Sie sind auf Seite 1von 207

Introduction to Path Integrals in Field Theory

U. Mosel1
Institut fuer Theoretische Physik, Universitaet Giessen
D-35392 Giessen, Germany
SS 02
July 6, 2002

1
http://theorie.physik.uni-giessen.de
1

This manuscript on path integrals is based on lectures I have


given at the University of Giessen. If you find any conceptual
or typographical errors I would like to learn about them.
In this case please send an e-mail to
mosel@theo.physik.uni-giessen.de
Contents

I Non-Relativistic Quantum Theory 6


1 PATH INTEGRAL IN QUANTUM THEORY 7
1.1 Propagator of the Schrödinger Equation . . . . . . . . . . . . 7
1.2 Propagator as Path Integral . . . . . . . . . . . . . . . . . . . 10
1.3 Quadratic Hamiltonians . . . . . . . . . . . . . . . . . . . . . 13
1.3.1 Cartesian metric . . . . . . . . . . . . . . . . . . . . . 14
1.3.2 Non-Cartesian metric . . . . . . . . . . . . . . . . . . . 15
1.4 Classical Interpretation . . . . . . . . . . . . . . . . . . . . . . 17

2 PERTURBATION THEORY 20
2.1 Free propagator . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Perturbative Expansion . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Application to Scattering . . . . . . . . . . . . . . . . . . . . . 27

3 GENERATING FUNCTIONALS 32
3.1 Groundstate-to-Groundstate Transitions . . . . . . . . . . . . 33
3.1.1 Generating functional. . . . . . . . . . . . . . . . . . . 37
3.2 Functional Derivatives
of Transition Amplitudes . . . . . . . . . . . . . . . . . . . . . 38

II Relativistic Quantum Field Theory 43


4 CLASSICAL RELATIVISTIC FIELDS 44
4.1 Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . 44
4.1.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2 Symmetries and Conservation Laws . . . . . . . . . . . . . . . 50
4.2.1 Geometrical Space–Time Symmetries . . . . . . . . . . 51
4.2.2 Internal Symmetries . . . . . . . . . . . . . . . . . . . 53

2
CONTENTS 3

5 PATH INTEGRALS FOR SCALAR FIELDS 56


5.1 Generating Functional for Fields . . . . . . . . . . . . . . . . . 57
5.1.1 Euclidean Representation . . . . . . . . . . . . . . . . 59

6 EVALUATION OF PATH INTEGRALS 62


6.1 Free Scalar Fields . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.1.1 Generating functional . . . . . . . . . . . . . . . . . . . 62
6.1.2 Feynman propagator . . . . . . . . . . . . . . . . . . . 64
6.1.3 Gaussian Integration . . . . . . . . . . . . . . . . . . . 68
6.2 Stationary Phase Approximation . . . . . . . . . . . . . . . . 71
6.3 Numerical Evaluation of Path Integrals . . . . . . . . . . . . . 74
6.3.1 Imaginary time method . . . . . . . . . . . . . . . . . 74
6.3.2 Real time formalism . . . . . . . . . . . . . . . . . . . 75

7 S-MATRIX AND GREEN’S FUNCTIONS 78


7.1 Scattering Matrix . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.2 Reduction Theorem . . . . . . . . . . . . . . . . . . . . . . . . 80
7.2.1 Canonical field quantization . . . . . . . . . . . . . . . 80
7.2.2 Derivation of the reduction theorem . . . . . . . . . . . 82

8 GREEN’S FUNCTIONS 87
8.1 n-point Green’s Functions . . . . . . . . . . . . . . . . . . . . 87
8.1.1 Momentum representation . . . . . . . . . . . . . . . . 88
8.1.2 Operator Representations . . . . . . . . . . . . . . . . 89
8.2 Free Scalar Fields . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.2.1 Wick’s theorem . . . . . . . . . . . . . . . . . . . . . . 91
8.2.2 Feynman rules . . . . . . . . . . . . . . . . . . . . . . . 92
8.3 Interacting Scalar Fields . . . . . . . . . . . . . . . . . . . . . 94
8.3.1 Perturbative expansion . . . . . . . . . . . . . . . . . . 96

9 PERTURBATIVE φ4 THEORY 99
9.1 Perturbative Expansion of the Generating Function . . . . . . 99
9.1.1 Feynman rules . . . . . . . . . . . . . . . . . . . . . . . 101
9.1.2 Vacuum contributions . . . . . . . . . . . . . . . . . . 102
9.2 Two-Point Function . . . . . . . . . . . . . . . . . . . . . . . . 103
9.2.1 Terms up to O(g 0 ) . . . . . . . . . . . . . . . . . . . . 103
9.2.2 Terms up to O(g) . . . . . . . . . . . . . . . . . . . . . 104
9.2.3 Terms up to O(g 2 ) . . . . . . . . . . . . . . . . . . . . 107
9.3 Four-Point Function . . . . . . . . . . . . . . . . . . . . . . . 109
9.3.1 Terms up to O(g) . . . . . . . . . . . . . . . . . . . . . 109
9.3.2 Terms up to O(g 2 ) . . . . . . . . . . . . . . . . . . . . 110
CONTENTS 4

10 DIVERGENCES IN n-POINT FUNCTIONS 113


10.1 Power Counting . . . . . . . . . . . . . . . . . . . . . . . . . . 114
10.2 Dimensional Regularization
of φ4 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
10.2.1 Two-point function . . . . . . . . . . . . . . . . . . . . 117
10.2.2 Four-point function . . . . . . . . . . . . . . . . . . . . 118
10.3 Renormalization . . . . . . . . . . . . . . . . . . . . . . . . . . 122
10.3.1 Renormalization of φ4 Theory . . . . . . . . . . . . . . 125

11 GREEN’S FUNCTIONS FOR FERMIONS 128


11.1 Grassmann Algebra . . . . . . . . . . . . . . . . . . . . . . . . 128
11.1.1 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . 130
11.1.2 Integration . . . . . . . . . . . . . . . . . . . . . . . . 132
11.2 Green’s Functions for Fermions . . . . . . . . . . . . . . . . . 138
11.2.1 Generating Functional for fermions . . . . . . . . . . . 138
11.2.2 Green’s Functions . . . . . . . . . . . . . . . . . . . . . 141

12 INTERACTING FIELDS 144


12.1 Feynman Rules . . . . . . . . . . . . . . . . . . . . . . . . . . 144
12.1.1 Fermion Loops . . . . . . . . . . . . . . . . . . . . . . 145
12.2 Wick’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 148
12.3 Removal of Degrees of Freedom:
Yukawa Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 150
12.3.1 Perturbative Expansion . . . . . . . . . . . . . . . . . 152

13 PATH INTEGRALS FOR GAUGE FIELDS 157


13.1 Gauge invariance in Abelian theories . . . . . . . . . . . . . . 158
13.2 Non-abelian gauge fields . . . . . . . . . . . . . . . . . . . . . 162
13.3 Path integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
13.3.1 Gauge Fixing . . . . . . . . . . . . . . . . . . . . . . . 169
13.4 Feynman Rules . . . . . . . . . . . . . . . . . . . . . . . . . . 171
13.4.1 Faddeev-Popov Determinant . . . . . . . . . . . . . . . 171
13.4.2 Ghost fields . . . . . . . . . . . . . . . . . . . . . . . . 175
13.4.3 Feynman Rules . . . . . . . . . . . . . . . . . . . . . . 176

14 EXAMPLES FOR GAUGE FIELD THEORIES 184


14.1 Quantum Electrodynamics . . . . . . . . . . . . . . . . . . . . 184
14.2 Quantum Chromodynamics . . . . . . . . . . . . . . . . . . . 184
14.3 Electroweak Interactions . . . . . . . . . . . . . . . . . . . . . 186
CONTENTS 5

A Units and Metric 189


A.1 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
A.2 Metric and Notation . . . . . . . . . . . . . . . . . . . . . . . 190

B Functionals 192
B.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
B.2 Functional Integration . . . . . . . . . . . . . . . . . . . . . . 193
B.2.1 Gaussian integrals . . . . . . . . . . . . . . . . . . . . 193
B.3 Functional Derivatives . . . . . . . . . . . . . . . . . . . . . . 196

C RENORMALIZATION INTEGRALS 198

D GRASSMANN INTEGRATION FORMULA 202

E BIBLIOGRAPHY 205
Part I

Non-Relativistic Quantum
Theory

6
Chapter 1

PATH INTEGRAL IN
QUANTUM THEORY

In this starting chapter we introduce the concepts of propagators and path


integrals in the framework of nonrelativistic quantum theory. In all these
considerations, and the following chapters on nonrelativistic quantum theory,
we work with one coordinate only, but all the results can be easily generalized
to the case of d dimensions.

1.1 Propagator of the Schrödinger Equation


We start by considering a nonrelativistic particle in a one-dimensional po-
tential V (x). The Schrödinger equation reads

h̄2 ∂ 2 ψ(x, t) ∂ψ(x, t)


Hψ(x, t) = − + V (x)ψ(x, t) = ih̄ . (1.1)
2m ∂x2 ∂t
This equation allows us to calculate the wavefunction ψ(x, t) at a later time,
if we know ψ(x, t0 ) at the earlier time t0 < t. For further calculations we
rewrite this equation into the following form
!

ih̄ − H ψ(x, t) = 0 . (1.2)
∂t

Next, we consider the function K (x, t; xi , ti ) which is defined as solution


of the equation
!

ih̄ − H K (x, t; xi , ti ) = ih̄δ(x − xi )δ(t − ti ) . (1.3)
∂t

7
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 8

K is the “Green’s function” of the Schrödinger equation (K is also often


called “propagator”) with the initial condition

K(x, ti + 0; xi , ti ) = δ(x − xi ) . (1.4)

The solution of the Schrödinger equation (1.2) can be written as


Z
ψ(x, t) = K (x, t; xi , ti ) ψ(xi , ti ) dxi (1.5)

for t > ti (Huygen’s principle). Relation (1.5) can be proven by inserting the
lhs into the Schrödinger equation
!Z

ih̄ − H K (x, t; xi , ti ) ψ(xi , t) dxi
∂t
Z
= ih̄ δ (t − ti ) δ (x − xi ) ψ (xi , t) dxi
= ih̄δ (t − ti ) ψ(x, t) = 0 for t > ti . (1.6)

Thus the ψ defined by (1.5) is indeed a solution of the Schrödinger equation


for all times t > ti . K (x, t; xi , ti ) is the probability amplitude for a transition
from xi , at time ti , to the position x, at the later time t. The restriction to
later times preserves causality.
We can find an explicit form for the propagator, if the solutions of the
stationary Schrödinger equation, ϕn (x), and the corresponding eigenvalues,
En , are known. Since the ϕn form a complete system, K can certainly be
expanded in this basis (for t ≥ ti )
X i
K (x, t; xi , ti ) = an ϕn (x)e− h̄ En t Θ (t − ti ) . (1.7)
n

Here the stepfunction Θ(t) = 0 for t < 0 and Θ(t) = 1 for t ≥ 0 takes
explicitly into account that we propagate the wavefunction only forward in
time. The expansion coefficients obviously depend on xi , ti

an = an (xi , ti ) . (1.8)

Because of the initial condition K(x, ti + 0; xi , ti ) = δ (x − xi ) we have


X i
δ (x − xi ) = an (xi , ti )ϕn (x)e− h̄ En ti . (1.9)
n

The lhs is time-independent; thus we must have


i
an (xi , ti ) = an (xi )e+ h̄ En ti , (1.10)
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 9

and consequently X
δ (x − xi ) = an (xi ) ϕn (x) . (1.11)
n
This can be fulfilled by
an (xi ) = ϕ∗n (xi ) (1.12)
(closure relation). Thus we have a representation of K (x, t; xi , ti ) in terms
of the eigenfunctions and eigenvalues of the underlying Hamiltonian
X i
K (x, t; xi , ti ) = Θ(t − ti ) ϕ∗n (xi ) ϕn (x)e− h̄ En (t−ti ) . (1.13)
n

It is easy to show that this propagator fulfills (1.3).


In Dirac’s bra and ket notation this result can also be written as
X i
K (x, t; xi , ti ) = ϕ∗n (xi ) ϕn (x)e− h̄ En (t−ti ) Θ (t − ti )
n
X i
= hn|xi ie− h̄ En (t−ti ) hx|ni Θ (t − ti )
n
X i i
= hn|e+ h̄ Ĥti |xi ihx|e− h̄ Ĥt |ni Θ (t − ti )
n
i
= hx|e− h̄ Ĥ(t−ti ) |xi i ≡ hx|Û (t, ti ) |xi i Θ (t − ti ) .(1.14)
Thus the propagator is nothing else than the time development operator
i
Û (t, ti ) = e− h̄ Ĥ(t−ti ) (1.15)
for t > ti in the x representation. It is also often written as
i
K (x, t; xi , ti ) = hx|e− h̄ Ĥ(t−ti ) |xi i Θ (t − ti ) ≡ hxt|xi ti i . (1.16)
The notation here is that of the Heisenberg representation of quantum
mechanics. In this representation the physical state vectors are time-inde-
pendent and the operators themselves carry all the time-dependence whereas
this is just the opposite for the Schrödinger representation. For example, for
the position operator x̂ in the Schrödinger representation we obtain the time-
dependent operator
i i
x̂H (t) = e h̄ Ĥt x̂e− h̄ Ĥt (1.17)
and
x̂H (t)|xti = x|xti (1.18)
with
i
|xti = e h̄ Ĥt |xi . (1.19)
The state |xti is thus the eigenstate of the operator x̂H (t) with eigenvalue x
and not the state that evolves with time out of |xi; this explains the sign of
the frequency in the exponent.
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 10

t1

ti
x
xi x

Figure 1.1: Possible paths from xi to x, corresponding to (1.22).

1.2 Propagator as Path Integral


We start by dividing the time-interval between ti and t by inserting the time
t1 . The wavefunction is first propagated until t1 and then, in a second step,
until t
Z
ψ (x1 , t1 ) = K (x1 , t1 ; xi , ti ) ψ (xi , ti ) dxi (1.20)
Z
ψ(x, t) = K(x, t; x1 , t1 )ψ (x1 , t1 ) dx1 .

Taking these two equations together we get


Z Z
ψ(x, t) = K (x, t; x1 , t1 ) K (x1 , t1 ; xi , ti ) ψ(xi , ti ) dxi dx1 . (1.21)

Comparing this result with (1.5) yields


Z
K (x, t; xi , ti ) = K (x, t; x1 , t1 ) K (x1 , t1 ; xi , ti ) dx1 . (1.22)

We can thus view the transition from (xi , ti ) to (x, t) as the result of a
transition first from (x, t) to all possible intermediate points (x1 , t1 ), which is
then followed by a transition from these intermediate points to the endpoint.
We could also say that the integration in (1.22) is performed over all possible
paths between the points (xi , ti ) and (x, t), which consist of two straight line
segments with a bend at t1 . This is illustrated in Fig. 1.1.
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 11

We now subdivide the time interval further into (n + 1) equal parts of


length ∆t = η. We then have in direct generalization of the previous result
Z Z
K (x, t; xi , ti ) = ... dx1 dx2 . . . dxn (1.23)
× [K (x, t; xn , tn ) K (xn , tn ; xn−1 , tn−1 ) . . . K (x1 , t1 ; xi , ti )] .

The integrals run here over all possible paths between (xi , ti ) and (x, t) which
consist of (n + 1) segments with boundaries that are determined by the time
steps ti , t1 , . . . , tn , t.
We now calculate the propagator for a small time interval ∆t = η from
tj to tj+1 . For this propagation we have according to (1.16)
i
K (xj+1 , tj+1 ; xj , tj ) = hxj+1 |e− h̄ Ĥη |xj i (1.24)
∼ i
= hxj+1 |1 − Ĥη|xj i

i
= δ (xj+1 − xj ) − ηhxj+1 |Ĥ|xj i

1 Z i p(xj+1 −xj ) iη
= e h̄ dp − hxj+1 |Ĥ|xj i
2πh̄ h̄
with the representation for the δ-function
1 Z ik(x−x0 )
δ(x − x0 ) = e dk . (1.25)

We now assume that Ĥ is given by

Ĥ = T̂ (p̂) + V̂ (x̂) . (1.26)

Here T̂ , p̂, V̂ , x̂ are all operators; we assume that T (p̂) and V (x̂) are Taylor-
expandable. In this case, where the p- and x-dependences separate, we can
also bring the last term in (1.24) into an integral form. We have

hxj+1 |Ĥ|xj i = hxj+1 |T̂ + V̂ |xj i . (1.27)

First, we consider the first summand


Z
hxj+1 |T̂ |xj i = dp0 dphxj+1 |p0 ihp0 |T̂ (p̂)|pihp|xj i (1.28)
Z
= dp0 dphxj+1 |p0 iδ(p0 − p)T (p)hp|xj i
Z
= dphxj+1 |piT (p)hp|xj i .
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 12

With the normalized momentum eigenfunctions


1 i
hx|pi = √ e h̄ px (1.29)
2πh̄
we thus obtain
1 Z i
hxj+1 |T̂ (p̂)|xj i = T (p) e h̄ p(xj+1 −xj ) dp . (1.30)
2πh̄
While there is an operator p̂ on the lhs of this equation there are only numbers
p on its rhs.
For the potential part an analogous transformation can be performed

hxj+1 |V̂ |xj i = V (xj ) δ (xj+1 − xj ) (1.31)


1 Z i p(xj+1 −xj )
= e h̄ dp V (xj )
2πh̄
Again, on the lhs V̂ is an operator, while the rhs of this equation contains
no operators.
In summary, we have for the propagator over a time-segment η
1 Z i
K (xj+1 , tj+1 ; xj , tj ) = dp e h̄ p(xj+1 −xj )

2πh̄ 
iη 1 Z i 1 Z i
− dp T (p)e h̄ p(xj+1 −xj ) + dp e h̄ p(xj+1 −xj ) V (xj )
h̄ 2πh̄ 2πh̄
Z  
1 i
p(xj+1 −xj ) iη
= dp e h̄ 1 − H (p, xj )
2πh̄ h̄
Z  
−→ 1 i
η→0 dpj exp [pj (xj+1 − xj ) − ηH (pj , xj )] . (1.32)
2πh̄ h̄
Here H = T + V is a function of the numbers x and p and no longer an
operator! In the last step we have renamed the integration variable into pj
to indicate that it may be viewed as the momentum of a classical particle
moving from xj to xj+1 between times tj and tj+1 .
We now insert (1.32) into (1.23) and obtain

K (x, t; xi , ti ) (1.33)
 
n
Z Y n
Z Y n
dpl i X 
= lim dxk exp  [pj (xj+1 − xj ) − ηH (pj , xj )] .
n→∞
k=1 l=0 2πh̄ h̄ j=0

(with x0 = xi and xn+1 = x). The asymmetry in the range of the products
over x- and p-integrations comes about because with n intermediate steps
between xi and x there are n + 1 intervals and corresponding momenta.
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 13

The integrand here is, for finite n, a complex function of all the coordi-
nates x1 , x2 , . . . , xn and the momenta p1 , p2 , . . . , pn . In the limit n → ∞ it
depends on the whole trajectory x(t), p(t). Here we note that p is not the
momentum canonically conjugate to the coordinate x, but instead just an
integration variable.
In the limit n → ∞ we obtain for the exponent
n
X
[pj (xj+1 − xj ) − ηH (pj , xj )]
j=0
n
!
X xj+1 − xj
= η pj − H (pj , xj )
j=0 η
Zt
−→ dt0 [p(t0 )ẋ(t0 ) − H(p(t0 ), x(t0 ))] . (1.34)
n→∞
ti

With this result we rewrite (1.33) in an abbreviated, symbolic form

i
Rt
Z Z h̄
dt0 [p(t0 )ẋ(t0 )−H(p(t0 ),x(t0 ))]
K (x, t; xi , ti ) = Dx Dp e ti
, (1.35)
Q Q
where Dx stands for dxk and Dp for dpl /(2πh̄). The integrals here are
limits of n-dimensional integrals over x and p for n → ∞, they are integrals
over all functions x(t) and p(t) and are defined by (1.33).
Equation (1.35) represents an important result. It allows to calculate the
propagator and thus the solution of the Schrödinger equation in terms of a
path integral over classical functions.

1.3 Quadratic Hamiltonians


Even though the propagator (1.35) looks like a path integral over an expo-
nential function of the action, this is in general not the case, because

pẋ − H(p, x) = L(x, ẋ, p) (1.36)

is not equal to the classical Lagrange function since p is not the canonical mo-
mentum as already stressed above. Therefore, in general one cannot express
the path integral (1.35) in terms of the action.
Such a simplification, however, is possible for a special p-dependence of
the Hamiltonian. If H depends at most quadratically on p, then the path
integration over the momentum p can be performed and the action appears
in the exponent. This will be discussed in the next 2 sections.
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 14

1.3.1 Cartesian metric


In the last section we have made the special ansatz H = T (p) + V (x) in
which the momenta and coordinates are separated. For the special case, in
which H depends only quadratically on p with constant coefficient, e.g.
p2
H= + V (x) , (1.37)
2m
we can further simplify the path-integral (1.33)
n
Z Y n
Z Y
dpl
K (x, t; xi , ti ) = lim dxk (1.38)
n→∞
k=1 l=0 2πh̄
 ! 
i n
X xj+1 − xj p2j 
× exp  η pj − − V (xj )  .
h̄ j=0 η 2m

Using the integral relation


Z r
+∞
−ap2 +bp+c π b2 +c
e dp = e 4a (1.39)
−∞ a
for Gaussian integrals, discussed in more detail in App. B.2.1, we obtain by
performing the p-integration
! n+1
m 2
K = lim (1.40)
n→∞ i2πh̄η
  !2 
n
Z Y i n
m X xj+1 − xj 
× dxk exp  η  − V (xj ) .
k=1 h̄ j=0 2 η

Thus in this special case (H = p2 /2m + V ) the propagator K is given (again


in abbreviated notation) by

i
Rt
Z h̄
L(x,ẋ)dt0 Z
i
K (x, t; xi , ti ) = N Dx e ti
=N Dx e h̄ S[x(t)] , (1.41)

with the Lagrangian


m 2
L(x, ẋ) = ẋ − V (x) (1.42)
2
and the action Z t
S[x(t)] = L(x(t0 ), ẋ(t0 )) dt0 . (1.43)
ti

N represents the factor in front of the integral in (1.40). There is a problem


with N : the factor N is complex and becomes infinite for n → ∞, η →
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 15

0. We will see, however, later, for example in Sect. 2.1, that the whole
pathintegral leads to a well-defined expression. In addition, this problem will
be bypassed in later developments where we show that physically relevant is
only a normalized propagator in which N has been removed.
The propagator K has thus been reduced to a one-dimensional path-
integral, which is only possible for Hamiltonians which are quadratic in p.
This is a quite important result that we will use throughout all the follow-
ing sections. Eq. (1.41) shows that the propagator is given by the phase
exp h̄i S[x(t)] summed over all possible trajectories x(t) with fixed starting
and end points. This is reminiscent of the partition function in statistical
mechanics which is obtained by summing the Boltzmann factor exp (−En /T )
over all possible states of the system.
At this point we should realize that in going from (1.38) to (1.40) we have
integrated an oscillatory integrand (eif (p) ) over an infinite interval. This was
only possible by a mathematical trick: in applying the Gaussian integration
formula (B.18) we have in effect used the quantity iη in (1.38) as if it were real.
In other words: we have analytically continued expression (1.38) into the
complex plane by setting the time interval η → −iη 0 with η 0 real and positive.
Then (1.38) becomes a well-behaved Gaussian integral. After performing
the integration we have then gone back to the original η. This analytical
continuation is in general possible only if no singularities are encountered
while going to the real variable η 0 . Also, one has to worry about phase
ambiguities connected with the appearance of the squareroot of a complex
number.

1.3.2 Non-Cartesian metric


The momentum integration is even possible for Hamiltonians of a much more
general form. As an example we consider
1
H= f1 (x)p2 + f2 (x)p + f3 (x) . (1.44)
2m
Here a problem arises because the canonical quantization of such a Hamilto-
nian is ambiguous. This is so because the classical coordinates and momenta
commute, so that H can be brought into various forms that are classically
all equal, but differ after quantization because the operators p̂ and x̂ do not
commute. Even though the time development operator (1.24) could still be
evaluated for any of these various forms, these would in general not lead to
a path integral of the form (1.35).
We, therefore, turn the question around and ask:
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 16

Given a path integral

i
Rt
Z Z h̄
dt0 (pẋ−H(p,x))
Dx Dp e ti
(1.45)

with the classical Hamiltonian (1.52), can this still be identified as a propa-
gator and, if so, for which Hamiltonian?
The answer to this question is given here without proof1 :

i
Rt
Z Z h̄
dt0 [pẋ−H(p,x)]
i
Dx Dp e ti
= hx0 |e− h̄ (t−ti )ĤW |xi . (1.46)

Here ĤW is the “Weyl-ordered” Hamiltonian


1  2 
ĤW = p̂ f1 (x) + 2p̂f1 (x)p̂ + f1 (x)p̂2 (1.47)
8m
1
+ [p̂f2 (x) + f2 (x)p̂] + f3 (x) .
2
In this form the momentum- and coordinate-dependent terms are symme-
trized. Note that the path integral in (1.46) is well determined because all
quantities on the rhs of this equation are classical, commuting quantities.
The lhs of (1.46) can be simplified by performing the integration over the
momenta in (1.33)
n
Z Y n
Y dpl
K = lim dxk (1.48)
n→∞
k=1 l=0 2πh̄
 " !#
n
i X p2j 
× exp  pj (xj+1 − xj ) − η f1 (xj ) + f2 (xj )pj + f3 (xj ) .
h̄ j=0 2m 

Using again the Gaussian integral formula (1.39) this gives


! n+1 Z n n
m 2 Y Y 1
K = lim dxk q (1.49)
n→∞ i2πh̄η k=1 l=0 f1 (xl )
  h i2 
(xj+1 −xj )
 n
i X m η
− f2 (xj ) 

× exp  η
 − ηf3 (xj )
 .
 h̄ j=0 2 f1 (xj ) 

1
The proof can be found in [LEE], p. 475
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 17

The exponent (in round brackets) is


h i2
xj+1 −xj
m η
− f2 (xj )
(. . .) = − f3 (xj ) (1.50)
2 f1 (xj )
2 2
−→ m ẋ − 2ẋf2 (x) + f2 (x) − f (x) .
η→0 3
2 f1 (x)
This last expression is just the Lagrangian
m
L= g1 (x)ẋ2 + g2 (x)ẋ + g3 (x) (1.51)
2
(a kinetic term of this form appears, for example, when rewriting the kinetic
energy of a particle from cartesian into polar coordinates). For this L the
corresponding classical Hamiltonian is given by
m
H = pẋ − L = (mg1 ẋ + g2 ) ẋ − g1 ẋ2 − g2 ẋ − g3
2
!2
m m p − g2
= g1 ẋ2 − g3 = g1 − g3
2 2 mg1
1 1 2 g2 1 g22
= p − p+ − g3 . (1.52)
2m g1 mg1 2m g1
With
1 g2 g22
, f2 = −
f1 = , f3 = − g3 (1.53)
g1 mg1 2mg1
this is just the Hamiltonian (1.44) that we started out with.
We thus have for the complete propagator in this case

K (x, t; xi , ti ) (1.54)
! n+1 Z n  
n q n
m 2 Y Y i X 
= lim dxk g1 (xl ) exp  η L(xj , ẋj ) .
n→∞ i2πh̄η k=1 l=0 h̄ j=1

Thus, in this case the path integral is changed. The square root of a function
that determines the metric of the system appears in the integrand. For
Hamiltonians even more general than (1.44) an additional potential term
appears in the Lagrangian [Grosche/Steiner].

1.4 Classical Interpretation


The simple form (1.41) for the path integral allows a very physical interpre-
tation of the classical limit to quantum mechanics. For the classical path the
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 18

variation of the action is, according to Hamilton’s principle, equal to zero,


i.e. the action is stationary2
Zt Zt
δS = L (xcl + δx, ẋ + δ ẋcl ) dt − L (xcl , ẋcl ) dt = 0 . (1.55)
ti ti

This implies that all the paths close to the classical path give about equal
contributions to the path integral. For each path (C1 ) somewhat more re-
moved from the classical one there will also be another one, (C2 ), whose
action differs from that on C1 just by πh̄. Then we have
i i i
e h̄ S(C1 ) = e h̄ S(C2 )+iπ = −e h̄ S(C2 ) , (1.56)

so that the contributions from these two paths cancel each other. Sizeable
contributions to the path integral thus come from paths close to the classical
one. Quantum mechanics then describes the fluctuations of the action in a
narrow range around the classical path.
This observation forms the basis for a semiclassical approximation. This
can be formulated by expanding the action functional S[x(t), ẋ(t)] in terms
of fluctuations δx around the classical path xcl (t). This gives
!
1 Z δ2L 2 δ2L δ2L
S[x, ẋ] = Scl + 2
(δx) + 2 δx δ ẋ + 2 (δ ẋ)2 + .....
2 δx δxδ ẋ δ ẋ
2
≡ Scl + δ S + . . . . (1.57)

Here all the derivatives have to be taken at the classical path. Because S is
stationary at the classical path, there is no first derivative in this equation.
The propagator (1.41) now becomes
Z Z  
i
S i
S 1i 2
K (x, t; xi , ti ) = N Dx e h̄ = Ne h̄ cl Dx exp δ S + .... . (1.58)
2 h̄
Note that this result (without higher order terms) is exact for Lagrangians
that depend at most quadratically on x. The second factor gives the effects
of quantum mechanical fluctuations around the classical path.
An interesting observation on the character of these fluctuations can be
made. The main contribution to the integrand in (1.40) for η → 0 comes
from exponents !2
η m xj+1 − xj
 1, (1.59)
h̄ 2 η
2
For an explanation of functionals and their derivatives see App. B
CHAPTER 1. PATH INTEGRAL IN QUANTUM THEORY 19

q
i.e. from velocities vj ≈ 2h̄/(ηm) which diverge with η → 0. This implies

that the average displacement d within a time-step η is proportional to η,
so that d2 ∼ η (not d2 = v 2 dt2 = v 2 η 2 !), just as for a random walk. The main
contribution to the path integral, and therefore to the quantum mechanical
fluctuation around the classical trajectory, thus comes from paths that are
continous, but have no finite derivative.
Chapter 2

PERTURBATION THEORY

In this chapter we discuss first how to calculate the propagator of a free


particle and derive its analytic form. In most cases, however, with a potential
included the exact propagator cannot be calculated. Thus one has to resort
to perturbation theory which will also be developed in this chapter.

2.1 Free propagator


We start with the free propagator K0 , given by
Z
i
K0 = N Dx e h̄ S0
! n+1 +∞  !2 
Z Yn n
m 2
i X m xj+1 − xj
= lim dxk exp  η  . (2.1)
n→∞ 2πih̄η k=1 h̄ j=0 2 η
−∞

This path-integral can be performed exactly. With (B.19) we obtain for the
free propagator
! n+1  1
2
m 2
in π n
K0 = lim   n  (2.2)
n→∞ 2πih̄η (n + 1) m
2h̄η
!
i m
× exp (x − xi )2 ,
n + 1 2h̄η

since xn+1 = x, x0 = xi . With (n + 1)η = t − ti this becomes


s 2 r
m i m(x−xi ) m i m∆x2
K0 (∆x, ∆t) = e h̄ 2(t−ti )
= e h̄ 2∆t (2.3)
2πh̄i (t − ti ) 2πh̄i∆t

20
CHAPTER 2. PERTURBATION THEORY 21

with ∆x = x−xi , ∆t = t−ti . This is the propagator of a free particle for ∆t ≥


0; for ∆t < 0 it has to be supplemented by the condition K = 0. Because of
Galilei invariance and time homogeneity the free particle propagator depends
only on the space- and time-distances.
The last step, from (2.2) to (2.3), shows nicely how in the limit n → ∞
the infinite normalization factor combines with another equally ill defined
factor from the path integral into a well defined product in (2.2).
Since a free particle has conserved momentum, it is advantageous to trans-
form K0 into the momentum representation
1 Z − i p∆x
K0 (p, ∆t) = √ e h̄ K0 (∆x, ∆t) d∆x
2πh̄
r
1 m Z − i p∆x i m ∆x2
= √ e h̄ e h̄ 2∆t d∆x . (2.4)
2πh̄ 2πh̄i∆t
We can now use again the integral relation (B.18) in the form
+∞
Z r
−a∆x2 +b∆x π b2
e d∆x = e 4a (2.5)
a
−∞

(with a = −im/(2h̄∆t) and b = −ip/h̄) to write


r s
1 m π2h̄∆t − i p2 ∆t
K0 (p, ∆t) = e h̄ 2m
2πh̄ i∆t −im
1 i p
2
= √ e− h̄ 2m ∆t , (2.6)
2πh̄
so that we obtain
1 Z i p∆x
K0 (x, t; xi , ti ) = √ e h̄ K0 (p, ∆t) dp
2πh̄  
p2
1 Z h̄i p∆x− 2m ∆t
= e dp Θ(∆t) , (2.7)
2πh̄
where in the last line the causality condition Θ(∆t) has been written explic-
itly; it takes care of the boundary condition K = 0 for ∆t < 0. Eq. (2.7) is
just the Fourier representation of the propagator (2.3).
The step function can be rewritten using the relation
+∞
1 Z eiω∆t
Θ(∆t) = dω (ε > 0) , (2.8)
2πi ω − iε
−∞
CHAPTER 2. PERTURBATION THEORY 22

which follows directly from the residue theorem: for ∆t > 0 the integral can
be closed in the upper half of the complex ω plane; Cauchy’s integral theorem
then gives 2πi in the limit ε → 0 so that Θ(t) = 1. If ∆t < 0, on the other
hand, then the loop integration can only be closed in the lower half-space
thus missing the pole at ω = +iε). Multiplying (2.6) with (2.8) gives
h   i
i p2
+∞
Z h̄
p∆x− 2m
−h̄ω ∆t
1 e
K0 (x, t; xi , ti ) = dp dω . (2.9)
(2π)2 h̄i ω − iε
−∞

We now substitute
p2
E= − h̄ω (2.10)
2m
and obtain
1 Z i
(p∆x−E∆t) ih̄
K0 (x, t; xi , ti ) = dp dE e h̄
p2
(2.11)
(2πh̄)2 E − 2m + iε

(the “−” sign coming from the substitution is cancelled by another sign
obtained by inverting the integration boundaries).
Since we will need these expressions later on in three space dimensions
we give them here in a straightforward generalization of (2.7) and (2.11)
Z i 0 p2

1 p
~ x
·(~ x
−~ )− (t 0 −t)
K0 (x0 , t0 ; x, t) = e h̄ 2m
d3p Θ(t0 − t) , (2.12)
(2πh̄)3

and
1 Z 3 i
(~
p·∆~
x−E∆t) ih̄
K0 (x, t; xi , ti ) = d p dE e h̄
p2
. (2.13)
(2πh̄)4 E − 2m + iε

The integrand on the right-hand side of (2.11) is the free propagator


in the energy-momentum representation. Since p and E are independent
variables in the integral, we see that propagation also takes place at energies
E 6= p2 /2m. The classical dispersion relation does show up as a pole in the
propagator.

2.2 Perturbative Expansion


We now assume that the unperturbed particle moves freely and that the
perturbing interaction is given by V (x, t). We furthermore assume that H
CHAPTER 2. PERTURBATION THEORY 23

has the special form H = p2 /2m + V (x, t), so that the propagator is given
by Z
i
K (x, t; xi , ti ) = N Dx e h̄ S (2.14)
with
Zt Zt  
0 m 2
S= L(x, ẋ) dt = ẋ − V (x, t) dt . (2.15)
2
ti ti

Since the integrand here is a classical function, we have

i
Rt Rt

m 2
2
ẋ dt − h̄i V (x,t) dt
i
S
e h̄ =e ti
e ti
. (2.16)

The second factor can now be expanded in powers of the potential, thus
yielding a perturbative expansion of the action
Rt  2
i
V (x,t) dt t t

∼ i Z 1 1 Z
e ti
=1− V (x, t) dt − V (x, t) dt + . . . . (2.17)
h̄ 2! h̄2
t i t i

When we substitute this expansion into the expression (2.14) we obtain


Z (
i
K=N Dx e h̄ S0
  2 
Zt Zt )
 i 1 1  
× 1 − V (x, t) dt − V (x, t) dt + . . .
h̄ 2! h̄2
ti ti

= K0 + K1 + K2 + . . . , (2.18)

i.e. a sum ordered in powers of the interaction.

First order propagator. We next determine the first-order propagator


K1 . According to (2.18) it is given by
tf
i Z i
Z
K1 = − N Dx e h̄ S0 V (x, t) dt (2.19)

ti
! n+1
i m 2
= − lim
h̄ n→∞ 2πh̄iη
+∞
 
Z Yn n n
X m X
× dxi V (xk , tk ) η exp i (xj+1 − xj )2  .
2h̄η
−∞ i=1 k=1 j=0
CHAPTER 2. PERTURBATION THEORY 24

Here the time-integral has been written as a sum. In a next step we now
split the sum in the exponent into two pieces, one running from j = 0 to
j = k − 1 and the other from j = k to j = n and separate the corresponding
integrals. This gives
n Z "
i X
K1 lim
= − n→∞ η dxk V (xk , tk ) (2.20)
h̄ k=1
 
k−1
P
m 2
Z i 2h̄η (xj+1 −xj )
 k 
N
× 2 dx1 dx2 · · · dxk−1 e j=0 

 n
P 
#
Z m
i 2h̄η (xj+1 −xj )2
n−k+1
×
N 2 dxk+1 · · · dxn e j=k 

with
m
N≡ . (2.21)
2πh̄iη
The term in the first bracket is nothing else than the propagator from ti to
tk (K0 (xk , tk ; xi , ti )) , and that in the second bracket is that from tk to t
(K0 (x, t; xk , tk )). Thus we have

K1 (xf , tf ; xi , ti ) =
+∞ Ztf
i Z
− dx dt K0 (xf tf ; x, t) V (x, t)K0 (x, t; xi , ti ) . (2.22)

−∞ ti

The time integral over the interval from ti to tf can be extended to ∞ by


noting that

K0 (x, t; xi , ti ) = 0 for t < ti (2.23)


K0 (xf , tf ; x, t) = 0 for tf < t .

This gives

K1 (xf , tf ; xi , ti ) =
+∞ +∞
i Z Z
− dt dx K0 (xf , tf ; x, t) V (x, t)K0 (x, t; xi , ti ) . (2.24)

−∞ −∞
CHAPTER 2. PERTURBATION THEORY 25

xf t f

x2 t 2
= + + + ...
x1 t 1 x1 t 1

xi t i

Figure 2.1: Born series for the propagator

Higher order propagators. Similarly, the higher order terms in the per-
turbation expansion (2.18) can be obtained. This yields finally

K (xf , tf ; xi , ti ) =
i Z
K0 (xf , tf ) − K0 (xf , tf ; x1 , t1 ) V (x1 , t1 ) K0 (x1 , t1 ; xi , ti ) dx1 dt1
Z h

1
− 2 K0 (xf , tf ; x1 , t1 ) V (x1 , t1 ) K0 (x1 , t1 ; x2 , t2 )
h̄ i
× V (x2 , t2 ) K0 (x2 , t2 ; xi , ti ) dx1 dt1 dx2 dt2
+ ··· . (2.25)

Note that the time-integrations in (2.25) are effectively time-ordered because


of the implicit Θ (tf − ti ) function in each of the propagators, leading to
tf > t1 > t2 > · · · > ti . This fact explains why, for example, the factor 1/(2!)
in the second order term of (2.18) no longer appears in (2.25)
Z 2
1 1 Z
dtV (x, t) = V (x, t)V (x, t0 ) dt dt0
2! 2!
1 Z
= [V (x, t)V (x, t0 )Θ (t0 − t) + V (x, t)V (x, t0 )Θ(t − t0 )] dt dt0
Z2!
= V (x, t0 )Θ (t0 − t) V (x, t) dt dt0 . (2.26)

The last line follows from a simple change of variables; the Θ function in it
is absorbed into the propagator in (2.25). A similar argument holds for all
CHAPTER 2. PERTURBATION THEORY 26

the higher-order terms in the interaction. In each case the prefactor (1/n!)
stemming from the expansion of the exponential in (2.17) is cancelled by the
time ordering inherent in the propagators.
Equation (2.25) is the Born series for the propagator; it can be represented
graphically as shown in Fig. 2.1. A time axis runs here from bottom to
top. The straight lines denote the free propagation of the particles (K0 )
and the dots stand for the interaction vertices (−iV /h̄) (the circles mark the
interaction range) and a space and time integration appears at each vertex.

Bethe-Salpeter equation. The expansion (2.25) can be formally summed.


This can be seen as follows (in obvious shortterm notation)

K = K 0 + K0 U K0 + K0 U K0 U K0 + · · · (2.27)
= K0 + K0 U (K0 + K0 U K0 + · · ·) with U = − h̄i V .

The expression in parentheses is just again K, so that we obtain the Bethe-


Salpeter equation
K = K 0 + K0 U K . (2.28)
The Bethe-Salpeter equation is an integral equation for the full interacting
propagator K as can be seen most easily from its space-time representation

K (xf , tf ; xi , ti ) = K0 (xf , tf ; xi , ti ) (2.29)


i Z
− K0 (xf , tf ; x, t) V (x, t)K (x, t; xi , ti ) dx dt .

We can also represent the Bethe-Salpeter equation in a diagrammatic
way. If we denote the so-called “dressed propagator” K, that includes all the
effects of the interactions, by a double line

= K (x2 , t2 ; x1 , t1 ) , (2.30)

then this equation can be graphically represented as shown in Fig. 2.2. Each
graph in Fig. 2.2 finds a one-to-one correspondence in (2.28): the single lines
represent the free propagator, the double line the dressed propagator and
the dot stands for the interaction U . (2.28) shows that the factors in the
second term of this equation have to be written from left to right against the
time-arrow in Fig. 2.2.
The Bethe-Salpeter equation can also be written in an equivalent form
for the interacting wavefunction
Z
Ψ (xf , tf ) = K (xf , tf ; xi , ti ) Ψ (xi , ti ) dxi
CHAPTER 2. PERTURBATION THEORY 27

= +

Figure 2.2: Bethe-Salpeter equation (2.28). The arrows indicate the time
direction.
Z
= K0 (xf , tf ; xi , ti ) Ψ (xi , ti ) dxi
i Z
− K0 (xf , tf ; x, t) V (x, t)K (x, t; xi , ti ) Ψ (xi , ti ) dxi dx dt
Z h̄
= K0 (xf , tf ; xi , ti ) Ψ (xi , ti ) dxi (2.31)
i Z
− K0 (xf , tf ; x, t) V (x, t)Ψ(x, t) dx dt .

This constitutes an integral equation for the unknown wavefunction Ψ(x, t).

2.3 Application to Scattering


Let us now apply the results of the last section to a scattering process. In this
case the particle is free at t = −∞, then undergoes the scattering interaction
and then, at t = +∞, is free again.
We treat this problem as usual by adiabatically switching on and off the
interaction V (x, t). The initial condition (for t → −∞) for the wavefunction
then is
~
Ψin (x, t) = N ei(ki ·~x−ωi t) . (2.32)
We choose here
√ a box normalization with periodic boundary conditions so
that N = 1/ V . The scattering state that evolves from this incoming state
is denoted by Ψ(+) (x, t). The superscript (+) indicates that the state evolves
forward in time, starting from Ψin at t = −∞; it thus fulfills the boundary
condition
Ψ(+) (x, t → −∞) = Ψin (x, t) . (2.33)
In a scattering experiment one looks at t → +∞ for a free scattered particle
CHAPTER 2. PERTURBATION THEORY 28

with definite momentum; the corresponding final state is denoted by


~
Ψout (x, t) = N ei(kf ·~x−ωf t) . (2.34)

The probability amplitude for the presence of Ψout in the scattered state Ψ(+)
is given by
Z
Sf i = Ψ∗out (xf , tf ) Ψ(+) (xf , tf ) d3xf for tf → ∞ . (2.35)

This is just the transition amplitude from the initial state i to the final state
f (S-matrix).
Expansion (2.29) yields the Bethe-Salpeter equation for the scattering
wavefunction
Z
Ψ(+) (xf , tf ) = K0 (xf , tf ; xi , ti ) Ψin (xi , ti ) d3xi (2.36)
i Z
− K0 (xf , tf ; x, t) V (x, t)K(x, t; xi , ti )Ψin (xi , ti ) d3xi d3x dt .

Inserting this into (2.35) gives for the S-matrix
Z
Sf i = Ψ∗out (xf , tf ) K0 (xf , tf ; xi , ti ) Ψin (xi , ti ) d3xi d3xf (2.37)
Z
i
− Ψ∗out (xf , tf ) K0 (xf , tf ; x, t) V (x, t)

× K(x, t; xi , ti )Ψin (xi , ti ) d3xi d3x dt d3xf .

Since Ψin is a plane wave, we know that


Z
φ (xf , tf ) = K0 (xf , tf ; xi , ti ) Ψin (xi , ti ) d3xi (2.38)

is also a plane wave state, since K0 is the free propagator so that no interac-
tion takes place. φ is actually the same wavefunction as Ψin , only taken at a
later time and space point
~
φ (xf , tf ) = N ei(ki ·~xf −ωi tf ) . (2.39)

This means that the first integral in (2.37) can be easily evaluated
Z  
Ψ∗out (xf , tf ) φ (xf , tf ) d3xf = δ 3 ~kf − ~ki . (2.40)

We thus have
~
3

~
 i Z 3
Sf i = δ k f − k i − d xf d3x d3xi dt (2.41)

×[Ψ∗out (xf , tf ) K0 (xf , tf ; x, t) V (x, t)K(x, t; xi , ti )Ψin (xi , ti ) ] .
CHAPTER 2. PERTURBATION THEORY 29

xf t f

xt

xi t i

Figure 2.3: First order scattering diagram, corresponding to the second term
in (2.41). Time runs from left to right.

The amplitude for a scattering process is given by the second term.


If K on the rhs of (2.41) is now represented by the Born series expan-
sion of the Bethe-Salpeter equation (2.27) this amplitude can be graphically
represented as shown in Fig. 2.3. This diagram can be translated into the
amplitude just given by writing for each straight-line piece

= K0 (x2 , t2 ; x1 , t1 ) (2.42)
xt t 1 x2 t 2

and for each interaction vertex


i
= − V (x, t) + integration over x, t . (2.43)
xt h̄

The rules are completed by multiplying Ψin and Ψ∗out at the corresponding
sides of the diagram and then integrating over the spatial variables of these
wavefunctions and over all intermediate times. The ordering of factors is
such that in writing the various factors from left to right one goes against
the flow of time in the figure.
These rules are illustrated for a second-order scattering process in Fig. 2.4.
In this figure the time runs from ti to tf from left to right. The corresponding
amplitude is given by
Z 
(2) 3 3 3 3
A = d xi d xf d x1 d x2 dt1 dt2 Ψ∗out (xf , tf ) K0 (xf , tf ; x2 , t2 )
    
i i
− V (x2 , t2 ) K0 (x2 , t2 ; x1 , t1 ) − V (x1 , t1 ) K0 (x1 , t1 ; xi , ti ) Ψin (xi , ti ) .
h̄ h̄
(2.44)
CHAPTER 2. PERTURBATION THEORY 30

xf t f

x1 t 1

x2 t 2

xi t i

Figure 2.4: Second order scattering diagram, corresponding to (2.44).

We now evaluate the first order scattering amplitude (cf. (2.41))

(1) i Z h ∗
A = − Ψout (xf , tf ) K0 (xf , tf ; x, t) V (x, t)K0 (x, t; xi , ti )
h̄ i
× Ψin (xi , ti ) d3x dt d3xi d3xf (2.45)

somewhat further by using expression (2.6) for the free propagator (extended
to three dimensions)
 
p2
1 Z h̄i p~·(~x0 −~x)− 2m (t0 −t)
0 0
K0 (x , t ; x, t) = 3
e d3p Θ(t0 − t) . (2.46)
(2πh̄)

Inserting this into the expression for A(1) gives


i 1 1 Z 3
A(1) = − d x dt d3xi d3xf d3p d3q (2.47)
h̄ V (2πh̄)6
  2
  2
 
p q
i
~·(~
p x)− 2m
xf −~ (tf −t) i
q~·(~
x−~
xi )− 2m (t−ti )
e − h̄i (p~f ·~xf −Ef tf ) e h̄
V (~x, t)e h̄
e
i
(~
p ·~
h̄ i i
x −Ei ti )
,

where the time-ordering tf > t > ti is understood and we have abbreviated


Ei = p2i /(2m) and Ef = p2f /(2m).
We integrate first over xi and xf . This yields
Z
d3xi −→ (2πh̄)3 δ 3 (~pi − ~q) (2.48)
Z
d3xf −→ (2πh̄)3 δ 3 (~pf − p~) .

Performing next the integrals over p and q gives for the amplitude

(1) i Z 3
A =− d x dt
h̄V
CHAPTER 2. PERTURBATION THEORY 31

   
" p2 p2 #
f
+ h̄i Ef tf
i

pf ·~
−~ x− 2m (tf −t) i

p
~i ·~ i (t−t )
x− 2m i
− h̄i Ei ti
× e e V (x, t)e e
i Z h̄i (−~pf ·~x+Ef t) i
= − e V (x, t)e h̄ (+~pi ·x−Ei t) d3x dt . (2.49)
h̄V
The time-integration is performed next. For that we assume that V (x, t)
acts only during a finite, but long time-interval ([−T, T ]), in which it is time-
independent; at the boundaries of the interval it is adiabatically, i.e. without
any significant energy-transfer, turned on and off. We then obtain
Z ZT
dt V (x, t)e (Ef −Ei )t dt =
i T →∞
h̄ V (x)ei(ωf −ωi )t dt −→ V (x) 2πδ (ωf − ωi ) ,
−T
(2.50)
with ωf − ωi = (Ef − Ei )/h̄. Thus A(1) becomes
Z
i
2πδ (ωf − ωi ) e h̄ (p~i −~pf )·~x V (~x) d3x .
i
A(1) = − (2.51)
h̄V
This is the well-known lowest-order Born-approximation result.
Chapter 3

GENERATING
FUNCTIONALS

In this section we consider the transition amplitude in the presence of an


external “source” J(t), so that the Hamiltonian reads

HJ (x, p) = H(x, p) − h̄J(t)x . (3.1)

A classical example is that of a harmonic oscillator with externally driven


equilibrium position x0 (t). Its Hamiltonian is

p2 1
H x0 = + k(x − x0 (t))2
2m 2
= H − kx0 (t)x + O(x20 ) . (3.2)

With h̄J(t) = kx0 (t) and for small amplitudes x0 we just have the form of
(3.1). It is evident that the states of this system will change as time develops,
because of its changing equilibrium position.
Suppose now that the system was in its groundstate at t → −∞ and
that x0 (t) acts only for a limited time period. We could then calculate the
transition probability for the system to be still in its groundstate at t → +∞
by using the techniques developed in sections 2.2 and 2.3. There we saw
(cf. (2.25) and (2.44)) that this probability is determined by matrixelements
of time-ordered products of the interaction, i.e. of the operator x̂ in the
present case. In this chapter we will show that these matrixelements can all
be generated once only the functional dependence of the probability for the
system to remain in its groundstate on an external source is known.

32
CHAPTER 3. GENERATING FUNCTIONALS 33

3.1 Groundstate-to-Groundstate Transitions


Generalizing the special example of the introduction we assume that an ar-
bitrary physical system is at first (at ti ) stationary and then changes under
the influence of an external source J(t)x of finite duration. After the source
has been turned off, the groundstate of the system is still the same (up to a
phase), but the system may be excited.
The propagator for this system, is quite generally, given by

i
Rtf
Z Z h̄
[pẋ−H(x,p)+h̄J(t)x] dt
hxf , tf |xi , ti iJ = Dx Dp e ti
. (3.3)

The value of this propagator depends obviously on the source function J(t);
it is a functional of the source J(t).
In the following paragraphs we will now discuss this functional depen-
dence for tf → +∞ and ti → −∞. We will also show that it determines the
vacuum expectation values of time-ordered x̂ operators.
We start by calculating the propagator (3.3) of a system under the influ-
ence of a source, i.e. for a system described by (3.1). We first assume that
the source is nonzero only for a limited time between −T and +T

J(t) = 0 for |t| > T . (3.4)

We can then write for the propagator (ti < −T , tf > T )


Z
hxf tf |xi ti iJ = dx0 dx hxf tf |x0 T ihx0 T |x −T iJ hx −T |xi ti i . (3.5)

Note that the two outer propagators are taken with the sourceless Hamil-
tonian (J = 0) because of condition (3.4). They are given by, e.g.,
i
hx −T |xi ti i = hx|e− h̄ Ĥ(−T −ti ) |xi i
X i
= ϕn (x)ϕ∗n (xi ) e− h̄ En (−T −ti ) . (3.6)
n

Similarly we get
i
hxf tf |x0 T i = hxf |e− h̄ Ĥ(tf −T ) |x0 i
X i
= ϕn (xf )ϕ∗n (x0 )e− h̄ En (tf −T ) . (3.7)
n

Here the ϕn are eigenstates of the Hamiltonian Ĥ without a source; we


assume that the spectrum of Ĥ is bounded from below with eigenvalues
En ≥ E0 . The dependence on ti and tf is now isolated.
CHAPTER 3. GENERATING FUNCTIONALS 34

Taking now the limits ti → −∞ and tf → +∞ is not straightforward


because both times appear as arguments of oscillatory functions. These func-
tions oscillate the more rapidly the higher the eigenvalues En are. We can
thus expect that the dominant contribution will come from the lowest eigen-
value E0 .

Wick Rotation. This can indeed be shown by by a mathematical trick,


the so-called Wick rotation. In this method one looks at the propagators
(3.6) and (3.7) as functions of ti and tf , respectively, and continues these
variables analytically from the real axis into the complex plane. One then
performs the limits and after that rotates back to real times. Mathematically
this is achieved by replacing the physical Minkowski time t by a complex time
τ
t −→ τ = te−iε . (3.8)
The direction of the rotation is mandated by the requirement that there
are no singularities of the integrand encountered in the rotation of the time-
axis. This is the case for
0≤ε<π . (3.9)
For larger angles the rhs of (3.6) and (3.7) develops an essential singularity
for ti/f → −/ + ∞.
This rotation corresponds to an analytic continuation in the variable ε;
the real, four-dimensional Minkowski space can be viewed as a subspace of
a four-dimensional complex space (or of a five-dimensional space with 3 real
space-coordinates and 1 complex time-coordinate). The substitution (3.8)
corresponds to a clockwise rotation of the time-axis out of the real subspace
into the larger complex space. For ε = +π/2 this corresponds to a rotation
t = +∞ → τ = −i∞, t = −∞ → τ = +i∞; only this special case is
often called a Wick rotation. Physical results are obtained by evaluating
the relevant expressions in the enlarged, complex space and then, at the
end, rotating the result back to the real time-axis (which works only if no
singularities are encountered on this way).
We now apply this rotation to the expression
i X i i
e− h̄ E0 ti hx −T |xi ti i = ϕn (x)ϕ∗n (xi ) e h̄ En T e h̄ (En −E0 )ti (3.10)
n

and thus perform the limit ti → −∞ on the rotated time axis


i
lim e− h̄ E0 τi hx −T |xi τi i (3.11)
τi →−∞(cos ε−i sin ε)
X i i
= lim ϕn (x)ϕ∗n (xi ) e h̄ En T e+ h̄ (En −E0 )τi .
τi →−∞(cos ε−i sin ε) n
CHAPTER 3. GENERATING FUNCTIONALS 35

Because we have separated out the lowest frequency the last factor vanishes
for ε > 0 , except for n = 0. Thus we get
i i
lim e− h̄ E0 τi hx −T |xi τi i = ϕ0 (x)ϕ∗0 (xi ) e h̄ E0 T . (3.12)
τi →−∞(cos ε−i sin ε)

Analogously we also obtain


i i
lim e+ h̄ E0 τf hxf τf |x0 T i = ϕ0 (xf )ϕ∗0 (x0 ) e h̄ E0 T . (3.13)
τf →+∞(cos ε−i sin ε)

Both expressions, (3.12) and (3.13), are the analytical continuations of the
corresponding limits on the real time axis from ε = 0 to ε > 0. Since the
right hand sides of these equations do not depend on ε they can be continued
back to the real time axis (ε = 0) without any change. By means of the
Wick rotation we have thus been able to make the expression (3.12) and
(3.13) convergent; in this process we have found that only the groundstate
contributes to the transition probability.
We now insert these expressions into (3.5) and get
i
lim e h̄ E0 (τf −τi ) hxf τf |xi τi iJ (3.14)
τi →−∞(cos ε−i sin ε)
τf →∞(cos ε−i sin ε)
Z
i
= ϕ0 (xf )ϕ∗0 (xi ) e h̄ E0 2T dx dx0 ϕ∗0 (x0 ) hx0 T |x −T iJ ϕ0 (x)

The integral in the last line can be rewritten, using ϕ0 (x) = hx|0i. This gives
Z Z
0
dx dx ϕ∗0 (x0 ) hx0 T |x −T iJ ϕ0 (x) = dx dx0 h0|x0 ihx0 T |x −T iJ hx|0i
= h0T |0 −T iJ (3.15)

so that we obtain for (3.14)


i
lim e h̄ E0 (τf −τi ) hxf τf |xi τi iJ
τi →−∞(cos ε−i sin ε)
τf →∞(cos ε−i sin ε)

i
= ϕ0 (xf )ϕ∗0 (xi ) e h̄ E0 2T h0T |0 −T iJ . (3.16)

We obtain by Wick-rotating back


i i
e h̄ E0 (tf −ti ) e− h̄ E0 2T
h0T |0 −T iJ = t lim hxf tf |xi ti iJ . (3.17)
i →−∞
t →+∞
ϕ∗0 (xi )ϕ0 (xf )
f

With (in the limit ti → −∞, tf → +∞)


i
hxf tf |xi ti iJ=0 = ϕ∗0 (xi )ϕ0 (xf )e− h̄ E0 (tf −ti ) (3.18)
CHAPTER 3. GENERATING FUNCTIONALS 36

(3.17) becomes
hxf tf |xi ti iJ − i E0 2T
h0T |0 −T iJ = t lim e h̄ . (3.19)
→−∞
i
tf →+∞
hxf tf |xi ti i0

This immediately gives for the free (J = 0) transition amplitude


i
h0T |0 −T i0 = e− h̄ E0 2T , (3.20)

as it should.
Equation (3.17) implies that the groundstate-to-groundstate transition
amplitude is – up to a factor – given by a path integral from arbitrary xi
to arbitrary xf and thus does not depend on these quantities, if only the
corresponding times are taken to infinity.

Gs-to-gs transition rate. The vacuum transition rates h0T |0 −T iJ de-


serve some explanation. Formally, they are given by
i
h0T |0 −T iJ = h0|e− h̄ (H−h̄Jx)2T |0i . (3.21)

The vacuum state |0i is assumed to be unique, if there is no source J present,


and normalized to 1. It is the vacuum of the theory before and after the
i
action of the source. On the other hand, e− h̄ (H−h̄Jx)2T |0i is the state that
the vacuum at t = −T has evolved into at t = +T under the influence of the
external source J(t). If this external perturbation has acted adiabatically
(very gently turned on and off again) the vacuum at t > T differs from that
of the source-free theory only by a phase. The matrixelement (3.21) is then
the probability amplitude to find the original vacuum in the time-evolved
vacuum state. The absolute value of this probability amplitude is surely 1,
so that the two states can differ only by a J-dependent real phase
i
h0T |0 −T iJ = e h̄ (S[J]−E0 2T ) (3.22)

where we have taken out the free propagation contribution. This phase is
the quantity determined by (3.17).
To conclude these considerations we note that instead of using the Wick-
rotation to make the transition rates well behaved for very large times we
could also have added a small negative imaginary term −iεEn to all eigen-
values En . This would have given a damping factor to the oscillating expo-
nentials that becomes larger with n and thus would have led to a suppression
of all higher excitations. This becomes apparent by looking at expressions
(3.6) and (3.7). At the end of the calculation the limit ε → 0 would have to
be performed.
CHAPTER 3. GENERATING FUNCTIONALS 37

3.1.1 Generating functional.


This groundstate-to-groundstate transition rate is a functional of the source
J(t) which we denote by

hxf tf |xi ti iJ − i E0 2T
W [J] = h0 +∞|0 −∞iJ = t lim e h̄ . (3.23)
→−∞ i
tf →+∞
hxf tf |xi ti iJ=0
T →∞

W [J] is called a generating functional for reasons that will become clear in
the next section.
In order to get rid of the phase that is produced already by a source-free
propagation (exp − h̄i E0 2T ) we define now a normalized generating functional

W [J] h0 +∞|0 −∞iJ hxf tf |xi ti iJ


Z[J] = = = t lim (3.24)
W [0] h0 +∞|0 −∞iJ=0 i →−∞ hxf tf |xi ti iJ=0
t →+∞ f

with Z[0] = 1. The functional Z[J] describes the processes relative to the un-
perturbed (J = 0) time-development. The numerator in (3.24) is a transition
amplitude and can therefore be written as a path integral
+∞
R
i
Z Z h̄
[pẋ−H(p,x)+h̄J(t)x] dt
hxf + ∞|xi − ∞iJ = Dx Dp e −∞
. (3.25)

If H is quadratic in p and of the form H = p2 /(2m)+V (x), the propagator


can be rewritten as (see Sect. 1.3)
+∞ R
i
Z h̄
[L(x,ẋ)+h̄J(t)x] dt
hxf + ∞|xi − ∞iJ = N Dx e −∞
. (3.26)

In the normalized functional Z[J] the (infinite) factor N cancels out because
it is independent of J

W [J] h0 +∞|0 −∞iJ


Z[J] = =
W [0] h0 +∞|0 −∞iJ=0
+∞
R
i

[L(x,ẋ)+h̄J(t)x] dt
R
Dx e −∞
= +∞
R
(3.27)
i

[L(x,ẋ)] dt
R
Dx e −∞
CHAPTER 3. GENERATING FUNCTIONALS 38

3.2 Functional Derivatives


of Transition Amplitudes
In this section we will show – by using the methods outlined in App. B – that
the groundstate expectation value of a time-ordered product of interaction
operators – in this present case of the operators x̂(t) – can be obtained as
functional derivatives of the functional W [J] with respect to J.
We start with the definition of a path integral as a limit of a finite di-
mensional integral (see (1.33))
n
Z Y n
Z Y
dpl
hxf tf |xi ti iJ = lim dxk (3.28)
n→∞
k=1 l=0 2πh̄
 
n
i X
× exp  [pj (xj+1 − xj ) − ηH (pj , xj ) + h̄Jj xj ] .
h̄ j=0

and calculate its functional derivative with respect to J. In order to become


familiar with functional derivatives we do this in quite some detail. Using the
definition (B.26) for the functional derivative we get (with the abbreviation
F (xj , pj ) = pj (xj+1 − xj ) − ηH(pj , x̄j ))

δhxf tf |xi ti iJ
δJ(t1 )
( Z Yn Z Yn
1 dpl
= lim lim dxk (3.29)
ε→0 ε n→∞
k=1 l=0 2πh̄
  
n
i X
× exp  {F (xj , pj ) + h̄xj [Jj + εδ(tj − t1 )]}
h̄ j=0
 
n
i X 
− exp  (F (xj , pj ) + h̄xj Jj )
h̄ j=0 
 
n
Z Y l
Z Y n
dpl i X
= lim dxk ix1 exp  (F (xj , pj ) + h̄xj Jj ) ,
n→∞
k=1 l=0 2πh̄ h̄ j=0

which can be written as

i
Rtf
Z Z [pẋ−H(x,p)+h̄J(t)x] dt
δhxf tf |xi ti iJ h̄
= i Dx Dp x(t1 )e ti
. (3.30)
δJ(t1 )
We now want to relate this derivative of a classical functional to quan-
tum mechanical expressions and thus understand its physical significance and
CHAPTER 3. GENERATING FUNCTIONALS 39

meaning. In order to do so, we go back to the definition of the propagator.


There we had (in (1.23))

hxf , tf |xi ti i (3.31)


Z
= dx1 . . . dxn hxf tf |xn , tn ihxn tn |xn−1 tn−1 i . . . hx1 t1 |xi ti i .

Now, in (3.30) for J = 0, we have one factor more on the righthand side of
this equation
Z
dx1 . . . dxn hxf tf |xn tn i . . . hx1 t1 |xi ti ix1
Z
= dx1 . . . dxn hxf tf |xn tn i . . . hx1 t1 |x̂(t1 )|xi ti i ; (3.32)

the last step is possible because |x1 i is an eigenstate of the x̂ operator with
eigenvalue x1 (cf. (1.18)). The last integral is obviously equal to

hxf tf |x̂(t1 )|xi ti i , (3.33)

i.e. to a matrixelement of the position operator. We thus have

i
Rtf
Z Z [pẋ−H(x,p)] dt
δhxf tf |xi ti iJ h̄
= i Dx Dp x(t1 )e ti

δJ(t1 ) J=0

= ihxf tf |x̂(t1 )|xi ti i . (3.34)

For the case that H is separable in x and p and quadratic in p, this relation
reads

i
Rtf
Z L(x,ẋ) dt
δhxf tf |xi ti iJ h̄
= iN Dx x(t1 )e ti

δJ(t1 ) J=0

= ihxf tf |x̂(t1 )|xi ti i . (3.35)

The functional derivative of the propagator with respect to the source thus
gives the transition matrix element of the coordinate x.
The higher order functional derivatives yield
δ n hxf tf |xi ti iJ
(3.36)
δJ(t1 )δJ(t2 ) . . . δJ(tn )
i
Rtf
Z Z h̄
[pẋ−H(x,p)+h̄J(t)x] dt
n
= (i) Dx Dp x(t1 )x(t2 ) . . . x(tn ) e ti
.
CHAPTER 3. GENERATING FUNCTIONALS 40

One might guess that



δ n hxf tf |xi ti iJ

= in hxf tf |x̂(t1 )x̂(t2 ) . . . x̂(tn )|xi ti i , (3.37)
δJ(t1 )δJ(t2 ) . . . δJ(tn ) J=0
but this equation is not quite correct. We see this by considering explicitly
the second derivative. We can proceed there exactly in the same way as for
the first. We have for the rhs of (3.36) in the case of the second derivative

i
Rtf
2 Z Z [pẋ−H(x,p)] dt
δ hxf tf |xi ti iJ h̄
= i2 Dx Dp x(tα )x(tβ )e ti

δJ(tα )δJ(tβ ) J=0


Z
= i2 dxi . . . dxn hxf tf |xn tn i · · · hxl tl |x̂(tα )|xl−1 tl−1 i
· · · hxk tk |x̂(tβ )|xk−1 tk−1 i · · · hx1 t1 |xi ti i . (3.38)

Here we have assumed that tα > tβ , since each of the infinitesimal Green’s
functions propagates only forward in time. In this case (3.38) is indeed equal
to
i2 hxf tf |x̂(tα )x̂(tβ )|xi ti i . (3.39)
However, if tα < tβ , then these two times appear in a different ordering in the
rhs of (3.38) and thus of the matrix element. The two cases can be combined
by introducing the time-ordering operator T
(
x̂(t1 )x̂(t2 ) t1 > t2
T [x̂(t1 )x̂(t2 )] = (3.40)
x̂(t2 )x̂(t1 ) t2 > t1 .
With the time-ordering operator we have

i
Rtf
[pẋ−H(x,p)] dt
δ 2 hxf tf |xi ti iJ 2
Z Z h̄
= i Dx Dp x(t1 )x(t2 )e ti

δJ(t1 )δJ(t2 ) J=0


= i2 hxf tf |T [x̂(t1 )x̂(t2 )] |xi ti i . (3.41)

The same reasoning leads to the following result for higher-order deriva-
tives
 n
1 δ n hxf tf |xi ti iJ


i δJ(t1 )J(t2 ) . . . δJ(tn ) J=0

i
Rtf
Z Z h̄
[pẋ−H(x,p)] dt
= Dx Dp x(t1 )x(t2 ) . . . x(tn ) e ti

= hxf tf |T [x̂(t1 )x̂(t2 ) . . . x̂(tn )] |xi ti i . (3.42)


CHAPTER 3. GENERATING FUNCTIONALS 41

This is the generalization of (3.41). We thus have


 n
1 δn hxf tf |xi ti iJ

i δJ(t1 )δJ(t2 ) . . . δJ(tn ) hxf tf |xi ti iJ=0 J=0
 n
1 δn
= Z[J]
i δJ(t1 )δJ(t2 ) . . . δJ(tn ) J=0

i
Rtf

[pẋ−H(x,p)] dt
R R
Dx Dp x(t1 )x(t2 ) . . . x(tn ) e ti

=
i
Rtf

[pẋ−H(x,p)] dt
R R
Dx Dp e ti

hxf tf |T [x̂(t1 )x̂(t2 ) . . . x̂(tn )] |xi ti i


= , (3.43)
hxf tf |xi ti i
where the limit ti → −∞, tf → +∞ is understood and all the times t1 , . . . , tn
lie in between these limits. If the Hamiltonian is quadratic in p and separates
in p and x, then we have
R i
hxf tf |T [x̂(t1 )x̂(t2 ) . . . x̂(tn )] |xi ti i Dx x(t1 )x(t2 ) . . . x(tn ) e h̄ S[x(t)]
= R i ,
hxf tf |xi ti i Dx e h̄ S[x(t)]
(3.44)
where S[x(t)] is the action that depends functionally on the trajectory x(t).
We now rewrite this equation. The numerator of the lhs becomes (in the
limit ti → −∞, tf → +∞, indicated by the arrow)
hxf tf |T [. . .]|xi ti iJ=0 −→ hxf tf |0ih0|T [. . .]|0ih0|xi ti iJ=0
i i
= h0|T [. . .]|0i ϕ0 (xf )e− h̄ E0 tf ϕ∗0 (xi )e+ h̄ E0 ti , (3.45)
while the denominator can be written as (c.f. (3.18))
i i i i
hxf tf |xi ti i = hxf |e− h̄ Ĥtf e+ h̄ Ĥti |xi i −→ ϕ0 (xf )e− h̄ E0 tf ϕ∗0 (xi )e+ h̄ E0 ti (3.46)
where we have used in both cases (1.19), inserted a complete set of states and
– through the limit of infinite times – projected out the groundstate. The
gs wavefunctions and the time-dependent exponentials cancel out so that we
obtain (for quadratic Hamiltonians)
R i
Dx x(t1 )x(t2 ) . . . x(tn ) e h̄ S[x(t)]
h0|T [x̂(t1 )x̂(t2 ) . . . x̂(tn )] |0i = R i
Dx e h̄ S[x(t)]
 n
1 δn
= Z[J]
i δJ(t1 )δJ(t2 ) . . . δJ(tn ) J=0
(3.47)
CHAPTER 3. GENERATING FUNCTIONALS 42

R +∞
with S[x(t)] = −∞ L(x(t), ẋ(t)) dt.
This is a very important result. It shows that the groundstate expectation
value of a time-ordered product of position operators, the so-called correlation
function, can be obtained as a functional derivative of the functional Z[J]
defined in (3.24). Note that |0i is the groundstate of Ĥ, as can be seen
from (3.45). Thus the groundstate appearing on the lhs of (3.47) and the
propagator Z[J] are linked together: if Z[J] contains, for example, only a
free Hamiltonian, then |0i is the groundstate of a free theory. If, on the
other hand, Z[J] contains interactions, then |0i is the groundstate of the full
interacting theory.
Since all the correlation functions can be generated from Z[J] this func-
tional is called a generating functional. For the remainder of this book we
will be concerned with these generating functionals.
Coming back to our example of the driven harmonic oscillator, discussed
at the start of this section, we see that the time-ordered vacuum expecta-
tion values of x̂ are just the matrix elements that would appear in a time-
dependent perturbation theory treatment of the groundstate of this system.
Thus, if all the correlation functions are known, then the perturbation series
expansion is also known.
Part II

Relativistic Quantum Field


Theory

43
Chapter 4

CLASSICAL RELATIVISTIC
FIELDS

In this chapter a few essential facts of classical relativistic field theory are
summarized. It will first be shown how to derive the equations of motion of
a field theory, for example the Maxwell equations of electrodynamics, from a
Lagrangian. Second, the connection between symmetries of the Lagrangian
and conservation laws will be discussed1 .

4.1 Equations of Motion


The equations of motion of classical mechanics can be obtained from a La-
grange function by using Hamilton’s principle that the action for a given
mechanical system is stationary for the physical space–time development of
the system.
The equations of motion for fields that determine their space–time depen-
dence can be obtained in an analogous way by identifying the field amplitudes
at a coordinate ~x with the dynamical variables (coordinates) of the theory.
Let the functions that describe the fields be denoted by

Φα (x) with xµ = (t, ~x) , (4.1)

where α labels the various fields appearing in a theory. The Lagrangian L of


the system is expressed in terms of a Lagrange density L, as follows:
Z
L= L(Φα , ∂µ Φα ) d3 x (4.2)
1
The units, the metric and the notation used in this and the following chapters is
explained in Appendix A.

44
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 45

where the spatial integration is performed over the volume of the system.
The action S is then defined as usual by
Zt1 Z
S= L dt = L(Φα , ∂µ Φα ) d4 x (4.3)
t0 Ω

with the Lorentz-invariant four-dimensional volume element d4 x = d3 x dt.


The, in general finite, space–time volume of the system is denoted by Ω.
As pointed out before the fields Φα (x) play the same role as the gener-
alized coordinates qi in classical mechanics; the analogy here is such that
the fields Φα correspond to the coordinates q and the points ~x and α to the
classical indices i. The corresponding velocities are given in a direct analogy
by the time derivatives of Φα : ∂t Φα . Lorentz covariance then requires that
also the derivatives with respect to the first three coordinates appear; this
explains the presence of the four-gradients ∂µ Φα in (4.2).
In order to derive the field equations from the action S by Hamilton’s
principle, we now vary the fields and their derivatives

Φα → Φ0α = Φα + δΦα
∂µ Φα → (∂µ Φα )0 = ∂µ Φα + δ(∂µ Φα ) . (4.4)

This yields

δL = L(Φ0α , (∂µ Φα )0 ) − L(Φα , ∂µ Φα )


∂L ∂L
= δΦα + δ(∂µ Φα )
∂Φα ∂(∂µ Φα )
∂L ∂L
= δΦα + ∂µ (δΦα ) . (4.5)
∂Φα ∂(∂µ Φα )

According to the Einstein convention a summation over µ is implicitly con-


tained in this expression. In going from the second to the third line differen-
tiation and variation could be commuted because both are linear operations.
The equations of motion are now obtained from the variational principle
Z " ! !#
∂L ∂L ∂L
δS = − ∂µ δΦα + ∂µ δΦα d4 x
Ω ∂Φα ∂(∂µ Φα ) ∂(∂µ Φα )
= 0 (4.6)

for arbitrary variations δΦα under the constraint that

δΦα (t0 ) = δΦα (t1 ) = 0 ,


CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 46

where t0 and t1 are the time-like boundaries of the four-volume Ω. The last
term in (4.6) can be converted into a surface integral by using Gauss’s law;
for fields which are localized in space this surface integral vanishes if the
surface is moved out to infinity. Since the variations δΦα are arbitrary the
condition δS = 0 leads to the equations of motion
!
∂ ∂L ∂L
− =0. (4.7)
∂xµ ∂(∂µ Φα ) ∂Φα

The relativistic equivalence principle demands that these equations have


the same form in every inertial frame of reference, i.e. that they are Lorentz
covariant. This is only possible if L is a Lorentz scalar, i.e. if it has the same
functional dependence on the fields and their derivatives in each reference
frame.
In a further analogy to classical mechanics, the canonical field momentum
is defined as
∂L ∂L
Πα = = . (4.8)
∂ Φ̇α ∂(∂0 Φα )
From L and Πα the Hamiltonian H is obtained as
Z Z
3
H= Hd x= (Πα Φ̇α − L) d3 x . (4.9)

The Hamiltonian H represents the energy of the field configuration.

4.1.1 Examples
The following sections contain examples of classical field theories and their
formulation within the Lagrangian formalism just introduced. We start out
with the probably best-known case of classical electrodynamics, then general-
ize it to a treatment of massive vector fields and then move on to a discussion
of classical Klein–Gordon and Dirac fields that will play a major role in the
later chapters of this book.

Electrodynamics
The best-known classical field theory is probably that of electrodynamics,
in which the Maxwell equations are the equations of motion. The two ho-
mogeneous Maxwell equations allow us to rewrite the fields in terms of a
four-potential
~ ,
Aµ = (A0 , A) (4.10)
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 47

defined via
~ = ∇
B ~ ×A
~,
~
~ = −∇A
E ~ 0 − ∂A . (4.11)
∂t
Note that the 2 homogeneous Maxwell equations are now automatically ful-
filled. The two inhomogeneous Maxwell equations2
~ ·E
∇ ~ = ρ,
~
~ ×B
∇ ~ − ∂E = ~j (4.12)
∂t
with external density ρ and current ~j can be rewritten as
∂F µν
∂µ F µν = = jν , (4.13)
∂xµ
with the four-current
j ν = (ρ, ~j) (4.14)
and the antisymmetric field tensor
∂Aν ∂Aµ
F µν = − = ∂ µ Aν − ∂ ν Aµ . (4.15)
∂xµ ∂xν
The tensor Fµν is the dyadic product of two fourvectors and thus a Lorentz-
tensor. Equation (4.13) is the equation of motion for the field tensor or the
four-vector field Aµ .
It is easy to show that (4.13) can be obtained from the Lagrangian
1
L = − Fµν F µν − j ν Aν (4.16)
4
by using (4.7); the fields Aν here play the role of the fields Φα in (4.7). We
have
∂L
= −j ν ,
∂Aν
∂L 1
= − 2(+F µν − F νµ ) = −F µν . (4.17)
∂(∂µ Aν ) 4
The last step is possible because F is an antisymmetric tensor. The equation
of motion is therefore
!
∂ ∂L ∂L ∂F µν
− =− + jν = 0 , (4.18)
∂xµ ∂(∂µ Aν ) ∂Aν ∂x µ

2
Here the Heaviside units are used with c = 1 and 0 = µ0 = 1.
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 48

in agreement with (4.13). It is now easy to interpret the two terms in L


(4.16): the first one gives the Lagrangian for the free electromagnetic field,
whereas the second describes the interaction of the field with charges and
currents.
The two homogeneous Maxwell equations can also be expressed in terms
of the field tensor by first introducing the dual field tensor
1
F̃ µν = µνρσ Fρσ ; (4.19)
2
here µνρσ is the Levi–Civita antisymmetric tensor which assumes the values
+1 or −1 according to whether (µνρσ) is an even or odd permutation of
(0,1,2,3), and 0 otherwise. In terms of F̃ the homogeneous Maxwell equa-
tions read
∂µ F̃ µν = 0 . (4.20)
Eqs. (4.13) and (4.20) represent the Maxwell equations in a manifestly
covariant form. The Lagrangian (4.16) is obviously Lorentz-invariant since
it consists of invariant contractions of two Lorentz-tensors (the first term)
and two Lorentz-vectors (the second term). It is also invariant under a gauge
transformation
Aµ −→ Aµ0 = Aµ + ∂ µ φ (4.21)
because F itself is gauge-invariant by construction and the contribution of
the interaction term to the action is gauge-invariant for an external conserved
current. The same then holds for the equation of motion (4.18). Symmetry
(Lorentz-invariance), gauge-invariance and simplicity (there are no higher
order terms in (4.16)) thus determine the Lagrangian of electrodynamics.
The gauge freedom can be used to impose constraints on the four compo-
nents of the vector field Aµ . In addition, for free fields this gauge freedom can
be used, for example, to set the 0th component of the four-potential equal to
zero. Thus, a free electromagnetic field has only two degrees of freedom left.

Massive Vector Fields


Vector fields in which – in contrast to the electromagnetic field – the field
quanta are massive are described by the so-called Proca equation:

∂µ F µν + m2 Aν = j ν . (4.22)

Operating on this equation with the four-divergence ∂ν gives, because F is


antisymmetric,
m 2 ∂ν A ν = ∂ ν j ν . (4.23)
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 49

For m 6= 0 and a conserved current, this reduces the equation of motion


(4.22) to  
2 + m 2 Aν = j ν ; ∂ν Aν = 0 . (4.24)
Thus for massive vector fields the freedom to make gauge transformations
on the vector field is lost. The condition of vanishing four-divergence of the
field reduces the degrees of freedom of the field from 4 to 3. The space-like
components represent the physical degrees of freedom.
The Lagrangian that leads to (4.22) is given by
1 1
L = − F 2 + m2 A 2 − j · A . (4.25)
4 2

Klein–Gordon Fields
A particularly simple example is provided by the so-called Klein–Gordon field
φ that obeys the equation of motion
   
∂µ ∂ µ + m 2 φ = 2 + m 2 φ = 0 ; (4.26)

such a field describes scalar particles, i.e. particles without intrinsic spin.
The Lagrangian leading to (4.26) is given by
1 
L= (∂µ φ) (∂ µ φ) − m2 φ2 (4.27)
2
since we have
∂L
= ∂ µφ (4.28)
∂ (∂µ φ)
and
∂L
= −m2 φ . (4.29)
∂φ
It is essential to note here that the Lagrangian density
R
(4.27) that leads to
(4.26) is not unique; unique is only the action S = L d4x. For localized
fields for which the surface contributions vanish we can perform a partial
integration of the kinetic term in (4.26)
Z Z
d4x (∂µ φ) (∂ µ φ) = − d4x φ2φ (4.30)

so that we obtain
1  
L = − φ 2 + m2 φ . (4.31)
2
The Lagrangians (4.31) and (4.27) are equivalent. Since both give the same
action they also lead to the same equation of motion (4.26).
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 50

An interesting case occurs if we consider two independent real scalar fields,


φ1 and φ2 , with the same mass m. The total Lagrangian is then simply given
by a sum over the Lagrangians describing the individual fields, i.e.
1  1 
L= (∂µ φ1 ) (∂ µ φ1 ) − m2 φ21 + (∂µ φ2 ) (∂ µ φ2 ) − m2 φ22 . (4.32)
2 2
On the other hand, we can also construct two complex fields from the two
real fields φ1 and φ2 , namely
1
φ = √ (φ1 + iφ2 ) (4.33)
2
and its complex conjugate. In terms of these the Lagrangian (4.32) can be
rewritten to

L = (∂µ φ)∗ (∂ µ φ) − m2 φ∗ φ
 
= − φ∗ 2 + m2 φ . (4.34)

Dirac Fields
A particularly simple example is provided by the Dirac field Ψ for which the
equation of motion is just the Dirac equation

(iγ µ ∂µ − m) Ψ = 0 , (4.35)

where the γµ are the usual (4 × 4) matrices of Dirac theory. Ψ itself is a


(4 × 1) matrix of four independent fields, a so-called spinor.
The corresponding Lagrangian is given by

L = Ψ̄ (iγ µ ∂µ − m) Ψ . (4.36)

This can be seen by identifying the fields Φα in (4.7) with the four components
of the Dirac spinor Ψ̄ = Ψ† γ0 . Since L does not depend on ∂µ Ψ̄ the equation
of motion is simply given by
∂L
= (iγ µ ∂µ − m) Ψ = 0 . (4.37)
∂ Ψ̄

4.2 Symmetries and Conservation Laws


As in classical mechanics there is also in field theory a conservation law as-
sociated with each continuous symmetry of L. The theorem which describes
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 51

the connection between the invariance of the Lagrangian under a continu-


ous symmetry transformation and the related conserved current is known
as Noether’s theorem. In the following, this will be illustrated for different
types of symmetries which then lead to the well-known conservation laws.
The common expression in the arguments to follow is the change of the
Lagrangian density under a change of the fields and their derivatives (see
(4.4)). According to (4.5) and the Lagrange equations of motion (4.7) this
change is given by !
∂L
δL = ∂µ δΦα . (4.38)
∂(∂µ Φα )

4.2.1 Geometrical Space–Time Symmetries


In this section we investigate the consequences of translations in four-dimen-
sional space–time, i.e. infinitesimal transformations of the form
xν → x0ν = xν + ν , (4.39)
where ν is a constant infinitesimal shift of the coordinate xν . Under such
transformations the change of L is given by
∂L
δL = ν = ν ∂ ν L , (4.40)
∂xν
since L is a scalar.
If now L is required to be form-invariant under translations, it does not
explicitly depend on xν . In this case, δL is also given by (4.38). The changes
of the fields Φα appearing there are for the space–time translation considered
here given by
∂Φα
δΦα = ν = ν ∂ ν Φα . (4.41)
∂xν
Inserting (4.41) into (4.38) yields
!
∂L
δL = ν ∂µ ∂ ν Φα . (4.42)
∂(∂µ Φα )
Equating (4.42) and (4.40) finally gives
!
∂L
∂µ ∂ ν Φα − L g µν =0 , (4.43)
∂(∂µ Φα )
since the ν are arbitrary. By defining the tensor T µν as
∂L
T µν ≡ ∂ ν Φα − L g µν (4.44)
∂(∂µ Φα )
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 52

this equation reads


∂µ T µν = 0 . (4.45)
Relation (4.45) has the form of a continuity equation. Spatial integration
over a finite volume yields
 
d Z 0µ Z
∂T iµ 3 I
~ (µ) · ~n dS .
T (x)d3 x = − d x = − S (4.46)
dt ∂xi
V V S

~ (µ)
Here ~n is a unit vector vertical on the surface S pointing outwards and S
is a three-vector:
~ (µ) = (T 1µ , T 2µ , T 3µ ) .
S (4.47)
The surface integral on the rhs of (4.46) is taken over the surface S of
volume V. For localized fields it can be made to vanish by extending the
volume towards infinity. It is then evident that the quantities
Z
Pµ = T 0µ d3 x (4.48)

are conserved. These are the components of the four-momentum of the field,
as can be verified for the zeroth component,
Z Z !
0 00 3 ∂L
P = T d x= ∂ 0 Φ α − L d3 x
∂(∂0 Φα )
Z
= (Πα Φ̇α − L) d3 x = H , (4.49)

according to (4.8) and (4.9). The spatial components of the field momentum
are Z Z
k 0k 3 ∂L
P = T d x= ∂ k Φ α d3 x . (4.50)
∂(∂0 Φα )
T µν , as defined in (4.44), has no specific symmetry properties. It can,
however, always be made symmetric in its Lorentz-indices because (4.45)
does not define the tensor T uniquely. We can always add a term of the form
∂λ Dλµν , where D λµν is a tensor antisymmetric in the indices λ and µ, such
that T becomes symmetric.3
3
In classical mechanics the form invariance of the Lagrangian under rotations leads to
the conservation of angular momentum. Analogously, in a relativistic field theory the form
invariance of L under four-dimensional space–time rotations (Lorentz covariance) leads to
the conservation of a quantity that is identified with the angular momentum of the field.
To obtain the same form for the angular momentum as in classical mechanics it is essential
that Tµν is symmetric.
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 53

Comparing (4.46) with (4.50) and assuming T to be symmetric we see that


the normal components of the vectors S ~ (µ) in (4.47) describe the momentum
flow through the surface S of the volume V and thus determine the “radiation
pressure” of the field.4 These properties allow us to identify Tµν as the energy-
momentum tensor of the field. For the Lagrangian (4.16) of electrodynamics
Tµν is just the well-known Maxwell’s stress tensor.
As already mentioned at the beginning of this chapter these conservation
laws are special cases of Noether’s theorem, which can be summarized for
the general case as follows:
Each continuous symmetry transformation that leaves the La-
grangian invariant is associated with a conserved current. The
spatial integral over this current’s zeroth component yields a con-
served charge.

4.2.2 Internal Symmetries


Relativistic field theories may contain conservation laws that are not conse-
quences of space–time symmetries of the Lagrangian, but instead are con-
nected with symmetries in the internal degrees of freedom such as, e.g.,
isospin or charge.
We therefore now allow for a mixture of the different fields under the
transformation
Φα (x) → Φ0α (x) = e−iεqαβ Φβ , (4.51)
where  is an infinitesimal parameter and the qαβ are fixed c-numbers. We
then have
δΦα (x) = Φ0α (x) − Φα (x) = −iεqαβ Φβ (x) . (4.52)
The change of the Lagrangian is given by (4.38)
!
∂L
δL = ∂µ δΦα . (4.53)
∂(∂µ Φα )
If L is invariant under this variation δΦα , then we have
!
∂L
δL = ∂µ δΦα = 0 . (4.54)
∂(∂µ Φα )
Equation (4.54) is in the form of a continuity equation for the “current”
∂L 1
j µ (x) = δΦα . (4.55)
∂(∂µ Φα ) ε
4
More precisely, S k(µ) denotes the flux of the µth component of the field momentum
in the direction xk .
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 54

Inserting the field variations δΦα yields for the current


∂L
j µ (x) = −i qαβ Φβ . (4.56)
∂(∂µ Φα )
Equations (4.54) and (4.56) imply that the “charge”
Z Z
0 3 ∂L
Q= j (x) d x = −i qαβ Φβ d3 x (4.57)
∂(∂0 Φα )
of the system is conserved. The physical nature of these “charges” and “cur-
rents” has to remain open. It depends on the specific form of the symmetry
transformation (4.63) and can be determined only by coupling the system to
external fields.

Example: Quantum Electrodynamics


To illustrate this conservation law, the theory of electromagnetic interac-
tions is used as an example. However, in contrast to the considerations in
Sect. 4.1.1 we now consider a coupled system of a fermion field Ψ(x) and the
electromagnetic field Aµ (x) to determine the physical meaning of the con-
served current. Together with a quantization procedure this theory is called
Quantum Electrodynamics (QED). The Lagrangian is given by
1
L = − Fµν F µν + Ψ̄ [iγ µ (∂µ + ieAµ ) − m] Ψ . (4.58)
4
L contains a part that describes the free electromagnetic field (first term).
The second term describes the fermion Lagrangian; it is obtained from the
free particle Lagrangian of (4.36) by replacing the derivative ∂µ by the co-
variant derivative
Dµ = ∂µ + ieAµ (4.59)
(minimal coupling). Here e is the electron’s charge (e = −|e|).
The Lagrangian (4.58) is obviously invariant under a variation of the
fermion fields of the form
Ψ → Ψ0 = e−ie Ψ . (4.60)
Comparison with (4.51) gives qαβ = e δαβ so that the conserved “current”
given by (4.56) is:
jµ (x) = e Ψ̄γµ Ψ . (4.61)
Note that this conserved current is exactly the quantity that couples to the
electromagnetic field in (4.58). This property allows one to identify the
current (4.61) as the electromagnetic current of the electron fields.
CHAPTER 4. CLASSICAL RELATIVISTIC FIELDS 55

Example: Scalar Electrodynamics


The Lagrangian for the case of a complex scalar field interacting with an
electromagnetic field is given by
1
L = − Fµν F µν + (Dµ φ)∗ (Dµ φ) − m2 φ∗ φ . (4.62)
4
This Lagrangian is simply the sum of the free electromagnetic Lagrangian
(4.16) and the Lagrangian for a complex scalar field (4.34), where in the latter
again the derivative ∂µ has been replaced – through minimal substitution –
by the covariant derivative Dµ (4.59).
The Lagrangian (4.62) is obviously invariant under the phase transforma-
tions

φ(x) −→ e−iεe φ(x)


φ∗ (x) −→ e+iεe φ∗ (x) (4.63)

The conserved current connected with this invariance can be obtained from
the definition (4.56)
!
µ ∂L ∂L
j = −i eφ + ∗
(−e)φ∗
∂ (∂µ φ) ∂ (∂µ φ )
= ie (φ D φ − φ (D φ)∗ ) .
∗ µ µ
(4.64)

The conserved charge is then given by


Z Z   ∗ 
3 0
Q= d x j (x) = ie d3 x φ ∗ D 0 φ − φ D 0 φ . (4.65)

It is remarkable that now the electromagnetic field Aµ appears in the con-


served current (through the covariant derivative D µ = ∂ µ + ieAµ ). Again the
conserved current provides the coupling to the electromagnetic field. If the
scalar field is real then it is invariant under the transformation (4.63) only for
e = 0. Eq. (4.65) then shows that in this case the conserved charge Q = 0.
Chapter 5

PATH INTEGRALS FOR


SCALAR FIELDS

In this chapter we apply the methods developed in chapter 3 to the case of


scalar fields. There we showed that the vacuum expectation values of time-
ordered products of x̂ operators could be obtained as functional derivatives of
a generating functional. All these results can be taken over into field theory
remembering that fields play the role of the coordinates of the theory and the
spatial locations ~x correspond to the indices of the classical coordinates. This
implies that we can obtain the vacuum expectation values of time-ordered
field operators by performing the derivatives on an appropriate functional.
To discuss this functional is the main purpose of this chapter.
It is easy to see that these vacuum expectation values of time-ordered
products of field operators play an important role in quantum field the-
ory. Each field operator creates or annihilates particles and a time-ordered
product of field operators can thus describe the probability amplitude for a
physical process in which particles are created and annihilated. The quanti-
tative information about such a process is contained in the S matrix which
can be obtained from the vacuum expectation values of time-ordered field
operators by means of the so-called reduction theorem which we will derive
in a later chapter. The rest of this manuscript will therefore be concerned
with calculating these expectation values and with developing perturbative
methods for their determination when an exact calculation is not possible.

56
CHAPTER 5. PATH INTEGRALS FOR SCALAR FIELDS 57

5.1 Generating Functional for Fields


We assume that the system is described by a Lagrangian of the form
1 µ 
L (φ, ∂µ φ) = ∂ φ∂µ φ − m2 φ2 − V (φ) . (5.1)
2
In order to obtain the functional W [J] for fields we note that the fields play
the role of the coordinates of the theory and that sums over the different
coordinates have to be replaced by integrals over the space-time coordinates.
In order to define more stringently what is actually meant by a path integral
for fields we write it down here in detail for a free field Lagrangian (V = 0).
In order to do so we bring it into a form as close as possible to the classical
definition (1.38).
We first Fourier-expand the field
1 X ~
φ(~x, t) = √ q~k (t)eik·~x . (5.2)
V ~k

For a real field φ we have



q− ~k = q~k . (5.3)
Inserting this expansion into the free Lagrangian gives1
Z
1Z  µ
3

L = Ld x = ∂ φ∂µ φ − m2 φ2 d3x (5.4)
2
1 XZ h   i ~ ~0
= qk qk0 ~k · ~k 0 − m2 + q̇k q̇k0 ei(k+k )·~x d3x .
2V kk0

Integration over d3x gives V δ~k,−~k0 so that we have

1 X 
L = q̇k q̇−k − (~k 2 + m2 )qk q−k
2 k
1X
= (q̇k q̇k∗ − ωk qk qk∗ ) (5.5)
2 k
q
with ωk = ~k 2 + m2 . Using (5.3), writing qk = Xk + iYk , and grouping then
the coordinates Xk and −iYk into a new vector xk gives
1 X 2 
L= ẋk − ωk2 x2k . (5.6)
2 k
1
For ease of notation we drop the vector arrows in the indices
CHAPTER 5. PATH INTEGRALS FOR SCALAR FIELDS 58

In the general spirit of field theory we introduce a source density J which


we also expand
1 X ~
J(~x, t) = √ j~k (t)eik·~x . (5.7)
V ~k
With this expansion we obtain for the source term
Z X X
d3x J(x)φ(x) = jk (t)q−k (t) = Jk (t)xk (t) (5.8)
k k

where we have grouped jk and −ijk into a new vector Jk .


The Lagrangian (5.6) has the structure of a Lagrangian with quadratic
momentum dependence and constant coefficient, discussed in Sect. 1.3.1; also
the souce term reads formally just the same as for the nonrelativistic systems
treated in chapter 3. We can thus again integrate the momentum dependence
out and obtain for the vacuum-to-vacuum transition amplitude (cf. (3.23))

W [J] = lim h0tf |0ti iJ


tf →+∞
ti →−∞
! n+1 Z n
P
n iη (Ll +Jk xkl )
1 2 YY
= lim dxkj e l=0 , (5.9)
η→0 2πh̄iη
k j=1

with
1 X 2 
Ll = L (xl , ẋl ) = ẋkl − ωk2 x2kl , (5.10)
2 k
where the first index (k) denotes the coordinate and the second (j) the time
interval. If we now identify the integration measure as
! n+1 n
1 2 YY
Dφ = lim dxkj , (5.11)
η→0 2πh̄iη
k j=1

we can write the generating functional also as


Z  Z   
1h   i
W [J] = Dφ exp i ∂µ φ∂ µ φ − m2 − iε φ2 + Jφ d4x
Z  Z 
2  
ε 2 4
= Dφ exp i L (φ, ∂µ φ) + Jφ + i φ d x . (5.12)
2
Here the volume element d4x is given by the Lorentz-invariant expression

d4x = d3x dt . (5.13)

The term iεφ2 /2 with positive ε has been introduced in an ad hoc manner
to ensure the convergence of W when taking the fields to infinity with the
CHAPTER 5. PATH INTEGRALS FOR SCALAR FIELDS 59

understanding that ultimately ε has to be taken to 0 (cf. the discussion at


the end of Sect. (3.1)).
The second line of (5.12) also gives the generating functional for an in-
teracting theory with the interaction V included in L. This can be easily
proven by including an additional interaction in the Lagrangian (5.4) and
Fourier-expanding it. The generating functional for a scalar field theory is
thus given by
Z R
d4x[ 12 φ(2+m2 −iε)φ+V (φ)−Jφ]
W [J] = Dφ e−i (5.14)

where the Lagrangian in the form (4.31) has been used.


In analogy to (3.24) we define a normalized functional
W [J]
Z[J] = . (5.15)
W [0]
Again the infinite normalization factor inherent in W [J] cancels out in this
definition. Since W [J] involves an exponential it will often be convenient
in the following discussions to introduce its logarithm iS[J] which is itself a
functional of J
W [J] ≡ eiS[J] =⇒ S[J] = −i ln W [J] = −i ln Z[J] − i ln W [0] . (5.16)

5.1.1 Euclidean Representation


In the preceding section we have ensured convergence of the generating func-
tional by introducing the ε-dependent term in the energies. In this subsection
we go back to the alternative method that relies on the Wick rotation that
we introduced in Sect. 3.21 and discuss the Euclidean representation of the
generating functional. This discussion also illustrates in some more detail the
remarks on integrating oscillatory functions made at the end of Sect. 1.3.1.
The real Euclidean space is obtained from Minkowski space by rotating
the real axis in the x0 plane by δ = −π/2 into the negative imaginary axis
(Wick rotation). We denote a space-time point in Euclidean space by xE ; it
is related to the usual space-time point x in Minkowski space by
xE = (~x, x4 ) with x4 = ix0 = it . (5.17)
Under the Wick rotation t → −it and x4 thus becomes real. With this
definition we can extend the usual Minkowski-space definitions of volume
element and space-time distance to Euclidean space
d4 xE ≡ d3 x dx4 = d3 x idt = id4 x
3
X
dx2E = dx2j + x24 = −dx2 . (5.18)
j=1
CHAPTER 5. PATH INTEGRALS FOR SCALAR FIELDS 60

The d’Alembert operator is then given by


4
∂2 2
~ 2 =− ∂ −∇~ 2=−
X ∂2
2= − ∇ ≡ −2E . (5.19)
∂t2 ∂x24 a=1 ∂xa
2

With these transformations the generating functional for a free scalar field
in Minkowski space (5.14) becomes in its Euclidean representation
Z R
d4xE { 21 [(∂E φ)2 +m2 φ2 ]−Jφ}
WE0 [J] = Dφ e− , (5.20)

with (∂E φ)2 = − (∂φ)2 . Because x4 is now real, (∂E φ)2 is always positive
and the exponent is negative definite; the integral thus converges and is well-
defined even without adding in the ε-dependent term. Since the exponent
is furthermore quadratic in the fields, WE0 [J] can be evaluated by using the
techniques for Gaussian integrals that are explained in App. B. Physical
results are then obtained by rotating backwards after all integrations have
been performed.
Remembering that in field theory the fields play the role of the coordinates
of a Lagrangian theory we can now directly generalize some of the results of
Chapt. 3 to field theory. In particular, we have that WE [J] of eq. (5.20) is the
transition amplitude from the vacuum state of the free theory at t → −∞
to that at t → +∞ under the influence of the external source J (cf. (3.27)
so that the normalized transition amplitude ZE is given by
WE0 [J] h0 + ∞|0 − ∞iJ
ZE0 [J] = 0
= , (5.21)
WE [0] h0 + ∞|0 − ∞i0
where |0i is the vacuum state of the free theory. Eq. (5.20) shows that the
normalized transitionR amplitude can be understood as an integration of the
source action exp (+ d4xE Jφ) with the weights
R
1
d4xE [(∂E φ)2 +m2 φ2 ]
wE0 (φ) = e− 2 (5.22)

over all fields φ R


R 4
Dφ wE0 (φ)e d xE Jφ
ZE0 [J] = R . (5.23)
Dφ wE0 (φ)
Eq. (3.47) then shows that for a function of the fields O(φ)
R Q
Dφ wE0 (φ) j Oj (φ(xj )) Y
R = h0|T [ Oj (φ̂(xj ))]|0i . (5.24)
Dφ wE0 (φ) j

The field operators on the rhs here are those of free fields.
CHAPTER 5. PATH INTEGRALS FOR SCALAR FIELDS 61

In the interacting case all these relations still hold if wE0 is replaced by a
weight function for the interacting theory and the vacuum state is now that
of the interacting theory which we will denote by |0̃i. To obtain the weight
function of the interacting theory we use a perturbative treatment of the
interaction V . By writing in analogy to (5.22)
R
d4xE { 21 [(∂E φ)2 +m2 φ2 ]+V (φ)}
wE (φ) = e− (5.25)
R
R R V (φ) d4xE
= e− V (φ) d4xE − 2
e
1
[(∂E φ)2 +m2 φ2 ] d4xE ≡ e− 0 (φ)
wE

we get with (5.24)


Y Z Y
h0̃|T [ Oj (φ̂)]|0̃i = Dφ wE (φ) Oj (φ)
j j
R
R V (φ) d4xE Q
Dφ wE0 (φ)e− j Oj (φ)
= R R
V (φ) d4x
Dφ wE0 (φ)e− E

R
Q ˆ − V (φˆ0 ) d4xE
h0|T [ j Oj (φ0 )e ]|0i
= R . (5.26)
V (φˆ0 ) d4xE
h0|T [e− ]|0i

Here |0i on the rhs is the vacuum state of the non-interacting free theory
(V = 0) and all the field operators on the rhs are free field operators φ̂0 at all
times if they were free at t → −∞. This can be seen from the path integral
which contains the free weight wE0 connected with free propagation.
In the following chapters it is often useful to think in terms of this prob-
abilistic interpretation even when we work in Minkowski metric. This is
possible by relating Euclidean and Minkowski space by analytic continua-
tion.
Chapter 6

EVALUATION OF PATH
INTEGRALS

Only a limited class of path integrals can be evaluated analytically [Grosche-


Steiner] so that one is forced to use either numerical or perturbative methods.
In this section we first calculate the generating functional for free scalar fields,
both by a direct reduction method and, to gain familiarity with this tech-
nique, with the the methods of Gaussian integration developed in App. B.2.1.
We then show how more general types of path integrals can approximately be
reduced to the Gaussian form. In this reduction we find a systematic method
for a semiclassical expansion in terms of the Planck constant h̄. Finally, we
briefly discuss methods for the numerical evaluation of path integrals.

6.1 Free Scalar Fields


The generating functional for a free scalar field theory plays a special role
in the theory of path integrals. This is so because of two reasons: first, a
free-field theory is the simplest possible field theory and, second, as we have
just seen in the last section the effects of the interaction term V (φ) can be
described with the help of perturbation theory in which the functional of the
full, interacting theory is expanded around that of the free theory.

6.1.1 Generating functional


The representation of the generating functional of a free field theory (V = 0 in
(5.14)) as a path integral is still quite cumbersome for practical applications.
For these it would be very desirable if we could factorize out the functional
dependence on J in form of a normal integral; the remaining path integral

62
CHAPTER 6. EVALUATION OF PATH INTEGRALS 63

would then disappear in the normalization.


In the following we will therefore separate the path integral (5.9) into 2
factors, one depending on J and the other one being an integral over φ, i.e.
just a number. For this purpose we start with the generating functional
Z R
d4x[ 12 φ(2+m2 −iε)φ−Jφ]
W [J] = Dφ e−i (6.1)

where the Lagrangian in the form (4.31) has been used. The field φ here is
an integration variable; it does not fulfill a Klein-Gordon equation! We now
introduce a field φ0 that does just that
h i
2 + (m2 − iε) φ0 (x) = J(x) . (6.2)
J thus plays the role of a source to φ0 . We now take this field φ0 as a reference
field and expand φ around it, setting φ = φ0 + φ0 . We thus obtain for the
integrand in the exponent in (6.1)
1  
(φ0 + φ0 ) 2 (φ0 + φ0 ) + (φ0 + φ0 ) m2 − iε (φ0 + φ0 ) − J (φ0 + φ0 )
2
1 0h  i
= φ 2 + m2 − iε φ0
2
1 h  i 1 h  i
+ φ0 2 + m2 − iε φ0 + φ0 2 + m2 − iε φ0
2 2
1 0h 
2
i
+ φ 2 + m − iε φ0 − Jφ0 − Jφ0 . (6.3)
2
The third and the fourth terms on the rhs give the same contribution when
integrated over. The second term gives, according to (6.2), 21 Jφ0 . Collecting
all terms we therefore have for the action in (6.1)
Z  
1 h  i
4
S[φ, J] = − d x φ 2 + m2 − iε φ − Jφ
2
Z 
1 h  i 1
= − d4x φ0 2 + m2 − iε φ0 + Jφ0
2h  i
2 o
2
+ φ 2 + m − iε φ0 − Jφ0 − Jφ0 .
0
(6.4)

In the last line of this equation we can again apply (6.2) to obtain
1Z 4 n 0h  i o
S[φ, J] = − d x φ 2 + m2 − iε φ0 − Jφ0 (6.5)
2
We are now very close to our aim to factorize out the J dependence. To
do so we solve (6.2) by writing
Z
φ0 (x) = − DF (x − y)J(y) d4y , (6.6)
CHAPTER 6. EVALUATION OF PATH INTEGRALS 64

where DF , the so-called Feynman propagator, fulfills the equation


h  i
2 + m2 − iε DF (x) = −δ 4 (x) (6.7)

in complete analogy to the nonrelativistic propagator K in chapter 1. Using


 4 Z
4 1
δ (x) = d4k e−ikx (6.8)

we obtain
1 Z e−ik(x−y)
DF (x − y) = d4k . (6.9)
(2π)4 k 2 − m2 + iε
Substituting (6.6) into the action (6.5) gives
 
1Z 4 h  i Z
S[φ, J] = − d x φ0 2 + m2 − iε φ0 + J(x) DF (x − y)J(y) d4y
2
1Z 4 n 0h  i o
= − d x φ 2 + m2 − iε φ0
2
1Z
− J(x)DF (x − y)J(y) d4x d4y . (6.10)
2
The exponential of the last term no longer depends on φ and can, therefore,
be pulled out of the pathintegral (6.1). The pathintegral involving the expo-
nential of the first term appears also in the denominator of the normalized
generating function (5.15) and thus drops out.
We thus obtain now for the normalized generating functional

W [J] i
R 4 4
Z0 [J] = = e− 2 J(x)DF (x−y)J(y) d x d y . (6.11)
W [0]

This is the vacuum-to-vacuum transition amplitude for a free scalar field


theory. Note that it no longer involves a path integral.

6.1.2 Feynman propagator


The imaginary part of the mass, originally introduced to achieve convergence
for the path integrals, appears here now in the denominator of the propagator
DF
1 Z e−ikx
DF (x) = d4k . (6.12)
(2π)4 k 2 − m2 + iε
It determines the position of the poles in DF , which are, in the k0 -integration,
at
k02 = ~k 2 + m2 − iε (6.13)
CHAPTER 6. EVALUATION OF PATH INTEGRALS 65

Im k0

–ω k+iδ

Re k0

+ω k–iδ

Figure 6.1: Location of the poles in the Feynman propagator.

or q
k0 = ± ~k 2 + m2 ∓ iδ . (6.14)
The poles are therefore located as indicated in Fig. 6.1.
The location of the poles, originally introduced only in an ad-hoc way to
achieve convergence of the path integrals, determines now the properties of
the propagator of the free Klein-Gordon equation, the Feynman propagator.
This can be seen by rewriting the Feynman propagator DF in the following
form
1 Z 4 e−ikx
DF (x) = d k (6.15)
(2π)4 k 2 − m2 + iε
1 Z 3 e−ikx
= d k dk 0
(2π)4 k02 − ~k 2 − m2 + iε
Z  
1 3 −ikx 1 1 1
= d k dk0 e −
(2π)4 2ωk k0 − ωk + iδ k0 + ωk − iδ

with ωk2 = ~k 2 + m2 .
We now first perform the integration over k0 . Since the exponential con-
tains a factor e−ik0 t , the path can be completed in the upper half plane for
negative times and in the lower half for t > 0. Cauchy’s theorem then gives
2πi × the sum of the residues of the enclosed poles
~
i Z 3 eik·~x  +iωk t −iωk t

DF (x) = d k −Θ(−t)e − Θ(t)e
(2π)3 2ωk
~
i Z 3 eik·~x  +iωk t −iωk t

= − d k Θ(−t)e + Θ(t)e (6.16)
(2π)3 2ωk
CHAPTER 6. EVALUATION OF PATH INTEGRALS 66

(in the second term here an extra “−” sign appears because of the negative
direction of the contour integral).
It can now be shown that DF propagates free fields with negative fre-
quencies backwards in time and those with positive frequencies forwards. To
demonstrate this we write
~
i Z 3 eik·~x  +iωk x0 −iωk x0

DF (x) = − d k Θ(−x 0 )e + Θ(x 0 )e
(2π)3 2ωk
Z
i 1 ikx
= − Θ(−x0 ) 3
d3k e
(2π) 2ωk
i Z 3 1 −ikx
− Θ(x0 ) dk e . (6.17)
(2π)3 2ωk

Here we have changed ~k → −~k in the first integral, which does not change
its value under this substitution.
The integrands are products of orthogonal, normalized solutions of the
Klein-Gordon equation
(±) 1
ϕ~k (x) = q e∓ikx (6.18)
(2π)3 2ω k

which fulfill the normalization and orthogonality conditions appropriate for


a scalar field (cf. (4.65) without an A field)
Z  
d3x ϕ~k0 (x)ϕ̇~k (x) − ϕ̇~k0 (x)ϕ~k (x) = ±δ 3 (~k 0 − ~k) .
(±)∗ (±) (±)∗ (±)
i (6.19)

and Z  
(±)∗ (∓) (±)∗ (∓)
i d3x ϕ~k0 (x)ϕ̇~k (x) − ϕ̇~k0 (x)ϕ~k (x) = 0 . (6.20)
We can thus write
1 Z 3 1 1
iDF (x) = Θ(−x0 ) 3
dk √ √ eikx (6.21)
(2π) 2ωk 2ωk
1 Z 3 1 1
+ Θ(x0 ) 3
dk √ √ e−ikx
(2π) 2ωk 2ωk
Z Z
3 (−)∗ (−) (+)∗ (+)
= Θ(−x0 ) dk ϕ~k (0)ϕ~k (x) + Θ(x0 ) d3k ϕ~k (0)ϕ~k (x) .

Comparison with (1.13) shows that negative-frequency solutions are prop-


agated backwards in time, and positive-frequency solutions forward. This
particular behavior is a consequence of the location of the poles relative to
the integration path. This location is fixed by the sign of the ε term which
in turn was needed to achieve convergence for the generating functional.
CHAPTER 6. EVALUATION OF PATH INTEGRALS 67

Euclidean representation. The Feynman propagator can also be given


in a Euclidean representation. We can define a corresponding Euclidean
momentum space by requiring that k 0 x0 = k4E xE4 ; this condition ensures that
a plane wave propagating forward in time does so both in Euclidean and in
Minkowski space. We thus get
kE = (~k, k4 ) with k4 = −ik0 . (6.22)
The eigenvalues of 2E become k 2 = k02 − ~k 2 = −(k42 + ~k 2 ) = −kE2 < 0.
This gives for the momentum space volume element and the energy-mo-
mentum distance in Euclidean space
d4 kE ≡ d3 k dk4 = −d3 k idk0 = −id4 k
3
X
dkE2 = dkj2 + dk42 = −dk 2 . (6.23)
j=1

Combining both of these definitions yields for the typical exponent of a plane
wave
kx = kµ xµ = k0 x0 − ~k · ~x = ik4 (−i)x4 − ~k · ~x
= k4 x4 − ~k · ~x 6= kE xE . (6.24)
Note that this is not equal to the Euclidean scalar product. However, in
Fourier transforms, where this expression often appears, we always have an
integration over d3 k and can thus change ~k → −~k; thus in these Fourier
integrals – and only there – we can replace kx by kE xE .
The Feynman propagator can now be rewritten. For that purpose we
choose a different path for the integration over k0 after we have Wick-rotated
the time axis. Instead of integrating along the real energy (k0 ) axis we
integrate along the imaginary energy axis. We then close the integration
path in the right half of the k0 plane for t > 0 and in the left half for t < 0.
Since in this way the same poles are included as on the original path the
value of the integral does not change. This fact makes this rotation in the
k0 -space different from the one in the t-space where the imaginary time-axis
has to be rotated back to the real one after the calculations in order to obtain
physical results.
The Euclidean propagator then reads
i Z e−ikE xE 4
DF (x) = − d kE
(2π)4 kE2 + m2
i Z e−ikE xE
= − d4kE . (6.25)
(2π) 4 ~ 2 2
k + k4 + m 2

Because k4 is real, the integral no longer contains any poles on its integration
path and is therefore well defined.
CHAPTER 6. EVALUATION OF PATH INTEGRALS 68

6.1.3 Gaussian Integration


In Sect. 1.3 we have already used a Gaussian integral relation to integrate
out the p-dependence of the path integral. In many cases the generating
functions appearing in field theory are of a form that contains the fields and
their derivatives only in quadratic form so that again a Gaussian method can
be used. In order to gain familiarity with this technique we derive in this
section again the generating functional for the free scalar field theory.
We thus apply the Gaussian integration formulas of Sect. B.2.1 to the
generating functional (6.1)
Z R R
− 2i φ(2+m2 −iε)φ d4x i Jφ d4 x
W [J] = Dφ e e . (6.26)

In order to make the matrix structure of the exponent more visible we use
two-fold partial integration and write
Z R R
i
φ(x)δ 4 (x−y)(2y +m2 −iε)φ(y) d4x d4y i Jφ d4 x
W [J] = Dφ e− 2 e
Z R R
− 2i φ(x)[(2y +m2 −iε)δ 4 (x−y)]φ(y) d4x d4y i Jφ d4 x
= Dφ e e
Z R
Dφ e− { 2 φ(xE )[(−2y +m )δ(xE −yE )]φ(yE ) d yE −Jφ} d xE . (6.27)
1 E 2 4 4

In the last step we have gone over to the Euclidean representation of the
generating functional (cf. (5.20)), using 2E = −2 and δ 4 (xE −yE ) = −iδ 4 (x−
y) (cf. Sect. 5.1.1).
The integrals appearing here are now of Gaussian type and can thus be
integrated by using the expressions developed in the last section. We first
identify the matrices A and B in the Gaussian integration formula (B.18) as

AE (x, y) = −(2E 2 4
y − m )δ (xE − yE )
B(x) = −J T (x) . (6.28)

AE is real with positive eigenvalues (kE2 +m2 > 0) as required by the derivation
in section B.2.1. It is also symmetric as can be seen by writing the d’Alembert
operator in a discretized form, e.g.

d2
1
2
φ(x) = lim [φ(xi + h) − 2φ(xi ) + φ(xi − h)]
dx h→0 h2
xi
= Aij φ(xj ) (6.29)

with φ(xi + h) = φ(xi+1 ) etc. and

Aij = δi,j−1 − 2δi,j + δi,j+1 .


CHAPTER 6. EVALUATION OF PATH INTEGRALS 69

Thus the necessary conditions for the application of (B.13) are fulfilled. Ap-
plying now (B.18) gives
h i− 1 R −1
1
J(xE )(AE (xE ,yE )) J(yE ) d4 xE d4 yE
W E [J] = det(−2E + m2 )δ(xE − yE ) 2
e2 .
(6.30)
The Wick rotation back to real times is easily performed by the transfor-
mation (5.19) 2E → −2 − iε; it reintroduces the term +iε to guarantee the
proper treatment of the poles. In this case A becomes

AE → A = (2 + m2 − iε)iδ(x − y) . (6.31)

and the volume element

d4xE d4yE → −d4 xd4 y (6.32)

so that we finally obtain1


1 R −1
e− 2 J(x)[i(2y +m +iε)δ (x−y)] J(y) d x d y .
1 2 4 4 4
W [J] = √ (6.33)
det A
The determinant of a matrix A is in general given by the product of its
eigenvalues αi . Therefore we have
!
Y X
ln(det A) = ln αi = ln αi = tr ln A . (6.34)
i i

In the present case, though, A contains an operator and the notation det A
deserves some explanation. The matrix is given by
    1 Z 4 ik(x−y)
A(x, y) = 2 + m2 iδ(x − y) = 2 + m2 i dke
(2π)4
i Z 4  2 2

= d k −k + m eik(x−y)
(2π)4
Z Z
1 0
  1
= i d k d4k 0 q
4
eik x −k 2 + m2 δ 4 (k − k 0 ) q e−iky .
(2π) 4 (2π) 4

Thus the momentum representation of the operator A is given by


 
A(k, k 0 ) = −k 2 + m2 δ 4 (k − k 0 ) . (6.35)

We now evaluate the trace of the logarithm of this matrix where – in ac-
cordance with (6.34) – the logarithm of a matrix is explained by taking the
1
Note that formally we could have obtained this also by using (B.18) with A = i(2 y +
m − iε)δ 4 (x − y), B = −iJ, C = 0.
2
CHAPTER 6. EVALUATION OF PATH INTEGRALS 70

logarithm of each of the diagonal elements, i.e. the eigenvalues, after the
matrix has been diagonalized. This yields in the present case
Z
4 d4k 4 0 4
4 0
tr ln A(x, y) = d xd y 4
d k δ (x − y)ei(k x−ky) ln(−k 2 + m2 )δ 4 (k − k 0 )
(2π)
Z 4
dk
= d4x 4
ln(−k 2 + m2 ) (6.36)
(2π)

Now we use that the inverse operator appearing in (6.33) is just the
Feynman propagator. This can be seen by deriving the equation of motion
for the inverse operator
Z
4
δ (x − y) = A(x, z)A−1 (z, y) d4z
Z h   i
= i 2z + m2 − iε δ 4 (x − z) A−1 (z, y) d4z
Z  
= i δ 4 (x − z) 2z + m2 − iε A−1 (z, y) d4z
 
= i 2x + m2 − iε A−1 (x, y) . (6.37)

This is just the defining equation for −DF (6.7), so that we have

A−1 = iDF . (6.38)

The propagator can thus be obtained as the inverse of the operator between
the two fields in the Lagrangian for the free field (4.31).
For the normalized generating functional we thus obtain

W [J] i
R 4 4
Z0 [J] = = e− 2 J(x)DF (x−y)J(y) d x d y . (6.39)
W [0]

Equation (6.39) is the result derived earlier in section 5.1 (cf. (6.1.1)). The
propagator that appears here is just given by the inverse of the Klein-Gordon
Operator  −1
DF (x − y) = − 2 + m2 δ(x − y) , (6.40)
i.e. of that operator that appears between the two fields in the Lagrangian
for a free Klein-Gordon field
1
L = − φ(2 + m2 )φ . (6.41)
2
CHAPTER 6. EVALUATION OF PATH INTEGRALS 71

6.2 Stationary Phase Approximation


If the path integral in question is not that over a Gaussian, it can be approxi-
mately brought into a Gaussian form by using the so-called stationary phase
or saddle point method. In this method one first looks for the stationary
point of the exponent in the path integral. As explained earlier this will give
a major contribution to the path integral. The remaining contributions are
approximated by expanding the exponent around the stationary point.
We illustrate this method here for the case of a scalar field with selfinter-
actions. The Lagrangian is given by
1 h i
L = − φ 2 + m2 φ − V (φ) (6.42)
2
and the action Z
S[φ, J] = d4x (L + Jφ) (6.43)
is a functional of the field φ and the source J. We next determine the
stationary point by looking for the zero of the functional derivative

δS[φ, J]  
!
= − 2 + m2 φ0 (x) − V 0 (φ0 (x)) + J(x) = 0 ; (6.44)
δφ(x) φ0

this is the classical equation of motion corresponding to the action S[φ, J].
The stationary field is just the classical field; the corresponding classical
action is
Z  
4 1  2

S[φ0 , J] = − d x φ0 2 + m φ0 + V (φ0 ) − Jφ0 . (6.45)
2
We now expand S[φ, J] around this stationary field (cf. (B.35),(B.37))
S[φ, J] = S[φ0 , J] (6.46)

1 Z
δ2S
+ d4x1 d4x2
[φ(x1 ) − φ0 (x1 )] [φ(x2 ) − φ0 (x2 )] + · · · .
2 δφ(x1 )δφ(x2 ) φ0

The second functional derivative appearing here can be obtained by varying


the first derivative (6.44). We thus get from the definition (B.26)

δ2S
δ h 2 0
i

= − (2 + m )φ + V (φ) − J (6.47)
δφ(x2 )δφ(x1 ) φ0 δφ(x2 ) 1
φ0

Using now (B.27) we get



δ2S h i
= − 2 + m2 + V 00 (φ0 ) δ 4 (x2 − x1 )

(6.48)
δφ(x2 )δφ(x1 ) φ0 1
CHAPTER 6. EVALUATION OF PATH INTEGRALS 72

which is an operator. The index 1 here means that the corresponding ex-
pressions are to be taken at the point x1 .
The action (6.45) is calculated at the fixed classical field φ0 . It can,
therefore, be taken out of the path integral so that we finally obtain
Z R
i
d4x (L+Jφ)
W [J] = Dφ e h̄ (6.49)
Z  Z
i i
= e h̄ S[φ0 ,h̄J] Dφ exp − d4x1 d4x2
2h̄ !
n nh i o
2 00 4
× [φ(x1 ) − φ0 (x1 )] 2 + m + V (φ0 ) δ (x2 − x1 ) [φ(x2 ) − φ0 (x2 )]}
1

+ ... . (6.50)
In order to facilitate the following discussion we have put the unit of action,
h̄, explicitly into this expression by setting i → i/h̄ and J → h̄J.
The path integral remaining here is now in a Gaussian form. It can be
evaluated after a Wick rotation, just as in the developments leading to (6.33).
After a “coordinate
√ transformation” φ → φ0 = φ − φ0 and after scaling the
fields by φ → h̄φ we get for the generating functional
i
n h   io− 1
W [J] = e h̄ S[φ0 ,h̄J] det i 2 + m2 + V 00 (φ0 ) δ 4 (x2 − x1 ) 2
. (6.51)

We now perform a normalization with respect to the free case (6.33), i.e. to
n h   io− 1
W0 [0] = det i 2 + m2 δ 4 (x2 − x1 ) 2

1
= {A(x1 , x2 )}− 2 (6.52)
with A from (6.31); the index 0 on W denotes V = 0. This gives for the
normalized generating functional
W [J] i
W̃ [J] = = e h̄ S[φ0 ,h̄J] (6.53)
W0 [0]
 Z − 1
2
4 −1 00 4
× det d z A (x2 , z) (A(z, x1 ) + iV (φ0 (x1 ))) δ (z − x1 ) .

With A−1 = iDF (6.38) we get


i
n h io− 1
W̃ [J] = e h̄ S[φ0 ,h̄J] det δ 4 (x1 − x2 ) − DF (x2 − x1 )V 00 (φ0 (x1 )) 2
(6.54)
Our aim is now to write the inverse root of the determinant as a correction
term to the classical action. For this purpose we use (6.34)
1
− 12 − tr ln[. . .]
{det[. . .]} =e 2 . (6.55)
CHAPTER 6. EVALUATION OF PATH INTEGRALS 73

The matrix is given by


h i
hx1 |1 − DF V 00 (φ0 )|x2 i = δ 4 (x1 − x2 ) − DF (x2 − x1 )V 00 (φ0 (x1 )) . (6.56)

The trace of its logarithm is then given by


Z
tr ln[. . .] = d4x ln [1 − DF (0)V 00 (φ0 (x))] . (6.57)

We can now write


i
W̃ [J] = e h̄ S[φ0 ,J] (6.58)
with
Z  
Z
1  
S[φ0 , J] = − d4x φ0 2 + m2 φ0 + V (φ0 ) + h̄ d4x Jφ0
2
i Z
+ h̄ d4x ln [1 − DF (0)V 00 (φ0 (x))] + O(h̄2 ) . (6.59)
2
The first line is just the classical action S[φ0 , J]. The two terms can be
summed with a resulting action
Z  
1  
S[φ0 , J] = − 4
d x φ0 2 + m2 φ0 + Veff (φ0 ) + h̄Jφ0 (6.60)
2
with the effective potential
i
Veff (φ(x)) = V (φ(x)) − h̄ ln [1 − DF (0)V 00 (φ0 (x))] . (6.61)
2
Expression (6.59) shows that the saddle point approximation amounts to
an expansion of the action in powers of h̄. This is in accordance with the
discussion in Sect. 1.4 that quantum mechanics describes the fluctuations
of the action around the classical path. The potential Veff incorporates the
effects of these fluctuations into a classical potential.
Eq. (6.59) suggests a perturbative treatment through an expansion of the
logarithm (ln(1 + x) = x − x2 /2 + x3 /3 − . . .) in terms of DF V 00 , i.e. the
strength of the potential. The trace corresponds to an integration over x
such that the initial and final space-time points in the individual terms in
the expansion are identical, i.e. to a closed loop integration. With higher
orders in the expansion of the logarithm more and more vertices appear,
but they are always located on this one closed loop. This loop expansion
approximates the quantum mechanical behavior, whereas the perturbation
treatment takes interactions into account.
If we take φ0 as a constant field, i.e. if we neglect its space-time de-
pendence through the d’Alembert operator in (6.44), then the operator 1 +
CHAPTER 6. EVALUATION OF PATH INTEGRALS 74

DF V 00 (φ0 ) becomes local in momentum space. In this case we can evaluate


the trace of its logarithm by integrating over the eigenvalues

hx1 | ln [1 − DF V 00 (φ0 )] |i
d4k −ik(x2 −x1 )
Z  
1 00
= e ln 1 − 2 V (φ0 ) . (6.62)
(2π)4 k − m2 + iε

This equation writes the original matrix in x-space as a result of a unitary


transformation of a diagonal matrix in k-space. The logarithm is taken of
this diagonalized matrix which is then transformed back to x-space.

6.3 Numerical Evaluation of Path Integrals


An alternative method for the evaluation of path integrals is that of direct
numerical computation; with rapidly increasing computer power this method
becomes more and more important nowadays.

6.3.1 Imaginary time method


The generating functional is in general given by

W [J] = h0|e−i(Ĥ+J)(tf −ti ) |0i (6.63)

for ti → −∞ and tf → +∞, as we have seen in chapter 3. After a Wick


rotation this becomes

W [J] = lim h0|e−β(Ĥ+J) |0i , (6.64)


β→∞

where β denotes the real Euclidean time.


It is immediately obvious that (6.64) also equals the groundstate expec-
tation value of the statistical operator of quantum statistics if we identify β
with the inverse temperature, i.e. β = 1/T .
Inserting the explicit definition (5.9) and performing the Wick rotation
gives
! n+1 Z n
P
n β
n+1 2 YY − n+1 (Hl +Jk xkl )
W [J] = lim dxkj e l=0 , (6.65)
n→0 2πh̄β
k j=1

with
1 Xh 2 i
Hl = Hl (xl , ẋl ) = ẋkl + ωk2 x̄2kl + V (xkl ) , (6.66)
2 k
CHAPTER 6. EVALUATION OF PATH INTEGRALS 75

so that W [J] can be written as


!
n + 1 Z YY β
P
W [J] = lim dxkj ρ(xkj )e− n+1 l V (xkl ) (6.67)
n→0 2πh̄β
k j

with β
P
ρ(x) = e− n+1 [ 1
l 2
(ẋ2kl +ωk2 x2kl )+Jk xkl ]. (6.68)
The function ρ(x) is of Gaussian shape and can, therefore, analytically be
normalized into
ρ(x)
P (x) = R . (6.69)
dx ρ(x)
The multiple integrals appearing here can be evaluated by a Monte-Carlo
technique which samples the integrand at a large number of points, where
each ‘point’ really corresponds to a full path x(t). Given a certain point
x = x1 , . . . , xn one randomly chooses a new point x0 = x01 , . . . , x0n , often by
just changing one single coordinate. One then evaluates
P (x0 )
r= . (6.70)
P (x)
If r is larger than 1, the new point is accepted. If r < 1, on the other hand,
then a random number ρ between 0 and 1 is picked. If ρ < r, then the new
point is also accepted, otherwise it is rejected. This method is repeated until
a large enough number of points is sampled. In this way the most important
regions in x space are sampled, thus generating finally M accepted points
xm . The integral is then approximated by
M
1 X β
Pn
W [J] = e− n+1 l=0 V ((xkl )m ) . (6.71)
M m=1

The sampling algorithm just described is known after its inventor as the
Metropolis algorithm; it plays an important role in numerical evaluations of
statistical physics expressions.
In the present case of evaluating the generating functional for gs to gs
transitions one has to choose a fixed value of β, i.e. the Euclidean time.
The calculations then have to be performed for several values of β with a
subsequent extrapolation to β → ∞ (or T → 0).

6.3.2 Real time formalism


The major difficulty in evaluating a path integral numerically stems from the
oscillatory character of the integrand. The discretized form of W [J] (5.9) can
CHAPTER 6. EVALUATION OF PATH INTEGRALS 76

– in an obvious abbreviation – be written as the limit of a multidimensional


integral Z
W = dx eiS(x) , (6.72)
where x is a n-dimensional vector. Importance sampling such as the one
just discussed in the last section cannot directly be used because there is no
positive probability weight function in the integrand.
A weight function can, however, be inserted by a mathematical trick.
Using (B.18) in the form
Z q 1 T A(x−x
dx0 det(A/2π)e− 2 (x−x0 ) 0)
=1 (6.73)

we can write
Z q 1 T A(x−x
W = dxdx0 det(A/2π)e− 2 (x−x0 ) 0)
eiS(x) . (6.74)

The Gaussian factor under the integral ensures that values of x close to x0
will contribute the most to the integral. The function S(x) can, therefore be
expanded around x0
1
S(x) = S(x0 ) + S1 (x)(x − x0 ) + (x − x0 )T S2 (x0 )(x − x0 ) + . . . (6.75)
2
with
dS d2 S
S1 (x0 ) = and S2 (x0 ) = . (6.76)
dx x0 dx2 x0
After inserting this expansion we can perform the x-integration and obtain
from (B.18)
Z !1
iS(x0 ) det(A) 2
1 T −1
W = dx0 e e− 2 S1 (x0 )[A−iS2 (x0 )] S1 (x0 )
. (6.77)
det[A − iS2 (x0 )]

If we now use that (6.73) is still approximately valid even if A is a function


of x0 , we can choose

A = A(x0 ) = iS2 (x0 ) + c−1 1 (6.78)

with c > 0 and obtain


Z h i− 1 c T
W = dx0 eiS(x0 ) det(1 − iS2 (x0 )A−1 ) 2
e− 2 S1 (x0 )S1 (x0 ) . (6.79)

The function
c T
P (x0 ) = e− 2 S1 (x0 )S1 (x0 ) (6.80)
CHAPTER 6. EVALUATION OF PATH INTEGRALS 77

thus provides a probability distribution for sampling the remaining integrand


and the expression can be evaluated with the Monte-Carlo method discussed
in the last section. We obtain, therefore,
M
1 X 1
W = eiS(xi ) [det(1 + icS2 (xi )]− 2 , (6.81)
M i=1

where the M points xi are taken randomly from the distribution P (x).
For large values of c the points with S1 ≈ 0 will contribute the most
to W . This is just the stationarity condition discussed in section 6.2 which
corresponds to the classical solution. Since only a few points close to this
configuration will contribute significantly, a good sampling with good statis-
tics can be obtained. On the other hand, for small c the probability dis-
tribution becomes broad and the statistics correspondingly worse; however,
quantum mechanical effects beyond the one-loop approximation now start to
contribute. Thus, by choosing the proper value of c quantum mechanics can
be switched on.
Chapter 7

S-MATRIX AND GREEN’S


FUNCTIONS

The ultimate aim of all our fieldtheoretical developments in this book is


to calculate reaction and transition rates for processes involving elementary
particles. In this chapter, we therefore, now derive a connection between
these transition rates and expectation values of field operators, the so-called
reduction theorem.

7.1 Scattering Matrix


The typical initial state of a scattering experiment is that of widely separated
on-shell particles at t → −∞. On-shell here means that these particles
fulfill the free energy-momentum dispersion relation. The groundstate of the
system is the state of lowest-energy, i.e. the state with no particles present.
At large times t → +∞ the final state is again that of free, non-interacting
on-shell particles. The vacuum state of the theory is unique and is therefore
the same as the initial vacuum state.
The transition rate of any quantum process, be it a scattering process
m + n → m0 + n0 is determined by the so-called S-matrix. We define the S-
matrix as the probability amplitude for a process that leads from an ingoing
state |α, in i to an outgoing state |β, out i. The particles are assumed to
move freely in these asymptotic states outside the range of the interaction;
both of these states can therefore be characterized by giving the momenta of
all participating particles and possibly other quantum numbers as well all of
which are denoted by α and β. The S-matrix is thus given by

Sβα = hβ, out|α, ini . (7.1)

78
CHAPTER 7. S-MATRIX AND GREEN’S FUNCTIONS 79

We can then also introduce an operator Ŝ that transforms in-states (bras)


into out-states (bras)
hβ, out| = hβ, in|Ŝ (7.2)
so that we have
Sβα = hβ, in|Ŝ|α, ini . (7.3)
Ŝ is a unitary operator. This can be seen by taking the hermitean conjugate
of (7.2) and writing

hβ, in|Ŝ Ŝ † |α, ini = hβ, out|α, outi = δα,β . (7.4)

Thus we have
Ŝ Ŝ † = 1. (7.5)
The field operators transform in the standard way under the unitary trans-
formation S
φout = Ŝ † φ̂in Ŝ . (7.6)
In Sect. 2.3 we have expressed the matrix element Sβα in terms of wave-
functions. There we showed that S (see (2.35)) could be written as
Z
Sβα = Ψ∗β (~x, t → +∞)Ψ(+)
α (~x, t → +∞) d3x . (7.7)

Here Ψ(+) was a wavefunction that fulfilled an “in” boundary condition, i.e.
it evolved forward in time, starting from an incoming free plane wave at
t → −∞. Ψβ , on the other hand, was a free plane wave for t → +∞.
We now generalize these considerations to fields and introduce the free
asymptotic fields φin and φout

φin = lim φ(x, t)


t→−∞
φout = lim φ(x, t) . (7.8)
t→+∞

Using now the time-development operator U on φin gives for the S- matrix

Sβα = hφout |U (+∞, −∞)|φin i (7.9)

with the time-development operator (1.1). With

U (+∞, −∞) = U (+∞, t0 )U (t0 , −∞)

we can write the S-matrix also as


(−) (+)
Sβα = hφout |U (+∞, t0 )U (t0 , −∞)|φin i = hφout (t0 )|φin (t0 )i . (7.10)
CHAPTER 7. S-MATRIX AND GREEN’S FUNCTIONS 80

This form corresponds to (7.7).


As is obvious from its definition the S-matrix determines all the transition
rates possible within the field theory. For example, a cross-section for a
1 + 1 → 1 + 1 collision is simply given by the absolute square of S, multiplied
with the available phase-space of the outgoing particles and normalized to
the incoming current. The Reduction Theorem to be derived in the following
section provides a link between the S-matrix and expectation values of time-
ordered products of field operators that can be calculated as derivatives of
generating functionals.

7.2 Reduction Theorem


Since the asymptotic states appearing in the S matrix are those of free, on-
shell particles we can describe them as non-interacting quantum excitations
of the vacuum of the theory with free dispersion relations. We, therefore, first
review the basic properties of free field creation and annihilation operators
in the next subsection.

7.2.1 Canonical field quantization


The field operators are obtained by quantizing the free asymptotic fields φin
and φout . This is done in the usual way by imposing commutator relations
for the fields and their momenta. We impose the canonical commutator
relations of quantum mechanics for coordinates and corresponding canonical
momenta. Remembering that in field theory the fields play the role of the
classical coordinates we thus impose1
[Π(~x, t), φ(~x0 , t)] = − iδ 3 (~x − ~x0 )
h i
Π(~x, t), φ̇(~x0 , t) = 0
[φ(~x, t), φ(~x0 , t)] = 0 . (7.11)
We now employ the normal mode expansion of the fields (see the discus-
sion at the start of section 5.1) and momenta
1 X 1  ~ ~

φ(~x, t) = √ √ a~k (t) eik·~x + a~†k (t) e−ik·~x
V ~k 2ωk
i X ωk  ~ ~

Π(~x, t) = − √ √ a~k (t) eik·~x − a~†k (t) e−ik·~x (7.12)
V ~k 2ωk
1
From here on all the fields φ in this section are operators; only for ease of notation we
do not write the ’operator-hats’ explicitly
CHAPTER 7. S-MATRIX AND GREEN’S FUNCTIONS 81

q
with ωk = ~k 2 + m2 . Since the fields φ and the momenta Π are now opera-
tors, the ’expansion coefficients’ a and a† are operators as well.
The inverse Fourier transformation is given by
Z
1 ~
a~†k (t)= √ eik·~x (ωk φ(~x, t) + iΠ(~x, t)) d3x
2ωk V
Z
† 1 ~
a~k (t) = √ eik·~x (ωk φ(~x, t) − iΠ(~x, t)) d3x (7.13)
2ωk V
Using the commutator relations (7.11) we obtain also the commutator
relations for the operators a~k and a~†k
h i
a~k (t), a~†k0 (t) = δ~k,~k0
h i h i
a~k (t), a~k0 (t) = a~†k (t), a~†k0 (t) = 0 . (7.14)

The asymptotic in and out fields are free fields so that the time-dependence
of their operators a and a† is harmonic

a~k (t) = a~k (0)e−iωk t


a~†k (t) = a~†k (0)e+iωk t (7.15)

as can be obtained from the Heisenberg equation of motion. The asymptotic


in fields φin are then given by
1 X 1  
φin (~x, t) = √ √ a~k (0) e−ikx + a~†k (0) e+ikx ; (7.16)
V ~k 2ωk

with ikx = i(ωk t − ~k · ~x). The fields φout can be represented in the same way.
Remembering that for free fields Π = Φ̇, the free field annihilation and
creation operators for the in and out states are given in terms of the fields
and momenta as
−i Z 3 −ikx
a~†k (0) = √ d xe (∂t φ(~x, t) + iωk φ(~x, t))
2ωk V V
Z
i
a~k (0) = √ d3x e+ikx (∂t φ(~x, t) − iωk φ(~x, t)) (7.17)
2ωk V V

with kx = ωk t − ~k · ~x.
The operators (7.17) are the same as the ones used in the well known
algebraic treatment of the harmonic oscillators for the normal field modes.
The a† are the creation and the a the annihilation operators for free field
quanta and the vacuum (groundstate) of the free field theory is given by
a|0i = 0.
CHAPTER 7. S-MATRIX AND GREEN’S FUNCTIONS 82

7.2.2 Derivation of the reduction theorem


In a realistic physics situation the scattering or decay processes that we
aim to describe involve interactions between the particles. Simulating the
situation in a scattering experiment we, therefore, now assume that the in-
teractions between the particles are adiabatically being switched on and off;
adiabatically here means without energy transfer. The asymptotic “in” and
“out” states are thus free states which can be described by the free field
operators (7.17). The operators a† (t) and a(t) in (7.13) have a harmonic
time-dependence (7.15) only for times t → ±∞. The free field operators
(7.17) acting on the vacuum of the full, interacting theory create and anni-
hilate field quanta only at times t → ±∞ while they do so at all times when
acting on the vacuum state of the noninteracting theory. They can thus be
used to describe the asymptotic states.
We can thus write for the S matrix element (7.1)

Sβα = hβ, out|α, ini = hβ, out|a†in (k)|α − k, ini , (7.18)

where we have assumed that the in-state |α, ini contained a free particle with
three-momentum k ; |α−k, ini is then that in state in which just this particle
is missing. We can further write this as

hβ, out|α, ini = hβ, out|a†out (k)|α − k, ini (7.19)


+ hβ, out|a†in (k) − a†out (k)|α − k, ini .

In the first term hβ, out|a†out (k) = hβ − k, out| (= 0, if hβ, out| does not
contain a particle with momentum k).
We now rewrite the in and out creation operator into a more compact
form Z ↔

ain,out (k) = −i d3x fk (x) ∂ t φin,out (~x, t) (7.20)
with
↔ ∂φ ∂f
f (t) ∂ t φ(~x, t) = f − φ (7.21)
∂t ∂t
and
1
fk (x) = √ e−ikx .
2ωk V
With this expression we can get the S-matrix element into the form

hβ, out |α, ini = hβ − k, out|α − k, ini (7.22)


Z ↔
− ihβ, out| d3x fk (x) ∂t [φin (x) − φout (x)] |α − k, ini .
CHAPTER 7. S-MATRIX AND GREEN’S FUNCTIONS 83

For only 2 particles in the in and out states the first term on the rhs represents
a single particle transition matrixelement. In this case it can obviously only
contribute if both particles do not change their energy and momentum, i.e. if
|outi and |ini are identical. It is then just the forward scattering amplitude.
The rhs of (7.22) is time-independent. We can see this by calculating the
time derivative of its integrand
   

∂t f (x) ∂ t φin (x) = f ∂t2 φin − ∂t2 f φin . (7.23)

Here f (x) = e−ikx and φin both solve the free Klein-Gordon equation with
the same mass. We therefore have
 

∂t f (x) ∂t ~ 2 φin − (∇
φin (x) = f ∇ ~ 2 f )φin . (7.24)

The integral over this expression vanishes after twofold partial integration of
one of the terms. The same holds, of course, for the term involving φout in
(7.22).
The rhs of (7.22) is thus indeed time-independent. We can, therefore,
take it at any time and in particular also at t → ±∞. Then we can replace
the in and out fields at these times by the limits of the field φ(x). This gives
for the S-matrix element
hβ, out|α, ini = hβ − k, out|α − k, ini
Z ↔
+ lim ihβ, out| d3x fk (x) ∂ t φ(x)|α − k, ini
t→+∞
Z ↔
− lim ihβ, out| d3x fk (x) ∂ t φ(x)|α − k, ini . (7.25)
t→−∞

We now write this expression in a covariant form by using


 Z  

lim − lim d3x f (x) ∂ t φ
t→+∞ t→−∞
+∞
Z Z   Z

3
↔ h   i
= dt dx f (x) ∂ t φ = d4x f ∂t2 φ − ∂t2 f φ
∂t
−∞
Z h  i
= ~ 2 − m2 )f
d4x f ∂t2 − (∇ φ. (7.26)

Note that here φ is an interacting field, since we integrate now over all times.
Thus, in contrast to (7.23) φ does not solve the free Klein-Gordon equation
and, consequently, this integral does not vanish. Twofold partial integration
in the second term on the rhs allows us now to roll the Laplace operator from
f over to φ. This gives
Z   Z
4
dx f ∂t2 φ− (∂t2 f )φ = d4x f (x)(2 + m2 ) φ(x) . (7.27)
CHAPTER 7. S-MATRIX AND GREEN’S FUNCTIONS 84

We thus have

hβ, out|α, ini = hβ − k, out|α − k, ini (7.28)


Z
+i d4x fk (x)(2 + m2 )hβ, out|φ(x)|α − k, ini .

In this expression we have removed one particle from the in state.


We now continue by removing one particle with the momentum k 0 from
the out state by going through exactly the same steps as before. Disregarding
the first term in (7.28), that contributes only to forward scattering, we get

hβ, out|φ(x)|α − k, ini = hβ − k 0 , out|aout (k 0 )φ(x)|α − k, ini


= hβ − k 0 , out|φ(x)ain (k 0 )|α − k, ini (7.29)
0 0 0
+ hβ − k , out|aout (k )φ(x) − φ(x)ain (k )|α − k, ini .

We next replace the annihilation operators by the corresponding field oper-


ators as in (7.20) (here we have to take the hermitian conjugate operator)
and obtain

hβ, out|φ(x)|α − k, ini = hβ − k 0 , out|φ(x)|α − k − k 0 , ini (7.30)


Z  

+i d3x0 hβ − k 0 , out| fk∗0 (x0 ) ∂ t0 [φout (x0 )φ(x) − φ(x)φin (x0 )] |α − k, ini .

Taking now the limits t0 → ±∞ gives, as above, for this expression

hβ − k 0 , out|φ(x)|α − k − k 0 , ini
Z  
4 0 ∂

0 ∗ 0 0
+ ihβ − k , out| d x 0 fk0 (x ) ∂ t0 T [φ(x )φ(x)] |α − k, ini
∂t
Z 
= . . . + ihβ − k 0 , out| d4x0 fk∗0 (x0 )∂t20 T [φ(x0 )φ(x)]
  
− ∂t20 fk∗0 (x0 ) T [φ(x0 )φ(x)] |α − k, ini . (7.31)

We now use again (7.27) and obtain


Z
0
= · · · + i hβ − k , out| d4x0 fk∗0 (x0 )(20 + m2 )T [φ(x0 )φ(x)] |α − k, ini .

Combining this result with (7.28) finally gives (neglecting the forward scat-
tering amplitude)
Z
hβ, out|α, ini = i2 d4x d4x0 fk∗0 (x0 )fk (x) (7.32)
×(20 + m2 )(2 + m2 )hβ − k 0 , out|T [φ(x0 )φ(x)] |α − k, ini .
CHAPTER 7. S-MATRIX AND GREEN’S FUNCTIONS 85

This reduction can obviously be continued on both sides until we have


(with n particles with momenta k 0 in the out state and m particles with
momenta k in the in state)

Sβα = hβn k 0 , out|αm k, ini (7.33)


m
Z Y n
Z Y
= im+n d4xi d4x0j fk∗j0 (x0j )fki (xi ) ×
i=1 j=1
2 2
(20j + m )(2i + m )h0|T [φ(x01 )φ(x02 ) . . . φ(x0n )φ(x1 )φ(x2 ) . . . φ(xm )] |0i .

This is the so-called Reduction Theorem that enables us to express the S-


matrix in terms of the (n + m)-point Green’s function, somestimes also called
correlation function

G(x01 , x02 , . . . , x0n , x1 , x2 , . . . , xm ) (7.34)


= h0|T [φ(x01 )φ(x02 ) . . . φ(x0n )φ(x1 )φ(x2 ) . . . φ(xm )] |0i .

Note that in the reduction theorem (7.33) the information about the in-
teraction of the particles is contained in the (interacting) field operators φ.
Also the vacuum appearing here is that of the full, interacting theory. The
n+m-point function appearing there is, therefore, also that of the interacting
theory!
The physical process described by the reduction theorem is that of m on-
shell particles in the initial state at the asymptotic space-time coordinates
x1 , . . . , xm and n on-shell particles at the space-time coordinates x01 , . . . , x0n
with an interaction region in between these sets of coordinates. As we will see
later, the Klein-Gordon operators 2 + m2 when acting on G just remove the
propagators from the interaction region out to the asymptotic points, creating
so-called vertex functions Γ(x01 , . . . , x0n , x1 , . . . , xm ). The reduction theorem
(7.33) then gives the transition rate Sβα as the Fourier transform of this
vertex function. The Fourier transform in (7.33) contains factors of the form
exp(ik 0 x0 ) for the outgoing particles and exp(−ikx) for the incoming ones.
This could be symmetrized by changing all outgoing momenta k 0 → −k 0 .
This gives

Sβα = hβn − k 0 , out|αm k, ini


m
Z Y n
Z Y
1 1 0 0
= i m+n 4
d xi d4x0j q q e−i(kj xj +ki xi )
i=1 j=1 2ωkj0 V 2ωki V
×Γ(x01 , x02 , . . . , x0n , x1 , x2 , . . . , xm ) . (7.35)

The connection between the S-matrix and the Green’s function as ex-
pressed by the reduction theorem has been derived here only for scalar fields,
CHAPTER 7. S-MATRIX AND GREEN’S FUNCTIONS 86

but it is valid in general. The only formal difference is that the Klein-Gordon
operator 2 + m2 has to be replaced by the corresponding free-field operator.
It is also important to note that the method of canonical quantization
used here to derive the reduction theorem has been used only for the asymp-
totic states. Thus, even for fields where this method runs into difficulties
when interactions are present, like, e.g., the gauge fields to be treated in
chapter 13, the reduction theorem holds in the form given above.
In the remainder of this book we will be concerned with calculating the
correlation functions by using path integral methods. Once these correlation
functions are known the reduction theorem allows us to calculate any reaction
rate or decay probability.
Chapter 8

GREEN’S FUNCTIONS

In chapter 7 we have found that all the S-matrix elements can be calculated
once the correlation, or Green’s, functions are known. In this chapter we
discuss how these functions can be obtained as functional derivatives of the
generating functionals of the theory.

8.1 n-point Green’s Functions


In chapter 7 we have seen that the correlation function, i.e. the vacuum
expectation value of time-ordered field operators,
h i
G(x1 , x2 , . . . , xn ) = h0|T φ̂(x1 )φ̂(x2 ) . . . φ̂(xn ) |0i . (8.1)

determines the transition rate for all physical processes. Remembering that
in field theory the field operators play the role of the coordinates in classical
quantum theory we can now directly use the results obtained in Chapt. 3
and Sect. 5.1.1 and write using (3.47)
h i
G(x1 , x2 , . . . , xn ) = h0|T φ̂(x1 )φ̂(x2 ) . . . φ̂(xn ) |0i
Z
Dφ φ(x1 )φ(x2 ) . . . φ(xn )eiS[φ]
= Z (8.2)
Dφ eiS[φ]

+∞
R
with S[φ] = L(φ, ∂µ φ)d4x. The vacuum here is that of the full, interacting
−∞
Hamiltonian.
The latter expression can also be obtained as a functional derivative of
the generating functional of the theory (cf. (3.47)) so that we can also equiv-

87
CHAPTER 8. GREEN’S FUNCTIONS 88

alently define the n-point Green’s function by


 n
1 δ n Z[J]

G(x1 , x2 , . . . , xn ) = . (8.3)
i δJ(x1 )δJ(x2 ) . . . δJ(xn ) J=0

We note that G(x1 , . . . , xn ) is a symmetric function of its arguments.


Therefore, according to (B.37) the following relation holds
X 1 Z
Z[J] = dx1 . . . dxn in G(x1 , x2 , . . . , xn )J(x1 )J(x2 ) . . . J(xn ) . (8.4)
n n!

Connected Green’s Functions. Guided by (3.22) we define a functional


S[J] by the relation
Z[J] = eiS[J] (8.5)
and introduce the so-called connected Green’s functions Gc in terms of S[J]
defined by the relation
 n−1
1 δnS

Gc (x1 , . . . , xn ) = . (8.6)
i δJ(x1 ) . . . δJ(xn ) J=0

The name of this correlation function and its physics content will become
clear later in this section.

8.1.1 Momentum representation


Very often it is advantageous to work in momentum space because the exter-
nal lines of Feynman graphs represent free particles with good momentum.
In general the transformation of the Green’s function into the momentum-
representation is given by
Z
e−i(p1 x1 +p2 x2 +...+pn xn ) G(x1 , x2 , . . . , xn ) d4x1 d4x2 . . . d4xn
= (2π)4 δ 4 (p1 + p2 + . . . + pn ) G(p1 , p2 , . . . , pn ) (8.7)

The δ-function here reflects the momentum conservation due to translational


invariance. This can be seen by performing pairwise transformations of two
space-time points to their cm. point and their relative coordinate. If we
then assume that G depends only on the latter, the integral over the c.m.
coordinate can be performed and yields the δ-function. As in our discussion
around (7.35) we take all the momenta as pointing into the vertex (see Fig.
8.1).
CHAPTER 8. GREEN’S FUNCTIONS 89

p1 p4
p5
p2

p6
p3 p7

Figure 8.1: Momentum representation of the n-point function. Note that all
momenta are pointing into the shaded interaction region.

8.1.2 Operator Representations


Operator representation of the generating functional. For complete-
ness, we now derive an alternative expression for the generating functional
Z[J]. We start by defining the operator functional
R
J(x)φ̂(x) d4x
Ẑ[J] = T ei (8.8)

where φ̂ is an operator! If we form the functional derivatives of Ẑ[J] we get,


in analogy to (3.43),
 n
1 δ n Ẑ h i
= T φ̂(x1 ) . . . φ̂(xn )Ẑ[J] , (8.9)
i δJ(x1 ) . . . δJ(xn )

so that, because of Ẑ[0] = 1,


n
1 δ n h0|Ẑ[J]|0i h i
= h0|T φ̂(x 1 ) . . . φ̂(x n |0i
)
i δJ(x1 ) . . . δJ(xn ) J=0
 n
1 δnZ

= . (8.10)
i δJ(x1 ) . . . δJ(xn ) J=0

Thus all the functional derivatives of h0|Ẑ[J]|0i agree with those of Z[J] at
J = 0. According to (8.3) and (8.4) the two expressions therefore have to be
equal
Z[J] = h0|Ẑ[J]|0i . (8.11)

Functional form of the scattering operator. After having seen that the
Green’s functions can be obtained as functional derivatives of a generating
functional in this section we show that the scattering operator Ŝ can also be
CHAPTER 8. GREEN’S FUNCTIONS 90

written in a functional form as


R δ
i φ̂in (x)(2+m2 ) δJ(x) d4 x
Ŝ = : e : Z[J]|J=0 (8.12)
∞ k Z k
X i δ
= d4x1 . . . d4xk : φ̂in (x1 ) . . . φ̂in (xk ) : Z[J] |J=0 .
k=0 k! δJ(x1 ) . . . δJ(xk )
Here the : : symbol denotes the so-called normal ordered product of field
operators. This normal-ordered product is defined in such a way that all
operators in it are reordered so that all the annihilation operators are moved
to the right. This reordering takes place without a sign change for boson
fields and with a sign-change for each pairwise exchange for fermion fields.
δ
The operator (2 + m2 ) δJ(x) in (8.12) acts only on Z[J].
The matrix elements of the operator (8.12) indeed agree with (7.33). This
can be seen by considering again a matrixelement with n particles in the out
states and m particles in the in state. In the expansion of the exponential in
(8.12) only that term can contribute that contains exactly m + n powers of
the fields. We thus have
m
Z Y Z m+n
Y
m+n 4
hn, out|Ŝ|m, ini = i d xi d4xj (8.13)
i=1 j=m+1
1
× hn| : φin (x1 ) . . . φin (xm+n ) : |mi
(m + n)!
× (21 + m2 )(22 + m2 ) . . . (2m+n + m2 ) im+n G(x1 , x2 , . . . , xm+n ) .
Here the in field operator φin (x) can simply be replaced by operators of the
free field (7.8). The normal product reorders the expansion (8.13) such that
all the annihilation operators are on the right. Since each field contains two
independent sums over positive and negative energy eigenstates, respectively,
we have in total 2m+n operator products; of these only the term with n
creation operators and m annihilation operators can contribute. This gives
with the expansion (7.12)
hn| : φin (x1 ) . . . φin (xm+n ) : |mi
n n+m
(m + n)! Y Y
= hn| fk∗k (xk )a†k fkl (xl )al |mi
m! n! k=1 l=n+1
n n+m
(m + n)! Y ∗
Y
= fkk (xk ) fkl (xl ) . (8.14)
m! n! k=1 l=n+1

The degeneracy factor in front of the matrix element follows from the bi-
nomial expansion of the individual terms in the normal mode expansion
(fk ak + fk∗ a†k )m+n . The integration in (8.13) over the xi and xj just gives an
extra degeneracy factor m!n!. Taking this result together with (8.13) is the
same as (7.33).
CHAPTER 8. GREEN’S FUNCTIONS 91

8.2 Free Scalar Fields


We consider first the case of free fields. In this case the generating functional
can be given analytically (6.11)
i
R
J(x)DF (x−y)J(y) d4x d4y
Z0 [J] = e− 2 . (8.15)
The first functional derivative vanishes at J = 0 because the integral (8.15)
is Gaussian. For the second functional derivative we obtain
δ 2 Z0 [J] i
R 4 4
= −iDF (x1 − x2 ) e− 2 J(x)DF (x−y)J(y) d x d y (8.16)
δJ(x1 )J(x2 )
Z R
i
2 J(x)DF (x−y)J(y) d4x d4y
+ (−i) d4x d4y DF (x1 − x)DF (x2 − y)J(x)J(y) e− 2 ,

so that we have

δ 2 Z0 [J]
G(x1 , x2 ) = − = iDF (x1 − x2 ) . (8.17)
δJ(x1 )δJ(x2 ) J=0
The two-point function is thus just the Feynman propagator. It is there-
fore also a solution of (cf. (6.7))
h  i
2 + m2 − iε G(x1 , x2 ) = −iδ 4 (x) (8.18)

8.2.1 Wick’s theorem


The higher order derivatives can be most easily obtained by expanding (8.15)
∞  n 
X 1 iZ
Z0 [J] = −
J(x)DF (x − y)J(y) dx dy (8.19)
n=0 n! 2

i nZ
 
X 1
= 1+ − dx1 . . . dx2n D12 D34 . . . D2n−1 2n J1 J2 . . . J2n
n=1 n! 2
with the shorthand notation Dij = DF (xi − xj ), Jk = J(xk ). Noting that Z
always contains even powers of J, it is immediately evident that all n-point
functions with odd n vanish because an odd functional derivative of an even
function always vanishes at J = 0.
Taking now the 2k-th functional derivative of Z0 and using (8.3) and
(B.36) we obtain
 2k
1 δ 2k Z0
G(x1 , x2 , . . . , x2k ) = |
i δJ1 . . . δJ2k J=0
(i)k X
= k Dp1 p2 . . . Dp2k−1 p2k (8.20)
2 k! P
CHAPTER 8. GREEN’S FUNCTIONS 92

where the sum runs over all permutations (p1 , p2 , . . . , p2k ) of the numbers
(1, 2, . . . , 2k). The factor in front of the sum removes the doublecounting
because of the symmetry Dp2 p1 = Dp1 p2 (2k ) and because of the random
order of factors under the sum (k!).
Equation (8.20) states that the n-point function of a system of free bosons
can be written as a properly normalized and symmetrized product of two-
point functions. This is the so-called Wick’s theorem.
As an example we consider the case n = 4. We then have
1 X
G(x1 , x2 , x3 , x4 ) = − Dp p Dp p . (8.21)
8 P ∈S4 1 2 3 4
Among the 24 terms in the sum, 12 are pairwise equal because the product
of the two propagators commutes. Furthermore, the propagator Dp1 p2 and
Dp2 p1 are pairwise equal. Thus, there are only 24 : 2 : 2 : 2 = 3 essentially
distinct terms in the sum; the factor 1/8 just takes care of all the others. We
thus have
G(x1 , x2 , x3 , x4 ) = − DF (x1 − x2 )DF (x3 − x4 ) (8.22)
− DF (x1 − x3 )DF (x2 − x4 )
− DF (x1 − x4 )DF (x2 − x3 ) .
The first few n-point Green’s functions are therefore – according to (8.17),
(8.20) and (8.22) – given by
G(x1 ) = 0 (8.23)
G(x1 , x2 ) = h0|T [φ(x1 )φ(x2 )] |0i = iDF (x1 − x2 )
G(x1 , x2 , x3 ) = 0
G(x1 , x2 , x3 , x4 ) = h0|T [φ(x1 )φ(x2 )φ(x3 )φ(x4 )] |0i
= − DF (x1 − x2 )DF (x3 − x4 )
− DF (x1 − x3 )DF (x2 − x4 )
− DF (x1 − x4 )DF (x2 − x3 )
etc. (8.24)
Here |0i is the vacuum state of the free Hamiltonian because it was obtained
as a functional derivative of the non-interacting functional Z0 (8.19). All
n-point functions with odd n vanish.

8.2.2 Feynman rules


As in the classical case (cf. Sect. 2.3) we can again represent these results in
a graphical way. The Feynman rules, that establish the connection between
CHAPTER 8. GREEN’S FUNCTIONS 93

the algebraic and the graphical representation, are for the case of free fields
still rather trivial.
They are given by
1) each Feynman propagator is represented by a line:
= iDF (x − y) .
x y
2) each source is represented by a cross: = iJ(x) .
x
3) There is an integration over all space-time coordinates of the currents

4) Each diagram has a factor that takes its symmetry into account.
For example, if there is an integration over the endpoints x and y, these
could be exchanged without changing the result; the symmetry factor
is, correspondingly, 1/2.
With rule 1) we get, for example, for the fourpoint function (8.22), i.e.
the two-particle Green’s function
1 2 1 2
1 2
G(x1 , x2 , x3 , x4 ) = + + (8.25)
3 4
3 4 3 4
Each line connecting the two points x and y denotes the free propagator and
gives a factor iDF (x − y).
We now set
Z0 [J] = eiS0 [J] (8.26)
with
iZ 4 4
iS0 [J] = − d x d y J(x)DF (x − y)J(y) (8.27)
2
By using rules 2), 3) and 4) we find immediately

iS0 = (8.28)

Now we expand

Z0 [J] = eiS0 [J] = e


 
i Z
= 1+ − J(x)DF (x − y)J(y) d4x d4y
2
 2
1 iZ
+ − J(x)DF (x − y)J(y) d4x d4y + · · ·
2! 2
CHAPTER 8. GREEN’S FUNCTIONS 94

1 1
= 1 + iS0 [J] + (iS0 [J])2 + (iS0 [J])3 + . . .
2! 3!
1 1
= 1+ + + + ... . (8.29)
2! 3!

We now call all graphs that hang together connected graphs and all the
others unconnected graphs. In our simple case here S0 [J] is represented
by only one connected graph (8.28), whereas Z0 [J] – through the power
expansion (8.29) of the exponential function – generates unconnected graphs
as well.
In the simple case of a free field discussed here there is only one connected
graph (see (8.25)). The connected Green’s function can be obtained from its
definition (8.6) as
1 δ 2 S0
Gc (x1 , x2 ) = | = iDF (x1 − x2 ) . (8.30)
i δJ(x1 )δJ(x2 ) J=0
All higher functional derivatives of S0 [J] vanish. Gc is in this free case thus
just given by the Feynman propagator.

8.3 Interacting Scalar Fields


In this section we consider Lagrangians of the form
L = L0 − V (φ) , (8.31)
where L0 is the free scalar Lagrangian (5.1) and V represents a selfinteraction
of the field. For such a Lagrangian the generating functional for the n-point
functions can no longer be given in closed form. In order to obtain the n-point
function one has to resort to perturbative methods.
The n-point function then follows from its definition (8.2)
Z
Dφ φ(x1 )φ(x2 ) . . . φ(xn ) eiS[φ]
G(x1 , x2 , . . . , xn ) = Z (8.32)
iS[φ]
Dφ e

with the exponential in the generating functional of the form


R
d4x (L0 −V +i 2ε φ2 )
eiS[φ] = ei . (8.33)
The action exponential can be Taylor-expanded in the interaction strength
∞  Z N
X 1
eiS[φ] = −i d4x V eiS0 [φ] (8.34)
N =0 N !
CHAPTER 8. GREEN’S FUNCTIONS 95

with the free action


Z  
ε
S0 [φ] = d x L0 + i φ2
4
. (8.35)
2
Inserting this into (8.32) gives the n-point function of the interacting theory
in terms of powers of the interaction and the free-field action
Z
Dφ φ(x1 )φ(x2 ) . . . φ(xn ) eiS[φ]
G(x1 , x2 , . . . , xn ) = Z (8.36)
iS[φ]
Dφ e
Z ∞  Z N
X 1 4
Dφ φ(x1 )φ(x2 ) . . . φ(xn ) −i d xV eiS0 [φ]
N =0 N !
= Z ∞  Z N
X 1
Dφ −i d4x V eiS0 [φ]
N =0 N !

By using (8.2) and the developments in Sects. 3.2 and 5.1.1 we can rewrite
this equation also in terms of normalized vacuum expectation values. The last
line of (8.36) involves the free action S0 and thus free propagation. Therefore,
if the field operators are those of free in fields at asymptotic times they remain
so even during propagation. Also the vacuum appearing in the quantum
mechanical vacuum expectation value is then that of the non-interacting
theory so that we get
h i
G(x1 , x2 , . . . , xn ) = h0̃|T φ̂(x1 )φ̂(x2 ) . . . φ̂(xn ) |0̃i
" ∞  Z N #
X 1 4
h0|T φ̂in (x1 ) . . . φ̂in (xn ) −i V̂ d x |0i
N =0 N!
= " ∞  Z N # . (8.37)
X 1 4
h0|T −i V̂ d x |0i
N =0 N!

As in Sect. 5.1.1 |0̃i denotes here the vacuum state of the full, interacting
Hamiltonian, whereas |0i is that of the free Hamiltonian.
With (8.37) we have achieved a remarkable result: (8.37) expresses the
expectation value of the time-ordered product of field operators in the vac-
uum state of the full, interacting theory by a perturbative expansion over free
field expectation values. The latter can be calculated as path integrals over
products of classical fields and powers of the interaction (8.36). This enables
us to calculate G, and ultimately also the scattering matrix S (Chapt. 7),
perturbatively up to any desired order in the interaction.
CHAPTER 8. GREEN’S FUNCTIONS 96

8.3.1 Perturbative expansion


Eq. (8.32) allows us to calculate the perturbative expansion of the full Green’s
function up to any desired order in V . Alternatively, these higher order terms
can also be obtained as functional derivatives of the free generating functional
Z0 [J] which is known.
According to our general considerations the functional for the Lagrangian
(8.31) is given by
Z R
d4x (L0 −V (φ)+Jφ+i 2ε φ2 )
Z[J] = Z0 Dφ ei (8.38)
Z R R
d4x V (φ) i d4x (L0 +Jφ+i 2ε φ2 )
= Z0 Dφ e−i e .

Here Z0 is just the inverse of the path integral for J = 0


Z R
d4x (L0 −V (φ)+i 2ε φ2 )
Z0−1 = Dφ ei . (8.39)

We now use the relation


1 δ R 4 R 4
ei d x (L0 +Jφ+i 2 φ ) = φ(y) ei d x (L0 +Jφ+i 2 φ ) .
ε 2 ε 2
(8.40)
i δJ(y)
This relation, read from right to left, will also be true for any function V (φ),
as can be seen by expanding V into a series in powers of φ. We thus have
also
" #
R
d4x (L0 +Jφ+i 2ε φ2 ) 1 δ R 4
ei d x (L0 +Jφ+i 2 φ )
ε 2
i
V [φ(y)] e =V (8.41)
i δJ(y)
and consequently, after exponentiation, also
R R
d4y V [φ(y)] i d4x (L0 +Jφ+i 2ε φ2 )
e−i e (8.42)
R R
−i d4y V ( 1i δ
) ei d4x ( L0 +Jφ+i 2ε φ2 ).
= e δJ(y)

This relation allows us to take the V -dependent factor in (8.38) out of the
path-integral
R Z R
−i d4y V ( 1i δJ(y)
δ
) i d4x (L0 +Jφ+i 2ε φ2 )
Z[J] = Z0 e Dφ e . (8.43)

The last factor in (8.43) has been expressed in terms of the free two-particle
propagator introduced in the last section. We thus have
R R
d4z V ( 1i δJ(z)
δ
) − 2i 4 4
Z[J] = Z0 e−i e J(x)DF (x−y)J(y) d x d y
R 4
= Z0 e−i d x V ( i δJ(x) ) eiS0 [J]
1 δ

R 4
= Z0 e−i d x V ( i δJ(x) ) Z0 [J] .
1 δ
(8.44)
CHAPTER 8. GREEN’S FUNCTIONS 97

Expanding the exponential that contains the interaction V then gives the
perturbative expansion for Z[J]
∞ Z !!N
X 1 4 1 δ
Z[J] = Z0 −i d xV Z0 [J] . (8.45)
N =0 N! i δJ(x)
Since we will later on be mostly interested in connected graphs we do
not need Z[J] directly but instead its logarithm. We, therefore, now expand
the functional iS[J] = ln Z[J] in powers of the interaction V . We start by
inserting a factor 1 = exp (+iS0 ) exp (−iS0 ) between Z0 and the exponential
in (8.44) and taking the logarithm
 R 
d4x V ( 1i δ
) eiS0 [J]
ln Z[J] = ln Z0 + ln 1 · e−i δJ

 R 
−iS0 −i d4x V iS0
= ln Z0 + iS0 + ln e e e
  R  
−iS0 [J] −i d4x V ( 1i δJ
δ
) iS0 [J]
= ln Z0 + iS0 [J] + ln 1 + e e −1 e
= iS[J] . (8.46)
A perturbation theoretical treatment is now based on a Taylor expansion of
the logarithm. For that purpose we abbreviate
 R 
d4x V
ε[J] = e−iS0 [J] e−i − 1 eiS0 [J] (8.47)
and obtain
iS[J] = ln Z[J] = ln Z0 + iS0 [J] + ln(1 + ε[J]) (8.48)
 
1 2 3
= ln Z0 + iS0 [J] + ε[J] − ε [J] + O(ε ) .
2
Equation (8.48) represents an expansion of S in powers of the (for V → 0)
small quantity ε. In order to obtain a perturbative expansion in terms of the
potential V we now rearrange the expansion (8.48). First we expand ε[J] of
(8.47) in terms of the strength of the interaction. This gives
( Z !
−iS0 [J] 4 1 δ
ε[J] = e −i d xV (8.49)
i δJ
" !#2 
Z
1 1 δ 
+ −i d4x V + · · · e+iS0 [J] .
2! i δJ

We now insert this expression into (8.48) and obtain


 
1
iS[J] = ln Z0 + iS0 [J] + ε[J] − ε2 [J] + . . .
2
CHAPTER 8. GREEN’S FUNCTIONS 98

" Z !#
−iS0 [J] 4 1 δ
= ln Z0 + iS0 [J] + e −i d xV eiS0 [J]
i δJ
" Z !#2
1 1 δ
+ e−iS0 [J] −i d4x V eiS0 [J]
2! i δJ
( Z" !# )2
1 −iS0 [J] 1 δ
− e −i d4x V e iS0 [J]
+ O(V 3 )
2 i δJ
1
= ln Z0 + iS0 [J] + iS1 [J] + iS2 [J] − (iS1 [J])2 + O(V 3 ) ,
2
(8.50)

with
iZ
iS0 [J] = − J(x)DF (x − y)J(y) d4x d4y
2 !
Z
−iS0 [J] 4 1 δ
iS1 [J] = e (−i) d x V e+iS0 [J]
i δJ
Z" !#2
1 −iS0 [J] 1 δ
iS2 [J] = e −i d4x V e+iS0 [J] . (8.51)
2! i δJ

Equation (8.50) represents a perturbative expansion of S in powers of V .


Note that the term of second order in the interaction V receives contributions
both from the linear and the quadratic term in the original expansion (8.48)
in ε. The term ∼ S1 obviously just contains an iteration of the linear term.
Chapter 9

PERTURBATIVE φ4 THEORY

In this chapter we apply the formalism developed in the preceding chapter


to the so-called φ4 theory whose Lagrangian is given by
g 4
L = L0 − V (φ) = L0 − φ . (9.1)
4!
Here g is a coupling constant. This φ4 theory is a prototype of a field theory
with selfinteractions. It serves as a didactical example which exhibits all
phenomena of more complex field theories.

9.1 Perturbative Expansion of the Generat-


ing Function
We start with the generating functional for connected Green’s functions
(8.50)
1
iS[J] = ln Z0 + iS0 [J] + iS1 [J] + iS2 [J] − (iS1 [J])2 + O(V 3 ) . (9.2)
2
Inserting the interaction of φ4 theory (9.1) into (8.51) we obtain
iZ 4 4
iS0 [J] = − d z d y J(z)DF (z − y)J(y) ,
2
ig −iS0 [J] Z 4 δ4
iS1 [J] = − e d x 4 eiS0 [J] (9.3)
4! δJ (x)
and
1 −ig 2 −iS0 [J] Z 4 4 δ4 δ4
 
iS2 [J] = e d xd y 4 eiS0 [J] .
2! 4! δJ (x) δJ 4 (y)

99
CHAPTER 9. PERTURBATIVE φ4 THEORY 100

For notational convenience in the following we now introduce the Si defined


by
ig
iS1 [J] = − S̃1 [J]
4!
−ig 2
 
iS2 [J] = S̃2 [J] . (9.4)
4!
The generating function for the connected Green‘s functions in φ4 theory
reads then
 2  2
−ig −ig 1 −ig
iS[J] = ln Z0 + iS0 [J] + S̃1 [J] + S̃2 [J] − S̃1 [J] + ... .
4! 4! 2 4!
(9.5)
Since each of the Si [J] contains functional derivatives of the known S0 [J] we
can now evaluate the functional derivatives of S[J] and obtain all the Green’s
functions.
First, we calculate the terms linear in g
−iS0 [J] δ4Z
S̃1 [J] = e 4
d4x
e+iS0 [J] (9.6)
δJ (x)
Z
δ4 i
R 4 4
= e−iS0 [J] d4x 4 e− 2 J(z)DF (z−y)J(y) d z d y .
δJ (x)
The fourth functional derivative is most easily obtained by going to a discrete
representation. Noting that
∂ 4 − i Ji Dij Jj
e 2 = [−3Dkk Dkk + 6iDkk (DJ)k (DJ)k (9.7)
∂Jk4
i
+ (DJ)k (DJ)k (DJ)k (DJ)k ] e− 2 Ji Dij Jj
we obtain (k =x)
ˆ
Z
S̃1 [J] = −3 DF (x − x)DF (x − x) d4x
Z
+ 6i DF (y − x)DF (x − x)DF (x − z)J(y)J(z) d4x d4y d4z
Z
+ [DF (x − y)DF (x − z)DF (x − v)DF (x − w)
i
× J(y)J(z)J(v)J(w) d4x d4y d4v d4w d4z . (9.8)
The first term has no sources and will, therefore, not contribute to any
Green’s function, the second term is quadratic in J and thus contributes
to the two-point function and the last term here has the structure of a point
interaction of four fields generated by independent sources at y, z, v, and w
and thus contributes only to the four-point function.
CHAPTER 9. PERTURBATIVE φ4 THEORY 101

9.1.1 Feynman rules


We can again represent these results in a graphical form by using the rules
given in section 8.2.2. These were
1) propagator: iDF (x − y) =
x y
2) source: iJ(x) =
x
3) Integration over the space-time coordinates of the sources

4) Symmetry factor for each diagram


We supplement these now for the interacting theory by the additional rules

−ig
5) Each interaction is represented by a dot: =
4!

R
6) Integration d4x for each loop.
If we represent the interacting connected functional iS[J] by a double
line
iS[J] = (9.9)
we can draw the graphs for
−ig
iS[J] = ln Z0 + iS0 [J] + S̃1 [J] + O(g 2 ) (9.10)
4!
as

= ln Z0 + (9.11)
 
w z
x 2
+
 x + + 
 + O(g )
y x z y v
The first graph on the right-hand side is again the zeroth order term (9.3),
whereas the graphs in the parentheses represent all the terms of first order in
the interaction (9.8). The first one of these describes a process without any
external lines; this is a vacuum process that takes place regardless if physical
particles are present or not. It constitutes a background to all physical
processes. The second graph with the single loop describes a mass change
due to the selfinteraction that we will discuss in the next section. The third
graph, finally, describes a true interaction process.
CHAPTER 9. PERTURBATIVE φ4 THEORY 102

The symmetry factors in these graphs are the factors in front of the
integrals in (9.8); they can be obtained as follows. The first graph carries
the factor 1/2 as explained in section 8.2.2.
To construct the vacuum graph we pick one of the 4 legs of the interaction
vertex and then connect with any of the other three free legs; there are always
2 pairwise equal loops. Thus in total we get a weight of 4 × 3/(2 × 2) = 3.
For the second term in parentheses in (9.11) we have four legs of the
vertex to connect with the external line to y; this gives 4 possibilities. The
external line to z can then still be connected with 3 remaining vertex legs.
Since there is an exchange symmetry between y and z we get an additional
factor 1/2 (as in S0 ), so that the weight of this vertex becomes 6.
The last graph, finally, is obtained by joining one of the four legs of
the vertex to one of the external points, say z. This generates 4 possibilities.
Next we join any one of the 3 remaining free legs of the vertex to the external
point y; there are obviously 3 ways to do this. The remaining 2 legs can be
joined in 2 different ways with the two external points v and w. Thus, there
are in total 4! = 24 possibilities. However, since all the external points v,w,y
and z can be exchanged without changing any of the physics (v,w,y and z
are integration variables), there are also 4! identical terms so that the last
graph in the parentheses in Fig. 9.11 carries the weight 1.
These weights (symmetry factors) have to be multiplied for each graph
to the analytical expression obtained by following the rules given above for
the translation of the pictorial representation into an analytical one. Indeed,
using the symmetry factors just given and following the Feynman rules for
the graphs (9.11) gives the expression (8.50) (together with (9.4) and (9.8)).

9.1.2 Vacuum contributions


We now consider the normalization term ln Z0 . Since we are working with
normalized generating functionals, Z0 is given by W [0]−1
Z R
i d4x(L0 −V +i 2ε φ2 )
Z0−1 = Dφ e (9.12)
R R
d4xV ( 1i δJ(x)
δ
) − 2i J(x)DF (x−y)J(y)d4x d4y |
= e−i e J=0 .

We can now treat this expression in exactly the same way as we just did for
S[J]; the only change being that we have to take the final result at J = 0.
This gives (see (9.10))

(−ig)
ln Z0 = −iS[0] = −iS0 [0] − S̃1 [0] + O(g 2 ) . (9.13)
4!
CHAPTER 9. PERTURBATIVE φ4 THEORY 103

In the graphical
R
representation this reads, using S0 [0] = 0 and S1 [0] =
−3DF2 (0) d4x (cf. (9.8)),

ln N = − (9.14)

The normalization constant, or – in other words – the denominator of the


generating functional, thus contains just the vacuum graph. Inserting (9.14)
into (9.11) then removes the vacuum contribution giving, finally, for Z[J] the
graphical representation up to terms of first order in the interaction

= + + (9.15)

Although we have shown here only for first-order coupling that the de-
nominator in Z[J] (see (5.15),(5.16)) just removes the vacuum contributions,
this is a general result that holds to all orders of perturbation theory.

9.2 Two-Point Function


The connected n-point function can now be obtained from its definition in
(8.6)
 n−1
1 δnS

Gc (x1 , . . . , xn ) = . (9.16)
i δJ(x1 ) . . . δJ(xn ) J=0
In this section we work out the connected two-point function in the lowest
orders of the coupling constant.

9.2.1 Terms up to O(g 0 )


There is only one connected Green’s function in the free case which is just
given by the Feynman propagator (8.30).

9.2.1.1 Momentum representation


We now evaluate the lowest order (in g) two-point function in momentum
space (cf. (8.7)). Taking the Fourier-transform of the two-point function
(8.30) gives
Z
e−i(p1 x1 +p2 x2 ) Gc (x1 , x2 ) d4x1 d4x2 (9.17)
Z " !Z #
i e−iq(x1 −x2 )
= e−i(p1 x1 +p2 x2 ) d4q d4x1 d4x2 .
(2π)4 2 2
q − m + iε
CHAPTER 9. PERTURBATIVE φ4 THEORY 104

p −p

Figure 9.1: Two-point function of a scalar theory.

We first perform the integrations over x1 and x2 and obtain, according to the
definition (8.1.1), for the rhs
i
(2π)4 δ 4 (p1 + p2 )Gc (p1 , p2 ) = (2π)4 δ 4 (p1 + p2 ) . (9.18)
p21 − m2 + iε
From this equation we can read off the momentum representation of the prop-
agator. The momenta p1 and p2 are incoming momenta that point towards a
vertex. Thus, the momentum representation of the free propagator is given
by
i
G0 (p, p0 = −p) = 2 , (9.19)
p − m2 + iε
pictured in Fig. 9.1. Note that here the second momentum appears with a
negative sign. This is due to our notation to take all momenta as incoming
(see Fig. 8.1).

9.2.2 Terms up to O(g)


Up to terms linear in the coupling strength we obtain from (9.5)

δ2 S

Gc (x1 , x2 ) = −i (9.20)
δJ(x1 )δJ(x2 ) J=0

δ 2 S0 (−ig) δ 2 S1
= −i −
+ O(g 2 ) .
δJ(x1 )δJ(x2 ) 4! δJ(x1 )δJ(x2 ) J=0

With S0 [J] from (9.3) and S1 [J] from (9.8) this gives for the two-point func-
tion

Gc (x1 , x2 ) = iDF (x1 − x2 )


  Z
ig
− − 12i d4xDF (x − x)DF (x − x1 )DF (x − x2 ) + O(g 2 )
4!
gZ 4
= iDF (x1 − x2 ) − d xDF (x2 − x)DF (x − x)DF (x − x1 )
2
+ O(g 2 ) . (9.21)
CHAPTER 9. PERTURBATIVE φ4 THEORY 105

This is the connected propagator, up to terms of O(g), of the interacting


theory.
We can represent this equation in the following graphical form, where
denotes the “dressed” propagator

= + (9.22)
x1 x2 x1 x2
x1 x2
with the rules developed above. Since the n-point functions involve deriva-
tives with respect to the source current, taken at zero source, the external
lines of all Feynman graphs do not contain crosses, that depict sources, any-
more. They are instead given by free propagators.
The weight factors for these diagrams have been explained at the end
of section 8.2.2. Since we deal here with Green’s functions with definite,
fixed external points, the exchange symmetry factors must not be divided
out here. Thus, the first diagram on the rhs in (9.22) carries the weight 1
and the second, the so-called tadpole diagram, the weight 12.

9.2.2.1 Momentum representation


In Sect. 8.1.1 we have introduced the momentum representation of the Green’s
function and in Sect. 9.2.1.1 we have already evaluated it for the free case.
Here we now determine it for the φ4 theory up to terms of order O(g).
Taking the Fourier-transform of the two-point function (9.21) gives
Z
e−i(p1 x1 +p2 x2 ) Gc (x1 , x2 ) d4x1 d4x2 (9.23)
Z " !Z #
i e−iq(x1 −x2 )
= e−i(p1 x1 +p2 x2 ) d4q d4x1 d4x2
(2π)4 2 2
q − m + iε
 !3
g Z  −i(p1 x1 +p2 x2 ) Z 4 1
− e dx
2 (2π)4
#
Z
4 4 4 e−iq2 (x2 −x) e−iq1 (x−x1 )
× d q1 d q2 d q3 2 d4x1 d4x2 .
(q1 − m2 + iε)(q22 + m2 − iε)(q32 − m2 + iε)
As in Sect. 9.2.1.1 we first perform the integrations over x1 , x2 and x and
obtain for the rhs
i
(2π)4 δ 4 (p1 + p2 )Gc (p1 , p2 ) = (2π)4 δ 4 (p1 + p2 ) 2
p1 − m2 + iε
g 1 1
− (2π)4 δ 4 (p1 + p2 ) 2 2
2 p1 − m + iε p2 − m2 + iε
2
Z
d4q 1
× . (9.24)
(2π) q − m2 + iε
4 2
CHAPTER 9. PERTURBATIVE φ4 THEORY 106

This equation is used to read off the momentum representation of the propa-
gator (see (8.7)). The momenta p1 and p2 are incoming momenta that point
towards a vertex. Thus, the momentum representation of (9.21) is given by
i
Gc (p, p0 = −p) = (9.25)
p2
− m2 + iε
!
i −ig Z d4q i i
+S 2 2
p − m + iε 4! (2π) q − m + iε p − m2 + iε
4 2 2 2

where the symmetry factor is S = 12. Equation (9.25) gives the momentum
representation of the two-point function up to terms of order g. The first
term on the rhs gives the free propagator (9.19) already obtained in Sect.
9.2.1.1, whereas the second term gives the contribution of the interaction to
this two-point function.

Momentum space Feynman rules. The Feynman rules for (9.25) are
now
i
1) each line gives a factor 2 .
q − m2 + iε
−ig
2) each vertex gives a factor .
4!
3) there is four-momentum conservation for the sum of all momenta flow-
ing into a vertex.
Z
d4q
4) each internal line gives an integration .
(2π)4
5) to each diagram a weight factor has to be multiplied as explained above.

Selfenergy. With the abbreviation


g Z d4 q i
Σ= (9.26)
2 (2π) q − m2 + iε
4 2

and the free two-point function G0 from (9.19 we can write the two-point
function (9.25) as
   −1
Σ Σ ΣG0
Gc (p, −p) = G0 + G0 G0 = G0 1 + G0 ≈ G0 1 −
i i i
i 1
= 2
p − m + iε 1 − Σ p2 −m1 2 +iε
2

i
= 2 2
. (9.27)
p − m − Σ + iε
CHAPTER 9. PERTURBATIVE φ4 THEORY 107

a) b) c)

Figure 9.2: Feynman graphs for the two-point function up to O(g 2 ).

Here we have consistently kept terms up to O(g).


The quantity Σ appears like an additional mass term in the final result.
It is, therefore, called a selfenergy and the second graph on the rhs in (9.22)
is called a selfenergy insertion. The appearance of this selfenergy is a first
indication that the mass m appearing in the Lagrangian is the mass of the
particle only in a classical theory. In quantum theory it gets changed by the
interactions.

9.2.3 Terms up to O(g 2 )


As noted at the end of Sect. 8.3.1 there are two distinct contributions to the
second order term, one being a genuine term of second order in V and the
other just being an iteration of the first order term. With the help of the
Feynman rules we can now construct the corresponding Feynman graphs up
to terms of O(g 2 ). For the two-point function these are given in Fig. 9.2.
Graph (a) in Fig. 9.2 obviously just represents an iteration of the first
order tadpole graph in (9.22). Its contribution to the two-point function is
given by
!
i −ig Z d4q i
Ga (p, −p) = Sa 2 2
(9.28)
p − m + iε 4! (2π) q − m2 + iε
4 2
!
i −ig Z d4q i i
× 2 2
,
p − m + iε 4! (2π) q − m + iε p − m2 + iε
4 2 2 2

Sa is the symmetry factor; it is simply given by the product of the corre-


sponding factors for the one-loop graphs, Sa = 12 · 12 = 144.
It is evident from Fig. 9.2a, as well as from its algebraic representation
in (9.28), that the graph can be cut into two parts, each representing a first
CHAPTER 9. PERTURBATIVE φ4 THEORY 108

order process. This reflects the appearance of the last term in (8.50) that
is simply the square of the first order term. Such a graph that falls apart
into 2 unconnected parts, if one internal line is cut, is called one-particle-
reducible; otherwise it is one-particle-irreducible (1PI). The reducible graph
here is generated by the square of the first order term ∼ S̃12 in (9.5).
In order to facilitate the following discussions we introduce now the ver-
tex function Γ(p1 , p2 , . . . , pn ), sometimes also called connected proper vertex
function, which describes only 1PI graphs and in which the propagators for
the external lines are missing. The n-point vertex function is, therefore, given
by

Γ(p1 , p2 , . . . , pn ) (9.29)
−1 −1 −1
= G (p1 , −p1 )G (p2 , −p2 ) . . . G (pn , −pn )Gc (p1 , p2 , . . . , pn ) .

The free 1PI 2-point function is defined by

Γ(p, −p) = p2 − m2 . (9.30)

Note that the product of inverse two-point Green’s functions and the n-body
function is just the combination that appears in the reduction theorem (cf.
(7.35)).
The 1PI part of the Green’s function for the graph 9.2a reads

Γa (p, −p) = G−1 (p, −p)G−1 (−p, p)Ga (p, −p)


−ig Z d4q i
= Sa (9.31)
4! (2π) q − m2 + iε
4 2

with Sa = 144.
We then get for the graph in Fig. 9.2b, which is one-particle irreducible,
−ig 2 Z d4q d4u i(2π)4 δ 4 (q − u)
 
i
Γb (p, −p) = Sb
4! (2π)4 (2π)4 q 2 − m2 + iε u2 − m2 + iε
Z 4
dr i
× (9.32)
(2π) r − m2 + iε
4 2

with Sb = 12 · 12 = 144. For the graph in Fig. 9.2c (also 1PI) we finally
obtain
2 Z
d4q d4r d4s

−ig
Γc (p, −p) = Sc δ(p − (q + r + s))
(2π)4 (2π)4 (2π)4
4!
i i i
× 2 , (9.33)
q − m + iε r − m + iε s − m2 + iε
2 2 2 2

with Sc = 4 · 4! = 96.
CHAPTER 9. PERTURBATIVE φ4 THEORY 109

3 4

1 2

Figure 9.3: Feynman graph for the four-point function.

9.3 Four-Point Function


It is easy to see that the three-point function vanishes for the model con-
sidered here since the third functional derivative (see (9.16)) of the action
(8.50) at J = 0 vanishes.

9.3.1 Terms up to O(g)


The four-point function up to terms of O(g) is given by
 3
1 δ4 S

Gc (x1 , x2 , x3 , x4 ) = (9.34)
i δJ(x1 ) . . . δJ(x4 ) J=0

δ 4 S0
(−ig) δ 4 S1

= i +
δJ(x1 ) . . . δJ(x4 ) J=0 4! δJ(x1 ) . . . δJ(x4 ) J=0
+ O(g 2 )
where S is given by (9.5) and Fig. 9.2. Because Gc involves the fourth
derivative with respect to the source and S0 depends on J only quadratically
(see (9.3)), only the last term of S1 in (9.8) can contribute to the Green’s
function. Thus we get
Z
Gc (x1 , . . . , x4 ) = −ig d4xDF (x − x1 )DF (x − x2 )DF (x − x3 )DF (x − x4 ) .
(9.35)
In momentum space this is simply the product of the four propagators (9.19)
times the factor −ig. The corresponding Feynman graph is given in Fig.
9.3. It carries the symmetry factor 4!, corresponding to a symmetry under
permutation of all external legs.

Unconnected graphs. From the connected graphs calculated so far, we


could reconstruct also the unconnected graphs. Up to terms linear in the
CHAPTER 9. PERTURBATIVE φ4 THEORY 110

coupling constant g we get in symbolic notation


1
Z[J] = e = 1+ + ( )2 + . . .
2! !
= 1+ + +
!2
1
+ + +
2!
+O(g 2 ) . (9.36)
The four-point function is generated by diagrams with four external legs
(each external leg corresponds to a factor J in the generating functional),
because G is given by a fourth functional derivative. Therefore, only the last
diagram in the second line of (9.36) and the square of the first term in the
last line can contribute to the four-point function in this order.
We thus have for the four-point function up to terms linear in the cou-
pling

G(x1 , x2 , x3 , x4 ) = + + (9.37)

In both of the first two diagrams the two particles just move by each other,
without interaction. These unconnected graphs thus do not contribute to
any interaction processes.

9.3.2 Terms up to O(g 2 )


.
In order to become more familiar with Feynman graphs, we now construct
the connected four-point function up to terms of order g 2 in a graphical way.
This four-point function is shown in Fig. 9.4.
The first line in Fig. 9.4 gives the four-point function just constructed,
with a symmetry factor S1 = 4!. The four diagrams on the second line
are just the basic vertices with self-energy insertions on each of the external
legs; these insertions carry the extra symmetry factor S2 = 12, as we have
seen in section 9.1.1. The last three diagrams in the third line are of a new
topological structure. They represent modifications of the basic interaction
vertex through the insertion of internal lines.
Each one of these graphs has the same external lines. We thus have an
overall factor for all graphs
4
Y i
S1 2
, (9.38)
k=1 pk − m2
CHAPTER 9. PERTURBATIVE φ4 THEORY 111

3 4
3 4

1 2
1 2
3 4 3 4 3 4 3 4

+ + + +

1 2 1 2 1 2 1 2
3 4
3 4
3 4
q1
+ + + + O(g 3 )
q2
1 2
1 2
1 2

Figure 9.4: Four-point function of φ4 theory up to terms ∼ g 2 .

with the external symmetry factor S1 = 4! so that the full four-point function
is given by
4
Y i
Gc (p1 , p2 , p3 , p4 ) = S1 2
(G1 + G2 + G3 ) , (9.39)
k=1 pk − m2

where Gi denotes the contribution from the i-th line in Fig. 9.4 without the
external line symmetry factor.
The first basic vertex then just gives the contribution
−ig
G1 = . (9.40)
4!
The graphs of the second line have one loop in addition. We thus have
4
−ig 2 Z d4q i X i
G2 = ( ) S2 2
(9.41)
4! (2π)4 q 2 − m2 l=1 pl − m2
CHAPTER 9. PERTURBATIVE φ4 THEORY 112

for their contribution, with S2 = 12. The internal momenta pl here are
those between the loop and the four-point vertex. Since the loop carries no
momentum away they are the same as the corresponding incoming momenta.
The three graphs of the third line, finally, have two internal lines, which
are, however, related by energy - and momentum conservation at the incom-
ing vertex. They give

−ig 2 Z d4q1 d4q2 i i


G3 = ( ) S3 2 2
4! (2π) (2π) q1 − m q2 − m2
4 4 2

X
× (2π)4 δ 4 (q1 + q2 − (pk + pl )) (9.42)
kl

where the last sum runs only over the pairs of numbers (1,2), (1,3) and (1,4),
i.e. the external legs at one of the vertices in each of the three graphs. The δ-
function appears because the net momentum running into the dressed vertex
has to be zero (see (8.7)).
The symmetry factor S3 can, for example for the middle graph of Fig.
9.4, be obtained as follows. The external leg 1 can be connected with the left
vertex in 4 different ways; the same holds for the leg 2 with the right vertex.
Once these connections (4 × 4 possibilities) have been done, each of the 3
free legs of the left vertex can be connected with the external leg 3; the same
holds for the connections of the right vertex to the external point 4 (3 × 3
possibilities). Finally, each of the remaining 2 legs of the left vertex can be
connected to the right vertex; for the remaining leg there is then no freedom
left (2 possibilities). Thus, the symmetry factor for the last 3 graphs in Fig.
9.4 is
(4 · 4)(3 · 3)2 (4!)2 4!
S3 = = = . (9.43)
S1 2S1 2
Chapter 10

DIVERGENCES IN n-POINT
FUNCTIONS

Many of the expressions obtained in the preceding sections for two- and four-
point functions are actually ill-defined because they diverge, as we will show
in this section. We start with a rather general discussion of divergences in
φ4 theory and then evaluate explicitly the two- and four-point functions.
To illustrate the divergence of the Green’s functions obtained we consider,
as an example, the two-point function
i
Gc (p, −p) = (10.1)
p2− m2 + iε
 
g i i
+ −i iD F (0) .
2 p2 − m2 + iε p2 − m2 + iε
The loop contribution between the two Feynman propagators in the second
term on the rhs is given by
g g Z d4q i
Σ = iDF (0) = . (10.2)
2 2 (2π) q − m2 + iε
4 2

The integral here diverges: after integrating over q0 we obtain integrals of


the form (cf. section 6.1.2)
Z
d3q
√ 2 . (10.3)
~q + m2
By introducing an upper bound Λ for the integral over |~q| and then taking
Λ → ∞ we see that the integral diverges as Λ2 ; this is called a quadratic
divergence. Because this divergence happens here for large q one also speaks
of a quadratic ultraviolet divergence.

113
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 114

Another way to see the degree of divergence is the so-called “power-


counting”: There are 4 powers of q in the integration measure, but only
two powers of q in the denominator of (10.2); the degree of divergence is
then given by the net power (2) of q.

10.1 Power Counting


The power-counting just illustrated for the case of the tadpole diagram can
be generalized to any Feynman graphs with an arbitrary number of loops.
In order to see this we consider a theory with an interaction ∼ φp in n
dimensions.
R
Since each loop contributes according to the Feynman rules an
integral dnq to the total expression and since each internal propagator gives
a power q −2 , we have for the degree of divergence D in a diagram with L
loops and I internal lines
D = nL − 2I . (10.4)
Note that here each loop has also at least 1 internal line. For example, the
tadpole diagram has L = 1 and I = 1 , giving D = 2 in four dimensions.
D > 0 clearly diverges, D = 0 corresponds to a logarithmic divergence, and
D < 0 seems to be convergent.
If a graph has V interaction vertices, then the total number of lines in φp
theory is pV , since each vertex has p legs. These legs can be either external
or internal lines. If they are internal, they count twice because each internal
line originates and disappears at a vertex. Thus we have

pV = E + 2I , (10.5)

where E is the number of external lines. In addition, the number of loops,


L, is related to the number of vertices, V , by

L=I −V +1 . (10.6)

Combining equations (10.5) and (10.6) with (10.4) allows us to eliminate L


and I to obtain
!  
n(p − 2) p
D =n+ −p V − −1 E . (10.7)
2 2
The degree of divergence of a connected graph in this φp theory in n dimen-
sions thus depends on the number of external legs and, in general, also on
the number of vertices.
In the perturbative treatment of field theory derived in Chap. 8.3 we
have seen that the order of perturbation theory directly equals the number of
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 115

D = 4 · 1 − 2 · 1 = +2

D =4·1−2·2=0

D = 4 · 0 − 2 · 1 = −2

D = 4 · 1 − 2 · 3 = −2

Figure 10.1: Examples for graphs with different degree of divergence. On the
right equation (10.4) is illustrated for n = 4.

vertices, V , in a Feynman diagram. Thus, the degree of divergence becomes


larger and larger with increasing order of a perturbative treatment, if the
factor of V in (10.7) is positive. On the other hand, D is independent of this
order if that factor is zero and D becomes even smaller when going to higher
orders of perturbation theory if the factor is negative. Thus the perturbative
treatment leads to a finite number of divergent terms if and only if

n(p − 2)
−p≤0 . (10.8)
2
If this factor is zero, then the total number of divergent diagrams in a per-
turbative expansion can still be infinite, but in each order of perturbation
theory only the same finite number of divergent diagrams appears. In this
case the theory is called renormalizable. If the factor is negative, then even
the total number of divergent terms is finite; in this case the theory is called
superrenormalizable. In both cases one can add a finite number of so-called
counter terms to the Lagrangian that just remove these divergences.
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 116

In order to become familiar with this power counting for φ4 theory in


four dimensions we give three examples in Fig. 10.1. According to our
earlier considerations we expect that D ≥ 0 diverges. However, this does
not guarantee that graphs with D < 0 actually converge. This is illustrated
by the lowest example, in Fig. 10.1, that has L = 1, I = 3, and thus D =
4 · 1 − 2 · 3 = −2, but diverges, because of the loop on the externals legs.
Only when each possible subgraph has also D < 0, then the whole expression
converges (Weinberg’s Theorem). Note that in the physical case n = 4
and p = 4 D does not depend on V , i.e. it is independent of the order of
perturbation theory. φ4 theory is thus renormalizable.

10.2 Dimensional Regularization


of φ4 Theory
Regularization serves to make all divergent expressions convergent, at the
expense of introducing a parameter that has no physical meaning and ulti-
mately has to be removed. If the integrals can be performed and evaluated as
a function of this parameter then the infinite contributions can be separated
from the finite ones.
We show this procedure by evaluating now the divergent integrals by a
modern technique called dimensional regularization. This technique starts
by considering the theory in n dimensions.
We first consider the dimensions in the Lagrangian
1h µ i g
L= ∂ φ∂µ φ − m2 φ2 − φ4 . (10.9)
2 4!
Since the action Z
S= L dnx (10.10)

is a dimensionless quantity (in units in which h̄ = 1), L must have the


dimension `−n (` is a length). The kinetic energy term in L thus has also
dimension `−n and – since [∂µ ] = `−1 – we get [φ] = `1−n/2 . Any potential
term of the form gφp must have the same dimension as the Lagrangian, i.e.
n
`−n ; we thus get [g] = `−n /`p(1−n/2) = `−(n+p(1− 2 )) . The mass term (p = 2)
thus has [m2 ] = `−2 in n dimensions, as it should.
Note that the dimension of g is just the factor of the number of vertices,
V , in (10.7). Thus, the dimension of the coupling constant and the renor-
malizability of a theory are closely connected: if the dimension of g is that
of a positive or zeroth power of mass, then the theory is superrenormalizable
or renormalizable, respectively.
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 117

These considerations show that for p = 4

[g] = `n−4 = mass4−n ; (10.11)

i.e. in four dimensions the coupling constant of a φ4 theory is dimensionless


and the theory is, therefore, renormalizable; in fact, only theories with p ≤ 4
are renormalizable. Therefore, according to the discussion in the preceding
section the φ4 theory contains only a finite number of divergent vertices. In
n 6= 4 dimensions, however, this is no longer true.
If we want to keep the coupling constant dimensionless in order to ensure
renormalizability also in n dimensions we have to modify the φ4 term in
the Lagrangian (8.51) such that an additional factor with the dimension of
(mass)4−n absorbs the dimension and g becomes dimensionless
g 4−n 4
L = L0 − µ φ . (10.12)
4!
Note that µ here is an arbitrary mass.

10.2.1 Two-point function


The two-point function is completely determined once we know the selfenergy.
We start by calculating this quantity in lowest order in the coupling by
evaluating explicitly the contribution of the tadpole diagram.
To do so we first go to n dimensions so that (10.2) becomes
g Z
dnq i
Σ = µ4−n . (10.13)
2 (2π)n q 2 − m2 + iε
This integral can be obtained analytically by going into a space of n-dimen-
sional polar coordinates (see Appendix. C). The result is
g µ4−n n−2 n
 
n
Σ= n
m π2Γ 1− . (10.14)
2 (2π) 2
The divergence of this expression is now manifest, since the Γ-function has
poles at 0 and the negative integers, and thus also for n = 4.
We now expand Γ around this pole. For that purpose we write
   
n ε
Γ 1− = Γ −1 + (10.15)
2 2
with ε = 4 − n and expand in powers of ε (cf. (C.4))
 
ε 2
Γ −1 + = − − 1 + γ + O(ε) , (10.16)
2 ε
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 118

where γ is the Euler-Mascheroni constant (γ ≈ 0.577..). We thus get for n


close to 4
g µε
 
2−ε 4−ε 2
Σ = m π 2 − − 1 + γ + O(ε) (10.17)
2 (2π)4−ε ε
!ε 
m2 4πµ2

2
2
= g − − 1 + γ + O(ε) .
32π 2 m2 ε

We now use
ε→0
xε = eε ln x −→ 1 + ε ln x (10.18)
and obtain
" !# 
gm2 4πµ2

ε→0 ε 2
Σ −→ 2
1 + ln 2
− − 1 + γ + O(ε)
32π 2 m ε
" !#
gm2 2 4πµ2
= − − 1 + γ − ln + O(ε)
32π 2 ε m2
" !#
gm2 1 gm2 4πµ2
= − − 1 − γ + ln + O(ε) . (10.19)
16π 2 ε 32π 2 m2

We have thus split Σ into two parts. The first one is clearly diverging as n
approaches 4 (ε → 0). The second term is finite and depends on the arbitrary
mass µ that was originally introduced only to keep the coupling constant free
of dimension. The appearance of the arbitrary mass µ in the finite part is
related to the arbitrariness in separating an overall infinite expression into a
sum of an infinite and a finite contribution.
Note that Σ is independent of p. In next higher order the same is true
for the selfenergy contribution of Fig. 9.2b on page 107 whereas that of Fig.
9.2c depends quadratically on p [Ramond].

10.2.2 Four-point function


In this section we will now evaluate the four-point function. By looking at
Fig. 9.4 on page 111 we expect that the four graphs in the second line just
contribute again to the selfenergy. On the other hand, we expect that the
three diagrams in the lowest line can all graphically be contracted such that
they contain only one interaction point with possibly modified interaction
strength.
As an example, we now evaluate the middle graph in the last line of Fig.
9.4. According to the Feynman rules its contribution to the 1PI vertex is
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 119

(see (9.42))
2
d4q
 Z
−ig 4−n i i
∆Γ(p1 , p2 , p3 , p4 ) = µ S3 S1
4! (2π) q − m (p − q)2 − m2
4 2 2

(10.20)
with p = p1 + p3 . The symmetry factor is S1 S3 = 12 · 4! (9.42).
The two denominators in the integrand can be combined into one by a
mathematical trick due to Feynman that is very often used for the evaluation
of such expressions. This trick starts from the elementary integral relation
Zb
dx 1 1 1 b−a
2
= − |ba = − + = . (10.21)
a
x x b a ab

By substituting now
x = az + b(1 − z) (10.22)
we obtain
Zb Z 0
dx dz
2
= (a − b) 2 . (10.23)
a
x [az + b(1 − z)]
1

Combining (10.21) and (10.23) gives


Z1
1 dz
= 2 . (10.24)
ab [az + b(1 − z)]
0

We now apply this to the integrand in (10.20) and obtain


Z1
1 1 dz
=
2 2 2
q − m (p − q) − m 2
0
{(q − m )z + [(p − q)2 − m2 ] (1 − z)}2
2 2

Z1
dz
= .
0
[q 2 − 2pq(1 − z) + p2 (1 − z) − m2 ]2
(10.25)

The first three terms in the denominator can be combined by substituting

q 0 = q − p(1 − z) . (10.26)

This gives for expression in the denominator

q 2 − 2pq(1 − z) + p2 (1 − z) − m2
= [q − p(1 − z)]2 − m2 − p2 (1 − z)2 + p2 (1 − z)
= q 02 − m2 − p2 z(z − 1) . (10.27)
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 120

The 1PI vertex now reads in n dimensions


1
1 2 2 4−n Z dnq 0 Z 1
∆Γ(p1 , p2 , p3 , p4 ) = g (µ ) dz 02
2 (2π) n
0
[q − m2 − sz(z − 1)]2
(10.28)
with s = p2 = (p1 + p3 )2 .
We now interchange the order of integration and evaluate the integral
over q 0 first with the help of (C.18) in Appendix C
Z
dnq 0 1
(2π)n [q 02 − m2 − sz(1 − z)]2
 
i h 2 i n−4 n Γ 2 − n
2
= n
m + sz(1 − z) 2 π 2 . (10.29)
(2π) Γ(2)
The four-point function thus reads
n  
1 2 2 4−n π 2 n
∆Γ(p1 , p2 , p3 , p4 ) = g (µ ) i n
Γ 2−
2 (2π) 2
Z1 h i n−4
× dz m2 + sz(1 − z) 2
. (10.30)
0

We now introduce again ε = 4 − n. This gives


  Z1 " 2 #− ε
1 2 ε 1 ε m + sz(1 − z) 2
∆Γ(p1 , p2 , p3 , p4 ) = g µ i Γ dz . (10.31)
2 16π 2 2 4πµ2
0

Now the four-point function is in a form that allows to take the limit ε → 0.
Using again
xε = eε ln x → 1 + ε ln x (10.32)
and (C.2)  
ε 2
Γ = − γ + O(ε) (10.33)
2 ε
gives
ig 2 µε 2
 
∆Γ(p1 , p2 , p3 , p4 ) = − γ + +O(ε)
32π 2 ε
 1 ! 
εZ m2 + sz(1 − z)
× 1 − ln dz 
2 4πµ2
0
 1 ! 
ig 2 µε 1 ig 2 µε  Z
m2 + sz(1 − z)
= − γ + ln dz 
16π 2 ε 32π 2 4πµ2
0
+ O(ε) with s = (p1 + p3 )2 . (10.34)
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 121

Here the four-point function has been separated into a divergent part and a
convergent term (for ε → 0). The integral appearing is a function of p, m,
and µ; the latter dependence remains even when ε → 0.
So far, we have only calculated the middle graph in the last line of Fig.
9.4 on page 111. It is evident, however, that the result can be directly taken
over also to the other 2 graphs in the last line by taking for p the appropriate
total momentum at one of the vertices. We, therefore, introduce now the
three Lorentz-invariant Mandelstam variables
s = (p1 + p3 )2 = p2
t = (p1 + p2 )2
u = (p1 + p4 )2 , (10.35)
with the property s + t + u = m21 + m22 + m23 + m24 . These variables represent
the total squared four-momentum at the vertex that involves p1 in each of
the graphs in the last line of Fig. 9.4.
We thus get for the sum of all three diagrams in the last line of Fig. 9.4,
the vertex correction diagrams,
3ig 2 µε 1
Γv (p1 , p2 , p3 , p4 ) = (10.36)
16π 2 ε
ig 2 µε
− [3γ + F (s, m, µ) + F (t, m, µ) + F (u, m, µ)] .
32π 2
Here F (s, m, µ) denotes the integral in (10.34).
Of the diagrams in Fig. 9.4 only the first and the three last ones are
1PI graphs. The four diagrams in the second line are all 1P reducible; they
just differ by the propagators on the external legs that are left out when we
consider the 1PI four-point function. The complete 1PI four-point function
is, therefore, given by
Γ(p1 , p2 , p3 , p4 ) = −igµε + Γv (p1 , p2 , p3 , p4 )
3ig 2 µε 1
= −igµε +
16π 2 ε
2 ε
ig µ
− [3γ + F (s, m, µ) + F (t, m, µ) + F (u, m, µ)]
32π 2
3g 1
= − igµε 1 − (10.37)
16π 2 ε 
g
+ [3γ + F (s, m, µ) + F (t, m, µ) + F (u, m, µ)] .
32π 2
Equation (10.37) gives the effective, “dressed” interaction vertex. By com-
parison with the free 1PI four-point function, its forms suggests to absorb all
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 122

effects of the loop graphs contained in the curly brackets into a new effective
coupling constant
g −→ g {1 + δg(s, t, u, m, ε, µ)} . (10.38)
This effective coupling constant depends on s, t, u, m and µ and, as the self-
energy, separates into a divergent and a finite term. It contains all the effects
of the loops.

10.3 Renormalization
In the preceding two subsections we have seen that both the 1PI two-point
and the 1PI four-point functions are (for ε 6= 0) regular functions of ε. The in
four dimensions diverging quantities have thus been regularized. For n → 4
both have divergent and finite contributions from higher order diagrams. In
this section we now show how to handle these divergences by the renormaliza-
tion technique. Any renormalization procedure requires first a regularization,
either by a cut-off for the upper bounds of diverging integrals or by going to
n 6= 4 dimensions, to be followed by a procedure in which the dependence on
these artifacts is removed.
Since the separation of a divergent quantity into a finite and an infinite
contribution is arbitrary there are various so-called renormalization schemes.
They all have in common that they either add terms to the Lagrangian
or scale the fields and coupling constants such that the original form of
the Lagrangian is maintained. The most obvious scheme is the so-called
minimal subtraction scheme that removes just the pole contributions in the
dimensional regularization, i.e. the terms that go like powers of 1/ε. This
scheme is straighforward and well-suited for the dimensional regularization,
but it leads to expressions in which the parameters m (mass) and g (coupling)
have no direct relation to measurable quantities.
Here we discuss another scheme, in which we require that the parame-
ters of the Lagrangian assume their physical, measured values, i.e. m is the
physical mass and g the physical coupling constant.
We start by looking at the structure of the most general two-point func-
tion. When we go to terms that depend on g 2 and higher orders of the
interactions we always encounter 1P reducible graphs, such as the one in
Fig. 9.2a. The general structure of the (reducible) two-point function is of
the form
Σ
Gc (p, −p) = G0 (p, −p) + G0 (p, −p) G0 (p, −p)
i
Σ Σ
+ G0 (p, −p) G0 (p, −p) G0 (p, −p) + . . . , (10.39)
i i
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 123

where Σ is the so-called proper self-energy. Note that (10.39) is an expansion


in terms of irreducible diagrams. Equation (10.39) can be summed and gives
 
Σ Σ Σ
Gc (p, −p) = G0 (p, −p) 1 + G0 (p, −p) + G0 (p, −p) G0 (p, −p) + . . .
i i i
 −1
Σ
= G0 (p, −p) 1 − G0 (p, −p)
i
Σ −1
 
−1
= G0 (p, −p) −
i
i
= 2 . (10.40)
p − m2 − Σ + iε
The self-energy Σ is in general a function of the momentum p of the
particle. We can therefore expand Σ(p2 ) around the on-shell point p2 = m2 ,
where m is the physical, observable mass

Σ(p2 ) = Σ(m2 ) + (p2 − m2 )Σ1 + Σ2 (p2 ) . (10.41)

Here
∂Σ
Σ1 = |2 2 and Σ2 (m2 ) = 0 . (10.42)
∂p2 p =m
In (10.41) only the first two terms of the expansion of Σ have been written
out explicitly; Σ2 (p2 ) denotes the whole remainder of the expansion. Σ, being
the selfenergy insertion of a two point function, is quadratically divergent.
Consequently, Σ1 as the first derivative of Σ with respect to p2 is logarith-
mically divergent and Σ2 as a second derivative is convergent since taking
the derivate always adds one more power of q 2 in the denominator of the
Feynman propagators.
Inserting this expansion into (10.40) gives for the propagator
i
Gc (p, −p) =
p2
− − m2− Σ(m2 ) (p2
− m2 )Σ1 − Σ2 (p2 ) + iε
1 i
= Σ(m 2 )+Σ (p2 ) . (10.43)
1 − Σ1 p − m −
2 2 2
+ iε
1−Σ1

The pole of this propagator should be at the physical mass and its residuum
should be i. However, this is in general not the case, if we start with the
observable, physical mass in the Lagrangian, because the selfinteractions con-
tribute to the self-energy.
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 124

Counter terms. In order to get the pole to the correct, physical location
we therefore have to change the mass in the Lagrangian by adding a so-called
counterterm to it
1 1
Lcm = − δm2 Zφ2 with Z= . (10.44)
2 1 − Σ1
Z is usually called “field renormalization constant” for reasons that will be-
come obvious a little later (see (10.56)). Note that this counterterm has
exactly the same form as the mass term in the original Lagrangian. It will
thus also add a term −Zδm2 in the denominator of the dressed propagator
(10.43).
We determine the unknown δm2 by the requirement
δm2 + Σ(m2 ) = 0 (10.45)
so that the new term just cancels the selfenergy contribution at the on-shell
point. With the counterterm added, the propagator then becomes
1 i
G(p, −p) = . (10.46)
1 − Σ1 p − m − ZΣ2 (p2 ) + iε
2 2

Since Σ̃2 (m2 ) = 0 by definition, this propagator has the correct pole at the
physical mass m.
Its residue, however, is – instead of being simply i –
i
= iZ . (10.47)
1 − Σ1
This deficiency can be cured by adding another, additional counterterm
1  
Lcφ = (Z − 1) ∂µ φ∂ µ φ − m2 φ2 (10.48)
2
to the Lagrangian. Then the propagator becomes
i
G(p, −p) = Z 2
p − m − ZΣ2 (p ) + (Z − 1)(p2 − m2 ) + iε
2 2

i
= 2 . (10.49)
p − m − Σ2 (p2 ) + iε
2

This propagator has the pole at the correct, physical mass m (because of
Σ2 (m2 ) = 0) and the correct residue i.
By adding the given counterterms we have thus removed the divergent
quantities Σ(m2 ) and Σ1 from the propagator. The one remaining quantity
Σ2 (p2 ) involves a second derivative of the selfenergy with respect to p2 and
is convergent; it vanishes at the on-shell point. Note that the counterterms
all have the structure of terms already present in the original Lagrangian. If
this is the case, in general a theory is called renormalizable.
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 125

10.3.1 Renormalization of φ4 Theory


We now specify all of these general considerations to the example of φ4 the-
ory. As we have discussed in Sect. 10.2 the φ4 theory is renormalizable,
i.e. only a finite number of elementary vertices diverges. These are just the
terms we have calculated in the last two subsections, namely the selfenergy
contribution (10.19) and the vertex function (10.37).
For the selfenergy we have, up to one-loop diagrams, (cf. (10.19)),
" !#
2 gm2 1 gm2 4πµ2
Σ(p ) = − − 1 − γ + ln + O(ε) , (10.50)
16π 2 ε 32π 2 m2

i.e. Σ1 = Σ2 = 0 and Z = 1. Equation (10.45) reduces to

δm2 = −Σ(0) = −Σ =⇒ Z = 1 . (10.51)

Thus in this one-loop approximation, there is no field renormalization, but


already in the two-loop approximation we would get Σ1 6= 0 because the
graph Fig. 9.2c is p-dependent. To be general we, therefore, keep the factor
Z in the following expressions.
So far we have not taken the change of the coupling constant due to higher
order loop diagrams into account. In (10.37) we have indeed already seen that
also the interaction vertex gets modified due to higher order loop corrections.
The structure there was such that the coupling constant g was replaced by
an effective coupling constant g(1 + δg(s, t, u, m, ε, µ)). If we again want to
have the physical, observable coupling constant in our Lagrangian we must
get rid of the modification by a proper counter term. We, therefore, define a
vertex renormalization constant Zg by

Zg = (1 + δg(s, t, u, m, ε, µ))−1 |r , (10.52)

where r denotes an in principle arbitrary renormalization point. Since we


want to relate the coupling to a measurable quantity we use the so-called
symmetric point  
2 4 1
pα · p β = m δij − (10.53)
3 3
for i, j = 1, . . . , 4. At this point we have s = t = u = −4m2 /3. We can then
introduce the additional counter term
gµε
Lcv = − (Zg − 1) φ4 . (10.54)
4!
In this way we can ensure that the coupling constant g in the Lagrangian
has its physical, observable value at r.
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 126

The Lagrangian with all the counterterms now reads


h i gµε
L = (∂µ φ)2 − m2 φ2 −
4!
1 h i 1 gµε
+ (Z − 1) ∂ µ φ∂µ φ − m2 φ2 − δm2 Zφ2 − (Zg − 1) φ4
2 2 4!
Zh i 1 gµε
= (∂µ φ)2 − m2 φ2 − δm2 Zφ2 − Zg φ 4 . (10.55)
2 2 4!
This Lagrangian leads to the correct physical behaviour for the 2-point
Green’s function with the physical mass m and the proper residue. The
fields φ, in terms of which the propagator is defined, are thus the physical
fields, which include already the effects of selfinteractions.
By introducing the “bare field” φ0 and the “bare mass” m0 by

φ0 = Zφ
m0 = m2 + δm2
2
(10.56)

and a “bare coupling constant” by


Zg
g0 = gµε , (10.57)
Z2
then the whole Lagrangian can be expressed in terms of bare quantities only
1h i g0
L= (∂µ φ0 )2 − m20 φ20 − φ40 . (10.58)
2 4!
This bare Lagrangian has the same form as the original one, because
all the counter terms had the same form as terms already appearing in the
original Lagrangian. As a consequence it leads to finite physical quantities
in all orders of perturbation theory. If this is the case, then the theory is well
defined, i.e. it is said to be renormalizable. The bare Lagrangian is really
considered to be the ‘true’ Lagrangian of the theory because it leads only to
finite physical quantities.
In the preceding considerations we have chosen 2 arbitrary renormaliza-
tion points; we have required the propagator to have a pole at p2 = m2 ,
where m is the physical mass (cf. (10.49)), and for the vertex we have chosen
the symmetric point r (s = t = u = −4m2 /3) (cf. (10.52)). Choosing other
renormalization points will lead to other values for the mass and the cou-
pling. This arbitrariness in the renormalization point reflects the fact that
the divergent expressions for the selfenergy and the vertex all separate into
a divergent and a finite term. Any scheme, that removes the infinite terms,
will lead to finite expressions for the physical observables, independent of
CHAPTER 10. DIVERGENCES IN N -POINT FUNCTIONS 127

what it does with the finite contributions. It will also lead to the same n-
point function, and – with the help of the reduction theorem – to the same
observable transition rates.
In the dimensional renormalization discussed here all the counter terms
and renormalization factors also depend on the arbitrary mass µ, even after
ε → 0 and after the infinite terms have been removed. This µ was originally
introduced to keep the coupling constant g dimensionless also for n 6= 4.
It is obvious from our considerations above that – for specified renormal-
ization points – choosing different values for µ will lead to different values
for the wavefunction renormalization Z and the coupling constant g. This
arbitrariness, however, is nothing else than the arbitrariness we have already
encountered in the choice of the renormalization point. The physics must be
independent of µ. This observation is the starting point for the development
of the renormalization group method.
It is essential to realize that even in a massless theory, which contains
no dimensional scale parameters, µ introduces a scale that determines the
momentum dependence of the coupling constant. Thus, at the quantum level,
a bare Lagrangian is not enough to specify a theory, but a renormalization
scheme must be added that introduces necessarily a scale into the theory.
Chapter 11

GREEN’S FUNCTIONS FOR


FERMIONS

For simplicity of notation we have so far in this book discussed only the
path integrals and generating functions for scalar fields. All this formalism
can be easily generalized also to vector fields which obey the Proca equation
(4.24) and thus fulfill component by component the Klein-Gordon equation.
For fermion fields, however, there is a problem. The main idea in using path
integrals is to express quantum mechanical transition amplitudes by integrals
over classical fields; the values of these fields at the discrete coordinate-
sites were taken to be commuting numbers. Such a formalism can, however,
not “know” about the Pauli principle. For example, with the formalism
developed so far, a fermion could be propagated to a point in configuration
space which is already occupied. In nature, however, this propagation is
Pauli-forbidden.
For the description of fermions it is, therefore, necessary to extend the
theory developed so far such that the Pauli principle is taken into account.
This can be achieved by using an anticommuting algebra for the classical
fields, the so-called Grassmann algebra.

11.1 Grassmann Algebra


In this section we outline the basic mathematical properties of the Grassman
algebra as far as we will need them in the later developments.
We define the n generators i , . . . , n of an n-dimensional Grassmann al-
gebra by the anticommutation relations

{i , j } ≡ i j + j i = 0 . (11.1)

128
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 129

Let us now consider series expansions in these variables. Since

21 = 22 = . . . = 2n = 0 , (11.2)

because of (11.1), any series in i must have the form


X X
φ() = φ0 + φ1 (i)i + φ2 (i, j)i j (11.3)
i i<j
X
+ φ3 (i, j, k)i j k + . . . + φn (1, 2, . . . n)1 2 . . . n .
i<j<k

Here the φi are ordinary, commuting c-numbers. Note that the expansion
actually terminates because of (11.2). Every element of the algebra can be
written in the form (11.3); thus, the various powers of  in this expansion
constitute a basis of the algebra.
If we choose the function φi to be totally antisymmetric in all their vari-
ables, then we can also write
X 1 X
φ() = φ0 + φ1 (i)i + φ2 (i, j)i j
i 2! ij
1 X
+ φ3 (i, j, k)i j k + . . . . (11.4)
3! ijk

From now on we can assume that the φi are indeed antisymmetric since any
symmetric part of φ would not contribute anyway.
As a consequence of this relation any analytical function in 1 dimension
has the form
φ(1) () = φ0 + φ1  , (11.5)
and in 2 dimensions

φ(2) () = φ0 + φ1 (1)1 + φ1 (2)2 + φ2 (1, 2)1 2 . (11.6)

These relations imply for the Taylor expansion of a Gaussian


2
e− = 1 (11.7)

and
e−i j = 1 − i j . (11.8)
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 130

11.1.1 Derivatives
Since functions of Grassmann variables can, because of (11.2), be at most
linear in ε, the derivative operator is defined as an algebraic operation that,
when applied to a Grassmann variable, simply replaces that variable by a 1;
when applied to a product the variable first has to be commuted to a position
next to the derivative before it can be removed. We then get, for example,
( )
∂ ∂ ∂1
,  i 1 = (i 1 ) + i
∂j ∂j ∂j
∂i ∂1 ∂1
= 1 − i +  i = δij 1 . (11.9)
∂j ∂j ∂j
The “−” sign in the second line appears, because the factor 1 must be
first brought to the left before the derivative can act. This result is valid
in general. We thus can define the derivative also by the anticommutation
relation ( )

, i = δij . (11.10)
∂j
This relation can be used to obtain the derivative of a function φ().
Applying the lhs of (11.10) to φ, which can always be written as polynomial
(see (11.4)), yields
( )
∂ ∂ ∂
, i φ() = [i φ()] + i φ() (11.11)
∂j ∂j ∂j
"
∂ X 1 X
= δij φ() + φ0 − φ1 (k)k + φ2 (k, l)k l
∂j k 2! kl
!#
1 X ∂
− φ3 (k, l, m)k l m + . . . i +  i φ() .
3! klm ∂j

Here all the terms in φ() with an odd number of 0 s have changed sign when
they were brought to the left so that the derivative can act on them. With

k l = δkj l − δlj k (11.12)
∂j
and

k l m = δkj l m − δlj k m + δmj k l (11.13)
∂j
we get
( ) "
∂ 1 X 1 X
, i φ() = δij φ() + −φ1 (j) + φ2 (j, l)l − φ2 (k, j)k
∂j 2! l 2! k
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 131

1 X 1 X
− φ3 (j, l, m)l m + φ3 (k, j, m)k m
3! lm 3! km
#
1 X ∂
− φ3 (k, l, j)k l + . . . i + i φ( )
3! kl ∂j
= δij φ()
" #
X 1 X
− φ1 (j) − φ2 (j, k)k + φ3 (j, k, l)k l − . . . i
k 2! kl

+ i φ( ) . (11.14)
∂j

Because of (11.10) this has to be equal to δij φ(). Commuting i through


to the left side of the square parentheses changes again the signs of all odd
terms. The last two terms in (11.14) then cancel each other if the expression
in the square parentheses equals the derivative of φ
∂ X 1 X
φ() = φ1 (j) + φ2 (j, k)k + φ3 (j, k, l)k l + . . . . (11.15)
∂j k 2! kl

This gives the derivative of a general function φ(). Eq. (11.15) could also
have been obtained by differentiating the expansion (11.4) directly.
The second derivatives can also be defined
∂ ∂
φ() = φ2 (j, i)
∂i ∂j
1 X 1 X
+ φ3 (j, i, l)l − φ3 (j, k, i)k + . . .
2! l 2! k
X
= φ2 (j, i) + φ3 (j, i, l)l + . . . . (11.16)
l

Because the φi are antisymmetric this gives immediately

∂2 ∂2
φ() + φ() = 0 , (11.17)
∂i ∂j ∂j ∂i
or ( )
∂ ∂
, =0. (11.18)
∂i ∂j
Thus we have, in particular,
∂2
=0. (11.19)
∂2i
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 132

This relation implies that there is no inverse to differentiation. This can


be seen by multiplying the defining equation for the inverse
!−1
∂ ∂
φ() = φ() (11.20)
∂ε ∂ε

from the left by ∂ε
. This gives
!−1 !−1
∂ ∂ ∂ ∂ ∂
φ() = 0 · φ() = φ() (11.21)
∂ε ∂ε ∂ε ∂ε ∂ε
and thus the inverse does not exist.

11.1.2 Integration
Integration in the space of Grassmann variables can, therefore, be defined
only in an operational sense. This is achieved by the following relations
Z
di = 0
Z
di i = 1 . (11.22)

The symbols di obey the commutation relations


{di , dk } = {di , k } = 0 , (11.23)
but note that d is not an infinitesimal interval and, in particular, not a
member of the Grassmann algebra of the i .
Integration over Grassmann variables has thus the same effect as dif-
ferentiation. The integral in (11.22) is the only nonvanishing integral over
functions of i . It has the property of translational invariance
Z Z
(1)
d φ () = d φ(1) ( + α) = φ1 (11.24)

with the definition of φ(1) in (11.5); α is also a Grassmann number.


For Grassmann variables we can also define a δ-function. In 1 dimension
we have
Z Z
0 0 0
− d ( −  )φ( ) = − d0 ( − 0 )(φ0 + φ1 0 )
= φ0 + φ1  = φ() . (11.25)
We can thus identify the δ function as
δ( − 0 ) = −( − 0 ) ; (11.26)
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 133

note that δ( − 0 ) is an odd function. There is also a Fourier-representation


for the δ-function
Z Z
0
0
δ( −  ) = −( −  ) = 0
dζ [1 − ζ( −  )] = 0
dζ e−ζ(− ) ; (11.27)

here (11.8) has been used.

11.1.2.1 Multiple integrals


In multiple integrals the integration variable first has to be commuted to a
position next to the integration measure. For example, we get
Z Z
−i j
di dj e = di dj (1 − i j )
Z Z Z 
= 0− di dj i j = + di dj j i
Z
= di i = 1 . (11.28)

We now determine the Jacobian J that appears when we perform a linear


transformation
0 = O (11.29)
of the integration variables in n dimensions; O is a general matrix. J is
defined by
Z Z
d01 . . . d0n 01 . . . 0n = d1 . . . dn J (O)1 (O)2 . . . (O)n (11.30)
Z
= d1 . . . dn J O1α O2β . . . Onν α β . . . ν .

Note that this equality holds for general matrices O. The product of the
Grassmann variables α , . . . , ν vanishes, if any of the indices appears twice.
Thus only the permutations of 1, . . . , n survive. We can therefore bring each
of them into the ordered form by writing

α β . . . ν = (−)P 1 . . . n , (11.31)

where (−)P is the sign of the permutation. We thus have


Z Z X
d01 . . . d0n 01 . . . 0n = d1 . . . dn J (−)P O1α . . . Onν 1 . . . n
P
Z
= d1 . . . dn J 1 . . . n det(O) . (11.32)
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 134

On the other hand, we can also evaluate the integrals over the ’s directly.
This gives Z Z
0 0 0 0
d1 . . . dn 1 . . . n = d1 . . . dn 1 . . . n , (11.33)

because of (11.22). Comparison with (11.33) yields

J = [det(O)]−1 . (11.34)

Note that this is the inverse of the ordinary result!

11.1.2.2 Gaussian integrals


We next consider the Gaussian integral
Z
T M
I(n) = d1 . . . dn e− (11.35)

where M is an antisymmetric (n×n) matrix with ordinary c-number elements


and  is a vector with n elements. Expanding the exponential function we
obtain
T 1 T 2
e− M  = 1 − T M  +  M − · · · . (11.36)
2!
For simplicity let us consider the case n = 2 first. We have for I(2)
Z Z  
−T M  T 1 T 2
I(2) = d1 d2 e = d1 d2 1 −  M +  M − . . .
2
Z
T 1Z  2
= 0 − d1 d2  M  + d1 d2 T M  − . . .
2
≡ −I1 (2) + I2 (2) − . . . (11.37)

For the first integral we obtain


Z Z
I1 (2) = d1 d2 T M  = d1 d2 i Mij j
Z
= d1 d2 (1 M11 1 + 1 M12 2 + 2 M21 1 + 2 M22 2 )
Z
= d1 d2 (1 M12 2 + 2 M21 1 ) , (11.38)

because the integrands of the other two terms contain squares of Grassmann
variables which vanish. We therefore have (because M is antisymmetric)
q
I1 (2) = −M12 + M21 = −2M12 = −2 det(M ) . (11.39)
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 135

The second integral in (11.37) is


Z  2
I2 (2) = d1 d2 T M  (11.40)
Z
= d1 d2 i Mij j k Mkl l .

The indices i, j, k and l all have to be different, because otherwise the product
of the 0 s vanishes. This, however, is impossible because we have only 2
different variables; the integral I2 (2) thus vanishes. The same holds for all
higher order terms in (11.37).
In summary we obtain for the integral (11.37)
q 2
q
I(2) = −I1 (2) = +2 det(M ) = 2 2 det(M ) . (11.41)

For 3 variables we obtain immediately


Z
T M
d1 d2 d3 e− =0, (11.42)

because the exponent is quadratic in  and, therefore, the series expansion


involves
R
only even powers. The term of second order in  gives 0, because
of d = 0; the term of fourth order also vanishes, because, with only three
variables, it has to contain a square of one of them, which vanishes.
A similar reasoning leads one to the conclusion that with four variables
only the term of fourth order in  can contribute. Thus we have
Z
T M
I(4) = d1 d2 d3 d4 e−
1Z  2
= d1 . . . d4 T M 
2Z
1
= d1 . . . d4 i Mij j k Mkl l
2
4 Z
1 X
= Mij Mkl d1 . . . d4 i j k l . (11.43)
2 ijkl=1

Here only the terms with all four indices different can contribute. We thus
get 4! = 24 nonvanishing terms. Of these 24 terms many are equal to each
other: the antisymmetry Mij = −Mji gives Mij Mkl = Mji Mlk and Mij Mlk =
Mji Mkl = −Mij Mkl and thus a factor 22 , the symmetry Mij Mkl = Mkl Mij
another factor 2, so that only 3 essentially different terms remain. We then
have
Z
I(4) = 4M12 M34 d1 . . . d4 1 2 3 4
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 136

Z
+ 4M13 M24 d1 . . . d4 1 3 2 4
Z
+ 4M14 M23 d1 . . . d4 1 4 2 3
= 4 (M12 M34 − M13 M24 + M14 M23 )
4
q
= 2 2 det(M ) (11.44)

The last step can be seen by explicitly expanding the (4 × 4) determinant of


M in terms of its elements in the first row.
The result just obtained holds in general
Z ( n
q
−T M  22 det(M ) n even
d1 . . . dn e = for . (11.45)
0 n odd

The proof is given in Appendix D. Equation (11.45) is the equivalent of


(B.10) for commuting numbers. Note that here the squareroot of the deter-
minant appears in the numerator whereas for commuting numbers it is in
the denominator.
The other integral relations for commuting variables, derived in section
B.2.1, can also be obtained for Grassmann variables. One gets
Z q
1 T 1 T
M −1 η
d1 . . . dn e− 2  M +η i i
= det(M ) e 2 η , (11.46)

where the ηi are also Grassmann variables with

{ηi , ηj } = 0 = {ηi , j } (11.47)

(there is no factor 2(2n)/2 here because the exponent contains an extra factor
1/2). Equation (11.46) corresponds directly to (B.10).

Complex integration. We now enlarge the algebra of the n by adding


the conjugate elements. Here conjugation is defined by the properties

(i )∗ = ∗i
(∗i )∗ = i
(λi )∗ = λ∗ ∗i
(1 2 · · · n )∗ = ∗n ∗n−1 · · · ∗1 . (11.48)

The latter condition ensures that ∗  is real. The n Grassmann genera-


tors ∗1 , . . . , ∗n are then combined with the n generators 1 , . . . , n into a 2n-
dimensional algebra with generators κ1 , . . . , κ2n such that

{κ1 , . . . , κ2n } = {∗1 , 1 , ∗2 , 2 , . . . , ∗n , n } . (11.49)


CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 137

All the κi anticommute with each other.


With this definition we evaluate a Gaussian integral containing also con-
jugate Grassmann numbers
Z
†M 
Z∼ d∗1 d1 d∗2 d2 . . . d∗n dn e− . (11.50)
The matrix M in (11.50) can now be a general, complex matrix without
any specific symmetries. In contrast to the c-number integrals discussed in
section B.2.1 it does not have to have special properties since convergence of
the integrals is guaranteed here by the Grassmann algebra. In order to be
able to apply the results obtained above (see (11.45)) to this integral we now
enlarge the general (n × n)-dimensional matrix M into an antisymmetric
(2n × 2n)-dimensional matrix N , so that the exponent in (11.50) can be
rewritten
n 2n
X 1 X 1
† M  = ∗i Mij j = κk Nkl κl = κT N κ . (11.51)
i,j=1 2 k,l=1 2
For example, for the case n = 2 the matrix N reads
 
0 M11 0 M12

 −M11 0 −M21 0 

N =  . (11.52)
 0 M21 0 M22 
−M12 0 −M22 0
One then gets in general
det(N ) = [det(M )]2 . (11.53)
Using now (11.45) gives
Z q
1 T Nκ
Z∼ dκ1 . . . dκ2n e− 2 κ = det(N ) , (11.54)
and with (11.53) we finally get
Z
†M 
Z∼ d∗1 d1 . . . d∗n dn e− = det(M ) . (11.55)
This result can also easily be obtained for a diagonal matrix M . Even though
the integral is a 2n-dimensional one, the matrix M on the rhs is only n-
dimensional. Note again that the matrix M here is a general complex matrix.
A slightly more general form follows when we add linear terms in the
exponent and write an expressively imaginary M . Then we obtain
Z Y
† M +iη † +i† η † M −1 η
d∗n dn e−i = det(iM ) eiη . (11.56)
n
In all these formulas for Grassmann integrals the determinant appears in
the numerator in contrast to the Gaussian integrals over ordinary numbers
(B.22).
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 138

11.2 Green’s Functions for Fermions


In this section we now apply the Grassmann algebra to a path integral de-
scription of fermions. We do so by introducing classical Grassmann fields
ψ for the fermions. These appear in the continuum limit when the number
of generators n of the Grassmann algebra goes to infinity. The relations
corresponding to (11.1),(11.10) and (11.22) are, respectively,

{ψ(x), ψ(y)} = 0
( )
δ
, ψ(y) = δ(x − y)
δψ(x)
Z
dψ(x) = 0
Z
dψ(x) ψ(x) = 1 . (11.57)

With these fields we can construct path integrals of the type


Z R
ψ̄M ψ d4x
Z∼ Dψ̄ Dψ ei (11.58)

where the ψ̄ and ψ are Grassmann fields.

11.2.1 Generating Functional for fermions


The Lagrangian for Dirac fields is

L = Ψ̄ (iγµ ∂ µ − m) Ψ − V (ψ̄, ψ) (11.59)

where Ψ and Ψ̄ are to be considered as independent fields. They are now


taken to be elements of an infinite-dimensional Grassmann algebra. The
path integral is then – in direct generalization of our results for boson fields
– given by
Z R
i d4x [L+η̄(x)Ψ(x)+Ψ̄(x)η(x)]
Z[η, η̄] = Z0 DΨ̄ DΨ e . (11.60)

Here Z0 is the normalization factor, equal to the inverse of the path integral
without sources. The functions η(x) and η̄(x) are four-component source
functions corresponding to the classical fields Ψ̄ and Ψ, respectively. They
are also taken to be Grassmann fields that anticommute among themselves
as well as with the fields Ψ and Ψ̄ .

{ψ(x), ψ(x0 )} = {ψ(x), ψ̄(x0 )} = {ψ̄(x), ψ̄(x0 )} = 0


CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 139

{η(x), η(x0 )} = {η(x), η̄(x0 )} = {η̄(x), η̄(x0 )} = 0


{η(x), ψ(x0 )} = {η(x), ψ̄(x0 )} = {η̄(x), ψ(x0 )} = 0 . (11.61)
While (11.60) is an obvious generalization of our earlier results obtained
in Chap. 5 for scalar fields, its actual (numerical) evaluation may not seem to
be straightforward, because of the Grassmann nature of the fields. In order
to define more stringently what is actually meant by the path integral in
(11.60) we can proceed as in chapter 5.1 for scalar fields and Fourier-expand
the fields. This is carried through in detail in [LEE]. For our present purpose
it is enough to mention that the fields Ψ and Ψ̄ can be expanded in terms of
Grassmann numbers α and ¯α such that
X
Ψ(x) = φα (~x)α (t)
α
X
Ψ̄(x) = φ̄α (~x)¯α (t) , (11.62)
α

where the φα are ~x-dependent Dirac spinors (u(~pα , sα )ei~pα ·~x or v(~pα , sα )e−i~pα ·~x ,
for positive and negative energies, respectively, see [LEE], p. 517-521).
Equation (11.62) just amounts to an ~x-dependent change of the basis from
α , ¯α to the Grassmann numbers Ψ and Ψ̄. The fermionic action, e.g. for
free fields, then reads in terms of the Grassmann numbers (with 6 k ≡ γµ k µ )
Z
S = Ψ̄ (i 6 ∂ − m) Ψ d4x
XZ
= φ̄α ¯α (i 6 ∂ − m) φβ β d4x , (11.63)
αβ

and the path-integral integration measure is really


YY
DΨ̄ DΨ = dΨ̄(xk , tj ) dΨ(xk , tj )
k j
YY
∼ d¯α (tj ) dα (tj ) . (11.64)
α j

Here the first index (α) denotes the Grassmann generator, the second (j) the
time interval. The last step here was possible because of the orthonormality
of the functions φα and φ̄α .

11.2.1.1 Feynman propagator for fermions


In the case of free boson fields we succeeded in writing the generating func-
tional as a term involving only space-time integrals over the sources (see
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 140

(6.11) and (6.39)). This had the advantage that the two-point function could
be directly read off from that expression.
We achieve here the same for fermions by using (11.56). This gives
Z R
d4x(L+η̄Ψ+Ψ̄η)
Z0 [η, η̄] = Z0 DΨ̄DΨ ei
−1 η
= Z0 det(iM ) eiη̄M , (11.65)

where the matrix M is obtained by identifying


Z h i
−i† M  ←→ i d4x Ψ̄(x)(i 6 ∂ − m)Ψ(x) (11.66)
Z h i
= −i d4 x d4 y Ψ̄(x) (i 6 ∂y + m)δ 4 (x − y) Ψ(y)

after partial integration. Since Z0 [0, 0] = 1, we have Z0 = [det(iM )]−1 . We


thus get R −1 4 4
Z0 [η, η̄] = ei η̄(x)M (x,y)η(y)d x d y . (11.67)
The inverse operator appearing here can be obtained as at the end of
section (6.1.3)
Z Z h i
M (x, z)M −1 (z, y) d4z = (i 6 ∂ z + m) δ 4 (x − z) M −1 (z, y) d4z
Z
= − δ 4 (x − z) (i 6 ∂ z − m) M −1 (z, y) d4z
!
= − (i 6 ∂ x − m) M −1 (x, y) = δ 4 (x − y) .
(11.68)

The last equation is fulfilled by

M −1 (x, y) = − (i 6 ∂ + m) DF (x − y) , (11.69)

since

−(i 6 ∂ − m)(i 6 ∂ + m)DF (x − y) = (2 + m2 )DF (x − y)


= − δ 4 (x − y) , (11.70)

because of (6.7). Here we have used {γµ , γν } = 2gµν .


We therefore have now the desired result. For free fields the normalized
generating functional is
R
η̄(x)SF (x−y)η(y) d4xd4y
Z0 [η, η̄] = e−i , (11.71)
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 141

with the Feynman propagator for fermions


SF (x − y) = (i 6 ∂ + m)DF (x − y) (11.72)
being the propagator of the Dirac equation
(i 6 ∂ − m) SF (x − y) = δ 4 (x − y) . (11.73)
Note that SF is a (4 × 4) matrix because of the Dirac matrix structure of γµ .
For SF we can also find an integral representation
1 Z 4 e−ik(x−y)
SF (x − y) = (i 6 ∂ + m) d k
(2π)4 k 2 − (m − i)2
Z
1 4 −ik(x−y) 6k + m
= d k e
(2π)4 k 2 − (m − i)2
Z
1 1
= 4
d4k e−ik(x−y) (11.74)
(2π) 6 k − (m − i)
because of (6 k + m)(6 k − m) = k 2 − m2 .
Note that SF (x−y) is not symmetric under exchange x ↔ y, while DF (x−
y) is. However, it is antisymmetric under an operation that corresponds in
Dirac theory to simultaneous x ↔ y exchange and hermitean conjugation
since
γ0 SF (y − x)† γ0 = SF (x − y) , (11.75)
which can easily be derived with the help of the relation
γ0 㵆 γ0 = γµ . (11.76)
As we have discussed in Sect. 6.1.2 for scalar fields the Feynman propaga-
tor moves positive-energy solutions forward in time and those with negative
energy backwards. The same is true for the propagator for fermions as can
be shown following steps very similar to those leading to (6.21).

11.2.2 Green’s Functions


The n-point functions for fermions can now be obtained, as in the case of
bosons, as functional derivatives of the functional Z[J]. For this purpose we
modify the definition (8.3) such that the presence of two fermion fields, Ψ
and Ψ̄, is taken into account.
We define the 2n-point function for fermions by
Gα,β,...,2ν (x1 , x2 , . . . , x2n ) (11.77)
 2n
2n
1 δ Z

= ,
i δη2ν (x2n ) · · · δην+1 (xn+1 )δ η̄ν (xn ) · · · δ η̄α (x1 ) η=η̄=0
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 142

where Z[η, η̄] is given by (11.60)


Z R
d4x [L+η̄(x)Ψ(x)+Ψ̄(x)η(x)]
Z[η, η̄] = Z0 DΨ̄ DΨ ei . (11.78)

Here the order of the functional derivatives with respect to η and η̄, respec-
tively, is fixed by convention (see discussion below). Note that the Green’s
function now depends also on the Dirac spinor-indices, in addition to the
space-time coordinates.
As in (8.2) for bosons, the Green’s function can be written as a vacuum
expectation value of a time-ordered product of field operators

G(x1 , x2 , . . . , x2n ) (11.79)


R
DΨ̄ DΨ Ψ(xn ) · · · Ψ(x1 )Ψ̄(x2n ) · · · Ψ̄(xn+1 ) eiS
= R
DΨ̄ DΨ eiS
h i
ˆ (x ) · · · Ψ̄
= h0|T Ψ̂(xn )Ψ̂(xn−1 ) · · · Ψ̂(x1 ) Ψ̄ ˆ (x ) |0i ,
2n n+1

where S denotes the action. The special ordering of fields in (11.79) is a


consequence of our ordering of the derivatives in (11.77); it ensures that there
is no additional phase present. This can be seen as follows: In calculating the
derivatives with respect to η̄ no phase appears, because first the derivative
is anticommuted through all field operators already pulled down from the
exponential, then acts without a further sign change on the exponential and
then is anticommuted back, thus creating the same phase backwards as on
the way forward. The extra phase (−) that appears from the derivative with
respect to η due to the necessary reordering in the exponent is cancelled by
the same phase caused by anticommuting the derivative operator δ/δη with
all the Grassmann fields Ψ and Ψ̄, that were already pulled down, and by
reordering them.
Thus, the n-point Green’s function, originally introduced as a functional
derivative of a generating functional, can in general also be defined as the
vacuum expectation value of a time-ordered product of field operators Φ̂(x)
that can describe either boson or fermion fields
h i
G(x1 , x2 , . . . , x2n ) = h0|T Φ̂(xn )Φ̂(xn−1 ) · · · Φ̂(x1 ) Φ̂(x2n ) · · · Φ̂(xn+1 ) |0i .
(11.80)
The only difference between boson and fermion fields appears in the reorder-
ing caused by the T operator; here a ”(−)” sign appears whenever a pair
of fermions changes its order whereas there is no such sign when the boson
operators are reordered.
CHAPTER 11. GREEN’S FUNCTIONS FOR FERMIONS 143

2-point function. The two-point function is given by



δ 2 Z[η, η̄]
Gαβ (x1 , x2 ) = − . (11.81)
δηβ (x2 )δ η̄α (x1 ) η=η̄=0
Using (11.65) for the generating functional of the free theory we obtain
R
R L d4x
DΨ̄DΨ Ψα (x1 )Ψ̄β (x2 )ei
Gαβ (x1 , x2 ) = R R
L d4x
(11.82)
DΨ̄DΨ ei
Here α and β are Dirac-Spinor indices.
This expression can, as for bosons, be written as an expectation value
over a time-ordered product of field operators
h i
ˆ (x ) |0i
G(x1 , x2 ) = h0|T Ψ̂(x1 )Ψ̄ (11.83)
2

where, now for fermions, we have



h i  ˆ (x )
Ψ̂(x1 )Ψ̄ t1 > t 2
ˆ (x ) =
T Ψ̂(x1 )Ψ̄ 2
2 ˆ (x )Ψ̂(x ) for t2 > t1
 −Ψ̄
. (11.84)
2 1

Note the extra minus sign compared to the boson case. This sign appears
because of the Grassmann nature of the fermion fields when we reorder the
fields as in the developments leading to (3.47).
We could have obtained the two-point function also from (11.71). This
gives
Gαβ (x1 , x2 ) = i[SF (x1 − x2 )]αβ . (11.85)
Again, as in scalar field theory, the 2-point function is – up to a phase – just
equal to the propagator of the relevant equation of motion.
The propagator of our fermionic theory is thus just the inverse of the
operator appearing between the fields Ψ̄ and Ψ in the Lagrangian
L = Ψ̄(i 6 ∂ − m)Ψ
G(x, y) = i(i 6 ∂ − m)−1 = iSF (x − y) . (11.86)
For boson fields this was also the case (see the discussion at the end of section
6.1.3)
1
L = − φ(2 + m2 )φ
2
G(x, y) = − i(2 + m2 )−1 = iDF (x − y) . (11.87)
We can thus read off the propagator directly from the Lagrangian1 .
1
The factor 1/2 in the Lagrangian for the bosons is a special feature of the uncharged
field and would not be there if we were working with charged fields.
Chapter 12

INTERACTING FIELDS

So far we have treated only the case of free fermion fields. In the case of an
interacting theory, expression (8.44) can directly be taken over for the case
of fermions so that we have
R R
−i d4x V ( 1i δ 1 δ
, ) e−i η̄(x)SF (x−y)η(y)d4x d4y
Z[η, η̄] = Z0 e δη i δ η̄ . (12.1)
Here the interaction V is defined by the Lagrangian of the interacting theory
L = L0 − V (Ψ̄, Ψ) . (12.2)
If the theory involves both boson and fermion fields and their couplings,
i.e. if the Lagrangian is given by
φ
L = LΨ
0 + L0 − V1 (Ψ̄, Ψ) − V2 (φ) − Vint (Ψ̄, Ψ, φ) , (12.3)
then Z[J, η, η̄] is given simply by (cf. (8.44))
R
d4x Vint ( 1i δη , i δη̄ , i δJ )
δ 1 δ 1 δ
Z[J, η, η̄] = Z0 e−i
R R
×e−i R d x V1 ( i δη , i δη̄ ) e−i d x V2 ( i δJ )
4 1 δ 1 δ 4 1 δ

4 4
×e−i Rη̄(x)SF (x−y)η(y)d x d y
i 4 4
×e− 2 J(x)DF (x−y)J(y)d x d y . (12.4)
This functional can generate all the n-point functions of the interacting the-
ory and these can again be graphically represented in the form of Feynman
diagrams.

12.1 Feynman Rules


The Feynman rules of Sect. 9.1 can be extended by the following rules (in
momentum space)

144
CHAPTER 12. INTERACTING FIELDS 145

Figure 12.1: Fermion-boson vertex (see (12.6)). The solid line denotes the
fermion, the dashed one the boson. The dot denotes the interaction vertex.

1) each internal fermion line gives a factor


i
. (12.5)
6 k − (m − i)
The fermion lines carry now an arrow that points into the direction
into which the fermion’s charge flows. This is a consequence of the fact
that the fermion propagator is no longer symmetric in its arguments
(cf. (11.75)) so that we have to give the direction of motion explicitly.
For fermions we deal with two fields, ψ and ψ̄, that carry different
charges. In a Feynman graph, in which time runs from left to right,
particle lines are always rightwards directed whereas antiparticle lines
move leftwards. In a loose way of speaking the fermion lines move from
ψ̄ to ψ, because ψ̄ creates particles and ψ annihilates them. Since the
opposite holds for antiparticles their lines run leftwards, reflecting the
fact that the Feynman propagator propagates them backwards in time.
In a coupled theory a new class of diagrams can appear because of the
coupling of fermion and boson fields. For example, a coupling term of the
form
Vint = g Ψ̄Ψφ = g Ψ̄α Ψα φ , (12.6)
the so-called Yukawa coupling, will generate vertices of the form shown in
Fig. 12.1. We, therefore, have a new Feynman rule dealing with fermion-
boson vertices:
2) each boson-fermion vertex gives a factor −ig. If the coupling term
involves other bilinear forms of the Dirac spinors, e.g. Ψ̄γµ Ψ, then an
additional factor γµ would appear at each vertex.

12.1.1 Fermion Loops


A change in the rules appears when we consider fermion loop insertions in
the boson propagator, e.g. the graph shown in Fig. 12.2, where the solid
CHAPTER 12. INTERACTING FIELDS 146

(x α) (y β)

Figure 12.2: Fermion loop contribution to the boson propagator.

line denotes the fermions and the dashed line the bosons. Such a term is
obviously of second order in the fermion-boson coupling constant (indicated
by the two vertices) and contributes to the bosonic 2-point function. It
can thus be generated by the second order term in the expansion of (12.4),
properly generalized to fermion fields. Because the graph in Fig. 12.2 is one-
particle irreducible only the true second order term (∼ S2 in (8.50),(8.51))
can contribute. This term has the structure

iS[η, η̄, J] = . . .
" Z !#2
1 −iS0 [η,η̄,J] 1 δ 1 δ 1 δ
+ e (−i) d4x Vint , , e+iS0 [η,η̄,J]
2 i δη(x) i δ η̄(x) i δJ(x)
+ ... . (12.7)

Here S0 [η, η̄, J] is given by


iZ 4 4
iS0 [η, η̄, J] = − d v d w J(v)DF (v − w)J(w)
2Z
−i d4x0 d4y 0 η̄(x0 )SF (x0 − y 0 )η(y 0 ) . (12.8)

We now remember that only those terms in the generating functional can
contribute to the boson propagator in Fig. 12.2 that are of second order in
the bosonic current J and of zeroth order in the fermionic currents η and η̄.
Therefore, the only contributing term in (12.7) with the interaction (12.6) is

(−ig)2 −iS0
iS2loop = − e (12.9)
2
Z
δ6
× d4x d4y e+iS0
δηα (x)δ η̄α (x)δJ(x)δηβ (y)δ η̄β (y)δJ(y)
(−ig)2 −iS0 Z 4 4 Z 4 4
= e d xd y d vd w
2
δ4
 
× DF (x − v)J(v)DF (y − w)J(w) e+iS0 ,
δηα (x)δ η̄α (x)δηβ (y)δ η̄β (y)
CHAPTER 12. INTERACTING FIELDS 147

after performing the derivative with respect to J.


We now concentrate on the fermionic derivative (for notational conve-
nience we drop here the index F for the propagators)
δ3 δ R 0 0 0 0 4 0 4 0
e−i η̄(x )S(x −y )η(y ) d x d y
δηα (x)δ η̄α (x)δηβ (y) δ η̄β (y)
δ2 δ Z R
= (−i) Sβγ (y − y 0 )ηγ (y 0 ) d4y 0 e−i ···
δηα (x)δ η̄α (x) δηβ (y)
"
δ δ
= − iSββ (y − y)
δηα (x) δ η̄α (x)
Z Z  #
R
0 0 4 0 0 0 4 0 −i ···
− Sβγ (y − y )ηγ (y ) d y η̄δ (x )Sδβ (x − y) d x e
" Z
δ
= − Sββ (y − y) Sαγ (x − y 0 )ηγ (y 0 ) d4y 0
δηα (x)
Z  #
R
0 0 4 0 −i ···
⊕ Sβγ (y − y )ηγ (y ) d y Sαβ (x − y) + O(η η̄) e . (12.10)

Here the symbol ⊕ denotes a ‘plus’ sign that is due to the Grassmann nature
of the source functions and would have been opposite, if we had worked
with bosons instead of fermions. The terms denoted by O(η η̄) stand for
expressions that involve products of the two source functions and would thus
be linear in η or η̄ after the last functional derivative has been taken. Since
the n-point function is given by a functional derivative at vanishing source,
these terms do not contribute to the 2-point function we are after.
Performing now the last derivative gives for the expression (12.10)
= [−Sββ (y − y)Sαα (x − x) ⊕ Sβα (y − x)Sαβ (x − y) + O(η η̄)]
R
η̄(x0 )SF (x0 −y 0 )η(y 0 ) d4x0 d4y 0
× e−i . (12.11)
The sign of the second term would have been opposite, if we had worked
with bosons.
We now combine this result with (12.9). Since we are interested in partic-
ular in the fermion loop insertion to the boson propagator shown above, we
need to construct a bosonic 2-point function with no external fermion lines.
Therefore, only those terms of the fermionic part (12.11) can contribute to
this graph that contain no sources η or η̄, except in the exponential. All
other terms ∼ η η̄ would vanish after setting η = η̄ = 0.
Disregarding the vacuum contributions (first term in (12.11)) we thus
have as the only term of (12.11) that contributes to the loop diagram
⊕ tr [SF (y − x)SF (x − y)] . (12.12)
CHAPTER 12. INTERACTING FIELDS 148

Inserting this result into (12.9) gives for the relevant part of S2

(−ig)2 Z 4 4 4 4
iS2loop = ⊕ d x d y d v d w tr [SF (y − x)SF (x − y)]
2
× DF (x − v)DF (y − w)J(v)J(w) (12.13)

and, correspondingly, for the 2-point function

δ 2 (iS2loop )
Gloop
boson (x1 , x2 ) =− (12.14)
δJ(x1 )δJ(x2 )
Z
= (−ig)2 d4x d4y DF (x1 − x)tr [SF (x − y)SF (y − x)] DF (y − x2 ) .

When we retrace this detailed calculation we find that the positive sign in
(12.12) and thus the negative sign in (12.14) is due to the fact that η and η̄
are Grassmann fields.
Thus, we get the additional Feynman rule, which holds in general,

3) Any fermionic loop graph is associated with an additional (−) sign and
a trace over the Dirac indices.

12.2 Wick’s Theorem


In Sect. 8.2.1 we had expressed the n-point function for bosons G(x1 , . . . , xn )
in terms of a symmetrized product of n/2 2-point functions (Wick’s theorem).
The same can now be done also for the general theory that involves bosons
and fermions.

Mixed n-point functions. First, it is clear that boson and fermion fields
commute; the corresponding field operators act in different Hilbert spaces so
that the groundstate expectation value separates into a product of vacuum
expectation values over fermion and boson operators separately, e.g.

h0|Ψ̂(1)φ̂(2)φ̂(3)Ψ̄ ˆ (4)|0ih0|φ̂(2)φ̂(3)|0i .
ˆ (4)|0i = h0|Ψ̂(1)Ψ̄ (12.15)

We, therefore, need to formulate here Wick’s theorem only for the fermions.

Wick’s theorem for fermions. We now derive a simple method, known


as Wick’s Theorem that we encountered for bosons already in Sect. 8.2.1, for
the evaluation of a vacuum expectation value of a time-ordered product of
free field operators. In that case we can use the form (11.67) of the generating
CHAPTER 12. INTERACTING FIELDS 149

functional for the evaluation of the functional derivative. Performing first the
functional derivatives with respect to η̄ on that expression gives1

δ 2n R 4 4
e−i η̄(x)SF (x−y)η(y) d x d y |η=η̄=0
δη(x2n ) · · · δη(xn+1 )δ η̄(xn ) · · · δ η̄(x1 )
δn Z 
n
= (−i) SF (xn − y)η(y) d4y
δη(x2n ) · · · δη(xn+1 )
Z Z
× SF (xn−1 − y)η(y) d4y · · · SF (x1 − y)η(y) d4y
!
R
−i η̄(x)SF (x−y)η(y) d4x d4y
×e

. (12.16)
n=η̄=0

Now the differentiation with respect to η is performed. Since the whole


expression has to be taken at η = η̄ = 0, only the prefactors of the exponential
can contribute. This gives
X
= (−i)n (−)P SF (x1 − xp2n )SF (x2 − xp2n−1 ) . . . SF (xn − xpn+1 ) , (12.17)
P

where the sum over P runs over all permutations of the indices n + 1, . . . , 2n.
The sign of the permutations is such that (−)P = +, if always the outermost
operators in (11.77) are combined into the 2-point function, i.e. if (1, 2n),
(2, 2n − 1) · · · (n, n + 1) are combined. The Dirac indices, that have not been
written down here, have to be combined in the same way.
Inserting this result into the definition of the Green’s function in (11.77)
gives
 2n
1 δ 2n Z
G(x1 , x2 , . . . , x2n ) =
i δη(x2n ) . . . δη(xn+1 )δ η̄(xn ) . . . δ η̄(x1 )
X
= in (−)P SF (x1 − xp2n ) · · · SF (xn − xpn+1 ) .
P
(12.18)

Equation (12.18) is Wick’s theorem for fermions. Written as a vacuum ex-


pectation value of time-ordered field operators it reads
h i
ˆ (x ) · · · Ψ̄
h0|T Ψ̂(xn ) · · · Ψ̂(x1 ) Ψ̄ ˆ (x
2n n+1 |0i (12.19)
h i h i
= in
X
ˆ (x ) |0i · · · h0|T Ψ̂(x )Ψ̄
(−)P h0|T Ψ̂(x1 )Ψ̄ ˆ (x
p2n n pn+1 ) |0i .
P

1
To facilitate the notation the Dirac indices are not written down here.
CHAPTER 12. INTERACTING FIELDS 150

Wick’s theorem for fermions (12.18) has a form that is analogous to that for
bosons (8.20). The essential difference is the appearance of the sign of the
permutation that reflects the antisymmetric of fermionic states. In addition
the degeneracy factor 1/2n is missing here because for fermions we have no
longer a symmetry under exchange: SF (x − y) 6= SF (y − x) (see (11.75)).
This form also exhibits clearly the important consequence of using anti-
commuting Grassmann fields for the fermions. The exchange of any two of
the coordinates appearing in the fields Ψ or of any 2 in the fields Ψ̄ gives
a “−” sign to the Green’s function. In particular, the 2-particle, 4-point
Green’s function has the property
Gαβγδ (x1 , x2 , y1 , y2 ) = −Gβαγδ (x2 , x1 , y1 , y2 ) = −Gαβδγ (x1 , x2 , y2 , y1 )
= Gβαδγ (x2 , x1 , y2 , y1 ) . (12.20)
This implies that for G not to vanish we must have x1 6= x2 and y1 6= y2 , if
all the Dirac-indices are the same, thus reflecting the Pauli principle.

12.3 Removal of Degrees of Freedom:


Yukawa Theory
In the beginning of Chap. 12 we have first met a coupled fermion-boson theory
with a simple Yukawa coupling. We now consider a theory that contains
free fermions coupled to a self-interacting boson field by means of a Yukawa
interaction. Its Lagrangian reads
1
L = − φ2φ − V (φ) + ψ̄ (i 6 ∂ − mF ) ψ − g ψ̄ψφ
2
1
= − φ2φ − V (φ) + ψ̄ (i 6 ∂ − mF − gφ) ψ . (12.21)
2
The term V (φ) may contain both the mass term and further selfinteraction
terms. Since the coupling constant g here is dimensionless we expect that this
theory is renormalizable (cf. the discussions in Sect. 10.2). The second line of
(12.21) shows that the Yukawa coupling contributes the fermion mass and,
for mF = 0, can even generate the whole mass of a fermion. This mechanism
plays an important role in the theory of electroweak interactions.
In the preceding sections we have discussed the perturbative treatment
of such a Lagrangian. Here we consider now an alternative treatment that is
based on the fact that the generating functional is given by
Z  Z   
Z[η, η̄, J] = N Dψ̄ Dψ Dφ exp i L[ψ̄, ψ, φ] + Jφ + η̄ψ + ψ̄η d4x .
(12.22)
CHAPTER 12. INTERACTING FIELDS 151

It is thus quadratic in the fermion fields and, therefore, of Gaussian form in


the fermion fields. The path integral over the fermionic degrees of freedom
can thus be performed. According to (11.56) and (11.65) this gives
Z R
Z[η, η̄, J] = N Dφ det(M ) e iη̄M −1 η −i
e [ 21 φ2φ−V (φ)−Jφ] d4x (12.23)

with (see (11.66))


M = i 6 ∂ − m∗ with m∗ = mF + gφ ; (12.24)
the factor ‘i’ before M i/n the determinant has been absorbed into the nor-
malization constant N . Note that the fermion determinant M is a function
of φ. Equation (12.23) is completely equivalent to (12.22).
We now concentrate on processes that contain only bosons on the external
legs and thus require only bosonic n-point functions for the calculation of
transition rates etc. We can then set η and η̄ to zero. In doing so we give
up the possibility to calculate any n-point function with fermionic external
legs because these are obtained as functional derivatives with respect to η.
Our procedure thus amounts to keeping only fermionic loops in our theory
for the scalar field φ. More specifically, as we will see below, our procedure
keeps fermionic effects only on the one-loop level.
Using now (6.55)
det(M (φ)) = etr ln M (φ) (12.25)
gives Z R
Dφ e−i [ 2 φ2φ+V (φ)−Jφ] d x+tr ln M (φ) .
1 4
Z[J] = N (12.26)
Equation (12.26) shows that now formally all the fermion degrees of freedom
have been removed. Alternatively, we could also say that we are now working
with a new, effective action for the bosonic sector
Z  
1
SB = − φ 2φ + V (φ) d4x − i tr ln (i 6 ∂ − m∗ (φ)) . (12.27)
2
The trace here has to be performed over Dirac indices and over space-time
coordinates (or, equivalently, in momentum space). The corresponding effec-
tive Lagrangian density reads
1
LB = − φ2φ − V (φ) − i Tr ln (i 6 ∂ − m∗ (φ)) , (12.28)
2
where we have used Z
tr = Tr d4 x ; (12.29)
Tr is a trace over the Dirac (and possibly other internal) indices. In this form,
which is still exact, the original Lagrangian (12.21) is said to be bosonized.
CHAPTER 12. INTERACTING FIELDS 152

12.3.1 Perturbative Expansion


We now develop a perturbative treatment of this bosonized Lagrangian that
uses an expansion around the groundstate of the bosonic sector.
The true groundstate (vacuum) of the bosonic sector is, of course, deter-
mined by the potential V (φ), but it may not coincide with the minimum of
this potential. Nevertheless, we may take here the constant, translationally
invariant field φ0 which minimizes V as our reference field. In order to nor-
malize the action, and thus the physics, to this state we subtract the action
belonging to φ0 . This gives (with m∗0 = mF + gφ0 )
Z  
1
SB = − φ2φ − V (φ) d4x − i tr ln (i 6 ∂ − m∗ ) + i tr ln (i 6 ∂ − m∗0 ) ,
2
(12.30)
where a constant term has been dropped. Next we use the identity
tr ln (i 6 ∂ − m∗ ) = tr ln (i 6 ∂ − m∗ )† = tr ln (−i 6 ∂ − m∗ ) (12.31)
which holds because of the hermiticity of the Dirac operator i 6 ∂ − m∗ ; here
we remember from our discussion in Sect. 6.1.3 that the determinant of an
operator is just the product of its eigenvalues. We thus get
1
tr ln (i 6 ∂ − m∗ ) = tr ln [(i 6 ∂ − m∗ ) (−i 6 ∂ − m∗ )]
2
1  
= tr ln 2 + m∗ 2 − ig 6 ∂φ . (12.32)
2
Thus the effective bosonic action becomes
Z  
1
SB = − φ2φ + V (φ) d4x (12.33)
2
i   i  
− tr ln 2 + m∗ 2 − ig (6 ∂φ) + tr ln 2 + m∗0 2
2 2 !
Z 
m∗ 2 − m∗0 2 − ig 6 ∂φ

1 4 i
= − φ2φ + V (φ) d x − tr ln 1 +
2 2 2 + m∗0 2
and the Lagrangian density now reads
!
1 i m∗ 2 − m∗0 2 − ig 6 ∂φ
LB = − φ2φ − V (φ) − Tr ln 1 −
2 2 −2 − m∗0 2
1 i
= − φ2φ − V (φ) − Tr ln (1 − D(0)W (x)) (12.34)
2 2
with the operator D and the field-dependent quantity W
1
D=− and W = m∗ 2 − m∗0 2 − ig 6 ∂φ . (12.35)
2 + m∗0 2
CHAPTER 12. INTERACTING FIELDS 153

Note that D is nothing else than the Feynman propagator (cf. (6.7)). So far,
no approximations have been made and the theory is still exact; note that
the reference field is in principle arbitrary.
The action is now in a form that is suitable for a perturbative treatment.
2
We thus expand the logarithm (− ln(1 − x) = x + x2 + . . .) in (12.33) and
obtain ∞
X 1
−tr ln(. . .) = tr (DW )n (12.36)
n=1 n
The trace in (12.36) indicates that each term in the expansion contains one
closed fermion loop (with n vertices W coupling to the field φ). Equation
(12.36) is therefore called the one-loop expansion, it represents a perturbative
expansion of the effective action. Integrating out the fermion degrees of
freedom has thus led to an effective theory that takes all fermionic one-loop
diagrams into account. The expansion is expected to converge the more
rapidly the closer the actual field φ is to the chosen field φ0 and the smaller
the gradients of φ are.
Our aim in the following developments is to write the trace term as an
integral over a space-time density which could then be interpreted as a cor-
rection term to the Lagrangian density, thus yielding an effective Lagrangian
which contains the effects of the fermions on the one-loop level. In order to
be able to do so we have to separate the terms under the trace into a product
of operators that are local in x and p. In this case the trace can be separated
Z
d4p
tr [F (p̂)G(x̂)] = Tr hp|F (p̂)G(x̂)|pi
(2π)4
Z
d4p Z d4p0 Z 4 Z 4 0
= Tr dx dx
(2π)4 (2π)4
×hp|F (p̂)|p0 ihp0 |x0 ihx0 |G(x̂)|xihx|pi
Z
d4p Z
= Tr F (p) d4x G(x) . (12.37)
(2π)4

The expressions in (12.36) are, however, not in a form that allows direct
application of (12.37), because the field φ, and thus also W , is dependent on
x and the propagator D acts on it. We thus first reorder each term such that
all propagators are moved to the left. To do so we start from the identity
h i
W D = DW + D D−1 , W D (12.38)

and iterate this equality by applying it again to the quantity W 0 = [D−1 , W ]


h i
W 0 D = DW 0 + D D−1 , W 0 D . (12.39)
CHAPTER 12. INTERACTING FIELDS 154

Combining now (12.38) and (12.39) gives


h i h h ii
W D = DW + D2 D−1 , W + D2 D−1 , D−1 , W D. (12.40)

This procedure can be continued and yields the operator identity


h i h h ii
W D = DW + D2 D−1 , W + D3 D−1 , D−1 , W + ... . (12.41)

The basic commutator appearing here is given by


h i
D−1 , W = [−2, W ] = −2W − 2 (∂ µ W ) ∂µ . (12.42)

Since W contains a derivative of the field this term is of order O((∂µ φ)3 ) and
the next higher order term in (12.41) would be of order O((∂µ φ)4 ). Equation
(12.41) thus represents an expansion in terms of gradients of the field. We
also see that it contains increasing powers of D so that we expect to encounter
only a finite number of ultraviolet, i.e. for large momenta, divergent terms.
We now use the identity (12.41) to rewrite the expansion (12.36). Up to
terms of second order in W the expansion reads
1
−tr ln(. . .) = tr(DW ) + tr(DW DW ) + O(W 3 ) (12.43)
2
1 1  h i 
= tr(DW ) + tr(D2 W 2 ) + tr D3 D−1 , W W + O(W 3 ) .
2 2
Note that all the terms on the rhs have the momentum dependence and the
x-dependence separated so that – according to (12.37) – they can be written
as an integral over the momentum times a spatial integral over a Lagrangian
density. Eq. (12.43) is an expansion in terms of powers of W .
We now discuss the terms in the expansion (12.43) and start with the
term tr(DW ). This is explicitly given by
h  i
tr(DW ) = tr D m∗ 2 − m∗0 2 − ig 6 ∂φ . (12.44)

Since
tr γµ = 0 (12.45)
and   
m∗ 2 − m∗0 2 = g 2mF (φ − φ0 ) + g φ2 − φ20 (12.46)
we get
n h  io
tr(DW ) = g tr D 2mF (φ − φ0 ) + g φ2 − φ20 . (12.47)
CHAPTER 12. INTERACTING FIELDS 155

The trace operation can now be performed. This gives

d4k Z
1 Z
4
h 
2 2
i
tr(DW ) = 4g d x 2m F (φ(x) − φ 0 ) + g φ (x) − φ 0 ,
(2π)4 k 2 − m∗0 2
(12.48)
where we have used (12.37) with the trace over the Dirac indices Tr = 4. This
term is now in the desired form, i.e. it is given by a space-time integral over
fields. However, the momentum-integral appearing here actually diverges
quadratically as power counting shows us.
Also the next term in (12.43) diverges. It is given by

2 2 2
Z
d4k 1
tr(D W ) = g Tr   (12.49)
(2π) k 2 − m∗ 2 2
4
0
Z    2
× d4x 2mF (φ(x) − φ0 ) + g φ2 (x) − φ20 − i 6 ∂φ(x)

and diverges logarithmically.


The next term in (12.43)
 h i  h i
tr D3 D−1 , W W = tr (D)3 (−2W − 2(∂ µ W )∂µ ) W (12.50)

and all higher order terms are clearly ultraviolet convergent because they
contain higher and higher powers of the propagator D.
This means that we have only two divergent terms ((12.48) and (12.49)) in
the expansion (12.36). These can be removed by introducing proper counter
terms. We can thus define a finite, renormalized effective Lagrangian density
in the boson sector of our theory by just adding the two divergent terms
(12.48) and (12.49) to the Lagrangian
1
L̃B (x) = − φ(x)2φ(x) − V (φ(x))
2
i h  i
− Tr ln 1 − D m∗ 2 (x) − m∗0 2 − ig 6 ∂φ(x)
2
Z
d4k 2mF (φ(x) − φ0 ) + g (φ2 (x) − φ20 )
− i2g
(2π)4 k 2 − m∗0 2
g 2 Z d4k 1
−i Tr   (12.51)
4 (2π) k 2 − m∗ 2 2
4
0
   2
2
× 2mF (φ(x) − φ0 ) − i 6 ∂φ(x) + g φ (x) − φ20 .
CHAPTER 12. INTERACTING FIELDS 156

Writing this expression in a somewhat more compact form gives


1 i
L̃B = − φ(x)2φ(x) − V (φ(x)) − Tr(1 − DW )
2  
2
i 1 2
− Tr DW + (DW ) (12.52)
2 2
1 1  3 h −1 i 
= − φ(x)2φ(x) − V (φ(x)) + Tr D D , W W ,
2 2
where Tr denotes a trace over the Dirac indices only and where it is un-
derstood that we have to use the expansion (??) for the Tr ln term. The
two counter terms have the structure of terms already present in the original
boson Lagrangian; they just remove the first two terms in the Taylor expan-
sion of the logarithmic term in (12.34). The Yukawa theory is thus indeed
renormalizable, provided the potential V (φ) is ‘well-behaved’ and contains
at most terms of fourth order in φ.
Chapter 13

PATH INTEGRALS FOR


GAUGE FIELDS

The fundamental interactions of nature like, e.g. the electroweak and the
strong interactions are described by so-called gauge field theories. In these
theories the action is invariant under certain space-time dependent unitary
“gauge” transformations. As a consequence for each field configuration there
are infinitely many others that can all be obtained by a gauge transformation
from the original one. The physics remains unchanged under such transfor-
mations.
Special problems arise when such gauge theories are quantized. The rea-
son is quite easy to see in the framework of path integrals. If we consider a
field Aµ , the path integral would be naively written down as
Z
DA eiS[Aµ ] (13.1)
where the action S is a functional of Aµ . The integration measure stands for
a straightforward generalization of the measure used for scalar fields
n Y
YY 4
DA ∼ dAµ (~x, tj ) (13.2)
x j=1 µ=1
~

in the limit n → ∞ (cf. (5.11)); the measure so defined is gauge invariant


as we will see later in this chapter. For a fixed field Āµ there are infinitely
many other fields, connected to Āµ by gauge transformations, that leave the
action S invariant and thus all give the same contribution to the integrand.
Thus the path integral as written down here cannot converge.
In this chapter we develop methods to deal with this difficulty. After a
discussion of gauge-invariance in electrodynamics we summarize the proper-
ties of a more general class of gauge-field theories and then develop the path
integral description for them.

157
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 158

13.1 Gauge invariance in Abelian theories


We start this section with a discussion of gauge invariance in electrodynamics
by considering a free field configuration without currents.
For the electromagnetic field the free Lagrangian is given by
1 1
L = − Fµν F µν = − (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) (13.3)
4 4
and is clearly invariant under a gauge transformation
Aµ (x) −→ A0µ (x) = Aµ (x) + ∂µ (x) . (13.4)
Here (x) is an arbitrary differentiable function of space-time. The corre-
sponding equation of motion for the fields Aµ is
∂ν F νµ = ∂ν (∂ ν Aµ − ∂ µ Aν ) = (δνµ 2 − ∂ν ∂ µ )Aν = 0
−→ (gµν 2 − ∂µ ∂ν ) Aν = 0 , (13.5)
which is also gauge invariant. If we now fix a gauge, e.g. the Lorentz gauge
∂µ Aµ = 0, we obtain the free wave equation for Aµ
2Aµ = 0 , (13.6)
which is no longer gauge invariant.
We now proceed to obtain the propagator for the electromagnetic field.
As mentioned at the end of section 11.2.2 the propagator of the theory could
be read off from the term in the Lagrangian that is quadratic in the field.
The Lagrangian (13.3) can indeed also be converted into a form quadratic in
the fields by writing
Z
4 1Z 4
Ld x = − d x (∂ µ Aν − ∂ ν Aµ )(∂µ Aν − ∂ν Aµ ) (13.7)
4 Z
1
= − d4x [ (∂ µ Aν )(∂µ Aν ) − (∂ µ Aν )(∂ν Aµ )
4
− (∂ ν Aµ )(∂µ Aν ) + (∂ ν Aµ )(∂ν Aµ )] .
By partial integration we obtain
Z
4 1Z 4
Ld x = d x (Aµ 2Aµ − 2Aµ ∂ν ∂µ Aν + Aµ 2Aµ ) (13.8)
4
1Z 4 µ
= d x A (gµν 2 − ∂µ ∂ν ) Aν ,
2
so that L can also be written as
1
L = Aµ (gµν 2 − ∂µ ∂ν ) Aν . (13.9)
2
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 159

This Lagrangian is very similar to that of a scalar field theory (4.31) so


that we expect the generating functional to have the form
R
d4x L
Z[0] = ei . (13.10)

Since the Lagrangian is quadratic in the fields it can be treated in analogy


to the developments in Sect. 6.1.3. According to our normal procedure (cf.
(11.87)) we thus expect to obtain the propagator of the electromagnetic field
as
−1
Dµν ∼ gµν 2 − ∂µ ∂ν . (13.11)
The problem arises when we realize that this operator D does not exist.
We see this by applying its inverse (13.11) to an arbitrary four-gradient ∂ ν G
−1 ν
Dµν ∂ G = (2∂µ − ∂µ 2) G = 0 . (13.12)

Thus, D−1 has a zero eigenvalue and, loosely speaking, its inverse D is in-
finite. This infinity is linked to that in the path integral over gauge fields
discussed at the start of this chapter. This can be seen by a straightforward
application of (B.18) to (13.10) which gives the equivalent of (6.33) with
det(A) = 0 in the denominator1 . In other words: we are integrating over too
many degrees of freedom when the gauge condition is not taken into account.
Therefore, we expect that one way to overcome this difficulty is to fix a
gauge, for example by imposing the Lorentz gauge condition

∂ ν Aν = 0 . (13.13)

If we then consider only potentials that fulfill this condition and integrate
only over them, we expect the path integral to be well-behaved.
This expectation is based on the fact that we can find a Lagrangian that
incorporates the gauge condition (13.13) at the price of loosing its manifest
gauge invariance
1
L̃ = − Fµν F µν + LGF
4
1 1
= − Fµν F µν − (∂µ Aµ )2 . (13.14)
4 2λ
This Lagrangian, that leads for λ = 1 to the wave equation (13.6) as we will
show below, contains a so-called gauge-fixing term as a quadratic constraint,
1
coupled in by means of a Lagrange-multiplier ( 2λ ). This additional term
1
Remember here that the determinant of an operator is given by a product of its
eigenvalues.
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 160

could be considered as a potential that becomes the more repulsive the more
the actual potential Aµ differs from the Lorentz-gauge (13.11); the constraint
thus drives the system towards that gauge.
Performing the steps leading to (13.9) once again, we find that (13.14) is
equivalent to the Lagrangian
   
1 1
L̃ = Aµ gµν 2 − 1 − ∂µ ∂ν A ν (13.15)
2 λ
which leads to the same action. The equations of motion following from
(13.15) are    
1
gµν 2 − 1 − ∂µ ∂ν A ν = 0 . (13.16)
λ
We can thus either start from a manifestly covariant Lagrangian, derive the
equations of motion from it and then impose a suitable gauge condition or we
can, alternatively, incorporate the gauge condition into the Lagrangian, at
the price of giving up its gauge invariance, and then obtain the gauge-fixed
equations of motion directly from it. The physics must obviously be the same
in both methods.
Reading again – as at the end of Chap. 11 on p. 143 – the inverse propa-
gator off from the Lagrangian (13.15) gives
 
−1 1
Dµν = gµν 2 − 1 − ∂µ ∂ν (13.17)
λ
with the momentum space representation
 
−1 2 1
Dµν (k) = −gµν k + 1 − kµ kν . (13.18)
λ
This operator no longer has a zero eigenvalue as we can easily see by perform-
−1 νλ
ing again the steps in (13.12). The propagator D, defined by Dµν D = δµλ ,
2
thus exists. In momentum space it is given by
!
1 kµ kν
Dµν (k) = − 2 gµν + (λ − 1) 2 . (13.19)
k k

This is the propagator for the electromagnetic field.


The constant λ in (13.19) is arbitrary, but note that the propagator does
not exist for λ → ∞; this is just the case of the unconstrained Lagrangian
discussed earlier in this section (cf. (13.3) and (13.14)). Choosing λ = 1
2
The denominator k 2 from here on should really read k 2 − iε in order to ensure the
proper boundary conditions for the propagator.
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 161

gives the so-called Feynman gauge, whereas λ = 0 describes the Landau-


gauge. The Feynman gauge is the one that we are used to from classical
electrodynamics; in it the equations of motion for the electromagnetic field
(13.16) read
2Aν = 0 . (13.20)
This is the equation of motion that we obtained earlier for the Lorentz gauge.

Generating functional. Now that we have constructed a propagator for


the electromagnetic field, we can write down the generating functional as
Z R
(L+LGF +Jµ Aµ )d4x
Z[J] = DA ei (13.21)
Q4
where DA = µ=1 DAµ and LGF is the gauge-fixing term
LGF = −1/(2λ)(∂µ Aµ )2 . (13.22)
The gauge-fixing term suppresses the contributions of fields that do not fulfill
the Lorentz gauge condition to the path integral.
We will see a little later, in section 13.3, that this functional can actually
be rewritten into a form that exhibits explicitly the integration over the gauge
degree of freedom. For a free field theory the generating functional can then
be obtained as usual (cf. (B.18))
R
− 2i J µ (x)Dµν (x−y)J ν (y) d4 xd4 y
Z[J] = e , (13.23)
and the two-point function is, as usual (see (11.86), (11.87)), found to be
Gµν (x, y) = iDµν (x − y) (13.24)
The path integral (13.21) does not suffer from the problems mentioned at
the start of this section. Now the matrix in the quadratic term of L (13.9)
can be inverted and the propagator exists. Furthermore, all fields that differ
from a specific field just by a gauge transformation that leads out of the
covariant Lorentz gauge have a higher action and thus contribute less to the
path integral.
The action itself is gauge-independent, whereas the integral over the
gauge-fixing term and the source terms do depend on the gauge. Thus, the
generating functional itself is gauge dependent and, consequently, also the
Green’s functions obtained from it by functional derivatives. All of the gauges
corresponding to the various values of λ, however, lead to the same physics
because in calculating a transition amplitude Dµν will couple to conserved
currents only which fulfill ∂µ j µ = 0 → kµ j µ (k) = 0. As a consequence the
λ-dependent term in (13.19) always drops out and the physically observable
transition amplitudes are all gauge independent.
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 162

13.2 Non-abelian gauge fields


In more general gauge field theories one demands invariance of the Lagrangian
under the transformation of the particle fields (see [MOSEL], chapter 7)

ψ(x) → ψ 0 (x) = U (x)ψ(x) . (13.25)

The particle fields are described by N -component spinors ψ where the com-
ponents represent the internal degrees of freedom. Here U is a unitary, local
gauge transformation
k k
U (x) = e−i (x)T (13.26)
where the T k are the generators of a Lie group. They are Hermitian so
that the Lie group is unitary and have vanishing trace so that the generated
Lie algebra is special, i.e. all group elements are represented by (N × N )
matrices that have unit determinant. The latter implies that the matrix
representations of the generators T k have vanishing trace. The generators
form a Lie algebra with the defining commutation relations

[T l , T m ] = if lmn T n (l = 1, 2, . . . N 2 − 1) . (13.27)

The generators are normalized such that


  1
tr T l T m = δ lm (13.28)
2
in the fundamental representation; the f lmn in (13.27) are the completely
antisymmetric structure constants of the group. Another important matrix
representation of the generators is the so-called regular representation
 kl
Tj = −if jkl . (13.29)

The special unitary groups in N dimensions with these properties are


called SU (N ). Since the generators do not commute the group SU (N ) is a
non-Abelian one. Only for the case N = 1 there is only one generator and
the group becomes Abelian, called U(1).
Invariance under the transformation (13.25) can be achieved only if the
derivative ∂µ is replaced by the so-called covariant derivative

Dµ = ∂µ + igT l Alµ (13.30)

where g is the coupling constant and Alµ is a vector (gauge) field; the super-
script l labels the internal degrees of freedom. This “minimal substitution”
fixes the interaction between particle and gauge fields.
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 163

The gauge transformation of these fields must then be given by


i
Aµ → AUµ = U Aµ U −1 − U ∂µ U −1 , (13.31)
g
if the Lagrangian in the particle sector is required to be invariant. Here we
have introduced the fields Aµ in (13.31) as contractions of the generators
with the fields Akµ
Aµ = T k Akµ . (13.32)
The change of A under an infinitesimal gauge transformation is given by
1  jl 1
δAjµ = − l f ljk Akµ + ∂µ j = il T k Akµ + ∂µ j
g g
1
= (Dµ )j (13.33)
g
kl
where we have used the regular representation of SU (N ), i.e. (T j ) = −if jkl .
Also for non-Abelian gauge fields the free Lagrangian is given by
1 l lµν
L = − Fµν F , (13.34)
4
but now the field tensor also carries the quantum numbers of the internal
gauge group and must have a more complicated structure
l
Fµν = ∂µ Alν − ∂ν Alµ − gf lmn Am n
µ Aν (13.35)

in order to be invariant under the gauge transformation (13.31). Gauge


invariance also requires that the gauge fields are massless, since a mass term
of the form m2 A · A would clearly violate gauge invariance.
l
The field tensor Fµν can also be contracted with the generators to form
l
a scalar in the intrinsic space (Fµν = Fµν T l ). These scalars change in the
gauge-transformation just as under any unitary transformation
U
Fµν → Fµν = U Fµν U −1 . (13.36)

Equation (13.34) together with (13.35) show that the free gauge field is
selfinteracting, with terms ∼ A3 and ∼ A4 in L . These terms are generated
by the non-Abelian piece of Fµν and are absent for an Abelian theory (f lmn =
0). Notice also that the coupling constant g, determining the interaction
strength, is the same for gauge field–gauge field (13.35) and gauge field–
particle (13.30) interactions.
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 164

The complete ‘generic’ Lagrangian of a non-Abelian gauge field theory


with many distinct fermion fields then reads
1 l l µν X h i
L = − Fµν F + ψ̄f (iγ µ Dµ − mf )ψf . (13.37)
4 f

The sum over f runs over all fermions. The fact that the coupling constant
appears both in the fermion–gauge field coupling and in the gauge field–gauge
field coupling has an important consequence: since there is a common gauge
field for all fermions, all these different fermions must couple with the same
g to the gauge fields. This universality is a special property of non-Abelian
gauge field theories.
An example for the relevance of a non-Abelian gauge field theory is pro-
vided by Quantum Chromodynamics (QCD), the theory of the strong in-
teractions between quarks. Here the relevant internal degree of freedom of
the quarks is color and the symmetry group is SU (3). Another example is
provided by the theory of the electroweak interaction where the gauge group
is U (1) × SU (2); in this case the relevant degrees of freedom are the electrical
charge (U (1)) and the weak isospin (SU (2)). Both of them will be discussed
in some more detail in Chapt. 14.

Abelian gauge field theories. For the case of QED U (x) = e−i(x) , T k =
1 and f lmn = 0; thus U is Abelian. Indeed (13.31) then reduces to the well
known gauge transformation of electrodynamics
1
Aµ → A0µ = Aµ + ∂µ  (13.38)
g
and also the field tensor assumes its well known form. The case of QED is
thus contained as a special case in the developments in this section. Since in
this case there is no coupling constant g in the field tensor F , universality
does not hold for non-Abelian gauge field theories.

13.3 Path integrals


We now want to develop a path integral formulation also for non-Abelian
fields. We start again with the naive expression for a path integral over
gauge fields Z
Z[0] = DA eiS[A] . (13.39)
Here the action is that of the original theory without any gauge-fixing term
Z
S= d4x L (13.40)
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 165

and the integration measure DA stands for


2
Y Y NY 4
−1 Y
DA = d Alµ (~x, tj ) (13.41)
~
x j l=1 µ=1

(cf. (5.11)). This integral, of course, diverges as in the case of the Abelian
theory because, as discussed in the last section, the path integral runs over
all field configurations, i.e. both over essentially distinct ones and those that
differ only by gauge transformations from each other. It is then tempting to
just add a gauge-fixing term to L as we have done in (13.21).
While this problem of the gauge-equivalent potentials appears for both
Abelian and non-Abelian gauge field theories, there is another difficulty that
is specific for non-Abelian theories. The path integral in the form given
above, in which an integration only over the fields appears, was shown in
chapter 1 to be valid only for quadratic Hamiltonians with constant coeffi-
cients in front of the kinetic term. This, however, is no longer the case for
a non-Abelian gauge theory, where the field tensor F depends not only on
the derivatives of the fields, but also on the fields themselves. Effectively
this leads to a kinetic energy term in which the coefficient of the momentum-
dependent terms depends on the fields themselves. When considering the
nonrelativistic analogon of this case in section 1.3.2 we have started from the
Hamiltonian formulation of the path integral. There we have found that a
new factor involving the squareroot of the coordinate dependent coefficient
appears under the path integral (see (1.54)). It is therefore natural that in
the treatment of non-Abelian gauge field theories we would also have to start
from the Hamiltonian representation of the path integral (1.32) (see [Lee] for
a detailed treatment). At the expense of some mathematical rigor, however,
the concept of a path integral over the fields only can be maintained and this
is what we are going to do in this chapter.
In the following we will now develop a method to split the path integral
up into two factors, one containing the divergent part and the other one being
convergent. The divergent part will be seen to be an infinite constant that
drops out when we work with the normalized functional. The divergence is
here – as in the Abelian case – connected with the presence of zero eigenvalues
of the matrix in the quadratic term of the gauge field action. It is just those
zero eigenvalues that we have to factor out by integrating only over those
fields that correspond to nonzero eigenvalues. We start by dividing up the
total configuration space of all possible Aµ (x) into equivalence classes. Each
of these equivalence classes contains all the possible fields AUµ that can be
obtained from an initial field Aµ (x) by gauge transformations. The integrand
of (13.39) is constant along the fields AUµ , i.e. in an equivalence class. This
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 166

means that the path integral (13.39) is proportional to an infinite constant


which is just the “volume” of the gauge group U . The physically relevant
part of the path integral then just runs over the essentially distinct fields
Aµ (x). This restriction can be achieved by requiring that the fields that
contribute to the integration all fulfill a gauge condition which we write in
the form
F l (Akµ (x)) = 0 . (13.42)
Examples of such gauge conditions are given by the
Lorentz gauge Fk = ∂ µ Akµ = 0
Coulomb gauge Fk =∇~ ·A ~k = 0
(13.43)
Axial gauge Fk = Ak3 = 0
Temporal gauge Fk = Ak0 = 0 .
The challenge is now to include these conditions in the path integrals such
that gauge-invariance of the theory is maintained.
With this aim in mind we now define a functional integration over the
elements of the symmetry group SU (N ), determined by the transformation
kT k
U (x) = e−i , (13.44)

and denote the integration measure by DU ; in Fig. 13.1 this integration over
the gauge group corresponds to an integration along the solid lines. This
integration measure is gauge invariant

DU 00 = D(U 0 U ) = DU (13.45)

because of the group-property of the transformations U ; this can also be seen


by noting that we integrate over all gauge transformations U . To factor out
an integral over DU is the aim in the following paragraphs.
With this integration measure we now consider the integral
Z
U
M−1 [Alµ ] = DU δ[F l (Akµ )] (13.46)

Here F l (Akµ ) = 0 is a gauge-fixing condition (13.42). The superscripts l, k


stand for internal degrees of freedom of the symmetry group SU (N ); for
notational convenience, we will write them out only when necessary. The
δ-functional in (13.46) is really a product of Dirac δ-functions at each space-
time point (~x, tj ). AU is the gauge-transformed field. Since the δ-functional
in (13.46) restricts the gauge transformations to U ≈ 1 it suffices to consider
only infinitesimal gauge transformations

U (x) ≈ 1 − il (x)T l with   1 . (13.47)


CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 167

In this case the measure can be written as


2
Y NY−1
DU = dl (~x, tj ) (13.48)
x,j l=1
~

in a discrete space-time representation. For the measure over infinitesimal


transformations in (13.48) we have
l l
U 00 = U 0 U ≈ (1 − i0 T l )(1 − il T l ) ≈ 1 − i(0 + l )T l
l
= 1 − i00 T l , (13.49)

so that the Jacobian from  to 00 = 0 +  is


!
δl
det =1. (13.50)
δ00 k
Thus the measure (13.48) is indeed gauge-invariant.

AUµ


Aµ(x)

Figure 13.1: Lines of constant integrands. The horizontal axis contains the
physically distinct fields, the vertical one the gauge-transformation degree of
freedom. The solid lines depict the fields within a given equivalence class;
the dashed line represents the gauge condition F (AUµ ) = 0 for an Abelian
theory, the dotted line for a non-Abelian theory exhibits the appearance of
Gribov copies for large fields.

The condition F (AUµ ) = 0 then defines a surface that crosses that of each
equivalence class (dashed line in Fig. 13.1); we assume that for an initial field
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 168

Aµ this is the case for one group element. This is certainly the case for an
Abelian gauge field theory, like e.g. QED. There, the gauge transformation
U reads (cf. (13.38))
1
Aµ (x) −→ AUµ (x) = Aµ (x) + ∂µ (x) (13.51)
g
and the gauge condition becomes
1
F (AUµ ) = ∂µ Aµ (x) + ∂µ ∂ µ (x) = 0 (13.52)
g
where we have chosen as an example the covariant gauge ∂µ Aµ = 0. Eq.
(13.52) represents an inhomogeneous wave equation for . With the boundary
condition Aµ (x) → 0 as |x| → ∞ it specifies a unique (x). As Gribov has
pointed out this is no longer the case if we are dealing with a non-Abelian
field theory. In this case there can be several equivalent fields and all fulfill
the same gauge condition (open points in Fig. 13.1). Thus, in this case one
also has to remove these so-called Gribov copies from the path integration.
However, since these copies appear only at large field strengths we can neglect
them in a perturbative treatment around either a vanishing or a classical field
configuration.
The integral (13.46) is gauge invariant. This can be seen by writing the
0
Jacobian as an integral and starting from a gauge-field AU
Z    
0 0 U
M −1
[AUµ ] = DU δ F AUµ . (13.53)

We now change the variables from U to U 00 = U U 0 and get, because the


integration measure is gauge-invariant,
Z h  i
0 00
M−1 [AUµ ] = DU 00 δ F AUµ = M−1 [Aµ ] , (13.54)

since U 00 is just an integration variable that could be renamed into U . M−1


is thus indeed gauge-invariant.
We can now transform the original path integral (13.39) by inserting the
1 from (13.46). This gives
Z Z Z h  i
Z[0] = DA eiS[A] = DA M[Aµ ] DU δ F AUµ eiS[A] (13.55)
| {z }
=1

with the action Z


S= d4x L . (13.56)
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 169

We now perform a gauge transformation AUµ → Aµ . Since S, M and DA


are invariant under this transformation the integrand no longer depends on
U and we can take the integral over U out of the remaining integral
Z Z
Z[0] = DU DA M[Aµ ] δ [F (Aµ )] eiS . (13.57)

This functional determines our theory in the F (Aµ ) = 0 gauge. The integral
Z
DU (13.58)

is just the infinite factor we wanted to pull out from the expression.
Since this factor is cancelled when normalizing the functional, we can
from now on work with the generating functional
Z h  i R
d4x (L+Jµl Al µ )
Z[J] = DA M[Aµ ] δ F l Akµ ei . (13.59)

This functional can be used for a perturbation theoretical treatment because


it is finite. This was not possible for the original functional (13.39) which
diverges because of the integration over physically equivalent gauge fields.

13.3.1 Gauge Fixing


For a derivation of the perturbation theory rules, i.e. the Feynman rules, the
functional (13.59) is still not easy to handle because of the presence of the
gauge-fixing δ-functional. As a guideline for a simplification we can use the
easier example of an Abelian Theory, treated in Sect. 13.1. There we had
succeeded in (13.21) to write down a meaningful generating functional that
did not contain such a δ-functional, but instead a gauge-fixing term in the
Lagrangian.
Taking this result as guideline we, therefore, in the following paragraphs
reformulate Z[J] into a more tractable form. To do this we consider some-
what more specialized gauge conditions

F m (Aµ (x)) − C m (x) = 0 , (13.60)

where C m (x) is an arbitrary function of space-time. Note that for given A,


C m and F m this condition determines a gauge-transformation. The definition
of M[A] changes correspondingly
Z h i
−1
M [Aµ ] = DU δ F m (AUµ (x)) − C m (x) . (13.61)
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 170

Suppose, we looked at a field for which the argument of the δ-functional


with given C m (x) vanishes and then looked at this argument again, but with
a different function C̃ m (x), so that the argument no longer vanishes. We
could then find a gauge transformation that makes the argument of the δ-
functional zero again. Since the rest of the integral is invariant under gauge
transformations (and we integrate anyway over all gauge transformations)
the value of the integral does not change. Thus, the FP determinant M does
not change under this replacement and is even independent of the specific
function C m (x)). Then, obviously, also
Z R
LJ d4 x
Z[J] = DA M[Aµ ] δ [F m (Aµ ) − C m (x)] ei (13.62)

(with LJ = L + J · A) is independent of C(x). We are thus free to change


the integrand by a weighting factor
i
R
[C m (x)]2 d4x
e− 2λ (13.63)

and integrate functionally over the function C; this will only change the
normalization of the integral. We then get
Z Z R
DA M[Aµ ] δ [F m (Aµ ) − C m (x)] ei (LJ − 2λ (C(x)) ) d x .
1 2 4
Z[J] = DC
(13.64)
Now the functional integration over C can be performed since C appears
in the δ-functional. This gives
Z R
DA M[Aµ ] ei (LJ − 2λ [F (Aµ )] ) d x .
1 m 2 4
Z[J] = (13.65)

In this form L has again picked up a ‘penalty potential’ (see discussion after
(13.14) on p. 159)). This generating functional is finite due to the presence
of the gauge fixing term and can, therefore, be used for a derivation of the
Feynman rules for perturbation theory. As we have seen in the last section
for the case of QED the divergence of the original path integral was linked
to the zero eigenvalues of the matrix in the quadratic part of the gauge-field
action so that no one-particle propagator could be defined. The vanishing
eigenvalues were connected to the infinite integration over the gauge trans-
formation degree of freedom. Here now this infinity has been removed by
fixing the gauge and the propagators are finite.
Notice that the appearance of the gauge-fixing term in the action is not
the only change. In a non-Abelian gauge-field theory an additional factor,
the FP determinant M, appears in the integrand of the path integral. In
non-Abelian theories the FP determinant in general depends on the fields and
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 171

thus cannot be pulled out of the path integral. The deeper reason for this
difference of Abelian and non-Abelian theories lies in the different structure
of the field tensors in both theories. Due to the presence of the selfinterac-
tion term in the non-Abelian field tensor (13.35) the kinetic part of the field
energy becomes field-dependent and fields and conjugate momenta no longer
decouple in the Hamiltonian. This is then exactly the situation that we have
encountered already in Sect. 1.3.2 in the framework of nonrelativistic quan-
tum mechanics where a coordinate dependent term also led to an additional
factor in the path integral (cf. (1.54)).
While everything else in the integrand of the path integral (13.65) is gauge
invariant, the source term and the gauge-fixing terms break this invariance.
Thus, the specific rules of the perturbation theory, e.g. the Feynman rules,
will depend on the gauge. In particular also the Green’s functions obtained
as functional derivatives of Z[J] are not gauge invariant. This, however, is
no problem since these Green’s functions only appear at intermediate stages
of the calculation. The S matrix elements, which are physical quantities, are
then gauge invariant again. We have seen an example for that in the last
section where we discussed the QED propagators; in calculating the S ma-
trix the gauge-dependent propagators are always contracted with conserved
currents and the gauge-dependent terms then drop out.

13.4 Feynman Rules


13.4.1 Faddeev-Popov Determinant
It now remains to evaluate M−1 [Aµ ]. In order to understand the meaning of
the integral in (13.46) better and to be able to actually evaluate it we write
it for n discretized space coordinates3
Z
M−1
n = d1 d2 . . . dn δ n [F (Aµ )] (13.66)

with n = (xn ). In order to evaluate this integral we now change the inte-
gration variables from dn to dF . This gives
d1 d2 . . . dn det Mij = dF (x1 ) dF (x2 ) . . . dF (xn ) , (13.67)
where det M is the Jacobian for this transformation of the integral measure.
We thus get
Z
M−1
n = dF (x1 ) dF (x2 ) . . . dF (xn ) δ n [F ] (det Mij )−1 . (13.68)
| {z }
=1
3
To facilitate the notation we drop here the group indices.
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 172

The integral is identically equal to 1 so that Mn is just the Jacobian


!
δF (xi )
Mn = det Mij = det . (13.69)
δ(xj ) F =0

We now generalize these equations to the continuum case. Since M[Aµ ]


is gauge invariant we can always choose a field Aµ that fulfills the condition
F (Aµ ) = 0. This allows us to replace the condition F = 0 for the derivative
in (13.69) by the simpler condition U = 1. We thus have to consider only
infinitesimal gauge transformations

U ≈ 1 − il T l (13.70)

for the evaluation of the derivative. The gauge condition F (A) can then be
Taylor-expanded
Z
F m (AUµ (x)) = 0 + d4y M ml (x, y)εl (y) + O(ε2 ) (13.71)

with  
δF m AUµ (x)
M ml (x, y) = |=0 . (13.72)
δl (y)
We thus get for M−1 at a field close to one fulfilling the gauge condition
Z h  i Z Z 
M−1 [Aµ ] = DU δ F m AUµ = DU δ d4y M ml (x, y)l (y) . (13.73)

For these integrations close to U = 1 we can use the integration measure


(13.48) and obtain (see (13.69))
" ! #−1
−1
h
ml
i−1 δF m (Aλ (x))
M [Aµ ] = det M (x, y) = det . (13.74)
δl (y) =0

The determinant here, the so-called Faddeev-Popov (FP) determinant, has


to be calculated with respect to both the space-time indices (x, y) and the
N 2 − 1 SU (N ) group indices (m, l). Note that the FP determinant depends
in general on the gauge fields Akµ .
We can now replace M[Aµ ] in (13.59) by the Faddeev-Popov determinant
(13.74) and obtain as the generating functional
Z   R
LJ d4x
Z[J] = DA det M ml (x, y) δ[F l (Aµ )] ei (13.75)

with LJ = L + J · A. This generating functional is the central result of this


chapter. The presence of the FP-determinant in (13.75) is a typical feature
of non-Abelian gauge field theories.
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 173

13.4.1.1 Explicit forms of the FP determinant


Abelian fields. We have seen earlier that the non-Abelian gauge field the-
ory contains the Abelian U (1) symmetry group as a special case. We therefore
start by analyzing M for an Abelian gauge field theory such as QED. We
obtain from the definition (13.72)
   
δF AUµ (x) δF Aµ (x) + g1 ∂µ (x)
M (x, y) = = . (13.76)
δ(y)
δ(y)

F =0 =0

Choosing, for example, the covariant gauge ∂µ Aµ = 0 gives

1 2δ(x) 1
M (x, y) = = 2δ 4 (x − y) . (13.77)
g δ(y) g

Thus, M is independent of Aµ , and therefore its determinant can be pulled


out of the path integral in (13.75), changing only the normalization. The
generating functional thus reduces to
Z R
d4 x L J
Z[J] = DA δ [F (Aµ )] ei . (13.78)

In this form the functional contains the free Lagrangian without any gauge-
fixing terms. It is nevertheless finite because of the presence of the δ-
functional. Note the difference of this functional to that written down in
(13.21); we will discuss this difference in the next Section.

Non-Abelian fields. For a non-Abelian field the first term in the gauge
transformation (13.31) has a more complicated form so that it will in general
contribute an Aµ -dependent factor. We start by again expanding F (AUµ )
around a field that fulfills the gauge condition, i.e. F m (Aµ ) = 0 and U ≈ 1
(cf. (13.71)). Under Aµ → AUµ = Aµ + δAµ the condition F changes into

m δF m (Aλ (x)) k
Z
F (AUµ ) = 0+ d4z δAµ (z)
δAkµ (z)
1 Z 4 δF m (Aλ (x))
= dz (Dµ (z))k (13.79)
g δAkµ (z)

Here we have used (13.33) to replace δA with the covariant derivative

Dµ = ∂µ + igT n Anµ . (13.80)


CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 174

Taking now the derivative according to (13.72) we get



ml δF m (Aλ (x))
M (x, y) =
δl (y)
=0
Z m
1 δF (Aλ (x)) δ
= d4z k l
(Dµ (z))k
g δAµ (z) δ (y)

1 Z 4 δF m (Aλ (x)) kl 4


= dz D µ (z)δ (z − y) . (13.81)
g δAkµ (z)
F =0

This is a more general form for the FP-matrix. Its actual value depends on
the gauge used.

Lorentz gauge. The Lorentz gauge is given by

∂ µ Anµ = 0 . (13.82)

The FP-matrix is then obtained from (13.81) as


 
λ m
1 Z 4 δ ∂ Aλ (x) kl

ml 4

M (x, y) = dz k
Dµ (z)δ (z − y) (13.83)
g δAµ (z)
F =0

1 Z 4 mk  µ 4 
∂x δ (x − z) Dµkl (z)δ 4 (z − y)

= dzδ .
g F =0

By using ∂xµ δ 4 (x − z) = −∂zµ δ 4 (x − z) and partial integration we obtain



ml 1 Z 4 mk 4
M (x, y) = d z δ δ (x − z)∂zµ Dµkl (z)δ 4 (z − y)
g F =0

1  µ ml 
= ∂x Dµ (x)δ 4 (x − y) . (13.84)
g F =0

Since we have to evaluate this expression for F m = ∂ µ Am


µ = 0 we can inter-
change the order of the two derivatives and obtain
1  ml µ 4 
M ml (x, y) = Dµ ∂x δ (x − y) . (13.85)
g
For the case of QED this expression indeed reduces to the form already found
earlier (see (13.77)). This is so because in this case δAµ = g1 ∂µ  in (13.79)
and thus in this case instead of Dµ simply ∂µ appears in (13.85).
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 175

Axial gauge. It is interesting to note that the FP-determinant is inde-


pendent of the fields Aµ not only for Abelian gauge theories, but also in a
non-Abelian theory, if the so-called axial gauge is chosen. This axial gauge
is defined by the constraint
n·A=0 , (13.86)
where n is a fixed four-vector; this form comprises the last two cases in
(13.43). In this case we obtain from (13.81)
 
1 Z δ nλ A m
λ (x)


M ml (x, y) = d4z kl 4
Dµ (z)δ (z − y)
g δAkµ (z)
F =0

1 mk µ Z 4 4  
d z δ (x − z) δ kl ∂µz + ig (T n )kl Anµ (z) δ 4 (z − y)

= δ n
g F =0
1 ml µ x 4
= δ n ∂µ δ (x − y) , (13.87)
g
because F = nµ Anµ = 0. Thus in this case the FP-determinant is again
independent of Aµ so that it can, as in the Abelian case, be pulled out of the
path integral and put into the normalization.

13.4.2 Ghost fields


We now continue with the discussion of the general non-Abelian case (13.65).
The trick now is to use (11.56) to express the Faddeev-Popov determinant by
another path integral over (hypothetical) Grassmann fields η(x). We have
Z R
η̄ m (x)M ml (x,y)η l (y) d4xd4y
det(iM ) = Dη̄ Dη e−i . (13.88)

Here the Grassmann fields η and η̄ carry the indices of the internal symmetry
group SU (N ). The generating functional of the theory then becomes (cf.
(13.65))
Z R
DA det(M ) ei (LJ − 2λ [F (Aµ )] )d x
1 m 2 4
Z[J] =
Z R R
DA Dη̄ Dη ei (LJ − 2λ [F (Aµ )] − η̄M η d4y )d4x
1 m 2
= . (13.89)

If we remember that the FP determinant owes its existence to the gauge


fixing condition, then we see that the introduction of the fields η(x) amounts
to a dynamical treatment of the gauge invariance. We therefore expect that
this treatment generates also new Feynman graphs involving the fields η(x).
In order to be able to calculate their Green’s functions we now supplement
this generating functional with sources also for the so-called ”ghost field” η
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 176

and obtain for the generating functional of the full theory in the gauge-ghost
sector Z R eff 4
¯
Z[J, ξ, ξ] = DA Dη̄ Dη ei LJ d x (13.90)
with
1 m
Leff
J (x) = L(x) − [F (Aµ )]2 (13.91)
Z 2λ
− η̄(x)M η(y) d4y + Jµn (x)Anµ (x) + ξ¯n (x)η n (x) + η̄ n (x)ξ n (x) .

The second term here is the gauge-fixing term. As stressed earlier the gauge-
fixing term breaks the gauge-invariance and thus causes the path integral to
be well defined. Note that this gauge fixing term is not enough to specify
the gauge for non-Abelian theories. In these the FP determinant appears in
addition and manifests itself in the presence of the fictitious “ghost field”,
introduced in the third term in (13.91). Since the ghosts carry the indices
of the underlying symmetry group SU (N ) there are as many ghost fields as
there are gauge fields. Because in general M = M (A) they couple to the
gauge fields, and only to them. The last three terms are the source terms for
the gauge fields and the ghost fields, respectively.
The ghost field η is clearly unphysical. It is a Lorentz-scalar, but anti-
commuting field and has thus the ”wrong” spin-statistics relation. Ghosts
can thus appear only on internal lines in a Feynman diagram.
As we have pointed out above, there is no need to introduce a ghost
field when we are dealing with an Abelian gauge theory. In this case M is
independent of the gauge field Aµ and the ghost-field η is totally decoupled
from Aµ ; it can, therefore, be left out from the beginning in this case. It
is worthwhile to remember that the same situation is encountered in a non-
Abelian theory if we work in the axial gauge. In this gauge, however, the
propagator is rather complicated because it involves explicitly the (arbitrary)
vector nµ .

13.4.3 Feynman Rules


Equations (13.90) and (13.91) give the final result for the generating func-
tional of a non-Abelian gauge-field theory. The generating functional now has
the same form as the one obtained earlier for QED (cf. (13.21)), except for the
presence of the ghosts which are a typical feature of non-Abelian gauge field
theories. If the system under study contains also physical fermions, then the
Lagrangian would have to be supplemented by a fermion term along the lines
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 177

developed in Chap. 11 together with the proper source terms. The fermion-
gauge field coupling is mandated by local gauge invariance (cf. (13.30)) and
appears through the covariant derivative.
Since the generating functional (13.90) is finite and free of the overcount-
ing of gauge fields discussed at the start of this chapter it can be taken as
a starting point to derive the Feynman rules for non-Abelian gauge field
theories.

Gauge field couplings. For the gauge field and its couplings we obtain
the Feynman-rules by isolating the selfinteraction terms in L. We get
1 l lµν
L = − Fµν F
4
1  
= − ∂µ Alν − ∂ν Alµ − gf lmn Am n
µ Aν ∂ µ Alν − ∂ ν Alµ − gf lop Aoµ Apν
4
1  
= − ∂µ Alν − ∂ν Alµ ∂ µ Alν − ∂ ν Alµ
4
1 lmn m n  µ lν 
+ gf Aµ Aν ∂ A − ∂ ν Alµ
2
1
− g 2 f lmn Am n lop oµ pν
µ Aν f A A . (13.92)
4
The first term alone just looks like the Lagrangian of an Abelian field. It
is the only term quadratic in the fields, so that the gauge boson propagator
is just that of the photon found in (13.19), except for the additional SU(N)
labels. We thus have for the gauge boson two-point function (cf. (13.24))

1)
!
µ k ν −iδ lm kµ kν
lm
= iDµν = 2 gµν + (λ − 1) 2 .
l m k + i k
(13.93)
Note that D has to be diagonal in the intrinsic quantum numbers be-
cause L involves a trace over the group indices.

The second and third line in (13.92) give the cubic self-coupling terms of
the gauge field. The cubic term
1 lmn m n  µ lν 
gf Aµ Aν ∂ A − ∂ ν Alµ = gf lmn Am n µ lν
µ Aν ∂ A (13.94)
2
can be represented by the graph shown in Fig. 13.2. Its Feynman rules can
be obtained by considering the generating functional of lowest order in the
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 178

coupling constant g. Quite analogous to the situation in the φ4 model (cf.


section 9.1, (9.8)) the only term of Z that contributes to the three-point
Green’s function is that of order O(J 3 ); it is obtained by just replacing the
fields by those generated by the external source. This gives
Z "Z Z
lmn mk
Z[J] = −igf Dµλ (x − y 0 )J kλ (y 0 )d4 y 0 nk
Dνλ (x − y 0 )J kλ (y 0 )d4 y 0
Z
µ
×∂ Dlk νλ (x − y 0 )Jλk (y 0 )d4 y 0
#
+ terms of lower order in J d4x
 
i Z jκ 0 jk 0
× exp − J (x )Dκλ (x − y 0 )J kλ (y 0 )d4x0 d4y 0 . (13.95)
2
Here the Latin superscripts refer to the intrinsic SU (N ) degrees of freedom,
the Greek subscripts and superscripts are those of the Lorentz group. The
three-point Green’s function can now be obtained by differentiating Z[J]
 3
1 δ3Z
Ghij

πρσ (x1 , x2 , x3 ) = hπ iρ jσ
. (13.96)
i δJ (x1 )δJ (x2 )δJ (x3 ) J=0
This gives

Ghij
πρσ (x1 , x2 , x3 )
Z (
lmn 4 mj ni
= igf d x iDµσ (x − x3 )iDνρ (x − x2 )∂ µ iDπlh ν (x − x1 )
)
+ permutations of external legs . (13.97)

Equation (13.97) becomes in momentum space (with G = iD)



Ghij
πρσ (p, q, k) = igf
lmn
Gmj ni µ lh ν
µσ (k)Gνρ (q)(−ip )Gπ (p)

+ permutations of external legs (13.98)

= gf hji pµ g νλ Gjj ii hh
µσ (k)Gνρ (q)Gλπ (p)

+ permutations of external legs .

In the Feynman gauge we obtain the gauge field three-point function


−i −i −i
Ghij
πρσ (p, q, k) = gf
hij
(13.99)
k 2 p2 q 2
 
× (k − q)π gρσ + (q − p)σ gρπ + (p − k)ρ gπσ .
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 179

σ3j

k
p
π1h
q

ρ2i

Figure 13.2: Cubic gauge coupling term. The first symbols at each gluon line
give the Lorentz indices, the second number the lines and the third denote
the SU (N ) labels.

From this expression the vertex factor can be read off. We then have the
rules
2) The vertex factor for the three-point vertex is
 
gf hij (k − q)π gρσ + (q − p)σ gρπ + (p − k)ρ gπσ . (13.100)

3) The four-point vertex in Fig. 13.3 gives similarly the factor


−ig 2 [ f his f jks (gπσ gρτ − gπτ gρσ ) (13.101)
+ f hjs f kis (gπτ gρσ − gπρ gστ )
+ f hks f ijs (gπρ gστ − gπσ gρτ ) ] .

τ 4k σ3j

π1h ρ2i

Figure 13.3: Quartic gauge coupling graph.


CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 180

λ l

Figure 13.4: Fermion-gauge boson vertex. The fermion line is represented by


a solid line.

Fermion-gauge coupling. If there are also particles present, then the


gauge bosons couple to them through the minimal substitution (13.30)

Dµ = ∂µ + igT l Alµ (13.102)

and the Feynman rules for these couplings can directly be obtained from our
considerations in Chaps. 8.3 and 11. In particular, any fundamental fermion
field couples to the gauge fields according to the rule

4) The fermion-gauge vertex (Fig. 13.4) for a gauge boson with Dirac
index λ and group index l is given by:
 
−ig (γλ )βα T l (13.103)
fi

Ghost couplings. We now discuss the rules for handling those terms in
the Lagrangian that involve the ghost field.
The ghost-part of the Lagrangian reads (cf. (13.91) and (13.85))
Z R R
¯ = (η̄M η)d4x d4y+i (ξ̄η+η̄ξ) d4x
Zghost [ξ, ξ] Dη̄ Dη e−i
Z
= Dη̄ Dη
(
iZ h m   i
× exp − η̄ (x) Dµml (x)∂xµ δ 4 (x − y) η l (y) d4x d4y
g
)
+ source terms
Z
= Dη̄ Dη
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 181

( )
iZ h m ml µ l
i
4
× exp − η̄ (x)Dµ ∂ η (x) d x + source terms
g
Z
= Dη̄ Dη
(
iZ h m   i
× exp − η̄ (x) δ ml ∂µ + ig (T n )ml Anµ (x) ∂ µ η l (x) d4x
g
)
+ source terms . (13.104)

Using the regular representation (13.29) for the generators we get for the
exponent (without the source terms)
iZ m Z  
{. . .} = − η̄ (x)2η m (x) d4x + η̄ m (x) −if nml Anµ (x) ∂ µ η l (x) d4x
g
iZ m Z
= − η̄ (x)2η m (x) d4x + if lmn η̄ m (x)Anµ (x) ∂ µ η l (x) d4x .
g

After rescaling the ghost field (η → gη) we are thus left with the following
ghost contribution to the generating functional
¯
Zghost [ξ, ξ] (13.105)
Z  Z h i
= Dη̄ Dη exp −i 4 l l
d x η̄ 2η − gf lmn
Anµ η̄ ∂ η + ξ¯l η l + η̄ l ξ l
m µ l
.

Note that the ghost field couples only to the gauge field and not to any of
the other fields, like e.g. fermions, in the theory. Also, we remember that
the ghost fields η and η̄ cannot appear on external lines because the ghosts
follow the wrong spin-statistics relation. This ensures that the ghosts are
integrated out so that the FP determinant is generated.
The Feynman rules for the ghost fields can now be read off from (13.105).

6) The ghost propagator is given by

k i
= δ lm . (13.106)
l m k2 + i

Note that the ghost line carries an arrow; this reflects the presence of ghosts
and antighosts, because the FP matrix is in general not hermitean. The
notation is then such that particles always run with the time axis from bottom
to top and antiparticles run against it. More loosely speaking, the ghost line
runs from η̄ to η.
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 182

2i

σ3j
k

1h

Figure 13.5: Ghost-gauge coupling. The wavy line denotes a gauge boson
whereas the dashed line denotes the ghost. The first symbol on the gauge
field line (σ) denotes the Lorentz index, the second just numbers the particle
and the third (j) gives the SU(N) label.

Finally, according to (13.105) the ghost can couple to the gauge field
as depicted in Fig. 13.5. The vertex factor is obtained by considering the
ghost-ghost-gauge boson three-point function. As for the gauge three-point
function just discussed it can be obtained by considering the lowest order (in
the gauge-ghost coupling) expression
Z 1
¯ J] = −i
 δ 1 δ 1 δ
Z[ξ, ξ, d4x −gf lmn ∂ µ ¯l eiS0 [ξ,ξ̄,J]
i δJ nµ (x) m
i δξ (x) i δ ξ (x)
Z "Z Z
= − igf lmn nk
Dµν (x 0 kν
− y )J (y ) d y 0 4 0
ξ¯k Dkm (x − y 0 ) d4 y 0
Z
µ
×∂ Dlk (x − y 0 )ξ k (y 0 ) d4 y 0
#
+ terms of lower order in J d4x eiS0 [ξ,ξ̄,J] (13.107)

This generating functional determines the following rule for the gauge boson-
ghost coupling vertex:

7) The gauge boson-ghost vertex (Fig. 13.5) has to be associated with a


factor
gf hij qσ (13.108)
where q is the outgoing momentum of the ghost. The four-momentum
conservation (p + k = q) is understood here; it is not written explicitly.
CHAPTER 13. PATH INTEGRALS FOR GAUGE FIELDS 183

In addition, we have, as for real, physical fermions, the rule

8) All ghost loops carry a sign (−1).

We also have to remember that ghosts cannot appear on external lines.


Chapter 14

EXAMPLES FOR GAUGE


FIELD THEORIES

The Feynman rules developed in the preceding chapter are of a ‘generic’


nature: they apply to all non-Abelian gauge field theories. The particu-
lar physics content of a gauge field theory is determined by specifying the
symmetry group SU (N ).
In the preceding chapter we have referred to Quantum Electrodynamics
(QED), Quantum Chromodynamics (QCD) and the theory of electroweak
Interactions. In this chapter we, very briefly, summarize the physical contents
and the Lagrangian of these theories.

14.1 Quantum Electrodynamics


Quantum Electrodynamics is the most simple gauge field theory. It de-
scribes the interactions between electrons and the electromagnetic field, i.e.
the Coulomb and Lorentz forces. Its ingredients are discussed at the end of
appendix 4 and in Sects. 13.1 and 13.2. It is based on the U (1) gauge group
and is thus an Abelian theory. Its Lagrangian is given in (4.58) in appendix
4.

14.2 Quantum Chromodynamics


Quantum Chromodynamics is the theory of the strong interactions. Its foun-
dations, first discussed in the 60’s, lie in the observation that the observed
hadron spectrum can be understood in terms of quarks, mesons being com-
posed of quark-antiquark pairs and baryons of three quarks; the quarks carry
spin 1/2 and a so-called flavor quantum number. For certain baryons, such

184
CHAPTER 14. EXAMPLES FOR GAUGE FIELD THEORIES 185

as the ∆++ , the total wavefunction, i.e. the product of space-part, spin-part
and flavor-part, one obtains in such a scheme (flavor SU(3)) a completely
symmetric wavefunction for the three quarks, in glaring contradiction to the
Pauli principle. This problem was solved by the introduction of a new, ad-
ditional intrinsic degree of freedom for the quarks, called color, in which the
quarks can differ (for a detailed discussion see [MOSEL]). The Pauli principle
then requires a completely antisymmetric wavefunction in this new intrin-
sic space. In experiments that produce mesons, i.e. quark-antiquark pairs,
through lepton-annihilation the number of colors has been found to be three.
The completely antisymmetric color state has therefore been identified with
the completely antisymmetric singlet state of a color SU(3) group. Gauging
this group has led to the non-Abelian theory of Quantum Chromodynamics
(QCD).
The Lagrangian of QCD is then very simply given by
1 c cµν X
LQCD = − Fµν F + [q̄f (iγ µ Dµ − mf )qf ] . (14.1)
4 f

Here the fundamental triplets of SU (3)C are given by the quark spinors
 
qrf
 f 
qf =  qg  , (14.2)
qbf

which transform according to the fundamental (= lowest-dimensional) rep-


resentation of SU (3)C . The indices r, g and b stand for the three colors (red,
green and blue), the index f for the flavor. The covariant derivative in (14.1)
is given by (cf. (13.30))
λc c
Dµ = ∂µ + ig G . (14.3)
2 µ
Here the eight matrices λc are the so-called Gell-Mann matrices, the gener-
ators of SU (3), and the fields Gc represent the eight gauge-fields, here called
the gluon fields, which form an SU (3) octett.
c
The field tensors Fµν have the general form of a non-Abelian theory
(13.35). The gluons thus also interact among themselves. One of the pre-
dictions of this theory is, therefore, the existence of so-called glueballs, that
consist only of gauge-field quanta.
The theory, furthermore, has the property of asymptotic freedom, i.e.
its running coupling constant (cf. the discussion at the end of Sect. 10.2.2)
becomes larger with decreasing gluon momentum. This leads to a quark-
quark potential that at small distances is Coulomb-like, but at large distances
CHAPTER 14. EXAMPLES FOR GAUGE FIELD THEORIES 186

increases linearly, and thus provides an explanation for the fact that free
quarks and gluons do not seem to exist, i.e. that they are confined inside the
hadrons.

14.3 Electroweak Interactions


While the Lagrangian of QCD has the rather simple, generic form (14.1), that
of the electroweak interactions is considerably more complicated. Its founda-
tions go back to the early seventies when a gauge-field theory was found that
comprised both ‘classical’ Quantum Electrodynamics and the phenomena of
weak interactions, responsible, for example, for the nuclear β-decay.
This theory is based on a product of two gauge-groups, one an Abelian
U (1) and the other one a non-Abelian SU (2). For the latter the left-handed
parts of the observed fermions, quarks and leptons,
1
ψL = (1 − γ5 ) ψ (14.4)
2
are assumed to form doublets
! !
νkL ukL
Ψk = or (14.5)
ekL d0kL

for the left-handed fields of the kth family of leptons and quarks. The right-
handed parts
Ψk = ekR or qkR (14.6)
are assumed to form SU (2) singlets. In (14.5) we have used generic abbrevi-
ations ! ! ! !
νk νe νµ ντ
for , or (14.7)
ek e− µ− τ−
and ! ! ! !
uk u c t
for , or . (14.8)
d0k d0 s0 b0
The U (1) gauge group is very similar to QED, but it is not connected
with the electric charge, but instead a so-called weak hypercharge y = 2q −2t 3
where q is the electrical charge of the fermion and t3 its weak isospin ± 12 ,
depending on the position of the particle in the two-component spinors (14.7),
(14.8), just as in the usual Pauli-spinors.
CHAPTER 14. EXAMPLES FOR GAUGE FIELD THEORIES 187

The complete Lagrangian of this theory is then given by


1 1
L = − Gµν · Gµν − F µν Fµν
4 4
   
2 † µ ϕ 2 1 2 µ ϕ 2
+ M W Wµ W 1 + + M Z Zµ Z 1 +
X
v 2 v
+ Ψ̄k iγ µ Dµ Ψk + LY (14.9)
k
 2 !
1 1 ϕ 1 ϕ
+ (∂ µ ϕ) (∂µ ϕ) − m2H ϕ2 1 + + .
2 2 v 4 v

The boldfaced field tensors are vectors in the internal space and their product
contains a summation over the group indices.
Since there are now two gauge groups present, we also have two gauge
fields appearing in the Lagrangian: Gµν for the non-Abelian SU (2)L and Fµν
for the Abelian U (1)Y . Thus, also the covariant derivative contains both of
these fields with two different coupling constants
 
l 01
Dµ Ψk = ∂µ + igt Wµl + ig yk Bµ Ψk , (14.10)
2
where yk is the hypercharge quantum number of Ψk and the tl are the gen-
erators of SU (2). The three fields W l and the single field B are the gauge
fields for SU (2)L and U (1)Y , resp. By linearly combining them one obtains
the 2 physical fields Wµ and Wµ† , the field Zµ as well as the photon field Aµ .
With the covariant derivative (14.10) the first term in the third line describes
massless fermions interacting with the gauge fields in the standard way.
The gauge fields W and B normally have to be massless, since an explicit
mass term breaks gauge invariance. Since these fields transmit the weak
interaction the latter would then be long ranged. This, however, presents
an immediate problem because the weak interaction, since Fermi’s theory, is
known to be very short-ranged. A way out of this difficulty is provided by
the so-called Higgs mechanism. The main ingredient of this mass generation
mechanism is the existence of a scalar Higgs field ϕ, yet to be discovered,
with non-linear self-couplings (last line in (14.9); these self-couplings lead to
a non-vanishing vacuum field. The second ingredient is a Yukawa coupling
of the gauge-fields to this Higgs field that generates their masses MW and
MZ in the same way as described in Sect. 12.3. Similarly, the term LY in
(14.9), generates masses for all the fermions through a Yukawa coupling to
the Higgs field. For the fermions it has the simple structure
 
v ϕ
L = −gf √ Ψ̄Ψ 1 + , (14.11)
2 v
CHAPTER 14. EXAMPLES FOR GAUGE FIELD THEORIES 188

from which the fermion mass can be read off as


v
mf = gf √ . (14.12)
2
Appendix A

Units and Metric

A.1 Units
While we work in this book in the first 3 chapters, the ’classical’ quantum
mechanics part, with the usual system of units, it is customary in elementary
particle physics and field theories to choose the system of units such that
the resulting expressions assume a simpler form. In particular, instead of
working with the usual three mechanical units for length, mass and time one
introduces instead three basic units for velocity, action and length.
The choice of the velocity of light, c, as the unit for the velocity implies
that
c=1. (A.1)
This in turn means that in such a system length and time have the same
dimensions and are equivalent units. With c = 1 the relativistic energy–
momentum relation assumes the simple form
E 2 = p2 + m2 . (A.2)
Choosing next the unit of action, h̄, such that
h̄ = 1 (A.3)
connects the dimensions of mass, time and length.
Since time and length are equivalent, mass assumes the dimension of an
inverse length; the same holds consequently for the momentum and the en-
ergy (A.2). The choice of 1 fm = 10−13 cm as a length unit, for example,
leads to masses, momenta and energies all given in fm−1 . Another often-used
unit for energy is that of 1 MeV, i.e. the energy that an electron acquires
when it is accelerated by the voltage of 1 MV. The transformation of the ex-
pressions back to the standard MKSA system can be achieved by multiplying

189
APPENDIX A. UNITS AND METRIC 190

all quantities with the appropriate combinations of h̄ and c. A particularly


useful relation for this conversion is

h̄c ≈ 197.3 MeV fm . (A.4)

A.2 Metric and Notation


All vectors, both ordinary three-component vectors and those in some inter-
nal space, are distinguished by boldface italic print in this book.
In relativistic expressions four-vectors are always written as
   
~ = A0 , A 1 , A 2 , A 3
Aµ = A0 , A (A.5)

Four-vectors with a superscript are called contravariant vectors. Covariant


vectors, denoted by subscripts, are then defined by

Aµ = gµν Aν (A.6)

with the metric tensor gµν that reads in Cartesian coordinates in Minkowski
space  
1 0 0 0
 0 −1 0 0 
 
  . (A.7)
 0 0 −1 0 
0 0 0 −1
The metric tensor with superscripts is used to raise the indices of four-vectors;
it is thus the inverse of g and is given by

g µν = gµν . (A.8)

The product
g µν gνλ = gλµ = 1 (A.9)
is often also abbreviated by the delta-function

δλµ = gλµ . (A.10)

A scalar product of 2 four-vectors is then defined by


~·B
A · B = A µ B µ = A µ Bµ = A 0 B0 − A ~ (A.11)

if Aµ and B µ are defined by


~ ,
Aµ = (A0 , A) ~ .
B µ = (B 0 , B) (A.12)
APPENDIX A. UNITS AND METRIC 191

The invariant square of the four-vector is thus given by


~2 .
A2 = Aµ Aµ = A20 − A (A.13)

Here, as always in the text, the Einstein convention of summing over double
indices, one lower and one upper, is used. The components of four-vectors
are generally labeled by Greek indices, whereas roman indices are used to
refer specifically to the last three (vector) components.
The space-time four-vector is

xµ = (x0 , ~x) = (t, ~x) , (A.14)

and the four-momentum is given by

pµ = (E, p~) . (A.15)

Finally, the components of the contravariant four-gradient are abbreviated


by !
µ ∂ ∂ ~ ,
∂ ≡ = , −∇ (A.16)
∂xµ ∂t
so that the four-momentum operator is given by

p̂µ = i∂ µ , (A.17)

and the Lorentz-invariant d’Alembert operator by

∂2 ~2=2 .
∂ µ ∂µ = −∇ (A.18)
∂t2
Other important four-vectors are the current

j µ = (ρ, ~) (A.19)

and the electromagnetic potential


 
~ .
Aµ = φ, A (A.20)
Appendix B

Functionals

B.1 Definition
Very generally speaking a functional is a mapping from a space of functions
into the real or complex numbers. For example, the integral
Z +∞
f (x)dx (B.1)
−∞

is a real or complex number that depends on the special function f (x) in the
integrand; other functions in general give other values for the integral. This
dependence of the integral on the function f is called a functional dependence
of the integral which itself is called a functional of f :
Z +∞
F [f ] = f (x)dx . (B.2)
−∞

In order to stress that not a special value of f , but instead the whole function
f is the argument of the functional square parentheses are used here to show
this functional dependence.
In physics one often deals with such functionals. For example, in classical
mechanics the action Z
S = L(q(t), q̇(t))dt (B.3)

integrated along a trajectory q(t) plays a special role. The action depends
obviously on the trajectory and is thus a functional integral of it: S = S[q(t)].
The Lagrange equations of motion are derived by investigating the changes
of S with the trajectory q(t): the physical trajectory leads to a stationary
value of S. To find this stationary value, and thus the physical trajectory,
requires taking the derivative of S[q] with respect to q. We thus need to
define also the so-called functional derivative.

192
APPENDIX B. FUNCTIONALS 193

B.2 Functional Integration


A functional integration can be introduced as an integration over a space of
functions Z
df F [f (x)] , (B.4)
which is defined as a limit n → ∞ of an integration over finite n-dimensional
subspaces. A typical example is given by the path integral for quadratic
Hamiltonians
i
Rt
Z h̄
L(x,ẋ)dt0
K (x, t; xi , ti ) = N Dx e ti

! n+1
m 2
= lim (B.5)
n→∞ i2πh̄η
  !2 
n
Z Y i n
m X xj+1 − xj 
× dxk exp  η  − V (x̄j ) .
k=1 h̄ j=0 2 η

B.2.1 Gaussian integrals


In this book we are often confronted with the evaluation of path integrals
over exponential functions whose exponents are quadratic in the coordinates.
It is thus useful to generalize the Gaussian integral formula
+∞ s
Z
− 21 ax2 2π
e dx = (a > 0) (B.6)
a
−∞

to this case.
From the definition of the path integral it is obvious that we have to
consider products of such integrals
! q
Z n
1X (2π)n
exp − ak x2k dx1 dx2 . . . dxn = Qn √ . (B.7)
2 k=1 k=1 ak
We next assume that the n numbers ak are all positive and form the elements
of a diagonal matrix A. We thus have
n
Y
det(A) = ak (B.8)
k=1

and n n
X X
ak x2k = xk Akk xk = xT Ax , (B.9)
k=1 k=1
APPENDIX B. FUNCTIONALS 194

where x is a column vector


 
x1

 x2 

x=
 ..  .

 . 
xn

Thus (B.7) becomes


Z n
− 21 xT Ax n (2π) 2
e dx= q . (B.10)
det(A)

So far, we have derived this equation only for a diagonal matrix A. It


is, however, valid for a more general class of matrices. This can be seen by
noting that for each real, symmetric matrix B there exists a real, orthogonal
matrix O that diagonalizes B to A

OT BO = A (O real, orthogonal) (B.11)

or, equivalently

B = OAOT (B real, symmetric) . (B.12)

We thus get
Z Z Z
− 21 yT By n − 21 yT OAOT y n 1 T Ax
e dy = e dy= e− 2 x dnx
n n
(2π) 2 (2π) 2
= q =q , (B.13)
det(A) det(B)

where we have substituted y = Ox and have used the fact that the Jacobian
of an orthogonal transformation is 1 (because det(O) = 1). The last step is
possible because the determinant of a matrix is invariant under an orthogonal
transformation. Equation (B.10) is thus valid also for general symmetric
matrices with positive eigenvalues; it can also be shown to hold for complex
matrices with positive real parts.
This result can also be extended to more general quadratic forms in the
exponent. For a one-dimensional integral of such type we have
+∞
Z r
2 +bp+c π b2 +c
e−ap dp = e 4a , (B.14)
a
−∞
APPENDIX B. FUNCTIONALS 195

We now assume an n-dimensional integral over such a form where the expo-
nential is given by
e−F (x) = e−( 2 x Ax+B x+C )
1 T T
(B.15)
where A is a real symmetric matrix, B is a vector and C a constant. We
bring F (x) into a quadratic form by writing
1
F (x) = (x − x0 )T A(x − x0 ) + F (x0 ) (B.16)
2
where x0 is given by
x0 = −A−1 B (B.17)
and F0 = F (x0 ) = C − 12 BT A−1 B. Setting now y = x − x0 gives
Z Z
−( 12 xT Ax+BT x+C ) n 1 T Ay−F
e dx = e− 2 y 0
dny
n
(2π) 2 1 T A−1 B−C
= q e2B . (B.18)
det(A)

Similar to the Gaussian integral (B.7) the following integral relation,


which can be proven by induction from n to n + 1, holds also
+∞
Z n h io
dx1 . . . dxn exp iλ (x1 − a)2 + (x2 − x1 )2 + . . . + (b − xn )2
−∞
!1 !
in π n 2

= exp (b − a)2 . (B.19)
(n + 1)λn n+1

Complex integrals. We can generalize these formulas now to complex


integration by noting that (B.6) can be squared and then be written as
π Z −ax2 Z −ay2 Z
2 2
= e dx e dy = e−a(x +y ) dx dy . (B.20)
a
Introducing now the complex variable z = x + iy gives
π 1 Z −az∗ z ∗
= e dz dz . (B.21)
a 2i
This can be generalized as before to many coordinates (by replacing orthog-
onal matrices by unitary ones). We obtain
Z
−z† Az n ∗ n
Z
∗ (2πi)n
e dz dz= e−zi Aij zj dnz ∗ dnz = , (B.22)
det(A)
APPENDIX B. FUNCTIONALS 196

where A is a Hermitian matrix with positive eigenvalues.


Another convenient, often used relation in this context is

ln det A = tr ln A , (B.23)

most easily proven for diagonal matrices. In this relation ln A is defined by


its power series expansion
(A − 1)2 (A − 1)3
ln A = ln(1 + A − 1) = A − 1 − + − ... . (B.24)
2 3!
With the help of this relation we have
Z
† Az
e−z dnz ∗ dnz = (2πi)n e−tr ln A
. (B.25)

B.3 Functional Derivatives


Suppose that F [f ] is a functional of the function f (x). The functional deriva-
tive of F is then defined by
δF [f (x)] F [f (x) + εδ(x − y)] − F [f (x)]
= lim . (B.26)
δf (y) ε→0 ε
For example, let us take
F [f ] = 2f (x) , (B.27)
where 2 is the d’Alembert operator. We get from the definition (B.26)
δF 1
= lim [2 (f (x) + εδ(x − y)) − 2f (x)] = 2δ(x − y) (B.28)
δf (y) ε→0 ε
Another example is
+∞
Z
F [f ] = f (x)dx . (B.29)
−∞

According to the definition just given we have


Z Z 
δF 1
= lim [f (x) + εδ(x − y)] dx − f (x) dx (B.30)
δf (y) ε→0 ε
Z
= δ(x − y) dx = 1

A second example is given by


R
F [f ] = ei f (x)x dx
(B.31)
APPENDIX B. FUNCTIONALS 197

with
δF 1  i R [f (x)+εδ(x−y)]x dx R 
= lim e − ei f (x)x dx
δf (y) ε→0 ε

1 R   R
= lim ei f (x)xdx eiεy − 1 = iy ei f (x)x dx . (B.32)
ε→0 ε

A general formula that we will need quite often involves functionals F [J]
of the form
Z Z
F [J] = dx1 . . . dxn f (x1 , . . . , xn ) J(x1 )J(x2 ) . . . J(xn ) (B.33)

with f symmetric in all variables. Then the functional derivative with respect
to J has the form
Z
δF [J]
= n dx1 dx2 . . . dxn−1 f (x1 , x2 , . . . , xn−1 , x)
δJ(x)
× J(x1 )J(x2 ) . . . J(xn−1 ) . (B.34)

In our later considerations the functional may also be defined by a power


series expansion

X 1 Z
φ[J] = dx1 . . . dxn φn (x1 , . . . , xn )J(x1 )J(x2 ) . . . J(xn ) . (B.35)
n=1 n!

Then the functional derivative is given by



δ k φ[J]
1 X
= φk (yp1 , yp2 , . . . , ypk ) , (B.36)
δJ(y1 )δJ(y2 ) . . . δJ(yk ) J=0 k! p

where the sum runs over all permutations p1 , . . . , pk of the indices 1, . . . , k.


If we assume that φk is a symmetric function under exchange of any of the
coordinates x1 , . . . , xk , then the functional derivative is given by

δ k φ[J]

= φk (y1 , . . . , yk ) , (B.37)
δJ(y1 )δJ(y2 ) . . . δJ(yk ) J=0

just as in a normal Taylor series.


Appendix C

RENORMALIZATION
INTEGRALS

In Chapt. 10 we encountered divergent integrals in the calculation of higher-


order two- and fourpoint functions. To evaluate these integrals analytically
in n dimensional Minkowski space is the purpose of this appendix. We start
with a discussion of the Gamma function whose properties play a role in the
evaluation of the integral and its expansion into n ≈ 4 dimensions.

Properties of the Gamma function. The Gamma function is defined


by an integral representation
Z ∞
Γ(x) = e−t tx−1 dt ; (C.1)
0

it is single-valued and analytic everywhere except at the points z = n =


0, −1, −2, . . . where it has a simple pole with residue (−1)n /n!. It can thus
be expanded around z = 0
1
Γ(z) = − γ + O(z) , (C.2)
z
where γ is known as the Euler-Mascheroni constant (γ ≈ 0.577 . . .). Since
the Gamma function also obeys the relation

Γ(z + 1) = zΓ(z) = z! (C.3)

we obtain
 
Γ(z) 1 1
Γ(−1 + z) = ≈ −(1 + z) −γ =− −1+γ . (C.4)
z−1 z z

198
APPENDIX C. RENORMALIZATION INTEGRALS 199

It obeys the relation


−1
Γ(z) = (z − 1)Γ(z − 1) ≈ (z − 1) , (C.5)
z+1
where the second part of this equation is correct close to the pole at z = −1.

Evaluation of integrals with powers of propagators in n dimensions.


The typical integral to be evaluated reads
Z Z Z
n 1 1
Il = dq l = d~q dq0 2 l . (C.6)
2 2
(q − m + iε) (q0 − ~q 2 − m2 + iε)
with one timelike (q 0 ) and n−1 spacelike (q k ) coordinates; the vector symbol
denotes a (n − 1)-dimensional vector ~q = (q 1 , q 2 , . . . , q n−1 ). l is an integer
parameter and m2 a real, positive number (mass squared).
We first consider the integration over q0 . The situation here is exactly
as in Sect. 6.1.2. The integrand has its only poles in the second and fourth
quadrant of the complex q0 plane at q0 = ± [(~q 2 + m2 ) − iδ] (cf. Fig. 5.1).
Since the integral behaves as 1/(q0 )2l−1 for large q0 the integration along
the q0 -axis can be closed in the lower half of the complex q0 plane without
changing the integral’s value.
According to Cauchy’s theorem this integration contour can now be chan-
ged into one that runs along the imaginary q0 axis and closes the contour in
the right half of the complex q0 plane. Because the contour still encloses the
same pole (the one in the fourth quadrant of the complex plane) the value
of the integral does not change.
This then gives the equality
Z Z
+∞ 1 +i∞ 1
dq0 l = dq0 l (C.7)
−∞ (q02 − ~q2 − m2 + iε) −i∞ (q02 − ~q2 − m2 + iε)
since in both cases the contribution of the half-circle that closes the contour
vanishes. In the integral on the rhs the integration runs along the imaginary
axis in the complex q0 plane. On that axis q0 is purely imaginary; the inte-
grand has thus been analytically continued from the originally purely real q0
to a complex one.
Using the transformations (6.22), (6.23)
qn = −iq0 dn qE ≡ dqn d~q = −idn q , (C.8)
with real qn we obtain the integral
Z Z Z
+∞ 1 1
Il = (−)l i d~q dqn l = (−)l i dn q E l (C.9)
−∞ (qE2 + m2 ) (qE2 + m2 )
APPENDIX C. RENORMALIZATION INTEGRALS 200

with n  
X 2
qE2 = qi = ~q 2 + qn2 . (C.10)
i=1

(C.8) is just the Wick rotation discussed in Sect. 5.1.1.


We now introduce “polar coordinates” by defining an n − 1 dimensional
solid angle element dΩn by the relation

dnqE = qEn−1 dqE dΩn . (C.11)

and get for the integral1


Z Z∞
l 1
Il = (−) i dΩn q n−1 dq . (C.12)
0
(q 2 + m2 )l

The integral over the solid angle can be analytically performed and yields
Z
2π n/2
dΩn =   .
n
(C.13)
Γ 2

This can be seen by considering the n-th power of the Gaussian integral
 √ n Z ∞ n Z Pn
−x2 x2k
π = dx e = dx1 dx2 . . . dxn e− k=1
0
Z Z Z

n−1 −x2 1Z ∞   n−2 2
= dΩn x e dx = dΩn d(x2 ) x2 2 e−x
0
 
2 0
Z
1 n
= dΩn Γ . (C.14)
2 2
Here the integral representation (C.1) of the Gamma function has been used
in the last step.
This gives for the integral
n ∞
l 2π 2 Z 1
Il = (−) i  n  q n−1 dq . (C.15)
Γ 2 0
(q 2 + m2 )l

The remaining integral can be evaluated with the help of Euler’s β func-
tion
Z∞
Γ(x)Γ(y)
B(x, y) = = 2 dt t2x−1 (1 + t2 )−x−y , (C.16)
Γ(x + y)
0

1
In order to simplify the notation we are no longer denoting the Euclidean vectors by
the subscript E.
APPENDIX C. RENORMALIZATION INTEGRALS 201

(see Abramowitz (6.2.1)); we obtain


   
n n
Z
q n−1 1 n n 1 Γ 2
Γ l− 2
l dq = mn−2l B( , l − ) = mn−2l . (C.17)
(q 2 + m2 ) 2 2 2 2 Γ(l)

Thus the complete integral is now given by


     
n n n n
2π 2 1 Γ 2
Γ l− 2 n Γ l− 2
Il = (−)l i  n  mn−2l = (−)l iπ m
2
n−2l
. (C.18)
Γ 2 Γ(l) Γ(l)
2

In this expression the dependence of the integral on the dimension n is ex-


plicit; it can analytically be continued to the physical case n = 4 where it
has a pole.
Appendix D

GRASSMANN
INTEGRATION FORMULA

A more general proof than the specific examples given in section (11.1.2.2)
uses the fact that any antisymmetric (n × n) dimensional matrix M can be
brought into a block-diagonal form through an orthogonal transformation

M −→ M 0 = OT M O (D.1)
 
0 M1 0 0 ··· 0 0
 −M1 0 0 0 ··· 0 0
 


 .. .. 



0 0 0 M2 · · · . . 

0
M = .. .. 
(D.2)

 0 0 −M2 0 ··· . . 

 .. .. .. .. .. .. 

 . . . . ··· . . 

 
 0 0 0 0 ··· 0 Mn 
0 0 0 0 · · · −Mn 0
for n even (for odd n the matrix M 0 has one more row and one more column
with all zeros, so that its determinant vanishes). The form of M 0 shows
clearly that
det(M ) = det(M 0 ) = M12 M22 · · · Mn2 . (D.3)
We thus get
Z Z
−η T M η T (OM 0 O T )η
I(n) = dη1 . . . dηn e = dη1 . . . dηn e−η
Z
T M 0
= d1 . . . dn J e− (D.4)

with  = O T η and J being the Jacobian. In our special case here, O is


orthogonal and therefore, [det(O)]−1 = J = 1.

202
APPENDIX D. GRASSMANN INTEGRATION FORMULA 203

We can now continue with the evaluation of I(n) in (D.4) and get
Z
n 1  n
I(n) = d1 . . . dn (−) 2  n  T M 0  2 , (D.5)
2
!

because only that term in the expansion of the exponental can contribute in
a n-dimensional integral that has exactly n factors of i . In detail, we have
n
(−) 2 Z 0
 n
2
I(n) =  n  d1 . . . dn α Mαβ β . (D.6)
2
!

Since M 0 has the special, block-diagonal form given above (D.2), we have
0
Mαβ = 0, except for α = 2α0 , β = α − 1 or α = 2α0 − 1, β = α + 1. Again,
there can be no two equal 0 s under the integral, if this integral is to be
nonzero. We thus have
n
(−) 2 Z 
0
I(n) =  
n
d1 . . . dn 2α0 M2α 0 ,2α0 −1 2α0 −1

2
!
n
0 2
− 2α0 −1 M2α 0 −1,2α0 2α0
n
2 2 (−)n Z 
0
n
2
=  
n
d1 . . . dn 2α0 −1 M2α 0 −1,2α0 2α0 , (D.7)
2
!

since M 0 is antisymmetric. Here the index α0 still runs from 1 to n/2; n is


even so that the phase disappears in the following.
We now perform the exponentiation
n
22 Z XX X
I(n) =  n  d1 . . . dn ... 2α0 −1 2α0 · · · 2ν 0 −1 2ν 0
2
! α0 β0 ν0
| {z }
n sums
0 0 0
× M2α 0 −1,2α0 M2β 0 −1,2β 0 · · · M2ν 0 −1,2ν 0 . (D.8)
q
0 0
The product of the factors M2α just gives det(M 0 ). Since
0 −1,2α0 · · · M2ν 0 −1,2ν 0

the i appear always pairwise, they can be commuted to normal ordering


(n · · · 1 ) without sign-change; the sums then give (n/2)! times the same
result. Thus
Z q
n
I(n) = 2 2 d1 . . . dn n · · · 1 det(M 0 )
n
q
= 22 det(M 0 ) . (D.9)
APPENDIX D. GRASSMANN INTEGRATION FORMULA 204

The last step is possible because we have for each blockmatrix Mβα separately

det(Mβα ) = − (M2α−1,2α ) (+M2α,2α−1 )


= + (M2α−1,2α )2 , (D.10)

and for det(M 0 ) Y


det(M 0 ) = 0
det(Mβα ). (D.11)
α

Using (D.3) we thus get the desired result


Z q
T M n
d1 . . . dn e− = 22 det(M ) . (D.12)

Equation (D.12) corresponds to (B.10) for the boson case. Note that here –
in contrast to the bosonic case – the determinant appears in the numerator!
Appendix E

BIBLIOGRAPHY

We list here a number of textbooks that provide for more information on the
topics treated in this manuscript.

Introductory texts

1. J.J. Sakurai, Modern Quantum Mechanics, Addison-Wesley, Redwood


City, 1985; contains a very nice introduction into the non-relativistic
path integral formalism.

2. U. Mosel, Fields, Symmetries, and Quarks, 2nd revised and enlarged


edition: Springer, Heidelberg, 1999; contains a short introduction into
fundamentals of field theories and gauge field theories.

3. T.D. Lee, Particle Physics and Introduction to Field Theory, Harwood,


Chur, 1981; very physical treatment of field theory and high-energy
phenomenology, somewhat annoying typesetting.

4. L.H. Ryder, Quantum Field Theory, Cambridge University Press, 1985;


very didactical presentation of modern field theories.

5. C. Itzykson and J-B. Zuber, Quantum Field Theory, McGraw-Hill, New


York, 1985; nearly comprehensive book on field theory, didactically not
very well done.

6. M.E. Peskin and D.V. Schroeder, An Introduction to Quantum Field


Theory, Addison-Wesley, Reading, 1995; also didactically very good,
rather comprehensive book on quantum field theory. The new standard
book on this topic!

205
APPENDIX E. BIBLIOGRAPHY 206

Specialized Literature

1. Ta-Pei Cheng and Ling-Fong Li, Gauge Theory of elementary particle


physics, Clarendon Press, Oxford, 1984; quite comprehensive book,
well-structured, easy to read, but not always complete in its arguments
and derivations.

2. P. Ramond, Field Theory, a modern Primer, Addison-Wesley, Redwood


City, 1990; very nice introduction to field theories, also modern aspects,
builds entirely on path integral formalism.

3. P.H. Frampton, Gauge Field Theories, Benjamin-Cummings, Menlo


Park, 1987; bad typesetting, not easy to read.

4. S. Pomorski, Gauge Field Theories, Cambridge University Press, 1987;


not very didactical, but quite deep.

5. D. Bailin and A. Love, Introduction to Gauge Field Theory, 2nd ed.,


Hilger, Bristol, 1994; very comprehensive treatment of gauge field theo-
ries based on path integral methods from the start, didactical, contains
modern aspects of field theory beyond Standard Model.

6. C. Grosche and F. Steiner, Handbook of Feynman Path Integrals, Sprin-


ger, Berlin, 1998; compact presentation of theory of path integrals and
their history. Unique in its description of evaluation techniques and
its tables of analytically calculable path integrals. Complete list of
references.

7. H. Kleinert, Path Integrals in Quantum Mechanics, Statistics and Poly-


mer Physics, World Scientific, Singapore 1995. Application of path
integrals to wide range of physics problems.

8. G. Roepstorff, Path Integral Approach to Quantum Physics, Springer,


Heidelberg, 1996

Das könnte Ihnen auch gefallen