253a: QFT1

Fall 2008
Matthew Schwartz

Lecture 10:
Spinors and the Dirac Equation

1 Introduction
1
From non-relativistic quantum mechanics, we already know that the electron has spin 2
. We
usually write it as a doublet  
 ↑
|ψ i =  (1)
↓
You might also remember the Pauli matrices
     
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 = (2)
1 0 i 0 0 −1
which simplify the treatment of spin.
For example, the interaction of the electron with a magnetic field appears in what is some-
times called the Schrodinger-Pauli equation:

K · LK 12×2 − 2µBBK · Kσ ψ
 2  
p
i∂tψ = Hψ = + V (r) − µBB (3)
2m
e
where µB = 2m is the “Bohr magneton” (the size of the electron’s orbital magnetic moment) and
K = xK × Kp is ethe angular momentum operator. We have also written Kσ = (σ1, σ2, σ3) to call
L
attention to the fact that these σi′s transform as the components of a vector, just like the mag-
netic field Bi. Thus (σ K · BK )ψ is rotationally invariant. This is non-trivial, and only works
because
[σi , σ j ] = 2iεijkσk (4)
which are the same algebraic relations satisfied by infinitesimal rotations (we will review this
shortly).
Keep in mind that σi do not change under rotations – they are always given by Eq. (2) in
any frame. ψ is changing and Bi is changing, and these changes cancel when we write (σ K · BK )ψ.
We could also have written down a rotationally invariant equation of motion for ψ
1 · ∂tψ − ∂iσiψ = 0 (5)
Since ∂i transforms like a 3-vector and so does σiψ, this equation is rotationally invariant. It
turns out it is Lorentz invariant too. In fact, this is just the Dirac equation!
If we write
σ µ = (12×2, σ1, σ2, σ3) (6)
Then it is
σ µ∂ µψ = 0 (7)
which is nice and simple looking. Actually, this is the Dirac equation for a Weyl spinor, which is
not exactly the same as the equation commonly called the Dirac equation.
By the way, it does not follow that this equation is Lorentz invariant just because we’ve
written it as σ µ∂ µ. For example,
(σ µ∂ µ + m)ψ = 0 (8)
is not Lorentz invariant. To understand these enigmatic transformation properties, we have to
know how to construct the Lorentz group out of the σ µ’s and see how those act on ψ. But
instead of just guessing the answer (or looking it up), we’re going to do something a little more
general, and motivate what kinds of transformation properties are possible.

1
2 Section 2

2 Representations of the Lorentz Group

We have already argued that particles in our world should transform under unitary representa-
tions of the Poincare group. These are characterized by mass m and spin j. We motivated this
in our study of the spin 1 fields. Mass is Lorentz invariant, so it is an obvious quantum number.
Momentum is also conserved, but it is Lorentz covariant. If choose a frame in which the
momentum has some canonical form, for example p µ = (m, 0, 0, 0) for m > 0, then the particles
are characterized by the group that holds this momentum fixed. That is the little group, in this
case the group of 3D rotations, SO(3). The group of 3D rotations provides the second quantum
number, j. The way the polarizations transform under the full Lorentz group is then induced by
the transformations under SO(3) and the way the momentum transforms under boosts.
We also saw that to do field theory, we have to write down Largrangians involving fields.
These are things like V µ or φ or T µν which don’t have to involve particles per se, although obvi-
ously in a physical theory they will. As we saw for spin 1, there’s a lot a trouble that comes
from having to embed particles of fixed mass and spin into fields like V µ. For example, V µ has 4
degrees of freedom which describes spin 0 and spin 1, so the Lagrangian has to be carefully
chosen to make sure the physical theory never excites the spin 0 component. In addition, when
we want the particle to have m = 0 and spin 1, we needed to go further and make sure the longi-
tudinal polarization is never produced. This led directly to charge conservation. The next log-
ical step to make these embeddings a bit more systematic is to see what kinds of Lorentz-
1
invariant fields we can write down at all. This will reveal the existence of the spin 2 states, and
help us characterize their embeddings into fields.
A group is a set of elements {gi } and a rule gi ⊗ gj → gk which tells how each pair of ele-
ments is multiplied to get a third. The rule defines the group, independent of any particular way
to write the group elements down as matrices. A representation is particular embedding of these
gi′s into matrices. Often we talk about the vectors on which the matrices act as being the repre-
sentation, but technically the matrix embedding is the representation. Any group has the trivial
representations r: gi → 1. But more generally we care about embeddings which are faithful , for
which each element gets it’s own matrix.
Recall that the Lorentz group is the set of rotations and boosts which preserve the
Minkoswki metric
ΛTηΛ = η (9)
Lorentz transformation can act on space-time 4-vectors as
V µ → ΛµνVν (10)

Where Λ is a combination of rotations and boosts. We saw that we could write a Lorentz trans-
formation as the product of 3 rotations and 3 boosts:
   
1 0 0 0 1 1
 0 cos θxy − sin θxy  cos θ xz − sin θ xz 
 1 
Λ = 
 0 sin θx y cosθx y 0 
 (11)
1  cosθx z − sin θx z 
0 0 0 1 sin θ xz cosθ xz sin θxz cos θxz
   
coshβx sinhβx coshβ y sinhβ y coshβz sinhβz
 sinhβx coshβx  1  1 
×    (12)
 1  sinhβ y coshβ y  1 
1 1 sinhβz coshβz

This is a particular representation. of the Lorentz group. That is, it is one embedding of the
group into a set of matrices.
The group itself is a mathematical object independent of any particular representation. To
extract the group away from its representations, it is easiest to look at infinitesimal transforma-
tions. We can always write these group elements as
Λ = exp(iθiλi) = 1 + iθiλi +  (13)
General representations 3

where θi are 6 numbers, corresponding to the 6 angles θxy , θxz , θ yz , βx y , βx z , β yz and Λi are
called generators. For the regular, vector, representation above, the generators are
     
0 0 0
 0 −1   0 −1   0 
λ1 = V12 = i , λ2 = V13 = i , λ3 = V23 = i 
 1 0   0   0 −1 
0 1 0 1 0
     
0 1 0 1 0 1
 1 0   0   0 
λ4 = V01 = i , λ5 = V02 = i , λ6 = V03 = i  (14)
 0   1 0   0 
0 0 1 0

The λ ′s and the V ′s are just different notations for the same objects. The generators satisfy:

[V µν , V ρσ ] = i(ηνρV µσ − η µρVνσ − ηνσV µρ + η µσVνρ) (15)

which means
[V01 , V12 ] = iV02 , [V12 , V23 ] = iV13 ,  (16)

So now we can define the Lorentz group as the set of transformations generated by these genera-
tors.
Another representation or the Lorentz generators is given

J µν = i(x µ∂ν − xν∂ µ) (17)

These are the classical generators of angular momentum generalized to include time. You can
check that J µν satisfy the commutation relations of the Lorentz algebra.
Technically, we say the generators make up the Lorentz Algebra so(1,3) which generates the
Lorentz Group SO(1, 3). A common convention is to use lowercase letters for the names of alge-
bras and uppercase letters for groups. As another technicality, we note that it is possible for two
different groups to have the same algebra. For example, the Proper Orthonchronous Lorentz
group and the Lorentz Group have the same algebra, but the Lorentz group has in addition the
discrete symmetries time reversal and parity reversal. It’s a small difference, but it’s a difference
and the groups are not identical.

3 General representations
There is a very nice way to study the representations of the Lorentz group. Start with the rota-
tion generators Ji and the boost generators K j . For example, you could take J1 = V23 , K1 = V01 ,
etc. as a particular representation of this basis. These satisfy
[Ji , J j ] = iǫijkJk (18)

[Ji , K j ] = iǫijkKk (19)

[Ki , K j ] = − iǫijkJk (20)

where ǫijk is defined by ε123 = 1 and being antisymmetric in all of its indices (it is called the
totally antisymmetric tensor .) As is probably clear to you already, [Ji , J j ] = iǫijkJk is the
algebra for rotations, so(3) so the Ji generate the subgroup of rotations in 3D.
Now take the linear combinations
1
Ji+ = (Ji + iKi) (21)
2
1
Ji− = (Ji − iKi) (22)
2
4 Section 3

which satisfy
[Ji+, J j+] = iǫijkJk+ (23)

[Ji−, J j−] = iǫijkJk− (24)

[Ji+, J j−] = 0 (25)

So we have found 2 commuting subgroups of the Lorentz group. The algebra of the J ′s is the
3D rotation algebra, so(3), more commonly called so(2). So we have shown that

so(1, 3) = su(2) × su(2) (26)

(technically, it should really be sl(2, R) = so(1, 1) not su(2), but these algebras are the same up
to some i’s, and physicists usually just say su(2).).
The names so(n) and su(n) stand for special orthogonal algebra and special unitary algebra.
Orthogonal means it preserves a norm defined with transpose: V TV is invariant under SO(n).
Unitary means it preserves a norm defined with adjoint V †V is invariant under SU(n). Special
means the determinant is 1, which just normalizes the elements and gets rid of phases.
The decomposition so(3, 1) = su(2) × su(2) makes studying the representations very easy. We
already know what the representations are of su(2), since this is the algebra of Pauli matrices,
which generates the 3D rotation group SO(3) (so(3) = su(2)). The representations are character-
ized by a quantum number j, and have 2j + 1 elements, labeled by m = jz, So representations of
the Lorentz group are characterized by two numbers A and B. The (A, B) representation has
(2A + 1)(2B + 1) degrees of freedom.
The regular rotation generators are J K = JK + + JK −, where I now use the vector superscript to
call attention the fact that the spins must be added vectorially (remember Clebsch-Gordon coef-
ficients?). The bottom line is that each (A, B) representation contains particles of spin J = A +
B , A + B − 1,  , |A − B |. For example,

rep of su(2) × su(2) rep of 3D rotation group

(A, B) = (0, 0) J =0
1 1
(A, B) = ( , 0) J=
2 2
1 1
(A, B) = (0, ) J= (27)
2 2
1 1
(A, B) = ( , ) J = 1, 0
2 2
(A, B) = (1, 0) J =1
(A, B) = (1, 1) J = 2, 1, 0

So we see that the regular 4D vector representation A µ contains spins 1,0 so it corresponds to
1 1
the ( 2 , 2 ) representation of the Lorentz group.

3.1 unitary representations

Representations of the Lorentz group are not always unitary. Unitarity means Λ †Λ = 1, which is
necessary to have Lorentz invariant matrix elements
hψ |ψi → ψ|Λ †Λ|ψ


(28)

Since the group element is the exponential of the generator Λ = eiλ, unitarity requires that λ † =
λ, that is, that λ be Hermetian. Looking back at our explicit 4D representation for the Lorentz
generators, we see that the rotations
     
0 0 0
 0 −1   0 −1   0 
J1 = i , J2 = i , J3 = i 
 1 0   0   0 −1 
0 1 0 1 0
 1
Spin 2
representation 5

are Hermetian. While the boosts

     
0 1 0 1 0 1
 1 0   0   0 
K1 = i

, K2 = i 1
  , K3 = i  (29)
0 0   0 
0 0 1 0

Are not. (They are anti-Hermetian λ † = − λ.)

Non-unitary representations of the Lorentz group are ok for fields, but not for particles. Par-
ticles must transform under unitary representations of the Poincare group, so that we can con-
struct physical amplitudes with them. As Wigner showed, and we discussed when we introduced
gauge invariance, unitary representations and are characterized by mass and spin. This spin
indicates a representation of the 3D rotation group, which is unitary.
Now, the generators for the su(2) × su(2) decomposition JK ± = JK ± i KK are Hermetian. This is
because su(2) is the special unitary group, so of course it has unitary representations. But this
doesn’t mean that the corresponding representations of the Lorentz group are unitary. You have
to be very careful about the factors of i and how these groups are defined. We said that a
Lorentz group element is
Λ = exp(iθiJi + iβiKi) (30)
where the θi are the rotation angles and βi the boosts “angles”. These are real numbers. If we
took the basis λi = J ±, then the corresponding θi′s would have to be complex. So the representa-
tions of the Lorentz group are determined by the algebra su(2) × su(2), but since the angles θ
are real, the corresponding representations of the Lorentz group are not unitary.

3.2 summary
In summary, the irreducible representations of the Lorentz group are characterized by two spins
(A, B). Physical particles are characterized by mass, m and spin j. Fields generally describe
particles of many spins, and one has to make sure that a physical theory only has causally prop-
agating particles of the spins we want.

1
4 Spin 2
representation
Now that we know that the irreducible representations of the Lorentz group are characterized by
1
two spins, (A, B), how do we find the explicit representations? In particular, what do the ( 2 , 0)
1
and (0, 2 ) representations look like? We need to find 2x2 matrices that satisfy
[Ji+, J j+] = iǫijkJk+ (31)

[Ji−, J j−] = iǫijkJk− (32)

[Ji+, J j−] = 0 (33)

But we already know such matrices: the Pauli matrices
[σi , σ j ] = 2iεij kσk (34)
rescaling we find
σi σ j σ
[ , ] = iεij k k (35)
2 2 2
Another useful fact is that
{σi , σ j } = σiσ j + σ jσi = 2δij (36)
σi 1
Thus we can set Ji+ = 2 . This is the ( 2 , ). What about Ji−? This should be the (, 0), so the
1
obvious thing to do is just take the trivial representation Ji− = 0. So the ( 2 , 0) representation is
1 σi
( , 0): Ji+ = , Ji− = 0 (37)
2 2
6 Section 4

1
Similarly, the (0, 2 ) representation is
1 σi
(0, ): Ji+ = 0, Ji− = (38)
2 2
K = JK + + JK −
What does this mean for actual Lorentz transformations? Well, the rotations are J
K
and the boosts are K = i(J− − J+ ) so
1 σi σi
( , 0): Ji = , Ki = − i (39)
2 2 2
1 σi σi
(0, ): Ji = , Ki = i (40)
2 2 2
Since the Pauli matrices are Hermetian σi† = σi we see that the rotations are Hermetian but the
boosts are anti-Hermetian. This is the same as what we found for the vector representation.
Also notice that these two representations are complex conjugates of each other. In contrast, the
vector representation was real .
1
Explicitly, if ψL is a ( 2 , 0) spinor, known also as a left-handed Weyl spinor, then under rota-
tions angles θi and boost angles βi
1 1
ψL = (1 + iθiσi + βiσi +  )ψL
1
(iθiσi +βiσi)
ψL → e 2 (41)
2 2
Similarly,
1 1
ψR → eiθiσi − βiσiψR = (1 + iθiσi − βiσi +  )ψR (42)
2 2
Infinitesimally,
1
δψL = (iθi + βi)σiψL (43)
2
1
δψR = (iθi − βi)σiψR (44)
2
Note again the angles θi and βi are real numbers. Although we mapped Ji− or Ji+ to 0, we still
have non-trivial action of all the Lorentz generators. So these are faithful irreducible representa-
tions of the Lorentz group. Similarly
1
δψL† = ( − iθi + βi)ψL† σi (45)
2
† 1
δψR = ( − iθi − βi)ψL† σi (46)
2

4.1 Lorentz invariants

Now the simplest thing to do would be to write down a Lagrangian with terms like
(ψL) †ψL + m2(ψL) † ψL (47)
However, we can see from the above that this is not Lorentz invariant:
1 1
δψL† ψL = (ψL) †[(iθi + βi)σiψL] + [(ψL) †( − iθi + βi)σi]ψL (48)
2 2
βi †
= ψLσiψL  0 (49)
2
This is just the manifestation of the fact that the representation is not unitary because the
boost generators are anti-Hermetian. If we allow ourselves two fields, ψL and ψR, we can write
†
down terms like ψR ψL. Under infinitesimal Lorentz transformations,
   
† † 1 † † 1
δ(ψRψL) = ψR ( − iθi − βi)σi ψL + ψR (iθi + βi)σi ψL = 0 (50)
2 2
 1
Spin 2
representation 7

Which is great. However, this term is not real. But then we can just add the Hermetian conju-
gate, so  
†
m ψR ψL + ψL† ψR (51)
is ok.
What about kinetic terms? We could try
†
ψR ψL + ψL† ψR (52)
which is both Lorentz invariant and real. But this is actually
 
not a very interesting Lagrangian.
ψ1
We can always split up our field into components ψL = ψ , where ψ1 and ψ2 are just regular
2

fields. Then we see that this is just the Lagrangian for a couple of scalars. So it’s not enough to
declare the Lorentz transformation properties of something, the Lagrangian has to force those
transformation properties. In the same way, a vector field is just 4 scalars until we contract it
with ∂ µ in the Lagrangian.
To proceed, let’s look at
ψL† σ jψL (53)
This transforms as
1 1
δψL† σ jψL = ψL† σ j [(iθi + βi)σiψL] + [ψL† ( − iθi + βi)σi]σ jψL (54)
2 2
iθi † β
= ψL(σ jσi − σ jσi)ψL + i ψL† (σiσ j + σ jσi)ψL (55)
2 2
= θiεi jkψL† σkψL + β jψL† ψL (56)
Thus we have found that
βi †
δψL† ψL = ψ σiψL (57)
2 L
δψL† σ jψL = θiεijkψL† σkψL + β jψL† ψL (58)
Similarly,
† βi †
δψR ψR = − ψ σiψL (59)
2 L
†
δψR σ jψR = θiεijkψL† σkψL + − β jψL† ψL (60)
Now, recall how a vector transforms
δV0 = βiV0 (61)
δV j = β jV0 + θiε ij kVk (62)

This is just the transformation of the vector V µL = (ψL† ψL , ψL† σ jψL). Thus

ψL† ∂tψL + ψL† ∂ jσ jψL (63)

† †
is Lorentz invariant. Similarly V µR = (ψR ψR , ψR σ jψR) transforms like a vector
Defining
σ µ = (1, Kσ ), σ̄ µ = (1, − Kσ ) (64)
We can write all the Lorentz-invariant terms we have found as
 
L = iψL† σ µ∂ µψL + iψR
† †
σ̄ µ∂ µψR + m ψR ψL + ψL† ψR (65)

I have added a factor of i on the kinetic terms for later convenience.

There’s an even shorter-hand way to write this. If we write our spinor as a doublet
 
ψL
ψ= (66)
ψR
Then, define  
†
ψ̄ = ψR ψL† (67)
8 Section 5

And using the 4x4 matrices  

σµ
γµ = (68)
σ̄ µ
Our Lagrangian becomes
L = ψ̄ (γ µ∂ µ − m)ψ (69)
Which is the conventional form of the Dirac equation.

5 Dirac matrices
Expanding them out, the Dirac matrices are
   
1 0 σi
γ0 = γi = (70)
1 − σi 0
Or,    
0 1 0 1
 0 1   0 1 
γ0 =
 1
, γ1 =  (71)
0   −1 0 
1 0 −1 0
   
0 −i 0 1
 0 i   0 −1 
γ2 = , γ3 =  (72)
 i 0   −1 0 
−i 0 1 0
They satisfy
{γ µ , γν } = 2η µν (73)
In the same way that the algebra of the Lorentz group is more fundamental than any particular
representation, the algebra of the γ ′s is more fundamental than any particular representation of
them. We say the γ ′s form the Dirac algebra, which is a special case of a Clifford algebra. This
particular form of the Dirac matrices is known as the Weyl representation.
The Lorentz generators are
i
S µν = [γ µ , γν ] (74)
4
They satisfy the Lorentz algebra for any γ ′s satisfying the Clifford algebra. That is, you can
derive from {γ µ , γν } = 2η µν that
[S µν , S ρσ ] = i(ηνρS µσ − η µρSνσ − ηνσS µρ + η µσSνρ) (75)
In the Weyl representation, the Lorentz generators are
   
1 σk i σi
Si j = εij k , Ki = S0i = − (76)
2 σk 2 − σi
Or, very explicitly
     
1 0 1 0 1
1 −1  i − , S23 = 1 1 0
1 0
  
S12 =  , S13 =   (77)
2 1  2 0 1  2 0 1 
−1 −1 0 1 0
     
0 −1 0 −1 −1
i −1 0 , S02 = 1 1 0 , S03 = i 1
    
S01 =   (78)
2 0 1  2 0 1  2 1 
1 0 −1 0 −1
1 1
These are block diagonal. These are the same generators we used for the ( 2 , 0) and (0, 2 ) repre-
sentations above. It makes it clear that the Dirac representation is reducible, it is the sum of a
left-handed and a right-handed spinor representation.
Rotations 9

Another important representation is the Majorana representation

! ! ! !
0 σ 2 iσ 3 0 0 − σ 2 − iσ 1 0
γ0 = , γ1 = , γ2 = γ3 = (79)
σ2 0 0 iσ 3 σ2 0 0 − iσ 1

In this basis the γ ′s are purely imaginary.

The Weyl spinors, ψL and ψR are more fundamental, because they correspond to irreducible
representations of the Lorentz group. But the electron is a Dirac spinor. Thus to do QED, it is
easiest just to stick to the γ ′s and to get used to manipulating them. Eventually, when you do
supersymmetry, or study the weak interactions, you will need to use the L and R representa-
tions again.

6 Rotations
Now let’s see what happens when we rotate by an angle θ in the xy plane. We use
Λ = exp(iθx yV12 ) (80)
How do we exponentiate a matrix? The standard trick is to first diagonalizing it with a unitary
transformation, do the rotation, then transform back. This unitary transformation is like
choosing a direction, except it is purely mathematical as the direction must be complex!
First, for the vector representation
   
0 0
 0 1  
= U −1 −1 
V12 = i U (81)
 −1 0   1 
0 0
So,  
0
 exp( − iθx y) 
Λ(θxy) = exp(iθx yV12 ) = U −1 U (82)
 exp(iθxy) 
0

Λ(2π) = 1 (83)
That is, we rotate 360 degrees and we’re back to where we started.
For the spinor representation
Λs = exp(iθ µνS µν ) (84)
the 12 rotation is already diagonal:
 
1
1 −1 
S12 =   (85)
2 1 
−1
So,  
i
exp( 2 θxy )
 i

 − exp( 2 θxy ) 
Λ(θxy) = exp(iθxyS12 ) = (86)
 
i 

 exp( 2 θxy ) 

i
− exp( 2 θxy )
 
−1
 1 
Λ(2π) =  (87)
 −1 
1
10 Section 7

Thus a 2π rotation does not bring us back where we started! If we rotate by 4π it would. So we
1
say spinors are spin 2 . What does this mean physically? I have no idea. There are plenty of
1
physical consequences of spinors being spin 2 , but this business of rotating by 2π is not actually
a physical thing you can do.
As an aside, note that the factor of 2 is determined by the normalization of the matrices,
which is set by the Lie Algebra. For each representation by itself, this would be arbitrary, but it
is important for expressions which combine the representations to be invariant.

7 Lorentz Invariants
The γ matrices themselves transform nicely under the Lorentz group.

Λs−1 γ µΛs = (ΛV ) µνγν (88)

where the Λs are the Lorentz transformations acting on each γ µ individually, as a matrix, and
the ΛV is the vector representation which mixes up the Lorentz indices. That is, writing out the
matrix indices γ µαβ , this means
(Λs−1)δαγ µαβ (Λs) βγ = (ΛV ) µνγναβ (89)

where µ refers to which γ matrix, and α and β index the elements of that matrix.
Then the equation
{γ µ , γν } = 2η µν (90)
really means
γ µαγγνγβ + γναγγ µγβ = 2η µνδ αβ (91)
And the equation
i
S µν = [γ µ , γν ] (92)
4
Should really be written as
αβ i  αγ γβ 
S µν = γ µ γν − γναγγ µγβ (93)
4
For an expression like
V µV µ = V µη µνVν = V µ {γ µ , γν }Vν (94)
To be invariant, it must be that µ transforms in the vector representation.
Next consider  
ψ † ψ → ψ †Λs† (Λsψ) (95)

For this to be invariant, we would need Λs† = Λs−1, that is, for the representation of the Lorentz
group to be unitary. The spinor representation, like any other representation is not unitary,
because the boost generators are anti-Hermetian.
It is useful to study the properties of the Lorentz generators from the Dirac algebra itself,
without needing to choose a particular basis for the γ µ. First note that

{γ µ , γν } = 2η µν ⇒ γ02 = 1, γi2 = − 1 (96)

So the eigenvalues of γ0 are ± 1 and the eigenvalues of γi are ± i. Thus if we diagonalize γ0, we
would find it Hermetian, and if we diagonalize γ1 γ2 or γ3 we would find they are anti-Herme-
tian. This is true in general.
γ0† = γ0, γi† = − γi (97)
Then,
 †
† i ih † † i ih † †i
(S µν ) = [γ µ , γν ] =− γ , γ = γ µ , γν (98)
4 4 ν µ 4
Which implies
† †
Sij = Sij S0i = − S0i (99)
Lorentz Invariants 11

Again, we see that the rotations are unitary and the boosts are not. You can see this from the
explicit representations above. But because we showed it algebraically, it is true in ANY repre-
sentation.
Now, observe that one of the Dirac matrices is Hermetian, γ0 (γ0 is the only Hermetian
Dirac matrix because the metric signature is (1,-1,-1-,1)). Moreover

γ0 γiγ0 = − γi = γi†, γ0 γ0 γ0 = γ0 = γ0† (100)

⇒ γ µ† = γ0 γµγ0 (101)

† ih † † i ih i i
⇒ γ0S µν γ0 = γ0 γ µ , γν γ0 = γ0 γ µ† γ0, γ0 γν† γ0 = [γ µ , γν ] = Sµν (102)
4 4 4
† †
(γ0Λsγ0) = γ0exp(iθ µνSµν ) † γ0 = exp( − iθ µνγ0S µνγ0) = exp( − iθ µνS µν ) = Λs−1 (103)
Then, finally,    
ψ † γ0 ψ → ψ †Λs† γ0(Λsψ) = ψ † γ0Λs−1Λsψ = ψ † γ0 ψ (104)

which is Lorentz invariant.

We have just been rederiving from the Dirac algebra point of view what we found by hand
from the Weyl point of view. The conjugate of ψ is not ψ † but

ψ̄ ≡ ψ † γ0 (105)

The point is that ψ̄ transforms according to Λs−1. Thus ψ̄ψ is Lorentz invariant.
We can also construct objects like like

ψ̄γ µψ, ψ̄γ µγνψ, ψ̄∂ µy (106)

all transform nicely under the Lorentz group. Also

L = ψ̄ (iγ µ∂ µ − m)ψ (107)

is Lorentz invariant. We abbreviate this with
L = ψ̄ (i∂ − m)ψ (108)
This is the Dirac Lagrangian.
The Dirac equation follows from this Lagrangian by the equations of motion

(i∂ − m)ψ = 0 (109)

To be explicit, this is shorthand for
(iγ µαβ∂ µ − mδ αβ )ψ β = 0 (110)

By multiplying on the right, we find

1 1
0 = (i∂ + m)(i∂ − m)ψ = ( − ∂ µ∂ν {γ µ , γν } − ∂ µ∂ν [γ µ , γν ] − m2)ψ (111)
2 2
= − (∂ 2 + m2)ψ (112)
So ψ satisfies the Klein-Gordon equation
( + m2)ψ = 0 (113)
It is in this sense that people sometimes say the Dirac equation is the square-root of the Klein-
Gordon equation.
We can integrate the Lagrangian by parts to derive the equations of motion for ψ̄ :

L = ψ̄ (i∂ − m)ψ = − i∂ µψ̄ (γ µ − m)ψ (114)

So,
− i∂ µψ̄γ µ − mψ̄ = 0 (115)
12 Section 8

This γ µ on the right is a little annoying, so we often hide it by writing

ψ̄ ( − i ∂ − m) = 0 (116)
where the derivative acts to the left. This makes the conjugate equation look more like the orig-
inal Dirac equation.

8 Coupling to the photon

Under a gauge transform ψ transforms just like a scalar, since the gauge invariance has nothing
to do with spin. So,
ψ → eiαψ (117)
Then we use the same covariant derivative ∂ µ − ieA µ as for a scalar. So
Dµψ = (∂ µ − ieA µ)ψ (118)
Then the Dirac equation becomes
(i∂ + e A − m)ψ = 0 (119)
Now we try to reproduce the Klein Gordon equation for a scalar field coupled to A µ.
[(i∂ µ + eA µ)2 − m2 φ

(120)
Following the same route as before
0 = (i∂ + eA + m)(i∂ + eA − m)ψ (121)
2
= (i∂ µ + eA µ)(i∂ν + eAν )γ µγ ν − m ψ
 
(122)
 
1 µ ν 1 µ ν 2
= {i∂ µ + eA µ , i∂ν + eAν }{γ , γ } + [i∂ µ + eA µ , i∂ν + eAν ][γ , γ ] − m ψ (123)
4 4
Before the antisymmetric combination dropped out, but now we find
1
[i∂ µ + eA µ , i∂ν + eAν ] = e[i∂ µAν − i∂νA µ] = eiF µν (124)
4
So we get  
2 ie µ ν 2
(i∂ µ + eA µ) + F µν [γ , γ ] − m ψ (125)
4
Which contains an extra term compared to the Klein-Gordon equation.
i
What is this term? Well, recall that 4 [γ µ , γ ν ] = S µν , our Lorentz generators. These have the
form (in the Weyl representation)
   
1 σk i σi
Sij = εijk , S0i = − (126)
2 σk 2 − σi
And since
F0i = Ei , Fij = εi jkBk (127)
We get
K + i EK )σK
( !)
2 2e (B
(∂ µ − ieA µ) + m −
2 K − iEK )σK
(B
ψ=0 (128)

This corresponds to a magnetic dipole moment.

This is pretty remarkable. For a free spinor, we reproduce the equation of motion of a scalar
field. But when the spinor is coupled to the photon, we find an additional interaction corre-
1
sponds to a magnetic dipole moment. We can read off that the electron has spin 2 . Note: the
coupling to the electric field is not an electric dipole moment – that would not have an i, but is
simply the effect of a magnetic moment in a boosted frame.
Weyl spinors – Helicity 13

9 Weyl spinors – Helicity

1 1
Dirac spinors, what we have been using, are 4 component complex fields in the ( 2 , 0) ⊕ (0, 2 )
representation of the Lorentz group. Let’s return for a moment to thinking about the 2-compo-
nent fields.
In the Weyl basis, the γ matrices have the form
 
0 σµ
γµ = (129)
σ̄ µ 0

and Lorentz generators are block diagonal

 
(iθi + βi)σi
δψ = ψ (130)
(iθi − βi)σi

We can write our spinor as the left and right helicity parts
 
ψL
ψ= (131)
ψR
The Dirac equation is   
− m iσ µDµ ψL
(132)
iσ̄ µD µ − m ψR
meaning
(iσ̄ µ∂ µ + eσ̄ µA µ)ψR = mψL (133)
(iσ µ∂ µ + eσ µA µ)ψL = mψR (134)
So the electron mass mixes the left and right handed states.
In the absence of a mass, this implies
0 = iσ µ∂ µψR = (E + Kσ · Kp )ψR (135)
0 = iσ̄ µ∂ µψL = (E − Kσ · Kp )ψL (136)
So the left and right handed states are eigenstates of the operator Kσ · Kp with opposite eigenvalue.
This operator projects the spin on the momentum direction. We call spin projected on the
direction of motion the helicity, so the left and right handed states have opposite helicity. Since
the mass mixes left and right handed states, massive fermions are not eigenstates of helicity.
This makes ssene – in the electron’s rest frame, there is no direction of motion so helicity is not
well definedl in the massless limit, there is no rest frame, and a left-handed electron can never
turn into a right-handed electron, and a right-handed electron can never turn into a left-handed
electron, even in the presence of electromagnetic fields.
The fact that projection of spin on the direction of momentum is a good quantum number
for massless particles works for massless particles of any spin. For any spin, we will always find
(E ± sJKKp )Ψs = 0, where JK are the rotation generators of spin s. For spin 1/2, JK = σK2 . For pho-
tons, the rotation generators are listed in section 2. For example, Jz = V23 has eigenvalues ± 1
with eigenstates (0, i, 1, 0) and (0, − i, 1, 0). These are the states of circularly polarized light in
the z direction. They are helicity eigenstates. So massless particles always have two helicity
states. It is true for spin 1/2 and spin 1, as we have seen, it is true for gravitons (spin 2),
Rarita-Schwing fields (spin 3/2) and spins s > 2 (although, as we have seen, it is impossible to
have interacting theories with massless fields of spin s > 2).
We have seen that the ψL and ψR states
• do not mix under Lorentz Transformation
• ψL and ψR each have two components on which the σ ′s act. These are the two spin
states of the electron – both left and right handed spinors have 2 spin states.
14 Section 10

• ψL and ψR have opposite helicity.

Using  
ψL
 
† †
ψ̄ = γ0 = ψR ψL† (137)
ψR
The Lagrangian
L = ψ̄ (iγ µ∂ µ + eγ µA µ − m)ψ (138)
becomes
L = iψL† σ̄ µD µψL + iψR
† µ
σ D µψR − m(ψL† ψR + ψR
†
ψL) (139)
Which is what we derived in the beginning. Note that ψL and ψR by themselves must be mass-
less . To right down a mass term, we need both a ψL and a ψR.
It is helpful to be able to project out the left or right handed Weyl spinors from a Dirac
spinor. We can do that with the γ5 matrix
γ 5 = iγ 0 γ 1 γ 2 γ 3 (140)
in the Weyl representation  
5 −1
γ = , (141)
1
So

1 + γ5 1 − γ5
   
0 1
PR = = , PL = = (142)
2 1 2 0

These are projection operators since PR2 = PR and PL2 = PL and

       
ψL 0 ψL ψL
PR = , PL = (143)
ψR ψR ψR 0

These projectors are useful because they are basis-independent.

10 Spin and statistics

Recall that we couldn’t write down a mass term ψL† ψL for just a left handed spinor because
1
δψL = (iθi + βi)σiψL (144)
2
1
δψL† = ( − iθi + βi)ψL† σi (145)
2
δψL† ψL = βiψL† ψL  0 (146)
This just indicates that the boosts are not unitary transformations.
To get something new, recall that for the Pauli matrices, σ1 and σ3 are real, and σ2 is imagi-
nary.      
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 = (147)
1 0 i 0 0 −1
We can see that
σ1⋆ = σ1, σ2⋆ = − σ2, σ3⋆ = σ3 (148)

σ1T = σ1, σ2T = − σ2, σ3T = σ3 (149)

For the i = 1 and i = 3, σiTσ2 = σiσ2 = − σ2σ1. For i = 2, σiTσ2 = − σ2σ2 = − σ2σi. So
σiTσ2 = − σ2σi (150)
Then
T 1 T T 1 T


δ(ψL σ2) = (iθi + βi)ψL σi σ2 = ( − iθi − βi) ψL σ2 σi (151)
2 2
Majorana fermions 15

Which cancels the transformation property of ψL. Thus

T
ψL σ2 ψ L (152)
is Lorentz invariant.  
−i
The only problem is that σ2 = i
so this is just
  
T

i ψ1
ψL σ2 ψL = ψ1 ψ2 = i(ψ1 ψ2 − ψ2 ψ1) (153)
−i ψ2
This is true for any spinor components ψ1 and ψ2. So,

ψ1 ψ2 − ψ2 ψ1 is Lorentz invariant (154)

This is an example of the spin-statistics theorem. I think it’s the simplest way to see the
connection between anti-symmetrization and Lorentz invariance.
You know it from quantum mechanics. Fermions come in anti-symmetric states. It may look
more familiar if we use arrows for the ψ1 and ψ2 states:
1
|ψi = √ (|↑i|↓i − |↓i|↑i) (155)
2
We see that wavefunctions are antisymmetric to the exchange of particles. This is the Pauli
exclusion principle. Here we see that it follows simply from Lorentz invariance.
Obviously if the fermion components commute this is just zero. So fermion components can’t
be regular numbers, they must be anticommuting numbers. Such things are called Grassman
numbers and satisfy a Grassman algebra. We will see that it’s easy to make the quantum fields
anti-commute, by demanding that the creation and annihilation operators anti-commute, a p† a q† =
− a q†a p† ,. And we’ll rederive spin-statistics from properties of the S-matrix. But here we are
seeing that if we are to make any sense of spinors classically, or in any way at all, they must
anticommute.

11 Majorana fermions
If we allow fermions to be Grassman numbers, then we can write down a Lagrangian for a single
Weyl spinor with a mass term
m
L = iψL† σ µ∂ µψL + i (ψL† σ2 ψL
⋆ T
− ψL σ2 ψL) (156)
2
These kinds of mass terms are called Majorana masses. Note that this mass term breaks the
symmetry under ψ → eiαψ, since
T T iα T
ψL σ2 ψL → ψLe σ2eiαψL = e2iαψL σ2 ψL (157)
So a particle with a Majorana mass cannot be charged.
The equation of motion for the Majorana fermion is
⋆
σ µ∂ µψL + mσ2 ψL =0 (158)
which follows from the Lagrangian above.
If we have a Majorana fermion, we can still use the Dirac algebra to describe it, but we have
to put it into a 4-component spinor
 
ψL
ψ= ⋆ (159)
− iσ2 ψL
⋆
This transforms like a Dirac spinor because σ2 ψL transforms like ψR.
Since (in the Weyl basis)
⋆
    
⋆ 0 σ2 ψL ⋆ iσ2 ψR
− iγ2 ψ = − i = ⋆ (160)
− σ2 0 ψR − iσ2 ψL
16 Section 12

We can see that a Majorana fermion is one satisfying

ψ = ψc ≡ − iγ2 ψ ⋆ (161)
We call ψc the charge conjugate fermion. A Majorana fermion is it’s own charge conjugate.
Since it is real, it is als its own antiparticle.
Finally, note that in the Weyl basis γ2 is imaginary and γ0, γ1 and γ3 are real. Of course, we
we could just as well have taken γ3 imaginary and γ2 real, but it’s conventional to pick out γ2.
We can also define a new representation of the γ matrices by γ̂ µ = γ2 γ µ⋆γ2. This satisfies the
Dirac Algebra because γ22 = − 1. Now define
ψc = − iγ2 ψ ⋆ ⇔ ψ ⋆ = − iγ2 ψc (162)
If we take the Dirac equation
(i∂ + e A − m)ψ = 0 (163)
and take the complex conjugate we get
( − iγ µ⋆∂ µ + eγ µ⋆A µ − m)ψ ⋆ = 0 (164)

⇒ γ2( − iγ µ⋆∂ µ + eγ µ⋆A µ − m)γ2 ψc = 0 (165)

⇒ (iγ̂ µ∂ µ − eγ̂ µA µ − m)ψc = 0 (166)

So ψc has the oppisite charge from ψ, which is another reason that Majorana fermions can’t be
charged.

11.1 summary
In summary,
We have seen three types of spinors
• Dirac Spinors: massive, left AND right handed
• Weyl Spinors: massless, left OR right handed
• Majorana spinors: real constrained Dirac spinors.
We are not sure if the neutrino is a Majorana or a Dirac spinor. Weyl spinors play a very
important role in supersymmetry, and in the theory of the Weak interactions. But for QED, we
can just stick with Dirac spinors.

12 Solving the Dirac equation

Let’s take a break from the Dirac equation for a moment, and recall how we discovered antipar-
ticles for complex scalar fields. The Lagrangian was
1
L = [(∂ µ + ieA µ)φ⋆][(∂ µ − ieA µ)φ] + m2 φ⋆ φ (167)
2
L = φ⋆( − ∂ µ + ieA µ)(∂ µ − ieA µ)φ] + m2 φ⋆ φ (168)
The equations of motion are
(∂ µ − ieA µ)2 φ + m2 φ = 0 (169)

(∂ µ + ieA µ)2 φ⋆ + m2 φ⋆ = 0 (170)

So we see that φ and φ⋆ have opposite charge, and we interpret them as particle and antipar-
ticle. Recall that when we quantized the field φ, it created a particle and destroyed and antipar-
ticle, and vice-versa for φ⋆. But at the classical level, we can just think of φ as particle and φ⋆
as antiparticle.
Solving the Dirac equation 17

How do we know if we have a particle or an antiparticle? There is an easy way to answer

this. The free equations for φ and φ⋆ are the same:

( + m2)φ = ( + m2)φ⋆ = 0 (171)

These have plane wave solutions
µ
φ = φ peipµx (172)
In the rest frame, p20 = m2, so p0 = ± m. The solution with p0 = − m is confusing. It is a legiti-
mate solution to the equation of motion, but it says that these particles are going backward in
time! But note that
φ = φ peip0 t ⇔ φ⋆ = φ⋆pe−ip0 t (173)

So we can just as easily interpret these solutions as anti-particles going forward in time. Obvi-
ously this interpretation is easier to swallow, but Feynman spent some time showing that there
really is no physically distinguishable difference.
Now back to spinors. The Dirac equation is

(i∂ + e A − m)ψ = 0 (174)

ψ̄ ( − i ∂ − eA − m) = 0 (175)
So ψ̄ is a particle with mass m and opposite charge to ψ: the positron.

12.1 free field solutions

Let’s expand the Dirac spinor in plane wave solutions
ψ p(x) = u β (p)eipx (176)
u β and v β are the polarizations. With this Ansatz, the Dirac equation becomes
 
− m p µσ µ
u β(p) = 0 (177)
p µσ̄ µ − m

Since the spinor also satisfies ( + m2)ψ = 0, it follows that p2 = m2. We already know that the
negative frequency solutions p0 = − m are not a problem since the antiparticles

ψ̄ = ūe− ipx (178)

have positive frequency. Still, it is somewhat convenient to keep talking about p0 < 0 for now.
For p0 < 0 we call the polarizations v β(p). Then the solutions are

ψ p,s(x) = us(p)eipx , p0 > 0 (179)

ψ p,s(x) = vs(p)e−ipx , p0 < 0 (180)

us and vs are the polarizations and s = 1, 2 labels the spin.
In the rest frame, p = (m, 0, 0, 0), the equations of motion reduce to
   
−1 1 −1 −1
us = 0 and vs = 0 (181)
1 −1 −1 −1
So solutions are constants  
ξs
us = (182)
ξs
and  
ηs
vs = (183)
− ηs
18 Section 12

For any two-component spinors

  
ξ and η. Thus for s = 1 we can take the basis spinor to be ξ1 =
1 0
0
and s = 2 may be ξ2 = 1 .
Now let’s boost in the z-direction. Peskin and Schroeder do the actual boost. But we’ll just
solve the equations again in the boosted frame and match the normalization. If p = (E , 0, 0, pz)
then    
E − pz 0 E + pz 0
p µσ µ = , pµσ̄ µ = (184)
0 E + pz 0 E − pz
√ √
Let x = E − pz and y = E + pz , then m2 = (E − pz)(E + pz)=xy.
Then the Dirac equation is
 
− xy 0 x2 0
 0 − xy 0 y2 
 
(i∂ − m)ψ = u β (p) = 0 (185)
 y2 0 − xy 0 
0 x2 0 − xy
The solutions are  
xξ1
 

x 0
1  yξ2  ξ 
us = u ξ (p) = √  = √1  0 y
y 0  (186)
m yξ1
  m ξ
0 x
xξ2
Check
− xy + x2 y 2 = 0,  (187)
Note also that in the rest frame pz = 0 in the rest frame, x2 = y 2 = m so we are back to the ξ
states.
So the solutions in the pz frame are
 √ !   √ ! 
E − pz 0 E − pz 0
√ ξs  √ ηs 
 0 E + pz  0 E + pz
us(p) = √ ! , vs(p) = √ !  (188)
 E + pz 0   E + pz 0 
√ ξs √ ηs
0 E − pz 0 E − pz
Using
√ ! √ !
√ E − pz 0 √ E + pz 0
p·σ = √ , p · σ̄ = √ (189)
0 E − pz 0 E − pz

We can write more generally

√ ! √ !
p · σ ξs p · σ ηs
us(p) = √ , vs(p) = √ (190)
p · σ̄ ξs − p · σ̄ ηs

where the square root of a matrix means diagonalize it and then take the square root. In prac-
tice, we will always pick p along the z axis, so we don’t really need to know how to make sense
of these formulas. But they are useful because they are explicitly Lorentz covariant so we can
manipulate them cleanly.

12.2 normalization and spin sums

We have chosen a normalization here, which amounts to
√ !  √ !
′ † p · σ ξs † 0 1 p · σ ξs ′
ψ̄ψ = us (p)γ0us ′(p) = √ √ (191)
p · σ̄ ξs 1 0 p · σ̄ ξs ′
s  
E − pz 0 E + pz 0
=2 ξs† ξs ′ (192)
0 E + pz 0 E − pz

= 2mδss ′ (193)
Solving the Dirac equation 19

We taken ξs† ξs ′ = δss ′ by convention. This is the normalization for the inner product .
For m = 0 this way of writing the condition is not useful. Instead we can just set the normal-
ization in the frame of the particle;s direction by normalizing
√ ! √ !
† p · σ ξs † p · σ ξs ′
us us = √ √ = 2Eξs† ξs ′ = 2Eδss ′ (194)
p · σ̄ ξs p · σ̄ ξs ′
√
which has the same 2E factors we’ve been using.
We can also compute the outer product
2
X
us(p)ūs(p) = p + m (195)
s=1
   
1 0
where the sum is over s = 1, 2 corresponding to ξ = 0
and ξ = 1
. For the antiparticles,
2
X
vs(p)v̄s(p) = p − m (196)
s=1
It is also true that
2
X 2
X
usv̄s = ūsvs = 0 (197)
s=1 s=1

You should try to verify these relations on your own.

To keep straight the inner and outer products, it may be helpful to compare to spin 1 parti-
cles. In that case we had for massive spin 1
[ǫiµ(p)]⋆ǫjµ(p) = − δ ij ↔ ūs(p)us ′(p) = 2mδss ′ (inner product) (198)
3 2
X p µpν X
[ǫiµ(p)]⋆ǫiν (p) = η µν − ↔ us(p)ūs(p) = p + m (outer product) (199)
m2
i=1 s=1

So when we sum Lorentz indices or internal spinor indices, we use an inner product and get a
number. When we sum over polarizations/spins, we get a matrix.

12.3 charge conjugation

In the rest frame,    
ξs ηs
us = , vs = (200)
ξs − ηs
where ξ and η are constants, for spin up and spin down.
Let us see how the spin states transform under charge conjugation
        
1 c 1 c 0 i 1 0
|↑ic = = − iσ2 = = = − i|↓i (201)
0 0 −i 0 0 −i
        
c 0 c 0 c 0 i 0 i
|↓i = = − iσ2 = = = i|↑i (202)
1 1 −i 0 1 0
So charge congugation flips the spin. Thus
 
ξ−s
(usp)c = (203)
− ξ−s
This will be relevant to studying charge-conjugation invariance.