Feynman Diagrams for Beginners∗

Krešimir Kumerički†
arXiv:1602.04182v1 [physics.ed-ph] 8 Feb 2016

Department of Physics, Faculty of Science, University of Zagreb, Croatia

Abstract
We give a short introduction to Feynman diagrams, with many exer-
cises. Text is targeted at students who had little or no prior exposure to
quantum field theory. We present condensed description of single-particle
Dirac equation, free quantum fields and construction of Feynman amplitude
using Feynman diagrams. As an example, we give a detailed calculation of
cross-section for annihilation of electron and positron into a muon pair. We
also show how such calculations are done with the aid of computer.

Contents
1 Natural units 2

2 Single-particle Dirac equation 4

2.1 The Dirac equation . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 The adjoint Dirac equation and the Dirac current . . . . . . . . . 6
2.3 Free-particle solutions of the Dirac equation . . . . . . . . . . . . 6

3 Free quantum fields 9

3.1 Spin 0: scalar field . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 Spin 1/2: the Dirac field . . . . . . . . . . . . . . . . . . . . . . . 10
3.3 Spin 1: vector field . . . . . . . . . . . . . . . . . . . . . . . . . 10

4 Golden rules for decays and scatterings 11

5 Feynman diagrams 13
∗
Notes for the exercises at the Adriatic School on Particle Physics and Physics Informatics, 11
– 21 Sep 2001, Split, Croatia
†
[email protected]

1
2 1 Natural units

6 Example: e+ e− → µ+ µ− in QED 16
6.1 Summing over polarizations . . . . . . . . . . . . . . . . . . . . 17
6.2 Casimir trick . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
6.3 Traces and contraction identities of γ matrices . . . . . . . . . . . 18
6.4 Kinematics in the center-of-mass frame . . . . . . . . . . . . . . 20
6.5 Integration over two-particle phase space . . . . . . . . . . . . . . 20
6.6 Summary of steps . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6.7 Mandelstam variables . . . . . . . . . . . . . . . . . . . . . . . . 22

Appendix: Doing Feynman diagrams on a computer 22

1 Natural units
To describe kinematics of some physical system or event we are free to choose
units of measure of the three basic kinematical physical quantities: length (L),
mass (M) and time (T). Equivalently, we may choose any three linearly indepen-
dent combinations of these quantities. The choice of L, T and M is usually made
(e.g. in SI system of units) because they are most convenient for description of
our immediate experience. However, elementary particles experience a different
world, one governed by the laws of relativistic quantum mechanics.
Natural units in relativistic quantum mechanics are chosen in such a way that
fundamental constants of this theory, c and ~, are both equal to one. [c] = LT −1 ,
[~] = M L−2 T −1 , and to completely fix our system of units we specify the unit of
energy (M L2 T −2 ):
1 GeV = 1.6 · 10−10 kg m2 s−2 ,
approximately equal to the mass of the proton. What we do in practice is:

• we ignore ~ and c in formulae and only restore them at the end (if at all)

• we measure everything in GeV, GeV−1 , GeV2 , . . .

Example: Thomson cross section

Total cross section for scattering of classical electromagnetic radiation by a free
electron (Thomson scattering) is, in natural units,

8πα2
σT = . (1)
3m2e

To restore ~ and c we insert them in the above equation with general powers α and
β, which we determine by requiring that cross section has the dimension of area
 1 Natural units 3

(L2 ):
8πα2 α β
σT = ~ c (2)
3m2e
1
[σ] = L2 = (M L2 T −1 )α (LT −1 )β
M2
⇒ α=2, β = −2 ,
i.e.
8πα2 ~2
σT = = 0.665 · 10−24 cm2 = 665 mb . (3)
3m2e c2
Linear independence of ~ and c implies that this can always be done in a unique
way. 
Following conversion relations are often useful:

1 fermi = 5.07 GeV−1

1 GeV−2 = 0.389 mb
1 GeV−1 = 6.582 · 10−25 s
1 kg = 5.61 · 1026 GeV
1m = 5.07 · 1015 GeV−1
1s = 1.52 · 1024 GeV−1

Exercise 1 Check these relations.

Calculating with GeVs is much more elegant. Using me = 0.511·10−3 GeV

we get
8πα2
σT = 2
= 1709 GeV−2 = 665 mb . (4)
3me
right away.

Exercise 2 The decay width of the π 0 particle is

1
Γ= = 7.7 eV. (5)
τ
Calculate its lifetime τ in seconds. (By the way, particle’s half-life is equal to
τ ln 2.)
4 2 Single-particle Dirac equation

2 Single-particle Dirac equation

2.1 The Dirac equation
Turning the relativistic energy equation

E 2 = p2 + m2 . (6)

into a differential equation using the usual substitutions

∂
p → −i∇ , E→i , (7)
∂t
results in the Klein-Gordon equation:

( + m2 )ψ(x) = 0 , (8)

which, interpreted as a single-particle wave equation, has pproblematic negative

energy solutions. This is due to the negative root in E = ± p2 + m2 . Namely, in
relativistic mechanics this negative root could be ignored, but in quantum physics
one must keep all of the complete set of solutions to a differential equation.
In order to overcome this problem Dirac tried the ansatz∗

(iβ µ ∂µ + m)(iγ ν ∂ν − m)ψ(x) = 0 (9)

with β µ and γ ν to be determined by requiring consistency with the Klein-Gordon

equation. This requires γ µ = β µ and

γ µ ∂µ γ ν ∂ν = ∂ µ ∂µ , (10)

which in turn implies

(γ 0 )2 = 1 , (γ i )2 = −1 ,
{γ µ , γ ν } ≡ γ µ γ ν + γ ν γ µ = 0 for µ 6= ν .
This can be compactly written in form of the anticommutation relations
 
1 0 0 0
 0 −1 0 0 
{γ µ , γ ν } = 2g µν , g µν = 
 0 0 −1 0  .
 (11)
0 0 0 −1

These conditions are obviously impossible to satisfy with γ’s being equal to usual
numbers, but we can satisfy them by taking γ’s equal to (at least) four-by-four
matrices.
∗
ansatz: guess, trial solution (from German Ansatz: start, beginning, onset, attack)
 2 Single-particle Dirac equation 5

Now, to satisfy (9) it is enough that one of the two factors in that equation is
zero, and by convention we require this from the second one. Thus we obtain the
Dirac equation:
(iγ µ ∂µ − m)ψ(x) = 0 . (12)
ψ(x) now has four components and is called the Dirac spinor.
One of the most frequently used representations for γ matrices is the original
Dirac representation
   
0 1 0 i 0 σi
γ = γ = , (13)
0 −1 −σ i 0

where σ i are the Pauli matrices:

     
1 0 1 2 0 −i 3 1 0
σ = σ = σ = . (14)
1 0 i 0 0 −1

This representation is very convenient for the non-relativistic approximation, since

then the dominant energy terms (iγ 0 ∂0 − . . . − m)ψ(0) turn out to be diagonal.
Two other often used representations are

• the Weyl (or chiral) representation — convenient in the ultra-relativistic

regime (where E  m)

• the Majorana representation — makes the Dirac equation real; convenient

for Majorana fermions for which antiparticles are equal to particles

(Question: Why can we choose at most one γ matrix to be diagonal?)

Properties of the Pauli matrices:

†
σi = σi (15)
i∗ 2 i 2
σ = (iσ )σ (iσ ) (16)
[σ i , σ j ] = 2iijk σ k (17)
i j ij
{σ , σ } = 2δ (18)
σ i σ j = δ ij + iijk σ k (19)

where ijk is the totally antisymmetric Levi-Civita tensor (123 = 231 = 312 = 1,
213 = 321 = 132 = −1, and all other components are zero).

Exercise 3 Prove that (σ · a)2 = a2 for any three-vector a.

6 2 Single-particle Dirac equation

Exercise 4 Using properties of the Pauli matrices, prove that γ matrices in the
Dirac representation satisfy {γ i , γ j } = 2g ij = −2δ ij , in accordance with the
anticommutation relations. (Other components of the anticommutation relations,
(γ 0 )2 = 1, {γ 0 , γ i } = 0, are trivial to prove.)
†
Exercise 5 Show that in the Dirac representation γ 0 γ µ γ 0 = γ µ .

Exercise 6 Determine the Dirac Hamiltonian by writing the Dirac equation in the
form i∂ψ/∂t = Hψ. Show that the hermiticity of the Dirac Hamiltonian implies
that the relation from the previous exercise is valid regardless of the representa-
tion.
/ ≡ aµ γ µ , is often used.
The Feynman slash notation, a

2.2 The adjoint Dirac equation and the Dirac current

For constructing the Dirac current we need the equation for ψ(x)† . By taking the
Hermitian adjoint of the Dirac equation we get
←
ψ † γ 0 (i ∂/ + m) = 0 ,

and we define the adjoint spinor ψ̄ ≡ ψ † γ 0 to get the adjoint Dirac equation
←
ψ̄(x)(i ∂/ + m) = 0 .

ψ̄ is introduced not only to get aesthetically pleasing equations but also because
it can be shown that, unlike ψ † , it transforms covariantly under the Lorentz trans-
formations.

Exercise 7 Check that the current j µ = ψ̄γ µ ψ is conserved, i.e. that it satisfies
the continuity relation ∂µ j µ = 0.
Components of this relativistic four-current are j µ = (ρ, j). Note that ρ =
j = ψ̄γ 0 ψ = ψ † ψ > 0, i.e. that probability is positive definite, as it must be.
0

2.3 Free-particle solutions of the Dirac equation

Since we are preparing ourselves for the perturbation theory calculations, we need
to consider only free-particle solutions. For solutions in various potentials, see the
literature.
The fact that Dirac spinors satisfy the Klein-Gordon equation suggests the
ansatz
ψ(x) = u(p)e−ipx , (20)
 2 Single-particle Dirac equation 7

which after inclusion in the Dirac equation gives the momentum space Dirac equa-
tion
(p/ − m)u(p) = 0 . (21)
This has two positive-energy solutions
 
(σ)
χ
u(p, σ) = N  , σ = 1, 2 , (22)
 
σ · p (σ) 
χ
E+m
where    
(1) 1 (2) 0
χ = , χ = , (23)
0 1
and two negative-energy solutions which are then interpreted as positive-energy
antiparticle solutions
σ·p
 
2 (σ)
(iσ )χ 
v(p, σ) = −N  E + m  , σ = 1, 2, E>0. (24)

2 (σ)
(iσ )χ

N is the normalization constant to be determined later. Spinors above agree with

those of [1]. The momentum-space Dirac equation for antiparticle solutions is
(p/ + m)v(p, σ) = 0 . (25)
It can be shown that the two solutions, one with σ = 1 and another with σ = 2,
correspond to the two spin states of the spin-1/2 particle.

Exercise 8 Determine momentum-space Dirac equations for ū(p, σ) and v̄(p, σ).

Normalization
In non-relativistic single-particle quantum mechanics normalization of a wave-
function is straightforward. Probability that the particle is somewhere in space is
equal to one, and this translates into the normalization condition ψ ∗ ψ dV = 1.
R

On the other hand, we will eventually use spinors (22) and (24) in many-particle
quantum field theory so their normalization is not unique. We will choose nor-
malization convention where we have 2E particles in the unit volume:
Z Z
ρ dV = ψ † ψ dV = 2E (26)
unit volume unit volume

This choice is relativistically covariant because the Lorentz contraction of the vol-
ume element is compensated by the energy change. There are other normalization
conventions with other advantages.
8 2 Single-particle Dirac equation

Exercise 9 Determine the normalization constant N conforming to this choice.

Completeness

Exercise 10 Using the explicit expressions (22) and (24) show that
X
u(p, σ)ū(p, σ) = p/ + m , (27)
σ=1,2
X
v(p, σ)v̄(p, σ) = p/ − m . (28)
σ=1,2

These relations are often needed in calculations of Feynman diagrams with unpo-
larized fermions. See later sections.

Parity and bilinear covariants

The parity transformation:

• P : x → −x, t → t

• P : ψ → γ 0ψ

Exercise 11 Check that the current j µ = ψ̄γ µ ψ transforms as a vector under par-
ity i.e. that j 0 → j 0 and j → −j.

Any fermion current will be of the form ψ̄Γψ, where Γ is some four-by-four
matrix. For construction of interaction Lagrangian we want to use only those
currents that have definite Lorentz transformation properties. To this end we first
define two new matrices:
 
5 0 1 2 3 Dirac rep. 0 1
γ ≡ iγ γ γ γ = , {γ 5 , γ µ } = 0 , (29)
1 0

i
σ µν ≡ [γ µ , γ ν ] , σ µν = −σ νµ . (30)
2

Now ψ̄Γψ will transform covariantly if Γ is one of the matrices given in the
following table. Transformation properties of ψ̄Γψ, the number of different γ
 3 Free quantum fields 9

matrices in Γ, and the number of components of Γ are also displayed.

Γ transforms as # of γ’s # of components

1 scalar 0 1
γµ vector 1 4
σ µν tensor 2 6
γ 5γ µ axial vector 3 4
γ5 pseudoscalar 4 1

This exhausts all possibilities. The total number of components is 16, meaning
that the set {1, γ µ , σ µν , γ 5 γ µ , γ 5 } makes a complete basis for any four-by-four
matrix. Such ψ̄Γψ currents are called bilinear covariants.

3 Free quantum fields

Single-particle Dirac equation is (a) not exactly right even for single-particle sys-
tems such as the H-atom, and (b) unable to treat many-particle processes such as
the β-decay n → p e− ν̄. We have to upgrade to quantum field theory.
Any Dirac field is some superposition of the complete set
u(p, σ)e−ipx , v(p, σ)eipx , σ = 1, 2, p ∈ R3
and we can write it as
XZ d3 p
u(p, σ)a(p, σ)e−ipx + v(p, σ)ac† (p, σ)eipx . (31)
 
ψ(x) = p
σ (2π)3 2E
p
Here 1/ (2π)3 2E is a normalization factor (there are many different conven-
tions), and a(p, σ) and ac† (p, σ) are expansion coefficients. To make this a quan-
tum Dirac field we promote these coefficients to the rank of operators by imposing
the anticommutation relations
{a(p, σ), a† (p0 , σ 0 )} = δσσ0 δ 3 (p − p0 ), (32)
and similarly for ac (p, σ). (For bosonic fields we would have a commutation
relations instead.) This is similar to the promotion of position and momentum
to the rank of operators by the [xi , pj ] = i~δij commutation relations, which is
why is this transition from the single-particle quantum theory to the quantum field
theory sometimes called second quantization.
Operator a† , when operating on vacuum state |0i, creates one-particle state
|p, σi
a† (p, σ)|0i = |p, σi , (33)
10 3 Free quantum fields

and this is the reason that it is named a creation operator. Similarly, a is an anni-
hilation operator
a(p, σ)|p, σi = |0i , (34)
and ac† and ac are creation and annihilation operators for antiparticle states (c in
ac stands for “conjugated”).
Processes in particle physics are mostly calculated in the framework of the
theory of such fields — quantum field theory. This theory can be described at
various levels of rigor but in any case is complicated enough to be beyond the
scope of these notes.
However, predictions of quantum field theory pertaining to the elementary
particle interactions can often be calculated using a relatively simple “recipe” —
Feynman diagrams.
Before we turn to describing the method of Feynman diagrams, let us just
specify other quantum fields that take part in the elementary particle physics inter-
actions. All these are free fields, and interactions are treated as their perturbations.
Each particle type (electron, photon, Higgs boson, ...) has its own quantum field.

3.1 Spin 0: scalar field

E.g. Higgs boson, pions, ...

d3 p
Z
a(p)e−ipx + ac† (p)eipx
 
φ(x) = p (35)
(2π)3 2E

3.2 Spin 1/2: the Dirac field

E.g. quarks, leptons
We have already specified the Dirac spin-1/2 field. There are other types: Weyl
and Majorana spin-1/2 fields but they are beyond our scope.

3.3 Spin 1: vector field

Either
• massive (e.g. W,Z weak bosons) or
• massless (e.g. photon)

XZ d3 p
µ
 (p, λ)a(p, λ)e−ipx + µ∗ (p, λ)a† (p, λ)eipx
µ 
A (x) = p
λ
(2π)3 2E
(36)
 4 Golden rules for decays and scatterings 11

µ (p, λ) is a polarization vector. For massive particles it obeys

pµ µ (p, λ) = 0 (37)

automatically, whereas in the massless case this condition can be imposed thanks
to gauge invariance (Lorentz gauge condition). This means that there are only
three independent polarizations of a massive vector particle: λ = 1, 2, 3 or λ =
+, −, 0. In massless case gauge symmetry can be further exploited to eliminate
one more polarization state leaving us with only two: λ = 1, 2 or λ = +, −.
Normalization of polarization vectors is such that

∗ (p, λ) · (p, λ) = −1 . (38)

E.g. for a massive particle moving along the z-axis (p = (E, 0, 0, |p|)) we can
take    
0 |p|
1  1   , (p, 0) = 1  0 
 
(p, ±) = ∓ √  (39)
2  ±i  m 0 
0 E

Exercise 12 Calculate X
µ∗ (p, λ)ν (p, λ)
λ

Hint: Write it in the most general form (Ag µν + Bpµ pν ) and then determine A
and B.

The obtained result obviously cannot be simply extrapolated to the massless

case via the limit m → 0. Gauge symmetry makes massless polarization sum
somewhat more complicated but for the purpose of the simple Feynman diagram
calculations it is permissible to use just the following relation
X
µ∗ (p, λ)ν (p, λ) = −g µν .
λ

4 Golden rules for decays and scatterings

Principal experimental observables of particle physics are

• scattering cross section σ(1 + 2 → 10 + 20 + · · · + n0 )

• decay width Γ(1 → 10 + 20 + · · · + n0 )

12 4 Golden rules for decays and scatterings

On the other hand, theory is defined in terms of Lagrangian density of quantum

fields, e.g.
1 1 g
L = ∂µ φ∂ µ φ − m2 φ2 − φ4 .
2 2 4!
How to calculate σ’s and Γ’s from L?
To calculate rate of transition from the state |αi to the state |βi in the pres-
ence of the interaction potential VI in non-relativistic quantum theory we have the
Fermi’s Golden Rule
   
α→β 2π 2 density of final
= |hβ|VI |αi| × . (40)
transition rate ~ quantum states
This is in the lowest order perturbation theory. For higher orders we have terms
with products of more interaction potential matrix elements h|VI |i.
In quantum field theory there is a counterpart to these matrix elements — the
S-matrix:
hβ|VI |αi + (higher-order terms) −→ hβ|S|αi . (41)
On one side, S-matrix elements can be perturbatively calculated (knowing the
interaction Lagrangian/Hamiltonian) with the help of the Dyson series
(−i)2
Z Z
4
S = 1 − i d x1 H(x1 ) + d4 x1 d4 x2 T {H(x1 )H(x2 )} + · · · , (42)
2!
and on another, we have “golden rules” that associate these matrix elements with
cross-sections and decay widths.
It is convenient to express these golden rules in terms of the Feynman invariant
amplitude M which is obtained by stripping some kinematical factors off the S-
matrix:
Y 1
hβ|S|αi = δβα − i(2π)4 δ 4 (pβ − pα )Mβα p . (43)
(2π) 3 2E
i=α,β i

Now we have two rules:

• Partial decay rate of 1 → 10 + 20 + · · · + n0
n
1 2 4 4 0 0
Y d3 p0i
dΓ = |Mβα | (2π) δ (p1 − p1 − · · · − pn ) , (44)
2E1 i=1
(2π)3 2Ei0

• Differential cross section for a scattering 1 + 2 → 10 + 20 + · · · + n0

n
1 1 1 2 4 4 0 0
Y d3 p0i
dσ = |Mβα | (2π) δ (p1 + p2 − p1 − · · · − pn ) ,
uα 2E1 2E2 i=1
(2π)3 2Ei0
(45)
 5 Feynman diagrams 13

where uα is the relative velocity of particles 1 and 2:

p
(p1 · p2 )2 − m21 m22
uα = , (46)
E1 E2

and |M|2 is the Feynman invariant amplitude averaged over unmeasured particle
spins (see Section 6.1). The dimension of M, in units of energy, is
• for decays [M] = 3 − n

• for scattering of two particles [M] = 2 − n

where n is the number of produced particles.
So calculation of some observable quantity consists of two stages:
1. Determination of |M|2 . For this we use the method of Feynman diagrams
to be introduced in the next section.

2. Integration over the Lorentz invariant phase space

n
Y d3 p0i
4 4
dLips = (2π) δ (p1 + p2 − p01 − ··· − p0n ) .
i=1
(2π)3 2Ei0

5 Feynman diagrams

Example: φ4 -theory

1 1 g
L = ∂µ φ∂ µ φ − m2 φ2 − φ4
2 2 4!
• Free (kinetic) Lagrangian (terms with exactly two fields) determines parti-
cles of the theory and their propagators. Here we have just one scalar field:
φ

• Interaction Lagrangian (terms with three or more fields) determines possible

vertices. Here, again, there is just one vertex:
φ
φ

φ
φ
14 5 Feynman diagrams

We construct all possible diagrams with fixed outer particles. E.g. for scatter-
ing of two scalar particles in this theory we would have
1
3

M(1 + 2 → 3 + 4) = + + + ...
2
4
t

In these diagrams time flows from left to right. Some people draw Feynman
diagrams with time flowing up, which is more in accordance with the way we
usually draw space-time in relativity physics.
Since each vertex corresponds to one interaction Lagrangian (Hamiltonian)
term in (42), diagrams with loops correspond to higher orders of perturbation
theory. Here we will work only to the lowest order, so we will use tree diagrams
only.
To actually write down the Feynman amplitude M, we have a set of Feynman
rules that associate factors with elements of the Feynman diagram. In particular,
to get −iM we construct the Feynman rules in the following way:

• the vertex factor is just the i times the interaction term in the (momentum
space) Lagrangian with all fields removed:

φ
φ
g removing fields g
iLI = −i φ4 ⇒ = −i (47)
4! φ
4!
φ

• the propagator is i times the inverse of the kinetic operator (defined by the
free equation of motion) in the momentum space:

Euler-Lagrange eq.
Lfree −→ (∂µ ∂ µ + m2 )φ = 0 (Klein-Gordon eq.) (48)

Going to the momentum space using the substitution ∂ µ → −ipµ and then
taking the inverse gives:

φ i
(p2 − m2 )φ = 0 ⇒ = (49)
p2 − m2

(Actually, the correct Feynman propagator is i/(p2 − m2 + i), but for our
purposes we can ignore the infinitesimal i term.)
 5 Feynman diagrams 15

• External lines are represented by the appropriate polarization vector or spinor

(the one that stands by the appropriate creation or annihilation operator in
the fields (31), (35), (36) and their conjugates):

particle Feynman rule

ingoing fermion u
outgoing fermion ū
ingoing antifermion v̄
outgoing antifermion v
ingoing photon µ
outgoing photon µ∗
ingoing scalar 1
outgoing scalar 1

So the tree-level contribution to the scalar-scalar scattering amplitude in this

4
φ theory would be just
g
−iM = −i . (50)
4!


Exercise 13 Determine the Feynman rules for the electron propagator and for the
only vertex of quantum electrodynamics (QED):
1
/ − m)ψ − Fµν F µν
L = ψ̄(i∂/ + eA F µν = ∂ µ Aν − ∂ ν Aµ . (51)
4
Note that also
P
p i σ u(p, σ)ū(p, σ)
= , (52)
p2 − m2
i.e. the electron propagator is just the scalar propagator multiplied by the polar-
ization sum. It is nice that this generalizes to propagators of all particles. This is
very helpful since inverting the photon kinetic operator is non-trivial due to gauge
symmetry complications. Hence, propagators of vector particles are

p µ pν
 
µν
−i g − 2
p, m m
massive: = 2 2
, (53)
p −m

p −ig µν
massless: = . (54)
p2
16 6 Example: e+ e− → µ+ µ− in QED

This is in principle almost all we need to know to be able to calculate the

Feynman amplitude of any given process. Note that propagators and external-line
polarization vectors are determined only by the particle type (its spin and mass)
so that the corresponding rules above are not restricted only to the φ4 theory and
QED, but will apply to all theories of scalars, spin-1 vector bosons and Dirac
fermions (such as the standard model). The only additional information we need
are the vertex factors.
“Almost” in the preceding paragraph alludes to the fact that in general Feyn-
man diagram calculation there are several additional subtleties:

• In loop diagrams some internal momenta are undetermined and we have

to integrate over those. Also, there is an additional factor (-1) for each
closed fermion loop. Since we will consider tree-level diagrams only, we
can ignore this.

• There are some combinatoric numerical factors when identical fields come
into a single vertex.

• Sometimes there is a relative (-) sign between diagrams.

• There is a symmetry factor if there are identical particles in the final state.

For explanation of these, reader is advised to look in some quantum field the-
ory textbook.

6 Example: e+e− → µ+µ− in QED

There is only one contributing tree-level diagram:

LNM TVU
     ./02143561)7

BDC4EF GDH4IKJ

 

  
!#"%$'&)( *+&-, 9#8 :%;=<?> @<A
OQP R#S
We write down the amplitude using the Feynman rules of QED and following
 6 Example: e+ e− → µ+ µ− in QED 17

fermion lines backwards. Order of lines themselves is unimportant.

−igµν
−iM = [ū(p3 , σ3 )(ieγ ν )v(p4 , σ4 )] [v̄(p2 , σ2 )(ieγ µ )u(p1 , σ1 )] ,
(p1 + p2 )2
(55)
or, introducing abbreviation u1 ≡ u(p1 , σ1 ),

e2
M= 2
[ū3 γµ v4 ][v̄2 γ µ u1 ] . (56)
(p1 + p2 )

Exercise 14 Draw Feynman diagram(s) and write down the amplitude for Comp-
ton scattering γe− → γe− .

6.1 Summing over polarizations

If we knew momenta and polarizations of all external particles, we could calculate
M explicitly. However, experiments are often done with unpolarized particles so
we have to sum over the polarizations (spins) of the final particles and average
over the polarizations (spins) of the initial ones:

sum over final pol.

1 1 X
z}|{
X
|M|2 → |M|2 = |M|2 . (57)
2 2 σσ σ3 σ4
| {z 1 }2
avg. over initial pol.

Factors 1/2 are due to the fact that each initial fermion has two polarization
(spin) states.
(Question: Why we sum probabilities and not amplitudes?)
In the calculation of |M|2 = M∗ M, the following identity is needed

[ūγ µ v]∗ = [u† γ 0 γ µ v]† = v † γ µ† γ 0 u = [v̄γ µ u] . (58)

Thus,

e4 X
|M|2 = 4
[v̄4 γµ u3 ][ū1 γ µ v2 ][ū3 γν v4 ][v̄2 γ ν u1 ] . (59)
4(p1 + p2 ) σ
1,2,3,4
18 6 Example: e+ e− → µ+ µ− in QED

6.2 Casimir trick

Sums P
over polarizations are easily performed using the following trick. First we
write [ū1 γ µ v2 ][v̄2 γ ν u1 ] with explicit spinor indices α, β, γ, δ = 1, 2, 3, 4:
X µ ν
ū1α γαβ v2β v̄2γ γγδ u1δ . (60)
σ1 σ2

We can now move u1δ to the front (u1δ is just a number, element of u1 vector, so
it commutes with everything), and then use the completeness relations (27) and
(28),
X
u1δ ū1α = (p/1 + m1 )δα ,
σ1
X
v2β v̄2γ = (p/2 − m2 )βγ ,
σ2

which turn sum (60) into

µ ν
(p/1 + m1 )δα γαβ (p/2 − m2 )βγ γγδ = Tr[(p/1 + m1 )γ µ (p/2 − m2 )γ ν ] . (61)
This means that

e4
|M|2 = Tr[(p/1 + m1 )γ µ (p/2 − m2 )γ ν ] Tr[(p/4 − m4 )γµ (p/3 + m3 )γν ] .
4(p1 + p2 )4
(62)
Thus we got rid off all the spinors and we are left only with traces of γ matri-
ces. These can be evaluated using the relations from the following section.

6.3 Traces and contraction identities of γ matrices

All are consequence of the anticommutation relations {γ µ , γ ν } = 2g µν , {γ µ , γ 5 } =
0, (γ 5 )2 = 1, and of nothing else!

Trace identities
1. Trace of an odd number of γ’s vanishes:
1
z}|{
Tr(γ µ1 γ µ2 · · · γ µ2n+1 ) = Tr(γ µ1 γ µ2 · · · γ µ2n+1 γ 5 γ 5 )
(moving γ 5 over each γ µi ) = −Tr(γ 5 γ µ1 γ µ2 · · · γ µ2n+1 γ 5 )
(cyclic property of trace) = −Tr(γ µ1 γ µ2 · · · γ µ2n+1 γ 5 γ 5 )
= −Tr(γ µ1 γ µ2 · · · γ µ2n+1 )
= 0
 6 Example: e+ e− → µ+ µ− in QED 19

2. Tr 1 = 4

3.
(2.)
Trγ µ γ ν = Tr(2g µν − γ ν γ µ ) = 8g µν − Trγ ν γ µ = 8g µν − Trγ µ γ ν
⇒ 2Trγ µ γ ν = 8g µν ⇒ Trγ µ γ ν = 4g µν
This also implies:
a/b = 4a · b
Tr/

4. Exercise 15 Calculate Tr(γ µ γ ν γ ρ γ σ ). Hint: Move γ σ all the way to the

left, using the anticommutation relations. Then use 3.
Homework: Prove that Tr(γ µ1 γ µ2 · · · γ µ2n ) has (2n − 1)!! terms.

5. Tr(γ 5 γ µ1 γ µ2 · · · γ µ2n+1 ) = 0. This follows from 1. and from the fact that γ 5
consists of even number of γ’s.

6. Trγ 5 = Tr(γ 0 γ 0 γ 5 ) = −Tr(γ 0 γ 5 γ 0 ) = −Trγ 5 = 0

7. Tr(γ 5 γ µ γ ν ) = 0. (Same trick as above, with γ α 6= µ, ν instead of γ 0 .)

8. Tr(γ 5 γ µ γ ν γ ρ γ σ ) = −4iµνρσ , with 0123 = 1. Careful: convention with

0123 = −1 is also in use.

Contraction identities
1.
1
γ µ γµ = gµν (γ µ γ ν + γ ν γ µ ) = gµν g µν = 4
2 | {z }
2g µν

2.
γ µ γ α γµ = −4γ α + 2γ α = −2γ α
| {z }
−γµ γ α +2gµ
α

3. Exercise 16 Contract γ µ γ α γ β γµ .

4. γ µ γ α γ β γ γ γµ = −2γ γ γ β γ α

Exercise 17 Calculate traces in |M|2 :

Tr[(p/1 + m1 )γ µ (p/2 − m2 )γ ν ] = ?
Tr[(p/4 − m4 )γµ (p/3 + m3 )γν ] = ?

Exercise 18 Calculate |M|2

20 6 Example: e+ e− → µ+ µ− in QED

6.4 Kinematics in the center-of-mass frame

In e+ e− coliders often pi  me , mµ , i = 1, . . . , 4, so we can take

mi → 0 “high-energy” or “extreme relativistic” limit

Then

8e4
|M|2 = [(p1 · p3 )(p2 · p4 ) + (p1 · p4 )(p2 · p3 )] (63)
(p1 + p2 )4
To calculate scattering cross-section σ we have to specialize to some particular
frame (σ is not frame-independent). For e+ e− colliders the most relevant is the
center-of-mass (CM) frame:

FHGJILKMNILO +,.-0/1214573 6

 

    !"#%$&"*'( )


89;:<=>%?&=ABD@ C

Exercise 19 Express |M|2 in terms of E and θ.

6.5 Integration over two-particle phase space

Now we can use the “golden rule” (45) for the 1+2 → 3+4 differential scattering
cross-section

1 1 1
dσ = |M|2 dLips2 (64)
uα 2E1 2E2
where two-particle phase space to be integrated over is

d3 p3 d3 p 4
dLips2 = (2π)4 δ 4 (p1 + p2 − p3 − p4 ) . (65)
(2π)3 2E3 (2π)3 2E4
 6 Example: e+ e− → µ+ µ− in QED 21

First we integrate over four out of six integration variables, and we do this in
general frame. δ-function makes the integration over d3 p4 trivial giving
1
dLips2 = δ(E1 + E2 − E3 − E4 ) d3 p3 (66)
(2π)2
4E3 E4 |{z}
p23 d|p3 |dΩ3
Now we integrate over d|p3 | by noting that E3 and E4 are functions of |p3 |
q
E3 = E3 (|p3 |) = p23 + m23 ,
q q
E4 = p24 + m24 = p23 + m24 ,
and by δ-function relation
(0)
δ(|p3 | − |p3 |)
q q
2 2 2 2
δ(E1 + E2 − p3 + m3 − p3 + m4 ) = δ[f (|p3 |)] = 0 . (67)
|f (|p3 |)||p3 |=|p(0) |
3

(0)
Here |p3 | is just the integration variable and |p3 | is the zero of f (|p3 |) i.e. the
actual momentum of the third particle. After we integrate over d|p3 | we put
(0)
|p3 | → |p3 |.
Since
E3 + E4
f 0 (|p3 |) = − |p3 | , (68)
E3 E4
we get
|p3 |dΩ
dLips2 = 2
. (69)
16π (E1 + E2 )
Now we again specialize to the CM frame and note that the flux factor is
q
4E1 E2 uα = 4 (p1 · p2 )2 − m21 m22 = 4|p1 |(E1 + E2 ) , (70)
giving finally
dσCM 1 |p3 |
= 2 2
|M|2 . (71)
dΩ 64π (E1 + E2 ) |p1 |
Note that we kept masses in each step so this formula is generally valid for any
CM scattering.
For our particular e− e+ → µ− µ+ scattering this gives the final result for dif-
ferential cross-section (introducing the fine structure constant α = e2 /(4π))
dσ α2
= 2
(1 + cos2 θ) . (72)
dΩ 16E
Exercise 20 Integrate this to get the total cross section σ.
Note that it is obvious that σ ∝ α2 , and that dimensional analysis requires
σ ∝ 1/E 2 , so only angular dependence (1 + cos2 θ) tests QED as a theory of
leptons and photons.
22 6 Example: e+ e− → µ+ µ− in QED

6.6 Summary of steps

To recapitulate, calculating (unpolarized) scattering cross-section (or decay width)
consists of the following steps:

1. drawing the Feynman diagram(s)

2. writing −iM using the Feynman rules

3. squaring M and using the Casimir trick to get traces

4. evaluating traces

5. applying kinematics of the chosen frame

6. integrating over the phase space

6.7 Mandelstam variables

Mandelstam variables s, t and u are often used in scattering calculations. They
are defined (for 1 + 2 → 3 + 4 scattering) as

s = (p1 + p2 )2
t = (p1 − p3 )2
u = (p1 − p4 )2

Exercise 21 Prove that s + t + u = m21 + m22 + m23 + m24

This means that only two Mandelstam variables are independent. Their main
advantage is that they are Lorentz invariant which renders them convenient for
Feynman amplitude calculations. Only at the end we can exchange them for “ex-
perimenter’s” variables E and θ.

Exercise 22 Express |M|2 for e− e+ → µ− µ+ scattering in terms of Mandelstam

variables.

Appendix: Doing Feynman diagrams on a computer

There are several computer programs that can perform some or all of the steps in
the calculation of Feynman diagrams. Here is a simple session with one such pro-
gram, FeynCalc [2] package for Wolfram’s Mathematica, where we calculate
the same process, e− e+ → µ− µ+ , that we just calculated in the text. Alternative
framework, relying only on open source software is FORM [3].
 REFERENCES 23

FeynCalcDemo.nb 1

FeynCalc demonstration
This Mathematica notebook demonstrates computer calculation of Feynman invariant amplitude for
e- e+ ® Μ- Μ+ scattering, using Feyncalc package.

First we load FeynCalc into Mathematica

In[1]:= << HighEnergyPhysics‘fc‘

FeynCalc 4.1.0.3b Evaluate ?FeynCalc for help or visit www.feyncalc.org

Spin−averagedFeynman amplitude squared È M È2 after using Feynman rules and applying the Casimir trick:


e4
4 Hp1 + p2L4
In[2]:= Msq = €€€€€€€€€€€€€€€€
€€€€€€€€€€€€€€€ Contract@Tr@HGS@p1D + meL.GA@ΜD.HGS@p2D - meL.GA@ΝDD

Tr@HGS@p4D - mmL.GA@ΜD.HGS@p3D + mmL.GA@ΝDDD

€€€€€€€€€€€€€€€€€€€€€€ He4 H64 mm2 me2 + 32 p3 × p4 me2 + 32 mm2 p1 × p2 + 32 p1 × p4 p2 × p3 + 32 p1 × p3 p2 × p4LL

1
4 Hp1 + p2L4
Out[2]= €€€€€€€€€€€€€€€€

Traces were evaluated and contractions performed automatically. Now we introduce Mandelstam variables by substitu-
tion rules,

mandelstam = 9prod@p1, p2D ® Hs - me2 - me2 L  2, prod@p3, p4D ® Hs - mm2 - mm2 L  2,

In[3]:= prod@a_, b_D := Pair@Momentum@aD, Momentum@bDD;

prod@p1, p3D ® Ht - me2 - mm2 L  2, prod@p2, p4D ® Ht - me2 - mm2 L  2,

!!!!
prod@p1, p4D ® Hu - me2 - mm2 L  2, prod@p2, p3D ® Hu - me2 - mm2 L  2, Hp1 + p2L ® s =;

and apply these substitutions to our amplitude:

In[5]:= Msq . mandelstam

€€€€€€€€€2€ €€ Ie4 I64 mm2 me2 + 16 Hs - 2 mm2 L me2 + 8 H-me2 - mm2 + tL + 8 H-me2 - mm2 + uL + 16 mm2 Hs - 2 me2 LMM
1 2 2
Out[5]=
4s

This result can be simplified by eliminating one Mandelstam variable:

In[6]:= Simplify@TrickMandelstam@%, s, t, u, 2 me2 + 2 mm2 DD

2 e4 H2 me4 + 4 Hmm2 - uL me2 + 2 mm4 + s2 + 2 u2 - 4 mm2 u + 2 s uL

Out[6]= €€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€
€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€
€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€
€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€
€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€
€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€€
€€€€€€€€€€€€€€€€
€€€€€€
s2

If we go to ultra−relativistic limit, we get result in agreement with our hand calculation:

In[7]:= Simplify@%% . 8mm ® 0, me ® 0<D

2 e4 Ht2 + u2 L
Out[7]= €€€€€€€€€€€€€€€€
€€€€€€€€€€€€€€€€
€€€€€€ €
s2

References
[1] D. Griffiths, Introduction to elementary particles, Wiley (1987)
24 REFERENCES

[2] V. Shtabovenko, R. Mertig and F. Orellana, New Developments in FeynCalc

9.0, arXiv:1601.01167 [hep-ph].

[3] J. A. M. Vermaseren, New features of FORM, math-ph/0010025.