e.g.

= one particle reducible

= one particle irreducible

Formally these come from the generating functional

Z
Γ[φcl ] = W [J] − d4 zJ(z)φcl (z)

where
δW [J]
φcl (z) =
δJ(z)
(φcl is the average field in the presence of a source. We assume that φcl [J]
is invertible, ie J = J[φcl ] and that J = 0 when φcl = 0.) This is like a
‘functional Legendre transform’

cf Hamiltonian case,
∂L
H = pq̇ − L p=
∂ q̇
where we can invert to find q̇ = q̇(p, q) (q is passive here.) Then
∂H ∂ q̇ ∂L
= q̇ + p − = q̇
∂p ∂p ∂p
|{z}
∂L ∂ q̇
∂ q̇ ∂q
|{z}
p

ie Hamiltonian’s equations. So here given φcl (and Γ) what is the correspond-

ing J?
Z Z
δΓ[φcl ] 4 δW [J] δJ(z) δJ(z)
= dz − d4 z φcl (x) − J(z)
δφcl (x) δJ(z) δφcl (x) δφcl (z)
or
δΓ[φcl ]
= −J(x)
δφcl (x)

37
The N-point function is defined in the usual way as

δN Γ
Γ(N ) (x1 , . . . , xN ) = (52)
δφcl (x1 ) . . . δφcl (xN ) φcl =0

We justify this (as usual) by looking at some examples for three and four
point functions.

Two point 1PI Green’s function

Z Z  
4 δ 2 W [J] δ 2 Γ[φcl ] 4 δφcl (x1 ) δJ(z)
dz · = dz −
δJ(x1 )δJ(z) δφcl (z)δφcl (x2 ) δJ(z) δφcl (x2 )
δφcl (x1 )
= − = −δ(x1 − x2 )
δφcl (x2 )
This is a functional matrix equation, so the 1P I Green’s function is the
inverse of the connected Green’s function with J, φcl set equal to zero, then
multiplying by (−i)2 i gives

(2) −1
Γ(2) (x1 x2 ) = iGC (x1 x2 ) (53)

[Inverse in a functional sense, cf free propagator.]

Three point 1PI Green’s function

∂
Similarly using ∂α M −1 (α) = −M −1 ∂M ∂α
M −1 with M → δ 2 Γ[φcl ]/δφcl (x1 )δφcl (x2 ),
2
or M → −δ W [J]/δJ(x1 )δJ(x2 ) and ∂/∂α → δ/δJ(x3 ) we have
−1

Z
δ 3 W [J] δ 2 W [J]
= + d4 ud4 v
δJ(x1 )δJ(x2 )δJ(x3 ) δJ(x1 )δJ(u)
δ δ 2 Γ[φcl ] δ 2 W [J]
×
δJ(x ) δφcl (u)δφcl(v) δJ(v)δJ(x2 )
| {z 3 }
δφcl (w) δ
d4 w
R
δJ(x ) δφcl (w)
| {z 3 }
δ 2 W [J ]
δJ (x3 )δJ (w)

Multiplying by (−i)3 i gives (J = 0, φcl = 0)

38
 Z
(3) (2) (2) (2)
GC (x1 , x2 , x3 ) = d4 ud4vd4 wGC (x1 , u)GC (x2 , v)GC (x3 , w)iΓ(3) (u, v, w)

giving

(2)
GC

iΓ(3)

=
(3)
Thus we see that in this case iΓ(3) = Gamp .

A similar derivation for iΓ(4) leads to

iΓ(4)

= + + crossings

So Γ(N ) are the fundamental Green’s functions we need to compute (i.e.

all other Green’s functions can be built from them). In fact they directly
give the (renormalised) coupling constants of the theory.
e.g. for 3!λ φ3 theory we have Γ(3) = −λ + . . .
for 4!λ φ4 theory we have Γ(4) = −λ + . . .
Special cases
(2)−1
iΓ(2) = −GC iΓ(3) = G(3)
amp

(N )
Note alternative notation G1P I ≡ iΓ(N ) .

39
A general proof of this result [ie Γ(N ) (x1 , . . . , xN ) is the sum of 1PI Feynman
graphs] is also given in Huang, (better Srednicki, p128) as a ‘thermodynamic
limit’ argument. Paraphrasing again, we consider a fictitious field theory
with action Γ[φ],
Z
1
R 4
−U [J,a]
e = [dφ]e− a [Γ[φ]+ d J(x)φ(x)]

where a ∼ ~ is taken as small. As

∞ Z
X 1
Γ[φ] = d4 x1 . . . d4 xN Γ(N ) (x1 , . . . , xN ) φ(x1 ) . . . φ(xN )
N =0
N!

then this field theory is non-local with vertices Γ(N ) . U[J, a] gives the con-
nected diagrams of this theory. Every propagator gives a factor a, every
source and every vertex gives a factor 1/a.

At tree level this is equal to a−1 . (A source a−1 will cancel the associated
propagator a, so we only need consider the propagators connecting the ver-
tices. There is always one less propagator than number of vertices, hence
result.)

In a diagram with IP propagators, IJ sources and V vertices then we have a factor of aIP −IJ −V . Note that using Euler’s

theorem (see later) then the expansion is actually aL−1 where L is the number of loops – the loop expansion. Here we

take L = 0.

Thus we have
 
 sum of tree level diagrams 
(2)
aU[J, a] = with propagator Γ(2)−1 ∼ GC + O(a)
 
and vertices Γ(N ) , N 6= 2
Alternatively as a → 0 (cf ~ → 0 and the classical limit) then the integral is
dominated by the saddle point (ie tree level), where the minimum, δΓ[φ]/δφ =
−J occurs when φ = φcl , so
Z
lim aU[J, a] = Γ[φcl ] + d4 x J(x)φcl (x)
a→0

= W [J] by definition
Hence equating the two aU[J, a]s gives the result. This is as shown above for
Γ(3) , Γ(4) .

40
8.4 Effective potentials
Alternatively to expanding Γ[φc ] in terms of 1PI Green’s functions, we may
expand as
Z
 
Γ[φcl ] = d4 x −V (φcl ) + 21 A(φcl )(∂µ φcl )2 + . . .

Note that V , A are functions of φcl not functionals, otherwise would have
products φcl (x1 )φcl (x2 ) etc.. . (This is actually an expansion in powers of
momentum about zero momentum.)

V (φcl ) is called the ‘effective potential’.

As δΓ[φcl ]/δφcl (z) = −dV (φcl (z))/dφcl (z) = −J(z) so if φcl is independent of
space-time z then
dV
=J
dφcl
Now when J = 0, φcl = hφi = hΩ|φH |Ωi ie the vacuum expectation value of
φ. So we can find the ground state of the system by minimising the effective
potential (justifies name).

For a free field, then

Z
W [J] = − 21 d4 xd4 y J(x)∆F (x − y)J(y)

and as
Z
δW [J]
φcl (z) = =− d4 y ∆F (z − y)J(y)
δJ(z)
which upon inverting gives
(∂ 2 + m2 )φcl (z) = J(z)
and thus
Z
Γ[φcl ] = W [J] − d4 z J(z)φcl (z)
Z
1
= 2 d4 xd4 y J(x)∆F (x − y)J(y)
Z


= d4 x 21 (∂µ φcl )2 − 21 m2 φ2cl

41
Hence V (φcl ) = 12 m2 φ2cl and so dV /dφcl = m2 φcl = 0 giving hΩ|φH |Ωi = 0.
This is a trivial result here, but not so in symmetry breaking (eg Higgs type)
models.

42
9 Momentum Green’s Functions
We take
Z
φ̃(p) = d4 x eip·x φ(x)

and then define momentum Green’s functions by

G̃(N ) (p1 , . . . , pN )(2π)4 δ(p1 + . . . + pN )

Z
= d4 x1 . . . d4 xN G(N ) (x1 , . . . , xN )e+i(p1 x1 +...pN xN ) (54)

All momenta are regarded as incoming. [often drop tilde]

The δ function factor occurs because translational invariance implies that

G(N ) (x1 . . . xN ) only depend on the differences of the xi .

So, for example we have for the free propagator

G̃(2) (p1 , p2 )(2π)4 δ(p1 + p2 )

Z Z 
4 4 d4 p −ip·(x1 −x2 ) i
= d x1 d x2 4
e 2 2
e+i(p1 ·x1 +p2 ·x2 )
(2π) p − m + iǫ
i
= 2 2
(2π)4 δ(p1 + p2 )
p1 − m + iǫ

Summary – Momentum space Feynman rules (for λφ4 interaction)

1. With each propagator carrying momentum p, we associate a factor

i
= i∆F (p) =
p2 − m2 + iǫ

2. With each vertex of four lines carrying momenta p1 , p2 , p3 , p4 associate

a factor
p4

p1

p2

p3 = −iλ with p1 + p2 + p3 + p4 = 0

d4 p
3. Integrate over internal momentum loops with a factor (2π)4

43
 4. Symmetry factor
# ways of joining vertices + external legs
S=
ν!
|{z} ×(4!)ν
λν (exp expansion)

[point 3 can be seen from

p
Z
d4 p i
with i∆F (0) =
(2π)4 p2 − m2 + iǫ

If p2 6= m2 particles are ‘off-shell’

If p2 = m2 particles are ‘on-shell’

Example of the symmetry factor

↔
S=
pick pick external x2 leg pick external x3 leg pick external x4 leg join up four
external x1 leg (same vertex) (other vertex) (same other vertex) remaining internal legs
z}|{ z}|{ z}|{ z}|{ z}|{
8 × 3 × 4 × 3 × 2
2!(4!)2
1
=
2

44