Advanced Quantum Field Theory

Prof Arttu Rajantie

2014-15 Second Term

Contents
I

Path Integral Quantization

Derivation of the Path Integral

1.1 Heuristic Argument . . . .
1.2 Quantum Mechanics . . .
1.3 Many Particles . . . . . .
1.4 Scalar Field Theory . . . .

II
5

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

3
3
4
7
8

Correlation Functions
2.1 Path Integral Representation
2.2 Generating Functional . . .
2.3 Scalar Propagator . . . . . .
2.4 Complex Scalar Field . . . .
2.5 Dirac Propagator . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

8
8
10
12
13
14

Gauge Fields
3.1 Abelian Symmetry . . . . . .
3.2 Non-Abelian Symmetry . . . .
3.3 Gauge Fixing . . . . . . . . .
3.4 Gauge and Ghost Propagators
3.5 Summary of Propagators . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

17
17
18
20
24
25

Interaction Vertices
4.1 Scalar Theory . . . . . . . . .
4.2 Connected and 1PI Correlators
4.3 Complex Scalar . . . . . . . .
4.4 QED . . . . . . . . . . . . . .
4.5 QCD . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

25
25
31
34
35
36

Renormalisation
Perturbative Renormalisation
5.1 Ultraviolet Divergences . .
5.2 Renormalised Couplings .
5.3 Field Renormalisation . . .
5.4 Power Counting . . . . . .

38
.
.
.
.

38
38
41
43
44

Renormalised Perturbation Theory

6.1 Counterterms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2 Dimensional Regularisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3 Minimal Subtraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47
47
50
54

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

Renormalisation Group
7.1 Callan-Symanzik Equation . . . . . . .
7.2 Wilsonian Renormalisation . . . . . . .
7.3 Renormalisation Group Transformation
7.4 Renormalisation Group Flow . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

55
55
58
62
63

Renormalisation of QCD
8.1 Counterterms . . . . . . . . .
8.2 Correlation Functions . . . . .
8.2.1 Gluon . . . . . . . . .
8.2.2 Quark . . . . . . . . .
8.2.3 Ghost . . . . . . . . .
8.2.4 Quark-gluon coupling
8.2.5 Summary . . . . . . .
8.3 Beta Function . . . . . . . . .
8.4 Renormalisation Group Flow .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

67
67
71
71
75
76
77
77
78
79

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

Index

81

Part I

Path Integral Quantization

1
1.1

Derivation of the Path Integral

Heuristic Argument

Double slit experiment:

A1
O

S
A2

- Light emitted at source S, observed at detector O

- Probability given by the modulus of the amplitude squared:
p(S O) = |A(S O)|2

(1)

- Superposition principle: Total amplitude is sum over two holes

A(S O) = A(S A1 O) + A(S A2 O)

(2)

- Drill n holes:
A1
O

S
An

A(S O) =

n
X

A(S Ai O)

i=1

- Add another screen:

(3)

A1,1

A2,1
O

S
A1,n

A(S O) =

n
X

A2,n

A(S A1,i A2,j O)

(4)

i,j=1

- Add infinite number of screens, drill infinite number of holes in each:

Sum over all paths:

A(S O) =

A(S (path) O)

(5)

all paths

1.2

Quantum Mechanics

- Non-relativistic particle in 1D potential:

2
= p + V (
H
q)
2m

(6)

- Amplitude from point qa to point qb in time T :

(T )|qa i,
U (qa , qb ; T ) = hqb |U

(7)

(t) = exp(itH/~)

(8)

where the time evolution operator is

(T ) = U
()N
- Split interval T into N shorter ones of duration  = T /N , so that U
()U
() U
()U
()|qa i,
U (qa , qb ; T ) = hqb |U

(9)

- Insert complete sets of states

Z
1=

dqk |qk ihqk |

(10)

between all factors and write qb = qN and qa = q0 :

Z
()|qN 1 ihqN 1 |U
()|qN 2 i hq1 |U
()|q0 i
U (qa , qb ; T ) =
dq1 dqN 1 hqN |U

Z NY
1
N
1
Y
()|qk i

=
dqj
hqk+1 |U
j=1

(11)

k=0

- If  is small, we can expand to linear order

i
() = exp(iH/~)

+ O(2 )
U
= 1 H
~
p, q) can contain terms that are products of p s and qs:
- In general H(
Use commutation relations [
p, q] = i~ to move qs to the left of p s
(Other choices possible: Symmetric ordering in P&S)
R
- Again, insert a complete set of states dpk |pk ihpk |,
Z
Z
p, q)|qk i = dpk hqk+1 |H(
p, q)|pk ihpk |qk i = dpk H(pk , qk+1 )hqk+1 |pk ihpk |qk i
hqk+1 |H(
- Remembering that the standard choice of the momentum operator p = i~/q implies
1
eip q/~ ,
hp |qi =
2~

(12)

(13)

(14)

we find
p, q)|qk i =
hqk+1 |H(

dpk
H(pk , qk+1 ) exp
2~


i
pk (qk+1 qk ) ,
~

(15)

and consequently for infinitesimal ,



i

hqk+1 |U ()|qk i = hqk+1 | 1 H |qk i + O(2 )

~


Z
i
=
dpk hqk+1 | 1 H |pk ihpk |qk i + O(2 )
~


Z
i
=
dpk 1 H(pk , qk+1 ) hqk+1 |pk ihpk |qk i + O(2 )
~




Z
dpk
i
i
=
1 H(pk , qk+1 ) exp
(pk (qk+1 qk )) + O(2 )
2~
~
~


Z
dpk
i
=
exp
(pk (qk+1 qk ) H(pk , qk+1 ) + O(2 ),
2~
~
which does not involve any operators
- We can therefore express the amplitude U (qa , qb ; T ) as a multidimensional integral

(16)

N
1
Y

Z
U (qa , qb ; T ) =

(
dqj

N
1
Y

j=1

k=0

dpk
2~

N 1
i X
exp
(pk (qk+1 qk ) H(pk , qk+1 ))
~

"

#
(17)

k=0

- Integral over all values of p and q at all times

- One more p integral than q integral
- Denote integral by shorthand Dq(t)Dp(t) and note that qk+1 qk q
#
" Z
Z q(T )=qb
i T
dt (p q H(p, q)) ,
Dq(t)Dp (t) exp
U (qa , qb ; T ) =
~ 0
q(0)=qa

(18)

- Boundary conditions: q(0) = qa and q(T ) = qb

- Hamiltonian path integral: Very general, works for any H

2
- Usually p only appears as a square: In our case p /2m
one can do the p integrals in Eq. (17) explicitly
Z



i
dp
exp
(p q H(p, q))
2~
~


Z



i
dp
i
2mq
exp
p
exp V (q)
p2
~
2~
2m~

Z
mq 2
2
m
i
i
dp
= e ~ ( 22 q V (q))
e 2m~ (p  )
2~


Z
2
i
m
i 2
d
p
(
2 q V (q))
~
2
= e
exp
p
2~
2m~
=

- We need to evaluate the Gaussian integral over p:

- Strictly speaking not defined, because does not converge
- Use analytic continuation


Z
dp
1 2
1
exp cp =
,
2
2
2c

c=

i
m~

(19)


 r
 

i
m
i m
dp
2
exp
(p q H(p, q)) =
exp
q

V
(q)
2~
~
2~i
~ 22

(20)

- Thus, the amplitude is

U (qa , qb ; T )

" N 1
#
1

 m N/2 Z NY
X m
i
2

dqj exp

(qk+1 qk ) V (qk+1 )
=
2~i
~
22
j=1
k=0
" Z

#
Z
i T
1 2

Dq(t) exp
dt
mq V (q)
~ 0
2
" Z
# Z
Z q(T )=qb
q(T )=qb
i T
=
Dq(t) exp
dtL(q,
q) =
Dq(t)eiS/~
(21)
~ 0
q(0)=qa
q(0)=qa

- This is the path integral expression of the amplitude

R
- Note: When action is large, S = dtL  ~, integrand highly oscillatory
- Stationary phase approximation:
Only paths where S is roughly stationary contribute significantly: S/q = 0
= Classical Lagrangian mechanics

1.3

Many Particles

- Generalise to M distinguishable particles:

- Operators q and p , with = 1, . . . , M
= (
= (
- For notational simplicity, define M -component vectors q
q 1 , . . . , qM ) and p
p1 , . . . , pM )
p, q
)
- Consider Hamiltonian H(
- Everything is more or less identical to the one-particle case:
()U
() U
()U
()|qa i
U (qa , qb ; T ) = hqb |U
(22)
commute, so a complete set of states is
- Different components of the position operator q
!
Z
M
Y
1=
dqk |qk ihqk |

(23)

=1

- The analogue of Eq. (11) is

Z
U (qa , qb ; T ) =

dqj

N
1 Y
M
Y
j=1 =1

N
1
Y

()|qk i
hqk+1 |U

(24)

k=0

- The matrix elements of the Hamiltonian are [cf. Eq. (13)]

!
Z
M
Y
p, q
)|qk i =
hqk+1 |H(
dp
H(pk , qk+1 )hqk+1 |pk ihpk |qk i
k

(25)

=1

and q
is
- The scalar product between eigenstates of p
M/2

hp |qi = (2~)

M
i X
p q
exp
~ =1

- Consequently, for an infinitesimal time evolution we find

!
"
!#
Z
M

Y
X
dp
i

k
()|qk i =
hqk+1 |U
exp
pk (qk+1 qk ) H(pk , qk+1
,
2~
~

=1

(26)

(27)

and for finite T , in analogy with Eq. (17)

U (qa , qb ; T ) =

Z Y

"
!#
Y
Y dp
iX X

k
dqj
exp
pk (qk+1 qk ) H(pk , qk+1 )
(28)
2~
~

j
k

- Assume that the Hamiltonian is

p, q
) =
H(

X (
p )2
2m

+ V (
q)

(29)

and do the p integrals using Eq. (19)

"
!#
r
Z Y r

2
Y
X m X (qk+1
q
)
m
m
i
k

U (qa , qb ; T ) =
dq exp

V (qk+1 )
2~i j
2~i j
~
2
2

k
" Z
!#
Z
i T
1 X

Dq(t) exp
dt
m
q q V (q)
(30)
~ 0
2

1.4

Scalar Field Theory

- Consider now the Hamiltonian of a 1+1D relativistic

scalar field

!2
Z

1
1
d

= dx
+ V ()
H
2 +
2
2 dx

(31)

- Discretize: Lattice spacing x x = x

(x)

d/dx
(+1 )/x
- Because the canonical commutation relation for fields is
x),
[(t,
(t, y)] = i~(x y) = i~

,
x

we define p =
(x)x
- Then we have [ , p ] = i~ as for particles
- This gives the Hamiltonian

2
X  1 
1 X
+1

H(
p, ) =
p p + V(),
V() =
x

+ V ( )
2x
2x2

(32)

(33)

which is of the same form as Eq. (29) but with m = x

- The amplitude from field configuration a (x) to b (x) in time T is therefore
!#
" Z
T
X 1
i
dt
x V()
D (t) exp
~ 0
2

"

#
Z
Z
i T X
1 1 (+1 )2

D (t) exp
x

V ( )
dt
~ 0
2
2
x2

" Z
!#
 2
Z
Z
i T
1 2 1
D (t, x) exp

V ((t, x))
dt dx
~ 0
2
2 x

 Z (T,x)=b (x)
 Z
Z
i
1
2

D exp
d x
V ()
=
D eiS/~ (34)
~
2
(0,x)=a (x)
Z

U (a , b ; T )

- The integral is over all functions (t, x) of two-dimensional Minkowski spacetime:

= 0, 1, = diag(1, 1)
R
- Action S = d2 xL, where L is the Lagrangian density
- Boundary conditions (0, x) = a (x), (T, x) = b (x)
- Spacetime integral from 0 to T
- Lorentz invariant apart from boundary conditions

2
2.1

Correlation Functions
Path Integral Representation

- Practical applications:
1 , x1 )(t
2 , x2 )|0i rather than amplitudes
Correlation functions such as DF (t1 t2 , x1 x2 ) h0|T (t

 x) = U
x)U
(t)(0,
(t)
- Heisenberg picture: Field operator (t,

2 , x2 )|a ; T i:
- Start with a slightly different problem: Calculate hb ; T |(t1 , x1 )(t
x) with eigenvalue a (x)
- |a ; ti denotes the eigenstate of (t,
x)|a ; ti = a (x)|a ; ti
(t,

(35)

- Note: The t-dependence of |a ; ti is NOT Schrodinger picture time evolution

(t)|a ; 0i
- Instead: |a ; ti = U
- At initial time t = T , the system is in field eigenstate with eigenvalue a (x),
and at final time t = T in another field eigenstate with eigenvalue b (x)
- To simplify notation, write |a i |a ; 0i
- Consider the matrix element
1 , x1 )(t
2 , x2 )|a ; T i = hb |U
1) U
2 )U
(T t1 )(x
(t1 t2 )(x
(t2 + T )|a i
hb ; T |(t

(36)

- Assume that t1 > t2 so that we can repeat the steps in Section 1 Path integral representation
- Assume now that ti = k:
In Eq. (11) we obtain a factor
U
()|k1 i = k (x)hk |U
()|k1 i
hk |(x)

(37)

- Thus, all it gives is an extra factor of (t, x) in the integrand,

Z (T,x)=b (x)

hb ; T |(t1 , x1 )(t2 , x2 )|a ; T i =

D eiS (t1 , x1 )(t2 , x2 ),

(38)

(T,x)=a (x)

- Note: RHS independent of the order of s, LHS not

We assumed t1 > t2 , so the path integral gives the time-ordered correlation function
Z (T,x)=b (x)
)(y
)|a ; T i
D eiS (x )(y ) = hb ; T |T (x

(39)

(T,x)=a (x)

]:
- Easy to generalise to arbitrary operators O[
Z (T,x)=b (x)
a ; T i
]|
D eiS O[] = hb ; T |T O[

(40)

(T,x)=a (x)

- We need correlation functions in vacuum |0i:

Take T (1 i)T , and T

- Insert a complete set of energy eigenstates |ni with eigenvalues En , i.e., H|ni
= En |ni:
|a ; T i

(T )|a i = eiHT
U
|a i =

eiEn T |nihn|a i

(+i)En T

|nihn|a i = e(+i)E0 T h0|a i|0i + O eE1 T

(+i)E0 T




h0|a i|0i 1 + O e(E1 E0 )T

- Inverting this, we obtain

e(+i)E0 T
|a ; T i
T h0|a i

|0i = lim
and correspondingly

(41)

e(i)E0 T
hb ; T |
T hb |0i

h0| = lim

expressed in terms of operators (x),

]
- Thus, we find that for arbitrary O[
Z b
2E0 T
e

]|0i
D eiS O[]
h0|T O[
= lim
T hb |0ih0|a i
a
- In particular, this is true for the identity operator 1, so we can write
R

]|0i
D eiS O[]
h0|T O[

h0|T O[]|0i =
hO[]i
= R
h0|0i
D eiS
- Constants hidden in the integration measure D cancel
- In principle, the action is integrated over complex time path
Z (1i) Z
S=
dt dxL

(42)

(43)

(44)

(45)

(1i)

- Dependence on the boundary conditions of has dropped out

(unless there are degenerate vacua, e.g. spontaneous symmetry breaking!)

2.2

Generating Functional

- In order to calculate correlators (44), it is useful to define a generating functional


Z
Z
Z[J] = D exp iS + i dd xJ(x)(x)
where J(x) is an arbitrary function of spacetime
- Correlation functions are given by functional derivatives with respect to J,


(i)N

h(x1 ) (xN )i =

Z[J]
Z[0] J(x1 )
J(xn )
J=0
= 1 m2 2
- Consider free scalar field (in d dimensions) with V ()
2
- The action is


Z
1
1 2 2
d

S = d x m
2
2
- By integrating by parts, the exponent in the
Z path integral becomes
1
iS =
dd xdd y(x)M(x, y)(y),
2

(46)

(47)

(48)

(49)

- where M(x, y) is the operator



M(x, y) = i(x y) + m2
- More generally the action of any free field can be written in the form (49)
- Eq. (46) is the continuum version of a multivariate 
Gaussian integral 
Z
1
N
Z(J) = d q exp qT Mq + JT q
2
where q is an N -component vector and M is a symmetric N N matrix
- The moments of the Gaussian integral,

10

(50)

(51)

R
hqi1 qin i

dNq qi1 qin exp 12 qT Mq


R
,
dNq exp 12 qT Mq

(52)

are given by derivatives



1

hqi1 qin i =

Z(J)
Z(0) Ji1
Jin
J=0

(53)

- We can evaluate Z(J) by diagonalising the matrix M:

T,
M = M

m
1

0
m
2

ij = m
M
i ij =

..
0

(54)

m
N

and writing q =
q, J = J.
Because is orthogonal, the integral becomes


 Y Z


1 T
1
T
2

M
=
Z(J) =
d q exp q
q+J q
d
qi exp m
i qi + Ji qi
2
2
i
"r
!#


Y
1 T 1
2
Ji2
(2)N/2
=
exp
=p
exp
J M J
m
i
2m
i
2

det M
i


(2)N/2
1 T 1
=
exp
J M J
2
det M
Z

- The second moment is






1
 1
1
=
(M )jk Jk Z(J) 
hqj qj i =
Z(J)
= (M1 )ij
Z(0) Ji Jj
Z(0)
J
i
J=0
J=0

(55)

(56)

- In an nth order moment hqi1 qin i

- Half of the derivatives must act on the exponent, bringing down factors of Ji
- The other half has to act on these factors, removing them
- Each pair i, j gives a factor (M1 )ij = hqi qj i
- Thus,
hqi1 qin i = hqi1 qi2 i hqin1 qin i + (all other pairings)

(57)

hqi qj qk ql i = hqi qj ihqk ql i + hqi qk ihqj ql i + hqi ql ihqj qk i

(58)

- For example,

- This means that to evaluate the moments of a Gaussian integral all we need to know is
the two-point function hqi qj i
- In the continuum, one finds in analogy with Eq. (56),
h(x)(y)i = (M1 )(x, y)

(59)

where the inverse M1 (x, y) is defined

as
Z

d4 yM(x, y)M1 (y, z) = (x z)

- Similarly, Eq. (57) becomes Wicks theorem:

The N -point correlator is given by the sum of all possible contractions in terms of products of
11

(60)

two-point functions

2.3

Scalar Propagator

- Consider a free real scalar field in 3+1 dimensions

1
1
L = m2 2
2
2

(61)

- By integrating by parts, we can write

Z the action as


i
d4 xd4 y (x)(x y) + m2 (y)
iS =
2

(62)

- This is of the quadratic form (49), with

M(x, y) = i(x y) + m2

(63)

- The two-point function (i.e. the Feynman propagator), is given by Eq. (59)
DF (x y) = h(x)(y)i = (M1 )(x, y)
- We can invert the matrix M by taking its Fourier transform
Z

M(k, q) =
d4 yd4 x eik x M(x , y )eiq y
Z

= i d4 x eik x + m2 eiq x
Z

= i d4 x ei(k +q )x q q + m2


= i(2)4 (k + q) k 2 m2
- Inserting this into the Fourier transform of Eq. (60),
Z
d4 q
M(k, q)M1 (q, p) = (2)4 (k p),
(2)4

(64)

(65)

(66)

we find
(M1 )(k, q) = (2)4 (k + q)

k2

i
m2

(67)

- Transforming back to coordinate space, we have

DF (x y)

d4 k d4 q ik x iq y 1
e
M (k, q)
(2)4 (2)4
Z
Z

d4 k ieik (x y )
d4 k ieik (x y )
=
=
(2)4 k 2 m2
(2)4 k02 ~k 2 m2
=

(M

)(x, y) =

- Two problems:
(i) M is imaginary Integral (51) does not converge
p
(ii) The integral over k0 in Eq. (68) crosses the poles at k0 = ~k 2 + m2
The complex time used in Eq. (45) solves these problems:
- When we rotate t (1 i)t, we also have to change

12

(68)

dt (1 i)dt,
- Therefore the action becomes

Z
S = (1 i) d4 x

t
1 i t

0 =

(69)


1
1 2 2
1
2
2
(
)

(
)

,
0
i
2(1 i)2
2
2

(70)

and the matrix M becomes

02
i2 + m2
M(x, y) = ( + i)(x y)
(1 i)2



(71)

- Real part  Gaussian integral becomes well defined

- Frequency k0 turns into k0 /(1 i) and the Feynman propagator becomes
DF (x y)

(M1 )(x, y) =

1
1 i

ieik (x y )
d4 k
(2)4 (1 i)2 k02 ~k 2 m2

(72)

p
- This shifts the poles away from the real axis to k0 = (1 i) ~k 2 + m2
- Note: We are taking  0
- The residues are finite, so we can ignore the prefactor 1/(1 i)
- For the poles, it only matters which way we go around them, so we can modify the expression to
Z
DF (x y)

d4 k ieik (x y )
,
(2)4 k 2 m2 + i

(73)

p
- This moves the poles to k0 = ~k 2 + m2 i
- Lorentz invariant
- Finally, it is illuminating to note that if we had made time fully imaginary in Eq. (70), we would have found that


Z
1
1
1 2 2
2
2
4
iS d x
(0 ) + (i ) + m SE
(74)
2
2
2
- The path integral becomes real
Z

DeSE

- Same form as classical canonical partition function

- 3+1D QFT 4D classical statistical mechanics

2.4

(75)
exp(H)

Complex Scalar Field

- Consider now a complex scalar field , with Lagrangian

L = m 2

(76)

(Note the different normalisation! No 1/2s!)

- Write in terms of real and imaginary parts
1
= (R + iI )
2
- The Lagrangian becomes
13

(77)

L=

1
1
1
1
R R + I I m2 2R m2 2I
2
2
2
2

(78)

- Two uncoupled real scalar fields!

- The Feynman propagator is
h (x)(y)i =
=

1
[hR (x)R (y)i + ihR (x)I (y)i ihI (x)R (y)i + hI (x)I (y)i]
2
1
[hR (x)R (y)i + hI (x)I (y)i]
2

where we have used hR (x)I (y)i = 0

- Fields R and I are identical and their two-point functions are given by Eq. (72), so we find that
the propagator for the complex field is the same as for the real field
Z

d4 k ieik (x y )
h (x)(y)i = DF (x y) =
(2)4 k 2 m2 + i

2.5

(79)

(80)

Dirac Propagator

- Fermions: Equal-time anticommutation relations

{ (~x), (~y )} = (~x ~y ) ,

{ (~x), (~y )} = { (~x), (~y )} = 0

(81)

- Cannot be represented by normal path integrals:

Real/complex numbers commute
h{ (~x), (~y )}i = 2h (~x) (~y )i =
6 0

(82)

- Instead: Use path integrals over Grassmann numbers

- Anticommuting: For any pair and ,
=
2 = 2 = 0
- Generators of Grassmann algebra:
- Vector space: Addition and scalar multiplication
- Multiplication rule fixed by the rule for single Grassmann numbers
- Elements of algebra not anticommuting in general: (1 1 )(2 2 ) = (2 2 )(1 1 ).
- General function of a Grassman number can be written as f () = A + B
(Taylor expansion: 2 = 0)
- A and B can be any elements of the algebra
- Derivative
dconst
d
d
= 1,
=0
f () = B
d
d
d

(83)

(84)

- Acts on the first factor

d
d
= =
d
d

14


d
=
d

(85)

R
- Define integral over as an analog of dx
- Should be linear: For any elements C and DZ
of the Grassmann
algebra

Z

Z
d [f ()C + g()D] =
df () C +
dg() D

(86)

- This implies that for f () = A + B



Z
Z
Z
d B
d A +
df () =
- Invariant under arbitrary shift
Z +
Z :
d =

d( + ) =

(87)

Z
d +

(88)

- This implies
Z


d = 0

(89)

R
and because is arbitrary, d = 0
- Finally, we choose the overall normalisation
Z to be
d = 1

(90)

so that for a general function f () = A +Z B we have

d f () = B

(91)

- Note: Integral = Derivative!

- Complex Grassman numbers
=

R + iI

,
2

R iI

(92)

- Integration

d d = i
- One-dimensional
Z Gaussian integral (with
Z complex b)

d d exp( b) =

dR dI

(93)

d d(1 + b) = b

d d(1 b) =

(94)

- n-dimensional complex Gaussian integral (i = 1, . . . , n, Hermitean Mij , sum over i,j )

Z

n
Y
i=1

!
di di

exp

i Mij j

X
1
k!

ij

di di

exp (j i Mij )

!
Y

k=0

- Only the k = n term survives:

For k > n some i has to appear twice Vanishes because i2 = 0
R
For k < n some i will not appear at Vanishes because di = 0

15

di di

(j i Mij )

(95)

di di

exp (i Mij j )

1
=
n!

!
Y

di di

(j i Mij )

(96)

- Each i can only appear once: n! such terms

Factors j i commute Put s in decreasing numerical order




1
n
(j i Mij ) = n in Min ,n 1 i1 Mi1 ,1
n!

(97)

- Put the i factors in the same order: Gives a factor i1 ,...,in
(Levi-Civita tensor, 12...n = 1, fully antisymmetric)
1
n
(j i Mij ) = i1 ,...,in Mi1 ,1 Min ,n n n 1 1
n!

(98)

and therefore
Z

!
Y

di di

exp (i Mij j ) = i1 ,...,in Mi1 ,1 Min ,n

!
Y

di di

n n 1 1

(99)

- The integral is equal to 1, and the prefactor is just the definition of a determinant
det M = i1 ,...,in Mi1 ,1 Min ,n

(100)

and therefore we have

Z

!
Y

di di

exp (i Mij j ) = det M

(101)

(cf. normal complex numbers (2)N / det M)

- External field (Grassmannian)
!
Z Y

Z( , )
di di exp (i Mij j + i i + i i )

(102)

- Shift i i + (M1 )ij j

i i + j (M1 )ji

Z( , )

!
Y

di di

exp i Mij j + i (M1 )ij j

= e

i (M1 )ij j

det M

(103)

- The second moments can be calculated as derivatives





i (M1 )ij j 
1


Z(
,
)
=
e
hi j i =


Z(0, 0) j i
j i
==0
==0
=

(M1 )ij

(104)

- The same result as for normal numbers (but keep track of the order)
- Dirac Lagrangian
L = (i/ m)

16

(105)

where = 0 , / = ,
and , = 1 . . . 4 are spinor indices
- The action is
Z
iS =

d4 xd4 y (x)i(x y)(i/ m) (y)

(106)

Matrix M
M (x, y) = i(x y)(i/ m)

(107)

- Feynman propagator
SF, (x y) h (x) (y)i = (M1 ) (x, y) =

3
3.1

d4 k
(2)4

ieik (x y )
k/ m + i


(108)

Gauge Fields
Abelian Symmetry

- The QED Lagrangian is

1
D
/ m),
L = F F + (i
4

(109)

where is the electron field (and = 0 )

F = A A is the field strength tensor
D = + ieA is the covariant derivative
- Path integral:
R
iS

]
DA DDe
O[A , ,
R
hOi =
.
iS

DA DDe

(110)

- To derive the photon propagator, let us focus on the Maxwell term


1

;
F = A A
d x F F
4
Z


1
d4 xA (x) 2 g A (x);
g = diag(1, 1, 1, 1)
2
Z


1
d4 k
A (k) k 2 g + k k A (k)
4
2
(2)

Z
S

=
=
=

(111)

- The matrix we need to invert is therefore

M (k, q) = i(2)4 (k + q) k 2 g k k

(112)

- Unfortunately, this matrix cannot be inverted:

- Zero eigenvalue:
M (k, q)k = 0
singular
- The reason is gauge invariance: Gauge transformations
17

(113)

(x)
A (x)

ei(x) (x)
1
A (x) (x)
e

(114)

which in momentum space is

i
A (k) A (k) k (k)
e

(115)

does not change the action (or indeed any physical observable)
- Momentum space: Longitudinal component AL
k unphysical
- The path integral will be infinite (even with complex time)
- Will be cured by gauge fixing

3.2

Non-Abelian Symmetry

- This is a brief reminder of the basic features of non-Abelian gauge theories.

For more details, see the Unification course, Chapter 15 in Peskin&Schroeder
or Chapter 9 in Bailin&Love
- SU(N ) gauge symmetry + fermions in fundamental representation (for QCD, N = 3)
- Other gauge groups and scalar fields follow the same lines
- The Lagrangian is
1
D
L = TrF F + (i
/ m),
2
where - is an N -component vector consisting of spinors i
(i = 1, . . . , N labels different colours)
- D = + igA is the covariant derivative
(be careful with the sign of g: Peskin&Schroder use -)
- The gauge field A is a traceless, self-adjoint(=Hermitean) N N matrix
(element of the SU(N ) Lie algebra)
- The field strength tensor is
i
F = [D , D ] = A A + ig[A , A ]
g

(116)

(117)

- The Lagrangian is invariant under transformation of the form

(x) U (x)(x),

(118)

where U (x) is an element of the Lie group SU(N )

[= unitary (U U = 1), special (det U = 1), N N matrices]
provided that the covariant derivative transforms as
D (x) U (x)D (x)U (x)
[because then F (x) U (x)F (x)U (x)]
- From this, we can determine how the gauge field has to transform
- Let us write

18

(119)

 + ig A
D D

(120)

- According to Eq. (119), this is equal to



i

D = U D U = U ( + igA )U = + ig U A U U U
g

(121)

- Therefore A must transform as

i
A U A U U U
g

(122)

(Check that A remains traceless and self-adjoint!)

- Let us then count the number of independent degrees of freedom in the field A :
- A general complex N N matrix has 2N 2 real degrees of freedom
- Self-adjoint: Removes half of them N 2
- Traceless: Removes one
Altogether 2N 2 N 2 1 = N 2 1 real degrees of freedom
- We can write A as a linear combination of matrices ta
A = Aa ta ,

a = 1, . . . , N 2 1

where Aa are real numbers

- The matrices ta are known as the generators of the group
- In principle any set of linearly independent, traceless, self-adjoint N N matrices would do
In practice, it is conventional to choose them in such a way that
1
Trta tb = ab
2

(123)

(124)

- The matrices do not commute, but the commutator is traceless and anti-self-adjoint, so we can write
[ta , tb ] = if abc tc ,

(125)

where the coefficients f abc are known as structure constants

- Note that
Tr [ta , tb ]tc = if abd Tr td tc =

i abc
f ,
2

(126)

from which it follows that f abc is cyclic and fully antisymmetric

- For SU(2) one can choose ta = a /2, where a are the Pauli matrices
Then the structure constants are given by the Levi-Civita tensor f abc = abc
- Furthermore, the structure constants satisfy the Jacobi identity
f ade f bcd + f bde f cad + f cde f abd = 0
- Other representations: Any set of p p matrices T a which satisfy Eq. (125)
- is then a p-component vector, and D = + igAa T a
a
- E.g. adjoint representation Tbc
= if abc
- For more details on groups, representations etc. see Peskin&Schroeder 15.4
(Note the standard set of SU(3) generators on p.502!)

19

(127)

- Yang-Mills Lagrangian
1
L = Tr F F
2

(128)

- In terms of the components Aa , the field strength tensor is

F

= A A + ig[A , A ]
=

( Aa Aa )ta + igAb Ac [tb , tc ]

( Aa Aa gf abc Ab Ac )ta

(129)

a a
- Thus, if we write F = F
t , we have
a
F
= Aa Aa gf abc Ab Ac

(130)

- The Lagrangian Eq. (128) is

L

=
=
=

1 a a
F
F
4
1
( Aa Aa gf abc Ab Ac )( Aa Aa gf ade Ad Ae )
4
1
( Aa Aa )( Aa Aa )
4
g2
g abc
+ f ( Aa Aa )Ab Ac f abc f ade Ab Ac Ad Ae
2
4

(131)

- Not purely quadratic: Interacting theory

- Limit g 0: Free theory
- Simply (N 2 1) copies of the Maxwell term (111)
Singular, cannot be inverted: We need to fix the gauge

3.3

Gauge Fixing

- Consider the integral

Z

DA O[A ]eiS[A ]

(132)

AU
= U A U U U
g

(133)

IO =
- Observables given by ratios hOi = IO /I1 .
- Denote gauge transformed field by

- Gauge invariance: S[AU

] = S[A ]
(Gaussian) integral diverges
- Fix the gauge:
- Impose a gauge condition G[A ] = 0 that removes the ambiguity
- G[A ] is some functional of A , which maps a field configuration A (x) to
a Lie-algebra-valued function of x, with N 2 1 real-valued components Ga [A ](x)
- For instance G[A ] = A (Lorenz gauge)
- For a given fixed field configuration A (x), G[AU
] defines a map from gauge transformations U (x)
20

to Lie-algebra-valued functions of x
- Write the unity as an integral over a delta function
Z
Z


1 = DG(G) = DU G[AU
] det

G[AU
]
U

!
,

(134)

where DU is the Haar measure (the unique invariant measure for the Lie group)
and the Jacobian determinant when changing the integration variable from G to U
- Invariant under multiplication by aZconstant group element
V
Z
DU f (U ) =

DU f (U V )

(135)

- This determines the measure uniquely up to a multiplicative constant

- The integral gets a contribution only from point U0 , where G[AU0 ] = 0
Consider infinitesimal region around U0
- We write U = (1 + i)U0 where is infinitesimal
self-adjoint (so that U is unitary)
and traceless (so that det U = 1)
- We can therefore write = a ta
- For ||  1, the integration measure becomes
Y
DU
Da

(136)

- We can write the Jacobian explicitly as

!
G[AU
]
det
det
U

(1+i)U0

Ga [A
b

!
(137)

- Let us now consider a particular choice of the gauge fixing functional

G[A ] = A (x) (x)

(138)

where (x) is some fixed Lie-algebra-valued function

- This means imposing the gauge condition A (x) = (x)
- To compute the Jacobian, we Taylor expand in
0
AU

0
= A(1+i)U0 AU

i
U0
0
(1 + i)AU
(1 i) (1 + i) (1 i) A
g

a
1
1
0
0 c
= i[, AU
a + gf abc b (AU
] =
) t
g
g
=

(139)

- Thus,
a
(AU
)

=
=

1
c
a + gf abc b (AU
)
g

b
1
1 U ab b
c
ab + gf cab (AU
) (D ) ,
g
g

(140)

2
where we can now think of AU
and as (N 1)-component vectors rather than N N matrices

21

- The operator (DU )ab is the covariant derivative in the adjoint representation
- Therefore we find
a
(AU
Ga [AU
1
)
]
=
= (DU )ab
b
b

(141)

- We will omit the indices ab from now on. This operator can be thought of as a multidimensional
matrix with indices labelled by a = 1, . . . , (N 2 1) and spacetime points.
- Thus, we have found that

 Z
Z






1 U
U
1 = DU G[AU
]
det

D
DU G[AU
(142)

] det i D ,
g
where we dropped the factor i/g because it gives an uninteresting constant factor
- Inserting Eq. (142) into Eq. (132)
Z (and changing the order of integrations) we obtain




U
iS[A ]
IO = DU DA G[AU
] det i D O[A ]e

(143)

- We can change the integration variable from A to A AU

, and the measure remains unchanged
DA = DA
(144)
because this only involves a unitary rotation and a shift

- Note that A = AU

- Because of gauge invariance, S[A ] = S[AU

] = S[A ]
- Assuming that the correlator we are calculating is also gauge-invariant,

i.e., O[A ] = O[AU

] = O[
ZA ], we have



O[A ]eiS[A ]
IO = DU DA G[A ] det i D

(145)

- We can now relabel A AZand find

IO =

DU DA (G[A ]) det (i D ) O[A ]eiS[A ]

(146)

- The integrand is now independent

of ZU , and the integral factorizes
Z
IO =

DU

DA (G[A ]) det (i D ) O[A ]eiS[A ]

(147)

- Integral over U gives a constant: Volume of the gauge group!

- Finite for compact groups such as SU(N )
(Gaussian integral diverges, but it is only an approximation)
- Will cancel from any correlator, because they are ratios of these integrals
- Writing the gauge constraint explicitly,
we have
Z
IO IO []

DA Aa a det (i D ) O[A ]eiS[A ]

- The integral IO [] is independent of a

- We average over a (x) with a Gaussian weight

22

(148)

 Z

a a
D IO [] exp i d4 x
2
 Z

Z
a a

DDA exp i d4 x
Aa a det (i D ) O[A ]eiS[A ]
2
!
Z
Z
a 2
( A )
4
det (i D ) O[A ]eiS[A ]
DA exp i d x
2
Z

IO

where we changed the order of the integrations

- We have exchanged the delta function to an extra Gaussian factor
- Can be absorbed into the action as a gauge fixing term


2
Z
R
( Aa
)
i S[A ] d4 x
2

IO DA det (i D ) O[A ]e

(149)

(150)

- Physical quantities are independent of the gauge fixing parameter : Useful check!
- According to Eq. (101), a determinant can be written as a Gaussian Grassmann integral
- Thus, we introduce a (N 2 1)-component Grassmann-valued field ca , and write
det (i D )

 Z

Dc Dc exp i d4 xc D c
 Z

Z


Dc Dc exp i d4 x ( ca ) ca gf abc Ab cc
Z

=
=

(151)

- This field is:

- Anticommuting fermion field
- Scalar (and thus violates the spin-statistics theorem)
- In the adjoint representation
- Known as the Faddeev-Popov ghost field
- Thus, we have been able to write the integral
IO as
Z
IO =

DA Dc DcO[A ]ei

d4 xL

(152)

with the gauge-fixed Lagrangian

L

1
= ( Aa Aa )( Aa Aa )
4
g abc
g2
+ f ( Aa Aa )Ab Ac f abc f ade Ab Ac Ad Ae
2
4
1 a 2
a
a
abc a b c
( A ) + c c gf c A c
2

- By construction, this Lagrangian gives the same expectation values for gauge-invariant operators
as the original Lagrangian, so it describes exactly the same physics, but it is not gauge invariant
- The result can be directly applied to the Abelian case, too
- The U(1) group has only one generator t1 = 1

23

(153)

- The colour index a has only one possible value a = 1

- The only structure constant vanishes f 111 = 0
- All the interaction terms vanish, including the photon-ghost interaction
- Therefore the ghost decouples and can be ignored

3.4

Gauge and Ghost Propagators

- For photons (i.e. Abelian gauge symmetry), the gauge-fixed action is



1
1

2
d x F F ( A )
4
2




Z
1
4 1
2

d x A g 1
A
2

Z
S

=
=

(154)

- Our matrix M becomes





1
2

M (k, q) = i(2) (k + q) k g 1
k k

(155)

- This is invertible and gives the photon propagator

1

hA (x)A (y)i =

(x, y) =

d4 k ieik (x y
(2)4
k 2 + i



k k
g (1 ) 2
k

F
D
(x y)

(156)

- Convenient choices for :

= 0 Landau gauge (DF transverse)
= 1 Feynman gauge (DF looks like scalar propagator)
- For gluons (i.e., non-Abelian gauge symmetry), each component has the same propagator as the photon


Z
d4 k ik (x y ) i ab
k k
F
hAa (x)Ab (y)i = ab D
(x y) =
e
g

(1

)
(157)

(2)4
k2
k2

- For ghosts, the free action is

Z
S=

d4 xca ab 2 cb ,

(158)

so the matrix to invert is

Mab (k, q) = i(2)4 (k + q) ab k 2

(159)

- Therefore, the propagator is

hca (x)cb (y)i =

d4 k ik (x y ) i ab
e
(2)4
k 2 + i

24

(160)

3.5

Summary of Propagators

- Propagators in momentum space:

- Scalar field:
DF (k) =

i
k 2 m2 + i

(161)

i
k/ m + i

(162)

- Dirac fermion:
SF (k) =
- Photon:
DF (k) =



i
k k

(1

)
k 2 + i
k2

(163)

- Gluon:
ab DF (k) = ab



i
k k

(1

)
k 2 + i
k2

(164)

i
+ i

(165)

- Ghost:
ab DF (k) =

4
4.1

k2

Interaction Vertices
Scalar Theory

- Consider now a (weakly) interacting scalar field theory

1
1
1
L = m2 2 4 = L0 + LI
2
2
4!

(166)

where L0 is the free Lagrangian (61) and LI is the interaction part

but this is now non-Gaussian
- We can write full expectation values hOiI in terms of free ones,
 R

DO[] exp i d4 x(L0 + LI )
 R

R
D exp i d4 x(L0 + LI )
 R

hO[] exp i d4 xLI i0
 R

hexp i d4 xLI i0
R

hOiI

where h i0 is the expectation value in the free theory

- Therefore, we need to calculate free expectation values of the form
 Z

Z
k
X ik
4
4
hO[]
d xLI i0 ,
hO[] exp i d xLI i0 =
k!
k

which are given by Gaussian path integrals

25

(167)

(168)

- If O is polynomial, all we need are terms like


k Z
1 i
d4 y1 d4 yk h(x1 ) (xn )(y1 )4 (yk )4 i0
k!
4!

(169)

- Gaussian expectation value:

Sum of all possible pairings into products of h(x)(y)i0 = DF (x y)
- In particular, we want to calculate n-point correlation functions in the interacting theory
Gn (x1 , . . . , xn ) h(x1 ) (xn )iI

(170)

- As an example, consider the two-point function G2 (x1 , x2 ) h(x1 )(x2 )iI

- To linear order in , we have
G2 (x1 , x2 )

=
=

R 4
d y(y)4 i0
h(x1 )(x2 ) 1 i
4!
R
+ O(2 )
4 y(y)4 i
h1 i
d
0
4!
Z
i
h(x1 )(x2 )i0 h(x1 )(x2 ) d4 y(y)4 i0
4!
Z
i
+ h(x1 )(x2 )i0 h d4 y(y)4 i0 + O(2 )
4!

(171)

- In the second term, we have to consider pairings of hh(x1 )(x2 )(y)(y)(y)(y)i0 :

- If x1 is connected to x2 ,
h(x1 )(x2 )(y)(y)(y)(y)i0

(172)

we obtain

i
3 DF (x1 x2 )
4!

d4 yDF (y y)2

(173)

where the factor 3 arises because the first (y) can choose between 3 other (y)s
- If x1 is connected to y,
h(x1 )(x2 )(y)(y)(y)(y)i0

(174)

we have

i
43
4!

d4 yDF (x1 y)DF (x2 y)DF (y y)

(175)

where the factor 4 3 = 12 arises because x1 can choose between 4 and x2 between 3 (y)s
- The third term only has one possible contraction,
h(x1 )(x2 )i0 h(y)(y)(y)(y)i0

(176)

and gives
i
3 DF (x1 x2 )
4!
which cancels Eq. (173)
- Thus the whole two-point function is

26

d4 yDF (y y)2

(177)









G2 (x1 , x2 ) = DF (x1 x2 )

i
DF (0)
2

d4 yDF (x1 y)DF (x2 y)

(178)

- We can give this result a simple pictorial representation in terms of Feynman diagrams:
- Spacetime points xi are represented by points in the diagram
- Propagator DF (x y) is represented by a dashed line connecting points x and y:
x

- The two-point function looks like

G2 (x1 , x2 )

i
3

4!

+43

i
+
4!

i
2

(179)

- Consider then the four-point function G4 (x1 , x2 , x3 , x4 )

- Two-to-two scattering process
- To linear order in , we have
G4 (x1 , x2 , x3 , x4 )

h(x1 )(x2 )(x3 )(x4 )i0

Z
i
h(x1 )(x2 )(x3 )(x4 ) d4 y(y)4 i0
4!
Z
i
+ h(x1 )(x2 )(x3 )(x4 )i0 h d4 y(y)4 i0 + O(2 )
4!

- The leading term gives

h(x1 )(x2 )(x3 )(x4 )i0

(180)

DF (x1 x2 )DF (x3 x4 ) + DF (x1 x3 )DF (x2 x4 )

+DF (x1 x4 )DF (x2 x3 )

- Particles propagating between points without interacting

27

(181)

- The second term consists of diagrams with five points: four lines end at y
- The ones in which all xi s disconnected from y,
h(x1 )(x2 )(x3 )(x4 )(y)(y)(y)(y)i0 ,
giving

+ permutations

(182)

(183)

are again cancelled by the third term: Generally the case for disconnected bubbles!
- There are six diagrams in which one pair of xi s is connected to each other and the rest to y,
h(x1 )(x2 )(x3 )(x4 )(y)(y)(y)(y)i0
corresponding to

+ permutations

Each gives a contribution

Z
i
4 3 DF (x1 x2 ) d4 yDF (x3 y)DF (x4 y)DF (y y)
4!

(184)

(185)

(186)

where the factor 4 3 arises because x3 can choose between 4 legs of y and
x4 can choose between 3 remaining legs
- Finally there is one diagram in which all xi s are connected to y,
h(x1 )(x2 )(x3 )(x4 )(y)(y)(y)(y)i0
corresponding to

and giving
Z
i
4 3 2 1 d4 yDF (x1 y)DF (x2 y)DF (x3 y)DF (x4 y)
4!
The factor 4 3 2 1 = 4! arises because x1 can choose between 4 (y)s etc.
- In summary, the diagrammatic expression for G4 (x1 , x2 , x3 , x4 ) is

28

(187)

(188)

(189)

G4 (x1 , x2 , x3 , x4 )


  


+

(190)

- It is easy to read the full expression from the diagrams using these rules:

The n-point function is given by the sum of all diagrams with n external legs, arbitrary number of
four-point vertices and no disconnected bubbles (as we will see later)
Each four-point vertex gives

Each line gives

i
4!

d4 y

DF (x y)

(191)

(192)

Multiply by the number of contractions leading to the same diagram

Multiply by 1/(number of vertices)! from the Taylor expansion of the exponential

- For example, the so-called sunset diagram

x1

corresponds to

x2

2
Z
i
8 4 3 2 d4 y1 d4 y2 DF (x1 y1 )DF (y2 y1 )3 DF (y2 x2 )
4!
Z
(i)2
=
d4 y1 d4 y2 DF (x1 y1 )DF (y2 y1 )3 DF (y2 x2 )
6
1
2!

- The numerical factors are usually combined to one symmetry factor:

- If the diagram consist of k vertices, the integral is divided by

29

(193)

(194)

k!(4!)k
number of different contractions

(195)

- Here the numerator counts the total number of possible permutations of the legs of the internal vertices:
Any such permutation will give the same topology (so represented by the same diagram)
- When this is divided by the number of different contractions, it gives the number of permutations
that give the same contraction:
This is the order of symmetry of the diagram
- For example, the symmetry factor for the sunset diagram (193) is 1/6 because it has a six-fold
symmetry under permutations of internal lines
- It is generally more convenient to work in momentum space, because the propagator is simpler
- Writing the interaction term in terms of the Fourier transformed field
Z
d4 k ikx
e
(k),
(x) =
(2)4

(196)

we find
Z

iSI

Z
i
d4 xLI =
d4 x(x)4
4!
Z 4
i
d k1 d4 k2 d4 k3 d4 k4
(2)4 (k1 +k2 +k3 +k4 )(k1 )(k2 )(k3 )(k4 ) (197)
=
4!
(2)4 (2)4 (2)4 (2)4
= i

- The free two-point function is

h(p)(q)i0 =

i
(2)4 (p + q)
p2 m2

(198)

- The delta functions simply enforce momentum conservation at all vertices

- Fixes all but loop momenta
- Sign choice: Positive momentum = inward
- Thus, repeating the same arguments as in the coordinate space, we find the Feynman rules




1. The n-point function is given by the sum of all diagrams with n external legs, arbitrary number of
four-point vertices and no disconnected bubbles
2. Each four-point vertex gives
k1
k2
i

(momentum conservation k1 +k2 +k3 +k4 = 0)

(199)

k3
k4
More generally, write iS in momentum space, and the vertex is the coefficient of the corresponding
term multiplied by the number of permutations of legs
3. Each line gives
p

i
p2 m2

(momentum conservation p+q = 0)

4. Multiply by the symmetry factor (i.e., divide by the order of symmetry)

30

(200)

5. Integrate over loop momenta

- Example: Sunset diagram (193)
(i)2
i
i
(2)4 (p + q)
6 p2 m2 q 2 m2

4.2

i
i
i
d4 k1 d4 k2
(201)
(2)4 (2)4 k12 m2 k22 m2 (pk1 k2 )2 m2

Connected and 1PI Correlators

- Consider the generating functional Z[J] defined in Eq. (46)

- Diagrammatic expansion:
- External field gives one-point vertex with coupling J(x)

iJ(x)

- Z[J] is the sum of all diagrams with

- Arbitrarily many separate pieces
- Arbitrarily many one-point vertices iJ(x)
- No external legs
- Any such diagram factorises:
- full diagram = (bubbles) (J-dependent piece)
where bubbles are diagrams with no one-point vertices
- Therefore the whole sum factorises, too
X
X
X
Z[J] =
(all diags) =
(bubbles)
(J-dependent)

(202)

(203)

- The sum of bubbles is simply Z[0], so Z[J]/Z[0] is given by all diagrams with no bubbles
- Therefore the same applies to correlators GN given by Eq. (47), as well
- Now, consider diagram with k separate pieces
- Whole integral = Product of integrals corresponding to each individual piece
- Symmetry: (permutations of pieces) (symmetry of each connected piece)
- Organise the sum as a sum over k:
Z[J]

=
=

sum over all diagrams

X
1
(sum over all connected diagrams)k
k!
k=1

exp(sum over all connected diagrams)

(204)

- Define generating functional of connected correlators

E[J] = i ln Z[J]

31

(205)

- Given by sum over all connected diagrams

- Analogous to free energy in statistical physics (F = kB T ln Z)
- Define connected correlator GN (x1 , . . . , xN ) as functional derivative



N +1
GN (x1 , . . . , xN ) = h(x1 ) (xN )iconn = (i)

E[J]
J(x1 )
J(xN )
J=0

(206)

- Given by diagrams in which all vertices are connected to each other

- Examples:
- Connected two-point correlator






G2 (x, y) = i
=
E[J]
ln Z[J]
J(x) J(y)
J(x)
J(y)
J=0
J=0


1 Z[J] 1 Z[J] 
1
Z[J] 
+
=
Z[J] J(x) J(y) J=0
Z[J] J(x) Z[J] J(y) J=0
= h(x)(y)i h(x)ih(y)i

(207)

- If the one-point function h(x)i vanishes, as it does in our theory, then G2 (x, y) = G2 (x, y)
- Connected four-point correlator

 !
"# $
G4 (k1 , k2 , k3 , k4 )

= G4 (k1 , k2 , k3 , k4 ) G2 (k1 , k2 )G2 (k3 , k4 )

G2 (k1 , k3 )G2 (k2 , k4 ) G2 (k1 , k4 )G2 (k2 , k3 )

(208)

- More generally, subtract all possible ways of expressing in terms of products of lower correlators
- To quadratic order in
G4 =

+4

+3

(209)

- The second diagram factorises

- In fact, any diagram with separate parts that are connected by only one propagator factorises
into a product of two separate integrals
k3
k6
k3
k6

k2

k5 = k2

k1

k4

k 0

i
k0
k 02 + m2

k1

k5

k4

with k 0 = k1 + k2 + k3 etc
- Define one-particle irreducible (1PI) diagrams:
Cutting any single line will not split the diagram into two disconnected pieces
- 1PI correlator n (k1 , . . . , kn ):
- Sum of all 1PI diagrams with n external legs
- Generating functional: Effective action [c (x)]
- Defined as the Legendre transform

32

(210)

Z
[c ] = E[J]

d4 xJ(x)c (x),

E[J]
= c (x)
J(x)

(211)

- Correlators as derivatives
n (x1 , . . . , xn ) = i

[c ]
c (x1 )
c (xn )

%
&'()

(212)

- The external leg propagators are not included, e.g. in Eq. (193)
i
i
2
2
2
p m q m2

(213)

- Also conventionally leave out the momentum conservation delta when in momentum space, e.g.
(2)4 (p + q)

(i)2
6

(214)

i
i
i
d4 k1 d4 k2
(2)4 (2)4 k12 m2 k22 m2 (pk1 k2 )2 m2

(215)

- Connected and full correlators can be expressed in terms of 1PI correlators

- When calculating them, remember to include
- A propagator for each external leg
- Overall momentum conservation delta function
- Connected n-point correlator Gn (black circle):
- Given by sum of all tree level (i.e. no loops) diagrams with 1PI correlators i with i n as vertices (shaded circles)
- First example: two-point function
- All diagrams consist of an arbitrary number of 2 s in a row
=

- Thus,

G2 (p, q)

k

X
i
i

(p)
2
p2 m2
p2 m2
k=0

1
i
i
= (2)4 (p + q) 2
1

(p)
2
p m2
p2 m2
i
= (2)4 (p + q) 2
2
p m i2 (p)
=

(216)

G2 (p, q) = (2)4 (p + q)

- Second example: four-point function

- Symmetry: Odd n-point correlators vanish
The only contributing 1PI vertices are 2 and 4
- The only way to make a 4-point diagram: One 4 and any number of 2 s in legs
Legs are just connected two-point functions

33

(217)

*+
,/-0.
1
G4 =

(218)

- If the 1PI three-point function 3 was non-zero, it would also contribute and we would have

G4 =

(219)

- The 1PI contribution is

+3

4 =

(220)

- The second diagram gives the integral

=

4.3

(i)2
2

d4 p
i
i
4
2
2
(2) p m (p + k1 + k2 )2 m2

(221)

2
( )
4

(222)

Complex Scalar

- Interacting Lagrangian
L = m2

- Note that the normalisation of the interaction term is different from the real scalar:
There are only 4 rather than 4! equivalent permutations of the factors and
- Interaction term
Z
Z
i
d4 x (x) (x)(x)(x)
i d4 xLI =
4
Z
i
d4 k1
d4 k4 i(k1 +k2 k3 k4 )x
(k1 ) (k2 )(k3 )(k4 )
=
d4 x

e
4
(2)4
(2)4
Z 4
i
d k1
d4 k4
=

(k1 + k2 k3 k4 ) (k1 ) (k2 )(k3 )(k4 ) (223)

4
(2)4
(2)4
- Feynman rules:
- Propagator
(q)
- Vertex

(k)

i
k 2 m2

34

(mom cons k q = 0)

(224)

3
4
(k1 )

(k2 )
i

(mom cons k1 + k2 k3 k4 )

(225)

(k3 )

(k4 )
- Note:
- Lines have a direction:
- On vertices, outgoing lines creation operators
incoming lines annihilation operators
- A propagator line goes from a creation operator to an annihilation operator
- Thus: The end of a line is attached to a leg of a vertex etc.
- Momentum is defined in the direction of the arrow
- Example: Sunset in complex theory
k1

p+k1 +k2

h (p)(q) i

k2

2
Z 4
1
i
d k1 d4 k2
i
i
i
=

422
2!
4
(2)4 (2)4 k12 m2 k22 m2 (p k1 k2 )2 m2
Z 4
(i)2
d k1 d4 k2
i
i
i
=
2
(2)4 (2)4 k12 m2 k22 m2 (p k1 k2 )2 m2

(226)

where we have again omitted the external propagators

- The numerical factor 1/2: Symmetry between the top and bottom arcs

4.4

QED

- Theory of fermions coupled to photon field

1
D
L = (i
/ m) F F
4

(227)

where D = + ieA and F = A A

- Free part
/ m) 1 F F
L0 = (i
4
- Interaction part (with spinor indices , written explicitly)
A = e A
LI = e

- Momentum space
Z
Z 4
d k1 d4 k2 d4 k3

i d4 xLI = ie
(2)4 (k1 k2 k3 ) (k1 ) (k2 )A (k3 )
(2)4 (2)4 (2)4
- Feynman rules:
- Interaction vertex
35

(228)

(229)

(230)

56
7
8

k1

k3

ie

(momentum cons k1 k2 k3 = 0)

(231)

k2

- Propagators

A (k)

A (k)

(k)

(k)

DF (k)



i
k k
= 2 g (1 ) 2
k
k

SF (k) =

i
k/ m

(232)


(233)

- Trace over each fermion loop

- Factor (1) for each fermion loop
(Need to anticommute one pair)
- Example:
p+k

A (p)

1 k
(2)
2!
Z
= (ie)2

Ai
A (p) hA (p)A (p) A




d4 k 

ie
SF (k) ie SF (p + k)
4
(2)


d4 k
i
i

tr

(2)4
k/ m p/ + k/ m

(234)

- Third line: Trace over spinor indices

Order of matrix factors opposite to the direction of the arrow in loop

4.5

QCD

- SU(N ) gauge field + fundamental fermions

- After gauge fixing, the Lagrangian
(116)

 becomes
 L=
 L0 + LI , with
1
1 a
2

A2 ca 2 ca + i (i/ m) i
L 0 = A g 1
2

(235)

and
LI

g abc
g2
f ( Aa Aa )Ab Ac f abc f ade Ab Ac Ad Ae
2
4
a a
gf abc ca Ab cc g
tij A i j

36

(236)

9:
;

- There are three different fields:

- Gluons (gauge field): Each component has the same propagator as the photon


i ab
k k
Aa (k)
ab DF (k) =
g

(1

)
Ab (k)
k2
k2
k
- Quarks (fermions): Each component has the same propagator as the electron


i
i (k)
ij SF (k) = ij
j (k)
k/ m
k

(237)

(238)

- Ghosts: Just like charged scalars

cb (k)

ca (k)

i ab
k2

(239)

- There are also four types of vertices:

- The three-gluon vertex arises from the term
Z
Z
ig
4 ig abc a
a
b c
iS = . . . + d x f ( A A )A A = . . . + d4 x f abc (g Aa g Aa )Ab Ac
2
2
(240)
- Going to momentum space Aa ik1 Aa (k1 ), and we obtain
Z

g
d4 k1 d4 k2 d4 k3
(2)4 (k1 + k2 + k3 ) f abc (g k1 g k1 )Aa (k1 )Ab (k2 )Ac (k3 )
4
4
4
(2) (2) (2)
2

<

- In the symmeric form, the integrand is

g abc
f [g (k2 k1 ) + g (k3 k2 ) + g (k1 k3 )] Aa (k1 )Ab (k2 )Ac (k3 )
6

(241)

(242)

- Including a factor 3! counting permutations of legs, the Feynman rule is

Aa (k1 )

Ab (k2 )

gf abc [g (k2 k1 ) + g (k3 k2 ) + g (k1 k3 )] (243)

Ac (k3 )

- The four-gluon vertex is given by the term

g 2 abe cde a b c d
f f A A A A
4

g 2 abe cde a b c d
f f g g A A A A
4


g 2  abe cde
=
f f
g g g g
24


+f ace f bde g g g g


+f ade f bce g g g g Aa Ab Ac Ad
=

and including the factors i and 4! we have the rule

37

(244)

Aa

Ac

=
>
?
Ab

Ad




ig 2 f abe f cde g g g g

+f ace f bde g g g g

+f ade f bce g g g g



(245)

- The ghost-gluon interaction term is

gf abc ca Ab cc igf abc k1 ca (k1 )Ab (k2 )cc (k3 )

and therefore we have

Ab (k2 )

gf abc k1

ca (k1 )
cc (k3 )
- Finally, the quark-gluon coupling is given by the term
g ta Aa i j
ij

whereby

Aa

a
ig
tij

Part II

Renormalisation

5.1

(247)

(248)

j
i
- In addition, we have the usual rules that we integrate over all loop momenta
and have a factor (1) for each fermion (quark or ghost) loop

(246)

Perturbative Renormalisation
Ultraviolet Divergences

- 1PI two-point correlator 2 at one loop

38

(249)

2 (p) =

i
2

i
d4 k
4
2
(2) k m2

- Need to evaluate the integral

Z
Z
Z
1
d4 k
dk0
4
3
I1 (m) =
=
(2)
d
k
(2)4 k 2 m2
k02 ~k 2 m2 + i
p
- The k0 integrand has two poles at k0 = ( ~k 2 + m2 i)
- Rotate the integration contour

(250)

(251)

Im

Re

dk0
k02 2 + i

iR
Z
+i

/2

dei
2

dkE
2

2
kE
Z /2

iR
Z
i



k0 = Rei

R2 e2i

[k0 = ikE ]

dei
R2 e2i 2
dkE
2 + 2
kE



k0 = Rei
(252)

- This implies
4

I1 (m) = i(2)

d k

dkE
= i
2
kE + ~k 2 + m2

Z
Eucl

d4 k
1
(2)4 k 2 + m2

where the subscript Eucl indicates a four-dimensional Euclidean integral

- In spherical polar coordinates
Z
Z
i
k 3 dk
I1 (m) =
d
4
4
2
(2)
k + m2
0
R
where d4 is the integral over the angles, i.e., the volume of S 3
- We can calculate it by considering

39

(253)

(254)

d/2

Z
=
=

x2

d

~
x2

dxe
= d xe
= dd
dr rd1 er

0
Z
Z
Z
1
(d/2)
d/21 z
dd
dz z
dd
e =
2
2
0
Z
2 d/2
dd =
(d/2)
d

(255)

where (x) is the gamma function and satisfies x(x) = (x + 1) and (n) = (n 1)! for n Z
- Since (2) = 1, we find
Z
k 3 dk
i
(256)
I1 (m) = 2
8 0 k 2 + m2
- The k integral is obviously divergent
- Introduce a ultraviolet cutoff
I1 (m)



Z
Z 2
Z 2
i
k 3 dk
i
i
udu
m2
=

du
1

8 2 0 k 2 + m2
16 2 0 u + m2
16 2 0
u + m2




i
2 + m2
i

4
2
2
2
2
2
=
+
O(m
/
)
(257)

m
ln
=

2m
ln
16 2
m2
16 2
m
=

A B

- Diverges at
- Has both quadratic (2 ) and logarithmic (ln ) divergences
- Therefore, the two-point function is



2
2
2 (p) = i
2m ln
32 2
m

(258)

- Let us then look at the four-point function 4

4 (k1 , k2 , k3 , k4 ) =

+3

(259)

where the factor 3 in the second term stands for the three inequivalent permutations of external legs
- The second diagram gives the integral
(i)2
=
2

d4 p
i
i
(i)2

I2 (k1 + k2 , m)
(2)4 p2 m2 (p + k1 + k2 )2 m2
2

(260)

R
- At high p, we have d4 p/p4 = log : Logarithmic divergece
- Momentum cutoff not well suited for this calculation because it breaks shift invariance p p + const
- Well use a better approach later
- To find the divergence, let us calculate this for the special case k1 + k2 = 0

40

Z
Z
p3 dp
d4 p
1
d4 p
1
i
=
i
=
4
2
2
2
4
2
2
2
2
2
(2) (p m )
8 0 (p + m2 )2
Eucl (2) (p + m )




i

1
I1 (m) =
log

+ O m2 /2
(261)
2
2
m
8
m 2

Z
I2 (0, m)

=
=

- Since the divergence comes from high p, it is independent of k1 and k2 , and we have



i
log
I2 (k, m) =
+
(finite
piece)
8 2
m
- The 1PI four-point function is therefore



3
2 i
log
log
= i 1
+ finite
4 = i + 3
2 8 2
m
16 2
m

5.2

(262)

(263)

Renormalised Couplings

- Our theory seems to give divergent results, but is this real?

- Look at physical, observable quantities such as scattering cross section
d
1
=
|M|2
2
d
64 2 Ecm

(264)

where the scattering amplitude is

iM = (sum of amputated 4-point diagrams) = 4

(265)

- This was indeed divergent,





3
+ (finite piece) + O(3 )
log
iM = i 1
16 2
m
- Assume there is a finite physical cutoff
- Theory not applicable above
- Makes everything finite
- The finite piece in Eq. (266) is a function of k1 , k2 , k3 , k4 and m
- Lorentz invariant, only dependent on Lorentz-invariant combinations (Mandelstam variables)
s = (k1 + k2 )2 , t = (k1 + k3 )2 , u = (k1 + k4 )2
2
- Dimensionless: Can only depend
on dimensionless
ratios s/m2 , t/m
and u/m2




3
s
t
u

iM = i 1
+f
, 2, 2
log
+ O(3 ),
2
2
16
m
m m m

(266)

(267)

where f is a function that we can calculate by evaluating the full integral I2 (k, m)
- M is now finite, but dependent on
- If an experimentalist want to test the theory, (s)he needs to know the coupling
- This can be determined by measuring M0 M(s0 , t0 , u0 ) for a given set of Mandelstam variables
- According to our theory, this is equal to


3

log
+ f0
+ O(3 )
(268)
iM0 = i 1
16 2
m

41

where we have defined


f0 = f

s0 t0 u0
,
,
m2 m2 m2


(269)

- We can find by inverting this,




3M0

= M0 1
log
+ f0
+ O(M30 )
16 2
m

(270)

- Once we have this, we can write Eq.

terms
 of M0 as 

 (267) in 
s
t
u
3M0
f
,
,

f
+ O(M30 ),
iM(s, t, u) = iM0 1 +
0
16 2
m2 m2 m2

(271)

- This expression is independent of , and M0 is finite: The divergence has disappeared!

- Since M0 rather than measures the real strength of the interaction, it makes sense to define
the physical or renormalised coupling
 R M
 0


t
u
3R
s
iM(s, t, u) = iR 1
f
,
,
f0
+ O(3R ),
16 2
m2 m2 m2
- The original coupling is called the bare coupling and denoted by B
- We can rewrite Eq. (268) as




s0 t0 u0

3B
+
f
,
,
log
+ O(3B )
R = B 1
16 2
m
m2 m2 m2
- Inverting this to order 2R , we have




s0 t0 u0
3R

B = R 1 +
+
f
,
,
log
+ O(3R )
16 2
m
m2 m2 m2

(272)

(273)

(274)

- Note that for fixed physical observables

B - depends on and diverges in the limit
- is the actual coeffient of 4 in the Lagrangian
- is given by the fundamental theory, if is a real cutoff
R - is finite and independent of
- depends on which values (s0 , t0 , u0 ) were used define it
= scale and scheme dependence
- Measures (roughly) the strength of interaction at low energies
- Consider then the two-point function
- The 1PI two-point function diverges [see Eq. (258)]


i
2 + m2
2
2
m log
+ O(2 )
2 (k) =
32 2
m2

(275)

- According to Eq. (217), the full two-point function is

G2 (k, q)

h(k)(q)i = (2)4 (k + q)

(2)4 (k + q)

k2

m2

k2

32 2

m2

i
i2 (k)

i
2 m2 log

2 +m2
m2

- The physical mass of a particle corresponds to the pole of its propagator,

42

 + O(2 )

(276)

so an experimentalist would measure



2 + m2

2
2

m
log
+ O(2 ) m2R
m2meas = m2 +
32 2
m2
- We call this the renormalised mass m2R and the original mass the bare mass m2B = m2 ,



2 + m2R
2
2
m2B = m2R

m
log
+ O(2 )
R
32 2
m2R

(277)

(278)

- Again, the bare mass is divergent but the two-point function G2 (k, q) is finite

5.3

Field Renormalisation

- In scalar theory, the leading-order expression for 2 (k) in Eq. (275) is independent of k
- Not true generally: O(2 ) term is k-dependent
O(e2 ) term in QED
- Lorentz invariance: Can only depend on k 2 , i.e., 2 = 2 (k 2 )
- The k 2 term is generally logarithmically divergent
- The general two-point function is
i
G2 (k, q) = h(k)(q)i = (2)4 (k + q) 2
k m2 i2 (k 2 )

(279)

- Again, define m2R as the location of the pole

m2R = m2 + i2 (k 2 = m2R )

(280)

- Expand 2 (k 2 ) around k 2 = m2R in powers of (k 2 m2R )

X
i2 (k 2 ) =
bi (k 2 m2R )i

(281)

i=0

where dimensional analysis tells us that generally b0 2 and b1 log

- Combining Eqs. (281) and (280), we find
b0 = i2 (k 2 = m2R ) = m2R m2

(282)

- In general bi for i > 0 are non-zero

- The full two-point function then has the expansion
k2

m2

1
i2 (k 2 )

=
=
=

1
P

+ b0 ) + (1 b1 )(k 2 m2R ) i=2 bi (k 2 m2R )i

1
P
(1 b1 )(k 2 m2R ) i=2 bi (k 2 m2R )i

X
Z
+
ai (k 2 m2R )i ,
(283)
k 2 m2R i=0
(m2

m2R

where ai are calculable coefficients, and Z = 1/(1 b1 )

- Thus, we have


iZ
G2 (k, q) = (2)4 (k + q) 2
+
(regular
terms)
k m2R
- In general the residue of the pole Z is not one
43

(284)

 Cannot identify the pole with a canonically normalised particle

- Rescale, or renormalise, the field
R = Z 1/2

(285)

- This implies
1
GR
G2 (k, q) = (2)4 (k + q)
2 (k, q) hR (k)R (q)i = Z


i
+
(regular
terms)
k 2 m2R

which has the correct normalisation

- Similarly, any n-point function Gn hn i is generally divergent,
n
n/2
but GR
Gn is finite
n hR i = Z
- In terms of the renormalised field, the Lagrangian is
1
1
1
L = Z R R Zm2B 2R Z 2 B 4R
2
2
4!

(286)

(287)

- To higher order in :
- We must solve the equations for R , m2R and Z simultaneously:
Can be messy because the results can be complicated functions of and m2
- The power series makes little sense because is large (divergent)
- There could also be more divergences to take care of, so we need to identify them systematically first

5.4

Power Counting

- Consider a 1PI diagram with E external legs, L loops, V vertices and I internal lines
in d spacetime dimensions

- Each internal line corresponds to a propagator D(k)

(use for quantities in momentum space)
- Each loop corresponds to integration over dd k
- Each vertex gives one power of coupling

I
ddL k D(k)

- In the UV (i.e. k ), D(k)

will go like some power of k, so the integrand will behave like
Z
V

dk k D1 V D ,
0

where we have defined the superficial degree of divergence D to parameterise the behaviour
- If D 0, the diagram is superficially divergent
(though not always actually divergent)
- We can determine D with dimensional analysis
- Denote the dimensionality of X by [X]: If X has dimensions of massn then [X] = n

44

(288)

(289)

 E (k), Eq. (289) implies

- First, because the diagram contributes
toithe momentum-space 1PI correlator
h
 V D

E (k) = = V [] + D,
(290)
from which we obtain
h
i
E (k) V []
D=

(291)

E (k)
- Now, we need to find the dimensionality of
E has trivial factors, which are absent from the 1PI correlator
E :
- The full momentum space correlator G
d
- Momentum conservation delta function (k1 + + kE )
i (k)
- External propagators D
E h
E h
h
i h
i 
i h
i
i
X
 X
E (k) =
E (k) + d (k1 + + kE ) +
i =
E (k) d +
i
G
D
D
(292)
i=1

i=1

- Therefore we find
h

E h
i h
i
i
X
E (k) = G
E (k) + d
i

(293)

i=1

- Take the Fourier transform to go to coordinate space


h
i Z
E (k) =
G
dd x1 dd xE eiki xi GE (x) = dE[x] + [GE (x)] = [GE (x)] dE

(294)

- This gives
h

E h
i
i
X
E (k) = [GE (x)] dE + d
i

(295)

i=1

- The dimensionality of the coordinate space correlator is

[GE (x)] = [h1 (x1 ) E (xE )i] =

E
X

[i (x)],

(296)

i=1

where we each i could be a different field

i ], we note that it is just the free two-point function,
- To calculate [D
2,i = hi (k)i (q)i = (2)d d (k + q)D
i (k)
G

(297)

- Using Eq. (294), this becomes

h i
h
i
i = d + G
2,i = d + [G2,i ] 2d = 2 [i ] d
D

(298)

- Substituting this into Eq. (295), we find

h
i
E (k) = d [GE (x)] ,

(299)

and finally, comparing with Eq. (290),

D = d [GE (x)] V []
- Furthermore, if one expands the correlator in powers of external momenta k,
each power reduces the superficial degree of divergence of the corresponding term by one:

45

(300)

 E (k)

k n Dn

(301)

- Three possibilities:
[] > 0: superrenormalisable
Degree of divergence decreases with increasing V
Only finite number of divergent diagrams (assuming [] > 0)
[] = 0: renormalisable
Degree of divergence independent of V
Infinite number of divergent diagrams
Finite number of divergent correlators (assuming [] > 0)
Examples: QED, QCD, EW theory in four dimensions
[] < 0: non-renormalisable
Degree of divergence grows with V
Infinite number of divergent diagrams
Infinite number of divergent correlators (in fact all of them)
Examples: gravity [GN ] = 2, Fermi theory of weak interactions [GF ] = 2

- Caveat: The actual degree of divergence may not be the same as D

- The leading divergence may cancel, especially if the theory has symmetries (e.g. SUSY)
- The actual divergence may be higher if there is a divergent subdiagram:

= 2

(302)

but the diagram has a divergent subdiagram (tadpole loop)

- The divergence comes from the two-point function, and disappears as it is renormalised
- Therefore it is enough to make sure all superficially divergent correlators are finite
- Example: Scalar theory
R
- The action is dimensionless 0 = [S] = [ dd xL] = d[x] + [L], so [L] = d
- The Lagrangian contains the derivative term [ ] = 2 + 2[] = [L] = d, so
[(x)] = d/2 1

(303)

- The Lagrangian also contains the interaction term [(x)4 ] = [] + 4[(x)] = [L] = d,
which implies
[] = d 4[(x)] = 4 d
- The theory is non-renormalisable in d > 4 dimensions
- In particular, the scalar coupling is dimensionless in d = 4, so the theory is renormalisable
- The divergent correlators are E 4, which means:
0 4 vacuum energy (cosmological constant)
2 2 + k 2 log two-point function

46

(304)

4 log four-point function

- In general, determine every parameter in the Lagrangian (including kinetic terms) from measurements
- If the number of divergent correlators is less than or equal to the number of coefficients,
the theory can be renormalised
- Symmetries: Fewer divergences, fewer parameters
- Renormalisable theories:
- Eq. (300) Superficial degree of divergence of expectation value hOi:
D (hOi) = d [O]

(305)

- Compare with the corresponding term in the Lagrangian:

L = + gO

(306)

- [L] = d [O] = d [g] D (hOi) = [g]

The superficial degree of divergence is the same as the dimensionality of the coupling
for the corresponding term in L
- Divergences are in one-to-one correspondence with renormalisable terms
- Just enough parameters to account for all divergences
- Non-renormalisable theories:
- Infinite number of divergences infinite number of parameters and measurements required
No predictive power!
- We conclude that we should write down the most general renormalisable Lagrangian subject to symmetries
- Number of couplings = Number of divergences
- Determine each coupling from experiment Absorbs all divergences
- When the theory has symmetries, there are fewer divergences and fewer couplings

6
6.1

Renormalised Perturbation Theory

Counterterms

- In principle we could follow Section 5 and do our calculations in terms of B and m2B
- Two problems:
- To renormalise, we would calculate R and m2R , and invert the expressions
but the expressions can be complicated if one works at high order in perturbation theory
- Our expansion parameter would be B which is divergent
- Therefore we want to organise the perturbative expansion in powers of R from the beginning
- The general philosophy in perturbative calculations:
- Take a simple model (L0 ) that you can solve and is a good approximation to the full theory
- Do a Taylor expansion in powers of LI = L L0
- This should be a good approximation if the subleading terms in the expansion are small
- In bare perturbation theory, the two-point function in the free theory is
47

G2 (k, q) = (2)4 (k + q)

i
k 2 m2B

(307)

where m2B is divergent, while the interacting theory gives Eq. (284), where m2R is finite but Z is divergent
L0 not a very good approximation!
- Clearly, it would be better to perturb around
1
1
L0 = R R m2R 2R
(308)
2
2
because then the free two-point function is
4
G2 (k, q) = ZGR
2 (k, q) = (2) (k + q)

k2

iZ
m2R

(309)

- This is a good approximation to Eq. (284), and the expansion parameter is R

- To do this, we write
Zm2B

= m2R + m2

Z 2 B

= R +

1 + Z

(310)

where m2 , and Z are known as counterterms

- Our Lagrangian then becomes
L

1
1
1
R R m2R 2R R 4R
2
2
4!
1
1
1
+ Z R R m2 2R 4R
2
2
4!

FGH

(311)

- Note that we havent changed the Lagrangian, we have only written it in a different way
- We can now use the first two terms as our free theory L0 and the rest as interaction LI
- The last three terms are known as counterterms
- The interaction vertices are

iR

i(k 2 Z m2 )

(312)

- The counterterms are treated as next order in R :

m2

O(R )

O(2R )

O(R )

- We also need to specify the renormalisation scheme, i.e., how we choose m2R , R and Z

48

(313)

- Our previous renormalisation scheme consisted of

(i) The scattering amplitude at (s0 , t0 , u0 ) is equal to R
4 (s0 , t0 , u0 ) = iR

(314)

(ii) The two-point function has a pole at k 2 = m2R

2 (m2R ) = 0

(315)

2 
2 (k )
=0
2
k
k2 =m2

(316)

(iii) The residue of the pole is one

IJ
R

- Use these conditions order by order to determine the values of the counterterms m2 , and Z
- As we will see later, there are other alternatives that can be more convenient
- Let us see how this works
- The 1PI two-point function is given by

2 (k 2 )

iR
=
32 2

+ O(2R )





2 m2R
2
2
mR log
+ i k 2 Z m2
2
mR

- This has to satisfy the renormalisation conditions

(316)



= iZ = 0
2 (k 2 )
k 2
2
2

(317)

(318)

k =mR

and (315)
2 (m2R )

K LM


2 m2R

log
im2 = 0
m2R


R
2 m2R
2
2
m2 =

m
log
R
32 2
m2R

iR

32 2

m2R

(319)

- The four-point function is given by

4 (s0 , t0 , u0 )

+3

+ O(3R )




3R

iR 1
+ f (s0 , t0 , u0 ) i = iR
log
16 2
mR
32R

log
=
+ f (s0 , t0 , u0 )
16 2
mR

- We could now, in principle, move to the next order

- Calculate 2 (k 2 ) and 4 to next order in R
- Determine the values of the counterterms to next order from Eqs. (314), (315) and (316)
- This would cancel all divergences to that order in R
49

(320)

- BHPZ (Bogoliubov-Parasiuk-Hepp-Zimmermann) theorem:

All correlators finite to all orders in R

6.2

Dimensional Regularisation

- The cutoff regularisation is concrete and easy to understand but usually impractical
- Multi-loop integrals become very cumbersome
- Incompatible with gauge invariance
- An elegant (but abstract) alternative: Dimensional regularisation
- Consider a loop integral In (m) in d dimensions (cf Eq. (251)
Z
1
dd k
= (1)n iInE (m)
In (m) =
(2)d (k 2 m2 )n
where we used the deformed complex integration path to turn it into a Euclidean integral
Z
Z
Z
dd k
1
k d1 dk
E
d
In (m) =
=
(2)
d
d
d
2
2 n
(k 2 + m2 )n
E (2) (k + m )
0
- Using Eq. (255), we can write this as
Z
Z 2 d/21 2
2
k d1 dk
1
(k )
dk
InE (m) =
=
2
2
n
2
2
d/2
d/2
(k + m )n
(4) (d/2) 0 (k + m )
(4) (d/2) 0

(321)

(322)

(323)

- Now, let us change the integration variable to

x=

k2

m2
,
+ m2

(324)

so that the integral becomes

InE (m) =

md2n
(4)d/2 (d/2)

xnd/21 (1 x)d/21 dx

(325)

- Now, recall the definition of the beta function

Z 1
()()
x1 (1 x)1 dx = B(, ) =
( + )
0

(326)

- This implies that

InE (m) =

md2n
(n d/2)(d/2)
md2n (n d/2)
=
(n)
(n)
(4)d/2 (d/2)
(4)d/2

- Consequently, the Minkowskian integral is equal to

md2n (n d/2)
In (m) = (1)n
(n)
(4)d/2
- The Gamma function (z) has poles at z = 0, 1, 2, . . .
In 4D, In (m) diverges for n = 1, 2, as we have seen before
- But for non-integer d it is always finite!
- We can use Eq. (328) to define the integral for d R
- In practice, we consider d = 4  with   1
- For integer l 0 one has

50

(327)

(328)



(1)l 2
(l + /2) =
+ 1 (l + 1) + O() ,
l!


(329)

where
1 (l + 1) =

l
X
1
k=1

=1+

1
1
+ . . . + ,
2
l

(330)

and 0.577216 is the Euler-Mascheroni constant

- For n 3, the result is finite (and therefore agrees with momentum cutoff)
- For n = 2, we have
I2 (m)

=
=

im
i(1  log m)


(/2) =
2/2
(4)
(4)2 1 2 log 4


2
i
4

+ O()
+
log
16 2 
m2


+ O()
(331)

- Comparison this with Eq. (261) shows that the two results agree if we identify
1
log + finite terms


(332)

- For n = 1, we find
I1 (m)

=
=

im2
im2 1  log m

(1)
(1
+
/2)
=

(4)2 1 2 log 4
(4)2/2


im2 2
4
+ log 2 + 1 + O()
2
16

m

2
+1



+ O()
(333)

- Comparing this with Eq. (257),

I1 (m) =

i
16 2



 2
2
m
2 m2 ln 2 + O
,
m
2

(334)

we see that the logarithmic terms agree again, but there is no analogue of the quadratic term 2
- Generally true: Only log divergences are present
- Simplifies calculations
- Can be sometimes misleading!
- In our results (333) and (331), we seem to have logs of dimensionful quantities
- Solution: In d = 4 , the coupling has dimensions [] = 
- Let us therefore write the 4  dimensional coupling as
 =  ,
where [] = 0 and [] = 1
- Consider a diagram with L loops and V vertices
- The integral is (schematically) proportional to
Z
V V

dd k
(2)d

(335)

L

- Using the results in Problem Sheet 4, we find that the number of loops L in a diagram is

51

(336)

L = V E/2 + 1

(337)

- Using this, our integral becomes

V

(L+E/21)

Z

dd k
(2)d

L

 Z

(E/21) V

dd k
(2)d

L

- Each loop comes therefore with a factor 

- Also an overall factor (E/21) , if the whole correlator is finite, this goes to 1 as  0
- Thus, when going to d = 4  dimensions, we should replace the integration measure by
Z
Z
d4 k
dd k


(2)4
(2)d

(338)

(339)

and we have


dd k
1
= (1)n
(2)d (k 2 m2 )n

In (m) =

42
m2

/2

im42n (n 2 + /2)
(4)2
(n)

(340)

- Then, Eq. (331) becomes



2

/2

i 1 + 2 log 4
m2
42
i
2
(/2)
=

+ O()
m2
(4)2
(4)2



2
42
i
+
log

+ O()
16 2 
m2


I2 (m)

=
=

(341)

and Eq. (331) becomes

/2




42
m2
im2

42
2
(1
+
/2)
=

1
+
log
(1)
+
1

+ O()
im2
(4)2
(4)2
2
m2



im2 2
42
+ log
+ 1 + O()
(342)
16 2 
m2


I1 (m)

=
=

- Finally, consider a massless (m = 0) integral InE (0)

- If d > 2n, Eq. (328) implies that InE (0) = 0
- If d < 2n, it diverges
- However, we define it by analytically continuing from high enough d
InE (0) = 0 for any n and d
- More generally, any scale-free integral is set to vanish
- This procedure is known as dimensional regularisation
- Go to d dimensions by replacing the integrals according to Eq. (339)
- Calculate the integral using Eq. (340)
- If the integral is scale-free, it vanishes
- This way power-law divergences disappear, and log divergences appear as
2
42
2
log 2 + log
m

m2

(343)

- To use dimensional regularisation in practice, we need to show that this can be applied to an arbitrary

52

Minkowskian integral of the form

Z
dd k
k 1 k m

2
d
2
(2) (k m1 )((k + p2 )2 m22 ) ((k + pn )2 m2n )

(344)

- First, use the identity

Z

1
2
2
(k m1 )((k + p)2 m22 )

(k 2

dx
((k + xp)2 + x(1 x)p2 x(m22 m21 ) m21 )2

dx
(k 02 )2

=
0

Z
=
0

+ x(2k p +

dx
m22 + m21 ) m21 )2

p2

(345)

where x is known as a Feynman parameter, k 0 = k + xp and

= x(1 x)p2 + x(m22 m21 ) + m21

(346)

- Furthermore, we can write

1

=
(k 02 )2

1
k 02

- Using these steps iteratively, we can write the general loop integral (344) in terms of
Z
dd k k 1 k m
In1 m () 
(2)d (k 2 )n

(347)

(348)

(Of course, the expression will involve complicated derivatives, and an n 1-dimensional integral
over Feynman parameters x1 , . . . , xn !)
- To deal with the numerator, note that I11 m () is a symmetric tensor of rank m
- Depends only on a scalar quantity , and therefore has to be Lorentz invariant
- The only symmetric tensor available is g , and therefore the tensor structure must
be a combination of g s only!
- Note, odd number of Lorentz indices vanish
- For example,
I1 () = Cg ,

(349)

with some Lorentz scalar C

- Contracting this with g , we find
g I1 ()

dd k
k2
= Cg g = dC,
(2)d k 2

(350)

and therefore
I1 () =

g 

dd k
k2

= g 
d
2
(2) k
d

dd k
1

= g I1 (1/2 ),
d
2
(2) k
d

where we used the result that any scale-free integral vanishes

- This way, we can write an arbitrary loop integral as a multi-dimensional integral over Feynman
parameters (which can still be extremely difficult to evaluate!)
- We must do it consistently in d = 4  dimensions and only take  0 at the very end

53

(351)

- As an example, let us calculate

I2 (p, m) = 

dd k
1
1
d
2
2
(2) k m (k + p)2 m2

(352)

- Using Feynman parameters (345) we can write this as

I2 (p, m)

Z 1
dx
dd k
d
2
(2) 0 ((k + xp) + x(1 x)p2 m2 )2
Z 1
Z
1
dd k 0
,
=
dx 
(2)d (k 02 (x))2
0
= 

(353)

where
(x) = x(1 x)p2 + m2

(354)

- Going to Euclidean space, this becomes

Z 1
dd kE
1
=
i
dx I2E (1/2 )
2 + (x))2
(2)d (kE
0
0

Z 1 
2
i
42
dx
+ log

16 2 0

(x)


Z 1
i
2
42
+

dx log
16 2 
x(1 x)p2 + m2
0
Z

I2 (p, m)

=
=
=

dx 

NO
P QR

(355)

- The integral over x can be done analytically, but leads to a long expression which we omit here

6.3

Minimal Subtraction

- In dimensional regularisation, the one-loop 1PI correlators are

2 (k 2 )

iR m2R
32 2

42
2
+ log 2 + 1

mR

+ i k 2 Z m2

(356)




3R 2
42
= iR 1
+ log 2 + f (s, t, u) i
32 2 
mR

(357)

and

+3

- To renormalise the theory, we need to specify how we split the bare parameters to
renormalised parameters and counterterms [see Eq. (310)]
- This split is completely artificial and does not affect the final result
(i.e. relationships between physical observables)

54

- Earlier we used the physical scheme (314)(316), which has a clear physical interpretation
but is rather clumsy
- Simpler alternative: Minimal subtraction
- Choose counterterms in such a way that they cancel the divergences, but only contain the 1/ pole
- Even more convenient: M S (modified minimal subtraction)
- For each pole, include the combination


1 1
42

(358)
+
log

2
M2
in the counterterm (because these terms always appear together)
- The mass scale M (renormalisation scale) is arbitrary
- The value does not affect physical results
- Analogous to the energy scales s0 , t0 and u0 in Eq. (314)
- One often identifies the scales M = , but we keep them separate:
- regularisation
- M renormalisation
- Using M S and Eqs. (356) and (357), we find
Z

m2

0


R m2R 2
42
+ log

32 2

M2


32R 2
42
+
log

32 2 
M2

(359)

- Substituting these back in Eqs. (356) and (357), we find



iR m2R
M2
2 (k ) =
log 2 + 1 + O(2R )
32 2
mR



M2
3R
4 = iR 1
log 2 + f (s, t, u) + O(3R )
32 2
mR
2

- We could even take a shortcut and simply subtract the divergences from Eqs. (356) and (357)
and replace M
- Now m2R and R do not have any direct physical interpretation
- They are dependent on the renormalisation scale M
- Their scale dependence cancels the explicit M -dependence in Eq. (360)

7
7.1

Renormalisation Group
Callan-Symanzik Equation

- Renormalised parameters m2R , R depend on the renormalisation scale M

but physical observables dont
Explicit dependence cancelled by the implicit dependence of the renormalised couplings
- For fixed bare coupling, the bare n-point function GB
n (p1 , . . . , pn ) is independent of M
- On the other hand, the renormalised n-point function GR
n (p1 , . . . , pn ) is a finite function of the
55

(360)

renormalised couplings and the scale M

2
GR
n (p1 , . . . , pn ; mR (M ), R (M ); M )

(361)

- Field renormalisation relates the two

n/2
2
GB
(M )GR
n (p1 , . . . , pn ) = Z
n (p1 , . . . , pn ; mR (M ), R (M ); M )

- The M dependence on the RHS must cancel

- Let us differentiate GB
n with respect to log M




GB
0 =
n
log M
B




m2R 

Z 1/2 
n
n/2
+
= Z (M )
log M B m2R R
Z 1/2 log M B



R 


2
+
+
GR
n (mR , R ; M ),
log M B R R log M

(362)

(363)

where |B and |R show whether we keep the bare parameters m2B or B or the renormalised
parameters m2R , R fixed when taking the partial derivative
- We then define



Z 1/2 
1
Z 
=
2Z log M B
Z 1/2 log M B

R 
log M B

m2R 
1
m2R log M B
1

where is known as the anomalous dimension and as the beta function

- This leads to the Callan-Symanzik
equation



2
2
+
+ 2 m R
+ n GR
n (mR , R ; M ) = 0,
log M
R
m2R
- It shows how the renormalised correlation functions depend on the renormalisation scale M
when the fundamental (bare) theory is kept fixed
- This is useful, because we can optimise perturbative calculations by a good choice of M :
- We want our renormalised perturbation theory to converge as fast as possible
- In general, this happens when M is comparable with the characteristic energy scale of the
process we are looking at
- For instance, the scattering amplitude for scattering is given by 4 in Eq. (360):
- If we choose M appropriately, the quantum correction becomes small, and R is
a good approximation of the actual scattering amplitude
- This typically means that M is close to the most relevant energy scales for the process
The beta function tells how the interaction strength depends on the characteristic energy scale
- According to Eq. (310), we can write the renormalised parameters as

56

(364)

(365)

Z(M )

1 + Z(M )

R (M )

= Z(M )2 B (M )

m2R (M )

= Z(M )m2B m2 (M )

- This means that to leading order we have



1
1 Z 
Z 
=

2Z log M B
2 log M R



Z 


= 2ZB

R
log M B
log M B
log M R



m2B
1 m2 
m2 
Z 
2 =

m2R log M B m2R log M B

log M R
- In the M S scheme, we obtain the counterterms in terms of the renormalised couplings
- The counterterms are are proportional to
4 2 2
2
+ log
,

M2

(366)

(367)

(368)

with no other (explicit) M dependence

- The coefficients , and 2 are therefore M -independent
- In fact, we can write the counterterms as

4 2 2
2
+ log
Z =


M2




1
4 2 2
2
m2 =
2 m2R
+ log

2

M2



1
4 2 2
2
=
2R
+ log

2

M2


(369)

- Comparing with Eq. (359), we find

0
32R
16 2
R
16 2

(370)

- Positive R grows with energy

- Explicit solution
R (M ) =

16 2
3 log(/M )

where is an integration constant

- Note that is really what parameterises the strength of the interaction
- R blows up at M = : Landau pole

57

(371)

7.2

Wilsonian Renormalisation

- So far we have taken the experimentalists view, expressing everything in terms of renormalised
parameters and essentially ignoring the bare ones
- But bare parameters are what actually appear in the Lagrangian, and they can give us important insight
- Therefore, let us look at the dependence of the bare theory on the cutoff,
keeping physical observables fixed
- Consider a fundamental theory with action S[] and a physical cutoff (e.g. Planck scale)
- Observables are given by derivatives of theZ generating functional
DeiS+i

Z[J] =

dd xJ(x)(x)

(372)

- Cutoff: Integrate only over those (x) whose Fourier transform (k) vanishes for |k 2 | > 2
- Does not work well in Minkowski space:
Allows arbitrarily high energies when momentum is nearly lightlike
- Wick rotation t = i

1
1
2
2
dtd x (t ) (i ) V ()
2
2


Z
1
1
d d3 x ( )2 + (i )2 + V ()
2
2


Z
1

d4 x ( )2 + V () SE
2
Eucl

Z
iS

=
=
=

- Write = x4 : Four-dimensional Euclidean integral

- Exactly the same rotation k0 ikE we did in Eq. (252)
- The cutoff becomes meaningful and we have

Z
Z
R d
Y
Z[J] =
DeSE d xJ(x)(x) =
d(k) eSE
0<k<

(373)

(374)

|k|<

- Compare with classical statistical

Z mechanics

eH

Z=

all configs

H SE

- 3+1D QFT = 4D classical statistical mechanics

- In statmech, the cutoff is physical: Atomic spacing = 1/ in ferromagnets
- Usually relevant length scales comparable to
- At critical (phase transition) point
Critical phenomena

Universality
(Independence of microscopic details)

QFT
m
Renormalisability
(Independence of UV physics)

- Now, we want to move the cutoff from to 0 < , but still describe the same physics
- This is known as coarse-graining, or integrating out momenta 0 < k <
- We can do this by splitting the integration into two pieces: k < 0 and 0 < k <
- Write + where
58

(375)


nonzero for k < 0
0
for 0 < k < ,
(k) =

0
for k >

for k < 0
0

nonzero for 0 < k <

(k) =

0
for k >

(376)

- Let us assume that we are only interested in observables with momenta less than the new cutoff 0 :
J(k) = 0 for k > 0
- Then we have
Z
Z

S[+]
De

Z[J] =
0<k<0

dd xJ(x)(x)

(377)

0 <k<

and defining the effective action Seff

[] as
0

Seff
[] = log

D eS[+]

(378)

0 <k<

we can write this as

Z
Z[J] =

D eSeff []

dd xJ(x)(x)

(379)

0<k<0

- Any correlation function can be obtained from Z[J]

0
Seff
gives an effective theory, with a lower cutoff 0 , with exactly the same low-energy (k < 0 ) physics
- We say that we integrate out the high-momentum (k > 0 ) modes to obtain an effective field theory
- The effective theory can then be used just as if it was the fundamental one
- Now, let us consider the scalar theory with Euclidean action

Z
1 2 2
1 4
1
2
d
S = d x ( ) + m +
2
2
4!

(380)

- The effective action (378) is then

0
Seff
[]

Z
= S[] log

S[]
De

(381)

0 <k<

where

S[]

S[]
= S[ + ]



Z
1
2 + 1 m2 2 + 1 3 +
1 2 2 + 1 3 + 1 4
=
dd x ( )
2
2
6
4
6
24

ST

and the quadratic cross terms cancel because of momentum conservation

- Eq. (381) is essentially the generating functional for connected correlators E[J] (205)

in theory with dynamical field and action S[]

- The low-energy field plays roughly the role of the external field J(x)
- The path integral can be calculated using same perturbative techniques as before:
Sum of connected diagrams
- Euclidean No imaginary units
- Momenta restricted to 0 < k <
- Represent by double line
: Internal lines
- Propagator:
(k)

k 2 + m2

59

(382)

(383)

UV
WX
Y
Z[

where (k) = 1 if 0 < k < and = 0 otherwise

- Vertices:

3
6

2
2

(384)

where dashed lines

represent factors of
- Note that what we denote by n is, strictly speaking
Z 0 d
dd kn
d k1

(k1 ) (kn )(2)d (k1 + k2 + k3 + k4 )

n = (x)n
d
(2)
(2)d
0

(385)

- To order O(), the integral in Eq. (381) is

(assuming m2  02 so that we can approximate m2 0)
Z
log

S
De

+ O(2 )

dd k 1
+ O(2 )
d k2
(2)
0

!
Z
Z
1 2
d
d3
=
(2)
dd
dk k
+ const + O(2 )
4
0


1
d2 0d2
2
=

+
const
+ O(2 )
2 (d 2)(4)d/2 (d/2)
= const

(386)

where the constant arises from the figure-eight diagram

- Taking the logarithm, we find the effective Lagrangian

d2 0d2
1
2

+
const
+ O(2 )
2 (d 2)(4)d/2 (d/2)


1
1
1
(1 + Z)( )2 +
m2 + m2 2 + 4 + O(2 )
2
2
4!


Leff

= L+
=

where we have dropped the uninteresting constant and defined

d2 0d2

d=4
m2 =

2 02
2
d/2
32
(d 2)(4) (d/2)

60

(387)

(388)

\^ _]`
ac b
d
f
e
gh

- Thus, we have found an effective theory:

- Lower cutoff 0 <
- Equivalent to the original one to linear order in
- To this order, we can compensate for the change in the cutoff by changing the bare mass
- To order O(2 ), we have
Z
log

S
De

... + +

(389)

- The first row gives an uninteresting constant term

- The sixth and eighth terms vanish because of momentum conservation
- The logarithm in Eq. (381) removes the disconnected diagrams
- Thus, the effective action to order O(2 ) is

Seff []

S[]

(390)

- The first three diagrams are proportional to 2 and give contributions to the mass term
- Because the two-loop diagrams are subleading, we ignore them
- The fourth diagram gives a correction to the 4 term (ignoring the external momenta):

Z d  2
Z
3
d k
1
3 2
2 d/2
= 2
=

dk k d5
d
2
d (d/2)
2
(2)
k
2
(2)
0
0

(


32
(4)d/2 (d4)(d/2)
d4 0d4 , when d 6= 4
=
32

16
when d = 4
2 log 0 ,

- The fifth diagram gives rise to a new, six-point interaction vertex

61

(391)

p1
p2

2 6
2 6
(p
+
p
+
p
)

(p1 + p2 + p3 )
1
2
3
(p1 + p2 + p3 )2
2

(392)

p3
- This may look worrying: We wanted to understand why QFTs in particle physics are renormalisable,
but instead, we found that even if we start from a renormalisable fundamental theory,
it creates a non-renormalisable effective interaction
Does it make any difference whether the fundamental theory is renormalisable
because it is equivalent to a non-renormalisable theory anyway?
- Note, however, that the effective 6 coupling is suppressed by 2
- In summary, the effective Lagrangian with cutoff 0 in 4D is
0
L
eff






2

1
32
2 6
1
2
2
2
02
4
log

(393)
( ) +
m +

+
#
2
2
32 2
4!
16 2
0
2

- Because we ignored the external momenta, we are missing derivative terms such as 2 ( )2
- It is easy to see that at higher order in , all terms allowed by symmetries appear
- Because is the only dimensionful parameter in loops, the coefficient of n is 4n

7.3

Renormalisation Group Transformation

- In the previous section, we changed the cutoff from to 0

- In general, this gives an effective Lagrangian of the form
Leff =

1
1
1
(1 + Z)( )2 + (m2 + m2 )2 + ( + )4 + C( )4 + D6 + . . . , (394)
2
2
4!

where we have included the contribution Z, which is closely related to field renormalisation and
two examples of non-renormalisable terms C and D
- Now, zoom out by the same amount: Rescale all lengths by factor b = 0 /
x

x0 = xb

k 0 = k/b

(395)

- This takes the cutoff back to its original value

0

0
=
b

- Basically, we are expressing everything in units of the cutoff

- Coarse-graining + rescaling = renormalisation group transformation
- Maps the theory to a different theory in the same space (i.e. same cutoff)
- The action becomes

62

(396)

Z
Seff

dd xLeff

Z
=

d 0 d

d xb

1
1
1
(1 + Z)b2 (0 )2 + (m2 + m2 )2 + ( + )4
2
2
4!

+Cb4 (0 )4 + D6 + . . . ,

- As before, it is convenient to rescale the field back to the canonical normalisation

q
0 = b2d (1 + Z)

(397)

(398)

so that the kinetic term (and therefore the free propagator) has the canonical normalisation
Z
Seff

d 0

d x

1 0 0 2 1 m2 + m2 2 02
1 + 4d 04
b
( ) +
b +
2
2 1 + Z
4! (1 + Z)2

C
D
d 0 0 4
2d6 06
+
b ( ) +
b
+ ...
(1 + Z)2
(1 + Z)3

- The coefficient G of a general term with n powers of and m derivatives is

G + G (d/21)n+md
G + G [G]
G0 =
b
=
b
(1 + Z)n/2
(1 + Z)n/2

7.4

(399)

(400)

Renormalisation Group Flow

- Often useful to consider infinitesimal transformations

- Write b = 0 / = 1 , with   1
- The corrections m2 , , Z etc. are then O()
- The mass parameter changes
m2 + m2
m02 = m2 + dm2 =
(1 + 2) ,
1 + Z

dm = m

m2
Z + 2
m2


(401)

- To linear order in we had (in 4D)

m2

(1 b2 )

2 =
2 
2
32
16 2
(402)

- Thus, writing
 = d log

1
b

(403)

we have
dm2 =


2m2 +

1
2

d log ,
2
16
b

which we can write as a differential equation

63

(404)

dm2

2
= 2m2 +
d log(1/b)
16 2

(405)

- For , we have
0 =

32
1
+
=

d log ,
2
2
(1 + Z)
16
b

(406)

which implies
d
32
=
d log(1/b)
16 2

(407)

d
32
= () + O(3 )
=
d log(1/b)
16 2

(408)

- Compare with Eq. (370):

- The leading term is generally the same:




R 
R 
B 
=
=

log M B
B M log M R




R 
B 
B 
B 
=

B M log R
log R
log 1/b R

(409)

where we used the fact that between the cutoff and dimensional regularisation, the divergences
correspond to each other as

log log
(410)
M
and that to leading order R = B
- Higher orders depend on the renormalisation scheme
- We have a similar equation for every coupling
dG
G (n/2)Z
= [G]G +
d log(1/b)

- Formally, we can write an infinite-dimensional vector ~g = (m2 , , C, D, . . .),
containing all the couplings, and write
d~g
= f (~g )
d log(1/b)

(411)

(412)

- This tells how the effective theory changes under gradual coarse-graining
- Flow in the space of all possible theories
= Renormalisation group flow
- First-order equation Trajectories cannot cross
- There can be fixed points ~g with f (~g ) = 0
- They define scale-invariant (conformal) field theories
- Example: Gaussian fixed point m2 = = C = D = . . . = 0, i.e.,
1
L = ( )2
2

64

(413)

- Let us look at the flow near the fixed point:

-  1, m2 /2  1 just as in our perturbative calculation
- Leading order: Ignore corrections m2 , etc
Scaling
m02

= m2 b2 m2 /b2 in 4D

= bd4

G0

= Gb[G]

(414)

- Following the flow toward smaller b, we find three different possibilities

- [G] > 0: Coupling grows (unstable direction)
= Relevant operator
- Corresponds to a superrenormalisable term
- Only m2 in our theory
- [G] = 0: Coupling does not change
= Marginal operator
- Corresponds to a renormalisable term
- Only in our theory
- [G] < 0: Coupling decreases (stable direction)
= Irrelevant operator
- Corresponds to a non-renormalisable term
- All other terms
- Thus, if we start with some fundamental theory with all couplings of order 1
- Relevant operators grow
- These must be fine tuned to remain small even at low energies in order to stay near the fixed point
- This is known as the hierarchy problem
- Irrelevant operators decrease
- Theory becomes automatically renormalisable
- This means that only relevant operators have to be fine-tuned!
- Sometimes no renormalisable terms are compatible with symmetries
- Then the least irrelevant term determines the strength of the interaction
- The strength of the interaction is suppressed by a power of b
- Examples: Gravity, Fermi theory of weak interactions
- There may also be other fixed points, which would also define valid low-energy theories
- However, these will be non-perturbative
- For example, second-order phase transitions define non-trivial fixed points
- Let us then include the leading quantum corrections
dm2
d log(1/b)
d
d log(1/b)

2m2 +

- Flow is somewhat more complicated

65

32
16 2

2
16 2
(415)

0.1

0.05

0
-1

-0.05

-0.1

- Fixed point still at m2 = = 0

- Unstable direction along = 0
- Stable direction along m2 = m2c () with
m2c () =

2
32 2

- The mass parameter of the fundamental theory has to be fine tuned to m2 m2c ()
- This is precisely what Eq. (278) does
- Analogous to critical phenomena T Tc
- Coupling constant runs: Interactions become stronger at high energies
16 2
() =
3 log(0 /)

(416)

(417)

- Diverges at a finite cutoff value = 0 (Landau pole)

The continuum limit cannot be taken if > 0
- Strictly speaking the perturbative expansion in becomes unreliable when 1
- It is possible that there is an ultraviolet fixed point at ( , m2 = m2c ( ))
- Non-perturbative: Could not be studied using perturbation theory
- This would make it possible to define a continuum theory:
RG flow trajectory that starts infinitesimally close to ( , m2 )
- No evidence for such a fixed point exists
- If no fixed point, the scalar theory is trivial:
Finite coupling in fundamental theory 0 at low energies
= Free theory
- The same conclusions apply to QED:
(e2 ) =

Nf 4
e
6 2

de2
Nf
= 2 e4
d log(1/b)
6

- Coupling e (=electron charge!) decreases at low energies (=long distances)

- Can be seen as a screening effect:

66

(418)

- Virtual electron-positron pairs get polarised by the electric field of an electron:

The positron is attracted and electron is repelled
- This screens some of the electric field
- If we calculate the electron change from the field strenght at long distances, we obtain a smaller value
- Again, we find a Landau pole
6 2
e2 () =
(419)
Nf log(0 /)
QED does not exist as a (perturbative) continuum theory
- Little experimental significance:
If = me (the electron mass), e2 (me ) = 0.09 and Nf = 1, then
0 10300 GeV

(420)

- QED can only exist as a part of some more fundamental theory (GUT?)
- Most 4D theories have a Landau pole and are trivial
- Main exception: Non-Abelian gauge field theories

Renormalisation of QCD

8.1

Counterterms

- The gauge-fixed Lagrangian (153) has six parameters:

- Three field normalisations: A , c and
- Parameters: g, and m
- Do we need to include more terms to renormalise the theory?
- Generally all terms compatible with symmetries
- Gauge invariance is broken by gauge fixing, so it cannot protect the Lagrangian
- The gauge fixed theory has a more complicated BRST (Becchi-Rouet-Stora-Tyutin) symmetry
- Introduce auxiliary field B a with Lagrangian
1
LB = B a B a + B a Aa
2
- Integrating over B a gives the gauge fixing term
67

(421)

DBei

dd xLB

= const ei

2
1
dd x( 2
( Aa
) )

- We can therefore replace in the Lagrangian (153)

1
1
( Aa )2 B a B a + B a Aa
2
2

(422)

(423)

- The resulting Lagrangian has a peculiar BRST symmetry

Aa

Aa + Dac cc

ca

i igtaij ca j
1
ca + gf abc cb cc
2
ca + B a

Ba

(424)

- This symmetry does not allow any new terms, and therefore it is enough to consider
Lagrangians of the form (153)
- All four interactions have coupling g (or g 2 )
- Related by gauge invariance, even in quantum theory
- We treat them as independent at first
- To set up renormalised perturbation theory, we first rescale the fields by renormalisation constants:
1/2

Aa ZA Aa ,

ca Zc1/2 ca ,

1/2

i Z i

(425)

- The Lagrangian (153) becomes

L

1
1 ZA a 2
ZA ( Aa Aa )( Aa Aa )
( A )
4
2 B
+Zc ca ca + Z i (i/ mB ) i
1 2 2 abc ade b c d e
1
3/2
Z A f f A A A A
+ gB ZA f abc ( Aa Aa )Ab Ac gB
2
4
1/2
1/2 a a
gB Zc ZA ca f abc Ab cc gB Z ZA
tij A i j ,

(426)

where we have also added the subscripts B to the parameters to indicate they are bare ones
- We define renormalised parameters by introducing multiplicative renormalisation constants
ZA
B
Z mB
3/2

ZA gB

= Zm m
=

= Z3 g

2 2
ZA
gB
1/2
Zc ZA gB

= Z4 g 2

1/2
Z ZA gB

= Z2 g

= Z1 g

- Note that the renormalisation constants Z1 , Z2 , Z3 and Z4 are related to each other, e.g.

68

(427)

Z2 =

Z
Z1
Zc

(428)

- In terms of the renormalised fields and parameters, the Lagrangian is therefore

L

1
1 Z a 2
= ZA ( Aa Aa )( Aa Aa )
( A )
4
2


+Zc ca ca + i iZ / Zm m i
1
1
+ Z3 gf abc ( Aa Aa )Ab Ac Z4 g 2 f abc f ade Ab Ac Ad Ae
2
4
a a
Z1 g ca f abc Ab cc Z2 g
tij A i j ,

(429)

- Next we write the renormalisation constants ZX in terms of counterterms, ZX = 1 + ZX

- This gives
L

1
1
= ( Aa Aa )( Aa Aa ) ( Aa )2
4
2


+ ca ca + i i/ m i
1
1
+ gf abc ( Aa Aa )Ab Ac g 2 f abc f ade Ab Ac Ad Ae
2
4
g ca f abc Ab cc g ta Aa i j

ij

1 Z a 2
1
( A )
ZA ( Aa Aa )( Aa Aa )
4
2


+Zc ca ca + i iZ / Zm m i
1
1
+ Z3 gf abc ( Aa Aa )Ab Ac Z4 g 2 f abc f ade Ab Ac Ad Ae
2
4
a a
Z1 g ca f abc Ab cc Z2 g
tij A i j

(430)

- The propagators were given in Eqs. (237)(239)

- The interaction vertices are Eqs. (243), (245), (247) and (249), together with the counterterm vertices

69

jkl
m
n
o
p




Z
2

ZA k g k k
k k

Aa (k)

Ab (k)

j (k)

i (k)

iij (Z k/ Zm m)

cb (k)

ca (k)

ab

i ab Zc k 2

Aa (k1 )

Ab (k2 )
Aa

Ac

gZ3 f abc [g (k2 k1 ) + g (k3 k2 ) + g (k1 k3 )]

Ac (k2 )
Ab

Ad

Ab (k2 )

ca (k1 )

Aa




ig 2 Z4 f abe f cde g g g g

+f ace f bde g g g g

+f ade f bce g g g g



Z1 gf abc k1

cb (k3 )

a
iZ2 g
tij

- To determine the counterterms, we have to find out which correlator gets a tree-level contribution
from each counterterm

70

(431)

ZA , Z

Zc

Z , Zm

Z1

Z2

Z3

Z4

qrs
tu
vw

- We need to compute all the two-point functions, but because Z1 Z4 are related, it is enough
to compute one of the last four correlators
- We choose the quark-gluon interaction Z2 , because it consists of fewer diagrams than
the gluon-gluon interactions, and it corresponds to a real physical process unlike
the ghost-gluon interaction
We need four correlation functions to determine the six counterterms

8.2

Correlation Functions

8.2.1

Gluon

- We start by looking at the gluon two-point function

71

(432)

x y| z} {
~
=

(433)

- The quark loop is

p+k
i

Aa (k)

Ab (k) =





1
a
b
(1) 2 ig
tij ig
tji
2!


 

i
i
dd p
(2)d p/ + k/ m p/ m

 

Z
i
i
dd p

2
a b 
Tr

(,434)
= g tr t t
(2)d
p/ + k/ m
p/ m


where tr is a trace over colour indices ij and Tr over spinor indices

- To proceed, we have to recall some identities involving gamma matrices

Tr

1
Tr { , } = g Tr 1 = 4g
2
= 0


= 4 g g + g g g g

Tr

Tr

2g

p/ + m
1
= 2
p/ m
p m2

{ , }

(435)

- Thus, we have

=
=

dd p Tr (p/ + k/ + m) (p/ m)
(2)d ((k + p)2 m2 )(p2 m2 )
Z 1
Z
dd p Tr (p/ + k/ + m) (p/ m)
2
a b

,
g tr t t
dx
(2)d
((p + xk)2 (x))2
0
2

a b 

g tr t t

(436)

where on the second line we have used the Feynman parameter trick (345) and defined
(x) = x(1 x)k 2 + m2
- We shift the integration variable p + xk p,

72

(437)

a b

= g tr t t

dd p Tr (p/ + (1 x)k/ + m) (p/ xk/ m)

(2)d
(p2 (x))2

dx
0

a b

= g tr t t

dd p (p + (1x)k) (p xk) Tr + m2 Tr
(2)d
(p2 (x))2

dd p (p p x(1x)k k )Tr + m2 Tr
,
(2)d
(p2 (x))2

dx
0

= g 2 tr ta tb

dx

(438)

where in the last step we used the fact that terms with one power of p in the numerator vanish
by Lorentz invariance
- As in Eq. (351), we have
Z
Z
p p
g 
p2
g
dd p
dd p

=
=

(I1 () + I2 ()) ,
d
2
2
d
2
(2) (p )
d
(2) (p )2
d

(439)

where


In () =

dd p
1
(2)d (p2 )n

(440)

- The values of the integrals are given by Eq. (328)

I1 ()

I2 ()

d/21
(1 d/2)
(4)d/2

d/22
(2 d/2)
(4)d/2

(441)

- The Gamma function satisfies z(z) = (z + 1), which implies

(2 d)I1 () = 2i

d/21
(4)d/2



d/21
d
(1 d/2) = 2i
(2 d/2) = 2I2 ()
1
2
(4)d/2

(442)

and further

dd p
p p
g
=
(2)d (p2 )2
d

2
I2 () + I2 ()
d2


=

g
I2 ()
d2

(443)

- Using Eqs. (435), we find

1


g
= 4g tr t t
I2 () x(1 x)k k I2 ()
dx
d2
0


g g + g g g g

+m2 g I2 ()
Z 1 
2 d
= 4g 2 tr ta tb
dx
g I2 ()
d2
0



x(1 x) 2k k k 2 g I2 () + m2 g I2 ()
Z 1




= 4g 2 tr ta tb dx + x(1 x)k 2 + m2 g 2x(1x)k k I2 ()
2

a b



= 4g 2 tr ta tb

dx 2x(1x) k 2 g k k I2 (),

73

(444)

where we used = m2 x(1 x)k 2

- Near four dimensions, when d = 4 , we have 

2
i
I2 ()
+ finite ,
16 2 

(445)

and therefore

Z 1
2
dx 2x(1 x)
+ finite

0



2

k k
+ finite


4ig 2
tr ta tb k 2 g k k
=
16 2
=

ig 2
4
tr ta tb k 2 g
16 2
3

(446)

- We could calculate the finite part, but for M S renormalisation it is not needed
- The group factor tr ta tb C(r) ab depends on the quark representation r
- For the fundamental representation (our case), Eq. (124) gives C(r) = 1/2
- For Nf fermion species, the total result is


ig 2 4
Nf C(r) ab k 2 g k k
=
16 2 3


2
+ finite


(447)

- The second diagram in Eq. (433) gives

=

ig 2 acd bcd
f f
16 2








25
17
2
2

k g
k k
+ finite
12 2
6








ig 2
25
3
ab
2


C2 (G)

k g k k + +
k k
16 2
12 2
4 2


2

+ finite ,
(448)


where we have written f acd f bcd = C2 (G) ab in terms of the quadratic Casimir invariant
of the adjoint representation (for SU(N ), C2 (G) = N )

- The third diagram is

because the integral is scale-free

- The ghost loop (4th diagram) gives

=
=

ig 2 acd bcd
f f
16 2

= 0,



1 2 1
2
k g + k k
+ finite
12
6





1
ig 2
1
2
ab
2

C2 (G)
k g k k + k k
+ finite ,
16 2
12
4


(449)

74

(450)

- Finally, the counterterm diagram gives

Z
= iZA k 2 g k k i
k k

(451)

- Putting all of these together, we find

=

ig 2 ab

16 2

iZA








4
13
Nf C(r) +

C2 (G) k 2 g k k
3
6
2


1
2

+
C2 (G)k k
+ finite
2



Z
k k
k 2 g k k i

(452)

- Following the M S scheme, we must therefore choose

ZA






4
42
g2
13
2
Nf C(r) +

+ log

C2 (G)
16 2
3
6
2

M2


g2 1
2
42
C
(G)
+
log

2
16 2 2

M2

(453)

- There are some interesting points to note:

- The counterterms depend on , but that is not a problem because they are not physical observables
- The gauge fixing parameter gets renormalised:
It is therefore not consistent to simply fix its value to something convenient

8.2.2

Quark

- The quark two-point function consists of only two diagrams

(454)

- The loop diagram is

p

i (k)

j (k)

i l

a
ig
tli

a
ig
tjl

l j

p+k

dd p
D (p)SF (p + k)
(2)d F
Z


dd p
g 2 (ta ta )ji 
DF (p) SF (p + k) (455)
d
(2)

75

- Note that the order of the matrices is opposite to the propagation of the quark
- We have here the quadratic Casimir of the quark representation
(ta ta )ij = C2 (r)ij

(456)

- For the fundamental representation in SU(N ), C2 (r) = (N 2 1)/2N

- Doing the algebra and evaluating the integral, one finally obtains

ig 2
ij C2 (r) ( k/ (3 + )m)
16 2


2
+ finite


(457)

- The counterterm diagram gives

= i (Z k/ Zm m) ij

(458)

- In the M S scheme, we therefore have

8.2.3

Ghost

Zm



g2
2
42

C2 (r)
+ log

16 2

M2


g2
2
42

C2 (r)(3 + )
+ log

16 2

M2

(459)

- The ghost two-point function consists of

where

ig 2 ab
3 5 2
=
C2 (G)
k
16 2
4

and

= i ab k 2 Zc

Therefore, we have

76


2
+ finite


(460)

(461)

(462) -

g2
3 5
Zc =
C2 (G)
16 2
4

2
42

+ log

M2

8.2.4

(463)

Quark-gluon coupling

- To compute Z2 , we need the quark-gluon interaction

Aa (q)

i (k)

(464)

j (k + q)

where

ig 3
=
16 2




1
2
a
C2 (r) C2 (G) tji
+ finite ,
2


ig 3
3(1 + ) a
C2 (G)
tji
16 2
4


2
+ finite ,


(465)

(466)

and

a
= iZ2 g
tji

(467)

- This means that the counterterm is

Z2

8.2.5




g2
3(1 + )

2
42

C
(G)

C
(r)
+
C
(G)
+
log

2
2
2
16 2
4
2

M2






g2
3
1
2
42

C
(G)
+
C
(G)
+
C
(r)

+
log

2
2
2
16 2 4
4

M2

Summary

- In summary, the values of the counterterms are

77

(468)

8.3

ZA

Zc

Zm

Z2





g2
13
4
2
42

C
(G)

N
C(r)
+
log
2
f
16 2
6
2
3

M2


g2
2
42

C2 (r)

+ log
2
16

M2


3 5 2
g2
42
C
(G)

+
log
2
16 2
4

M2


g2 1
2
42

C
(G)
+
log
2
16 2 2

M2


g2
2
42

C
(r)(3
+
)

+
log
2
16 2

M2



 
g2
3
1
2
42

C2 (G) +
C2 (G) + C2 (r)
+ log
16 2 4
4

M2

(469)

Beta Function

- We can now compute the QCD beta function

(g) = M


g 
M B

- In terms of the bare coupling and the counterterms, the renormalised coupling is


1/2
Z ZA
1
5
gB = 1 + Z + ZA Z2 gB + O(gB
)
g=
Z2
2

(470)

(471)

- Therefore, we have


(g)


Z
1 ZA
Z2
+

log M 2
2 log M 2
log M 2
 2
g
= 2g
C2 (r)
16 2



1 g2
13
4

C2 (G) Nf C(r)
2 16 2
6
2
3


 
2
g
3
1

C2 (G) +
C2 (G) + C2 (r)
16 2 4
4




2g 3
13
3
2
=
( ) C2 (r) + +
C2 (G) + Nf C(r)
16 2
12 4 4 4
3


3
g
11
4
=
C2 (G) + Nf C(r)
2
16
3
3
=

2g

- The gauge () dependence has now disappeared

- For SU(N ) and fundamental quarks we have C2 (G) = N and C(r) = 1/2, and therefore


g3
11N
2
(g) =

+ Nf
16 2
3
3
- If Nf < (11/2)N , the beta function is negative

78

(472)

(473)

 The interaction gets weaker at higher energies (=asymptotic freedom)

- This is precisely what deep inelastic scattering results showed already in 1970s
- Interaction becomes strong at low energies: This is why we cannot see individual quarks and gluons
- This calculation gave Gross, Politzer and Wilczek the 2004 Nobel prize

8.4

Renormalisation Group Flow

- Consider the flow in the Wilsonian approach (Section 7.4)

- Use g and m/ as coordinates
- How does the effective bare coupling g change as we lower the cutoff?
- It is given by the equation


11N
dg
g3
2

= (g) =
+ Nf
d log
16 2
3
3

(474)

- If Nf < (11/2)N , (g) < 0, and g is marginally unstable

- This is easy to solve


1
2
1
11N

=
Nf log
,
g2
8 2
3
3
QCD

(475)

where QCD is an integration constant

m/

0
0

0.5

- Because the beta function is negative (not positive as in QED or scalar theory), the flow is directed
away from the Gaussian fixed point (circle)
- Continuum limit corresponds to following the flow trajectory backwards
- We can go arbitrarily far The theory has a continuum limit (unlike QED or scalar theory)
- Following the flow towards low energies, g diverges at = QCD
- Analogue of the Landau pole but at low energies
- Physics does not really become singular:
It just becomes non-perturbative, so our assumptions fail
- Below QCD we need a different effective theory:
- The relevant degrees of freedom are protons and neutrons, not quarks and gluons
- In a sense, the flow does not actually hit g = but it leaves the two-dimensional plane
- The appropriate theory is known as chiral perturbation theory
- This explains why we dont see massless gluons or individual quarks
- The hadron masses given by QCD
79

- Account for most of the mass of everyday objects

- Nothing to do with the Higgs field!
- In the Callan-Symanzik approach, the same equation tells how the renormalised coupling depends on M ,


11N
1
2
M
1
Nf log
(476)
2 = 8 2
gR
3
3
QCD
- Note that gR depends only on how far above QCD we are:
- There is no free dimensionless parameter
- LEP:
g 2 (MZ )
s (MZ ) =
= 0.1176 0.002
4

QCD 250 MeV

- Compare with classical theory

- Classical Yang Mills:
- scale-invariant
- interaction strength is given by dimensionless parameter g
- Quantised Yang-Mills:
- scale-dependent
- interaction strength given by dimensionful scale QCD
- If we assume the quarks are massless, QCD is the only scale in the theory
There is nothing we could compare it with
- This means that Yang-Mills or QCD with massless quarks is a unique theory:
There is no free parameter that we could vary

80

(477)

Index
1PI correlator, 32
anomalous dimension, 56
asymptotic freedom, 79
beta function, 56
BHPZ theorem, 50
BRST symmetry, 67
Casimir invariant, 74
coarse-graining, 58
conformal field theory, 64
connected correlator, 32
continuum limit, 66
counterterm, 48
critical phenomena, 66
dimensionality, 44
effective action, 59
effective field theory, 59
Faddeev-Popov ghost, 23
Feynman diagram, 27
Feynman gauge, 24
Feynman parameter, 53
Feynman propagator, 12
Feynman rules, 30
fixed point, 64

marginal operator, 65
modified minimal subtraction (M S), 55
non-renormalisable, 46
one-particle irreducible, 32
relevant operator, 65
renormalisable, 46
renormalisation constant, 68
renormalisation group transformation, 62
renormalisation scale, 55
renormalisation scheme, 49
renormalised coupling, 42
renormalised mass, 43
running coupling, 66
scale-free integral, 52
scattering cross section, 41
structure constants, 19
superficial degree of divergence, 44
superrenormalisable, 46
symmetry factor, 29
trivial theory, 66
ultraviolet cutoff, 40
Wick rotation, 58

gauge condition, 20
gauge fixing parameter, 23
gauge invariance, 17
Gaussian fixed point, 64
Grassmann algebra, 14
Grassmann numbers, 14
Haar measure, 21
irrelevant operator, 65
Landau gauge, 24
Landau pole, 66
Lorenz gauge, 20
Mandelstam variables, 41

81