Dirac Matrices and Lorentz Spinors

1
Background: In 3D, the spinor j = 2 representation of the Spin(3) rotation group is
constructed from the Pauli matrices σx, σ y , and σ z , which obey both commutation and
anticommutation relations

[σ i , σ j ] = 2iijk σ k and {σ i , σ j } = 2δ ij × 12×2 . (1)

Consequently, the spin matrices

S = − 4i σ × σ = 1
2σ (2)

commute with each other like angular momenta, [S i , S j ] = iijk S k , so they represent the
generators of the rotation group. Moreover, under finite rotations R(φ, n) represented by


M (R) = exp −iφn · S , (3)

the spin matrices transform into each other as components of a 3–vector,

M −1 (R)S i M (R) = Rij S j . (4)

In this note, I shall generalize this construction to the Dirac spinor representation of the
Lorentz symmetry Spin(3, 1).

Dirac Matrices γ µ generalize the anti-commutation properties of the Pauli matrices σ i to

3 + 1 Minkowski dimensions:

γ µ γ ν + γ ν γ µ = 2g µν × 14×4 . (5)

The γ µ are 4 × 4 matrices, but there are several different conventions for their specific form
In my class I shall follow the same convention as the Peskin & Schroeder textbook, namely
the Weyl convention where in 2 × 2 block notations
! !
0 12×2 0 +~σ
γ0 = , ~γ = . (6)
12×2 0 −~σ 0

Note that the γ 0 matrix is hermitian while the γ 1 , γ 2 , and γ 3 matrices are anti-hermitian.
Apart from that, the specific forms of the matrices are not important, the Physics follows
from the anti-commutation relations (5).

1
Lorentz spin matrices generalize S = − 4i σ × σ rather than S = 12 σ. In 4D, the vector
product becomes the antisymmetric tensor product, so we define

def
S µν = −S νµ = i µ ν
4 [γ , γ ]. (7)

Thanks to the anti-commutation relations (5) for the γ µ matrices, the S µν obey the commu-
tation relations of the Lorentz generators Jˆµν = −Jˆνµ . Moreover, the commutation relations
of the spin matrices S µν with the Dirac matrices γ µ are similar to the commutation relations
of the Jˆµν with a Lorentz vector such as P̂ µ .

Lemma:

[γ λ , S µν ] = ig λµ γ ν − ig λν γ µ . (8)

Proof: Combining the definition (7) of the spin matrices as commutators with the anti-
commutation relations (5), we have

γ µγ ν = 1 µ ν
2 {γ , γ } + 1 µ ν
2 [γ , γ ] = g µν × 14×4 − 2iS µν . (9)

Since the unit matrix commutes with everything, we have

[X, S µν ] = i µ ν
2 [X, γ γ ] for any matrix X, (10)

and the commutator on the RHS may often be obtained from the Leibniz rules for the
commutators or anticommutators:

[A, BC] = [A, B]C + B[A, C] = {A, B}C − B{A, C},

(11)
{A, BC} = [A, B]C + B{A, C} = {A, B}C − B[A, C].

In particular,

[γ λ , γ µ γ ν ] = {γ λ , γ µ }γ ν − γ µ {γ λ , γ ν } = 2g λµ γ ν − 2g λν γ µ (12)

and hence

[γ λ , S µν ] = i λ µ ν
2 [γ , γ γ ] = ig λµ γ ν − ig λν γ µ . (13)

Quod erat demonstrandum.

2
Theorem: The S µν matrices commute with each other like Lorentz generators,
 κλ µν 
S ,S = ig λµ S κν − ig κν S µλ − ig λν S κµ + ig κµ S νλ . (14)

Proof: Again, we use the Leibniz rule and eq. (9):

 κ λ µν 
= γ κ γ λ , S µν + γ κ , S µν γ λ
   
γ γ ,S
= γ κ ig λµ γ ν − ig λν γ µ + ig κµ γ ν − ig κν γ µ γ λ

= ig λµ γ κ γ ν − ig κν γ µ γ λ − ig λν γ κ γ µ + ig κµ γ ν γ λ
(15)
= ig λµ g κν − 2iS κν − ig κν g λµ + 2iS λµ

− ig λν g κµ − 2iS κµ + ig κµ g λν + 2iS λν

= 2g λµ S κν − 2g κν S λµ − 2g λν S κµ + 2g κµ S λν ,

and hence
 κλ µν   κ λ µν 
S ,S = i
2 γ γ ,S = ig λµ S κν − ig κν S µλ − ig λν S κµ + ig κµ S νλ . (16)

Quod erat demonstrandum.

In light of this theorem, the S µν matrices represent the Lorentz generators Jˆµν in a
4-component spinor multiplet.

Finite Lorentz transforms:

Any continuous Lorentz transform — a rotation, or a boost, or a product of a boost and a
rotation — obtains from exponentiating an infinitesimal symmetry

X 0µ = X µ + µν Xν (17)

where the infinitesimal µν matrix is antisymmetric when both indices are raised (or both
lowered), µν = −νµ . Thus, the Lµν matrix of any continuous Lorentz transform is a matrix
exponential

1 µ λ 1 µ λ κ
Lµν = exp(Θ)µν ≡ δνµ + Θµν + 2 Θ λΘ ν + 6 Θ λΘ κΘ ν + ··· (18)

of some matrix Θ that becomes antisymmetric when both of its indices are raised or lowered,
Θµν = −Θνµ . Note however that in the matrix exponential (18), the first index of Θ is raised

3
while the second index is lowered, so the antisymmetry condition becomes (gΘ)> = −(gΘ)
instead of Θ> = −Θ.

The Dirac spinor representation of the finite Lorentz transform (18) is the 4 × 4 matrix

MD (L) = exp − 2i Θαβ S αβ .

(19)

The group law for such matrices

∀L1 , L2 ∈ SO+ (3, 1), MD (L2 L1 ) = MD (L2 )MD (L1 ) (20)

follows automatically from the S µν satisfying the commutation relations (14) of the Lorentz
generators, so I am not going to prove it. Instead, let me show that when the Dirac matrices
γ µ are sandwiched between the MD (L) and its inverse, they transform into each other as
components of a Lorentz 4–vector,

−1
MD (L)γ µ MD (L) = Lµν γ ν . (21)

This formula makes the Dirac equation transform covariantly under the Lorentz transforms.

Proof: In light of the exponential form (19) of the matrix MD (L) representing a finite Lorentz
transform in the Dirac spinor multiplet, let’s use the multiple commutator formula (AKA
the Hadamard Lemma ): for any 2 matrices F and H,

1 1
        
exp(−F )H exp(+F ) = H + H, F + 2 H, F , F + 6 H, F , F , F + · · · . (22)

−1
In particular, let H = γ µ while F = − 2i Θαβ S αβ so that MD (L) = exp(+F ) and MD (L) =
exp(−F ). Consequently,

−1
(L)γ µ MD (L) = γ µ + γ µ , F +
  1
 µ   1
 µ   
MD 2 γ ,F ,F + 6 γ , F , F , F + · · · (23)

where all the multiple commutators turn out to be linear combinations of the Dirac matrices.

4
Indeed, the single commutator here is

= Θαβ g µα γ β = Θµλ γ λ , (24)

 µ 
γ , F = − 2i Θαβ γ µ , S αβ = 1
g µα γ β − g µβ γ α
 

2 Θαβ

while the multiple commutators follow by iterating this formula:

γ , F , F = Θµλ γ λ , F = Θµλ Θλν γ ν , γ , F , F , F = Θµλ Θλρ Θρν γ ν , . . . . (25)

 µ      µ   

Combining all these commutators as in eq. (23), we obtain

−1 µ
γ MD = γ µ + γ µ , F +
 µ  
1 1
γ µ, F
     
MD 2 γ ,F ,F + 6 ,F ,F + ···
= γ µ + Θµν γ ν + 21 Θµλ Θλν γ ν + 1 µ λ ρ ν
+ ···
6 Θ λΘ ρΘ ν γ
  (26)
= δνµ + Θµν + 12 Θµλ Θλν + 61 Θµλ Θλρ Θρν + · · · γ ν
≡ Lµν γ ν .

Quod erat demonstrandum.

Dirac Equation and Dirac Spinor Fields

History:
Originally, the Klein–Gordon equation was thought to be the relativistic version of the
Schrödinger equation — that is, an equation for the wave function ψ(x, t) for one relativistic
particle. But pretty soon this interpretation run into trouble with bad probabilities (negative,
or > 1) when a particle travels through high potential barriers or deep potential wells. There
were also troubles with relativistic causality, and a few other things.

Paul Adrien Maurice Dirac had thought that the source of all those troubles was the
p
ugly form of relativistic Hamiltonian Ĥ = p̂2 + m2 in the coordinate basis, and that he
could solve all the problems with the Klein-Gordon equation by rewriting the Hamiltonian
as a first-order differential operator

∂ψ
Ĥ = p̂ · α
~ + mβ =⇒ Dirac equation i = −i~
α · ∇ψ + mβψ (27)
∂t

where α1 , α2 , α3 , β are matrices acting on a multi-component wave function. Specifically, all

5
four of these matrices are Hermitian, square to 1, and anticommute with each other,

{αi , αj } = 2δij , {αi , β} = 0, β 2 = 1. (28)

Consequently

2
~ · p̂
α = αi αj × p̂i p̂j = 1
2 {αi , αj } × p̂i p̂j = δij × p̂i p̂j = p̂2 , (29)

and therefore
 2
2
2
ĤDirac = α~ · p̂ + βm = ~ · p̂
α + {αi , β} × p̂i m + β 2 × m2 = pˆ2 + 0 + m2 . (30)

This, the Dirac Hamiltonian squares to p̂2 + m2 , as it should for the relativistic particle.

The Dirac equation (27) turned out to be a much better description of a relativistic
electron (which has spin = 12 ) than the Klein–Gordon equation. However, it did not resolve
the troubles with relativistic causality or bad probabilities for electrons going through big
potential differences e∆Φ > 2me c2 . Those problems are not solvable in the context of a
relativistic single-particle quantum mechanics but only in quantum field theory.

Modern point of view:

Today, we interpret the Dirac equation as the equation of motion for a Dirac spinor field
Ψ(x), comprising 4 complex component fields Ψα (x) arranged in a column vector
 
Ψ1 (x)
 Ψ2 (x) 
 
Ψ(x) = 
 Ψ (x)  ,
 (31)
 3 
Ψ4 (x)

and transforming under the continuous Lorentz symmetries x0µ = Lµν xν according to

Ψ0 (x0 ) = MD (L)Ψ(x). (32)

The classical Euler–Lagrange equation of motion for the spinor field is the Dirac equation

∂
i α · ∇Ψ − mβΨ = 0.
Ψ + i~ (33)
∂t

6
To recast this equation in a Lorentz-covariant form, let

β = γ 0, αi = γ 0 γ i ; (34)

it is easy to see that if the γ µ matrices obey the anticommutation relations (5) then the α
~
and β matrices obey the relations (28) and vice verse. Now let’s multiply the whole LHS of
the Dirac equation (33) by the β = γ 0 :

  
0 = γ 0 i∂0 + iγ 0~γ · ∇ − mγ 0 Ψ(x) = iγ 0 ∂0 + iγ i ∂i − m)Ψ(x), (35)

and hence

iγ µ ∂µ − m Ψ(x) = 0.


(36)

2
As expected from ĤDirac = p̂2 + m2 , the Dirac equation for the spinor field implies the
Klein–Gordon equation for each component Ψα (x). Indeed, if Ψ(x) obey the Dirac equation,
then obviously

−iγ ν ∂ν − m × iγ µ ∂µ − m Ψ(x) = 0,



(37)

but the differential operator on the LHS is equal to the Klein–Gordon m2 + ∂ 2 times a unit
matrix:

−iγ ν ∂ν − m iγ µ ∂µ − m = m2 + γ ν γ µ ∂ν ∂µ = m2 + 12 {γ µ , γ ν }∂ν ∂µ = m2 + g µν ∂ν ∂µ .

(38)

The Dirac equation (36) transforms covariantly under the Lorentz symmetries —
its LHS transforms exactly like the spinor field itself.
Proof: Note that since the Lorentz symmetries involve the xµ coordinates as well as the
spinor field components, the LHS of the Dirac equation becomes

iγ µ ∂µ0 − m Ψ0 (x0 )


(39)

where
∂ ∂xν ∂
ν
∂µ0 ≡ 0µ
= 0µ
× ν = L−1 µ
× ∂ν . (40)
∂x ∂x ∂x

7
Consequently,

ν
∂µ0 Ψ0 (x0 ) = L−1 µ
× MD (L) ∂ν Ψ(x) (41)

and hence

ν
γ µ ∂µ0 Ψ0 (x0 ) = L−1 µ
× γ µ MD (L) ∂ν Ψ(x). (42)

But according to eq. (23),

−1
MD (L)γ µ MD (L) = Lµν γ ν =⇒ γ µ MD (L) = Lµν × MD (L)γ ν

ν (43)
=⇒ L−1 µ × γ µ MD (L) = MD (L)γ ν ,

so

γ µ ∂µ0 Ψ0 (x0 ) = MD (L) × γ ν ∂ν Ψ(x). (44)

Altogether,

iγ µ ∂µ − m Ψ(x) −−−−→ iγ µ ∂µ0 − m Ψ0 (x0 ) = MD (L) × iγ µ ∂µ − m Ψ(x),

(45)
Lorentz

which proves the covariance of the Dirac equation. Quod erat demonstrandum.

Dirac Lagrangian

The Dirac equation is a first-order differential equation, so to obtain it as an Euler–

Lagrange equation, we need a Lagrangian which is linear rather than quadratic in the spinor
field’s derivatives. Thus, we want

L = Ψ × iγ µ ∂µ − m Ψ


(46)

where Ψ(x) is some kind of a conjugate field to the Ψ(x). Since Ψ is a complex field, we
treat it as a linearly-independent from the Ψ, so the Euler–Lagrange equation for the Ψ

∂L ∂L
iγ ν ∂ν − m Ψ − ∂µ 0



0 = − ∂µ = (47)
∂Ψ ∂∂µ Ψ

immediately gives us the Dirac equation for the Ψ(x) field.

8
 To keep the action S = d4 xL Lorentz-invariant, the Lagrangian (46) should transform
R

as a Lorentz scalar, L0 (x0 ) = L(x). In light of eq. (19) for the Ψ(x) field and covariance (45)
of the Dirac equation, the conjugate field Ψ(x) should transform according to

0 −1
Ψ (x0 ) = Ψ(x) × MD (L) =⇒ L0 (x0 ) = L(x). (48)

−1
Note that the MD (L) matrix is generally not unitary, so the inverse matrix MD (L) in
†
eq. (48) is different from the hermitian conjugate MD (L). Consequently, the conjugate
field Ψ(x) cannot be identified with the hermitian conjugate field Ψ† (x), since the latter
transforms to

† −1
Ψ0† (x0 ) = Ψ† (x) × MD (L) 6= Ψ† (x) × MD (L). (49)

Instead of the hermitian conjugate, we are going to use the Dirac conjugate spinor, see below.

Dirac conjugates:
Let Ψ be a 4-component Dirac spinor and Γ be any 4 × 4 matrix; we define their Dirac
conjugates according to

Ψ = Ψ† × γ 0 , Γ = γ 0 × Γ† × γ 0 . (50)

Thanks to γ 0 γ 0 = 1, the Dirac conjugates behave similarly to hermitian conjugates or

transposed matrices:

• For a a product of 2 matrices, (Γ1 × Γ2 ) = Γ2 × Γ1 .

• Likewise, for a matrix and a spinor, (Γ × Ψ) = Ψ × Γ.

• The Dirac conjugate of a complex number is its complex conjugate, (c × 1) = c∗ × 1.

∗
• For any two spinors Ψ1 and Ψ2 and any matrix Γ, Ψ1 ΓΨ2 = Ψ2 ΓΨ1 .

◦ The Dirac spinor fields are fermionic, so they anticommute with each other, even

†
in the classical limit. Nevertheless, Ψ†α Ψβ = +Ψ†β Ψα , and therefore for any

∗
matrix Γ, Ψ1 ΓΨ2 = + Ψ2 ΓΨ1 .

9
 The point of the Dirac conjugation (50) is that it works similarly for all four Dirac
matrices γ µ ,

γ µ = +γ µ . (51)

Proof: For µ = 0, the γ 0 is hermitian, hence

γ 0 = γ 0 (γ 0 )† γ 0 = +γ 0 γ 0 γ 0 = +γ 0 . (52)

For µ = i = 1, 2, 3, the γ i are anti-hermitian and also anticommute with the γ 0 , hence

γ i = γ 0 (γ i )† γ 0 = −γ 0 γ i γ 0 = +γ 0 γ 0 γ i = +γ i . (53)

Corollary: The Dirac conjugate of the matrix

MD (L) = exp − 2i Θµν S µν

(19)

representing any continuous Lorentz symmetry L = exp(Θ) is the inverse matrix

−1
(L) = exp + 2i Θµν S µν .


M D (L) = MD (54)

Proof: Let

X = − 2i Θµν S µν = + 18 Θµν [γ µ , γ ν ] = + 41 Θµν γ µ γ ν (55)

for some real antisymmetric Lorentz parameters Θµν = −Θνµ . The Dirac conjugate of the
X matrix is

1 ∗ ν µ ν µ µ ν
X = 1 µ ν
4 Θµν γ γ = 4 Θµν γ γ = 1
4 Θµν γ γ = 1
4 Θνµ γ γ = − 14 Θµν γ µ γ ν = −X. (56)

Consequently,

X 2 = X × X = +X 2 , X 3 = X × X 2 = X 2 × X = −X 3 , ..., X n = (−X)n ,
(57)

10
and hence
X 1 X 1
exp(X) = Xn = (−X)n = exp(−X). (58)
n
n! n
n!

In light of eq. (55), this means

exp − 2i Θµν S µν = exp + 2i Θµν S µν ,

(59)

that is,

−1
M D (L) = MD (L). (60)

Quod erat demonstrandum.

Back to the Dirac Lagrangian:

Thanks to the theorem (60), the conjugate field Ψ(x) in the Lagrangian (46) is simply the
Dirac conjugate of the Dirac spinor field Ψ(x),

Ψ(x) = Ψ† (x) × γ 0 , (61)

which transforms under Lorentz symmetries as

0 −1
Ψ (x0 ) = Ψ0 (x0 ) = MD (L) × Ψ(x) = Ψ(x) × M D (x) = Ψ(x) × MD (L). (62)

Consequently, the Dirac Lagrangian

L = Ψ × iγ µ ∂µ − m Ψ = Ψ† γ 0 × iγ µ ∂µ − m Ψ



(46)

is a Lorentz scalar and the action is Lorentz invariant.

11
 Hamiltonian for the Dirac Field

Canonical quantization of the Dirac spinor field Ψ(x) — just like any other field — begins
with classical Hamiltonian formalism. Let’s start with the canonical conjugate fields,

∂L
iΨγ 0 = iΨ†α


Πα = = α
(63)
∂(∂0 Ψα )

— the canonical conjugate to the Dirac spinor field Ψ(x) is simply its hermitian conjugate
Ψ† (x). This is similar to what we had for the non-relativistic field, and it happens for the
same reason — the Lagrangian which is linear in the time derivative.

In the non-relativistic field theory, the conjugacy relation (63) in the classical theory
lead to the equal-time commutation relations in the quantum theory,

 †
ψ̂ (x, t), ψ̂ † (y, t) = 0, ψ̂(x, t), ψ̂ † (y, t) = δ (3) (x − y). (64)
    
ψ̂(x, t), ψ̂(y, t) = 0,

1
However, the Dirac spinor field describes spin = 2 particles — like electrons, protons, or
neutrons — which are fermions rather than bosons. So instead of the commutations rela-
tions (64), the spinor fields obey equal-time anti-commutation relations


Ψ̂α (x, t), Ψ̂β (y, t) = 0,
Ψ̂α (x, t), Ψ̂†β (y, t) = 0,
 †
(65)
Ψ̂α (x, t), Ψ̂†β (y, t) = δαβ δ (3) (x − y).


Next, the classical Hamiltonian obtains as

Z
H = d3 x H(x),

H = iΨ† ∂0 Ψ − L
(66)
† † 0 0


= iΨ ∂0 Ψ − Ψ γ iγ ∂0 + i~γ · ∇ − m Ψ
= Ψ† −iγ 0~γ · ∇ + γ 0 m Ψ

where the terms involving the time derivative ∂0 cancel out. Consequently, the Hamiltonian

12
operator of the quantum field theory is
Z
d3 x Ψ̂† (x) −iγ 0~γ · ∇ + γ 0 m Ψ̂(x).


Ĥ = (67)

Note that the derivative operator (−iγ 0~γ · ∇ + γ 0 m) in this formula is precisely the 1-particle
Dirac Hamiltonian (27). This is very similar to what we had for the quantum non-relativistic
fields,
 
−1 2
Z
3 †
Ĥ = d x ψ̂ (x) ∇ + V (x) ψ̂(x), (68)
2M

except for a different differential operator, Schrödinger instead of Dirac.

In the Heisenberg picture, the quantum Dirac field obeys the Dirac equation. To see
how this works, we start with the Heisenberg equation
Z
∂
d3 y Ψ̂α (x, t), Ĥ(y, t) ,
   
i Ψ̂α (x, t) = Ψ̂α (x, t), Ĥ = (69)
∂t

and then evaluate the last commutator using the anti-commutation relations (65) and the
Leibniz rules (11). Indeed, let’s use the Leibniz rule

[A, BC] = {A, B}C − B{A, C} (70)

for
A = Ψ̂α (x, t),
B = Ψ̂†β (y, t), (71)
−iγ 0~γ · ∇ + γ 0 m βγ Ψ̂γ (y, t),


C =

so that BC = Ĥ(y, t). For the A, B, C at hand,

{A, B} = δαβ δ (3) (x − y) (72)

while

−iγ 0~γ · ∇y + γ 0 m βγ {Ψ̂α (x, t), Ψ̂γ (y, t)} = (diff.op.) × 0 = 0.

{A, C} = (73)

13
Consequently

 
Ψ̂α (x, t), Ĥ(y, t) ≡ [A, BC]
= {A, B} × C − B × {A, C}
(74)
= δαβ δ (3) (x − y) × −iγ 0~γ · ∇ + γ 0 m βγ Ψ̂γ (y, t)

− 0,

hence
Z
d3 y δ (3) (x − y) × −iγ 0~γ · ∇ + γ 0 m αγ Ψ̂γ (y, t)
 

Ψ̂α (x, t), Ĥ =
(75)
−iγ 0~γ · ∇ + γ 0 m αγ Ψ̂γ (x, t),


=

and therefore

−iγ 0~γ · ∇ + γ 0 m Ψ̂(x, t).

i∂0 Ψ̂(x, t) = (76)

Or if you prefer,

iγ µ ∂µ − m Ψ̂(x) = 0.


(77)

14