Dirac Trace Techniques

Consider a QED amplitude involving one incoming electron with momentum p and spin
s, one outgoing electron with momentum p′ and spin s′ , and some photons. There may
be several Feynman diagrams contributing to this amplitude, but they all have the same
external legs and the corresponding factors u(p, s) and ū(p′ , s′ ). Consequently, we may write
the amplitude as

e−′ , . . . M e− , . . . = ū(p′ , s′ )Γu(p, s) (1)

where Γ is comprises all the other factors of the QED Feynman rules; for the moment, we
don’t want to be specific, so Γ is just some kind of a 4 × 4 matrix.

In many experiments, the initial electrons come in un-polarized beam, 50% having spin
s = + 21 and 50% having s = − 12 . At the same time, the detector of the final electrons
measures their momenta p′ but is blind to their spins s′ . The cross-section σ measured by
such an experiment would be the average of the polarized cross-sections σ(s, s′ ) with respect
to initial spins s and the sum over the final spins s′ , thus

1 XX
σ = σ(s, s′ ). (2)
2 s ′
s

Similar averaging / summing rules apply to the un-polarized partial cross-sections,

dσ 1 X X dσ(s, s′ )
= , (3)
dΩ 2 s ′ dΩ
s

etc., etc. Since all total or partial cross-sections are proportional to mod-squares |M|2 of
amplitudes M, we need to know how to calculate

def 1 XX 2
|M|2 = M(s, s′ ) (4)
2 s ′
s

for amplitudes such as (1).

1
 To do such a calculation efficiently, we need to recall two things about Dirac spinors.
First,

If M = ū(p′ , s′ )Γu(p, s) then M∗ = ū(p, s)Γu(p′ , s′ ) (5)

where Γ = γ 0 Γ† γ 0 is the Dirac conjugate of the matrix Γ; for a product γλ · · · γν of Dirac

matrices, γλ · · · γν = γν · · · γλ . Second,
X
uα (p, s) × ūβ (p, s) = (6 p + m)αβ (6)
s

and likewise
X
uγ (p′ , s′ ) × ūδ (p′ , s′ ) = (6 p′ + m)γδ . (7)
s′

Combining these two facts, we obtain

X X
|M = ū(p′ , s′ )Γu(p, s)|2 = ū(p′ , s′ )Γu(p, s) × ū(p, s)Γu(p′ , s′ )
s,s′ s,s′
 
X X X
=  ūδ (p′ , s′ ) Γδα uα (p, s) × ūβ (p, s) Γβγ uγ (p′ , s′ )
s,s′ δ,α β,γ
!
X X
= Γδα Γβγ × uα (p, s) ūβ (p, s)
α,β,γ,δ s
!
X
× uγ (p′ , s′ ) ūδ (p′ , s′ )
s′
X
= Γδα Γβγ × (6 p + m)αβ × (6 p′ + m)γδ
α,β,γ,δ
X 
′
= matrix product (6 p + m) Γ (6 p + m) Γ
γγ
γ
 
= tr (6 p′ + m) Γ (6 p + m) Γ
(8)
and hence

1X 1  ′ 
e−′ , . . . M e− , . . . = ū(p′ , s′ )Γu(p, s) =⇒ |M|2 = tr (6 p + m) Γ (6 p + m) Γ .
2 ′ 2
s,s
(9)

A similar trace formula exists for un-polarized scattering of positrons. In this case, the

2
amplitude is

e+′ , . . . M e+ , . . . = v̄(p, s)Γv(p′ , s′ ) (10)

(note that v(p′ , s′ ) belongs to the outgoing positron while v̄(p, s) belongs to the incoming
s+ ), and we need to average |M|2 over s and sum over s′ . Using

X
vα (p, s) × v̄β (p, s) = (6 p − m)αβ (11)
s

and working through algebra similar to eq. (8), we arrive at

1X 1  
e+′ , . . . M e+ , . . . = v̄(p, s)Γv(p′ , s′ ) =⇒ |M|2 = tr (6 p − m) Γ (6 p′ − m) Γ .
2 ′ 2
s,s
(12)

Now suppose an electron with momentum p1 and spin s1 and a positron with momentum
p2 and spin s2 come in and annihilate each other. In this case, the amplitude has form

h. . .| M e− +
1 , e2 , . . . = v̄(p2 , s2 )Γu(p1 , s1 ) (13)

for some Γ, and if both electron and positron beams are un-polarized, we need to average the
|M|2 over both spins s1 and s2 . Again, there is a trace formula for such averaging, namely

1X 1  
h. . .| M e− +
1 , e2 , . . . = v̄(p2 , s2 )Γu(p1 , s1 ) =⇒ 2
|M| = tr (6 p2 −m) Γ (6 p1 +m) Γ .
4 s ,s 4
1 2

(14)

Finally, for a process in which an electron-positron pair is created, the amplitude has
form

e−′ +′ ′ ′ ′ ′
1 , e2 , . . . M |. . .i = ū(p1 , s1 )Γv(p2 , s2 ), (15)

and if we do not detect the spins of the outgoing electron and positron but only their
momenta, then we should sum the |M|2 over both spins s1 and s2 . Again, there is a trace

3
formula for this sum, namely

X  
e−′ +′ ′ ′ ′ ′
1 , e2 , . . . M |. . .i = ū(p1 , s1 )Γv(p2 , s2 ) =⇒ |M|2 = tr (6 p′1 + m) Γ (6 p′2 − m) Γ .
s1 ,s2
(16)

Processes involving 4 or more un-polarized fermions also have trace formulae for the spin
sums. For example, consider an e− + e+ collision in which a muon pair µ− + µ+ is created.
There is one tree diagram for this process,

µ− µ+

p′1 p′2

q (17)

p1 p2

e− e+

and it evaluates to

−ig λν
i µ− , µ+ M e− , e+ = × ū(µ− )(ieγλ )v(µ+ ) × v̄(e+ )(ieγν )u(e− )
q2
(18)
ie2
= × ū(µ− )γ ν v(µ+ ) × v̄(e+ )γν u(e− )
s

where

s = q 2 = (p1 + p2 )2 = (p′1 + p′2 )2 = Ec.m.

2
(19)

is the square of the total energy in the center-of-mass frame.

The amplitude (18) depends on the spins of all 4 particles involved. To get the partial
cross-section for an experiment using un-polarized beams of initial electrons and positrons
and a spin-blind muon detector, we need to average the |M|2 over both initial spins s1 , s2
and sum it over both final spins s′1 , s′2 , thus

|M|2
 
dσ 1 X
= where |M|2 ≡ |M|2 . (20)
dΩ c.m. 64π 2 s 4
s1 ,s2 ,s′1 ,s′2

4
 For the amplitude (18) at hand,

∗ e4  − ν + + −
 
+ λ − − +

M×M = 2 × ū(µ )γ v(µ ) × v̄(e )γν u(e ) × v̄(µ )γ u(µ ) × ū(e )γλ v(e )
s
— note sums over two separate Lorentz indices ν and λ —
e4    
= 2 × ū(µ− )γ ν v(µ+ ) × v̄(µ+ )γ λ u(µ− ) × v̄(e+ )γν u(e− ) × ū(e− )γλ v(e+ )
s
(21)
and consequently

 
e4 X
|M|2 = × ū(µ− )γ ν v(µ+ ) × v̄(µ+ )γ λ u(µ− ) ×
4s2
s′1 ,s′2
!
X (22)
× v̄(e+ )γν u(e− ) × ū(e− )γλ v(e+ )
s1 ,s2
e4 
′ ν ′ λ
  
= × tr (6 p1 + Mµ ) γ (6 p2 − Mµ ) γ × tr (6 p2 − me ) γν (6 p1 + me ) γλ .
4s2

Calculating Dirac Traces

Thus far, we have learned how to express un-polarized cross-sections in terms of Dirac
traces (i.e., traces of products of the Dirac γ λ matrices). In this section, we shall learn how
to calculate such traces.

Dirac traces do not depend on the specific form of the γ 0 , γ 1 , γ 2 , γ 4 matrices but are
completely determined by the Clifford algebra

{γ µ , γ ν } ≡ γ µ γ ν + γ ν γ µ = 2g µν . (23)

To see how this works, please recall the key property of the trace of any matrix product:
tr(AB) = tr(BA) for any two matrices A and B. This symmetry has two important corol-
laries:

• All commutators have zero traces, tr([A, B]) = 0 for any A and B.

5
 • Traces of products of several matrices have cyclic symmetry

tr(ABC · · · Y Z) = tr(BC · · · Y ZA) = tr(C · · · Y ZAB) = · · · = tr(ZABC · · · Y ).

(24)

Using these properties it is easy to show that

tr(γ µ γ ν ) = 4g µν =⇒ tr(6 a 6 b) = 4(ab) ≡ 4aµ bµ . (25)

Indeed,

tr(γ µ γ ν ) = tr(γ ν γ µ ) = tr 1 µ ν
= tr(g µν ) = g µν × tr(1) = g µν × 4,


2 {γ , γ } (26)

where the last equality follows from Dirac matrices being 4 × 4 and hence

tr(1) = 4. (27)

Next, all products of any odd numbers of the γ µ matrices have zero traces,

tr(γ µ ) = 0, tr(γ λ γ µ γ ν ) = 0, tr(γ λ γ µ γ ν γ ρ γ σ ) = 0, etc., (28)

and hence

tr(6 a) = 0, tr(6 a 6 b 6 c) = 0, tr(6 a 6 b 6 c 6 d 6 e) = 0, etc. (29)

To see how this works, we can use the γ 5 matrix which anticommutes with all the γ µ and
hence with any product Γ of an odd number of the Dirac γ’s, γ 5 Γ = −Γγ 5 . Combining this
observation with γ 5 γ 5 = 1, we have

Γ = γ 5 γ 5 Γ = −γ 5 Γγ 5 = − 12 [γ 5 Γ, γ 5 ] (30)

and therefore

tr(Γ) = − 12 tr([γ 5 Γ, γ 5 ]) = 0. (31)

Products of even numbers n = 2m of γ matrices have non-trivial traces, and we may

calculate them recursively in n. We already know the traces for n = 0 and n = 2, so consider

6
a product γ κ γ λ γ µ γ ν of n = 4 matrices. Thanks to the cyclic symmetry of the trace,
 
tr(γ κ γ λ γ µ γ ν ) = tr(γ λ γ µ γ ν γ κ ) = tr 21 {γ κ , γ λ γ µ γ ν } (32)

where the anticommutator follows from the Clifford algebra (23) and the Leibnitz rule,

{γ κ , γ λ γ µ γ ν } = {γ κ , γ λ }γ µ γ ν − γ λ {γ κ , γ µ }γ ν + γ λ γ µ {γ κ , γ ν }
(33)
= 2g κλ × γ µ γ ν − 2g κµ × γ λ γ ν + 2g κν × γ λ γ µ .

Consequently

tr(γ κ γ λ γ µ γ ν ) = g κλ × tr(γ µ γ ν ) − g κµ × tr(γ λ γ ν ) + g κν × tr(γ λ γ µ )

(34)
= 4g κλ g µν − 4g κµ g λν + 4g κν g λµ

and hence

tr(6 a 6 b 6 c 6 d) = 4(ab)(cd) − 4(ac)(bd) + 4(ad)(bc). (35)

Note that in eq. (34) we have expressed the trace of a 4–γ product to traces of 2–γ products.
Similar recursive formulae exist for all even numbers of γ matrices,

tr(γ ν1 γ ν2 · · · γ νn ) = tr 12 {γ ν1 , γ ν2 · · · γ νn }

for even n only

n
(36)
X
(−1)k g ν1νk × tr γ ν2 · · · γ νk · · · γ νn .


=
k=2

For example, for n = 6

tr(γ κ γ λ γ µ γ ν γ ρ γ σ ) = g κλ × tr(γ µ γ ν γ ρ γ σ ) − g κµ × tr(γ λ γ ν γ ρ γ σ ) + g κν × tr(γ λ γ µ γ ρ γ σ )

− g κρ × tr(γ λ γ µ γ ν γ σ ) + g κσ × tr(γ λ γ µ γ ν γ ρ )
 
= 4g κλ × g µν g ρσ − g µρ g νσ + g µσ g νρ
 
− 4g κµ × g λν g ρσ − g λρg νσ + g λσ g νρ
 
κν λµ ρσ λρ µσ λσ µρ
+ 4g × g g − g g + g g
 
− 4g κρ × g λµ g νσ − g λν g µσ + g λσ g µν
 
κσ λµ νρ λν µρ λρ µν
+ 4g × g g − g g + g g .
(37)

For products of more γ matrices, the recursive formulae (36) for traces produce even

7
more terms (105 terms for n = 8, 945 terms for n = 10, etc., etc.), so it helps to reduce
n whenever possible. For example, if the matrix product inside the trace contains two 6 a
matrices (for the same 4–vector aµ ) next to each other, you can simplify the product using
6 a 6 a = a2 , thus

tr(6 a 6 a 6 b · · · 6 c) = a2 × tr(6 b · · ·6 c). (38)

Also, when a product contains γ α and γα with the same Lorentz index α which should be
summed over, we may simplify the trace using

γ α γα = 4, γ α6 aγα = −2 6 a, γ α6 a 6 bγα = +4(ab), γ α6 a 6 b 6 cγα = −2 6 c 6 b 6 a, (39)

etc., cf. homework set#6, problem 1(a).

In the electroweak theory, one often needs to calculate traces of products containing the
γ 5 matrix. If the γ 5 appears more than once, we may simplify the product using γ 5 γ 5 = 1
and γ 5 γ ν = −γ ν γ 5 . For example,
   
tr γ µ (1 − γ 5 ) 6 p γ ν (1 − γ 5 ) 6 q = tr γ µ (1 − γ 5 ) 6 p (1 + γ 5 )γ ν 6 q
 
µ 5 5 ν
= tr γ (1 − γ )(1 − γ ) 6 p γ 6 q
 
= 2 tr γ µ (1 − γ 5 ) 6 p γ ν 6 q
(40)
because (1 − γ 5 )2 = 1 − 2γ 5 + γ 5 γ 5 = 2(1 − γ 5 )
 
5 µ ν
= 2 tr (1 + γ )γ 6 p γ 6 q
   
= 2 tr γ µ 6 p γ ν 6 q + 2 tr γ 5 γ µ 6 p γ ν 6 q .

When the γ 5 appears just one time, we may use γ 5 = iγ 0 γ 1 γ 2 γ 3 to show that

tr(γ 5 ) = 0, tr(γ 5 γ ν ) = 0, tr(γ 5 γ µ γ ν ) = 0, tr(γ 5 γ λ γ µ γ ν ) = 0, (41)

while

tr(γ 5 γ κ γ λ γ µ γ ν ) = −4iǫκλµν . (42)

For more γ ν matrices accompanying the γ 5 we have

tr(γ 5 γ ν1 · · · γ νn ) = 0 ∀ odd n, (43)

8
while for even n = 6, 8, . . . there are recursive formulae based on the identity

γ 5 γ λ γ µ γ ν = g λµ × γ 5 γ ν − g λν × γ 5 γ µ + g µν × γ 5 γ λ − iǫλµνρ × γρ . (44)

Muon Pair Production

As an example of trace technology, let us calculate the traces (22) for the muon pair
production. Let’s start with the trace due to summing over muons’ spins,

 
tr (6 p′1 + Mµ )γ λ (6 p′2 − Mµ )γ ν = tr(6 p′1 γ λ6 p′2 γ ν )
+ Mµ × tr(γ λ6 p′2 γ ν ) − Mµ × tr(6 p′1 γ λ γ ν ) (45)

− Mµ2 × tr(γ λ γ ν ).

On the second line here, we have three γ matrices inside each trace, so those traces vanish.
On the third line, tr(γ λ γ ν ) = 4g λν . Finally, the trace on the first line follows from eq. (34),

tr(6 p′1 γ λ6 p′2 γ ν ) = p′1α p′2β × tr(γ α γ λ γ β γ ν )

 
= p′1α p′2β × 4 g αλ × g βν − g αβ × g λν + g αν × g λβ (46)
λν
= 4p′λ ′ν ′ ′
1 × p2 − 4(p1 p2 ) × g + 4p′ν ′λ
1 × p2 .

Altogether,

 
tr (6 p′1 + Mµ )γ λ (6 p′2 − Mµ )γ ν = 4p′λ ′ν ′ν ′λ 2 ′ ′
1 p2 + 4p1 p2 − 4(Mµ + (p1 p2 )) × g
λν
(47)
λν
= 4p′λ ′ν ′ν ′λ
1 p2 + 4p1 p2 − 2s × g

where on the last line I have used

s ≡ (p′1 + p′2 )2 = p′2 ′2 ′ ′ 2 ′ ′

1 + p2 + 2(p1 p2 ) = 2Mµ + 2(p1 p2 ). (48)

9
 Similarly, for the second trace (22) due to averaging over electron’s and positron’s spins,
we have
 
tr (6 p2 − me ) γν (6 p1 + me ) γλ = tr(6 p2 γν 6 p1 γλ )
+ me × tr(6 p2 γν γλ ) − me × tr(γν 6 p1 γλ )
− m2e × tr(γν γλ )
= 4p2ν × p1λ − 4(p2 p1 ) × gνλ + 4p2λ × p1ν (49)

+ me × 0 − me × 0 − m2e × 4gνλ
= 4p2ν p1λ + 4p2λ p1ν − 4((p2 p1 ) + m2e ) × gλν
= 4p2ν p1λ + 4p2λ p1ν − 2s × gλν ,

where on the last line I have used

s ≡ (p1 + p2 )2 = p21 + p22 + 2(p2 p1 ) = 2m2e + 2(p2 p1 ). (50)

It remains to multiply the two traces and sum over the Lorentz indices λ and ν:
   
ν λ
tr (6 p′1 + Mµ )γ (6 p′2
− Mµ )γ × tr (6 p2 − me ) γν (6 p1 + me ) γλ =
   
νλ
= 4p′ν
1 2p ′λ
+ 4p ′λ ′ν
p
1 2 − 2s × g × 4p p
2ν 1ν + 4p p
2λ 1ν − 2s × g νλ

= 16(p′ν ′λ ′λ ′ν
1 p2 + p1 p2 ) × (p2ν p1λ + p2λ p1ν )

− 8s g νλ × (p2ν p1λ + p2λ p1ν )

− 8s gνλ × (p′λ ′ν ′ν ′λ
1 p2 + p1 p2 )

+ 4s2 × g λν gλν
 
= 16 × 2 × (p′1 p1 )(p′2 p2 ) + (p′2 p1 )(p′1 p2 )
− 8s × 2(p1 p2 ) − 8s × 2(p′1 p′2 ) + 4s2 × 4
= 32(p′1 p1 )(p′2 p2 ) + 32(p′2 p1 )(p′1 p2 ) + 16s × (Mµ2 + m2e )
(51)
where the last line follows from eqs. (48) and (50). Hence, in eq. (22) we have

1 X 2 4e4  ′ ′ ′ ′ 2 2

|M|2 ≡ |M| = 2 × 2(p1 p1 )(p2 p2 ) + 2(p2 p1 )(p1 p2 ) + s(Mµ + me ) . (52)
4 s
all spins

10
 Finally, let’s work out the kinematics of pair production. In the center-of-mass frame,
1
pν1,2 = (E, ±p) and p′ν ′ ′
1,2 = (E, ±p ), same E = 2 Ecm but p 6= p. Therefore,

(p′1 p1 ) = (p′2 p2 ) = E 2 − p′ · p,
(p′2 p1 ) = (p′1 p2 ) = E 2 + p′ · p,
s = 4E 2 ,
(53)
2(p′1 p1 )(p′2 p2 ) + 2(p′2 p1 )(p′1 p2 ) = 2(E 2 − p′ · p)2 + 2(E 2 + p′ · p)2
= 4E 4 + 4(p′ · p)2
= 4E 4 + 4p′2 p2 × cos2 θ,

and consequently

!
p′2 p2 Mµ2 + m2e
|M|2 = e4 1+ 2
× cos θ + . (54)
E4 E2

where p′2 = E 2 − Mµ2 and p2 = E 2 − m2e .

We may simplify this expression a bit using the experimental fact that the muon is much
heavier than the electron, Mµ ≈ 207me , so we need ultra-relativistic e∓ to produce µ∓ ,
E > Mµ ≫ me . This allows us to neglect the m2e term in eq. (54) and let p2 = E 2 , thus

! ! !
4 Mµ2 Mµ2 2
|M|2 = e 1+ 2 + 1− 2 × cos θ , (55)
E E

and consequently the partial cross-section is

! ! ! s
α2 Mµ2 Mµ2 Mµ2
 
dσ
= × 1+ + 1− × cos2 θ × 1− (56)
dΩ c.m. 4s E2 E2 E2

where the root comes from the phase-space factor |p′ |/|p| for inelastic processes.

Looking at the angular dependence of this partial cross-section, we see that just above the
energy threshold, for E = Mµ + small, the muons are produced isotropically in all directions.

11
On the other hand, for very high energies E ≫ Mµ when all 4 particles are ultra-relativistic,

 
dσ
∝ 1 + cos2 θ. (57)
dΩ c.m.

In the homework set#8 (problem 4) you will see that the polarized cross sections depend
on the angle as (1 ± cos θ)2 where the sign ± depends on the helicities of initial and final
particles; for the un-polarized particles, we average / sum over helicities, and that produces
the averaged angular distribution 1 + cos2 .

Finally, the total cross-section of muon pair production follows from eq. (56) using

4π
Z Z
2
d Ω = 4π, d2 Ω cos2 θ = , (58)
3

hence
!s
4π α2 Mµ2 Mµ2
σ tot (e− e+ → µ− µ+ ) = × 1+ 1− . (59)
3 s 2E 2 E2

Well above the threshold

4π α2
σ tot ≈ σ0tot ≡ . (60)
3 s

At the threshold the cross-section is zero, but it rises very rapidly with energy and for
E = 1.25Mµ , σ tot reaches about 80% of σ0tot .

σ tot /σ0tot

E
Mµ

12