The Dirac Equation

Asaf Pe’er1

February 11, 2014

This part of the course is based on Refs. [1], [2] and [3].

1. Introduction

So far we have only discussed scalar fields, such that under a Lorentz transformation
′ ′
x → xµ = λµ ν xν the field transforms as
µ

φ(x) → φ′ (x) = φ(Λ−1 x). (1)

We have seen that quantization of such fields gives rise to spin 0 particles. However, in the
framework of the standard model there is only one spin-0 particle known in nature: the
Higgs particle.
Most particles in nature have an intrinsic angular momentum, or spin. To describe
particles with spin, we are going to look at fields which themselves transform non-trivially
under the Lorentz group.
In this chapter we will describe the Dirac equation, whose quantization gives rise to
Fermionic spin 1/2 particles. To motivate the Dirac equation, we will start by studying the
appropriate representation of the Lorentz group.

2. The Lorentz group, its representations and generators

The first thing to realize is that the set of all possible Lorentz transformations form
what is known in mathematics as group.
By definition, a group G is defined as a set of objects, or operators (known as the
elements of G) that may be combined, or multiplied, to form a well-defined product in G
which satisfy the following conditions:

1. Closure: If a and b are two elements of G, then the product ab is also an element of
G.

1
Physics Dep., University College Cork
 –2–

2. The multiplication is associative: (ab)c = a(bc).

3. There is a unit element I in G, such that Ia = aI = a for every element a in G.

4. For every element a in G there is an inverse element a−1 , such that aa−1 = a−1 a = I.

Clearly, the set of all Lorentz transformations Λµ ν fulfill these conditions, and thus they
form a group, known as the (inhomogeneous) Lorentz group, or Poincaré group.1
While the group is an abstract mathematical entity, it is often very useful to represent
each group element by a matrix. (Technically, group representations describe abstract groups
in terms of linear transformations of vector spaces). This is particularly useful, as the group
operation can be represented by matrix multiplication. In fact, this is what we did so far,
without calling it as such !. The important point to note is that there could be more than a
single representation for each group element. We will see the implications of this below.
Let us now return to fields. We already saw an example of a field that transforms non-
trivially under Lorentz transformation - this is the vector field Aµ (x) of electromagnetism,

Aµ (x) → Λµ ν Aν (Λ−1 x). (2)

We will deal with this particular field later.

In general, under Lorentz transformation a field can transform as

φa (x) → D[Λ]a b φb (Λ−1 x), (3)

where the φa , φb represent different fields, and D[Λ]a b are matrices. The matrices D[Λ] form
a representation of the Lorentz group, namely it reflects all the multiplication properties of
the Lorentz transformations. As explained above, this means that (i) D[Λ1 ]D[Λ2 ] = D[Λ1 Λ2 ];
(ii) D[1] = 1; and (iii) D[Λ−1 ] = D[Λ]−1 .

2.1. Generators

The next question is: how do we find the different representations?

In order to answer this question, we first generalize our discussion to continuous groups,
known as Lie groups. In these groups, the parameters of a product element are analytical

1
Lorentz transformations can be split into 3 categories: translations, rotations and boosts. If we omit
translations and we consider only rotations and boosts, the resulting group is known as homogeneous
Lorentz group.
 –3–

functions (= having derivatives of all orders) of the parameters of the factors. The analytic
nature of the functions (differentiability) allows us to develop the concept of generators,
and reduce the study of the whole group to a study of the group elements in the neighborhood
of the identity element.
Roughly speaking, a set of generators is a set of group elements (all infinitesimally close
to the identity) such that possibly repeated application of the generators on themselves and
each other is capable of producing all the elements in the group. Thus, in order to find
all the representations, we look at the Lie algebra of infinitesimal transformations.
Clearly, the Lorentz group is a Lie group, so let’s look at infinitesimal transformations
of the Lorentz group as a demonstrator. Consider a Lorentz transformation that is infinites-
imally close to the identity
Λµ ν = δνµ + ǫω µ ν + O(ǫ2 ) (4)
for infinitesimal ǫ. We saw (see “Classical Fields”, Equation 49) that the condition for a
Lorentz transformation: Λµ σ Λν ρ η σρ = η µν is fulfilled (to first order in ǫ) provided that ω is
anti-symmetric:
ω µν + ω νµ = 0. (5)
Note that an antisymmetric 4 × 4 matrix has 4 × 3/2 = 6 independent components, which
agrees with the 6 transformations of the Lorentz group: 3 rotations and 3 boosts.
Let us now introduce a basis of these infinitesimal Lorentz transformations, namely a
basis of these six 4 × 4 anti-symmetric matrices. We could call them (MA )µν , with
A = 1, ..., 6. But in fact, it’s going to turn out better (although initially a little confusing)
to replace the single index A with a pair of antisymmetric indices [ρσ], where ρ, σ = 0, ..., 3.
We thus call our six basis matrices (Mρσ )µ ν .
The antisymmetry on the ρ and σ indices means that, for example, M01 = −M10 , etc.
Recall again that ρ and σ label six different matrices. The indices µ and ν label the
components of each matrix. Note that the matrices are also antisymmetric on the µν indices
because each one is by itself an antisymmetric matrices.
With this notation in place, we can write a basis of six 4 × 4 antisymmetric matrices as
(at the moment, we just pulled it out of a hat)

(Mρσ )µν = η ρµ η σν − η σµ η ρν , (6)

where the indices µ and ν are those of the 4 × 4 matrix, while ρ and σ denote which basis
element we are dealing with.
If we use these matrices for anything practical (for example, if we want to multiply them
 –4–

together, or act on some field) we will typically need to lower one index, so we have
(Mρσ )µ ν = η ρµ δνσ − η σµ δνρ . (7)
Since we lowered the index with the Minkowski metric, we pick up various minus signs which
means that when written in this form, the matrices are no longer necessarily antisymmetric.
Plugging directly in Equation 7, we can give two examples of these basis matrices:
   
0 1 0 0 0 0 0 0
1 0 0 0 0 0 −1 0
   
(M01 )µ ν =   and (M12 )µ ν =  (8)
0 0 0 0 0 1 0 0

0 0 0 0 0 0 0 0
The first, M01 , generates (infinitesimal) boosts in the x1 direction. It is real and symmetric.
The second, M12 , generates (infinitesimal) rotations in the (x1 , x2 )-plane. It is real and
anti-symmetric.
Being the basis of anti-symmetric matrices, we can now write any ω µ ν as a linear com-
bination of Mρσ ,
1
ω µ ν = Ωρσ (Mρσ )µ ν , (9)
2
where Ωρσ are just six numbers (again, we use the same notation, namely antisymmetric in
the indices), that tell us what Lorentz transformation we are doing.
The six basis matrices Mρσ are the generators of the Lorentz transformations. They
obey the Lorentz Lie algebra of Lorentz transformations,
[Mρσ , Mτ ν ] = η στ Mρν − η ρτ Mσν + η ρν Mστ − η σν Mρτ (10)
where we have suppressed the matrix indices. It should be stressed that these commutation
relations encapsulates all that is important about the Lorentz group; Any matrices that obey
these commutations relations form a representation of the Lorentz group. (You may be
familiar with that from SU(2) rotation group in QM). So what we really do is find matrices
which obey this commutation relations - the Lie algebra - and then we can use them to
express any Lorentz transformation.
The transformations ω µ ν are infinitesimally close to unity. We can express any finite
Lorentz transformation as the exponential2
1
Λ = exp( Ωρσ Mρσ ). (11)
2

2
This follows from the following: we first write ω = ǫω̃. Let ǫ = 1/n. Any finite transformation can be
written as Λ = limn→∞ (δ + (1/n)ω̃)n .
 –5–

Let me stress again what each of these objects are: the Mρσ are six 4 × 4 basis elements
of the Lorentz group; the Ωρσ are six numbers telling us what kind of Lorentz transformation
we are doing (for example, they say things like rotate by θ = π/7 about the x3 -direction and
run at speed v = 0.2 in the x1 direction).

3. The Spinor Representation

Having discussed (briefly) the generators of the Lorentz group, let us return to our
original question: how do we find the different representations of the Lorentz group ?
Basically, we need to find matrices which satisfy the Lorentz algebra commutation re-
lations, Equation 10.
We are going to construct the spinor representation. To do this, we start by defining
something which, at first sight, has nothing to do with the Lorentz group.

3.1. Clifford Algebra

Any set of objects γ µ which obey the anti-commutation relations is said to form a
Clifford algebra, namely

{γ µ , γ ν } ≡ γ ν γ ν + γ ν γ µ = 2η µν 1, (12)

where γ µ with µ = 0, 1, 2, 3 are a set of four matrices, and the 1 on the right hand side
denotes the unit matrix. This means that we are looking for 4 matrices such that

γ µ γ ν = −γ ν γ µ for µ 6= ν, (13)

and
(γ 0 )2 = 1 , (γ i )2 = −1 i = 1, 2, 3. (14)
It’s not hard to convince yourself that there are no representations of the Clifford algebra
using 2 × 2 or 3 × 3 matrices. The simplest representation of the Clifford algebra is in terms
of 4 × 4 matrices. There are many such examples of 4 × 4 matrices which obey Equation 12.
For example, we may take

0 σi
   
0 0 1 i
γ = , γ = (15)
1 0 −σ i 0
 –6–

where each element is itself a 2 × 2 matrix, with the σ i the Pauli matrices,
     
1 0 1 2 0 −i 3 1 0
σ = , σ = , σ = . (16)
1 0 i 0 0 −1

It is easy to check that these matrices themselves satisfy {σ i , σ j } = 2δ ij .

One can construct many other representations of the Clifford algebra by taking V γ µ V −1
for any invertible matrix V . However, up to this equivalence, one can prove that there is a
unique irreducible representation of the Clifford algebra. The matrices in Equation 15 provide
one example, known as the Weyl or chiral representation (for reasons that will soon
become clear). We will soon restrict ourselves further, and consider only representations of
the Clifford algebra that are related to the chiral representation by a unitary transformation
V.
So what does the Clifford algebra have to do with the Lorentz group? As we are going
to show shortly, the Lorentz group can be represented by the six matrices S ρσ , which are
defined by the commutator of two γ µ ,
 
ρσ 1 ρ σ 0 ρ=σ 1 1
S = [γ , γ ] = 1 ρ σ = γ ρ γ σ − η ρσ . (17)
4 2
γ γ ρ 6= σ 2 2

(You can verify this result directly, using the anti-commutation relations of the σ matrices).
Clearly, S µν = −S νµ , resulting in six such matrices.

3.1.1. Algebraic properties of S µν matrices.

Claim [1]. [S µν , γ ρ ] = γ µ η νρ − γ ν η ρµ .
Proof. For µ 6= ν we have

[S µν , γ ρ ] = 21 [γ µ γ ν , γ ρ ]
= 21 γ µ γ ν γ ρ − 21 γ ρ γ µ γ ν
= 12 γ µ {γ ν , γ ρ } − 21 γ µ γ ρ γ ν − 12 {γ ρ , γ µ } γ ν + 21 γ µ γ ρ γ ν
= γ µ η νρ − γ ν η ρµ .

Claim [2]. The matrices S µν form a representation of the Lorentz algebra

(Equation 10), in the sense that

[S µν , S ρσ ] = η νρ S µσ − η µρ S νσ + η µσ S νρ − η νσ S µρ (18)
 –7–

Proof. Taking ρ 6= σ and using the previous claim proven above, we have
1
[S µν , S ρσ ] = 2
[S µν , γ ρ γ σ ]
1
= 2
[S µν , γ ρ ]γ σ + 21 γ ρ [S µν , γ σ ] (19)
1 µ σ νρ
= 2
γ γ η − 21 γ ν γ σ η ρµ + 21 γ ρ γ µ η νσ − 1 ρ ν σµ
2
γ γ η

We can use now Equation 17 to write γ µ γ σ = 2S µσ + η µσ , and get

[S µν , S ρσ ] = S µσ η νρ − S νσ η ρµ + S ρµ η νσ − S ρν η σµ (20)

which completes the proof.

3.2. Dirac Spinors

The γ µ are 4 × 4 matrices, and therefore the S µν are also 4 × 4 matrices. We will call
the indices of the rows and columns of these matrices α, β = 1, 2, 3, 4.3
The matrices (S µν )α β act on a field. We introduce the Dirac spinor field ψ α (x), which
is an object with four complex components, which we label by α = 1, 2, 3, 4. Under Lorentz
transformation, we have
ψ α (x) → S[Λ]α β ψ β (Λ−1 x), (21)
where  
1
Λ = exp Ωρσ Mρσ (22)
2
and  
1 ρσ
S[Λ] = exp Ωρσ S . (23)
2
(Again: Mρσ are the generators of the Lorentz transformation Λ which transforms the
coordinates x, while S ρσ are the generators of the representation of Lorentz algebra which
determines the transformation of the fields ψ).
Although the basis of the generators Mρσ and S ρσ are different, we use the same six
numbers Ωρσ in both Λ and S[Λ]. This ensures that we are doing the same Lorentz trans-
formation on x and ψ. Note that we denote both the generator S ρσ and the full Lorentz
transformation S[Λ] as “S”. To avoid confusion, the latter will always come with the square
brackets [Λ].

3
We don’t use the labeling 0,1,2,3, as these have got nothing to do with space and time.
 –8–

Both Λ and S[Λ] are 4×4 matrices. So how can we be sure that the spinor representation
is something new, and isn’t equivalent to the familiar representation Λµ ν ? To see that the
two representations are truly different, let’s look at some specific transformations.
Example I: Rotations.
The generators S ij become

0 σi 0 σj i ijk σ k 0
    
ij 1
S = =− ǫ (for i 6= j) (24)
2 −σ i 0 −σ j 0 2 0 σk

If we define the rotation parameters as Ωij = −ǫijk ϕk (namely, Ω12 = −ϕ3 , etc.), then the
rotation matrix becomes
 +i~ϕ·~σ/2 
1 ρσ e 0
S[Λ] = exp( Ωρσ S ) = (25)
2 0 e+i~ϕ·~σ/2

where we need to remember that Ω12 = −Ω21 = −ϕ3 when following factors of 2.
Consider as an example a rotation by 2π about the x3 -axis. This is achieved by ϕ
~ =
(0, 0, 2π). The spinor rotation matrix becomes
3
e+iπσ
 
0
S[Λ] = 3 = −1. (26)
0 e+iπσ

Thus, under a 2π rotation we get

ψ α (x) → −ψ α (x) (27)

which is not the same as the transformation law of a vector!. There is this extra minus
sign (which connects to the Fermi-Dirac statistics, as we will see later).
To check that we haven’t been cheating with factors of 2, let’s see how a vector would
~ = (0, 0, ϕ3 ). We have:
transform under a rotation by ϕ
 
0 0 0 0
0 0 ϕ3 0
 
1 ρσ
 
Λ = exp Ωρσ M = exp  (28)
2 0 −ϕ3 0 0

0 0 0 0

(with Ω12 = −ϕ3 ). Thus, when we rotate a vector by ϕ3 = 2π, we learn that Λ = 1, as one
expects. We therefore conclude that S[Λ] is definitely a different representation from the
familiar vector representation Λµ ν .
 –9–

Example II: Boosts.

0 σi −σ i 0
    
0i1 0 1 1
S = = (29)
2 1 0 −σ i 0 2 0 σi

Writing the boost parameter as Ωi0 = −Ω0i = χi , we have

 +~χ·~σ/2 
e 0
S[Λ] = (30)
0 e−~χ·~σ/2

Representations of the Lorentz group are not unitary.

For the rotations given in Equation 25, S[Λ] is unitary, namely S[Λ]† S[Λ] = 1. However, for
the boosts given in Equation 30, S[Λ] is not unitary (there was no “i” in the exponent).
In fact, there are no finite dimensional unitary representations of the Lorentz group. We
have demonstrated this explicitly for the spinor representation using the chiral representation
(Equation 15) of the Clifford algebra.
We can get a feel for why it is true for a spinor representation constructed from any
representation of the Clifford algebra. Recall that
 
1 ρσ
S[Λ] = exp Ωρσ S
2

and therefore the representation is unitary if S µν are anti-hermitian, i.e., (S µν )† = −S µν .

However, this can never happen: using equation 17 we have
1
(S µν )† = − [(γ µ )† , (γ ν )† ], (31)
4
which can be anti-hermitian if all γ µ are hermitian or all are anti-hermitian. This, though,
can never happen since

(γ 0 )2 = 1 ⇒ Real Eigenvalues
(32)
(γ i )2 = −1 ⇒ Imaginary Eigenvalues

So we could pick γ 0 to be hermitian, but we can only pick γ i to be anti-hermitian. Indeed,

in the chiral representation (Equation 15), the matrices have this property: (γ 0 )† = γ 0 and
(γ i )† = −γ i . This can be traced back to the different signs in time and space in Minkowski
space-time. In general there is no way to pick γ µ such that S µν are anti-hermitian.
 – 10 –

4. Constructing an Action for the Dirac Spinor Field

We now completed our task and found a representation of the Lorentz group. The
representation we found work on a new field: the Dirac spinor field, ψ, introduced in Equa-
tion 21. Our next goal is to construct a Lorentz invariant equation of motion. We do this
by constructing a Lorentz invariant action.
We begin in a naive way which will not work, but will give us a clue how to proceed.
Define
ψ † (x) = (ψ ⋆ )T (x) (33)
which is the usual adjoint of a multi-component object. We could then try to form a Lorentz
scalar by taking the product ψ † ψ, with the spinor indices summed over. Let’s see how this
transforms under Lorentz transformations,

ψ(x) → S[Λ]ψ(Λ−1 x)
(34)
ψ † (x) → ψ † (Λ−1 x)S[Λ]† .

Thus, ψ † (x)ψ(x) → ψ † (Λ−1 x)S[Λ]† S[Λ]ψ(Λ−1 x). However, we have seen that for some
Lorentz transformations S[Λ]† S[Λ] 6= 1, since the representation is not unitary. This means
that ψ † ψ is not good: it doesn’t have any nice transformation under the Lorentz group, and
certainly it is not a Lorentz scalar. Thus, the fact that S[Λ] is not unitary is what failed
our attempt.
But now we see why it fails, we can also see how to solve the problem. Let’s pick a
representation of the Clifford algebra which, like the chiral representation (Equation 15),
satisfies (γ 0 )† = γ 0 and (γ i )† = −γ i . Then, for all µ = 0, 1, 2, 3 we have:

γ 0 γ µ γ 0 = (γ µ )† (35)

which, in turn, gives

1  ν † µ †
(S µν )† = (γ ) , (γ ) = −γ 0 S µν γ 0 , (36)
4
and thus we have  
1
†
S[Λ] = exp Ωρσ (S ) = γ 0 S[Λ]−1 γ 0 .
ρσ †
(37)
2
Thus, the reason why we failed producing a Lorentz scalar out of ψ † (x)ψ(x) is because
S[Λ]† 6= S[Λ]−1 , but S[Λ]† = γ 0 S[Λ]−1 γ 0 .
With this in mind, we can now define the Dirac adjoint (or Dirac Conjugate),

ψ̄(x) ≡ ψ † (x)γ 0 . (38)

 – 11 –

Let us now see what Lorentz covariant objects we can form out of a Dirac spinor ψ and its
adjoint ψ̄.
Claim [3]: ψ̄ψ is a Lorentz scalar.
Proof. under a Lorentz transformation,

ψ̄(x)ψ(x) = ψ † (x)γ 0 ψ(x)

→ ψ † (Λ−1 x)S[Λ]† γ 0 S[Λ]ψ(Λ−1 x)
(39)
= ψ † (Λ−1 x)γ 0 ψ(Λ−1 x)
= ψ̄(Λ−1 x)ψ(Λ−1 x),

which is indeed the transformation law of a Lorentz scalar.

Claim [4]: ψ̄γ µ ψ is a Lorentz vector, namely

ψ̄(x)γ µ ψ(x) → Λµ ν ψ̄(Λ−1 x)γ ν ψ(Λ−1 x). (40)

This equation implies that we can treat the µ = 0, 1, 2, 3 index on the γ µ matrices as a true
vector index. In particular we can form Lorentz scalars by contracting it with other Lorentz
indices.
Proof. Under Lorentz transformation we have

ψ̄γ µ ψ → ψ̄S[Λ]−1 γ µ S[Λ]ψ. (41)

Thus, if ψ̄γ µ ψ is to transform as a vector, we must have

S[Λ]−1 γ µ S[Λ] = Λµ ν γ ν . (42)

Let us show this. Infinitesimally, we have

 
1 ρσ 1
Λ = exp Ωρσ M ≈ 1 + Ωρσ Mρσ + ... (43)
2 2

and  
1 1
S[Λ] = exp Ωρσ S ρσ ≈ 1 + Ωρσ S ρσ + ... (44)
2 2
and thus the requirement in Equation 42 becomes

−[S ρσ , γ µ ] = (Mρσ )µ ν γ ν (45)

(omitting the α, β indices on γ µ and S µν , but otherwise leaving all other indices explicit).
 – 12 –

Equation 45 follows claim [1], [S ρσ , γ µ ] = γ ρ η σµ − γ σ η µρ . This can be seen by first

expanding M in the right hand side of Equation 45 using Equation 7:

(Mρσ )µ ν γ ν = (η ρµ δνσ − η σµ δνρ )γ ν

(46)
= η ρµ γ σ − η σµ γ ρ .

But in claim [1] we showed that the right hand side of Equation 46 is just equal to −[S ρσ , γ µ ],
which proves Equation 45, and completes the proof that ψ̄γ µ ψ transforms as a vector.
Claim [5]. ψ̄γ µ γ ν ψ transforms as a Lorentz tensor. More precisely, the symmetric part
is a Lorentz scalar, proportional to η µν ψ̄ψ, while the antisymmetric part is a Lorentz tensor,
proportional to ψ̄S µν ψ.
The proof of this claim follows the same steps as above.
Following the above discussion, we have now the three bilinears of the Dirac field, ψ̄ψ,
ψ̄γ ψ and ψ̄γ µ γ ν ψ, each of which transforms covariantly under the Lorentz group. We can
µ

now build a Lorentz invariant action out of these; in fact, we need only the first two. We
choose Z
S = d4 xψ̄(x)(iγ µ ∂µ − m)ψ(x). (47)

This is Dirac action. (ψ̄ψ is a Lorentz scalar; ψ̄γ µ ψ is a Lorentz vector, so contracting
it with ∂µ produces another scalar). The factor i is added to make the action real; upon
complex conjugation, it cancels a minus sign that comes from integration by parts.
As we will show shortly, after quantization this theory describes particles and anti-
particles of mass |m| and spin 1/2. Notice that the Lagrangian is first order, rather than the
second order Lagrangians we were working with for scalar fields. Also, the mass appears in
the Lagrangian as m, which can be positive or negative.

5. The Dirac Equation

The Equation of motion follows the action in Equation 47 by varying with respect to ψ
and ψ̄ independently. Varying with respect to ψ̄ gives

(iγ µ ∂µ − m)ψ = 0. (48)

This is the Dirac Equation.

Varying with respect to ψ gives the conjugate equation

i∂µ ψ̄γ µ − mψ̄ = 0 (49)

 – 13 –

The Dirac equation is first order in derivatives, yet (miraculously) Lorentz invariant.
Note that if we tried to write down a first order equation of motion for a scalar field,
it would have to look like v µ ∂µ φ = ..., which necessarily includes a privileged vector in
spacetime v µ and is not Lorentz invariant. However, for spinor fields, the magic of the γ µ
matrices implies that the Dirac Lagrangian is Lorentz invariant.
The Dirac Equation mixes up different components of ψ through the matrices γ µ . In-
terestingly, each individual component itself solves the Klein-Gordon Equation. To see this,
we write
(iγ ν ∂ν + m)(iγ µ ∂µ − m)ψ = −(γ µ γ ν ∂µ ∂ν + m2 )ψ = 0, (50)
and use the fact that γ µ γ ν ∂µ ∂ν = 12 {γ µ , γ ν }∂µ ∂ν = ∂µ ∂ µ , to get

−(∂µ ∂ µ + m2 )ψ = 0. (51)

There are no γ µ matrices in the last equation, and so it applies to each component ψ α with
α = 1, 2, 3, 4.
The Slash Notation
We often encounter 4-vectors contracted with γ µ matrices. Following Feynman, we
define:
/ ≡ γ µ Aµ
A (52)
Using this notation, the Dirac Equation can be written as

(i∂/ − m)ψ = 0. (53)

6. Chiral Spinors

So far, when we needed an explicit form of the γ µ matrices, we have used the chiral
representation given in Equation 15:

0 σi
   
0 0 1 i
γ = , γ =
1 0 −σ i 0

In this representation, the spinor rotation transformation S[Λrot ] and boost transformation
S[Λboost ] were computed in Equations 25 and 30, respectively. Both are block diagonal:

e+i~ϕ·~σ/2 e+~χ·~σ/2
   
0 0
S[Λrot ] = and S[Λboost ] = (54)
0 e σ/2
+i~
ϕ·~
0 e σ/2
χ·~
−~
 – 14 –

This means that the Dirac spinor representation of the Lorentz group is reducible: it de-
composes into two irreducible representations, acting separately on two-component spinors
u± .4 In the chiral representation, u± are defined by
 
u+
ψ= (55)
u−
The two-component objects u± are called Weyl spinors, or chiral spinors. (In some
textbooks, you find a different notation, u+ (u− ) written as ψL (ψR )). They transform in
the same way under rotations:
u± → ei~ϕ·~σ/2 u± , (56)
but oppositely under boosts:
u± → e±~χ·~σ/2 u± (57)

In group theory language, u+ is in the ( 21 , 0) representation of the Lorentz group, while

u− is in the (0, 12 ) representation. The Dirac spinor ψ lies in the ( 12 , 0) ⊕ (0, 12 ) representation.

6.1. The Weyl Equation

Let us see how the decomposition in Equation 55 into Weyl spinors affects the Dirac
Lagrangian. Using Equations 55, 38 and 15 we can write
∂0 + σ i ∂i
     
† † 0 1 0 u+
L = ψ̄(i∂/ − m)ψ = (u+ , u− ) i i −m
1 0 ∂0 − σ ∂i 0 u− (58)
† µ † µ † †
= iu− σ ∂µ u− + iu+ σ̄ ∂µ u+ − m(u+ u− + u− u+ ) = 0
where we have used the notation

σ µ = (1, σ i ) and σ̄ µ = (1, −σ i ) (59)

to describe the Pauli matrices with µ = 0, 1, 2, 3.

From Equation 58, we find that for a massive fermion, both u+ and u− are required,
since they couple through the mass term. However, for a massless fermion the coupling
vanishes, and only u+ (or u− ) is required for the equation of motion, which becomes
iσ̄ µ ∂µ u+ = 0;
(60)
or iσ µ ∂µ u− = 0.

4
Note that the γ µ matrices themselves were irreducible representations of the Clifford algebra; however,
we constructed from them a reducible representation of the Lorentz group.
 – 15 –

These are known as Weyl equations. Note that in the standard model of particle physics,
all fermions are massless, and therefore they are described by the Weyl spinors; Parti-
cles obtain their mass by the Higgs mechanism, which, unfortunately, is learned only in a
more advanced course on the standard model. We also allow ourself to call these particles
“Fermions”, although we haven’t shown it yet; though, when we will quantize the Dirac
spinor in the next chapter, we will see that it gives rise to spin-1/2 particles.
Degrees of Freedom
Let us look at the degrees of freedom in a spinor. First note that the Dirac fermion (=Dirac
spinor field) has 4 complex components = 8 real components (see Equation 21).
How do we count degrees of freedom? In classical mechanics, the number of degrees of
freedom of a system is equal to the dimension of the configuration space or, equivalently, half
the dimension of the phase space. In field theory we have an infinite number of degrees of
freedom, but it makes sense to count the number of degrees of freedom per spatial point: this
should at least be finite. For example, in this sense a real scalar field φ has a single degree
of freedom. At the quantum level, this translates to the fact that it gives rise to a single
type of particle. A classical complex scalar field has two degrees of freedom, corresponding
to the particle and the anti-particle in the quantum theory.
What about the Dirac spinor ? One might think that there are 8 degrees of freedom.
But this isn’t right. Crucially, and in contrast to the scalar field, the equation of motion is
first order rather than second order. In particular, for the Dirac Lagrangian, the momentum
conjugate to the spinor ψ is given by

πψ = ∂L/∂ ψ̇ = iψ † . (61)

It is not proportional to the time derivative of ψ. This means that the phase space for a
spinor is therefore parameterized by ψ and ψ † , while for a scalar it is parameterized by φ and
π = φ̇. So the phase space of the Dirac spinor ψ has 8 real dimensions and correspondingly
the number of real degrees of freedom is 4. We will see in the next chapter that, in the
quantum theory, this counting manifests itself as two degrees of freedom (spin up and down)
for the particle, and a further two for the anti-particle.
A similar counting for the Weyl fermion tells us that it has two degrees of freedom.

6.2. The γ 5 matrix

We have seen that the Lorentz group matrices S[Λ] came out to be block diagonal in
Equation 54. This is due to our choice of the specific representation in Equation 15. In fact,
 – 16 –

this is why the representation in Equation 15 is called the chiral representation: it’s because
the decomposition of the Dirac spinor ψ is simply given by Equation 55.
But we could choose a different representation γ µ of the Clifford algebra, so that

γ µ → U γ µ U −1 and ψ → U ψ. (62)

In this representation, S[Λ] will not be block diagonal. However, we can still define the chiral
spinors in an invariant way, by introducing the “fifth” gamma matrix,

γ 5 ≡ −iγ 0 γ 1 γ 2 γ 3 . (63)

(Note that by adding the (-) sign, I follow David Tong’s convention, rather than Peskin and
Schroeder). You can check that this matrix satisfies

{γ 5 , γ µ } = 0 and (γ 5 )2 = +1. (64)

(The reason why it is called γ 5 is because the set of matrices γ̃ A = (γ µ , iγ 5 ) with A =

0, 1, 2, 3, 4 satisfy the five-dimensional Clifford algebra, {γ̃ A , γ̃ B } = 2η AB ).
You can also check that [Sµν , γ 5 ] = 0, which means that γ 5 is a scalar under rotations
and boosts. Since (γ 5 )2 = 1, this means that we may form the Lorentz invariant projection
operators
1
P± = (1 ± γ 5 ), (65)
2
such that P+2 = P+ , P−2 = P− and P+ P− = 0. Further, one can check by direct substitution
in Equation 63 that for the chiral representation (Equation 15),
 
5 1 0
γ = , (66)
0 −1

from which we see that the operators P± project onto the Weyl spinors, u± . However, for an
arbitrary representation of the Clifford algebra, we may use γ 5 to define the chiral spinors,

ψ± = P± ψ, (67)

which form the irreducible representations of the Lorentz group. ψ+ is often called a “right-
handed” spinor, while ψ− is “left-handed”. These are thus the analogues of the Weyl spinors,
but in arbitrary basis. In particular, in the chiral representation, one finds
   
u+ 0
ψ+ = and ψ− = . (68)
0 u−
 – 17 –

6.3. Discrete Symmetries of the Dirac Theory: Parity

The spinors ψ± are related to each other by parity. Let us first define this concept.
The Lorentz group (=the group of all Lorentz transformations) is defined by xµ →
Λµ ν xν , such that
Λµ ν Λρ σ η νσ = η µρ . (69)
So far we have only considered continuous transformations Λ, which are continuously con-
nected to the identity. Only these transformations have an infinitesimal form. However there
are also two discrete symmetries which are part of the Lorentz group:
Time reversal T : x0 → −x0 ; xi → xi
(70)
Parity P : x0 → x0 ; xi → −xi
We are not going to discuss much time reversal in this course (see Peskin & Schroeder for
further discussion). However, we do want to discuss parity, as it plays an important role in
the standard model (and in particular in the theory of weak interactions). As we saw, parity
sends (t, ~x) to (t, −~x), reversing the handedness of space.
Under parity, the left and right-handed spinors are exchanged. This follows from the
transformation of the spinors under the Lorentz group. In the chiral representation, we
saw that the rotation (Equation 56) and boost (Equation 57) transformations for the Weyl
spinors u± are
rot boost
u± → ei~ϕ·~σ/2 u± and u± → e±~χ·~σ/2 u± (71)

Under parity, both the spatial coordinates and the momentum change sign, and hence
rotations don’t change sign. But boosts do flip sign. This confirms that parity exchanges
right-handed and left-handed spinors, P : u± → u∓ , or in the notation ψ± = 12 (1 ± γ 5 )ψ, we
have
P : ψ± (t, ~x) → ψ∓ (t, −~x). (72)

Using this knowledge of how chiral spinors transform, and the fact that P 2 = 1, we see
that the action of parity on the Dirac spinor itself can be written as
P : ψ(~x, t) → γ 0 ψ(−~x, t). (73)

Also, note that if ψ(~x, t) satisfies the Dirac equation, then the parity transformed spinor
0
γ ψ(−~x, t) also satisfies the Dirac equation:
(iγ 0 ∂t + iγ i ∂i − m)γ 0 ψ(−~x, t) = γ 0 (iγ 0 ∂t − iγ i ∂i − m)ψ(−~x, t) = 0 (74)
where the extra minus sign from passing γ 0 through γ i is compensated by the derivative
acting on −~x instead of +~x.
 – 18 –

6.4. Chiral Interactions and Dirac Field Bilinears

We would like to ask now the following question. Consider the general expression ψ̄Γψ,
where Γ is any 4 × 4 constant matrix. Can we decompose this expression into terms that
have definite transformation properties under the Lorentz group? Once we know how to do
that, we could build Lorentz invariant Lagrangians from these terms.
In order to carry this calculation, let us look at how the different possible interaction
terms change under parity. We can look at each of the spinor bilinears from which we built
the action,
P : ψ̄ψ(~x, t) → ψ̄ψ(−~x, t). (75)
This is the transformation under parity we expect from a scalar. Furthermore, we saw in
Equation 39 that ψ̄ψ transforms as a scalar under continuous Lorentz transformation.
For the vector ψ̄γ µ ψ, we can look at the temporal and spatial components separately,
P : ψ̄γ 0 ψ(~x, t) → ψ̄γ 0 ψ(−~x, t);
(76)
P : ψ̄γ i ψ(~x, t) → ψ̄γ 0 γ i γ 0 ψ(−~x, t) = −ψ̄γ i ψ(−~x, t).
This implies that ψ̄γ µ ψ transforms as a vector, with the spatial part changing sign. Again,
this is the transformation law under parity we expect from a vector. You can also check that
ψ̄S µν ψ transforms as a suitable tensor.
However, we can use γ 5 to form another Lorentz scalar and Lorentz vector:
ψ̄γ 5 ψ and ψ̄γ 5 γ µ ψ. (77)
These have the appropriate transformation law as Lorentz scalar and Lorentz vector under
continuous Lorentz transformation. However, under parity, these object transform like:
P : ψ̄γ 5 ψ(~x, t) → ψ̄γ 0 γ 5 γ 0 ψ(−~x, t) = −ψ̄γ 5 ψ(−~x, t)
−ψ̄γ 5 γ 0 ψ(−~x, t) µ = 0

5 µ 0 5 µ 0 (78)
P : ψ̄γ γ ψ(~x, t) → ψ̄γ γ γ γ ψ(−~x, t) =
+ψ̄γ 5 γ i ψ(−~x, t) µ = i.
Thus, both these objects obtain an extra minus (-) sign under parity transformation. We
say that ψ̄γ 5 ψ transforms as pseudo-scalar, while ψ̄γ 5 γ µ ψ transforms as an axial vector
(also known as pseudo-vector).
To summarize, we have the following spinor bilinears:
ψ̄ψ : scalar
ψ̄γ µ ψ : vector
ψ̄S µν ψ : tensor (79)
ψ̄γ 5 ψ : pseudoscalar
ψ̄γ 5 γ µ ψ : axial vector
 – 19 –

The total number of bilinears is 1 + 4 + (4 × 3/2) + 4 + 1 = 16, which is all we could

hope for from a 4-component object (Γ is a 4 × 4 matrix).
Armed now with new terms involving γ 5 , we can start adding terms to our Lagrangian
to construct new theories. Typically such terms will break parity invariance of the theory,
although this is not always true. (For example, the term φψ̄γ 5 ψ doesn’t break parity if ψ is
itself a pseudoscalar).
Nature makes use of these parity violating interactions by using γ 5 in the weak force.
A theory which treats ψ± on an equal footing is called a vector-like theory. A theory in
which ψ+ and ψ− appear differently is called a chiral theory.

7. Majorana Fermions

Our spinor ψ α is a complex object. It has to be because the representation S[Λ] is

typically also complex. This means that if we were to try to make ψ real, for example by
imposing ψ = ψ ⋆ , then it wouldn’t stay that way once we make a Lorentz transformation.
However, there is a way to impose a reality condition on the Dirac spinor ψ. To motivate
this possibility, let us look at a novel basis for the Clifford algebra, known as the Majorana
basis.
0 σ2
 3
0 −σ 2 −iσ 1
      
0 1 iσ 0 2 3 0
γ = , γ = , γ = , γ = . (80)
σ2 0 0 iσ 3 σ2 0 0 −iσ 1
These matrices satisfy the Clifford algebra.
All these matrices have a special property: they are all pure imaginary, (γ µ )⋆ = −γ µ .
This means that the generators of the Lorentz group S µν = 14 [γ µ , γ ν ], and hence the matrices
S[Λ] are all real. So with this basis of the Clifford algebra, we can work with a real spinor
simply by imposing the condition,
ψ = ψ⋆ (81)
which is preserved under Lorentz transformation. Such spinors are called Majorana spinors.
Let us now look at a general basis for the Clifford algebra. We only ask that the basis
satisfies (γ 0 )† = γ 0 , and (γ i )† = −γ i . We define the charge conjugate of a Dirac spinor ψ
as
ψ (c) = Cψ ⋆ . (82)
Here, C is a 4 × 4 matrix satisfying

C † C = 1 and C † γ µ C = −(γ µ )⋆ . (83)

 – 20 –

Let us see first that Equation 82 is a good definition, in the sense that ψ (c) transforms nicely
under a Lorentz transformation. We have:
ψ (c) → CS[Λ]⋆ ψ ⋆ = S[Λ]Cψ ⋆ = S[Λ]ψ (c) , (84)
where we used equation 83 in taking the matrix C through S[Λ]⋆ .
In fact, not only does ψ (c) transform nicely under the Lorentz group, but if ψ satisfies
the Dirac equation, then ψ (c) does too. This follows from
⋆
(i∂/ − m)ψ = 0 ⇒ (−i∂/ − m)ψ ⋆ = 0
⋆
⇒ C(−i∂/ − m)ψ ⋆ = (+i∂/ − m)ψ (c) = 0.

Finally, we can now impose the Lorentz invariant reality condition on the Dirac spinor,
to yield a Majorana spinor,
ψ (c) = ψ. (85)

After quantization, the Majorana spinor gives rise to a fermion that is its own anti-
particle. This is exactly the same as in the case of scalar fields, where we have seen that a
real scalar field gives rise to a spin 0 boson that is its own anti-particle. (Note that in many
texts an extra factor of γ 0 is absorbed into the definition of C).
So what is this matrix C? For a given representation of the Clifford algebra, it is
something that we can find fairly easily. In the Majorana basis, where the gamma matrices
are pure imaginary, we have simply CMaj = 1 and the Majorana condition ψ = ψ (c) becomes
ψ = ψ⋆.
In the chiral basis (Equation 15) , only γ 2 is imaginary, and we may take Cchiral = iγ 2 =
iσ 2
 
0
. (Note that the matrix iσ 2 that appears here is simply the anti-symmetric
−iσ 2 0
matrix ǫαβ ).
It is interesting to see how the Majorana condition (Equation 85) looks in terms of
the decomposition into left and right handed Weyl spinors (Equation 55). Plugging in the
various definitions, we find that u+ = iσ 2 u⋆− and u− = −iσ 2 u⋆+ . In other words, a Majorana
spinor can be written in terms of Weyl spinors as
 
u+
ψ= . (86)
−iσ 2 u⋆+

Note that it is not possible to impose the Majorana condition ψ = ψ (c) at the same time
as the Weyl condition (u− = 0 or u+ = 0). Instead the Majorana condition relates u− and
u+ .
 – 21 –

An Aside: Spinors in Different Dimensions. The ability to impose Majorana

or Weyl conditions on Dirac spinors depends on both the dimension and the signature of
spacetime. One can always impose the Weyl condition on a spinor in even dimensional
Minkowski space, basically because you can always build a suitable “γ 5 ” projection matrix by
multiplying together all the other γ-matrices. The pattern for when the Majorana condition
can be imposed is a little more sporadic. Interestingly, although the Majorana condition
and Weyl condition cannot be imposed simultaneously in four dimensions, you can do this
in Minkowski spacetimes of dimension 2, 10, 18, ...

8. Symmetries and Conserved Currents

The Dirac Lagrangian enjoys a number of symmetries. Here we list them and compute
the associated conserved currents.
Spacetime Translations
Under spacetime translations the spinor transforms as

δψ = ǫµ ∂µ ψ (87)

The Lagrangian depends on ∂µ ψ, but not ∂µ ψ̄, so the standard formula (“Classical
Fields”, Eq. 41: T µ ν = (∂L/∂(∂µ φ))∂ν φ − δνµ L) gives us the energy-momentum tensor

T µν = iψ̄γ µ ∂ ν ψ − η µν L. (88)

Since a current is conserved only when the equations of motion are obeyed, we can
impose the equations of motion already on T µν . In the case of a scalar field this didn’t really
buy us anything because the equations of motion are second order in derivatives, while the
energy-momentum is typically first order. However, for a spinor field the equations of motion
are first order: (i∂/ − m)ψ = 0. This means we can set L = 0 in T µν , leaving

T µν = iψ̄γ µ ∂ ν ψ. (89)

In particular, we have the total energy:

Z Z Z
E = d xT = d xiψ̄γ ψ̇ = d3 xψ † γ 0 (−iγ i ∂i + m)ψ,
3 00 3 0
(90)

where, in the last equality, we have again used the equations of motion.
Lorentz Transformations
Under an infinitesimal Lorentz transformation, the Dirac spinor transforms as in Equation 21
 – 22 –

which, in infinitesimal form, reads

1
δψ α = −ω µ ν xν ∂µ ψ α + Ωρσ (S ρσ )α β ψ β . (91)
2

Using Equation 9, we can write ω µ ν = 21 Ωρσ (Mρσ )µ ν , where the generators of the Lorentz
algebra Mρσ are given in Equation 7,

(Mρσ )µ ν = η ρµ δνσ − η σµ δνρ .

Direct substitution thus yields ω µν = Ωµν . We thus get

 
α µν α 1 α β
δψ = −ω xν ∂µ ψ − (Sµν ) β ψ (92)
2

The conserved current arising from Lorentz transformations now follows from the same
calculation we saw for the scalar field (“Classical Fields”, Equation 54: j µ = −ω ρ ν T µ ρ xν )
with two differences: firstly, as we saw above, the spinor equations of motion set L = 0;
secondly, we pick up an extra piece in the current from the second term in Equation 92. We
have
(J µ )ρσ = xσ T µρ − xρ T µσ − iψ̄γ µ S ρσ ψ (93)

After quantization, when (J µ )ρσ is turned into an operator, this extra term will be
responsible for providing the single particle states with internal angular momentum, telling
us that the quantization of a Dirac spinor gives rise to a particle carrying spin 1/2.
Internal Vector Symmetry
The Dirac Lagrangian is invariant under rotating the phase of the spinor, ψ → e−iα ψ. This
gives rise to the current
jVµ = ψ̄γ µ ψ (94)
(where “V ” stands for “vector”, reflecting the fact that the left and right-handed components
ψ± transform in the same way under this symmetry).
We can easily check that jVµ is conserved under the equations of motion,

∂µ jVµ = (∂µ ψ̄)γ µ ψ + ψ̄γ µ (∂µ ψ) = imψ̄ψ − imψ̄ψ = 0. (95)

Where, in the last equality, we have used the equations of motion i∂/ψ = mψ and i∂µ ψ̄γ µ =
−mψ̄.
The conserved quantity arising from this symmetry is
Z Z
Q = d xψ̄γ ψ = d3 xψ † ψ
3 0
(96)
 – 23 –

We will see shortly that this has the interpretation of electric charge, or particle number, for
fermions.
Axial Symmetry
When m = 0, the Dirac Lagrangian admits an extra internal symmetry which rotates left
and right-handed fermions in opposite directions,
5 5
ψ → eiαγ ψ and ψ̄ → ψ̄eiαγ . (97)
5 5
(The second transformation follows the first after noting that e−iαγ γ 0 = γ 0 e+iαγ ). This
gives the conserved current,
jAµ = ψ̄γ µ γ 5 ψ (98)
where A stands for “Axial”, since jAµ is an axial vector. This is conserved only when m = 0.
Indeed, with the full Dirac Lagrangian we may compute

∂µ jAµ = (∂µ ψ̄)γ µ γ 5 ψ + ψ̄γ µ γ 5 ∂µ ψ = 2imψ̄γ 5 ψ, (99)

which vanishes only for m = 0.

However, in the quantum theory things become more interesting for the axial current.
When the theory is coupled to gauge fields (in a manner we will discuss later on), the axial
transformation remains a symmetry of the classical Lagrangian. But it doesn’t survive the
quantization process. It is the archetypal example of an anomaly: a symmetry of the classical
theory that is not preserved in the quantum theory.

9. Plane Wave Solutions to Dirac Equation

Let us now study solutions to the Dirac equation (48),

(iγ µ ∂µ − m)ψ = 0.

We start by making a simple ansatz:

ψ = u(~p)e−ip·x , (100)

where u(~p) is a four-component spinor, independent of spacetime x. As the notation

suggests, it can depend on the 3-momentum p~. The Dirac equation then becomes

−m pµ σ µ
 
µ
(γ pµ − m)u(~p) = u(~p) = 0 (101)
pµ σ̄ µ −m
 – 24 –

where again we use the definition,

σ µ = (1, σ i ) and σ̄ µ = (1, −σ i ). (102)

Claim. The solution to Dirac Equation 101 (in the chiral representation) is
√ 
p · σξ
u(~p) = √ , (103)
p · σ̄ξ
for any 2-component spinor ξ which we will normalize to ξ † ξ = 1.
Proof. Let us write u(~p)T = (u1 , u2 ). Then Equation 101 reads
(p · σ)u2 = mu1 and (p · σ̄)u1 = mu2 (104)
Either one of these equations implies the other, a fact which follows from the identity (p ·
σ)(p · σ̄) = p20 − pi pj σ i σ j = p20 − pi pj δ ij = pµ pµ = m2 .
Thus, let’s try the ansatz u1 = (p · σ)ξ ′ for some spinor ξ ′ . The second Equation in 104
implies u2 = mξ ′ . So we learn that any spinor of the form
(p · σ)ξ ′
 
u(~p) = A (105)
mξ ′
with some constant A is a solution to Equation 101. To make this more symmetric, we choose
√ √ √
A = 1/m and ξ ′ = p · σ̄ξ, with constant ξ. Then we obtain u1 = (p · σ) p · σ̄ξ = m p · σξ,
which leads to result in Equation 103.
Negative Frequency Solutions
We can get further solutions to the Dirac Equation using the ansatz
ψ = v(~p)e+ip·x . (106)

Solutions of the form given by Equation 100, which oscillate in time as ψ ∼ e−iEt are
called positive frequency solutions. Those of the form given by equation 106, which oscillates
as ψ ∼ e+iEt , are negative frequency solutions. Note though, that both are solutions to
the classical field equations, and both have positive energy (see Equation 90). The Dirac
equation requires that the 4-component spinor v(~p) satisfies
m pµ σ µ
 
µ
(γ pµ + m)v(~p) = v(~p) = 0, (107)
pµ σ̄ µ m
which is solved by √ 
p · ση
v(~p) = √ , (108)
− p · σ̄η
for some 2-component spinor η, which we take to be constant and normalized to η † η = 1.
 – 25 –

9.1. Examples

Consider the positive frequency solution to Dirac’s equation with mass m and 3-momentum
p~ = 0,
√
 
ξ
u(~p) = m , (109)
ξ
where ξ is any 2-component spinor. Spatial rotations of the field act on ξ by Equation 25,
ξ → e+i~ϕ·~σ/2 ξ. (110)
The 2-component spinor ξ defines the spin of the field.
This should be familiar from quantum mechanics. A field with spin up (down) along
a given direction is described by the eigenvector of the corresponding Pauli matrix with
eigenvalue +1 (-1 respectively). For example, ξ T = (1, 0) describes a field with spin up along
the z-axis. After quantization, this will become the spin of the associated particle. In the rest
of this section, we will indulge in an abuse of terminology and refer to the classical solutions
to the Dirac equations as “particles”, even though they have no such interpretation before
quantization.
Consider now boosting the particle with spin ξ T = (1, 0) along the x3 direction, with
pµ = (E, 0, 0, p). The solution to the Dirac equation becomes
     
√ 1 p 1
 p·σ   E − p3
0   0 

u(~p) =  = (111)
    
√ 1  p 1 

p · σ̄ E+p 3
0 0
In fact, this expression also makes sense for a massless field, for which E = p3 . (We picked
the normalization in Equation 103 for the solutions so that this would be the case).
For a massless particle we have
 
0
√  0

u(~p) = 2E   . (112)
1
0

Similarly, for a boosted solution of the spin down ξ T = (0, 1) field, we have
       
√ 0 p 0 0
 p·σ   E + p3 √
1 1 1
  
m→0
u(~p) =   = p  → 2E   (113)
     
√ 0   0  0
p · σ̄ E − p3
1 1 0
 – 26 –

Helicity
The helicity operator is defined to be the projection of the angular momentum along the
direction of momentum:  i 
i i jk 1 σ 0
h = ǫijk p̂ S = p̂i (114)
2 2 0 σi
where S ij is the rotation generator given in Equation 24. The massless field with spin
ξ T = (1, 0) in Equation 112 has helicity h = 1/2; we say that it is right-handed. Similarly,
the field in Equation 113 has helicity h = −1/2, and is called left-handed.

9.2. Some Useful Formulae: Inner and Outer Products

There are a number of identities that will be very useful in the following, regarding the
inner (and outer) products of the spinors u(~p) and v(~p).
In deriving these identities, it is convenient to introduce a basis ξ s and η s , with s = 1, 2
for the two-component spinors such that

ξ r† ξ s = δ rs and η r† η s = δ rs . (115)

For example, we can take    

1 1 0
2
ξ = and ξ = , (116)
0 1
and similarly for η s .
Let us look first at the positive frequency plane waves. We write the two independent
solutions as √
p · σξ s

s
u (~p) = √ (117)
p · σ̄ξ s

We can take the inner product of four-component spinors in two different ways: either
†
as u · u, or as ū · u. Of course, only the latter will be Lorentz invariant, but it turns out
that the former is needed when we come to quantize the theory. Here we state both:

√p · σξ s
 
r† s r †√ r †√
u (~p) · u (~p) = ξ p · σ, ξ p · σ̄ √
p · σ̄ξ s
= ξ r † p · σξ s + ξ r † p · σ̄ξ s (118)
r† s
= 2ξ p0 ξ
= 2p0 δ rs
 – 27 –

while the Lorentz invariant inner product is

 √
r† √ p · σξ s
 
r† √
r s

0 1
ū (~p) · u (~p) = ξ p · σ, ξ p · σ̄ √ = 2mδ rs (119)
1 0 p · σ̄ξ s

We have analogous results for the negative frequency solutions, which we may write as
√
p · ση s

s
v (~p) = √ , (120)
− p · σ̄η s

with v r† (~p) · v s (~p) = 2p0 δ rs , and similarly v̄ r (~p) · v s (~p) = −2mδ rs .

We can also compute the inner product between u and v. We have

0 √p · ση s
 
r s r †√ r †√
ū (~p) · v (~p) = ξ p · σ, ξ p · σ̄ γ √
p p − p · σ̄η s (121)
= ξ r † (p · σ̄)(p · σ)η s − ξ r † (p · σ̄)(p · σ)η s = 0
and similarly, v̄ r (~p) · us (~p) = 0.
However, when we come to u† · v, it is a slightly different combination that has nice
properties (and this same combination appears when we quantize the theory). We look at
ur† (~p) · v s (−~p), with the 3-momentum in the spinor v taking the opposite sign. Defining the
4-momentum (p′ )µ = (p0 , −~p), we have
√ ′ s

† √ † √ p · ση
ur† (~p) · v s (−~p) = ξ r p · σ, ξ r p · σ̄


√
p − pp′ · σ̄η s (122)
r† ′ s r† ′ s
= ξ (p · σ)(p · σ)η − ξ (p · σ̄)(p · σ̄)η

Now the terms under the square-root are given by (p · σ)(p′ · σ) = (p0 + pi σ i )(p0 − pi σ i ) =
p20 − p~2 = m2 . The same expression holds for (p · σ̄)(p′ · σ̄), and thus the two terms cancel.
We thus find that
ur† (~p) · v s (−~p) = v r† (~p) · us (−~p) = 0 (123)

Outer products
The last spinor identity that we need before going into quantum theory is
2
X
us (~p)ūs (~p) = p/ + m, (124)
s=1

where the two spinors are not contracted, but instead placed back to back to give a 4 × 4
matrix. Also,
X2
v s (~p)v̄ s (~p) = p/ − m. (125)
s=1
 – 28 –

Proof.
2 2 √
p · σξ s √

X
s s
X √
ξ s† p · σ̄, ξ s† p · σ .


u (~p)ū (~p) = √ s (126)
s=1 s=1
p · σ̄ξ

We can now use s ξ s ξ s † = 1 (a 2 × 2 unit matrix), and get

P

2  
X
s s m p·σ
u (~p)ū (~p) = (127)
s=1
p · σ̄ m

which is the desired result.

P2
A similar proof holds for s=1 v s (~p)v̄ s (~p).

REFERENCES

[1] D. Tong, Lectures on Quantum Field Theory, part 4. The Dirac Equation
(http://www.damtp.cam.ac.uk/user/dt281/qft.html)
[2] M. Peskin and D. Schroeder, An Introduction to Quantum Field Theory (Westview Press),
chapter 3.
[3] G. Arfken and H. Weber, Mathematical Methods for Physicists (Academic Press), chap-
ter 4.