is

a(A)-V<z(A) = A" „ 7" (17) \4>(x), P»\ = id^(x)

y>(x),Jllu] = i{xlldv-xl,dtl)<i)(x)
which is easy to prove for infinitesimal A using (14). This formula implies that
M'{x)i (}il'(x) is a scalar. Mx), PJ = id^(x)
(23)
c) Unitary representations

The transformations of dynamical variables 0 - n J ' , etc., discussed above are "canon- where cr^ = ^[7^, -yj\ according to (14). These formulae are obtained from (19) by
ical" transformations, from the classical point of view. In the quantum theory such comparing the coefficients of terms linear in ¥> and e1"1 = -e"1'. Note that Pa is just the
canonical transformations can be realized by unitary transformations in the space of state Hamiltonian of previous lectures.
vectors. Corresponding to each Poincare transformation (A, b) there is a unitary operator, Some important properties of the generators, PH and J ^ , can be deduced from the
t/(A,6), acting in the Hilbert space, such that composition rule (20). Firstly, P^ is a vector in the sense,

= U(\,b)-l4>(x)U(A,b) (24)
= U(A,bylipa(z)U(\,b) (18)
Secondly, J^ is a tensor if b = 0,
etc, This is a way of saying that the change of coordinate system, as it affects matrix
(25)
elements, may be written alternatively as a change in the operators or as a change in the
state vectors, From (24) and (25) we can deduce the commutation rules

where

|a'} = t/(A,6)|a) (26)

62 63
by taking A and b infinitesimal. Almost all of the relevant properties of the representation is invariant. Now the equation (19) or, equivalently,
C/(A, b) can be obtained from a study of the commutation rules (26).
U(h.b)4>m(x)U(A, b)-1 = tt>in{\x + b)
(1) Spin and helicity
implies
Among the homogeneovis transformations (A, 0) we have the purely space rotations, , b)am(k)U(\, b}'1 = ain(M)e-iW> (30)
A° „ = 6°, or infinitesimaily, e°" = 0. These transformations are generated by three
operators so that am(k) is a scalar with respect to homogeneous transformations (A,0) and also,
M^J23, M2 = J3l, M3 = J]2 (27)
\Pl,,a,n(k)] = -kua,n(k) (31)
The commutation rules for those operators can be extracted from (26),
The fermionic case is somewhat more complicated because of spin. Firstly, (19) implies
[Mi, Mi] = iM3, [M2, M3] = [M3, Mi] = iM2 (28)

which will be recognized as the familiar commutators of angular momentum. Also, from
(26), these operators are conserved: they commute with the Hamiltonian so that
(29)
[P)i,dln(k,\)+] = +klldln(k,\)+ (32)
They do not commute with the linear momenta
exactly as for bosons. For the homogeneous Lorentz transformations we need

a(A)u{k, \) = £>(Ajt, \')DXX

etc. We cannot diagonalize Pf and M( simultaneously. However they do have some eigen-
states in common, those for which P;| ) = 0. Such states, where the total momentum
where D = D(k, A) is a 2 x 2 unitary matrix. Using this formula in the plane wave
vanishes are called "rest states". On these states we can diagonalize M2 = MjMj and M3
to define their "intrinsic angular momentum" or spin. expansion it is straightforward to extract the result,
Any particle with mass can be brought to rest by transforming to an appropriate (33)
(/(A, 0)Mfc,
coordinate system. Likewise a state with P; ^ 0 can be transformed to a rest state by
applying a suitable operator, (7(A,0). Its spin can then be defined by the eigenvalue
of M2. The operator P_ • M = P,hU commutes with P^ and we can diagonalize them and similarly for din(k, A).
simultaneously. The eigenvalues of ( P 2 ) - 1 / ^ . M are known as helicity.

e) Free fields

It is important to know the transformation properties of the creation and annihilation

operators defined by the plane wave expansions of 0, n (i) and il>in{x). Firstly, observe that
the integration measure d3k/2u/ is Lorentz invariant because we can write

where the factor 8(ko)5(k2 — M2) serves to restrict the 4-dimensional integral to the 3 -
dimensional sheet withfc0= uj(fc) > 0. Under a Lorentz transformation k' = A^ "kv we
have

because detA = ±1 according to (3). The factor 6(k2 — M2) is clearly invariant and it
restricts k^ to timelike directions for which 0(fco) is also invariant. Hence, the integral

64 65
 I I);. >- K | | j •MtiMiMMiM'Hfi"' i,

Exercises 11 Symmetries in quantum field theory II (Inter-

nal)
10.1) Write the infinitesimal Dirac spinor representative (10.14) in the form
Having described the inhomogeneons Lorentz transformations, (A,6), and the unitary
operators, U(A.b), that represent them in quantum field theories, we now consider the
transformations associated with internal symmetries. These are represented by unitary
where operators that commute with U(A,b).

a) Complex fields
Using the matrices (7.27) verify the following
The simplest internal symmetry consists in multiplication by a phase factor, e1Q. where
a is real, for example, the Dirac field p(x) could undergo the transformation
b)ffl(A)+ 7oa(A) = 7o, for A = 1 +e (la)
c) a(A}-'fva(X) = Atf "-fu, +
The conjugate V-' = </' 7o would then transform according to
10.2) The chiral spinors 4>i and ipn are projected from a Dirac spinor, ijt,
(lb)
tyL = 1—-—%\
- 7s , , = 1—-—ip
yj R
+ 75 ,
so that, both U)0 and xln /dtp are "invariant", i.e. unchanged. It is easy to find a unitary
operator, U[a), that will produce this effect,
where 75 = 'J7D7I7273- Prove the following.
a) Lorentz invariance: a(A)"'7sa(A) =75
b) i'(i /) -m)t = V>Lt M'L + V-'flJ fliiR - m{i)Lil>R + -H'RIPL), (2)
where ipL = V^7o and i{iR = !p^f0
We can write
10.3) Charge conjugation. Find an antisymmetric unitary matrix, C, that satisfies U(a) = ,•.-""> (3)
and prove that (11.2) is satisfied if Q is hermitian and if
C'17,iC=-7j (transpose)

The conjugate field, ?|!>c(x) is defined by

(4)
If $(x} is a free field then the annihilation operators, 6 and d must satisfy
Prove the following

a) r(x) = -ipr(x)C~l
b) I/J C ^ C = tpxi} and ty'l^lf = -ip"iuH>
Hence it is sufficient to choose
c) i'c\i J9 - m ) ^ c = V(i fi - rri]i> - v ^

d) i'l = Ci>l, WR = C¥j Q= fc, A)+rf(fc, A)) (6)

e) Eliminate ipR and TJ>R in favour of ipi and i>\ to express the Dirac Lagrangian
in terms of left handed fields. so that Q|0) = 0. The value of Q on a multiparticle state is therefore equal to the number
of d-quanta minus the number of 6-quanta, i.e. b+ creates a particle with "charge" -1 and
10.4) Substitute the infinitesimal version of L'(A,6) into (24) and (25) collect the first d+ creates one with charge +1. The two particles are alike in other respects (energy and
order terms to prove the commutator formulae (26). helicity). They are said to be "antiparticles".

66 67
 The real scalar field does not admit phase transformations like (11.1) but the complex and
scalar does. Again there is a particle-ant iparticle phenomenon. The plane wave expansion (16)
for a complex scalar field would be
The operators Qa, are generalizations of the simple charge, Q, introduce above. Their
(7) eigenvalues can be useful in characterizing states and transition amplitudes.

where a(k) and c.{k) are distinct annihilation operators satisfying c) Noether's theorem
To determine whether a particular group of symmetries is relevant, it is necessary to
examine the Lagrangian. If the Lagrangian is invariant then the symmetry is relevant.
[c(k),Q] = Indeed, one of the first tasks in establishing the general properties of a model is to find the
(8)
maximum group of transformations that leave the Lagrangian invariant. The number of
so that independent transformations is equal to the dimension of the group and to each of them
0) there corresponds a charge operator.
+ + The central idea is expressed in the fundamental theorem of Noether: in local field
so that a creates a particle with Q = - 1 and c creates an antiparticle with Q = +1. theories, to each invariance of the Lagrangian there corresponds a current density, Jh{x),
that is conserved,
b) Non-Abelian symmetries (sketch)
Many kinds of generalization of the above idea are possible. Usually we are given a The space integral of this density - if it converges - defines a charge
set of fields &(x), i = 1, - • •, N that transform according to a linear formula
=
Q = f d*x,P(xJ.)
<pi\X] —* u> I X I / L/ij(pAX\ ^ 1UJ
j that is independent of time.
where DtJ is an N x N unitary (or orthogonal) matrix belonging to a family which To prove the theorem consider a Lagrangian density, ,drp), invariant under the
represents some group, D,j = Dij(g), group of transformations (11.10),

D(g}D(g') = D(gg') C(<p',d<p') = (17)

Exactly as for the Lorentz group we can examine the infinitesimal transformations, Infinitesimally, this reads
0 =
D(l + £•) = 1 + isaqa (e™ = real) (11)
dC
where qa is an N x N hermitian (or antisymmetric imaginary) matrix and a = 1,... d
where d is the "dimension" of the group. It is characteristic of non-Abelian groups that dC dC
the matrices, qa, satisfy commutation rules of the form (18)

where 6<J>< is given by (11.10), (11.11), and summation over repeated indices is understood,
7
where the coefficients cag are known as "structure constants". As before, we construct (50s =
a family of unitary operators, U(g), such that
The equation (11.18) must be satisfied for arbitrary fields, 4>i{x), provided only that 64>i
:) (13) is given by (11.19). It is an identity. However, if 4>i{x) is a solution of the equations of
motion, then (11.18) simplifies to
Inflnitesimally they take the form
(20)
(14)
where the generators Qa, analogous to Pv and J^ of Lecture 10, are hermitian. They Since the infinitesimal parameters e° are arbitrary we define the currents
satisfy dC , , ,
r^. /"I ] h
If) (i*\\
(21)

68 69
 «'• W W ill I

and (11-20) shows that these currents are conserved as a consequence of the equations of Exercises
motion. This is Xoether's theorem. The charge operators are defined by

(22) 11.1) The orthogonal group S0(N) comprises the Ar x A7 real matrices, fl^. that, satisfy

The lime derivative of this operator R-[ = Rl and det R = 1

Q« = Infinitesimally, R = 1 + <• where c7' = - e . Define the infinitesimal generators,

Q.H = -Qga, by

Using the composition rule, U{R) U (fl'J = U(RR'). show that QaB is a tensor,
where we use Gauss' theorem to express the volume integral of divj as an integral of U{R-x)Qa3U(R) = Rai
the normal component of J_ on the boundary surface. If the volume includes all of space
then the surface integral should vanish, implying that. Q,, is independent, of time. (It and obtain the commutation rules
can happen in some cases that the boundary integral does not vanish. In these cases the
integral (11.28) fails to converge and the charge operator does not exist. This is known fiatQ^ - t>tt
as "spontaneous symmetry breaking" and will be studied in Part 2 of this course).
It follows from the canonical commutation rules (since dC/O(t>, is the canonical mo-
mentum, TTJ and the commutation rule (11.12) for the matrices gn, that, the operators Qn
obey the rules (11.15) and (11.16).
The proof of Noether's theorem is easily extended to include fcrmions as well as bosons.
For example, in the Lagrangian

L=

where 0 is real, we have the symmetry (11.1) and the conserved current is

Finally, we remark that Noether's theorem also applies to the group of inhomogeneous
Lorentz transformations, if the Lagrangian density is a scalar.

then there exists a symmetric tensor, ^""(i), that is conserved, dj}"" = 0, as a conse-
quence of the equations of motion. The component of 8"" are interpreted as follows
(/'"di^lx^dT* — energy contained in volume ifx
(i'udx'dx2dTt = momentum contained in volume d 3 i
0vldx'iis%dx° = energy crossing dx2dx'J in time di° (flux)
lfl(h:idTidxfl = momentum crossing dx'2dx3 in time dx° (flux)
Since $'l) — #01 we have energy flux equivalent to momentum density: this is a funda-
mental consequence of Lorentz invariance.

70
71
12 Cross sections It is useful to define a function T{x) by writing
Methods for computing the S-matrix will be considered in subsequent lectures. For the S = 1 + i f d'xT(x) (9)
present we shall assume that these matrix elements are given functions, and turn to the
problem of extracting observable quantities from them, the reaction rates. The main tech-
Comparison with (4), (5) gives
nical features to be considered here are energy-momentum conservation and the appro-
priate normalization of states, leading to normalized probability distributions for various
processes. T(x) = Cinl(4>in(x)) +
Before the interactions take place we typically have a 2-particle state,
so that
a in) = a in ( (1) idpTix) = [T(x), P.]
This vector represents a state which, in the remote past comprises two independent and Since [P^, Pv] = 0 it is easy to integrate these equations to obtain
freely moving quanta. In the course of time there will be interactions that modify their
motion (scattering) or, more generally, cause the production of new quanta. The final T(x) = e'plT(0)e-*Px (11)
state will be represented by a superposition of vectors like where Px = P^x1*. Matrix elements of T(x) between eigenstates of P^ therefore depend
on x only through a phase factor,

corresponding to states which, in the remote future comprise JV independent and freely {0 m\T(x)\a in) = (3 in\T(Q)\a
moving quanta. The amplitude for the transition between states (1) and (2) is the
5-matrix element, (fi out\a m). The probability for this transition is proportional to and, according to (9)
\{0 aut\a in)|'2. Our task now is to find the proportionality factor and define a correctly
(y3 out\a in) = (0 tn\S\a in)
normalized probability distribution for the various transitions.
= {0in\am)+i(2K)ibi(Pa-Pli){0i7i\T(Q)\olin) (12)
a) Energy-momentum conservation
The same kind of logic that led to the translation invariance of 5, expressed in (8),
In relativistic field theories the Lagrangian is translation invariant. This implies the also implies that it is Lorentz invariant,
existence of an operator, P^, such that
[S,J^] = U. (13)
(3)
Because the asymptotic states \a in) and {0 in | are constructed by applying covariant
We want to show that PM commutes with the S operator given by creation or annihilation operators to a Poincare invariant vacuum,

(4) = 0 (14)

where it follows that the matrix elements 0 in\T{0)\a in) have very simple transformation
(5) properties with respect to the Potncare group. The same is true for internal symmetries
The "interaction Lagrangian density", Ctnt, consists of the cubic and higher order terms (at least for those that are not spontaneously broken). If Qa generates such a symmetry
contained in the complete Lagrangian. Generally, it is a polynomial in the fields. From then
(3) it follows that \S,Qa]=0
[£m((0,n(l)). P»] = idpCinAfanix)) + O(s) (6) and
r
where the small term, O(f), is due to the damping factor, e~ ''l, in Cint. Hence Qa\0) = 0 (15)

b) Normalization

The same argument applies to each term in the expansion of (4). Hence in the limit In order to obtain a normalized probability distribution it is necessary to employ
e —> 0, we must have normalized states. Since it is much simpler to use plane wave states - eigenstates of
[S, PJ = 0 (8) momentum and energy - rather than wave packets, we shall confine the system to a box

72 73
of volume, V. with periodic boundary conditions. This destroys the Lorentz invariance
f We already know that in the limit V —> DO,
but, at the end,, we shall take the limit V —> oc and recover it.
The creation and annihilation operators in the box version satisfy the commutation
rule £k
{a{k}.a{k') + } = 2uiV8^
A normalized 1-particle state is obtained from |fc) = a(fc) + |0) by multiplying with Hence we arrive at the probability distribution,
(2wV)~''' i . Hence, the transition amplitude between normalized states in
i flu i ,fil'1 1
1 1 1 (20)
= (Jt', ... in) (16) \ " ' (27r
for N "identical" particles, which represents the probability per unit volume, per unit
2
time for a transition to one of the states in the specified range 3.
where u\ = kw = \fkj + M , etc. To obtain the transition probability we must square Now we face the problem of taking the limit V —t oo where dv clearly vanishes.
this amplitude. But here there is a problem: how to interpret the quantity This means that the original two particles in the state \a in) become less likely to have
any interaction as the volume grows. In order to see an event we have to increase the
((2TT)%(P Q - Pa)f = (2n)i6l(0)(2n)%(Pa - PB)
"intensity" of the initial state, i.e. take many copies of the 2-paxticle state. In practice,
because (5(0) is infinite. In the box version the problem is less severe because 6P j,^ = 1 experiments involve the collision of a beam of particles with momentum kt with another
beam comprising particles of momentum £ 2 . If the number of particles in each beam is
for p^ —Vv 0 otherwise,
proportiona! to V then the expected number of transitions per unit time, the reaction
rate, will approach a finite limit as V —> oo.

c) Cross sections
Only the energy conserving delta function is troublesome. The simplest way to solve the
problem is by temporarily restricting the system to a finite time interval, X, with periodic The density of particles in a monochromatic beam is characterized by a 4-vector J*1 =
boundary conditions so that 2TT4(P" - Pf) is replaced by T6paPa where P° is now a discrete pu1* where u? is a timelike unit vector, the 4-velocity, and p is the rest frame density. The
variable. In effect, we are making the interpretation components of 3^ are interpreted as follows:
J°dxldxidxi — number of particles in the volume dz1dx2dzi,
(2TT)%(0) = VT , (17) 1 2 i 0
J dx dx dx = number of particles crossing the area dx2dx3 in the time interval dx°,
etc. The components of the 4-velocity are expressed in terms of the 3-velocity v by the
the total volume of spacetime in which the system exists.
familiar relativistic formulas,
Then choosing a ^ p so that {/3 in\a in) = 0, the transition probability between
normalized states equals

1 1 1
VT(2ir)%{Pa-P3), (18)
Hence, in any frame of reference we have a simple relation between the number density,
J°, and the flux density, J\ viz.
where (27r)%(P o - P$) stands for ^
J' = v'J° (21)
The expression (18) clearly explodes in the limit T —• oo. What this means is that we
In the 1-particle state, (2tjjV)~1/2ain(k)+[Q), for example, we have
must divide it by T to define the transition probability per unit time, which remains finite
a s f - M {and Tipa Po —> 2KO{P^ - P$)). In fact, we shall divide also by V to define
and (22)
the "transition probability per unit volume, per unit time".
We next remark that the values k'^^,.-., tiN although discrete, are very close together
=v
if V is large. In practice what we want is the probability for a transition to any of the where v' = k'/u.
states in some range, d?k[... d3fc'N. Now, the number of states in the range d?k is given Since the distribution (20) represents the transition probability, per unit volume, per
by the formula unit time, from the initial state with one particle of momentumfciand one of momentum
3
V d*k (19) The factor l/Af! appears in (20) if the out-state particles are identical, |fci,ij,.,} = k'2k[ . . . ) , etc. We
V
(2_? (19) must count the number of independent states with momenta in the range d?k[ .. .d^k'^j. If the particles
ate distinguishable by some other property (mass, charge, etc.) then |*J,*2 • • •) # |fcj,*i • • •) and the
factor 1/JV! will not be present.

74 75
k-i we obtain the number of transitions per unit volume, per unit time, in the collision of limited resolving power and will not be able to distinguish between events for which the
two beams by scaling it up: momenta fep. ..k'N are too close. Hence it is a practical necessity to integrate dtr over
dN = NlN-1dv (23) finite regions of phase space.
where h\ and N2 represent the numbers of particles of momenta kx and k2, respectively, Integration on phase space, while conceptually simple, can become unwieldy for large
in the beams. Those numbers are each proportional to the volume, N. We illustrate such an integral for the relatively trivial case, N = 2, in the centre of
mass frame, fcl + k2 = 0. In this case phase .space has dimension 6, but, because of energy
and N2=-VJ?2 momentum conservation, the integrand is non-vanishing only on a subspace of dimension
?2) 2. consider the integral
The differential cross section, da, is defined by writing the distribution (20) in the
form k'.2 - fc, - k-
dv = ~2da (24)
^, -k\) (30)
where the factor /, introduced for convenience, is defined by the Lorenta invariant expres-
sion, where UJ\ = k^ + Mf, u/7 = yky + A/.|. In the centre of mass frame we have
(25) hence
It is straightforward to write this in terms of 3-velocities, w\du\ = u'2(L/2 = k'dk'

(26) where*;'= |*iI = l&il- Hence

~ U2)2 - (M\ X
V.1)

(One has only to substitute k0 = w, k, = unitforfc,,in (25)).

The number of transitions per unit volume, per unit time now takes the form = (ui + ui2)k'dk'
J
u o

Therefore
The volume factors are all accounted for and we can safely take the limit V —> oo. On
comparing (24) with (20) we obtain the cross section distribution, = k''2dk'd9.'
l Jy + J2)di1' (31)
da = (27
where dQ' = sin 8'dffdip' is an element of solid angle.
Because of the delta function in (30) it is trivial to integrate over u>[ + <JT Hence

This distribution is made entirely from Lorentz covariant factors: it is "frame indepen-
dent" .
For confrontation with experiment it may be appropriate to express the reaction rate
(27) in a frame of reference where v2 — 0, the "lab frame". It reduces in this frame to
dN = (29) JdW... (32)

The event rate is proportional to the incident flux density, |jZ(i)|, and the target density,
ja where k' is to be written as a function of uj\ +LJ2 by solving the equation ui\ +LJ'2 =
The result (32) is usually expressed in the compact form
d) Phase space k!
1 JQt
da =
The expression (28) is a distribution. This means that its rote is to be integrated over
some chosen domain in the space of the variables £, fc'N, "phase space". The region (33)
•dn'
to be integrated over is dictated by experimental arrangements. Detectors always have a

76 77
where {7") = (iiijITfOJIfc,,^), evaluated at kt + k7 = 0. In (33) the distribution is with
respect to the angular coordinates, fff'. of the vector k\. The matrix element (T) is a
Exercises
function of these angles and the total energy, u,'i +^>2. The factor / is given by (26) in
the centre of mass frame, ui\V_] + ui->v2 = 0, 12.1) Scattering from a heavy target. In the scattering of a light particle by a heavy one,
it can be a good approximation to assume that the motion of the heavy particle is
unaffected by the collision. Choose its rest hame. In this frame energy is conserved
but not 3-momentum. The S matrix elements take the form
• \u2\) =
(34) (!) out\a in} = i2irh(E3 - En)((i\T\a)
where k = |fc]| = \k2\. Finally Suppose |'i) is a 1-particle plane wave state \k) representing the incoming light
particle, and \3) is a general rnultiparticle out state [£',.. .k'N).
k'\{T)\2
da = 2
(35) a) Write an expression for the probability per unit time, dxv, of a transition to one
647T k s
of the states in the element, dJk[ ... <f fw. Use box quantization ami verify
where s = (u>i + W2)2 is a standard notation for the square of the centre of mass energy. that dw contains one power of l/V.
Since £, + k2 = 0 we can write b) Define the cross section by the formula dw = f,da where v is the velocity of the
incoming particle. Give the formula for the number of events, dN, in a beam
of flux J.
c) Suppose the out-state |fc') has only one particle. Derive the formula
which is Lorentz invariant. If the scattering is "elastic", i.e. M[ — M,, Ik'
k' = k and (35) is simplified accordingly. In (33) and (35) it is tacitly assumed that the
2-particle oui.-state involves distinguishable particles. Otherwise a factor of ^ should be
where u/ = itt, i.e. k'1 + M'2 = k1 + M2, and dil' is the element of solid angle
included. corresponding to d:>}c.

12.2) Decay of unstable particles. Suppose the in-state, \k), consists of one unstable
particle. The S-matrix element for the decay of this particle into a multiparticle
out-state, [ t j . . . k'N), is
(13 out\a in) = ji(2Jr)li44(fcj + . . . + k'%,)(k[ . .. k^\T\k)
a) Show that the probability per unit time, dw, for a transition to one of the states
in the element, d3k\ ... d3^,, is given by

dw = {2n)46i{k[

in the limit V —» oc.

b) Suppose the final state contains two distinguishable particles. Show that, in the
rest frame of the decaying particle,

where M = <*J[ + w'2 is the mass of the unstable particle.

c) If the final state particles are identical scalar bosons of mass m, express k'
in terms of m and M. Correct the result (b) so that it describes identical
particles in the final state. If the initial (unstable) particle has spin 5 and
.z-component of spin equal to A, give the angular dependence of the matrix
element (k'1,k'2\T\k, A). For which values of S is such a decay allowed?

78
79
13 Perturbation theory (I) b) Perturbation series
Having seen how reaction rates are expressed in terms of beam fluxes and matrix elements In Lecture 9 we showed that the Heisenberg Held. Q(x). could be expressed in terms
of T(0) we now consider how to compute such matrix elements. We first recall the of the asymptotic field, <&„ by the formula
reduction formulae of Lecture 9 that gives the S-matrix elements in terms of Green's
functions. We then turn to the computation of the Green's functions themselves. In t>(x) =
Lecture 9 it was shown that these functions are given by a series expansion in powers
of the interaction Lagrangian. To compute the terms of these series we employ Wick's and that, more generally.
theorem (Lecture 8) and show that they can be put into correspondence with Feyntnan
diagrams. ) -•••?) (4)

The operator S is given by the series,

a) Reduction formulae for bosons

Consider, for simplicity, a system with only one kind particle, a real scalar boson of 5 = T
mass A/. (At a later stage it will be relatively easy to make the generalizations needed for
(5)
systems with several particle types, including fermions). In Lecture 9 it was shown that
the matrix elements of S between multiparticle asymptotic states are expressible in terms
of the vacuum expectation values of chronologically ordered products of the Heisenberg where Cint(y) = C,nt(4>ln{y)) is a polynomial in ipm{y).
picturefield,<f>{x). Since the vacuum is required to be stable, S|0) = |0), it follows that the Green's
function is given by the series,
(fc, . . . out\k\ . . . in) = (fc;...tn|S|A;i
/''))!0) (0)
f2)... J dx\eu
)...^;)..-j|0> (1)
where it is assumed that the out-state momenta, k\,..., are all different from the in-state This important formula (and its generalizations to include fermions, etc.) defines the
momenta, fc[,.... Otherwise there would be additional terms containing ^(Jjj — k'j). See perturbation series on which the rest of this course is based, Our task is to construct, a
equations (9-15). In the function {0\T(d>(x\).. .)|0) there is one field for each particle in systematic procedureforcomputing the terms of this series.
the in- and out-states. (We shall use the name, "Green's function", for this quantity but
c) Wick's theorem
it is sometimes called "r-function" in the literature.)
The Fourier transform of the Green's function is defined by Since Cinl(y) is a polynomial in <i>m{y) it appears that, we need to compute the vacuum
expectation value of the T-products of finite numbers of free fields. These functions arc
all comprised in the generating functional (8.17)

(2)

defined for arbitrary momenta, puPz, • • • Since the T-product is symmetric with respect where J{x) is an arbitrary source function and A F is the causal Green's function, or Feyn-
to permutations of the bose fields, it is clear that F(pi,pj,...) is a symmetric function. man propagator, (8.11). By expanding both sides of (7) in powers of */, and comparing
The S-matrix element (1) is now given by coefficients, we obtain the required formulas. The terms of order 2N give
r
k[... out\kx ...in) = lim \-{kj - M2) ...-{k?- M7)... F(-Jk,,... k[,... (3) j dxl.. - dx2NJ{xi) ••• 10} =

1 ( ]\N f
where the limit means kf —> M 2 , . . . , fcf —» M 5 , . . . This limit does not vanish because, as = -yi f - j ) j dx
i • ••dx1NJ(xi)t\F(x1 - x2)J{x2)...
we shall see, the function F ( p i , . . . ) contains factors i{p\ - M2)'1 for each of the momenta
(8)
(Lectures 16, 20).
Since J{xl)... J(X2N) is symmetric under the (2iV)! permutations of xi,...x2fj we must
symmetrize its coefficient on the right hand side of (8) before making the comparison.

80 81
This is achieved by relabelling of the integration variables in all possible ways. The The details of the integration measure and the functional F(J) are not important. If
number of independent ways is
F{<t>) is polylocal then the integral is an ordinary inulti-dimensional Fourier transform
with respect to the variables 0(xi),d(T 2 ),... etc. The quantity we want is
(2.V)'
2lV:V!
(o|r(F^,,,))|o) =
Thus, not all of the (2A')! permutations give a different coefficient because Ay(x, - x3) =
AF(3;j — J:,), hence divide by 2' v , and the order of the N factors A f -... Af- is irrelevant, = J{dJ)F(J)f-"'}<l^'J'I^^-'J^'1 (13)
hence divide by A'!.The result of this symmetrization is
according to Wick's theorem (7). Now, using the elementary Fourier identity.
1 , ] , „[ •
dx
Jf\}~2> J > •••dx2?

— ( yv /
NV T (2N)\J which is valid for any functional, f(iJ), we can write
lV r
where the sum over "contractions"' means the sum over all (2A')!/2 A ! independent pair-
ings of the points r.i,.. ., X^N- Thus we obtain the formula

so that
The right hand side contains just enough terms to make it fully symmetric. For example,
with N = 2 it gives

(14)
- X4)
(10) where the function (p(x) is set equal to zero after the differentiations have been carried
out. The formula (14) is only another version of Wick's theorem. It simplifies drastically
The formula (9) is adequate for evaluating the terms of the perturbation series (G). A when F{ij)) is polylocal. Suppose
typical term,
~(0\T(4>m(xi) • • • 4>m(xN)dr,tiyi)- • • Cint(yn))\D) (11)
Derivatives are defined, as usual, by
involves a finite number of free fields and will be expressed as a finite sum over all possible
pairings. The result will be a function that is symmetric in the labels x\,.. . xN and also in SF
j/i,. ..!/„. However, the number of pairings grows quite rapidly with the number of fields 6F(4>) = JdxSd>(x)-
and we need to exercise a good deal of care to be sure that none is overlooked. There are
several ways to organize the calculation so as to make it mere mechanical and automatic.
We shall discuss two: ;v differential operator method, and a diagrammatic method (next
lecture). which implies
6F
<•) Differential operator method
On substituting for 6/6<j> in (14) we obtain
Suppose we are to compute the vacuum expectation value of some chronologiclaly or-
dered functional, T(F(<j>)) where the functional F(4>) may be "polylocal", i.e. an ordinary
function of <p(xt) and/or its derivatives at a set of points Xi,x?,..., or it may be a more
general functional. We require only that it makes sense to write the functional Fourier 01, 02, • • -J^^G (15)
integral,
d J { ) i ) which is yet another version of Wick's theorem. It is easy to verify that it reproduces the
J i (12) formula (9) when / = <j>(xi).. .<J>{XIN). But it is particularly suitable for evaluating the
perturbative contributions (11).

82 83
 To illustrate, consider the example e) Reduction formulae for fenmons
1
(16) Perturbative formula for Green's functions and S-matrix elements for Dirac fermioiis
and their antiparticles are developed in the same way. T h e reduction formulae ((J. 16') for
which contains six fields and therefore should yield 5.3.1 = 15 contractions. According to
the annihilation operators. b(k. A) and il{k. A), are
(15) (he function (16} is given by

ri 0=0 = / dxeikl\n{k, \)(i

t(k,\rT{...)?T(...)<
3dv + cycl23}\ -
= fdxe-iilxv(k,X)-{i (20b)
- y)
where the -(+) sign applies if T{. ..) includes an even (odd) number of fermionic operators.
A second example is Similarly

J
36

(j 4

It is not difficult to see that this formula can also be obtained from (17) by allowing the
^i-i ^ -m)v(k, A) (20d)
three locations, .;:,, x-2, x^ to coalesce. Repeated applications of these formulae lead to the fermionic analogue of (1) or (3).
d) Normal ordering The results are somewhat more complicated because of the need to distinguish between
particles and antiparticles. We illustrate their application to matrix elements where two
The function A / ( i - - x1) is very singular at x ~ x'. Formulae like (17) and (18) are asymptotic fermioiis (or antifermions ) are reduced out. define the function
t herefore problematic. To solve this problem it is necessary to perform a 'renormalization".
Technically, this involves smoothing the Green's function by introducing an ultraviolet
cutoff. A, so that Af-(0 becomes a finite function of A. At the end of any computation one
removes the cutoff by taking the limit A —» oo. At an intermediate stage, however, one • — (>. 0i — m){A (21)
modifies the interaction Lagrangian by introducing cutoff-dependent terms (counterterms)
whose job is to cancel AF(Q) wherever it appears. The limit A —> oo will then yield where (A out] and \B in) are any multiparticle asymptotic states.
something finite. Now consider two fermioiis (k\, X\) and (fca,A-,>) such t h a t k{ ^ k2 a n d / o r Ai i and
It. is clear from (15) that the troublesome terms in the expressions (1) are given by {A out\bma(k2\2) +
- 0
e 5 ^ 1 0 1 '"•<«>' C,nt(y) &in(/tiAi)|B in) = 0 (22)
n a
which is a polynomial in A F (l)). To cancel them we have to replace £; n t y new We construct the 5-matrix element.
interaction Lagrangian.
(A, fciAi out\B, k-iXi in)
ajr(0) = {A out\bout(k-[A])|B,k2\2 in)
Cint{y) + ACint(y) = e""* ^£m(&) (19)
where A £ m t is the counterterm. = j dx\eiklIiS(fciA]) — (; ^i — m){A out\yi(xi)b,n{k2X2) +
\B in)
From now: on we shall assume that this correction has been made. In effect, this
means that we shall work with the original interaction Lagrangian but discard any term = f dxldx1e'klX'-'klX2u{kl\l)-{i fa - m) -
J i
that contains Af(0). Equivalently, we may assume Ap(O) = 0 in the formula (15) when
it is applied to the computation of the perturbative terms (11). This simplification is •A out\T(ip(Xi)ip(x-2))B in)-( — i 02 —frt)u(fejA2)
known as "normal ordering the interactions"'. (23a)

84
using first (20a) and then (20c) together with (22). The same kind of logic gives the Exercise
imtifcrmion reduction.

13.1) Pauli -Villars regularization. A Lumitz invaiiuut method to make the propagator
(23b) finite fit .T = .(' is by deh'ning a ••regularized" propagator. This means, in Wick's
theorem replace AF(x — x') by
using (20b) and (20d).
Similarly for pair creation,

-4,A"|Ai, fcaA2 out\B, in) =

= (A, k,Xu out\dmt(k2X2)\B, in) where
I.C-2

= - fdx2c lk2I2
(A, kiXi out\i!(x2)\ n)-(-a k2 - .HP k2 - A?
with Cj and r 2 chosen such that Arrg ~ (l/y,:?) for A'2 —+ oo.
I +lk
= - Idxldx2c''" ' ^u(k]\l)-(i - m) •
a) Find c\ and c^.
{A t\TW)4())iB i){ b) Show that A rf , 3 (x — .r') is finite* at x = r'.
(23c)

Finally, for i>air annihilation

(A out \k1\].k2X2,B in)

f dxt
( #
(23d)

86 87
14 Perturbation theory (II) - Feynman graphs hi the expansion of the exponential operator in (2) it is necessary to pick out all lhe terms
that contain djdoix,) once, i = X....S -Mid (cl/cM.ij,,))"' where rn is the power of d{tj)
We now consider the graphical reptt -sent at ion of perturbative contributions to the Green's
functions. After some generalities we illustrate the main features with examples. in £,„,, — a = 1.. . . . u. Ench term torresponds to a graph.
For simplicity of illustration we shall assume that C,,,t is a monomial.
a) Lines and vertices

The free field vacuum expectation value, {Q\T{4>ln(x\} - • • &n(;c2/v))|0} was shown in
Lecture 13 (Eq. (13.9)) to equal a sum over all the (2N)\/N\2N independent, pairings with m = 3 or 4. (Generalization to polynomials with several different powers is not
of the locations Xi x-2y, each pair x,,Xj corresponding to the factor A F (i, — Xj). difficult to include.)
Each term in the sum is now represented by a graph with 2N "vertices", labelled by the The function (2) is necessarily symmetric with respect to permutations of the x, and.
coordinates i , , . . . , j:^,v and each factor Ap(xt -x}) is represented by a line joining vertex separately, the ij u . Hence many of the diagrams can be obtained from each other simply
.j.-j to vertex x.j. For example, with N = 2. the three terms in (13.10) are represented by by permuting these labels. However, it can happen that some diagrams are invariant with
three graphs; respect to some of the permutations. One must be careful not to include more than one
copy of each diagram.
There are four important features:

X 1) Because we are assuming that £ m ( is normal ordered, we have A(.(0) = 0 and hence
can exclude graphs where any line begins and ends on the same vertex.
2) Because we are going to integrate over y\ yn these coordinates become "dummy
lables" and gTaphs obtained from each other by a permutation of the yn will yield
identical contributions to the Green's function. Usually, it vvill be sufficient, to
evaluate only one of these graphs and discard the factor 1/nl However if a particular
gTaph has some permutation invariance then it will belong to a a set with fewer than
Next. \\v. must, consider the graphical representation of the perturbative term JI! copies. In such cases the factor 1/n! does not quite cancel and we are left with
a coefficient 1/5, where 5 is known as the "symmetry factor", [t is equal to the
.. .4>m(xN)Cml(yi)... Cinl{yn))\0} (1) number of independent permutations that leave the graph invariant.

Now we have graphs with N + it vertices. x\,... ,xN and j / i , . . . y n . In a typical graph one 3) Since, for the interaction (4), we have [dj'd<p)'nC,nl((}>) = g we can ignore the factors
line starts from the vertex, x, ami must go to some other vertex, either x} or ya. From 1/rn! Each 1/,,-vertex effectively contributes the factor, ig, to the Green's function.
each vertex i/n several lines emerge the number depending on the power of 4>,n(ya) in
£ m l - going f 0 other vertices either x3 or y^. There can be no free ends, every line must
4) For any graph in which two vertices ?/„ and ylt are joined by more than one line, there
go from one vertex to another. To represent the function (1) correctly it is necessary to is a factor (l/£!)A F (j/ a - yfl)f according to the formula (2). The factor l/f\ must
include every diagram that satisfies these rules. These will exhaust the terms to be found be included.
in the formal expression (see (13.15)). The Feynman Titles for evaluating the rjth order contribution to the Green's function
(0\T(<j>(xi)... <l>{xff))\0) may be summarized as follows:
(2)
» ' •

A) Draw all graphs with vertices i'i r^, yi •, ,yn where one line emerges from each Xi
where and m lines from each ya. Exclude graphs where any line begins and ends on the
same vertex. Exclude copies, i.e. graphs obtained by permuting the vertex labels
Add
'
B) For each of the graphs construct the corresponding contribution to the Green's func-
tion as a product of the following factors

d d 1) For each interaction vertex: ig

E (3)
2) For each line: A/-
3) For multiple lines joining a pair of vertices: \ji\

89
 , ? * * • m « H is::* W

4) For graphs with permutation invariance, the symmetry factor:l/S

C) Integrate over y\,..., yn

li) Examples —
To begin we illustrate the first few terms in the expansion of the 2-point Green's
function (0\T((p(x[}<fi(x2))\Q). We examine two models, one with cubic interaction,

C - V (5)

and one with quart ic interaction

(6)
n
To save writing,we shall not explicitly show the factor (ig) or the integration over
yi yn. Also, we denote the Feynman propagators as follows.

A;..(i t - x2) = A n , A F ( i , - y,} = A n , AF(yi ~- y2) = A r (7)

etc.
Firstly, for the cubic interaction we find one contribution for n = 0, two for n = 2 and
six for 11 = 4.

n = 0: • -; = A;;

1 1 2 3 4 2
n = 2;
The first of these is invariant with respect to the 4! = 24 permutations of j/i, yi, y%, j / 4 .
Hence the factor 1/24. The second has three lines joining j/i to y2 and two joining j/- t to
2/4- It is also invariant under y\ «-» j / 2 . Hence the factor

1 1 1 1_ 1
2
- 2 *' 12 22 3!" 2 ' 2 = 24

In the first ?< = 2 diagram there are three lines joining j/i to yi and the diagram is The third has two lines joining j/i to y2 and symmetry under yy <-> y2. Hence the factor
unchanged by the interchange of t/i and y2. Hence the numerical factor is 1 1_ 1
1_ 1 _ 1 2 ' 2 ~ 4
3! "2 ~\2
The fourth has two lines joining 7/3 to y4 and no symmetries. Hence the factor 1/2.
In the second there are two lines joining yi to y-i but no permutation invariance. Hence
the factor is 1/2. The fifth is invariant under 1/3 <-• 1/4. Hence 1/2.

n = 4: 4 3
The sixth has one pair joining jh to y2 and another joining t/3 to y4. Hence,
X 1 1_ 1
2'2 ~ 4
There are no other diagrams up to this order, a fact that can be verified by the
differential operator formula (2).

90 91
 Now consider the quartic contributions. We find,

3 4 1 4 2 4 3 5 5 T 1 3

n - 2
1 2
°TT &12 +
6 V, -
22

+ 5 terms

n = 3 The last three are given by

T 3 2 1 1

^+^ ^
h 2>
Of the two diagrams for n = 2, the first has 4 lines joining y^ to j/ 2 and is invariant
under i/i «-» j/ 2 , hence it carries the factor 1/4! • 1/2 = 1/48. The second has 3 lines
joining yi to y-2, hence the factor 1/3! = 1/6.
Of the three diagrams for n = 3, the first has three pairs of lines joining vertices and
is invariant under the 3! permutations of yu y2, j/ 3 . Hence the factor (1/2)3 • |! = ^ . The
second has three lines joining j/i to j/2 a n d is invariant under y^ *-+ j/a- Hence the factor Again the symmetry with respect to permutations of X\,... ,14 is achieved by inte-
5T ' 5 ~ tl' '^' le ^'TCI has two pairs of lines joining j/3 to yi and y?, respectively, hence grating 2/i,t/2.
factor (\? = 1/4. The theory with quartic interactions develops in the same way. There is one graph at
It will be noticed that some of the above contributions to the 2-point Green's function n = 1:
do not appear to be symmetric under Xy «-» Xi. This symmetry is recovered when the
interaction coordinates, 2/1,3/2,... are integrated.
Now consider the 4-point Green's function, {0|Tf</>(zi),.. 0(x4))|O). Firstly we take
the cubic interaction. There are 3 graphs in zeroth order _ _ 4
11 21 31

-4 +- 1 4 + i-
n =0
At n = 2 there are twelve graphs. Three are like the n = 0 graphs accompanied by
2 3 3 1 the vacuum factor

= AjjAjj + A23A14 + AsiAji

IS hi
There are no graphs in first order. There are twelve graphs in second order. Three of
them are like the n = 0 diagrams except that they are accompanied by the vacuum factor
1 3 The next six are like the n = 0 graphs except that one !ine in each pair is corrected,
12
~ 12
The next six are like the n = 0 graphs except that one Sine in each pair includes a
correction:

92
93
 pairs t/4,... giving \)N\. After integration over the j/ ; s the contribution is
+ 5 graphs
nOr (8)

i 4 A 4?T A +5 terms These contributions can be summed over N to give

6
34 Tl 1 2 22

The last three are (9)

The effect is quite general. The totality of vacuum graphs can be summed to give an

'XX'. XX'
exponential of the connected vacuum graphs:
<0|S!0) = exp (10)
It can be shown that the contributions of connected vacuum graphs are pure imaginary
+ - A , , A5l A ^ +- so that (0|S|0) is a phase factor. In fact this phase is unobservable and we shall always
renorraalize the theory so as to make it vanish, i.e. give vacuum stability.
The vacuum graphs will inevitably appear as factors in the perturbative expansion of
any Green's function. Since we now know that they sum up to give a multiplicative factor
c) Connectedness equal to unity, we shall henceforth ignore them. That is, we have another Feynman rule;

A great strength of the graphical method is that it makes very clear some general
features of the perturbation series that are difficult to recognize in the analytic expressions. D) Exclude graphs containing a disconnected vacuum part.
One of these is symmetry with respect to permutations. Another is "connectedness". A
Finally, consider the generating functional for Green's functions
graph is said to be connected if there is no subset of vertices that is not connected to
the remaining vertices by at ieast one line. If there is such a subset then the graph
is disconnected and its analytical expression will factorize (in spite of the integrations
over 2/1,... yn). The simplest manifestation of this in the above examples is the "vacuum (11)
graphs"
where <j>(x) is the Heisenberg picture field. The perturbation series for this functional is
1 = 2 : A
1^2 2 3! 12 given by

1 .3 1 ,3 = {li\T(eiIdli:'""M)\0) (12)
3T "ll 31" A3«
where the new interaction Lagrangian CJinl incorporates the source term, e.g.

m
6 24
From the viewpoint of the new interaction, the functional (12) is just the vacuum am-
n = 2N plitude. It is given by the sum of all vacuum graphs with three kinds of interaction
"12 2N-1.2N
vertex:

2N
: ig

The symmetry factor here is 1/W2" because the graph is invariant with respect to JV
(14)
independent exchanges t/i <-> j / 2 , . • • giving 2~N, and with respect to N\ exchanges of the

94 95
At each order, i.e. for specified powers of J, g and A, the contributions will factorize Exercises
(15) 14.1) Consider the model with two real scalars of the same mass.
m,! m2!
:
where Wt denotes the contribution of some connected graph, which occurs m t times and 4 ^ —0\ + -ov<fetf"02 $
1/7^! is the corresponding symmetry factor, W2 represents another connected graph, etc.
On summing the graphs we obtain

|0)=e H/ ( J » (16) This model has 50(2) symmetry

a) Give the infinitesimal 50(2) transformations and find the Noether current.
where
W,(J) + W2(J) + ... (17) b) Replace 0] and ip'i by the complex variables
is a functional that represents the contribution of all connected vacuum graphs. It serves
as the generating functional for "connected Green's functions". ), 0' = -j=(

writing the Lagrangian in the form

=£ • • drHJ(xx)... J{xN)G{xu ...txN) (18)
C = d^d"^ - M24>'0 - -
A
The name, Green's function, is commonly applied only to these functions which are rep- Show that in the free (A = 0) theory
resented by connected graphs,
<j4(z'))l<)> = 0
= {Q\T(4>(xl)...4>{xN))\ (19) x')')|O) = A,,(x - x')
By expanding both sides of the formula (16) in powers of J we can express {\T{<p(x\)... c) In the real formulation there are two propagators
<7)(x,v))jO) in terms of the connected Green's functions, G(x\,..., XJH), M < N. For
example, x x'
2
and three kinds of vertex

+G(x2 -
+G(x3 - xi)G(z, - T 4 ) 1
2 .
+G(xi,xi,x3,xt) (20)

etc. (Here we have made the simplifying assumption that G ( l , . . . , N) vanishes if N is

odd). From now on we shall be considering only connected graphs since, using formulae Explain the numerical facts | and j .
such as (16) or (20), it is easy to construct the complete functions from them. Compute the n — 2 contribution to the connected Green's function
There are two graphs.
d) In the complex formulation there is one propagator,
(0\T{<f>{x)<j>{x'y)\0) = - •

and one vertex v J

Explain.
X =~iX
Compute the n = 2 contribution to (T{<jxj>')}con. There is one graph. Compare the
result with the computation from part c).

96
15 Fermions and Yukawa theory T h e result (5) is antisymmetric with respect t o t h e p e r m u t a t i o n s of x,,... ,XN (along
with t h e spinor labels r i | , . . . ,<i. v ) a n d . independently, x\,... , x ' v ( a n d a\ ...a'K).
We repeat the discussion for systems that include Dirac spinors, i.e. we consider the
For N = 2 t h e formula (5) gives
application of Wick's theorem to chronological products of free fermion fields and develop
Feynman rules tor interactions involving fermions. These rules are then illustrated by
their application to the Yukawa interaction.

a) Wick's theorem
This expression is represented by the two graphs
Tlie fermionic analogue of the generating formula (13.7) is given by
1 , z' 1 ,z l
1 d r 1(i8) ( i |ll+ n(l) 1 r i 1 dil x) z
(GIT (t / *- <' '» *' i ' " ) 10) = e"/ " '^ S'-te-*") i( '> M)

where >l{x) is an anticommuting c-number source spinor and fj is its conjugate (see
i1 i i' z
Eq.(8.18)), Expanding both sides of (1) to order N in j] and fj,
In contrast with the real scalars of the last lecture we now have complex fields, ip and ip.
The propagator 5f (ii — xi) is not symmetric with respect to the exchange x, «-> Xj and
M ) IdXl •••dXN dx 1
' • --dx>N (0|T(f
'(ll) 4 n ( l l )
• • • we must therefore represent it by a line with an arow,

n . .. dx'N Tj(ri) S F (rr L - x\) r j f x ' , ) . . . fj(xN) - x'N) ? } « ) . (2)

The left hand side can be written

1 2

i . ..tlx'N n(xN)... q{Xi) (3) Just as in the scalar case we can give a differential operator version of Wick's theorem.
Let / be a polylocal function of i) and !•>. Then
where it is understood that the spinor index on fj{xi) is contracted with the corresponding -Z ssfc
index on v!7,(x,), etc. The order in which the various factors appear in (3) is significant (6)
because they antitominute. The right hand side of (2) can be written
It is necessary to be careful with signs when using this formula. On the right-hand
xi ... dx'f, 7}(xN).. -f - x'N) r){x\)... TI(X'N)
side the variables ib(x\),iji(x2},- •• are anticommuting (.-numbers. This means that the
differential operators d/d^>(x-l),O/d41{x2),. • • also anticommute. We use the convention
(-)N f that they always act from the left, e.g.
= - r r p / (fai . dx'x f}{xN) F(xN - x\) ij(x' v ) . - . T){x\)

= -r^r- / dxl...dx'f! fi(xN) XI ~ X'PK) ... SF(xN - x'Pt)

:\ . J

b) External field
alter relabelling the integration variables and using the antisymmetry under odd permu-
tations, (br = ±1) It must always be kept in mind that 4/ andtyare 4-component fields and Sp is a 4 x 4
matrix. In the above formulae this matrix structure is implicit. The right-hand side of
Comparing (3) with (4) we obtain the formula (5),forexample, is a multispinor of rank 2N, In many practical applications the spinor
indices are contracted and the resulting formulae have a simpler structure. To illustrate
(0|T (ip,n{xi) • • • i/Jm(z'v)) K') - S *J> ' i-TpJ.-.S^xn-x'^) (5) the kind of arrangement that frequently arises we consider the interaction Lagrangian
perm
C/a{y) = i>{y) M{y) il>(y] (7)
where the sum ranges over the /V! permutations
where M(y) is an arbitrary 4 x 4 matrix - an "external" field. We construct the vacuum
(1,2 JV)-»(A.f graphs.

98 99
 In second order, If we integrate (9) over the coordinates j/i... . , y.v then each graph makes the same con-
2
tribution, so that
l

-l— Jdyt ... dyN tr (M(yi) SF(yi - m) • SF{yN - j/0) • (1")

The factor l/N is a symmetry factor of the kind discussed in the last lecture. The result
(10) is represented by a fermion loop with N vertices (N > 2)
- y2) M(y2) SF{y2 -

(8)

where the notation tr{ ) means trace over Dirac indices. This expression is represented
by the graphs

Notice the minus sign in (10). This is characteristic of the fermion loop. It arises because
of the rearrangement,

M(yN) i% w )) =
= -tr [T yN) M(yN) 4>(yN) 0 ^ ) ) ]
in which the wavy line represents the external field, which involves an odd number of exchanges.
The full vacuum amplitude is obtained by exponentiating the connected vacuum
graphs as explained in the last lecture
From now on we shall use the normal ordered form,

(7')

so that, in effect, 5p(0) = 0 and the "tadpole" graphs are absent. This formula is actually quite important for Part 2 of this course where it will be derived
It is easy to generalize (8) to order A', again by another method. It will be shown that if M is independent of y, the sum over
N can be evaluated exactly.
To complete our discussion of the external field interaction we now consider the con-
— {OlT(£M(y})...CM(yN))\0}con nected Green's functions. There is only one,

G(x,x'\M) = (0| (iUx) An(x') e'/"" (12)

(9) The first order contribution is
where the sum ranges over the (N - 1)! permutations, ijdy <0|T (An{x) : i>Uy) M(y) iPin(y

= i j dy SF(x - y) M(y) SF(y - x')

r
The result is symmetric under the A ! permutations of yu..., yN. The reason we find
(A'' — 1)! and not N\ terms is that each graph is invariant with respect to the N cyclic _ - A - -1
permutations, x
(1,2, 3 , . . . , AT) -> (AT, 1,2, . . . , # - 1) .

100 101
where matrix multiplication is understood. At order A\ where normal ordering is understood. This theory involves three parameters: boson mass
fi, fermion mass m and Yukawa coupling, g. We wish to develop series expansions in g
for the connected Green's functions.

= <N Jdy, . . dyN SF{x - JA) A/(?,,) SF(yi - y2) M(y2)... M(yN) SF(yN - x1)

where
* V, V2 yN £mt(y) = g • j>in(v) 4>{y) ikUv) •• • (16)
The connected Green's functions will be described by graphs made from two kinds of line.
because each of the ,V! graphs makes the same contribution when y\..... y^ are integrated,
hence the series for G is represented by the graphs

corresponding to the analytical formula

G(x,x'\M) = SF(x-x') and the interaction vertex

+ i j liyt Sy(x- ,(/,) M(yt) Sfiyi -x)
i g =
l dy2 SF(x - Vl) M{Vl) 5 F (y, - y2) M(y2) Sf-{y2 - x')
+ ... (13)
It is possible to see this series as the solution of a differential equation. Using the We illustrate the construction with some low order examples.
basic equation
(»(? - m) SF(x -x') = i 6i(x - x) (1) boson 2-point function
we tind, A'F(x - x') = (0\T (fcn(x) 4>M) e'J'dy £ »') [OJ^
(i$ - m) G(.T, x'\M) = i iA(T - x') - M{x) SF(z - x') = AF(x - x') - {igf j dy dy' Ay(x - y) •
-Mix) ijdy SF(x - y) M(y) SF{y - x') - ... •tr (SF(y - y') SF(y' - y)) AF{y' - i') + ...
= i6i(x-x')-M{x)G{x!x'\M) .
= • • •+• * -i V • +• •••
Thus. xx1 x vV«/V x1 (17)
ifi -m + M{x)} G(x,x'\M) = i S^x - x) . (14)
There is no symmetry factor but there is a minus sign due to the fermion loop,

c) Yukawa theory (2) fermion 2-point function

The interaction with an external field is almost trivial in the sense that the connected S'F(x - x') = (0|T ( , U z ) ^in(x) e1^" £ "") \0)m =
graphs are very limited. The connected vacuum graphs have only one loop. There is
only one connected Green's function and its graphs have no loops. The simplest non- = SF(x - x') + (ig)11 dy dy1 SF(x - y) SF{y - y1) SF(y' - x') AF(j/ - y) + • • •
trivial theory describes the so-called Yukawa interaction between Dirac ferraions and
scalar bosons. It is governed by the Lagrangian

0 - ^ T <t>2 (15) x ¥
(18)

102 103
 In the next order there are many 1-loop graphs that contribute to this Green's function.
(3) boson 4-point Junction For example,

(0|T ( ^ ( z i ) • • • 4>in(x4) e<Jd» c«") 10)™ = I I

4
= (ig) jdyi... dy4 AF(xx - i n ) . . . AF(x4 -yt)-
t i
11
<0|T ( i / U y , ) i,[y{)... j>m(yi) AniVi)) |0)™ + ...
4 d
= -(*3) / Vi • • • dy4 A F { z i - !/i) • • • AF{xt - yt) •
and
- y4) SF(Vi -!/,)} + ... (19)

where the sum runs over the 3! permutations of 1/2,1/3, VA- There are six 1-loop graphs
together with their permutations.

(5) fermion-fermion scattering

4 3 2
t tt jdVl dy2 AF(Vl - y2){(SF(xi - Vl) SF(y, - x'2)) (SF(x2 - y2) SF{y2 - x\))
- (5F{Xi - j/,) SF(Vl - x\)) {SF{x2 - jto) SF(y2 - 4 ) ) } + ...
1N
S

+ i
*1 Vi »,

x2 v 2 x^
(21)
This is already a non-trivial computation. It describes the scattering of two bosons
where matrix multiplication is implied at the vertex y\ and, separately, at y2. Notice
through their conversion into intermediate fermion-antifermion pairs. Much simpler is
that the two graphs ensure the antisymmetry of the Green's function with respect to
the scattering of two fermions or one boson with one fermion because in these processes
interchanges of i i , i 2 (together with their spinor indices) and likewise for x\,x'2. Again
there is a low order contribution from single particle intermediate states which is described
there would be many 1-loop graphs if we were to continue to the next order.
by graphs without loops ("tree" graphs).
(4) boson fermion scattering
Exercise

15.1) Compute the 1-loop contribution to {Q\T($(xi) </>(z3))|0)con i n Yukawa the-

= [is)2 j SF{y2 - x2) • ory. Draw the graphs and label the vertices.
yi) AF(x3 - i
x4 x
t *

(20)

104
105
 •.•^t. « • " < • • • " •

1.

16 Momentum space One cau think of 4-momentum as something that Hows along the lines, coming out
at the external vertices, The amount of 4 momentum that comes out of the graph as a
For practical work such as the computing of cross sections it is necessary to obtain the whole is zero because of the factor (6). On the internal lines the flow is arbitrary, subject
Fourier transforms of Green's functions;. Some labour is saved by formulating the Feynman only to the condition that it be conserved at each interaction vertex
rules directly in momentum space. We do this here, illustrating the steps with examples.
We illustrate with some examples from Lecture 14 with the interaction
The method is then applied to calculate matrix elements using LSZ reduction formulae.
A ,
a) Feynman rules in momentum space 1711 (7)
~ 24

Consider firstly the connected Green's functions for scalar fields. Their Fourier trans- The 4-point connected Green's function receives a contribution at n = 1,
forms are defined by
i A j dy AF(x, - y) A/(J: 2 - y) A F (i 3 - y) AF(x4 - y) (8)
= Jdx1...dxNe«"x<+ •+lPKlw
represented by the graph with no internal lines

X
. (1)
A typical contribution will be represented by a connected graph with JV external lines,
terminating at the points X\, • •. x,\ and n interaction vertices, i/i ... yn. The external lines
represent, the factors AF(X, - ya) and the internal lines represent Af (ya - yp). Since the
graph is connected, each external line must arrive at one of the interaction vertices. The The Fourier transform of (8) is
integrals over xi . - .XN in (1) therefore give the factors, +
""IJ iXJdy A,.{n - y) • - y)
(2)
dy e»« ... &(Pi) e^
where ... + p4) A(p,)...A(p 4 ) (9)
A(p) = l (3)
- \i + IE and is represented by the graph
Hence G{]>\ ... ]>$) contains the factor A(pi).. . A(pjy) For each internal line write

(4)

It is then straightforward to integrate the coordinates J/I, ..., yn since the y-dependence
is contained in phase factors. For each interaction vertex we have

dya (5)
There axe three connected graphs at n = 2
where the sum, Sq, contains the 4 momentum vectors of all lines arriving at the vertex
yn. At this point the graph represents an integral over the </'s associated with internal
lines. There is a factor A(q) for each internal line and A(p) for each external line. For
each vertex there is a delta function ensuring 4-momentum conservation. Because of the
' * "XX"
n delta functions it is possible to directly integrate n - 1 of the q's. There always remains
one delta function that depends only on the external momenta, (to)

The Fourier transform of (a) is

that ensures overall 4-momentum conservation. The number of q's that remain is equal
to the number of loops in the graph.

106 107
 Again the internal lines carry 4-momentum to the left.
Since A(g) = A(-(j) it does not matter which direction we choose for the flow so long
dxx ...dx, f""*'+•••+•'«*•'i^ as it is conserved at each vertex. In theories containing fermions we have to be more
careful.
• A K (y, - y2)2 A F (y 2 - .en) A F (j/ a - I,,)
If two interaction vertices are connected by a fermiori line, write

A(Pl)... (2TT) 4
-yti) (12)

where
- q)
(13)
(10) - m + ie
We also need the fermion analogue of (2),
and is represented by the graph
Jdxe"1 SF{x-y) =
J dx ^ SF(y - x) = S(-p)el™ . (14)

Consider the n — 2 contribution to the fermion 2-point function

The graphs (b) and (c) go over in the same way to their momentum space versions.
One haa only to take cyclic permutations in P1.p3.p3 to obtain them. Conservation of
4-momentum at. the two vertices requires that the internal lines carry their respective
. \ . .
4 momenta out, of the right vertex and into the left.
Consider the n = 2 contribution to the 2-point function Its Fourier transform is
jdx, dx-i e<n*'+w (igf jdy, SF(Vl - y2) SF{y2 - x2) - y2)

= (rgffdy, dy2 S(Pl) e^ jft) S(-p2) e"™

The Fourier transform is

•(2TT)4
dy2 e !pi!
A(p, - 9) S ( - (15)

represented by the graph,

•(2TT)4 6 + 93 - Pi) ( 2lr ) 4 + 93 P,-q
A( P l ) +pa)/ ~ ^ A ( P l - qx - q2)

(11)

corresponding to the graph

The internal momenta flow to the left. The external lines represent the factors S(pi) and
S(-p2)- In general we use S(q) if the 4-momentum q flows in the direction of the arrow
and S(-q) if it flows in the opposite direction.

108 109
 . .>-,!•. III.••>•••• i

As a final example consider the 2 boson + 2 fermion function at n = 2

I .1 .

The Fourier transform is

.ipix,+...+,r,x4 fj^ja I dyi dy2 SF(xi - yt) Spiyi - y2)

dy2 Sip,} Klpm SF(y, - y2) S(-

• (A{p 3 ) <•"»«' A(p4) a"""' + A(p 4 ) e">«

1
= (ig) A(p3) A(p4) / S{p\) S(q) S{-p2) I dyi dy2 •

A(p4) Sip, S(px + p4)) 5 ( - p b) Reduction formulae and matrix elements

(16)
We can now apply the reduction formulae of Lecture 13 to obtain matrix elements
represented by the graphs
in the basis of asymptotic states. It is clear from the above disucssion that the momen-
I tum space Green's functions contain factors A(p) or S(±p) for each external line. The
pi pi reduction formulae {13.3} for bosons and (13.23) for fermions prescribe that these factors
i . i
31
should be removed {"amputated"). The external 4-momenta are then put on-shell, i.e.

=± (17)
Having gone through these examples we arrive at a prescription for constructing the
where ui = Jp2 + fi2 for bosons and u> = Jp1 -t- m 2 for fermions. If p0 = +w then the
Fourier transforms directly from the graphs. There is no longer any need to use the
external line is associated with a particle of 4-momentum p^ in the out-state. If pa = — u
coordinate space versions of the diagrams. To summarize, the momentum space Feynman
then it corresponds to a particle of 4-momentum -pv in the in-state. If the particle is a
rules for Green's functions are:
fermion then we must multiply by the appropriate Dirac spinor, u or v if the arrow leads
A) Draw all connected graphs with appropriate lines and vertices. Include couplings into the graph, u or v if the arrow leads out. There are four possibilities,
ig, i\. etc. for each vertex and the numerical factors, 1/^!, 1/5 as described in
Lectures 14, 15.
i) u(p) for out-state fermion
B) Assign 4-momenta to each line choosing a flow pattern such that 4-momentum is
conserved at each interaction vertex and flows out at each external vertex. Associate ii) u(— p) for in-state fermion
propagators A(<j) with each boson line carrying 4-momentum q and 5(g) (or S(-<?)) iii) v(p) for out-state antifermion
with each fermion line carrying q in the same (or opposite) direction as the arrow.
iv) vi~p) for in-state antifermion
C) Form the product and integrate over the independent loop momenta, contract the
Dirac indices of propagators that meet at a vertex. To illustrate, define the amputated, connected, 4-point Green's function, G, for the
scalar field,
A systematic way to assign the momenta is borrowed from electric circuit theory.
Independent momenta circulate around each loop of the graph. The lines that border a.
particular loop carry the loop momenta (in the sense of the circuit) plus or minus the
j dxi,,,dx i

external momenta that flows through the network. Two examples: (18)

110
111
Then the 5'-matrix element for the reaction 2 boson —* 2 boson is given by described by the various on-shell limits p l 0 —» ± w j . . . . of a single amputated Green's
function.
(*3,/c4 out|fc,,fca in) = <5(-A:i,-A>!,fc3,fei) (19)
The 4-point Green's functions are usually expressed as functions of the Mandelstam
where fci0 = wi, k20 = u 2 , fao = wa, k4n = UJ4. invariants,
Next, define the amputated, connected 2 boson + 2 fermion Green's function, F,
* = (P\ H = (j>i+ PA)7

Jdx,...dxi e""^-+'P^
u = (pi + p 4 ) 2 = (p2 + Px)2 • (23)
= A(p3) A(p4) S2 (20)
The conservation of 4-momentum, pi + p 2 + p3 + p 4 = 0, implies a relation
Then four distinct reactions are described:
s + c+ u = Pi + P2 + P3 + p 4
1) fermion + boson —» fermion + boson = m\ + m22 + ml + m\ (24)
a so that the on-shell amplitudes are functions of only two independent variables: energy
<Mi,*3 oul\k2X2,k4 in) =u (k (16.21a)
and scattering angle, or energy and momentum transfer. (It is also possible to make
2) antifermion + boson —» antifermion + boson partial wave expansions to obtain the amplitudes as functions of energy and angular
momentum.)
( k^\i,k3 out\ kJT, ,k4 in) = e Q {Mi) 3, -fc,) i;3(fc2A2) (16.216) However, it is instructive to retain the three Mandelstam variables, regarding them as
homogeneous coordinates on a plane
3) boson + boson —> fermion + antifermion

(kjX^k^out^iki in) =u a (fc,,A,) F 2, A2) (16.21c)

4) fermion + antifermion —> boson + boson

{/c.,,fc4out| M 7 (16 Md)

N
In all cases the energies are positive, kw = Wi, etc. The amputated Green's function is of
course given by the perturbation series discussed before. Its second order aproximation,
for example, is obtained from (16),

(xgf (Sa +p3) + Sa "(p! +p 4 )) + ... (22)

c) Mandftlstam variables
In the 3-space of s, t and u the Mandelstam plane (24) is defined by the dashed lines.
The formulae (21) illustrate a fundamental property of relativistic quantum field the- The "physical region" for the reaction (I) is the region in the s,t,u plane for which
ory known as crossing symmetry (or sometimes as the substitution law). It means that
amplitudes for physically distinct processes (where incoming particles are exchanged for PlO = -dlu P20 = —1^2, PiO ~ + ^ 3 , P40 = +^4 •
outgoing antiparticles, and vice versa) are obtained from the same Green'a function.
The other physical regions are defined analogously. To illustrate we consider the case
In our example the bosons, particles 3 and 4, are identical so that the amplitudes are
symmetric with respect to the interchange, p 3 *-» p4, but this need not always be the case. mi = m-3 = m and m2 = m4 = p, . (25)
In general one may consider the reactions
In the centre of mass frame for the reaction (I) we can choose the axes so that
(I) l+2~3 +4 f UJ2 \ / Wi
(It) 1 + 3 ^ 2 + 4 0 psmO
-P\ = -Pi = 0 - P3 = 0 Pt =
(III) l + 4 » 3 + 2
\-p2} \pCOS0
112 113
where i^i = sly1 + m2, W2 = s/v2 + U2 a n d P ^ 0. The Mandelstam invariants can now be
expressed as functions of the centre of mass momentum, ;), and scattering angle 8

(I)

t = -2p2(l-t-

(26)

Similarly, for the reaction (III) one obtains

III)

-2p2(l+cose)
t = 0

(27)

But it must lie kept in mind that p and 8 are different in (26) and (27).
For the reaction (II) in its centre of mass system write

—p sin 6
Pi = PA =
-v\ =
KV I K —p cos 8,
The three physical regions, parametrized by (26)-(28) are shaded.
where w = v V + '»2 = \/P V + MJ- The Mandelstam variables are now The second order result (16) gives, for the 2 fermion + 2 boson amputated Green's
function

F(p,...P4) = (2TT) 4 <5(Pl + . . . + p 4 ) ( z s 2

s = -{pi - 2ppc,os 8 + p2)
t = 4w 2
u = -{p2 + 2ppcas8 + p2) . (28) (29)
where pi and p 3 refer to the fermions, p% and P4 to the bosons. The amplitudes are
In these formulae 0 < 9 < n and p > 0, j) > '/mr^J? ( assuming m > singular on the lines s = rri2 and u = m2, but these lines do not intersect any of the
physical regions. The on-shell amplitudes are finite and regular in this aproximation but
their analytic continuation has pole singularities in unphysical regions. In higher order
approximations it can be shown that the amplitudes develop branch point singularities
(thresholds) inside the physical regions.
d) Optical theorem
Unitarity of the S-matrix implies a relation between forward elastic amplitudes and
total cross sections. To derive this relation write S = 1 + iT and impose SS+ = S+S = 1.
This reads,
T - T+ = i TT+ = i T+T . (30)

114 115
Denote the basis of in-states by [n). assuming they are complete and orthonormal, For- 17 Electrodynamics (I)
mally,
EI«)M = 1. (31) The dynamics of electric and magnetic fields interacting with charged fields is the subject,
n of quantum electrodynamics. We begin with ft description of the classical theory and
Assume also that they are eigenstatcs of 4-momentum, so that then consider its quantization, The central idea here is invariance with respect to gauge
transformations.
{m\T\n) = f dx (m\T(x)\n)
a) Maxwell's equations
= (2ir)i6i(pm-pn) { (32)
The classical field variables comprise an antisymmetric tensor F^x). The components
The matrix elements of (30) then give
are identified as follows,
(m\T(0)\n) - <n|T(0)W = i - Pt)
F01 — Ei, F02 = E-2, F0;i = E3
4
after discarding the factor (2n) S(pm - pn). In particular, for m = n, (1)
where the 3-vectors Et and B, are the electric and magnetic field strengths, respectively.
Im (n\T(0)\n) = L\ -?,) \W{Q)\n)\2 (33) Maxwell's equations are
9" i > = -j* (2a)
Suppose (n) is chosen as some 2-particle in-state |fcL k?) while \t} ranges over all possible £*""" 9A F^ = 0 (2b)
multipart icle states that can be created. For example in the scalar boson theory discussed A i 1
in Lecture 12, the precise form of (33) would read, where c <"' is the Levi -Civita tensor and j , , is the current density 4-vector. In 3-
dimensional notation these equations read.
7m (
=j0,

= -J2
:2T
and
: 2/ (optical theorem) (34)
- d-iE, - dBB, = 0
where we have used (12.28) for dan with the flux factor, /, given by (12.25). If the bosons -~ c/jxvi — 3{)B\ — 0
have a mass fi > 0 then the sum over JV is actually finite because
There are eigiit equations for six unknowns but, of course, they are not all independent.
One way to solve them is by representing the field strengths in terms of a vector potential,
As the total energy u\ + UJ2 is increased, thresholds are crossed at which TV —> N + 1,
a new channel opens, It can be shown that the forward amplitude (kik2\T(0)\k^k2) is
an analytic, function, of a = (w^ + u2)'J (at t — 0) with branch point singularities at the This solves the equations (2b) and reduces the number of unknowns to four. Equations
thresholds s = {N^f- TV = 2,3,... (2a) now take the form
The study of the analytic properties of 5-matrix elements is an important branch of d2Au — dud^A^ = —jv . (4)
quantum field theory but it lies outside the scope of this course.
However, these equations are not all independent, Since dM<?" F^ = 0 identically, the
equations (2a) or (4) are compatible only if the current is conserved,
Exercise <3M f = 0 - (5)

16.1) Give the Fourier transform of the 1-loop contribution obtained in Exercise (15.1). There remain only three independent equations to solve for the four components, A^.
Draw the graphs and label them but do not evaluate the integrals.

116 117
 *• -r
•mm. t. f •"-
. » • » • • * • • " . •

One should not. be surprised that the Maxwell equations are not sufficient to determine The answer is that fl should be the Xocther current for a system of charged fields.
the potential. This is because the representation (3) is not unique. A particular F^x) Consider, for example, a Dirac theory with the Lagrangian
can be represented by many different vectors. The transformation
A^x) + dp A(x) C = -- F'"1 (13)
(6)
where A(x) is arbitrary, leaves FIM{x) unchanged. Hence it is reasonable that. (4) should where e is a coupling constant. The equations of motion are
not give a unique solution for A^. Instead we are free to pick out some convenient ones
from among the equivalent solutions. For example, we may look for that solution of (4) (14a)
which also satisfies the equation
- TII)ii> = -<; (14b)
3W A11 = 0 (Landau gauge) . (7)
(14c)
In this case (4) reduces to the simple form We have here a "model" for the electromagnetic current,
d2 A. = -3» (8) f = c1pYiii . (15)
which can be solved by Green's function methods. For example, To show that it is conserved we must use the equations of motion (14b) and (14c),

A^x) = A™(x) - ijdx' DF(x - x') j^x') (9)

c -
where /I"1' is a solution of the homogeneous equations and Df is the Feynman propagator.
= 0,
^-^W^T^-e-*'"''. (10) Hence the equations (14) are compatible. They are compatible because the Lagrangian
(13) is "gauge invariant". Recall that the vector potential A^ contains redundant, unphys-
At this level the theory seems trivial. Complications enter only through the boundary ical information. We could use A^ + 0flA in place of A^ without affecting FIIV. Notice,
conditions: how to find Ai<3\ etc. Of course we had to assume that a conserved vector. however, that the equations (14b) and (14c) would be affected by such a change. To
j,t(x), is given, and this is not a realistic assumption. The motion of charges is affected maintain the form of these equations we should transform ip and %> as well as A^. This is
by the fields and cannot be specified in advance. What is needed is equations of motion known as a gauge transformation,
for the charges, involving Ful, or Alz, to be solved jointly with Maxwell's equations. This
problem is non-linear and highly non-trivial. We turn next to the question of how to set.
up consistent equations of motion. Alt(x) = Ap{x)

b) Gauge invariant, Lagrangians

Maxwell's equations in the form (4) can be derived from the Lagrangian (16)
where A(i) is an arbitrary function. It is easy to verify that (13) is invariant,

where Av(z) is the independent, variable and F)U/(A) means the representation (3). Under
an infinitesimal variation, 6/4,,, (11) gives and that j*1 is the Noether current associated with the subgroup, A = constant.
The Lagrangian (13) defines the theory of "spinor electrodynamics". It is the simplest
example - and the most important one - of a large class of gauge theories. There is a
F"" simple method for constructing such Lagrangians that we now describe.
Suppose we are given a Lagrangian, £\(<j>,dtj>), for a set of fields (pa{x), scalars and
(12) spinors. Suppose, moreover, that this Lagrangian is invariant with respect to spacetime-
Discarding the total derivative as usual, and setting 6C = 0 we obtain the equations (2a) independent phase transformations, i.e. it describes a theory with the internal symmetry,
or, since the representation (3) is understood, the equations (4). For these equations to U(l). With respect to the transformations
make sense it is necessary to have d^ j " = 0. How can this be ensured?
4>a(x) - <t>'a(x) = e">"e 4>a(x) (17)
118
119
where B is real and qa is the [/(I) charge of 0 O , the Lagrangian is invariant, where fi is another coupling parameter,

(18) c) Quantization
Define the a/vtiiiant derivative. A full understanding of the quantum theory cannot be achieved by the standard canon-
ical methods we are using in this course. This is because one component of the canonical
V,j <prj(x) = clj, rpn(x) - i e qa a{x) (19)
momentum is missing. Consider the free Maxwell Lagrangian,
and write the new Lagrangian
L = -l-FtluPu'
(20)

i.e. replace the ordinary derivative d^a by the covariant derivative V , , ^ and add the
Maxwell term. Tile Lagrangian (20) is invariant with respect to the gauge transformations The canonical momentum conjugate to A^ is given by
fi f

(21) dc = 0,
d(8QA0)
To prove this we have only to show that VM <pa is indeed covariant, viz.
Clearly we cannot impose the usual equal time commutation rule
- i e qtt

A - ie
Since the time derivative of Ao does not appear in the Lagrangian we cannot treat Ao
= r:1""* V , , «j>o as an independent variable. This is a general problem with gauge theories and there are
many ways to solve it. An elaborate methodology, known as BRST quantization, has
since the terms containing <9M A cancel. been developed for treating gauge theories. This method is particularly useful for the
non-Abelian gauge theories and will be discussed in Part 2 of this course. The problem
Another important example of a gauge theory is scalar electrodynamics, originates in the gauge invariance which itself reflects the fact that A^ carries redundant
(22) information. Here we shall use an old approach to the probiem which is adequate for
— tnu t -T V,,0 V 0 — M <U tp — — 10 ©1 the U{\) gauge theories. It involves modifying the Lagrangian so as to break the gauge
4 4 invariance.
where 0 is u complex scalar and Consider the Lagrangian

0* = (9M + ieA^ip* . (23)

«")--; (25)
(The (7(1) charges here are = l,q(<j>'} = - 1 . ) In this model the electromagnetic
£«
current is given by where a is a free parameter. This Lagrangian does not have the full gauge invariance
because
3V = i e(0' \ A» - d» A" + a3 A .
(24) In the limit a —• oo, however, this Lagrangian reduces to the original invariant one. We
shall therefore quantize this modified theory and examine what happens in the limit. In
The coupling of A,, to charged fields that is generated by the prescription dip —> V4> is
the modified theory 7r? is unchanged but
known as "minimal" coupling. It is the least that can be done to obtain consistent equa-
tions. In addition, however, one may include non-minimal couplings by adding further
gauge invariant interaction terms to the Lagrangian. For example, in spinor electrody-
= - I d A"
a
namics one could add the magnetic moment interaction,
so the standard methods should work.

120 121
 In the presence of current, J)L, the modified equations of motion are
P,-q
d2 A^+f-- l\ d,t dv A" = -./„ . (26)

These equations are independent and compatible even if Ju is not conserved, They are v
solved formally by P, " Q P,

= (ie)2(2Tr)46(Pl+p2) (34)
or. more precisely, by
Note the matrices 7" and 7". Matrix multiplication is understood in the product,
(27)
S(Pl)r S(q)Y S{-p2).
(x - x') J V Secondly the 2 fermion + 2 photon function at n — 2,
lr<;
A)t mOui * solutions of the homogeneous equation and Dvv is the Feynman prop-
K
agator for the vector potential, \ X
dk + —+~i—t—1—<—
Ax - x') = / (28)
(2;r)4 k2 +
•"1 "i '« z
The asymptotic fields Am<oul are quite complicated in this theory and we shall postpone
discussing them, and the associated problem of finding the LSZ reduction formulae. In DKll(Pi) DkL,(Pi)
order to compute Green's functions we need only the propagator (28). Just as in the
(35)
earlier lectures we can obtain Wick's theorem in the form
(0|T (<:'!dx -'"I'M"^) |0) = c~"Jdx **' •'"<I) D-l*-*'> •/"<*'> (29) Again matrix multiplication is implied. In both (34) and (35) the indices fj. and v are
summed over.
where J''(x) is an arbitrary external source. Alternatively, for a polylocal function
Exercise
'2 Y. jntfer "..-<'<- i
-)| A = o - (30) 17,1) Find the propagator for A^ in the Coulomb gauge, djA} = 0. This gauge choice
Feytimau rules for Green's functions are like those discussed previously except that we can be implemented by adding the Lagrange multiplier, C d3 A,, to the Maxwell
now have "photon'' lines term, treating C(x) as a new field.
v^wvuv* = D^x - x) (31)
or, in momentum space,

(31')

In spinor electrodynamics the interaction Lagrangian is

£.*((!/) = K iiv) 4( (32)

and the interaction vertex is represented

= i e 7" . (33)

To illustrate the rules we give firstly the n = 2 contribution to the fermion 2-point
function,

122 123
18 Electrodynamics (II) Since Wj(t)' = vfj(-i), this Hamiltonian is real. Notice, however, that it is not of the
usual oscillator form. To see this clearly we pick out the modes corresponding to the wave
In the last lecture we gave a prescription for quantizing the Maxwell theory. The pre- vectors
scription is to modify the Lagrangian, C —> C,,, breaking the gauge invariance, and apply, i=(0,0.±^), u,- = |fc|
standard canonical methods. This is known as "fixing a gauge" and there are many
different ways to do it. What we need now is to find ways of extracting gauge inde- for which (4) gives
pendent information, i.e. quantities that are the same in all gauge fixings {for example,
independent of the parameter, a).
The basic idea here is that of '"physical state". The Hilbert space of the modified
theory contains a sector of physical states and a complementary space of unphysical - TT' A
0A03)
(5)
states. Operators corresponding to observabies of the theory should be gauge invariant
and they should act entirely within the physical sector. Non-invariant operators, such The two transverse components, {n^Ai) and (TT,, A2), give oscillator terms and we shall
define the corresponding annihilation operators in the usual way,
as A^x), wiil carry physical states into unphysical ones and vice versa. The details of
the unphysical sector depend on how the gauge is fixed and are not very interesting. The
physical sector includes states representing the electromagnetic quanta (photons) and we
0.2 = <JjA-z + lit? . (C)
must, find the appropriate LSZ reduction formulae to compute matrix elements involving
incoming or outgoing photons. But the other two components (w3, A3) and {ira, Aa) are completely different. The oscilla-
Finally, it must be shown that physical sector matrix elelments of gauge invariant tor potentials, m2A\ and OJMJJ, are absent and the eigenvalue spectrum will therefore not
operators are in fact independent of the choice of gauge fixing. resemble an oscillator spectrum.
The Heisenberg equations of motion are obtained from (4) using the canonical com-
a) Physical and uuphysical states mutation rules,
Canonical quantization proceeds in the usual way. In the gauge specified by the
Lagrangian etc. They are
Ca{A) = -- F^ F"" - —(d^)2 , (1) di Aj = itj - ikj An
4 la
dt Aa ~ —Q7r0 ~ tkj Aj
the conjugate momenta are
==
Cft TTj \f± Vjf ^ KjKf) J\% — lKj TTQ
'f- = d0 A, - d3 Ao dt vf0 = -ikj TTJ (7)

and they can be solved with a little effort. The results are
(2)
d(d0A0) a
and the Hamiltonian is
Aj{k, 0 = Aj cosut + (iij - ikj Ao)
Ha = I d3x (TTJ % A3 + 7T(1 d0 A o - Ca) i , 1— QI sinajt\ 1
—tcosuit -\ I >
I
= 2
j d'x (- TTj - - JTO + -{dj At - dt Ajf - A$ dj itj + ir0 dj A3 (3) (8a)
sin a1/
It is convenient to change over to Fourier transforms, writing ,0 = AfjCOSUlt - ((

i 1-a sinu;t \ 1
{"" ("' (8b)
• — (

I S i n W + JW7To ( —ECOSWi -I I ?

where k • x = k3x' = —kjX1. The Hamiltonian (3) now reads 7Tj(fc, 0 = 7TjCOSUt — |lfcj 7f0 + it?
sinwt (8c)

2 '
7T0(fc, 0 = 7T0COSWt-ikj W, (Sd)

+?: A'o kjTTj-i TTQ kj A, . (4) where u = \k\ and the coefficients A^TTJ,... on the right-hand side are evaluated at t = 0.
Notice the peculiar, non-periodic, time depenence in (8a) and (8b). This results from the

124 125
noil oscillator structures in Ha. (Notice also the big simplification if a = 1, a gauge If the original interacting theory was gauge invariant then the current must be conserved
choice that is often used.) Neither the u-dependent nor the non-periodic terras appear as a result of the other equations of motion. Hence (16) tells us that 0t, A* is a free
in the gauge invariant quantities, FIH,. field even in the interacting theory. The solution (12) must be true in general and the
definition (14) makes sense. If the in state is physical then so will be the out-state. The
sin cjt S -matrix will act within the physical sector.
F}e{k,t) = — i(kj He — ki ij) (9a) The physical sector K a Fock space like the others we have been dealing with in
previous lectures. To prove this we must construct the photon creation and annihilation
F«Ak,t) = (9b)
operators and show that they commute with i)lt A1'. As a first step consider the operators
The general solution (8) of the Heisenberg equations of motion for the free, gauge
fixed theory, is surprisingly complicated. To interpret it we must now define the physical (tdt j(k, t) = iri" k} Ao)
states. We are guided by the requirement that Maxwell's equations should hold in the (idt 0(k, t) = t-r^
physical sector. Consider the equation OjE, — 0 (Gauss law). From (9b) and (8c) the
where the dots represent linear cominations of 7i"0 and k} TT, that need not be written
Fourier transfrom of djE} is given by
explicitly since they commute with dh J4" and they vanish in the physical sector. Equations
(17) come directly from (8a) and (8b) using the formulae
-ikj FOj = —ikj 7Tj(fe,t)
{idt +Lj)co&ijt = we'"1, (i0t + w)sinwt = itje^"^' .
= — ik3 frj (k) cos wt — na (k)u sin u>t . (10)
It is easy to verify that the linear combination,
This suggests that the physical states of the free Maxwell theory should be eigenvectors
of S 0 (i) and kjirjik) corresponding to the eigenvalue zero. These operators all commute
with each other so that their eigenvectors are common. It therefore makes sense to define
k, A,) w, + k} A,,)} (IS)
the physical states by
is physical, i.e. it commutes with i)^ A*1, if and only if the polarization vector f satisfies
Jf o (i)| phys) = 0 and k3 Sj(fc)| phys) = 0 . (11)
* • „ € * = <]• (19)
Another way of writing these equations is more suggestive. From (2) and (8ci) we obtain
There are three linearly independent solutions of (19) but one of them must be E, ' ~ k>l.
1
0IL A"{k,t) = duAa + ik, Aj
On substituting E,1' = k^ in (18) wu obtain an operator that vanishes in the physical
= -a 7ru(fc) cosut + i a hj Jfj(fc) ' (12) sector. Hence there are two significant solutions, ^(k. A), A = ±1, for which we choose
orthonormal spacelike vectors,
so that 0,, AIL{x) is a linear combination of the commuting operators w0 and k3 x,-. For
arbitrary points, x and x', we have £''(k,X)* £;.(*> ^ ' ) = -6>,y - (20)

The corresponding photon annihilation operators are defined

(13)
a(k, A) = -t1"'' E'*(k,\)'(idi +LJ) A^(k,t)
and the physical sector is defined by
= £j'(uj Aj + m}) - E,') kj Aj + ... (21)
6V A>l{x)\ phys) = 0 (all i") . (14)
They reduce to the form (6) in the physical sector if k is directed along the 3 axis. They
It is easy to check that t)M A*1 commutes with the field strength tensor F^ which means satisfy the usual commutation rules,
that F^ acts within the physical sector as required of all gauge invariant operators.
Going back to the interacting theory for a moment, the Heisenberg equations of motion [a{k, A), a(£, A')+] = 6Xy 2ij(27r)3 S:,(k - &') .
include the equations
Finally, the LSZ reduction formulae for the interacting theory are derived as in earlier
/.-*" + - d" a, A" = -f (15) lectures,
a
which imply, in particular,
A") = - c (16)

126 127
 Applying Wick's theorem, and considering only the photon propagators.

*. A ) " <c> (22a) < • • • »

Hence, to first order in 6a, this implies

(22b)

We can use these formulae to extract asymptotic state matrix elements from the = - - jdx dx' 6D^(x - x') (G\T(j>l (x)f(x') .. . ) \ Q ) c o (25)
Green's functions of Lecture 17. In momentum space each external A^ line carries the
factor where we have ignored any .^-dependence in the fields symbolized by 0{zj). (The formula
(25) is valid in spinor electrodynamics because £, nl is linear in A,
d
On multiplying by a polarization vector this gives = J"(V)

but this is not true in scalar electrodynamics where (25) would be modified.)
When bDuv in (25) is replaced by the expression (23) we can integrate by parts arriving
Applying the formulae (22) then removes the factor -i/k2 and we are left with the simple at the quantities
rule: amputate the photon propagators and contract with physical polarization vectors. d» (0\T(j"(x)...)\(l)con
which can be simplified using current conservation. This is the important topic known as
b) Gauge dependence and Ward identities
Ward identities, which we now discuss.
Gauge dependence in Green's functions and matrix elements derives from the a- Consider firstly the 2-current case
dependent longitudinal part of the photon propagator, Duv. We have just seen that the
propagators associated with external photon lines are amputated in the passage to physical
state matrix elements. Hence we can ignore their contribution to gauge dependency. What - x'a) f(x) f(x') + 8(x'o - x0) f(x') j"
about the photon propagators associated with internal lines? (x) ftf))+6{xo - £K)\3°(x),f(x')\
Under a small change of the gauge parameter, a, the propagator changes according to = 0. (26)

The result is zero because the current is conserved, 9ujM = 0, and because j"(i') is
electrically neutral, i.e. commutes with j°(x) at equal times. This can be generalized to
any number of currents,
A(x — x") (23)
c^ TW{x) j"(yCi... j"( j/n)) = 0 . (27)
where A (x — x') is a scalar whose precise structure is not of any interest. To see how
Green's functions respond to the variation (23) we go back to the perturbation series, Now consider a charged field, ty(x'). As before, one writes out the T-product using
^-functions and finds, using canonical commutation rules,
fo) •.. s)\o)m
iN
{0\T(<t>in{xx),.. 4n.
= -e &i{x - x') M>{x') (28a)
where 4>(xi)... is generic, standing for any set of fields. In spinor electrodynamics the
interaction Lagrangian is simply, and, similarly
T(f{x) i>{x')) = +e 6A(x - x') t(x') . (28b)
A* = fiv) (24)

128 129

T
 Consider the 3-point problem. dtl T(J'1{T) tjj{y) 4>{y'))- To save writing, we start by 19 Compton scattering
taking i/0 > l/o so that the T-product contains only three terms,
To illustrate the application of some of the formulae developed in these lectures we consider
T(jll(x) M'!j) My')) = 0{rn - y0) / ( * ) My) My') the historically important case of electron photon scattering in the lowest approximation.
+f)(,Al - ,r0) 9{xa - y'c) Mv)
+6[y'a - xa) My) My1) j"(x) a) In spinor electrodynamics there are two second order graphs.

Hence, using canonical rules, x a

0M T{f(x) My) My')) = K^ - yo){J^

\ a, \ tv v i M

(-e 6i{x - y) + e 6<(x - y')) My) My')

These graphs are the same as those of Lecture 17 but we have changed the labelling,
for j/o > ?/o- Going through the same steps with j/o < y'a we arrive at the complete result
(Pi.pj.p3.p4) -^ (p\~p,k',-k)
o,. T(f(x) My) My')) = (-*• -y) - y')} T{My) My')) • (29) so that 4 momentum conservation now reads
Generalization is immediate:
p' + k' =p + k .
oit T(f{x) f(x')... MyO My-,) • • • My\) My'*) •••) = All particles carry positive energy. The Green's function corresponding to these graphs
= (-e. E *4(x - ya) + e E * 4 (i - y^)} T(f{x'). .. M ) • • • Mv\) • • •) • ( 3 0 ) is given by (17.35),
The collection i>f formulae (26) (30) are operator identities derived from current con- f(2TT) 4 V P ' + k ' - p - k)
servation and canonical commutation rules. One can take matrix elements of these iden- •S(p') ( 7 "S(p' + fc'b" S(p) . (1)
tities to obtain various Ward identities.
An immediate appliction of these ideas is the proof that 5-matrix elements are gauge According to the rules developed in earlier lectures we can extract the S matrix el-
independent. From the general formula (25) we obtain, using LSZ, the response of S- ements by amputating all external line propagators, replacing them with polarization
tnatrix elements to changes in the photon propagator, spinors and vectors,
(p'X\ k'a' out|pA, k(T in) =
A(out|in} = - i J dx dx' WtJx - i') (aul\T(f(x) ,
= (le)2{2n)i 6A(p' + k' ~p-k)-

= I dx dx a,, A(i - x ) 9(4 j"(a;'))|in) •e(kW)' u(p'X'){y S(p' + *'h" + -y" S(p' - kW)u.(p\)Mka)
= 0 (31) (2)

using (23) and (2C). The argument clearly generalizes to matrix elements of netural where A, A' = ±1/2 are electron helicities and a, a' — ±1 are photon helicities. The
operators, spinors satisfy the Dirac equation
«<out|T(j*{j)...)|in> = 0 . (32) 0> - m) u(p. A) = 0, A = ±1/2
To summarize, electrodynamics is quantized byfixinga gauge. The gauge-fixed theory
and are normalized such that
ia treated by standard methods to obtain Green's functions and matrix elements. The
Hilbert space is found to contain a physical sector within which the observable gauge (3)
invariant, operators are acting. Matrix elements of such operators, between physical states,
arc independent of the gauge (for example, they make sense in the limiting case, a —» oo). Tile vectors are spacelike, orthonormal and orthogonal to kv,
Green's functions in general do depend on the gauge: they contain information about k" ^{k,a) = 0, o = ±l
unphysica] states.
(4)

130 131
 Factoring out the delta functions from (2) we obtain the electron-photon scattering where diY is the visual element of solid angle. Thus (9) becomes
amplitudes.
(p'\',k'n'\T(Q)\p\,ka) = Ida = f dQ' dE' 2 ~ ' +• E' - m - E)\Tf
.! J G4IT
64w
1 E'
' \T\2 • (12}
(5)

The cross sections are defined by Using (10) to evaluate the derivative.

• E' - Ecosfl
• £ ( 1 -CQS8) . (13)
where u/ = Jp'2 + m 2 and £ ' = £'|. As discussed in Lecture 12, the delta functions can
be eliminated from (6) by integrating four of the six variables p',k'. This was carried out In practice it is more convenient to use E' as an independent variable instead of 8. There
explicitly in Lecture 12 for the centre of mass frame, p + fc = 0, Now we illustrate this is a simple relation that follows from (10) on using energy conservation, a/ + E' = m + E,
calculation once again but in the lab frame, p = 0. i.e.
Choose the coordinate system such that (m-!-F - £")2 = £ " - 2 F £ " c o s 0 + E ' 2 2

On simplifying this gives the historic Compton formula,

and k" = (7) 1 1 - cos 0

(14)
E' E m
Substituting this expression for 1 - cos# into (13) gives
In this frame the flux factor, / , becomes
, C/LJ 7TT.E
(15)
dE' E'
= m E . (8)
which can be used in (12). The final expression for the lab frame differential cross sections
The integrated cross section in the frame (7) reduces as follows. is
m' (i6)
where {T) denotes one of the matrix elements (5). These matrix elements depend on the
(9) polarizations (spins) of the incoming and outgoing photon and electron. The explicit forms
AmE
could be worked out by constructing eigenstates u(p, A) and £(fc,<r) of suitably defined
where spin operators. This is quite laborious, Here we shall consider only the simpler problem
of "unpolarized" particles. This is appropriate for experiments in which the polarizations
J = <J(k - k'f + m'2 are not measured, In effect, one sums over the helicities of outgoing particles {since they
are not distinguished) and averages over the helicities of incoming particles (since they are
+ m2 . (10)
assumed to be equally likely). To carry out these steps we need some technical formulae
In obtaining (9) we have used 3-momentum conservation to integrate p'. This gives for sums over helicities.
p' = k - k!. The lab frame scattering angle is denned by
b) Polarization sums
fc •£' = £ £ ' c o s 0 (11)
The electron sums are easily obtained from the formulae derived in Lecture 7. Recall
The energy conservation delta function in (9) can be used to eliminate the integral over that u and v are eigenvectors of the hermitian matrix (7.23)
\k'\ = E', Write
H{k) = 707 • k + m -y0 (17)
d'k! = d<Y\k!\2d\k'
= dtt E"2 dE'

132 133
i.e. i/n = aju and i / r = —wv. The completeness of these eigenvectors is expressed and compute the divergence.
through (7.32).
k'lt M"v = w1^' S(P' + xy? + -:" sip' -
Y("(t-X) (((i- r ( - i . A) r(-*,A) + ) = 2w 1 . (18)

Multiply both sides of this equation by ^ + H and then by u - H to obtain

. r _ „, l
(19)
'
• + jf - m v+v -p-m
Now take the matrix element between w(p'A') and u(p\). Use the Dirac equations to
Multiply by 70 from the right to obtain the covariant version of (19)
obtain the Ward identity,
Y u{k,X) u(k,X) = jt + m t; u{p'\') M"" u(p\) = 0 . (25)
A In the same way one finds the other identity.
Y v{k,X) v(k,X) = p-m. (20)
u(p'X') M«" u(pX) = U . (26)
The sum over photon hclidtics depends on how the polarization vectors are defined.
We can now develop formulae for the polarization sums. If \ is some Dirac spinor and
They arc necessarily orthogonal to the 4-momentum, A*1, but they can also be selected
X — X+1o i s 't s adjoint then the sum of helicity components is
orthogonal to some fixed timelike vector, it*1. Thus, if we require

= x(j> + m)x (27)

then the polaii/tition sum is given by on using (20). Suppose V is a vector that satisfies kfl V — 0 Then the sum over helicity
l components is
k^i,, + kviip ni
(21)
fcn (kn)
It is necessary to show that amplitudes and cross sections do not depend on the choice of
»(J. This brings us back to Ward identities.
(28)
The amplitudes (5) are related by L3Z to matrix elements of the electromagnetic
current. For example, on using (21). We can now apply this method to the matrix elements (5),
{p'\'.k'o'\T(a)\p\,krj) =^{k\u')" (p'X'\f{0)\p\,ka) . (22)
\(T)\2 = u(pX)
But the electromagnetic current is conserved, hence

*;<p'A'|j*(0)|pA,to> = 0 . (23)
(29)
T h t amplitude is therefore unaffected by the replacement
where the adjoint matrix M*1" is defined by
A
^- &+K
AT" =
and since by this we can always bring ^ to satisfy the auxiliary condition vf £M = 0, we
= +ie 2 (7" S{p' +fc')V+ 7"
shall do it.
= Mwti . (30)
The formulae (22) and (23) are valid to all orders but it is instructive to see how (23)
is realized m the second order approximation (5). Define the matrix, We have used a simple property of the Dirac matrices,
S(p' S{p' - (24)
7 o ( 7 w 7 W • • - T^n ) + 7 o = 7 ^ • . . (31)

134 135
which can be easily proved using the representation (7.27). To evaluate the trace (29) is where rc is the "classical electron radius'" defined by the (non-quantum) formula identi-
a long and tedious job. involving repeated use of the formulae fying rest energy with electrostatic energy

(38)
An re
and higher versions. (Computer programmes are now available for these calculations.)
In the non-relativistic regime the total cross section, obtained by integrating (37) over all
The result is angles, is a simple multiple of the classical cross section,

pal - UTIJI — ,, " • e '

(39)

At higher energies one must use (34)- (36) to compute the total cross section.
where, in the second version we have eliminated 8 using the Compton formula (14). m (E'V\E E1 (40)
<:) Klein-Nishina formula
Putting together the formulae (16) and (32) we obtain the unpolarized cross section This integrai can be evaluated exphcitly. We give only the leading term for E 3> m:

diV ( E'
£[#••••
(41)
dW ( E' .4 I® , E E \m
(33)
which tends to zero.
The factor 1 /4 in front of the polarization sum is due to averaging over the four in -states,
A = ±l/2,d = ±1.
The solid angle is given by dil' = sinS d6 dp but the unpolarized cross section does
not depend on <f so we can integrate it, writing (33) in the form
V

(34)
<lnm E ' E'
This is the Klein Nisliina formula (1929) for the scattering of photons on relativistic
elect roiis.
We should eliminate Q from (34) using (14) to write

d(coB0) = ^ dE . (35)

The energy of the outgoing photon is restricted to the range,

E > E' > ^ ^ (36)

the upper limit corresponding to 6 — 0 and the lower to 6 = v. In the low energy regime,
E « m, the recoil electron is non-relativistic and (36) implies E1 SJ E. The formula (34)
then simplifies to
domv a 7r rl(l + cos29) d(cos8) (Thomson) (37)

136 137
 r
f

Exercise 20 Ultraviolet divergences

Some of the loop integrals diverge at large momenta. This "ultraviolet problem" can be
l'J.l) Compute the cross section for the scattering of unpolarized electrons from a sta- traced back to the local structure of the equations of motion. This structure in turn was
tionary chars". %<'•• working to first order in the interaction, based in the need for relativistic eovariance. It is generally believed to signal a failure in
our understanding of short, distance physics. Fortunately, however, there exists a small
Ci,t, = W "rfl <P class of models - the renormalizable field theories - where the divergences are confined to
a finite set of parameters. In these theories it is possible to make many predictions that
a) Obtain the 5- matrix element are independent of the short distance effects,
We give a brief description of the general features of the ultraviolet problem including
how to identify the dangerous graphs and isolate their divergent parts. We give some
(p'X out|pA in) = i 27r 6((J - CJ) —— u(p'\')-ya «(p> -M examples. A sketch of the renormahzation programme and its ideology concludes the
lecture.
w h e r e q — p' — p.
h) Derive the lorniula for the polarized cross section a) Divergent graphs

Divergences arise in loop integrals. It is therefore a good idea to restrict attention to

da = {^- - \ [u(p'A')7o u(p\)\2 dil' . parts of graphs that contain loops. Ignore external lines or other lines that are not in a
loop, i.e. consider only graphs that are connected, amputated and irreducible. Irreducible
means every line is in at least one loop. Any contribution to a Green's function can
c) Sum and average to obtain the unpolarized cross section (Mott formula)
be broken down into a product of factors associated with irredducible graphs. The full
collection or irreducible graphs with a given set of external lines defines a new kind of
dahl,ltl = (~ 1) [4^ - q1) dil' . Green's function known as the irreducible vertex function.
To find the degree of divergence in irreducible graphs it is helpful to use dimensional
analysis. Let us tise a natural system and measure everything in energy units or powers
d) Express this result in terms of the scattering angle.
of energy units. For example, time is measured in (energy units)"1; it has dimension. —1.
We start from the remark that action has dimension zero, hence the Lagrangian density
has dimension four,
[£] = 4 (1)
since the integration measure dAx has dimension, - 4 . We can easily work out the dimen-
sion of everything in any given model such as

£ = iy

Masses and spacetime derivatives all have dimension 1,

[M] = M = [du\ = 1

Since [dfpdip\ = 4 it follows that

[0(x)} = 1 (3)
Since [ij>d-<l>] = 4 it follows that
[i."(^)j = 3/2 (4)
Now that the dimensions of tj> andtyare known we can examine each term in the interaction
Lagrangian and determine the dimensioris of the coupling parameters,

139
 - m2)((q + k)? ~ m2)
Is! = W = o
[ft] = i
[G] = -2 (5)
where we have expanded the integrand in powers of Ijq1 to expose its large q behaviour.
Next, the dimension of Green's functions is given by The
The integral
integral clearly
clearly diverges. To make sense of such a contribution we need to impose
2
an ultraviolet
an ultraviolet cutoff
cutoff |gM < A in which case the integral is proportional to A , (quadratic
[(O|T(0(j'i) • • • t'(A).. .)|0» = +- divergence)
2 ) ^ = 4, £^ = 0 (D r =0)
where £0 and £^ are the numbers of scalar and spinor fields (external lines) respectively.
The Fourier transform has dimension
J - kx)S{q - k, - k3)S(q - kt - k - k3))

2 3
~ (nA

(6)
The integral diverges, but only logarithmically.

because each integral d4x has dimension, - 4 . Finally, we can extract the irreducible part
of this Green's function. Firstly, amputate the propagators associated with external lines:
- remove E0 factors A(k) = i{k2 - M 2 )" 1 * i \
remove E$ factors S(k) = i(/c - m)~l - ky)S(q)S(q + - k, - k2)
where [A(/t)] = - 2 and [S(Jfc)] = - 1 . Next, remove the energy-momentum conserving
delta function, noting *3 = 9^
[A.(fci + - . . ) ] = - 4
Finally, discard all graphs that are not irreducible. In this way one arrives at the irre- which converges.
ducible vertex function T(ku ...) with E$ scalar and E4, spinor external lines. Its dimen-
In each example we see that the large q behaviour of the integrand is ~ qD~*. If the
sion is given by
integral is regularized by cutting off at the scale A, then
[F(V..)1 = - 3 ^ - ^ 4
- ADtnA

so that D is a measure of the degree of divergence. Although the examples all involve one
loop only, the phenomenon is general.
This is a general result. Any connected, amputated graph with E$ external bosons and It is perhaps less clear that the "most divergent part" is always a constant, independent
of the external momenta. This is strictly true of the example 2 where the divergence is
Et, external fermions corresponds to a function of dimension, DT. (It is instructive to
logarithmic. In example 1, where the divergence is quadratic, the divergent terms are of
obtain the formula for Dr in spacetimes of arbitrary dimensionality, i.e. with [£] = d.)
the form ~ A2 + k?(nA, i.e. there is a less divergent piece proportional to k2. Taking
Since the contributions to r(Jfci,...} are obtained by integrating products of propaga- derivatives with respect to the (external) momenta or masses (or coupling parameters with
tors, rational functions of the loop momenta, we can take DT as an indicator of divergence. positive dimension if there are any) will generally lower the dimension of the function and
If Dr > 0 we expect the integrals to diverge but, if Dr < 0 they should coverge, at least make it less divergent.
superficially. To see how this works, consider a few examples:
In other words, the divergent or cut-off dependent terms in the irreducible vertex
= 0 (Dr = 2) functions are polynomials of degree Dr in the external momenta. They appear only in
those F for which Dj- > 0, i.e.
£\ + Iff, < 4 (8)

140 141
a very short list. The cut-off dependent terms can now be cancelled by adding appropriate
counterterms to the Lagrangian.

b) Renormalizability
• • +• *—V/A--» + •--•

The perturbative contributions to an irreducible vertex function define the expansion

in powers of coupling parameters. Successive terms correspond to graphs with more
vertices (more loops). The nature of this expansion depends strongly on the dimensions A'(fc) = A{k) + A(k)itl{k)A(k) + A(k)in(k)A{k)iU(k)A(k) + ... (9)
of the coupling parameters. There are three general categories according to whether a where the shaded circle represents the sum of all irreducible graphs with two external
coupling dimension is positive, zero or negative. scalar lines,
If the dimension is positive (like h<pz in (2)) then the successive terms have lower
dimension and so are more convergent. Divergences will be limited to some finite order.
Such interactions are called "super-renormalizable".
ff the coupling dimension is zero (like Xip* or gftpip) then the successive terms all have -O -- + --*_>- + •••
the same dimension, the same degree of divergence. These interactions are "renormaliz-
ablo". (10)
If the coupling dimension is negative (like G(^>t^)2) then the successive terms have The geometric series (9) can be summed,
higher dimension, are more divergent. These divergences are uncontrollable because they
appear in every vertex function. Even if D\- < 0 so that the lower order graphs converge,
it is always possible to make graphs with a sufficient number of G's to give an effectively
positive dimensionality and hence be divergent. Such interactions are "unrenormalizable". A'(fc) = - iTl(k)A(k)y
It can happen that unrenormalizability arises in a more subtle way. For example in
field theories containing a massive vector, the term kPA^A", which does not look like an
interaction, in fact gives rise to an effective coupling parameters, 1/M2, of dimension - 2 .
This makes the theory unrenormalizable. The quantization of vector meson theories was This suramability illustrates the role of irreducible vertex functions, of which U(k) is
•i puzzle for many years until the discovery of the spontaneously-broken-gauge symmetry an example. Generally they can only be obtained in approximate form by evaluating
(Higgs) mechanism. individual graphs. Stringing them together to make the connected Green's functions is
At the present time, the only known renormalizable field theories in 4-dimensional usually a trivial algebraic operation. No further integration is involved.
spacetmie are built from scalars, Dirac spinors and guage vectors such that the Lagrangian The assumption made in deriving LSZ formulae is A'(Jfc) w A(fc) near k2 = M2, or
contains no parameters of negative dimension. In addition to the kinetic and mass terms of precisely
the free field theories the interactions may include Yukawa, ip<jnl>, minimal gauge, ip p^s, j ™ ((Jfc3 - M2)A'{k)) = i (12)
|V0| 2 , scalar cubic. 03, and quartic, 04, Nothing else.
This implies that the irreducible part must be a function of the form
r) Renormahzation
II(ft) = (k2 - M2ff(k) , [/(*)] = - 2 (13)
In deriving the LSZ reduction formulae in Lecture 9 it was tacitly assumed that the 2 2
where f(k) is regular at fc = M . The condition (13) must be enforced.
masses of quanta in the asymptotic states are the same as the masses that appear in
the Lagrangian. In effect, it was assumed that the asymptotic fields, 4>in and tfw, are The integrals (10) that contribute to fl(fc) are quadratically divergent since [H(k)] = 2.
canonically normalized free fields with masses determined by the free Lagrangian. This We shall therefore add to the interaction Lagrangian a counter-term
assumption needs to be justified.
Since the LSZ prescription requires the amputation of propagators associated with
external lines followed by the taking of a limit p2 —> M 2 , etc. it means, in effect: extract where a and b are chosen so as to cancel the cutoff dependent terms in IT. The contribution
the residue of the pole in the external variable p2, for each external line. Hence we of AC to the irreducible 2-point graphs includes the shift
have been assuming that the full Green's functions have poles exactly at the free particle
rnas.sses. in general this looks implausible. The loop graphs will modify the propagators n{Jt) -* n(Jt) + ak2 + b (15)
and shift their poles. Consider, for example, the full, connected 2-point function, A'(fc), so that it should be possible to eliminate the cutoff dependence in this way. At the same
for the scalar field, it is given by the graphs time, we can choose a and b so that fl and its first derivative vanish at the point fc3 = M2,
i.e. we can indeed enforce the normalization condition (13).

142 143
 The coefficients a and 6 are expressed as power series in the coupling parameters. They
Hi-fi determined in principle by demanding that the cutoff dependence in II be cancelled
lit each order. Of course, the lower order terms in A£ will make contributions in the
internal lines in ir. These must be included along with the shift (15).
Tlif same can be done with all 2-point functions. When the correct pole structures
are enforced by the addition of appropriate counterterms, they wil! be represented by
ultraviolet convegent integrals.
Further counterterms are needed to cancel the divergences of irreducible graphs with
more than two external lines. The functions with D > 0 are very limited:
(E6,EV) = (1.2), (3,0), (4,0)

Their cutoff dependence will be cancelled by /\g, Ah and AA respectively.

There are many different ways to choose counterterms. Different choices give different
parametrizations of the irreducible vertex functions. The difficulties involved in comparing
one prescription with another can be non-trivial and lie beyond the scope of these lectures.

Exercise

20.1) Compute Dr, [g], [li],[A] in d-dimensional spacetime.

20.2) Derive the Ward identities for irreducible vertex funcitons in spinor electrodynam-

144