First Semester Course

Introduction to relativistic quantum field theory

(a primer for a basic education)

Roberto Soldati

July 5, 2010
Contents

0.1 Prologo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1 Basics in Group Theory 7

1.1 Groups and Group Representations . . . . . . . . 7
1.1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.2 Theorems . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.3 Direct Products of Group Representations . . 12
1.2 Continuous Groups and Lie Groups . . . . . . . . 14
1.2.1 The Continuous Groups . . . . . . . . . . . . . . . 14
1.2.2 The Lie Groups . . . . . . . . . . . . . . . . . . . . 16
1.2.3 An Example : the Rotations Group . . . . . . . 17
1.2.4 The Infinitesimal Operators . . . . . . . . . . . . 19
1.2.5 The Canonical Coordinates . . . . . . . . . . . . 22
1.2.6 The Special Unitary Groups . . . . . . . . . . . . 25
1.3 The Inhomogeneous Lorentz Group . . . . . . . . 31
1.3.1 The Lorentz Group . . . . . . . . . . . . . . . . . . 31
1.3.2 Semisimple Groups . . . . . . . . . . . . . . . . . . 40
1.3.3 The Poincaré Group . . . . . . . . . . . . . . . . . 46

2 The Action Functional 52

2.1 The Classical Relativistic Wave Fields . . . . . . 52
2.1.1 Field Variations . . . . . . . . . . . . . . . . . . . . 52
2.1.2 The Scalar and Vector Fields . . . . . . . . . . . 55
2.1.3 The Spinor Fields . . . . . . . . . . . . . . . . . . . 59
2.2 The Action Principle . . . . . . . . . . . . . . . . . . . 69
2.3 The Noether Theorem . . . . . . . . . . . . . . . . . . 72
2.3.1 Problems . . . . . . . . . . . . . . . . . . . . . . . . 77

1
3 The Scalar Field 79
3.1 Normal Modes Expansion . . . . . . . . . . . . . . . 79
3.2 Quantization of a Klein-Gordon Field . . . . . . . 87
3.3 The Fock Space . . . . . . . . . . . . . . . . . . . . . . . 92
3.4 Special Distributions . . . . . . . . . . . . . . . . . . . 99
3.4.1 Wick Rotation : Euclidean Formulation . . . . 102
3.5 The Generating Functional . . . . . . . . . . . . . . 104
3.5.1 The Symanzik Functional Equation . . . . . . . 104
3.5.2 The Functional Integral . . . . . . . . . . . . . . . 106
3.5.3 The Zeta Function Regularisation . . . . . . . . 109
3.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

4 The Spinor Field 128

4.1 The Dirac Equation . . . . . . . . . . . . . . . . . . . . 128
4.2 Noether Currents . . . . . . . . . . . . . . . . . . . . . . 136
4.3 Quantization of a Dirac Field . . . . . . . . . . . . . 141
4.4 Covariance of the Quantized Spinor Field . . . . 146
4.5 Special Distributions . . . . . . . . . . . . . . . . . . . 150
4.5.1 The Euclidean Fermions . . . . . . . . . . . . . . 153
4.6 The Fermion Generating Functional . . . . . . . . 157
4.6.1 Symanzik Equations for Fermions . . . . . . . . 157
4.6.2 The Integration over Grassmann Variables . . 161
4.6.3 The Functional Integrals for Fermions . . . . . 164
4.7 The C, P and T Transformations . . . . . . . . . 169
4.7.1 The Charge Conjugation . . . . . . . . . . . . . . 169
4.7.2 The Parity Transformation . . . . . . . . . . . . . 171
4.7.3 The Time Reversal . . . . . . . . . . . . . . . . . . 172
4.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

5 The Vector Field 198

5.1 General Covariant Gauges . . . . . . . . . . . . . . . 198
5.2 Normal Modes Decomposition . . . . . . . . . . . . 204
5.2.1 Normal Modes of the Massive Vector Field . . 205
5.2.2 Normal Modes of the Gauge Potential . . . . . 208
5.3 Covariant Canonical Quantization . . . . . . . . . 212
5.3.1 The Massive Vector Field . . . . . . . . . . . . . . 212
5.3.2 The Gauge Vector Potential . . . . . . . . . . . . 218

2
 5.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226

0.1 Prologo
The content of this manuscript is the first semester course in quantum field
theory that I have delivered at the Physics Department of the Faculty of
Sciences of the University of Bologna since the academic year 2007/2008.
It is a pleasure for me to warmly thank all the students and in particular
Dr. Pietro Longhi and Dr. Fabrizio Sgrignuoli for their unvaluable help
in finding mistakes and misprints. In addition it is mandatory to me to
express my gratitude to Dr. Paola Giacconi for her concrete help in reading
my notes and her continuous encouragement. To all of them I express my
deepest thankfullness.

Roberto Soldati

3
Introduction : Some Notations
Here I like to say a few words with respect to the notation adopted. The
components of all the four vectors have been chosen to be real. The metric
is defined by means of the Minkowski’s tensor

 0 for µ 6= ν
gµν = 1 for µ = ν = 0 µ, ν = 0, 1, 2, 3
−1 for µ = ν = 1, 2, 3


i.e. the invariant product of two four vectors a and b with components
a0 , a1 , a2 , a3 and b0 , b1 , b2 , b3 is defined in the following manner

a · b ≡ gµν aµ bν = a0 b0 − a1 b1 − a2 b2 − a3 b3 = a0 b0 − a · b = a0 b0 − ak bk

Summation over repeated indices is understood unless differently stated –

Einstein’s convention – bold type a, b is used to denote ordinary space three–
vectors. Indices representing all four components 0,1,2,3 are usually denoted
by greek letters, while indices representing the three space components 1,2,3
are denoted by latin letters.
The upper indices are, as usual, contravariant while the lower indices are
covariant. Raising and lowering indices are accomplished with the aid of the
metric Minkowski’s tensor , e.g.

aµ = gµν aν aµ = g µν aν gµν = g µν

Throughout the notes the natural system of units is used

c=~=1

for the speed of light and the reduced Planck’s constant as well as the
Heaviside–Lorentz C. G. S. system of electromagnetic units. In this system
of units energy, momentum and mass have the dimensions of a reciprocal
length or a wave number, while the time x0 has the dimensions of a length.
The Coulomb’s potential for a point charge q is
q q α
ϕ(x) = = 2
4π | x | e r

and the fine structure constant is

e2 e2 1
α= = = 7.297 352 568(24) × 10−3 ≈
4π 4π~c 137

4
The symbol −e (e > 0) stands for the negative electron charge. We generally
work with the four dimensional Minkowski’s form of Maxwell’s equation :

ε µνρσ ∂ν Fρσ = 0 ∂µ F µν = J ν

where
A µ = (ϕ, A) F µν = ∂µ Aν − ∂ν Aµ
in which

E = (E 1 , E 2 , E 3 ) , E k = F k0 = F0k , ( k = 1, 2, 3 )

B = (F23 , F31 , F12 )

are the electric and magnetic fields respectively while
∂  ∂ 
∂µ = = , ∇ = (∂ t , ∇ )
∂x µ ∂t
We will often work with the relativistic generalizations of the Schrödinger
wave functions of 1-particle quantum mechanical states. We represent the
energy momentum operators acting on such wave functions following the
convention:
P µ = (P 0 , P) = i ∂ µ = (i∂ t , − i∇)
In so doing the plane wave exp {− i p · x} has momentum p µ = ( p 0 , p) and
the Lorentz covariant coupling between a charged particle of charge q and
the electromagnetic field is provided by the so called minimal substitution

i ∂ t − q ϕ(x) for µ = 0
P µ − q A µ (x) =
i ∇ − q A(x) for µ = 1, 2, 3

The Pauli spin matrices are the three hermitean 2 × 2 matrices

     
0 1 0 −i  1 0
σ1 =  σ2 =  σ3 = 
 
  
 

1 0 i 0 0 −1
     

which satisfy
σi σj = δ ij + i ε ijk σk
hence
1 1
[ σi , σj ] = i ε ijk σk {σi , σj } = δ ij
2 2

5
The Dirac matrices in the Weyl, or spinorial, or even chiral representation
are given by
   
0 0 1 k 0 σ k
γ = γ = ( k = 1, 2, 3 )
 
  

1 0 − σk 0
   

with the hermiticity property

γ µ† = γ 0 γ µ γ 0 {γ µ , γ ν } = 2 g µν I

6
Chapter 1

Basics in Group Theory

1.1 Groups and Group Representations

1.1.1 Definitions
A set G of elements e, f, g, h, . . . ∈ G that satisfies all the four conditions
listed below is called an abstract group or simply a group.
1. A law of composition, or multiplication, is defined for the set, such that
the multiplication of each pair of elements f and g gives an element h
of the set : this is written as

fg = h

Element h is called the product of the elements f and g , which are

called the factors. In general the product of two factors depends upon
the order of the factors, so that the elements f g and gf can be different.

2. The multiplication is associative : if f , g and h are any three elements,

the product of the element f with gh must be equal to the product of
the element f g with h
f (gh) = (f g)h

3. The set G contains a unit element e called the identity giving the
relation
ef = f e = f ∀f ∈ G

4. For any element f ∈ G there is an element f −1 ∈ G called the inverse

or reciprocal of f such that

f −1 f = f f −1 = e ∀f ∈ G

7
 If the number of elements in G is finite, then the group is said to be
finite, otherwise the group is called infinite. The number of elements in a
finite group is named its order.
If the multiplication is commutative, i.e. , if for any pair of elements f
and g we have f g = gf , then the group is said to be commutative or abelian.
Any subset of a group G , forming a group relative to the very same law
of multiplication, is called a subgroup of G .
The one-to-one correspondence between the elements of two groups F
and G
f ↔ g f ∈F g∈G
is said to be an isomorphism iff for any pair of relations

f1 ↔ g1 f 2 ↔ g2 f1 , f2 ∈ F g1 , g2 ∈ G

then there follows the relation

f1 f2 ↔ g1 g2

Groups between the elements of which an isomorphism can be established

are called isomorphic groups. As an example, consider the set of the n−th
roots of unity in the complex plane

z k = exp{ 2πi k/n} ( k = 0, 1, 2, . . . , n − 1, n ∈ N)

which form the commutative group

Z n ≡ z k ∈ C | z kn = z0 = e 2πi ,  ∈ Z , k = 0, 1, 2, . . . , n − 1 , n ∈ N


where the composition law is the multiplication

z k · z h = z k+h = z h+k ∀ k, h = 0, 1, 2, . . . , n − 1

the identity is z0 = e 2πi (  ∈ Z ) and the inverse z k− 1 = z̄ k is the complex

conjugate. This finite group with n elements is isomorphic to the group of
the counterclockwise rotations around the OZ axis (mod 2π), for instance,
through the n angles
2πk
ϕk = ( k = 0, 1, 2, . . . , n − 1, n ∈ N)
n
in such a manner that
 
 cos ϕ k sin ϕ k
Rk =  ( k = 0, 1, 2, . . . , n − 1, n ∈ N)


− sin ϕ k cos ϕ k
 

8
The isomorphism of these groups follow from the correspondence

ϕk ↔ zk = e iϕk ( k = 0, 1, 2, . . . , n − 1, n ∈ N)

Homomorphism between two groups differs from isomorphism only by the

absence of the requirement of one-to-one correspondence : thus isomorphism
is a particular case of homomorphism. For example, the group S3 of the
permutations of three objects is homomorphic to Z 2 = {1, − 1} , the law of
composition being multiplication. The following relationships establish the
homomorphism of the two groups : namely,
     
1 2 3 1 2 3 1 2 3
→1 →1 →1
1 3 2 2 3 1 3 1 2
     
1 2 3 1 2 3 1 2 3
→ −1 → −1 → −1
2 1 3 1 3 2 3 2 1

The results of one branch of the theory of groups, namely the theory of
group representations, are used in the overwhelming majority of important
cases in which group theory is applied to Physics. The theory of the group
representations studies the homomorphic mappings of an arbitrary abstract
group on all possible groups of linear operators.

• We shall say that a representation T of a group G is given in a certain

linear space L , iff to each element g ∈ G there is a corresponding linear
operator T (g) acting in the space L , such that to each product of the
elements of the group there is a corresponding product of the linear
operators, i.e.

T (g1 )T (g2 ) = T (g1 g2 ) ∀g1 , g2 ∈ G

T (e) = I T (g −1 ) = T −1 (g)
where I denotes the identity operator on L . The dimensionality of the
space L is said to be the dimensionality of the representation. A group
can have representations both of a finite and of an infinite number of
dimensions. By the very definition, the set of all linear operators T (g) :
L → L ( g ∈ G ) is closed under the multiplication or composition
law. Hence it will realize an algebra of linear operators over L that will
be denoted by A(L) . If the mapping T : G → A(L) is an isomorphism,
then the representation T is said faithful. In what follows, keeping
in mind the utmost relevant applications in Physics, we shall always
assume that the linear spaces upon which the representations act are

9
 equipped by an inner product 1 , in such a manner that the concepts
of orthonormality, adjointness and unitarity are well defined in the
conventional way.
One of the problems in the theory of representations is to classify all the
possible representations of a given group. In the study of this problem
two concepts play a fundamental role : the concept of equivalence of
representations and the concept of reducibility of representations.

• Knowing any representation T of a group G in the space L one can

easily set up any number of new representations of the group. For
this purpose, let us select any nonsingular invertible linear operator
A , carrying vectors from L into a space L 0 with an equal number d of
dimensions, and assign to each element g ∈ G the linear operator

TA (g) = A T (g)A −1 ∀g ∈ G

acting in the vector space L 0 . It can be readily verified that the map
g 7→ TA (g) is a representation of the group G that will be thereby said
equivalent to the representation T (g) .
All representations equivalent to a given one are equivalent to each
other. Hence, all the representations of a given group split into classes
of mutually equivalent ones. Accordingly, the problem of classifying all
representations of a group is reduced to the more limited one of finding
all mutually inequivalent representations.

• Consider a subspace L1 of the linear vector space L . The subspace

L1 ⊆ L will be said an invariant subspace with respect to some given
linear operator A acting on L iff

A`1 ∈ L1 ∀`1 ∈ L1

Of course L itself and the empty set ∅ are trivial invariant subspaces.

• The representation T of the group G in the vector space L is said

to be reducible iff there exists in L at least one nontrivial subspace
L1 invariant with respect to all operators T (g) , g ∈ G . Otherwise the
representation is called irreducible. All one dimensional representations
are evidently irreducible.
1
The inner product between to vectors a, b ∈ L is usually denoted by (a , b) in the
mathematical literature and by the Dirac notation h a | b i in quantum physics. Here I
shall often employ both notations without loss of clarity.

10
• A representation T of the group G in the vector space L is said to be
unitary iff all the linear operators T (g) , g ∈ G , are unitary operators

T † (g) = T −1 (g) = T (g −1 ) ∀g ∈ G

1.1.2 Theorems
There are two important theorems concerning unitary representations.

• Theorem 1. Let T a unitary reducible representation of the group G

in the vector space L and let L1 ⊆ L an invariant subspace. Then the
subspace L2 = {L1 , the orthogonal complement of L1 , is also invariant.
Proof.
Let `1 ∈ L1 , `2 ∈ L2 , then T −1 (g)`1 ∈ L1 and ( T −1 (g)`1 , `2 ) = 0 . On
the other hand the unitarity of the representation T actually implies
( `1 , T (g)`2 ) = 0 . Hence, T (g)`2 is orthogonal to `1 and consequently
T (g)`2 ∈ L2 , ∀g ∈ G , so the theorem is proved. 
Hence, if the vector space L transforms according to a unitary reducible
representation, it decomposes into two mutually orthogonal invariant
subspaces L1 and L2 such that L = L1 ⊕ L2 . Iterating this process we
inevitably arrive at the irreducible representations.

• Theorem 2. Each reducible unitary representation T (g) of a group G

on a vector space L decomposes, univocally up to equivalence, into the
direct sum of irreducible unitary representations τ a (g) , a = 1, 2, 3, . . . ,
acting on the invariant vector spaces La ⊆ L in such away that
M
L = L1 ⊕ L2 ⊕ L3 ⊕ . . . = La La ⊥ Lb for a 6= b
a

τ a (g) `a ∈ La ∀a = 1, 2, . . . ∀g ∈ G
Conversely, each reducible unitary representation T (g) of a group G
can be always composed from the irreducible unitary representations
τ a (g) , a = 1, 2, 3, . . . , of the group.
The significance of this theorem lies in the fact that it reduces the
problem of classifying all the unitary representations of a group G ,
up to equivalent representations, to that of finding all its irreducible
unitary representations.

11
As an example of the decomposition of the unitary representations of the
rotation group, we recall the decomposition of the orbital angular momentum
which is well known from quantum mechanics. The latter is characterized
by an integer ` = 0, 1, 2, . . . and consist of (2` + 1) × (2` + 1) square matrices
acting on quantum states of the system with given eigenvalues λ` = ~2 `(`+1)
of the orbital angular momentum operator L2 = [ r × (− i~∇) ]2 . The latter
are labelled by the possible values of the projections of the orbital angular
momentum along a certain axis, e.g. Lz = −~`, −~(` − 1), . . . , ~(` − 1), ~` ,
in such a manner that we have the spectral decomposition
∞
X `
X
2 2
L = ~ `(` + 1) Pb` Pb` = | ` mih` m |
`=0 m=− `

with
h ` m | ` 0 m 0 i = δ `` 0 δ mm 0 tr Pb` = 2` + 1
where [ tr ] denotes the trace (sum over the diagonal matrix elements) of the
projectors Pb` over the finite dimensional spaces L` (` = 0, 1, 2, . . .) spanned by
the basis { | ` mi | m = −`, −` + 1, . . . , ` − 1, ` } of the common eigenstates of
L2 and Lz . Thus, for each rotation g , which corresponds to a 3×3 orthogonal
square matrix such that g −1 = g > , there exists a (2` + 1) × (2` + 1) unitary
matrix τ ` (g) that specifies how the (2` + 1) quantum states transform among
themselves as a result of the rotation g ∈ G and which actually realize all the
irreducible unitary finite dimensional representations of the rotation group
in the three dimensional space. The Hilbert space H for a point-like spinless
particle is thereby decomposed according to
∞
M
H= L` L` ⊥ Lm for ` 6= m dim (L` ) = 2` + 1
`=0

τ ` : L` → L` ` = 0, 1, 2, . . .

1.1.3 Direct Products of Group Representations

It is very useful to introduce the concepts of characters and the definition of
product of group representations.

• Let T (g) any linear operator corresponding to a given representation

of the group G . We define the characters χ(g) of the representation to
be the sum of the diagonal matrix elements of T (g)

χ(g) ≡ tr T (g) ∀g ∈ G (1.1)

12
 In the case of infinite dimensional representations we have to suppose
the linear operators T (g) to be of the trace class.
Two equivalent representations have the same characters, as the trace
operation does not depend upon the choice of the basis in the vector
space.
• Definition. Consider two representations T1 (g) and T2 (g) of the group
G acting on the vector (Hilbert) spaces L1 and L2 of dimensions n1 and
n2 respectively. Let
{e1j ∈ L1 | j = 1, 2, . . . , n1 } {e2r ∈ L2 | r = 1, 2, . . . , n2 }
any two bases so that
e1j 2r ≡ {e1j ⊗ e2r | j = 1, 2, . . . , n1 , r = 1, 2, . . . , n2 }
is a basis in the tensor product L1 ⊗ L2 of the two vector spaces with
dim(L1 ⊗L2 ) = n1 n2 . Then the matrix elements of the linear operators
T1 (g) and T2 (g) with respect to the above bases will be denoted by
[ T1 (g) ]jk ≡ h e1j , T1 (g) e1k i [ T2 (g) ]rs ≡ h e2r , T2 (g) e2s i
The direct product
T (g) ≡ T1 (g) × T2 (g) ∀g ∈ G
of the two representations is a representation of dimension n = n1 n2
the linear operators of which, acting upon the tensor product vector
space L1 ⊗ L2 , have the matrix elements, with respect to the basis
e1j 2r , which are defined by
( e1j 2r , T (g) e1k 2s ) ≡ [ T1 (g) ]jk [ T2 (g) ] rs
= ( e1j , T1 (g) e1k ) ( e2r , T2 (g) e2s ) (1.2)
where j, k = 1, 2, . . . , n1 and r, s = 1, 2, . . . , n2 . From the definition
(1.1) it is clear that we have
χ(g) = χ1 (g) χ2 (g) ∀g ∈ G
because
n1 X
X n2
χ(g) ≡ ( e1j 2r , T (g) e1j 2r )
j=1 r=1
Xn1 n2
X
= ( e1j , T1 (g) e1j ) ( e2r , T2 (g) e2r )
j=1 r=1
= χ1 (g) χ2 (g) ∀g ∈ G (1.3)

13
1.2 Continuous Groups and Lie Groups
1.2.1 The Continuous Groups
A group G is called continuous if the set of its elements forms a topological
space. This means that each element g ∈ G is put in correspondence with
an infinite number of subsets Ug ⊂ G, called the neighbourhoods of any
g ∈ G . This correspondence has to satisfy certain conditions that guarantee
the full compatibility between the topological space structure and the group
associative composition law - see the excellent monographies [5] for details.
To illustrate the concept of neighbourhood we consider the group of the
rotations around a fixed axis, which is an abelian continuous group. Let
g = R(ϕ) , 0 ≤ ϕ ≤ 2π , a rotation through an angle ϕ around e.g. the
OZ axis. By choosing arbitrarily a positive number ε > 0 , we consider the
set Ug (ε) consisting of all the rotations g 0 = R(ϕ 0 ) satisfying the inequality
| ϕ − ϕ 0 | < ε . Every such set Ug (ε) is a neighbourhood of the rotation
g = R(ϕ) and giving ε all its possible real positive values we obtain the
infinite manifold of the neighbourhoods of the rotation g = R(ϕ) .
The real functions f : G → R over the group G is said to be continuous
for the element g0 ∈ G if, for every positive number δ > 0 , there exists such
a neighborhood U0 of g0 that ∀g ∈ U0 the following inequality is satisfied
| f (g) − f (g0 ) | < δ ∀g ∈ U0 g0 ∈ U0 ⊂ G
A continuous group G is called compact if and only if each real function
f (g) , continuous for all the elements g ∈ G of the group, is bounded. For
example, the group of the rotations around a fixed axis is compact, the
rotation group in the three dimensional space is also a compact group. On
the other hand, the continuous abelian group R of all the real numbers is not
compact, since there exist continuous although not bounded functions, e.g.
f (x) = x , x ∈ R . The Lorentz group is not compact.
A continuous group G is called locally compact if and only if each real
function f (g) , continuous for all the elements g ∈ G of the group, is bounded
in every neighborhood U ⊂ G of the element g ∈ G . According to this
definition, the group of all the real numbers is a locally compact group and
the Lorentz group is also a locally compact group.
• Theorem : if a group G is locally compact, it always admits irreducible
unitary representations in infinite dimensional Hilbert spaces.
In accordance with this important theorem, proved by Gel’fand and Raikov,
Gel’fand and Naı̈mark found all the unitary irreducible representations of
the Lorentz group and of certain other locally compact groups.

14
In general, if we consider all possible continuous functions defined over a
continuous group G , we may find among them some multi-valued functions.
These continuous multi-valued functions can not be made single-valued by
brute force without violating continuity, that is, by rejecting the superflous
values for each element g ∈ G . As an example we have the function
1
f (ϕ) = e 2 iϕ
over the rotation group around a fixed axis. Since each rotation g = R(ϕ)
through an angle ϕ can also be considered as a rotation through an angle
ϕ + 2π , this function must have two values for a rotation through the same
angles : namely,
1 1 1
f+ (ϕ) = e 2 iϕ f− (ϕ) = e 2 iϕ+iπ = − e 2 iϕ = − f+ (ϕ)
Had we rejected the second of these two values, then the function f (ϕ) would
become discontinuous at the point ϕ = 0 = 2π .
The continuous groups which admit continuous multi-valued functions
are called multiply connected. As the above example shows, it turns out
that the rotation group around a fixed axis is multiply connected. Also the
rotation group in the three dimensional space is multiply-connected, as I
will show below in some detail. The presence of multi-valued continuous
functions in certain continuous groups leads us to expect that some of the
continuous representations of these groups will be multi-valued. On the one
hand, these multi-valued representations can not be ignored just because of
its importance in many physical applications. On the other hand, it is not
always possible, in general, to apply to those ones all the theorems valid for
single-valued continuous representations.
To overcome this difficulty, we use the fact that every multiply connected
group G is an homomorphic image of a certain simply connected group G. e
It turn out that the simply connected group G e can always be chosen in
such a manner that none of its simply connected subgroups would have the
group G as its homomorphic image. When the simply connected group G e is
selected in this way, it is called the universal covering group or the universal
enveloping group of the multiply connected group G – see [5] for the proof.
Consider once again, as an example, the multiply connected abelian group
of the rotations around a fixed axis. The universal covering group for it is
the simply connected commutative group R of all the real numbers. The
homomorphism is provided by the relationship 2
x → ϕ = x − 2π [ x/2π ] −∞<x<∞ 0 ≤ ϕ ≤ 2π
2
We recall that any real number x can always be uniquely decomposed into its integer
[ x ] and fractional {x} parts respectively.

15
where x = [ x ] + {x} . It turns out that every continuous representation of
the group G , including any multi-valued one, can always be considered as a
single-valued continuous representation of the universal enveloping group G
e.
The representations of the group G obtained in this manner do exhaust all
the continuous representations of the group G .

1.2.2 The Lie Groups

The Lie groups occupy an important place among continuous groups.
Marius Sophus Lie
Nordfjordeid (Norway) 17.12.1842 – Oslo 18.02.1899
Vorlesungen über continuierliche Gruppen (1893)
Lie groups occupy this special place for two reasons : first, they represent
a sufficiently wide class of groups, including the most important continuous
groups encountered in geometry, mathematical analysis and physics ; second,
every Lie group satisfies a whole series of strict requirements which makes
it possible to apply to its study the methods of the theory of differential
equations. We can define a Lie group as follows.
• Let G some continuous group. Consider any neighborhood V of the unit
element of this group. We assume that by means of n real parameters
α1 , α2 , . . . , αn we can define every element of the neighborhood V in
such a way that :
1. there is a continuous one-to-one correspondence between all the
different elements g ∈ V and all the different n−ples of the real
parameters α1 , α2 , . . . , αn ;
2. suppose that g1 , g2 and g = g1 g2 lie the neighborhood V and that

g1 = g1 (α10 , α20 , . . . , αn0 ) g2 = g2 (α10 0 , α20 0 , . . . , αn0 0 )

g = g (α1 , α2 , . . . , αn )
where

αa = αa (α10 , α20 , . . . , αn0 ; α10 0 , α20 0 , . . . , αn0 0 ) a = 1, 2, . . . , n

then the functions αa (a = 1, 2, . . . , n) are analytic functions of

the parameters αb0 , αc0 0 ( b, c = 1, 2, . . . , n) of the factors.
Then the continuous group G is called a Lie group of dimensions n.
We shall always choose the real parameters αa (a = 1, 2, . . . , n) in such
a way that their zero values correspond to the unit element.

16
 1.2.3 An Example : the Rotations Group
• The proper rotation group SO(3) .
Any rotation in the three dimensional space can always be described
by an oriented unit vector n̂ with origin in the center of rotation and
directed along the axis of rotation and by the angle of rotation α .
Therefore we can denote rotations by g(n̂, α) . The angle is measured
in the counterclockwise sense with respect to the positive direction of
n̂ . By elementary geometry it can be shown that any active rotation
g(n̂, α) transforms the position vector r into the vector

r 0 = r cos α + n̂ (r · n̂)(1 − cos α) + n̂ × r sin α

Every rotation is defined by three parameters. If we introduce the

vector α ≡ α n̂ (0 ≤ α < 2π), then we can take the projections of the
vector α on the coordinate axes as the three numbers α = (α1 , α2 , α3 )
that parametrize the rotation : namely

α1 = α sin θ cos φ α2 = α sin θ sin φ α3 = α cos θ

0 ≤ α < 2π 0≤θ≤π 0 ≤ φ ≤ 2π (1.4)

These angular parameters are called the canonical coordinates of the

rotation group and are evidently restricted to lie inside the 2–sphere

α12 + α22 + α32 < (2π)2

Thus, the rotation group is a three dimensional Lie group.

Any (abstract) rotation g can be realized by means of a linear operator
R(g) that transforms the position vector r into the new position vector
r 0 = R(g) r . This corresponds to the active point of view, in which the
reference frame is kept fixed while the position vectors are moved. Of
course, one can equivalently consider the passive point of view, in which
the position vectors are kept fixed while the reference frame is changed.
As a result, the linear operators corresponding to the two points of
view are inverse one of each other. Historically, the eulerian angles ϕ, θ
and ψ were firstly employed as parameters for describing rotations. A
rotation can be represented as a product of three orthogonal matrices:
the rotation matrix R 3 (ϕ) about the OZ axis, the rotation matrix R1 (θ)
about the OX 0 axis, which is called the nodal line, and the rotation
matrix R 3 (ψ) about the OZ 0 axis, i.e.

R(g) = R(ϕ, θ, ψ) = R 3 (ψ) R1 (θ) R 3 (ϕ)

17
 We have
cos ϕ − sin ϕ 0 
   
  1 0 0 
0 cos θ − sin θ
   
R 3 (ϕ) = 

 sin ϕ cos ϕ 0 
 R1 (θ) = 





   
0 0 1 0 sin θ cos θ

cos ψ − sin ψ 0 
 

 
R 3 (ψ) = 

 sin ψ cos ψ 0 

 
0 0 1
From the parametrization in terms of the eulerian angles it immediately
follows that every rotation matrix is univocally identified by a tern

g ↔ R( ϕ, θ, ψ ) 0 ≤ ϕ < 2π ; 0 ≤ θ ≤ π ; 0 ≤ ψ < 2π (1.5)

and that

R (g −1 ) = R −1 (g) = R 3 (−ϕ) R1 (−θ) R 3 (−ψ) = R > (g)

with det [ R( ϕ, θ, ψ ) ] = 1 . It follows thereby that an isomorphism exists

between the abstract rotation group and the group of the orthogonal
3 × 3 matrices with unit determinant : this matrix group is called the
3–dimensional special orthogonal group and is denoted by SO(3) .
The discrete transformation that carries every position vector r into the
vector − r is called inversion or parity transform. Inversion I commutes
with all rotations Ig = gI , ∀g ∈ SO(3) ; moreover I 2 = e , det I = −1 .
If we add to the elements of the rotation group all possible products
Ig , g ∈ SO(3) we obtain a group, as it can be readily checked.
This matrix group is called the 3–dimensional full orthogonal group
and is denoted by O(3) . The group O(3) splits into two connected
components : the proper rotation group O(3)+ = SO(3) , which is
the subgroup connected with the unit element, and the the component
O(3)− connected with the inversion, wich is not a group. Evidently the
matrices belonging to O(3)± have determinant equal to ±1 respectively.

• The full orthogonal group O(3) is the group of transformations that

leave invariant the line element

d r2 = d x2 + d y 2 + d z 2 = d xi d xi i = 1, 2, 3

in the 3–dimensional space. In a quite analogous way one introduces,

for odd N , the N −dimensional full orthogonal groups O(N ) and the
proper orthogonal Lie groups SO(N ) of dimensions n = 12 N (N − 1) .

18
 1.2.4 The Infinitesimal Operators
• Infinitesimal operators : in the following we shall consider only those
representations T of a Lie group G , the linear operators of which are
analytic funtions of the parameters α1 , α2 , . . . , αn in such a way that

T (g) = T (α1 , α2 , . . . , αn ) ≡ T (α) : V → A(V ) (V ⊂ G)

are analytic functions of their arguments.

The derivative of the operator T (g) with respect to the parameter αa
taken for g = e , i.e. at the values α1 = α2 = . . . = αn = 0 , is called
the infinitesimal operator or generator Ia of the representation T (g)
corresponding to the parameter αa .
Thus, any representation T (g) has n generators, i.e. , the number of the
infinitesimal operators is equal to the dimensionality of the Lie group.

• As an example let us consider the orthogonal matrices referred to the

canonical coordinates (1.4) of the rotation group. Since the orthogonal
matrices R ∈ SO(3) are linear operators acting on the three–vectors
r ∈ L = R3 , it follows that the orthogonal matrices do indeed realize
a representation of the abstract rotation group with the same number
of dimensions as the group itself.
Thereby, the orthogonal matrices do realize the so–called vector or
adjoint representation τ A (g) of the rotation group.
According to the passive point of view, the rotation g with the canonical
coordinates (α1 , 0, 0) corresponds to the rotation of the reference frame
around the positive direction of the OX axis through an angle α1 . Then
the operator τ A (α1 , 0, 0) is represented by the matrix
 
 1 0 0 
 
τ A (α1 , 0, 0) = 

 0 cos α1 sin α1  

0 − sin α1 cos α1
 

Hence  
0 0 0
∂  
−
 
τ A (α1 , 0, 0) = 
 0 sin α1 cos α1 

∂α1  
0 − cos α1 − sin α1
 

and consequently  
 0 0 0 
 
I1 = 

 0 0 1 

0 −1 0
 

19
 In a quite analogous way we obtain

0 0 −1 
   
  0 1 0 
−1 0 0 
   
I2 = 

 0 0 0  
 I3 = 

 

   
1 0 0 0 0 0

The following commutation relations among the infinitesimal operators

of the rotation group SO(3) hold true : namely,

Ia Ib − Ib Ia = − εabc Ic (a, b, c = 1, 2, 3) (1.6)

where εabc is the completely antisymmetric Levi–Civita symbol with

the normalization ε123 = 1 .
The crucial role played by the generators in the theory of the Lie groups
is unraveled by the following three theorems :

• Theorem I . Let T1 and T2 any two representations of a Lie group

G acting on the same vector space L and suppose that they have the
same set of generators. Then T1 and T2 are the same representation.

• Theorem II . The infinitesimal operators Ia (a = 1, 2, . . . , n) that

correspond to any representation T (g) of a Lie group G do satisfy the
following commutation relations

Ia Ib − Ib Ia ≡ [ Ia Ib ] = Cabc Ic a, b, c = 1, 2, . . . , n (1.7)

where the constant coefficients Cabc = − Cbac do not depend upon the
choice of the representation T (g) . The constant coefficients Cabc of a
Lie algebra fulfill the Jacobi’s identity

Cabe Cecd + Cbce Cead + Ccae Cebd = 0 (1.8)

because of the identity

[ [ Ia Ib ] Ic ] + [ [ Ib Ic ] Ia ] + [ [ Ic Ia ] Ib ] = 0

which can be readily checked by direct inspection.

• Theorem III . Suppose that a set of linear operators A1 , A2 , . . . , An

acting on a certain vector space L is given and which fulfill the same
commutation relations

[ Aa Ab ] = Cabc Ac a, b, c = 1, 2, . . . , n

20
 as the infinitesimal operators of the group G . Then the operators
Aa (a = 1, 2, . . . , n) are the generators of a certain representation T (g)
of the group G in the vector space L .
The details of the proofs of the above three fundamental theorems of
the theory of the Lie groups can be found in the excellent monographies
[5]. Theorems I, II and III are very important because they reduce the
problem of finding all the representations of a Lie group G to that
of classifying all the possible sets of linear operators which satisfy the
commutation relations (1.7).
• Lie algebra : the infinitesimal operators Ia (a = 1, 2, . . . , n) do generate
a linear space and the commutators define a product law within this
linear space. Then we have an algebra which is called the Lie algebra G
of the Lie group G with dim G = dim G = n . The constant coefficients
Cabc = − Cbac are named the structure constants of the Lie algebra G .
Suppose that the representation T (g) of the Lie group G acts on the
linear space L and let A any invertible linear operator upon L . Then the
linear operators A Ia A−1 ≡ Ja (a = 1, 2, . . . , n) do realize an equivalent
representation for G , i.e. the infinitesimal operators Ja (a = 1, 2, . . . , n)
correspond to a new basis in G because
[ Ja Jb ] = [ A Ia A−1 A Ib A−1 ] = A [ Ia Ib ] A−1
= A Cabc Ic A−1 = Cabc A Ic A−1
= Cabc Jc a, b, c = 1, 2, . . . , n

As a corollary, it turns out that, for any representation, the collections

of infinitesimal operators
Ia (g) ≡ T (g) Ia T −1 (g) | a = 1, 2, . . . , n

∀g ∈ G (1.9)
do indeed realize different and equivalent choices of basis in the Lie
algebra. We will say that the operators T (g) of a given representation
of the Lie group G generate all the inner automorphisms in the given
representation of the Lie algebra G .
• It is possible to regard the structure constants as the matrix elements
of the n–dimensional representation of the generators : namely,
k Ia k bc = Cacb (a, b, c = 1, 2, . . . , n) (1.10)
in such a way that we can rewrite the Jacobi identity (1.8) as the matrix
identity
− || Ic || de || Ia || eb + || Ia || de || Ic || eb − || Ic || ea || Ie || db = 0 (1.11)

21
 or after relabeling of the indices

|| [ Ia Ib ] || de = Cabc || Ic || de

Thus, for each Lie algebra G and Lie group G of dimensions n , there is
a representation, called the adjoint representation, which has the very
same dimensions n as the Lie group itself.
Evidently, an abelian Lie group has vanishing structure constants, so
that its adjoint representation is trivial, i.e. it consists of solely the unit
element.
As shown above the rotation group has three generators Ia (a = 1, 2, 3)
and structure constants Cabc equal to − εabc . Consequently, the adjoint
representation of the rotation group is a three dimensional one with
generators Ia given by

|| Ia || bc = εabc (a, b, c = 1, 2, 3)

In any neighborhood of the identity operator T (e) = T (0, 0, . . . , 0) ≡ 1

we can always find a set of parameters such that
n
X
T (α) = exp {Ia αa } = 1 + Ia αa + O(α2 ) (1.12)
a=1

Thus, in order to find the representations of the group G it is sufficient

to classify the representations of its Lie algebra G . More precisely, one
can prove the following theorem – see [5] for the proof.

1.2.5 The Canonical Coordinates

• Exponential representation : let Ia (a = 1, 2, . . . , n) a given base of the
Lie algebra G of a Lie group G . Then, for any neighborhood U ⊂ G
of the unit element e ∈ G there exists a set of canonical coordinates
αa (a = 1, 2, . . . , n) in U and a positive number δ > 0 such that

T (g) ≡ T (α1 , α2 , . . . , αn ) = exp {Ia αa } g ∈ U ⊂ G , | αa | < δ

where the exponential is understood by means of the so called Baker–

Campbell–Hausdorff formula

T (α) T (β) = T (γ) T (α) = exp {Ia αa } T (β) = exp {Ib βb }

22
 T (γ) ≡ exp Ia αa + Ib βb + 21 αa β b [ Ia Ib ]

1

+ 12 (αa αb βc + βa βb αc ) [ Ia [ Ib Ic ] ] + · · ·

in which the dots stand for higher order, iterated commutators among
generators. To elucidate these notions let me discuss two examples.

1. Consider the translations group along the real line

x → x0 = x + a (∀x ∈ R, a ∈ R)

The translations group on the real line is a one dimensional abelian

Lie group. Let us find the representation of this group acting on
the infinite dimensional functional space of all the real analytic
functions f : R → R with f ∈ C∞ (R) . To do this, we have to
recall the Taylor-McLaurin formula
∞
X 1 k (k)
f (x + a) = a f (x)
k=0
k!

Now, if we define

df d kf
T f (x) ≡ T k f (x) ≡ = f (k) (x)
dx dx k
then we can write

f (x + a) = exp{ a · T } f (x)
∞
! ∞
X 1 k k X 1 k (k)
= a T f (x) = a f (x)
k=0
k ! k=0
k !

in such a manner that we can actually identify the generator of

the infinite dimensional representation of the translations group
with the derivative operator acting upon the functional space of
all the real analytic functions.
2. As a second example, consider the rotations group on the plane
around a fixed point, which is an abelian and compact Lie group
O(2, R) i.e. the two dimensional real orthogonal group. A generic
group element g ∈ O(2) is provided by the 2×2 orthogonal matrix
 
cos ϕ sin ϕ 
g(ϕ) =  ( 0 ≤ ϕ ≤ 2π )

 
− sin ϕ cos ϕ


23
 which corresponds to a passive planar rotation around the origin.
If we introduce the generator
 
0 1
t≡
 

−1 0
 

with t2 = − I , then we readily find

∞
X 1 k k
g(ϕ) ≡ exp{ t ϕ } = ϕ t
k=0
k!
∞  2k
ϕ 2k+1

X
k ϕ
= (−1) I + t
k=0
(2k) ! (2k + 1) !
= I cos ϕ + t sin ϕ ∀ ϕ ∈ [ 0 , 2π ]

The exponential representation of the Lie group elements implies some

remarkable features.

• The generators of a unitary representation are antihermitean

T (g −1 ) = T † (g) ⇔ Ia = − Ia† (a = 1, 2, . . . , n)

• All the structure constants of an abelian Lie group are equal to zero.

• For a compact Lie group G with dim G = n it is possible to show

that all the sets of canonical coordinates span a bounded subset of
Rn . Moreover any element of any representation of a compact group
can always be expressed in the exponential form. In a non-compact
Lie group G with dim G = n the canonical coordinates run over an
unbounded subset of Rn and, in general, not all the group elements
can be always expressed in the exponential form.

• Any two Lie groups G1 and G2 with the same structure constants are
locally homeomorphic, in the sense that it is always possible to find two
neighborhoods of the unit elements U1 ⊂ G1 and U2 ⊂ G2 such that
there is an analytic isomorphism f : U1 ↔ U2 between the elements
of the two groups ∀ g1 ∈ U1 , ∀ g2 ∈ U2 . Of course, this does not imply
that there is a one–to–one analytic map over the whole parameter space,
i.e. the two groups are not necessarily globally homeomorphic. As an
example, consider the group R of all the real numbers and the unitary
group
U (1) ≡ {z ∈ C | z z̄ = 1}

24
 of the complex unimodular numbers. These two groups are abelian
groups. In a neighborhood of the unit element we can readily set up
an homeomorphic map, e.g. the exponential map

z = exp{iα} z̄z = 1 −π <α≤π

so that the angle α is the canonical coordinate. However, there is no

such mapping in the global sense, because the unit circle in the complex
plane is equivalent to a real line of which all the elements modulo 2π
are considered as identical. In other words, we can associate any real
number x = 2kπ + α (k ∈ Z , −π < α ≤ π) to some point of the
unit circle in the complex plane with a given canonical coordinate α
and a given integer winding number k = [ x/2π ] . If this is the case,
the complex unit circle is not simply connected, because all the closed
paths on a circle have a nonvanishing integer number of windings and
can not be continuously deformed to a point, at variance with regard
to the simply connected real line.

1.2.6 The Special Unitary Groups

• Example : the unitary group SU (2) . Consider the group of the 2 × 2
complex unitary matrices with unit determinant denoted as SU (2) ,
Special Unitary 2-dimensional matrices
 
v̄ −ū 
g= ūu + v̄v = 1

 
u v


which evidently realize a Lie group of dimensions n = 3 . Notice that

we can always set
4
X
u = −x2 + ix1 v = x4 + ix3 x2i = 1
i=1

whence it manifestly follows that the group SU (2) is topologically

homeomorphic to the three dimensional hypersphere S3 of unit radius
plunged in R4 .
As all the SU (2) matrices are unitary, the three generators are anti-
hermitean 2 × 2 matrices, which can be written in terms of the Pauli
matrices

τa ≡ 12 i σa ( a = 1, 2, 3 ) (1.13)

25
      
0 1  0 −i   1 0
σ1 =  σ2 =  σ3 =  (1.14)

  
  

1 0 i 0 0 −1
   

and therefore

[ τa τ b ] = − εabc τc (a, b, c = 1, 2, 3) (1.15)

It follows that SU (2) has the very same Lie algebra (1.6) of the rotation
group. Using the canonical coordinates (1.4), in a neighborhood of
the unit element we can write the SU (2) elements in the exponential
representation. From the very well known identities

σa σ b + σ b σa = 2 δab (σa αa )2 = | α | 2 = α12 + α22 + α32

where the 2 × 2 identity matrix is understood, we obtain

∞
X 1
g(α) = exp{τa αa } ≡ (τa αa )k
k=0
k!
∞
(  2k  2k+1 )
X 1 i 1 i σa αa
= |α| + |α|
k=0
(2k)! 2 (2k + 1)! 2 |α|
∞
( 2k 2k+1 )
(−1)k 1 i (−1)k
 
X 1 σ a αa
= |α| + |α|
k=0
(2k)! 2 (2k + 1)! 2 |α|
   
1 αa 1
= cos | α | + i σa sin |α|
2 |α| 2
= 1 x4 + i σa xa (x21 + x22 + x23 + x24 = 1) (1.16)

Notice that, by direct inspection,

g † (α) = g(−α) = g −1 (α)

so that g(α) are unitary matrices. Furthermore

 
 x4 + ix3 ix1 + x2
g(α) = 


ix1 − x2 x4 − ix3
 

hence det g(α) = 1 . From the explicit formula (1.16) it follows that the
whole set of special unitary 2 × 2 matrices is spanned if and only if the
canonical coordinates α are restricted to lie inside a sphere of radius
2π
| α | 2 = α12 + α22 + α32 < (2π)2

26
 it follows that the group SU (2) realizes the lowest dimensional non-
trivial, faithful, irreducible and unitary representation of the rotation
group, which is thereby named its fundamental representation and
denoted by τ F (g) . Since the fundamental representation acts upon
complex two–components column vectors, the Pauli spinors of nonrel-
ativistic quantum mechanics that describe particles of spin 12 , it is also
named as the spinorial representation of the rotation group and further
denoted by τ 1 (g) ≡ τ F (g) .
2

• A comparison between the fundamental τ F (g) = SU (2) and the adjoint

τ A (g) = SO(3) representations of the rotation group is very instructive.
Since these two representations of the rotation group share the same
Lie algebra they are locally homeomorphic. It is possible to obtain
all the finite elements of SO(3) in the exponential representation. As
a matter of fact, for the infinitesimal operators (1.6) of the adjoint
representation we have (the identity matrix is understood)

(Ia αa )2 = − | α | 2 + Ξ(α) || Ξ(α) || ac ≡ αa αc

(Ia αa )3 = − | α | 2 Ia αa
and consequently

[ Ξ(α) ]2 = | α | 2 Ξ(α) αa Ia Ξ(α) = 0 = Ξ(α) Ia αa

 
−2
(Ia αa ) 2k 2 k
= (− | α | ) 1 − | α | Ξ(α)

(Ia αa ) 2k+1 = (− | α | 2 ) k Ia αa
in such a way that
∞
X (−1)k
τ A (α) ≡ exp{Ia αa } = | α | 2k
k=0
(2k)!
∞
X (−1)k
+ | α | 2k Ia αa
k=0
(2k + 1)!
∞
X (−1)k
+ | α | 2k Ξ(α)
k=0
(2k + 2)!
∞
sin | α | Ξ(α) X (−1)k
= cos | α | + Ia αa + | α | 2k
|α| | α | 2 k=1 (2k)!
sin | α | cos | α | − 1
= cos | α | + Ia αa + Ξ(α) (1.17)
|α| | α |2

27
Now the manifold of the canonical coordinates can be divided into two
parts : an inside shell 0 ≤ | α | < π and an outer region π ≤ | α | < 2π .
To each point in the inside shell we can assign a point in the outer
region by means of the correspondence
2π − | α |
α 0 = (− α) 0 ≤ |α| < π, π ≤ | α 0 | < 2π (1.18)
|α|

The SU (2) elements corresponding to α and α 0 are then related by

τ F (α 0 ) = − τ F (α)

All points on the boundary of the parameter space, a 2–sphere of radius

2π , correspond to the very same element i.e. τ F ( | α | = 2π ) = − 1 .
Conversely, it turns out that for any pair of canonical coordinates
(α, α 0 ) connected by the relation (1.18) we have

τ A (α) = τ A (α 0 )

As a consequence the adjoint exponential representation (1.17) is an

irreducible and unitary representation of the rotation group that is not
a faithful one on the whole domain of the canonical coordinates of the
rotation group

D = {α = (α1 , α2 , α3 ) | α12 + α22 + α32 < (2π)2 }

the adjoint SO(3) representation being an homomorphism – the map

τ A (α) : D → SO(3) is not iniective – at variance with respect to the
fundamental SU (2) representation which is faithful.
This implies in turn that, for 0 ≤ | α | < 2π , there are paths which are
closed in the SO(3) adjoint representation and which can not deformed
into a point. Conversely, any closed path in the SU (2) fundamental
representation can always be deformed into a point. This means that
the spinor representation of the rotation group is simply connected,
though the vector representation is not.
Owing to this SU (2) is said to be the universal covering of SO(3) .
It is worthwhile to remark that, on the one hand, had we used the
eulerian angles (1.5) to parametrize the manifold of the proper rotation
group SO(3) , then the irreducible, orthogonal adjoint representation
τ A (ϕ, θ, ψ) turns out to be manifestly faithful. On the other hand,
the generic element of the irreducible and unitary fundamental spinor

28
 representation τ F (ϕ, θ, ψ) can be expressed in terms of the eulerian
angles as [5]

τ 1 (ϕ, θ, ψ) =
2 

 exp{i (ψ + ϕ)/2} cos θ/2 −i exp{i (ψ − ϕ)/2} sin θ/2 

−i exp{i (ϕ − ψ)/2} sin θ/2 exp{−i (ψ + ϕ)/2} cos θ/2
 

so that for instance

τ 1 (0, θ, ψ) = − τ 1 (2π, θ, ψ)
2 2

Hence it follows that to any rotation g(ϕ, θ, ψ) there correspond two

opposite matrices ± τ 1 (ϕ, θ, ψ) .
2

Thus, in terms of the eulerian angles, the adjoint vector representation

τ A (ϕ, θ, ψ) appears to be a real, faithful, orthogonal, irreducible single-
valued representation of the rotation group, whereas the fundamental
spinor representation τ F (ϕ, θ, ψ) turns out to be a unitary irreducible
double-valued representation of the rotation group. Hence, the rotation
group SO(3) in the three dimensional space is not simply connected,
its universal enveloping group being SU (2) .

• Unitary representations : as it is well known from the theory of the

angular momentum in quantum mechanics, it turns out that all the
irreducible, unitary, finite dimensional representations of the proper
rotation group are labelled by their weight, which is a positive semidef-
inite integer or half–integer number : namely,

{τ j | j = n/2 (n = 0, 1, 2, . . .)} dim τ j = 2j + 1

From the composition law of the angular momenta, it turns out that
the product τ j × τ k of two irreducible unitary representations of the
three dimensional rotation group of weights j and k contains just once
each of the irreducible unitary representations

τi (i = |j − k|, |j − k| + 1, . . . , j + k − 1, j + k)

Thus the following formula is valid

j+k
M
∀ j, k = 0, 21 , 1, 32 , 2, . . .


τj × τk = τi
i=|j−k|

29
• Other important examples of Lie groups are :

1. the full orthogonal groups O(N ) of the N × N orthogonal real

matrices with determinant equal to ±1 and their special or proper
subgroups SO(N ) with unit determinant, the dimensions of which
are n = 12 N (N − 1), ;
2. the groups U (N ) of the N × N unitary complex matrices, the
dimensions of which are n = N 2 ; in fact the generic U (N ) matrix
depends upon 2N 2 real parameters, but the request of unitarity
entails N 2 real conditions so that the dimension of the Lie group
U (N ) is n = N 2 .
3. the groups SU (N ) of the N ×N special unitary complex matrices,
i.e. with unit determinant, of dimensions n = N 2 − 1 .
Notice that U (N ) is homeomorphic to the product SU (N )×U (1) .
4. The special linear groups SL(N, R) of the N × N real matrices of
unit determinant, the dimensions of which are n = N 2 − 1 ;
5. the special linear groups SL(N, C) of the N ×N complex matrices
of unit determinant, the dimensions of which are n = 2N 2 − 2 ;
The distinguished case of the six dimensional special linear group
SL(2, C) will be of particular relevance, as it turns out to be the
universal covering spinorial group of the Lorentz group, i.e. it will
play the same role in respect to the Lorentz group as the unitary
group SU (2) did regard to the three dimensional rotation group.

30
1.3 The Inhomogeneous Lorentz Group
1.3.1 The Lorentz Group
Consider a spacetime point specified in two inertial coordinate frames S and
S 0 , where S 0 moves with constant relative velocity v with respect to S. In
S the spacetime point is labelled by (x, y, z, t) and in S 0 by (x 0 , y 0 , z 0 , t 0 ).
The transformation that relates the two inertial coordinate frames is called
a Lorentz transformation: according to the postulates of the special theory
of relativity, it has the characteristic property (c = 299 792 458 m s−1 is the
velocity of light in vacuum)
c2 t2 − x2 − y 2 − z 2 = c2 t 0 2 − x 0 2 − y 0 2 − z 0 2
where we are assuming that the origins of both inertial frame coordinate
systems do coincide for t = t 0 = 0 – for the moment we do not consider
translations under which the relative distances remain invariant; if these
are included one finds inhomogeneous Lorentz transformations, also called
Poincaré transformations. We shall use the following standard notations and
conventions [2] – sum over repeated indices is understood
x µ ≡ (x0 = ct, x, y, z) = (x0 , x) = (x0 , xk ) µ = 0, 1, 2, 3 k = 1, 2, 3
so that
x2 ≡ gµν x µ x ν = gµν x 0 µ x 0 ν ≡ x 0 2
with  
 1 0 0 0 
−1
 

 0 0 0 

gµν =
 0 0 −1 0
 



 

0 −1
 
0 0
Owing to spacetime homogeneity and isotropy the Lorentz transformations
are linear
x 0 µ = Λµν x ν
which implies
gρσ = gµν Λµρ Λνσ (1.19)
or even in matrix notations
x µ xµ = x> · g x = x 0 µ xµ0 x0 = Λ x g = Λ> g Λ
where > denotes transposed matrix. Eq. (1.19) does indeed define the Lorentz
group L as a group of 4×4 square–matrices acting upon Minkowski spacetime
point column four vectors. As a matter of fact we have

31
 1. composition law : Λ , Λ0 ∈ L ⇒ Λ · Λ0 = Λ00 ∈ L
matrix products : (Λ00 ) µρ = Λ µν (Λ0 ) νρ

2. identity matrix : ∃! I
 
 1 0 0 0 
 

 0 1 0 0 

I=
 



 0 0 1 0 


 
0 0 0 1

3. inverse matrix : from the defining relation (1.19) we have

det g = det (Λ> g Λ) ⇒ det Λ = ±1

and thereby ∃ ! Λ−1 = g · Λ> · g ∀Λ ∈ L

which means that L is a group of matrices that will be called the homogeneous
full Lorentz group. From the relation
2
1 = g 00 = g µν Λ µ0 Λ ν0 = Λ00 − Λ k0 Λ k0 ⇒  Λ00  ≥ 1
  

it follows that the homogeneous full Lorentz group splits into four categories
called connected components

• proper orthochronus L↑+ = {Λ ∈ L | det Λ = 1 ∩ Λ00 ≥ 1}

• improper orthochronus L↑− = {Λ ∈ L | det Λ = −1 ∩ Λ00 ≥ 1}

• proper nonorthochronus L↓+ = {Λ ∈ L | det Λ = 1 ∩ Λ00 ≤ −1}

• improper nonorthochronus L↓− = {Λ ∈ L | det Λ = −1 ∩ Λ00 ≤ −1}

Among the four connected components of the homogeneous full Lorentz

group there is only L↑+ which is a subgroup, i.e. the component connected
with the identity element, which is also called the restricted subgroup of
the homogeneous full Lorentz group. Other quite common notations are as
follows. The homogeneous full Lorentz group is also denoted by O(1, 3) ,
the proper orthochronus component L↑+ by SO(1, 3)+ or O(1, 3)+ + and is also
called the restricted homogeneous Lorentz group. The remaining connected
components are also respectively denoted by
↓ ↑ ↓
O(1, 3)−
+ = L+ O(1, 3)+
− = L− O(1, 3)−
− = L−

Examples :

32
1. special Lorentz transformation, or even boost, in the OX direction with
velocity v > 0 towards the positive OX axis

cosh η − sinh η 0 0 
 

− sinh η cosh η 0 0 

 
 
Λ(η) = 
 



 0 0 1 0  

 
0 0 0 1
v
cosh η = (1 − β 2 )−1/2 sinh η = β (1 − β 2 )−1/2 β=
c
2. spatial rotation around the OZ axis
 
 1 0 0 0 
 

 0 cos θ sin θ 0 

Λ(θ) =  
0 − sin θ cos θ 0

 


 

 
0 0 0 1

boosts and rotations belong to the special homogeneous Lorentz group.

3. parity transformation with respect to the OY axis

 
 1 0 0 0 
 
 0 1
 0 0 
ΛyP = 


−1 0 
 


 0 0 

 
0 0 0 1

full spatial reflection or full parity transformation

 
 1 0 0 0 
0 −1 0
 

 0 

ΛP =   (1.20)
0 0 −1 0 

 

 

0 −1
 
0 0

spatial reflections or parity transforms belong to L↑− = O(1, 3)+

−

4. time inversion or time reflection T transformation

−1 0 0
 
 0 
 
 0 1 0 0 
 ∈ L↓− = O(1, 3)−
 
ΛT =  
−


 0 0 1 0 


 
0 0 0 1

33
 5. full inversion or P T transformation
−1 0
 
 0 0 
0 −1 0
 
 0 
∈ L↓+ = O(1, 3)−
 
ΛP T =  
+
0 −1 0
 


 0 


0 −1
 
0 0

Any Lorentz transformation can always been decomposed as the product of

transformations of the four types belonging to the above described connected
components. Since there are three independent rotations as well as three
independent boosts, of for each spatial direction, the special homogeneous
Lorentz transformations are described in terms of six parameters.
The homogeneous Lorentz group is a six dimensional Lie group so that
each element can be labelled by six real parameters. For example, one can
choose the three eulerian angles and the three ratios between the components
of the relative velocity and the light velocity : namely,
g ∈ O(1, 3) → (ϕ, θ, ψ, β1 , β 2 , β 3 ) ( βk ≡ vk /c )
0 ≤ ϕ ≤ 2π , 0 ≤ θ < π , 0 ≤ ψ ≤ 2π , −1 < βk < 1 ( k = 1, 2, 3 )
The most suitable parametrization is in terms of the canonical coordinates
(α , η ) = (α1 , α2 , α3 ; η1 , η 2 , η3 ) (1.21)
where the angles αk (k = 1, 2, 3) with α2 < (2π)2 are related to the spatial
rotations around the orthogonal axes of the chosen inertial frame, whilst the
2 −1/2
hyperbolic arguments ηk ≡ Arsh β k (1 − β k ) (k = 1, 2, 3) with η ∈ R3
are related to boosts along the spatial directions. More specifically
 
 1 0 0 0 
 

 0 1 0 0  
Λ(α1 ) = 
 



 0 0 cos α1 sin α1  

0 0 − sin α1 cos α1
 

represents a rotation of the inertial reference frame in the counterclockwise

sense of an angle α1 around the OX axis whereas e.g.
cosh η3 0 0 − sinh η3 
 

 

 0 1 0 0 

Λ(β3 ) = 
 



 0 0 1 0 


− sinh η3 0 0 cosh η3
 

corresponds to a boost with a rapidity parameter

 
2 −1/2
η 3 = Arsh β 3 (1 − β 3 ) (1.22)

34
associated to the OZ direction. Notice that the inverse transformations can
be immediately obtained after sending αk 7→ − αk and η k 7→ − η k . Since
the domain of the canonical coordinates is the unbounded subset of R6

D ≡ {(α , η ) | α2 < (2π)2 , η ∈ R3 }

it follows that the Lorentz group is noncompact.

It is also very convenient to write the elements of the special homogeneous
Lorentz group in the exponential form and to introduce the infinitesimal
generators according to the standard convention [5]

Λ(α1 ) = exp{α1 I1 } Λ(β3 ) = exp{η 3 J3 }

where
 
 0 0 0 0 
dΛ
 
 0 0 0 0 
≡
 
I1 (α1 = 0) =  

dα1 

 0 0 0 1 


0 0 −1
 
0
−1
 
 0 0 0 
dΛ
 
 0 0 0
 0 
≡

J3 (β 3 = 0) = 
 

dβ3 

 0 0 0 0 


−1 0 0
 
0
et cetera . It is very important to gather that the infinitesimal generators
of the space rotations are antihermitean Ik† = −Ik (k = 1, 2, 3) whereas the
infinitesimal generators of the special Lorentz transformations turn out to be
hermitean Jk = Jk† (k = 1, 2, 3) . One can also check by direct inspection that
the infinitesimal generators do fulfill the following commutation relations:
namely,

[ Ij Ik ] = − εjkl Il [ Jj Jk ] = εjkl Il [ Ij Jk ] = − εjkl Jl (1.23)

( j, k, l = 1, 2, 3)
The above commutation relations univocally specify the Lie algebra of the
homogeneous Lorentz group.
Together with the infinitesimal generators Ik , Jk (k = 1, 2, 3) it is very
convenient to use the matrices

Ak ≡ 21 (Ik + i Jk ) Bk ≡ 12 (Ik − i Jk ) k = 1, 2, 3

It is worthwhile to remark that all the above matrices are antihermitean

A†j = − Aj Bk† = − Bk ( j, k = 1, 2, 3)

35
The commutation relations for these matrices have an especially simple form:

[ Aj Ak ] = − εjkl Al [ Bj Bk ] = − εjkl Bl [ Aj Bk ] = 0

( j, k, l = 1, 2, 3)
which follow from the commutation relations of the infinitesimal generators.
We stress that the commutation relations for the operators Ak (k = 1, 2, 3)
are the same as those for the generators of the three dimensional rotation
group SO(3) and of its universal covering group SU (2) . This is also true for
the operators Bk (k = 1, 2, 3) .
The infinite dimensional reducible unitary representations of the angular
momentum Lie algebra are very well known. Actually, one can diagonalize
simultaneously the positive semidefinite operators

A†j Aj = Aj A†j = − Aj Aj = − A2
Bk† Bk = Bk Bk† = − Bk Bk = − B2

the spectral resolutions of which read

X X
− A2 = m (m + 1) Pbm − B2 = n (n + 1) Pbn
m n

tr Pbm = 2m + 1 tr Pbn = 2n + 1
m , n = 0 , 21 , 1 , 32 , 2 , 52 , 3 , . . . . . .
Those operators are called the Casimir’s operators of the Lorentz group, the
latter ones being defined by the important property of commuting with all
the infinitesimal generators of the group. It follows therefrom that

• all the irreducible finite dimensional representations of the Lorentz

group are labelled by a pair (m, n) of positive semidefinite integer or
half–integer numbers

• each irreducible representation τ m n contains (2m + 1) (2n + 1) linearly

independent states characterized by the eigenvalues of e.g. A3 and B3
respectively

• we can identify Sk = Ik = Ak + Bk (k = 1, 2, 3) with the components

of the spin angular momentum operator, so that we can say that τ m n
is an irreducible representation of the Lorentz group of spin angular
momentum s = m + n

36
• the lowest dimensional irreducible representations of the Lorentz group
are: the unidimensional scalar representation τ 0 0 , the two dimensional
left spinor Weyl representation τ 1 0 , the two dimensional right spinor
2
Weyl representation τ 0 1 (the handedness is conventional)
2

Hermann Klaus Hugo Weyl

one of the most influential personalities for Mathematics
in the first half of the XXth century
Elmshorn, Germany, 9.11.1885 – Zurich, CH, 8.12.1955
Gruppentheorie und Quantenmechanik (1928)

• under a parity transform ΛP = ΛP−1 we find

Ik 7→ ΛP−1 Ik ΛP = Ik Jk 7→ ΛP−1 Jk ΛP = − Jk (1.24)

Ak 7→ ΛP−1 Ak ΛP = Bk Bk 7→ ΛP−1 Bk ΛP = Ak
i.e. the two inequivalent irreducible Weyl’s representations interchange
under parity. Hence, to set up a spinor representation of the full Lorentz
group out of the two Weyl’s representations, one has to consider the
direct sum
τD = τ 1 0 ⊕ τ0 1
2 2

which is a reducible four dimensional representation of the restricted

Lorentz group and is called the Dirac representation

• all the other irreducible representations of higher dimensions can be

obtained from the well known Clebsch–Gordan–Racah multiplication
and decomposition rule
M
τmn × τpq = τrs
r,s

|m − p| ≤ r ≤ m + p |n − q| ≤ s ≤ n + q
In particular, the spin 1 representation does coincide with the above
introduced four vector representation, which we actually used to define
the homogeneous Lorentz group

τ 1 0 × τ0 1 = τ 1 1
2 2 2 2

The antisymmetric Maxwell’s field strength Fµν transforms according

to the reducible parity symmetrical representation τ 1 0 ⊕ τ 0 1 .

37
 • All the irreducible finite dimensional representations of the Lorentz
group are nonunitary. As a matter of fact we have e.g.

τ m n (β3 ) = exp{ β3 J3 } = exp{ − iβ3 (A3 − B3 ) }

and thereby
† −1
τm n (β3 ) = τ m n (β3 ) 6= τ m n (β3 ) = τ m n (−β3 )

All the unitary irreducible representations of the Lorentz group are

infinite dimensional and have been classified by
I.M. Gel’fand and M.A. Naimark
Unitary representations of the Lorentz group
Izv. Akad. Nauk. SSSR, matem. 11, 411, 1947.

Exercise. A second rank tensor t µν with µ, ν = 0, 1, 2, 3 transforms according to the

reducible representation T = τ 21 12 × τ 12 12 of the Lorentz group O(1, 3) .
1. Decompose the representation T into the sum of irreducible representations of
O(1, 3)
2. Specify the dimensions of T and of each irreducible representations appearing in
the above decomposition
3. Among the sixteen matrix elements t µν find the components belonging to each
irreducible representations appearing in the decomposition.

Solution. From the general decomposition rule

M
τmn × τpq = τrs
r,s

|m − p| ≤ r ≤ m + p |n − q| ≤ s ≤ n + q
1
we obtain for m = n = 2
M M M
T = τ00 τ10 τ01 τ11

so that the dimensions correctly match because

16 = 1 + 3 + 3 + 9

The one dimensional irreducible representation τ 0 0 corresponds to the trace of the second
rank tensor t µν i.e.
t = g µν t µν
which is a Lorentz scalar. The six components of the antisymmetric part of the second
rank tensor t µν that is
A µν = 12 (t µν − t νµ )

38
 L
transform according to the parity invariant representation τ 1 0 τ 0 1 , while the nine
components of the traceless and symmetric part of the second rank tensor t µν viz.,

S µν = 1
2 (t µν + t νµ ) − 1
4 g µν t

transform according to the irreducible representation τ 1 1 . Therefore we can always write

the most general decomposition

t µν = A µν + S µν + 1
4 g µν t ( t = g µν t µν )

Exercise. Find the explicit form of the structure constants of the Lorentz group.
Solution. The commutation relations (1.23) can be written in explicit form as

[ I1 I2 ] = − I3 [ I1 I3 ] = I2 [ I1 J1 ] = 0 [ I1 J2 ] = − J3 [ I1 J 3 ] = J 2

[ I2 I3 ] = − I1 [ I2 J1 ] = J3 [ I2 J2 ] = 0 [ I2 J3 ] = − J1
[ I3 J1 ] = − J2 [ I3 J 2 ] = J 1 [ I3 J3 ] = 0
[ J1 J2 ] = I3 [ J1 J3 ] = − I2
[ J 2 J 3 ] = I1
Now it is convenient to slightly change the notation and denote the boost generators by

J1 = I4 J2 = I5 J3 = I6

in such a manner that all the nonvanishing structure constants of the Lorentz group

Cabc = − Cbac ( a, b, c, . . . = 1, 2, . . . , 6 )

can be read off the above commutation relations: namely,

C123 = −1 C132 = 1 C156 = −1 C165 = 1

C231 = −1 C246 = 1 C264 = −1
C345 = −1 C354 = 1
C453 = 1 C462 = −1
C561 = 1

whence it manifestly appears that the structure constants are completely antisymmetric,
as it will be proved later on.

39
1.3.2 Semisimple Groups
Lie groups and their corresponding Lie algebras can be divided into three
main categories depending upon the presence or absence of some invariant
subgroups and invariant subalgebras.
By its very definition, an invariant subgroup H ⊆ G satisfies the following
requirement : for any elements g ∈ G and h ∈ H there always exists an
element h0 ∈ H such that
gh = h0 g
Of course, any Lie group G has two trivial invariant subgroups, G itself and
the unit element.
Concerning the infinitesimal operators Ja ∈ G (a = 1, 2, . . . n) and Tb ∈
H (b = 1, 2, . . . m ≤ n) of the corresponding Lie algebra and subalgebra, we
shall necessarily find

[ Ja Tb ] = Cabc Tc (a = 1, 2, . . . n , b, c = 1, 2, . . . m ≤ n) (1.25)

• Groups that do not possess any nontrivial invariant subgroup are called
simple.

• A weaker requirement is that the group G had no nontrivial invariant

abelian subgroups : in such a circumstance G is said to be semisimple.
In this case it can be proved that any semisimple Lie group G is locally
isomorphic to the so called direct product or cartesian product of some
mutually commuting simple nonabelian groups G1 , G2 , . . . , Gs . This
means that ∀ gα ∈ Gα , gβ ∈ Gβ we have

˙ G1 × G2 × · · · × Gs
G=

gα gβ = gβ gα α 6= β (α, β = 1, 2, . . . , s)
s
X
dim(G) = dim(Gα )
α=1

where the symbol = ˙ means that the analytic isomorphism is true in

some suitable neighborhoods of the unit elements of the involved Lie
groups.
For the corresponding Lie algebras we have that ∀ Iα ∈ Gα , Iβ ∈ Gβ
s
M
G= Gα [ Iα Iβ ] = 0 α 6= β (α, β = 1, 2, . . . , s)
α=1

˙ SO(3) × SO(3) .
An example of a semisimple group is SO(4) =

40
• Groups that do contain some invariant abelian nontrivial subgroups are
said to be nonsemisimple.
Such groups do not always factorize into the direct product of an
abelian invariant subgroup and a semisimple group. As an example,
let us consider the two dimensional euclidean group or inhomogeneous
orthogonal group IO(2) , which is the group of the rototranslations in
the plane without reflections with respect to an axis of the plane. We
shall denote a translation by the symbol T (a) , where a = (ax , ay ) is a
general displacement of all points in the OXY plane. Evidently

T (a) T (b) = T (a + b)

It is elementary to proof that every element g of the group IO(2) can be

represented in the form of a product of a rotation around an arbitrary
point O of the plane and a certain translation :

g = T (a) R O g ∈ IO(2)

It is also straightforward to prove the following identities

g T (a) g −1 = T (g a) g 6= T (b) (1.26)

T (a) R O T (− a) = R O+a (1.27)

where g ∈ IO(2) , g a is the point in the plane obtained from a by

the displacement g ∈ IO(2) , R O any rotation around the point O and
O + a the point to which O moves owing to the translation of a . Now,
if we choose

g(α) = exp{α Iz } 0 ≤ α ≤ 2π T (a) = exp {a · P}

where
   
0 1   cos α sin α
Iz =  g(α) = 

  

−1 0 − sin α cos α
  

from eq. (1.26) and for very small α, ax , ay we readily obtain the Lie
algebra among the generators of IO(2) , i.e.

[ Px Py ] = 0 [ Iz Px ] = Py [ Iz Py ] = − Px

which precisely corresponds to the condition (1.25). Hence, translations

do constitute an invariant abelian subgroup of IO(2) and consequently
IO(2) is nonsemisimple.

41
A very important quantity is the Cartan–Killing metric of a Lie group G ,
which is defined to be

gab ≡ Cacd Cbdc (a, b, c, d = 1, 2, . . . , n) (1.28)

where Cabc are the structure constants of the Lie algebra G . According to
the correspondence (1.10) we can also write

tr (Aa Ab ) = gab (a, b = 1, 2, . . . , n)

where n × n square matrices Aa (a = 1, 2, . . . , n) denote the generators in

the adjoint representation. For instance, in the case of SU (2) we find

gab = (− εacd )(− εbdc ) = − εacd εbcd = − 2 δab

It is important to remark that the Cartan–Killing metric of a Lie group G

is a group invariant. As a matter of fact, for any inner automorphism of the
adjoint representation, see the definition (1.9), we find
   
−1 −1
tr Aa (g) Ab (g) = tr TA (g) Aa TA (g) TA (g) Ab TA (g)
   
= tr TA (g) Aa Ab TA−1 (g) = tr Aa Ab TA−1 (g) TA (g)
= tr (Aa Ab ) = gab
∀ g∈G (a, b = 1, 2, . . . , n) (1.29)

where TA (g) (g ∈ G) are the linear operators of the adjoint representation

and I have made use of the cyclicity property of the trace operation, that is

tr (A1 A2 · · · An ) = tr (A2 · · · An A1 ) = tr (An A1 A2 · · · An−1 ) = · · ·

In particular, for an infinitesimal group transformation

TA (g) = IA + αc Ac | αc |  1

where IA is the identity operator in the adjoint representation, the above

equalities entail
   
0 = δ gab = αc tr [ Ac Aa ] Ab + tr Aa [ Ac Ab ] αc
= αc Ccad tr (Ad Ab ) + Ccbd tr (Aa Ad ) αc
= (Ccad g bd + Ccbd g ad ) αc = (Ccab + Ccba ) αc (1.30)

whence

Ccab = − Ccba (1.31)

42
which means antisymmetry of the structure constants also respect the last
two indices. As a consequence the structure constants of any Lie group turn
out to be completely antisymmetric with respect to the exchange of all three
indices : namely, Cabc = − Cbac = Cbca = − Ccba .
The Cartan–Killing metric is non-singular, i.e. det || g || =
6 0 , if and only if
the group is semisimple. Here k g k indicates the n × n matrix of elements
gab . For example we have
−2 0
 
 0 
0 −2 0 
 
g ab = 
 
 for SU (2) (1.32)
0 −2
 
0
 
 0 0 0 

 0 0 0 
g ab = 



 for IO(2)
0 0 −2
 

The Cartan–Killing metric is a real symmetric matrix : therefore, it can

be diagonalized by means of an orthogonal transformation that reshuffles the
infinitesimal operators.
1. If there are null eigenvalues in the diagonal form of the metric, then
the Lie group is nonsemisimple.
2. If the Cartan–Killing metric is negative definite, then the Lie group is
compact.
3. If the Cartan–Killing metric has both positive and negative eigenvalues,
as in the case of the homogeneous Lorentz group or the group SL(2, R),
then the Lie group is noncompact.
Since the Cartan–Killing metric is a nonsingular n × n matrix for semisimple
Lie groups it is possible to define its inverse matrix g ab , which can be used
to raise and lower the group indices a, b, c, . . . = 1, 2, . . . , n . For example, the
Lie product of two generators in the representation R, or commutator in the
quantum physics terminology, will be written in different though equivalent
forms like
[ IaR IbR ] = Cabc IcR = Cabc g cd IdR = Cabc IcR = Cabc g cd IdR
because contractions over lower group indices have to be performed by means
of the inverse Cartan-Killing metric tensors.
For any semisimple Lie group G it is always possible to select a special
element for any representation of its semisimple Lie algebra G, that is
CR ≡ g ab IRa IRb = g ab IaR IbR IRa ≡ g ab IbR (a, b = 1, 2, . . . , n) (1.33)

43
the subscript R labelling some particular representation of the infinitesimal
operators. The quadratic operator CR is called the Casimir’s operator of the
representation R and turn out to be group invariant : namely,
TR (g) CR TR−1 (g) = CR ∀g ∈ G
or else
[ TR (g) CR ] = 0 ∀g ∈ G

Proof : first we notice that

[ IcR CR ] = g ab [ IcR (IaR IbR ) ]
= g ab [ IcR IaR ] IbR + g ab IaR [ IcR IbR ]
= g ab Ccad g de IeR IbR + g ab IaR Ccbd g de IeR
= g ab Ccad g de IeR IbR + g ab Ccbd g de IaR IeR
Reshuffling indices in the second addendum of the last line we obtain
 
[ IcR CR ] = Ccad g ab g ed + g ea g db IeR IbR (1.34)

and owing to complete antisymmetry of the structure constants with respect

to all indices we eventually find
[ IcR CR ] = 0
By iterating the above procedure it is immediate to verify that
[ ( αa IaR )k CR ] = 0 ∀k ∈ N αa ∈ R ( a = 1, 2, . . . , n )
Hence, from the exponential representation,
∞
X 1
TR (g) CR = exp{αa IaR } CR = ( αa IaR )k CR
k=0
k!
∞
X 1
= CR ( αa IaR )k
k=0
k!
= CR exp{αa IaR } = CR TR (g) ∀g ∈ G
which completes the proof . 
The general structure of the Casimir’s operators is strongly restricted by
• Schur’s lemma : if a linear operator commutes with every element of
any irreducible representation τ (g) of a group, then it is proportional
to the unit operator.
For a Lie group this lemma may also be rephrased as follows.

44
 • If a linear operator commutes with all the generators in an irreducible
representation of a Lie algebra G , then it must be proportional to the
unit operator.
See [5] for the proof. According to Schur’s lemma we come to the conclusion
that in the irreducible representation labelled by R we have
CR = dR IR
where dR is a number, depending upon the irreducible representation, which
is called the Dynkin’s index.
• For the adjoint representation we have
tr CA = g ab tr (Aa Ab ) = dA tr IA = n dA = g ab g ab = n
so that dA = 1 .
• In the case of the special unitary group SU (2) , i.e. in the case of
the fundamental representation of the rotation group, from the very
definition (1.13) and the Cartan-Killing metric (1.32), the Casimir’s
operator (1.33) reads – remember that g ab = − 12 δ ab
tr CF = 2 dF = − 12 δ ab tr (τa τb ) = − 12 δ ab − 14 2 δab = 34

3
so that dF = 8
and dA = 1 for the rotation group.
• For any irreducible representation of the rotation group it is possible
to show that
CR = 12 J (J + 1) IR
where J is the weight of the irreducible representation, which is related
to the eigenvalue of the total angular momentum by J2 = ~2 J(J + 1) .
Hence  
dJ = 21 J (J + 1) J = 0, 12 , 1, 32 , 2, 52 , 3, . . .
are the Dynkin’s indices for the irreducible unitary representations of
the rotation group.
• For a nonsemisimple Lie group G, just like the Poincaré group for
example, owing to the singular nature of the Cartan–Killing metric
tensor, the Casimir operators in any irreducible representation R are
just defined by the very requirement of being invariant under the whole
group of linear transformations TR (g) (g ∈ G) : namely,
CR = TR (g) CR TR −1 (g) ∀g ∈ G

45
1.3.3 The Poincaré Group
The quite general symmetry group of all the relativistic classical and quantum
field theories obeying the special principle of relativity is the (restricted)
inhomogeneous Lorentz group, also named the Poincaré group, under which

x 0 µ = Λµν x ν + a µ

where a µ is an arbitrary constant four vector. Hence the Poincaré group is a

nonsemisimple ten parameters Lie group, the canonical coordinates of which,
in accordance with (1.21), can be identified with
 
(α , η , a ) = α1 , α2 , α3 ; η1 , η 2 , η3 ; a0 , a1 , a2 , a3

The spacetime translations T (a) do constitute an abelian four parameters

subgroup and fulfill

T (a) T (b) = T (a + b) = T (b) T (a)

However, spacetime translations do not commute with the Lorentz group

elements. Consider in fact two Poincaré transformations with parameters
(Λ, a) and (Λ 0 , a 0 ) so that
 
x µ 7→ Λµν x ν + a µ 7→ Λ 0ρµ Λρν x ν + a ρ + a 0 µ

aµ 7→ Λ 0ρµ a ρ + a 0 µ
whence we see that the translation parameters get changed under a Lorentz
transformation. Owing to this feature, which is called the soldering between
the Lorentz transformations and the spacetime translations, the Poincaré
group is said to be a semidirect product of the Lorentz group and the space-
time translation abelian group.
The generators of the spacetime translations are the components of the
four gradient operator. As a matter of fact, for any analytic real function
f : M → R we have

T (a) f (x) = exp {a µ ∂µ } f (x) = f (x + a)

[ ∂µ , ∂ν ] ≡ ∂µ ∂ν − ∂ν ∂µ = 0
Thus, the infinitesimal operators of the spacetime translations turn out to
be differential operators acting of the infinite dimensional space C∞ (M) of
the analytic functions on the Minkowski’s spacetime.

46
 It is necessary to obtain an infinite dimensional representation of the
generators of the Lorentz group acting on the very same functional space.
Introduce the six antihermitean differential operators

`µν ≡ g µρ x ρ ∂ν − g νρ x ρ ∂µ ≡ x µ ∂ν − x ν ∂µ (1.35)
†
`µν = − `νµ = − `µν
By direct inspection it is straightforward to verify the commutation relations

[ `µν , `ρσ ] = − g µρ `νσ + g µσ `νρ − g νσ `µρ + g νρ `µσ (1.36)

so that, if we make the correspondences

1
Ij ↔ 2
ε jkl `kl Jk ↔ `0k ( j, k, l = 1, 2, 3 )

it can be readily checked that the above commutation relations among the
differential operators `µν do realize an infinite dimensional representation of
the Lie algebra (1.23) of the Lorentz group. Nonetheless, it is important to
remark that all six generators in the above infinite dimensional representation
of the Lorentz group are antihermitean, at variance with the infinitesimal
operators of all the finite dimensional irreducible representations, in which
only the three generators of the rotation subgroup are antihermitean.
Moreover we find

[ `µν , ∂ρ ] = − g µρ ∂ν + g νρ ∂µ

hence we eventually come to the infinite dimensional representation of the

Lie algebra of the Poincaré group

[ ∂µ , ∂ν ] = 0 [ `µν , ∂ρ ] = − g µρ ∂ν + g νρ ∂µ
[ `µν , `ρσ ] = − g µρ `νσ + g µσ `νρ − g νσ `µρ + g νρ `µσ (1.37)

However, in view of the applications to quantum mechanics and quantum

field theory, it is convenient and customary to introduce a set of hermitean
generators of the Poincaré group: namely,

Pµ ≡ i ∂µ = (i ∂ 0 , i ∇) (1.38)

Lµν ≡ i `µν = x µ Pν − x ν Pµ (1.39)

and the corresponding Lie algebra

[ Pµ , P ν ] = 0 [ Lµν , Pρ ] = − i g µρ Pν + i g νρ Pµ

47
 [ Lµν , Lρσ ] = − i g µρ Lνσ + i g µσ Lνρ − i g νσ Lµρ + i g νρ Lµσ (1.40)

One can easily recognize that the six hermitean differential operators Lµν
do actually constitute the relativistic generalization of the orbital angular
momentum operator of nonrelativistic quantum mechanics. We find indeed

L = (L23 , L31 , L12 ) = r × (−i ∇) [ L  , Lk ] = i ε  k` L`

On the other hand, it turns out that the most general infinite dimensional
representation of the Poincaré Lie algebra is given by

[ Pµ , P ν ] = 0 [ Mµν , Pρ ] = − i g µρ Pν + i g νρ Pµ
[ Mµν , Mρσ ] = − i g µρ Mνσ + i g µσ Mνρ − i g νσ Mµρ + i g νρ Mµσ

where

Mµν ≡ Lµν + Sµν (1.41)

in which the relativistic spin angular momentum operator Sµν must satisfy

[ Pµ , S νρ ] = 0 = [ S µν , L ρσ ] (1.42)

[ Sµν , Sρσ ] = − i g µρ Sνσ + i g µσ Sνρ − i g νσ Sµρ + i g νρ Sµσ (1.43)

Also the above construction is nothing but the relativistic generalization of

the most general representation for the Lie algebra of the three-dimensional
euclidean group IO(3) in the nonrelativistic quantum mechanics

p = − i~ ∇ J=L+S=r×p+S

where p, L, S are respectively the self–adjoint operators of the momentum,

orbital angular momentum and spin angular momentum of a point–particle
with spin.
Since the translation infinitesimal operators Pµ do evidently constitute
an abelian invariant subalgebra, the restricted Poincaré group turns out to
be a nonsemisimple and noncompact Lie group, the diagonalization of the
Cartan–Killing ten–by–ten real symmetric matrix leading to three positive,
three negative and four null eigenvalues.
In order to single out the Casimir operators of the Poincaré group, it is
useful to introduce the Pauli-Lubanski operator
1 1
Wµ ≡ 2
ε µνρσ Pν M ρσ = 2
ε µνρσ Pν S ρσ (1.44)

48
as well as the Shirkov operator
µ
V ≡ M µρ Pρ

which satisfy the relationships

Wµ P µ = 0 = Vµ P µ

P 2 M µν = Vµ Pν − Vν Pµ − ε µνρσ W ρ P σ
It is immediate to check that the following commutation relations hold true

[ Wµ , P ν ] = 0 [ M µν , Wρ ] = − i g µρ Wν + i g νρ Wµ

so that we can readily recognize the two Casimir’s operators of the Poincaré
group to be the scalar and pseudoscalar invariants

C m = P 2 = P µ Pµ Cs = W 2 = W µ Wµ (1.45)

which are respectively said the mass and the spin operators. Now, according
to Wigner’s theorem [6]
Eugene Paul Wigner (Budapest 17.11.1902 – Princeton 1.1.1995)
Gruppentheorie und ihre Anwendung auf die Quantenmechanik der Atom-
spektrum, Fredrick Vieweg und Sohn, Braunschweig, Deutschland, 1931, pp.
251–254, Group Theory and its Application to the Quantum Theory of Atomic
Spectra, Academic Press Inc., New York, 1959, pp. 233–236
any symmetry transformation in quantum mechanics must be realized only
by means of unitary or antiunitary operators. The representation theory for
the Poincaré group has been worked out by Bargmann and Wigner [7]. The
result is that all the unitary irreducible representations of the Poincaré group
have been classified and fall into four classes.

1. The eigenvalue of the Casimir’s operator Cm ≡ m2 is real and positive,

while the eigenvalues of the Casimir’s operator Cs ≡ − m2 s(s + 1) ,
where s is the spin, do assume discrete values s = 0, 21 , 1, 23 , 2, . . .
Such a kind of unitary irreducible representations τ m , s are labeled by
the rest mass m > 0 and the spin s . The states belonging to this kind
of unitary irreducible representations are distinguished, for instance,
by e.g. the component of the spin along the OZ axis

sz = −s, −s + 1, . . . , s − 1, s ; s = 0, 12 , 1, 32 , 2, . . .

49
 and by the continuous eigenvales of the spatial momentum p , so that
p20 = p2 + m2 : namely,

| m, s ; p, sz i m>0 s = 0, 12 , 1, 23 , 2, . . .

p ∈ R3 sz = −s, −s + 1, . . . , s − 1, s
Physically, these states will describe some elementary particle of rest
mass m , spin s , momentum p and spin projection sz along the OZ
axis. Massive particles of spin s are described by wave fields which
correspond to 2s + 1 real functions on Minkowski’s spacetime.

2. The eigenvalues of both the Casimir’s operators vanish, i.e. P 2 = 0 and

W 2 = 0 , and since P µ Wµ = 0 it follows that Wµ and Pµ are light–
like and proportional, the constant of proportionality being named the
helicity, which is equal to ± s , where, again, s = 0, 21 , 1, 23 , 2, . . . is the
spin of the representation.
The states belonging to this kind of unitary irreducible representations
τ 0 , s are distinguished by e.g. two possible values of the helicity h = ± s
and by the continuous eigenvales of the momentum p , so that p20 = p2

| s ; p, h i m=0 s = 0, 12 , 1, 32 , 2, . . .

p ∈ R3 h = ±s
Thus, this kind of unitary irreducible representations of the Poincaré
group will correspond to the massless particles that, for s 6= 0 , are
described by two independent real functions on the Minkowski’s space-
time. Examples of particles falling in this category are photons with
spin 1 and two helicity states for e.g. circular polarization, the spin 21
massless left–handed neutrinos ( h = − 12 ) and massless right–handed
antineutrinos ( h = + 21 ) and maybe the graviton with spin 2 and two
polarizations.

3. The eigenvalue of the Casimir’s mass operator is zero, P 2 = 0 , while

that one of the Casimir’s spin operator is continuous, i.e. W 2 = −w2
where w > 0 . Thus, this kind of unitary irreducible representations
τ 0 , w of the Poincaré group would correspond to elementary particles of
zero rest mass, with an infinite and continuous number of polarizations.
These objects have never been detected in Nature.

4. There are also tachyon–like unitary irreducible representations which

P µ Pµ < 0 which are not physical as they drive to acausality.

50
There are further irreducible representations of the Poincaré group but they
are neither unitary nor antiunitary. As already remarked, Wigner’s theorem
[6] generally states that all symmetry transformations – just like Poincaré
transformations – in quantum mechanics can be consistently realized solely
by means of some unitary or antiunitary operators.

• All the unitary representations of the Poincaré group turn out to be

infinite dimensional, corresponding to particle states with unbounded
momenta p ∈ R3 .

• Elementary particles correspond to the irreducible representations, the

reducible representations being naturally associated to the composite
objects. For example, the six massive quarks with spin s = 12 and
with fractional electric charge q = 32 or q = − 13 are considered as
truly elementary particles, although not directly detectable because of
the dynamical confinement mechanism provided by Quantum Chromo
Dynamics (QCD), while hadrons, like the nucleons and π mesons, are
understood as composite objects.

• As already seen, all finite dimensional irreducible representations of the

Lorentz group are nonunitary.

• The relativistic quantum wave fields on the Minkowski spacetime do

constitute the only available way to actually implement the unitary
irreducible representations of the Poincaré group, that will be therof
associated to the elementary particles obeying the general principles of
the quantum mechanics and of the special theory of relativity.

References

1. G.Ya. Lyubarskii
The Application of Group Theory in Physics
Pergamon Press, Oxford, 1960.

2. Mark Naı̈mark and A. Stern

Théorie des représentations des groupes
Editions Mir, Moscou, 1979.

3. Pierre Ramond
Field Theory: A Modern Primer
Benjamin, Reading, Massachusetts, 1981.

51
Chapter 2

The Action Functional

2.1 The Classical Relativistic Wave Fields

2.1.1 Field Variations
As we shall see in the next chapters, in order to build up explicitly the
infinite dimensional Hilbert spaces that carry on all the irreducible unitary
representations of the Poincaré group, which describe the quantum states of
the elementary particles we detect in Nature, we will perform the canonical
quantization of some classical mechanical systems with an infinite number of
degrees of freedom. These mechanical systems consist in a collection of real
or complex functions defined on the Minkowski’s spacetime M

R
uA (x) : M −→ (A = 1, 2, . . . , N )
C
with a well defined transformation law under the action of the Poincaré
group IO(1, 3) . We shall call these systems the classical relativistic wave
fields. More specifically, if we denote the elements of the Poincaré group by
g = (Λ, a) ∈ IO(1, 3) , where Λ ∈ L is an element of the Lorentz group while
a µ specify a translation in the four dimensional Minkowski spacetime M ,
we have
g
uA (x) 7−→ uA0 (x 0 ) ≡ uA0 (Λx + a)
uA0 (x 0 ) = [ T (Λ) ] AB uB (x)
 
= [ T (Λ) ] AB uB Λ −1 (x 0 − a) (2.1)

(Λ, a) ∈ IO(1, 3) (A, B = 1, 2, . . . , N )

where T (Λ) are the operators of a representation of the Lorentz group of finite
dimensions N . This means that, if the collection of the wave field functions

52
at the point P ∈ M of coordinates x µ is given by uA (x) (A = 1, 2, . . . , N ) in
a certain inertial frame S , then in the new inertial frame S 0 , related to S by
the Poincaré transformation (Λ, a) ∈ IO(1, 3) , the spacetime coordinates of
P will be changed to x 0 = Λx+a and contextually the wave field functions will
be reshuffled as uA0 (x 0 ) (A = 1, 2, . . . , N ) because the functional relationships
will be in general frame dependent.
We can always represent the collection of the classical relativistic wave
field functions as an N –component column vector
 

 u1 (x)  


 u 2 (x) 


.
 
. x∈M
 
u (x) = 

 .




u (x) 
 
 N −1

 
 
uN (x)


Then we can suitably introduce some finite quantities which are said to be
the total variation, the local variation and the differential of the classical
relativistic wave field u (x) according to

u 0 (x 0 ) − u (x) total variation (2.2)

u 0 (x 0 ) − u (x 0 ) local variation (2.3)
u (x 0 ) − u (x) differential variation (2.4)

so that we can write

u 0 (x 0 ) − u (x) = [ u 0 (x 0 ) − u (x 0 ) ] + [ u (x 0 ) − u (x) ]

that is the total variation is equal to the sum of the local and of the differential
variations for any classical relativistic wave field.

Consider a first order infinitesimal (passive) Lorentz transformation

Λ µν = δ µν + ε µν = δ µν + ε µρ g ρν | ε µν |  1 (2.5)

On the one hand, from the defining relation (1.19) of the Lorentz matrices
we obtain
0 = g µρ ε ρν + g νρ ε ρµ ⇔ ε µν + ε νµ = 0
that is, the infinitesimal parameters ε µν constitute an antisymmetric matrix
with six independent entries.
On the other hand, from the exponential formula (1.12) for the Lorentz
matrices we can write

Λ µν = δ µν + δαk (Ik ) µν + δβ k (Jk ) µν (2.6)

53
 | δαk |  1 | δβ k |  1 ( k = 1, 2, 3 )
If we set

δαj = − 21 εjkl δ ω kl δ ω kl = δαj ε jkl = δ ω kl = − δαj ε jkl

δβ k = δ ω k0 = δ ω 0k
ω ρσ + ω σρ = 0
where εjkl is the Levi-Civita symbol, totally antisymmetric in all of its three
indices and normalized as ε123 = + 1 = − ε123 , together with

Ij = 21 i εjkl Skl Jk = − i S 0k S ρσ + S σρ = 0 (2.7)

then we readily obtain

Λ µν = δ µν − 21 i δ ω ρσ (S ρσ ) µν (2.8)

For example, a passive rotation around the OZ axis corresponds to

α1 = α2 = 0 , α3 = − δ ω 12 = − ε12
Λ µν = δ µν + δ ω 12 (δ µ1 g 2ν − δ µ2 g 1ν )
 
 1 0 0 0 
 

 0 1 α3 0  
Λ(α3 ) =   (2.9)
−
 

 0 α 3 1 0 


 
0 0 0 1
Moreover, a passive boost along the OX axis is described by

β1 = δ ω 01 , β 2 = β3 = 0
Λ µν µ 01 µ
= δ ν + δ ω (δ 0 g 1ν − δ µ1 g 0ν )
1 −β1 0 0 
 

−β1 1 0 0 

 
 
Λ(β1 ) = 
 
 (2.10)


 0 0 1 0  

 
0 0 0 1

A comparison between the expressions (2.5) and (2.8) yields

ε µλ g λν = − ε ρσ δ µσ g ρν
= 21 ε ρσ (δ µρ g σν − δ µσ g ρν )
= − 12 i δ ω ρσ (S ρσ ) µν (2.11)

Hence

ε ρσ = δ ω ρσ (S ρσ ) µν = i δ µρ g σν − δ µσ g ρν


(2.12)

54
The six matrices S ρσ realize the relativistic spin angular momentum tensor of
the irreducible vector representation τ 1 1 of the Lorentz group. By the way,
2 2
it is worthwhile to remark that the components Sjk are hermitean matrices
and generate rotations, while the components S 0k are antihermitean matrices
and generate boosts.

Turning back to the three kinds of infinitesimal variation we can write

x 0 µ ≈ x µ + δ x µ = x µ + δ ω µν g νρ x ρ + δ ω µ (2.13)

δ ω µν + δ ω νµ = 0 | δ ω µν |  1 | δωµ |  | xµ |

∆ u (x) ≡ u 0 (x + δ x) − u (x)
≡ [ u 0 (x + δ x) − u (x + δ x) ] + [ u (x + δ x) − u (x) ]
≡ δ u (x + δ x) + d u (x)
= δ u (x) + δ x µ ∂µ δ u(x) + · · · + d u (x)
= δ u (x) + δ x µ ∂µ u (x) + O (δ u δ x) (2.14)

so that we can safely write the suggestive symbolic relation

∆ = δ + d = δ + δ x µ ∂µ (2.15)

among the first order infinitesimal variations together with

∆ u (x) ≈ δ u (x) + d u (x) = δ u (x) + δ x · ∂ u (x) (2.16)

It is important to remark that, by definition, the local variations do commute

with the four gradient differential operator

∂µ δ u (x) = δ ∂µ u (x) ⇔ [ δ , ∂µ ] = 0 (2.17)

Notice that the infinitesimal form of the Poincaré transformations for the
spacetime coordinates can be written in terms of the hermitean generators
(1.39), that is
δ x µ = 12 i δ ω ρσ L ρσ x µ − i δ ω ρ P ρ x µ
Here below we shall analyse the most relevant cases.

2.1.2 The Scalar and Vector Fields

1. Scalar field: the simplest case is that of a single invariant real function

φ : M −→ R φ 0 (x 0 ) = φ (x) (2.18)

55
 so that

∆ φ (x) = 0 ⇔ δ φ (x) = − d φ (x) = − δ x · ∂ φ (x)

From the infinitesimal change (2.13) we get the local variation

δ φ (x) = − δ ω µν g νρ x ρ ∂µ φ(x) − δ ω µ ∂µ φ(x)

= − 12 i δ ω µν L µν φ (x) + i δ ω µ Pµ φ (x) (2.19)

where, see the definition (1.39),

Pµ = i ∂ µ L µν ≡ x µ Pν − x ν Pµ

whence it follows that for a scalar field we find by definition

M µν φ (x) ≡ L µν φ (x) ⇔ S µν φ (x) ≡ 0

i.e. the scalar field carries relativistic orbital angular momentum but
not relativistic spin angular momentum.
It is worthwhile to consider also the pseudoscalar field, which are odd
with respect to improper orthochronus Lorentz transformations, i.e.

φe 0 (Λx + a) = (det Λ) φe (x) = − φe (x) ∀ Λ ∈ L↑− (2.20)

Complex scalar (pseudoscalar) field are complex invariant functions

with scalar (pseudoscalar) real and imaginary parts.

2. Vector field : a relativistic (covariant) vector wave field is defined by

the transformation law under the Poicaré group that reads

Vµ0 (x 0 ) = Vµ0 (Λx + a) ≡ Λ µν Vν (x) (2.21)

The above transformation law can be obviously generalized to the aim

of defining the arbitrary relativistic tensor wave fields of any rank with
many contravariant and covariant Minkowski’s indices
0 αβ...
Tµν... (x 0 ) = Tµν...
0 αβ...
(Λx + a)
≡ Λ µρ Λ νσ . . . Λ αλ Λ βκ . . . Tρσ...
λκ...
(x) (2.22)

It follows therefrom that the components of a vector field transform

according to the irreducible vector representation τ 1 1 of the Lorentz
2 2
group. For infinitesimal Lorentz transformations we can write

Λ µν = δ µν + δω µρ g ρν δω µρ + δω ρµ = 0

56
and consequently
∆ Vµ (x) = δ ω ρσ g µρ Vσ (x)
≡ − 21 i δ ω ρσ ( S ρσ )µν Vν (x) (2.23)
the generators of the total variation of the covariant vector field under a
Poincaré transformation being the relativistic spin angular momentum
matrices (2.7), the action of which actually reads
( S ρσ )µν Vν (x) = i g µρ Vσ (x) − i g σµ Vρ (x)
≡ S ρσ ∗ Vµ (x) (2.24)
in which
( S ρσ )µν = i gµρ δ σν − i gµσ δ ρν (2.25)
The above expression can be readily checked taking the four gradient
of a real scalar field. As a matter of fact we obtain
∂µ φ (x) ≡ Vµ (x) (2.26)
so that from eq. (2.19) we find the infinitesimal transformation law
∆ Vµ (x) = ∆ ∂µ φ (x)
= [ ∆ , ∂µ ] φ (x) + ∂µ ∆ φ (x)
= [ ∆ , ∂µ ] φ (x) (2.27)
Now we have
[ ∆ , ∂µ ] = [ δ , ∂µ ] + [ δx · ∂ , ∂µ ]
= δ ω λν g νρ [ x ρ , ∂µ ] ∂λ
= − δ ω λν g νρ δ µρ ∂λ
= δ ω λν g λµ ∂ ν (2.28)
and thereby
∆ ∂µ φ (x) = δ ω ρσ g µρ ∂ σ φ (x)
whence eq. (2.23) immediately follows.
From the symbolic relation (2.15) we obtain the expression for the local
variation of a relativistic covariant vector wave field
δ Vρ (x) = ∆ Vρ (x) − δ x µ ∂µ Vρ (x) = − 12 i δ ω µν S µν ∗ Vρ (x)
− δ ω µν x ν ∂µ Vρ (x) − δ ω µ ∂µ Vρ (x)
= − 12 i δ ω µν M µν ∗ Vρ (x) + i δ ω µ Pµ Vρ (x) (2.29)

57
 M µν = L µν + S µν = x µ Pν − x ν Pµ + S µν

Well–known examples of vector and tensor wave fields are the vector
potential and the field strength of the electromagnetic field

A 0µ (x 0 ) = Λ µν A ν (x)
∆ A µ (x) = δ ω λν g µλ A ν (x)
0
F µν (x 0 ) = Λ µρ Λ νσ F ρσ (x)
i
∆ F µν (x) = δ ω λρ [ g µλ F ρν (x) + g νλ F µρ (x) (2.30)

A tensor wave field of any rank with many contravariant and covariant
indices will exploit a local variation in accordance to the straightforward
generalization of the infinitesimal change (2.29). In particular, the
action of the relativistic spin matrix on a tensor wave field will be
the (algebraic) sum of expressions like (2.7), one for each index. For
instance, the action of the spin matrix on the electromagnetic field
strength is given by

S ρσ ∗ F µν (x) = i g ρµ F σν (x) − i g σµ F ρν (x)

+ i g ρν F µσ (x) − i g σν F µρ (x) (2.31)

If we consider a full parity transformation (1.20) ΛP ∈ L↑− we have

A 0µ (Λ P x) = (Λ P ) µν A ν (x) (2.32)

that yields

A 00 (x0 , −x) = A 0 (x0 , x) A 0 (x0 , −x) = − A (x0 , x) (2.33)

Conversely, a covariant pseudovector wave field will be defined by the

transformation law

Veµ0 (Λ x) = (det Λ) Λ µν Veν (x) ∀ Λ ∈ L↑− (2.34)

so that under parity

Ve00 (x0 , −x) = − Ve0 (x0 , x) e 0 (x0 , −x) = V

V e (x0 , x) (2.35)

58
2.1.3 The Spinor Fields
The two irreducible fundamental representations τ 1 0 and τ 0 1 of the
2 2
homogeneous Lorentz group can be realized by means of SL(2, C) ,
i.e. the group of complex 2 × 2 matrices of unit determinant. The
SL(2, C) matrices belonging to τ 1 0 act upon the so called left Weyl’s
2
two–component spinors, whilst the SL(2, C) matrices belonging to τ 0 1
2
act upon the so called right Weyl’s two–component spinors.
In any neighbourhood of the unit element, the SL(2, C) matrices can
always be presented in the exponential form

Λ L ≡ exp 21 i σk (αk − i η k )

(2.36)
1
Λ R ≡ exp 2 i σk (αk + i η k ) (2.37)

in which
vk  
αk , β k = , η k = Arsh β k (1 − βk2 )−1/2 ( k = 1, 2, 3 )
c
are respectively the angular canonical coordinates, the relative velocity
components (1.21) and rapidity (1.22) parameters of the Lorentz group,
whereas σ k (k = 1, 2, 3) are the Pauli matrices. Notice that

σ2 σk σ2 = − σk∗ = − σk> (k = 1, 2, 3) (2.38)

σj σk = δjk + i ε jkl σl {σj σk } = 2δjk

[ σj σk ] = 2 i ε jkl σl ( j, k, l = 1, 2, 3 ) (2.39)

It is clear that for βk = η k = 0 (k = 1, 2, 3) , i.e. for rotations, we have

that Λ L (α) = Λ R (α) ∈ SU (2) whereas for αk = 0 (k = 1, 2, 3) , i.e. for
boosts, the matrices Λ L , R (η) are hermitean and nonunitary.
Let us therefore introduce the relativistic two–component Weyl’s spinor
wave fields
   
ψL 1 (x)  ψR 1 (x) 
ψL (x) ≡  ψR (x) ≡  (2.40)

  
 
ψL 2 (x) ψR 2 (x)
 

which, by definition, transform according to

ψL0 (x 0 ) = ΛL ψL (x) ψR0 (x 0 ) = ΛR ψR (x) (2.41)

59
The infinitesimal form of the above transformation laws give rise to the
total variations
∆ ψL (x) = 12 i (σj δα j − i σk δβ k ) ψL (x)
= − 21 i σj 12 εjkl δ ω kl + i σk δ ω 0k ψL (x)

≡ − 21 i δ ω ρσ (Sρσ )L ψL (x) (2.42)

and quite analogously
∆ ψR (x) = 21 i (σj δα j + i σk δβ k ) ψR (x)
= − 21 i σj 21 εjk` δ ω k` − i σk δ ω 0k ψR (x)

≡ − 12 i δ ω ρσ (Sρσ )R ψR (x) (2.43)

whence we identify
1
(Sk` )L = ε
2 jk`
σj = (Sk` )R (2.44)
1
(S0k )L = 2
i σk = (S 0k )R (2.45)
From the symbolic relation (2.15) we can easily obtain the expression
for the local variation of both relativistic Weyl’s spinor fields.

The SL(2, C) matrices do satisfy some important properties :

† †
(a) Λ−1
L = ΛR Λ−1
R = ΛL
(b) σ2 ΛL σ2 = ΛR∗ σ2 ΛR σ2 = ΛL∗
(c) ΛL> = σ2 Λ−1
L σ2 ΛR> = σ2 Λ−1
R σ2

Proof
(a) We have
 
i
ΛR† = exp − σk (αk − i ηk )
2
 
i
= exp σk (− αk + i ηk )
2
= ΛL (− α, − η) = ΛL−1
 
i
ΛL† = exp − σk (αk + i ηk )
2
 
i
= exp σk (− αk − i ηk )
2
= ΛR (− α, − η) = ΛR−1

60
(b) From the definition of the exponential of a matrix we get
 
i
σ2 ΛL σ2 = σ2 exp σk (αk − i ηk ) σ2
2
∞  n
X i 1
= σ2 σk1 σk2 · · · σkn σ2
n=0
2 n!
× (αk1 − i η k1 ) (αk2 − i η k2 ) · · · (αkn − i η kn )
∞  n
X i 1 ∗ ∗
= σk1 σk2 · · · σk∗n
n=0
2 n!
× (− αk1 + i η k1 ) (− αk2 + i η k2 ) · · · (− αkn + i η kn )
 
i ∗
= exp σ (− αk + i ηk )
2 k
  ∗
i
= exp σk (αk + i ηk ) = ΛR∗
2

and in the very same way we prove that σ2 ΛR σ2 = ΛL∗ .

(c) Finally we get
†
σ2 Λ−1
L σ2 = σ2 ΛR σ2
 
i
= σ2 exp − σk (αk − i ηk ) σ2
2
  >
i
= exp σk (αk − i ηk ) = Λ>
L
2

and repeating step–by–step we prove that σ2 Λ−1 >

R σ2 = ΛR q.e.d.

The above listed relations turn out to be rather useful to single out
Lorentz invariant combinations out of the Weyl’s spinors. Consider for
instance
 0
σ2 ψL∗ = σ2 (ΛL ψL )∗ = σ2 Λ∗L σ2 σ2 ψL∗ = ΛR σ2 ψL∗

which means that σ2 ψL∗ ∈ τ 0 1 and correspondingly σ2 ψR∗ ∈ τ 1 0 .

2 2

We show now that the antisymmetric combination of the two Weyl

representations of the Lorentz group transforms according to the scalar
representation. As a matter of fact, on the one hand we find

χ>
L σ2 ψL = −i (χL 1 ψL 2 − χL 2 ψL 1 ) (2.46)

61
On the other hand we have

(χ> 0 > > >

L σ2 ψL ) = χL ΛL σ2 ΛL ψL = χL σ2 ψL ∈ τ 0 0 (2.47)

From the antisymmetry of the above invariant combination, it follows

that ψL> σ2 ψL ≡ 0 for ordinary c–number valued Weyl’s left spinors.
If we now take χL ≡ σ2 ψR∗ then we obtain the complex invariants

I = ψR>∗ σ2> σ2 ψL = − ψR† ψL I∗ = − ψL† ψR (2.48)

We can build up a four vector out of a single Weyl left spinor fields. To
this purpose, let me recall the transformation laws of a contravariant
four vector under passive infinitesimal boosts and rotations respectively

δ V 0 = ε0k V k = − ε 0k V k = − δ βk V k (2.49)
δ V j = ε j0 V 0 = ε j0 V 0 = − δ βj V 0 (2.50)
δ V j = ε jk V k = − ε jk V k = − ε jk` V k δ α` (2.51)

Consider in fact the left combination

ψL† (x) σ µ ψL (x) σ µ ≡ (1, − σk ) ( k = 1, 2, 3 )

Under a passive infinitesimal Lorentz transformation the total variation

(2.42) yields
 
∆ ψL† (x) σ µ ψL (x) =
− 1
2
i ψL† (x) σk σ µ ψL (x) (δ α k + i δ β k )
+ 1
2
i ψL† (x) σ µ σk ψL (x) (δ α k − i δ β k )
= i ψL† (x) [ σ µ , σk ] ψL (x) δ α k + 21 ψL† (x) {σk , σ µ } ψL (x) δ β k
1
2
ψL† (x) σk ψL (x) δ β k

(µ = 0)
=
− ψL† (x) ψL (x) δ β j + ε jk` ψL† (x) σ` ψL (x) δαk (µ = j)
= − 21 i δ ω ρσ (S ρσ ) µν ψ L† (x) σ ν ψL (x) (2.52)

which means that the left-handed combination

VLµ (x) ≡ ψL† (x) σ µ ψL (x) = VLµ † (x) σ µ = (1, − σk )

transforms under the Lorentz group like a contravariant real vector

field, that is

VL0 µ (x 0 ) = Λµν ψL† (x) σ ν ψL (x) = Λµν VLν (x) (2.53)

62
In the very same way one can see that the right combination

VRµ (x) ≡ ψR† (x) σ̄ µ ψR (x) = VRµ † (x) σ̄ µ = (1, σk )

transform under the Lorentz group like a contravariant real vector field

VR0 µ (x 0 ) = Λµν ψR† (x) σ̄ ν ψR (x) = Λµν VRν (x) (2.54)

In conclusion we see that the following relationships hold true : namely,

Λ L† σ µ Λ L = Λ µν σ ν Λ R† σ̄ µ Λ R = Λ µν σ̄ ν (2.55)

It becomes now easy to build up the Lorentz invariant real kinetic terms

TL = 1
2
ψL† (x) σ µ i ∂µ ψL (x) − 1
2
i ∂µ ψL† (x) σ µ ψL (x)
↔
≡ 1
2
ψL† (x) σ µ i ∂ µ ψL (x) (2.56)
TR = 1
2
ψR† (x) σ̄ µ i ∂µ ψR (x) − 1
2
i ∂µ ψR† (x) σ̄ µ ψR (x)
↔
≡ 1
2
ψR† (x) σ̄ µ i ∂ µ ψR (x) (2.57)

When the full Lorentz group is a concern, we have already seen (1.24)
that it is necessary to consider the direct sum of the two inequivalent
irreducible Weyl’s representation. In so doing, we are led to the so
called four components bispinor or Dirac relativistic spinor wave field
 
   ψL 1 (x) 
 
ψL (x)   ψL 2 (x) 
ψ(x) ≡ 
 
= (2.58)

  
 

ψR (x) 
 ψR 1 (x) 


 
ψR 2 (x)

The full parity transformation, also named space inversion,

P : ψL , R ↔ ψR , L

is then represented by the 4 × 4 matrix

 
   0 0 1 0 
 
 0 1   0 0 0 1 
γ0 ≡ 
 
=
  

1 0 1 0 0 0
 
 


 

0 1 0 0

63
so that
 
0 0 0 0 0 ψR (x)
ψ (x ) = ψ (x , − x) = (P ψ)(x) = γ ψ (x) =   (2.59)

 

ψL (x)

The left and right Weyl’s components can be singled out by means of
the two projectors
1 1
PL ≡ 2
(I − γ 5 ) PR ≡ 2
(I + γ 5 )

where
 
   1 0 0 0 
 
 1 0  
 0 1 0 0 

I= =
  
 (2.60)
0 1 0 0 1 0
 
 


 

0 0 0 1
−1 0 0
 
   0 
 −1 0   0 −1 0
 
 0 
γ5 = γ 5 ≡ 

=
  
 (2.61)
0 1 0 0 1 0
 
 


 

0 0 0 1

Starting from the Lorentz invariant complex bilinears in the Weyl

spinors (2.48), one can easily build up the Lorentz and parity invariant
real bilinear in the Dirac spinors, viz.,

I + I∗ = − ψR† ψL − ψL† ψR = − ψ † γ 0 ψ ≡ − ψ̄ψ (2.62)

where ψ̄ ≡ ψ † γ 0 is said to be the adjoint spinor. In the very same way

we can construct the parity even and Lorentz invariant Dirac kinetic
term
↔
TD = 12 ψ̄(x) γ µ i ∂ µ ψ(x) = TL + TR (2.63)

which is real, where we have eventually introduced the set of matrices

 
   0 0 1 0 
 
0 0 1  
 0 0 0 1  
γ =   =  (2.64)

  
 

1 0 
 1 0 0 0 


 
0 1 0 0
 
   0 0 0 1 
 
1 0 σ1  
 0 0 1 0  
γ =   =  (2.65)

   
−σ1 0 −1



 0 0 0 


−1 0 0 0
 

64
 0 −i 
 
   0 0
 
0 σ2   0 0
 i 0 
γ2

=   =  (2.66)
  
−σ2 0
 
0 i 0 0 
 
 
 
−i 0
 
0 0
 
   0 0 1 0 
0 −1 
 
0 σ3 
 0 0
γ3
 
=   =  (2.67)
  
−σ3 0 −1 0
 
0 0 
 
 

 

0 1 0 0
−1 0
 
   0 0 
 −1 0  0 −1
 

 0 0 

γ5 =   = 
  
 (2.68)
0 1 0 0 1 0 
 
 

 

0 0 0 1
The above set of five 4 × 4 matrices are said to be the Dirac matrices
in the Weyl or chiral or even spinorial representation. The gamma
matrices do satisfy the so called Clifford algebra

{γ µ , γ ν } = 2g µν {γ µ , γ 5 } = 0 (2.69)

in which
γ 5 ≡ i γ 0γ 1γ 2γ 3
Notice that we have the hermitean conjugation properties

γ µ † = γ 0γ µγ 0 γ 5† = γ 5 (2.70)

and moreover
 µ

0 σ̄
γµ =  σ̄ µ ≡ (1, σ) σ µ ≡ (1, − σ)
 

 µ
σ 0

 µ 
0  σ
µ 0
γ γ =  ≡ αµ


0 σ̄ µ


We have already seen the transformation laws of the left–handed (2.42)

and right–handed (2.43) Weyl’s spinors under the restricted Lorentz
group. If we consider an infinitesimal boost δ ω kl = 0 (k, l = 1, 2, 3)
both transformation laws can be written in terms of a Dirac spinor as

∆ ψ(x) = 41 δ ω 0k (γ 0 γ k − γ k γ 0 ) ψ(x) (k = 1, 2, 3) (2.71)

Analogously, the transformation laws under an infinitesimal rotation

δ ω 0k = 0 (k = 1, 2, 3) can be written together in the form

∆ ψ(x) = 18 δ ω jk (γ j γ k − γ k γ j ) ψ(x) ( j, k = 1, 2, 3) (2.72)

65
This means that, if we introduce
i µ ν
( S µν )D ≡ σ µν = [γ , γ ] (2.73)
4
we eventually obtain that the transformation law for the Dirac spinors
under the restricted Lorentz group becomes
i µν
∆ ψ(x) = − σ ψ(x) δ ω µν (2.74)
2
which leads to identify the boost and rotation generators for the Dirac
bispinors as
 
0k i k 1  σ k 0
( S )D = [ β , γ ] = 


4 2i 0 − σ
 
k

and respectively
 
i  σ` 0
( S jk )D = [ γ j , γ k ] = 12 ε jk`   ≡ 21 ε jk` Σ `


4 0 σ`


Hence the covariant spin angular momentum tensor operator for the
relativistic Dirac spinor wave field reads

σ µν = 1
4
i g µλ g νκ [ γ λ , γ κ ] (2.75)

the spatial components being hermitean whilst the spatial temporal

components being antihermitean, that is

( σ µν )† = γ 0 σ µν γ 0 σ µ ν = γ 0 σ †µ ν γ 0 (2.76)

By construction, the six components of the spin angular momentum

tensor of the Dirac field enjoy the Lie algebra of the Lorentz group

[ σ µν , σ λκ ] = − i g µλ σ νκ + i g µκ σ νλ − i g νκ σ µλ + i g νλ σ µκ

It is worthwhile to remark that the above construction keeps true in

any D-dimensional space with a symmetry group O(m, n) in which

D =m+n m≥0 n≥0

m
X n
X
2
x = x2k − x2m+j
k=1 j=1

66
For any finite passive transformation of the restricted Lorentz group
we have
ψ 0 (x 0 ) = Λ 1 (ω) ψ (x) = exp − 21 i σ µν ω µν ψ (x)

(2.77)
2

As an example, for a boost along the positive OY axis we find

η η
Λ 1 (η) = exp − i σ 02 ω 02 = cosh − γ 0 γ 2 sinh = Λ†1 (η)

2
 2 2 2

 cosh η/2 i sinh η/2 0 0 
−
 

 i sinh η/2 cosh η/2 0 0 

= 
 
cosh η/2 − i sinh η/2 



 0 0 

 
0 0 i sinh η/2 cosh η/2

where sinh η = vy (1 − vy2 )−1/2 (vy > 0) .

From the hermitean conjugation property (2.76) we can write
Λ †1 (ω) = exp 12 i σ †µν ω µν

2
∞ n  
X 1 Y i
= γ 0 σ µ ρ γ 0 ω µ ρ
n=0
n! =1 2
∞  n
X 1 i
= γ 0
( σ µν ω µν )n γ 0
n=0
n! 2
= γ 0 exp 12 i σ µν ω µν γ 0


= γ 0 Λ−1
1 (ω) γ
0
= γ 0 Λ 1 (− ω) γ 0 (2.78)
2 2

which entails in turn the two further relations

γ 0 Λ †1 (ω) γ 0 = Λ−1
1 (ω) = Λ 1 (− ω) (2.79)
2 2 2
 †
†
Λ−1
1 (ω) γ 0 Λ−1 0 −1
1 (ω) = Λ 1 (− ω) γ Λ 1 (ω) = γ
0
(2.80)
2 2 2 2

From the Lorentz invariance of the Dirac real kinetic term (2.63) it
follows that the bilinear
ψ̄(x) γ µ ψ(x) ≡ J µ (x)
transforms as a contravariant four vector and is thereby named the
Dirac vector current. Hence, by making use of equation (2.79), the
finite transformation law immediately follows, viz.,
Λ−1
1 (ω) γ
λ
Λ 1 (ω) = Λλκ γ κ (2.81)
2 2

γ λ = Λλκ Λ 1 (ω) γ κ Λ−1

1 (ω) (2.82)
2 2

67
As we have seen above, it turns out that σ2 ψL∗ ∈ τ 0 1 while σ2 ψR∗ ∈
2
τ 1 0 . Thus, we can build up the charge conjugated spinor of a given
2
relativistic Dirac wave field ψ(x) as follows :
 ∗

c ∗ σ 2 ψ
ψ 7→ ψ = Cψ =  R 
C = γ2 (2.83)
 
− σ2 ψL∗
 

Notice that (ψ c ) c = γ 2 (γ 2 ψ ∗ )∗ = ψ . As a consequence, starting from a

solely left or right handed Weyl spinor, it is possible to build up charge
selfconjugated Majorana bispinors
 
ψ L
χL =   = χLc
 

− σ2 ψL∗

 ∗

σ2 ψ
χR =   = χRc
R 
(2.84)
 
ψR


Ettore Majorana
(5 Agosto 1906, Catania, via Etnea 251 – 26 marzo 1938, ?)
Teoria simmetrica dell’elettrone e del positrone
Il Nuovo Cimento, volume 14 (1937) pp. 171-184

The Majorana charge selfconjugated spinors are Weyl spinors in a four

components form and thereby their field degrees of freedom content is
half of that one of a Dirac spinor wave field.

68
2.2 The Action Principle
In the previous sections we have seen how to build up Poincaré invariant
expressions out of the classical relativistic wave fields corresponding to the
irreducible tensor and spinor representations of the inhomogeneous Lorentz
group. The requirement of Poincaré invariance will ensure that these classical
field theories will obey the axioms of the special theory of relativity.
The general properties that will specify the action for the collection of
the classical relativistic wave field functions uA (x) (A = 1, 2, . . . , N ) will be
assumed in close analogy with the paradigmatic case of the electromagnetic
field.
1. The action integrand L(x) is called the Lagrange density or lagrangian
for short: in the absence of external preassigned background fields,
the lagrangian can not explicitely depend on the coordinates x µ , so
as to ensure spacetime translation invariance, and must be a Lorentz
invariant to ensure that the corresponding theory will obey the axioms
of special relativity.

2. To fulfill causality, the differential equations for the field functions must
be at most of the second order in time, in such a way that the related
Cauchy problem has a unique solution. Classical field theories described
by differential equations of order higher than the second in time will
typically develop acausal solutions, a well known example being the
Abraham–Lorentz equation 1 of electrodynamics, which is a third order
in time differential equation that encodes the effects of the radiation
reaction and shows acausal effects such as preacceleration of charged
particles yet to be hit by radiation.

3. The wave equations for all the fundamental fields that describe matter
and radiation are assumed to be partial differential equations and not
integro–differential equations, which do satisfy Lorentz covariance in
accordance with the special theory of relativity : as a consequence
the lagrangian must be a Lorentz invariant local functional of the field
functions and their first partial derivatives
 
L(x) = L uA (x), ∂µ uA (x)

In the classical theory the action has the physical dimensions of an

angular momentum, i.e. g cm2 s−1 , or equivalently of the Planck’s
1
M. Abraham, Ann. Phys. 10, 105 (1903), H. A. Lorentz, Amst. Versl. 12, 986 (1904);
cfr. [1] pp. 578–610.

69
 constant. In the natural unit system the action is dimensionless and
the lagrangian in four spacetime dimensions has natural dimensions of
cm−4 .

4. According to the variational principle of classical mechanics, the action

must be real and must exhibit a local minimum in correspondence of
the Euler–Lagrange equations of motion : in classical physics complex
potentials lead to absorption, i.e. disappearance of matter into nothing,
a phenomenon that will not be considered in the sequel – it will be found
a posteriori that a real classical action is crucial to obtain a satisfactory
quantum field theory where the total probability is conserved.

5. The astonishing phenomenological and theoretical success of the gauge

theories in the construction of the present day standard model of the
fundamental interactions in particle physics does indeed suggest that
the action functional will be invariant under further symmetry groups
of transformations beyond the inhomogeneous Lorentz group. Such a
kind of transformations does not act upon the spacetime coordinates
and will be thereby called internal symmetry groups of transformations.
These transformations will involve new peculiar field degrees of freedom
such as the electric charge, the weak charge, the colour charge and
maybe other charges yet to be discovered. In particular, gauge theories
are described by action functionals wich are invariant under local –
i.e. spacetime point dependent – transformations among those internal
degrees of freedom.

Consider the action

Z tf Z  
S(ti , tf ; [ u ]) = d x0 d x L u(x), ∂µ u(x)
ti

where we shall denote by

u(x) = {uA (x) | A = 1, 2, . . . , N , x = (x0 , x) , ti ≤ x0 ≤ tf , x ∈ R3 }

the collection of all classical relativistic local wave fields. The index A =
1, 2, . . . , N < ∞ runs over the Lorentz group as well as all the internal
symmetry group representations, so that we can suppose the local wave field
component to be real valued functions.
We recall that, by virtue of the principle of the least action, the field
variations are assumed to be local and infinitesimal

u(x) 7→ u0 (x) = u(x) + δu(x) | δu(x) |  | u(x) |

70
and to vanish at the initial and final times ti and tf

δu(ti , x) = δu(tf , x) = 0 ∀ x ∈ R3 (2.85)

A local variation with respect to the wave field amplitudes gives

Z tf Z  
δS = d t d x δ L u(x), ∂µ u(x)
t
Z itf Z  
δL δL
= dt dx δu(x) + δ ∂µ u(x)
ti δu(x) δ∂µ u(x)
The local infinitesimal variations do satisfy by definition

δ ∂µ u(x) = ∂µ δu(x) ⇒ [ δ , ∂µ ] = 0

so that
Z tf Z  
δL δL
δS = d x0 d x − ∂µ δu(x)
t δu(x) δ∂µ u(x)
Z itf Z h δL i
+ d x 0 d x ∂µ δu(x)
ti δ∂µ u(x)
The very last term can be rewritten, using the Gauß theorem, in the form
Z h δL itf Z tf Z h δL i
dx δu(x) + d x0 d x ∇ · δu(x)
δ ∂0 u(x) ti ti δ ∇u(x)
Z tf Z
2
h δL i
= lim R d x0 d Ω b r· δu(x0 , R, Ω)
R→∞ ti δ ∇u(x0 , R, Ω)

where R is the ray of a very large sphere centered at x = 0 , Ω = (θ, φ) is the

solid angle in the three dimensional space and b
r is the radial unit vector, i.e.
the exterior unit normal vector to the sphere. If we assume the asymptotic
radial behaviour

lim R2 [ δL/δ ∂ r u(x0 , R, Ω) ] δu(x0 , R, Ω) = 0 ∀x0 ∈ [ ti , tf ] (2.86)

R→∞

where ∂ r ≡ br·∇ is the radial derivative, then the above boundary term indeed
disappears and consequently, from the arbitrariness of the local variations
δu(x) , we eventually come to the Euler–Lagrange equations of motion for
the classical relativistic wave field
δL δL
∂µ − = 0 (2.87)
δ∂µ u(x) δu(x)

71
2.3 The Noether Theorem
For the construction of the constants of motion in field theory we shall use
Noether theorem :
Amalie Emmy Noether
Erlangen 23.03.1882 – Brynn 14.04.1935
Invariante Varlationsprobleme, Nachr. d. König. Gesellsch. d. Wiss.
Göttingen, Math–phys. Klasse (1918), 235–257
English translation M. A. Travel
Transport Theory and Statistical Physics 1 (1971), 183–207 .
This theorem states that to every continuous transformation of coordinates
and fields which makes the variation of the action equal to zero there always
corresponds a definite constant of motion, i.e. a combination of the field
functions and their derivatives which remains conserved in time. Such a
transformation of coordinates and fields will be called a continuous symmetry
and will correspond to some representation of a Lie group of transformations
of finite dimensions n .
In order to prove Noether theorem we shall consider the infinitesimal
transformation of coordinates

xµ 7→ x 0 µ = x µ + δx µ δx µ = X aµ δ ω a (2.88)

with coefficients Xaµ that may depend or not upon the spacetime points and
s spacetime independent infinitesimal parameters

δ ωa (a = 1, 2, . . . , s)

which, for example, in the case of a transformation from the inhomogeneous

Lorentz group include the infinitesimal translations δ ω µ and homogeneous
Lorentz transformations δ ω µν :

δx µ = δ ω µ + x ν δ ω µν δ ω µν = − δ ω νµ

In general we shall suppose to deal with a collection of N field functions

uA (x) (A = 1, 2, . . . , N )

with a well defined infinitesimal transformation law under the continuous

symmetry (2.88) that may be written in the form

uA (x) 7→ uA0 (x 0 ) = uA (x) + ∆ uA (x)

∆ uA (x) = uA0 (x 0 ) − uA (x) = Y AB

a
uB (x) δ ω a (2.89)

72
where the very last variation is just the already introduced total variation
(2.2) of the field function. For example, in the case of an infinitesimal Lorentz
transformation, the definitions (2.25), (2.42), (2.43), (2.74) and (2.75) yield
a
∆ uA (x) = Y AB uB (x) δ ω a = − 12 i (Sµν )AB uB (x) δ ω µν (2.90)

The jacobian that does correspond to the infinitesimal transformation of

coordinates is
δJ = δ det k ∂x 0 µ /∂x ν k = ∂µ δx µ
so that we can eventually write
Z tf Z  
∆S = d x0 d x L uA (x), ∂µ uA (x) ∂ λ δx λ
ti
Z tf Z  
+ d x0 d x ∆ L uA (x), ∂µ uA (x)
ti
Z tf Z  
= d x0 d x L uA (x), ∂µ uA (x) ∂ λ δx λ
ti
Z tf Z  
+ d x0 d x ∂ λ L uA (x), ∂µ uA (x) δx λ
ti
Z tf Z  
+ d x0 d x δL uA (x), ∂µ uA (x)
ti

where use has been made of the relation (2.15). If the Euler-Lagrange wave
field equations (2.87) are assumed to be valid besides the radial asymptotic
behaviour (2.86), then we can recast the local variation of the lagrangian in
the form  
  δL
δ L uA (x), ∂µ uA (x) = ∂µ δuA (x)
δ ∂µ uA (x)
Hence we immediately obtain
Z tf Z h δL i
∆S = d x0 d x ∂µ L(x) δx µ + δuA (x) (2.91)
ti δ ∂µ uA (x)

Alternatively, we can always recast the local variations (2.3) in terms of the
total variations (2.2) so that
Z tf Z   
µ δL ν
∆S = d t d x ∂µ δ ν L(x) − ∂ν uA (x) δx
ti δ ∂µ uA (x)
Z tf Z  
δL
+ d t d x ∂µ ∆uA (x) (2.92)
ti δ ∂µ uA (x)

73
Consider now the infinitesimal symmetry transformations depending upon
constant parameters, i.e. spacetime independent, so that δ ω a = ∆ ω a :
∂ xµ
 
µ
δx ≡ a
δ ω a = Xaµ δ ω a (a = 1, 2, . . . , s)
∂ω
∆ uA (x) ≡ ( Y a ) AB uB (x) δ ω a (A = 1, 2, . . . , N )
then we have
Z tf Z
∆S
+ d x0 d x ∂µ Jaµ (x) = 0 (2.93)
∆ωa ti

where
h δL i
Jaµ (x) ≡ ∂ν uA (x) − L(x) δ νµ Xaν
δ ∂µ uA (x)
δL
− ( Y a ) AB uB (x) (2.94)
δ ∂µ uA (x)
are the Noether currents associated to the parameters ω a (a = 1, 2, . . . , s)
of the Lie group of global symmetry transformations. Suppose the action
functional to be invariant under this group of global transformations
(∆S / ∆ ω a ) = 0 (a = 1, 2, . . . , s)
Then from Gauß theorem we get
Z h i Z tf Z
0 0
0 = d x Ja (tf , x) − Ja (ti , x) + d x0 d x ∇ · J (x0 , x)
ti
Z h i
= d x Ja0 (tf , x) − Ja0 (ti , x)
Z tf Z
2
+ lim R d x0 d Ω b r · Ja (x0 , R, Ω)
R→∞ ti

where R is the ray of a very large sphere centered at x = 0 , Ω = (θ, φ) is the

solid angle in the three-dimensional space and br is the exterior unit normal
vector to the sphere, i.e. the radial unit vector. Once again, if we assume
the radial asymptotic behaviour
lim R 2 b
r · Ja (x0 , R, Ω) = 0 ∀x0 ∈ [ ti , tf ] (2.95)
R→∞

then the above boundary term indeed disappears and we eventually come to
the conservation laws
∆S d
=0 ⇔ Qa (t) = 0 (2.96)
∆ωa dt
74
where the conserved Noether charges are defined to be
Z
Qa ≡ d x Ja0 (x) (a = 1, 2, . . . , s) (2.97)

This is Noether theorem. Notice that the Noether current is not univocally
identified : as a matter of fact, if we redefine the Noether currents (2.94)
according to
J˜aµ (x) ≡ Jaµ (x) + ∂ ν A µa ν (x)
where A µν
a (x) (a = 1, 2, . . . , s) is an arbitrary set of antisymmetric tensor
fields
A µa ν (x) + A νa µ (x) = 0
then by construction

∂µ J˜aµ (x) = ∂µ Jaµ (x) = ∂ t Ja0 (t, r) + ∇ · Ja (t, r)

and the same conserved Noether charges (2.97) are obtained. Let us now
examine some important examples.

1. Spacetime translations

δx µ = δ ω µ a ≡ ρ = 0, 1, 2, 3

Xaν ≡ δ ρν ∆ uA (x) ≡ 0
because there is no change under spacetime translations for any classical
relativistic wave field. In this case the corresponding Noether’s current
yields the energy momentum tensor

Jaµ (x) 7→ T µρ (x)

δL
T µρ (x) = ∂ ρ uA (x) − L(x) g µρ (2.98)
δ ∂µ uA (x)

the corresponding conserved charge being the total energy momentum

four vector of the system
Z
Pµ = d x T µ0 (x) (2.99)

75
2. Lorentz transformations

δ x ν = δ ω νρ x ρ a ≡ { ρ σ } = 1, . . . , 6

and from (2.90)

Xaν ≡ 1
∆ uA (x) ≡ − 12 i ( S ρ σ ) AB uB (x) δ ω ρσ


2
x σ δ ρν − x ρ δ σν

In this case the corresponding Noether current yields the relativistic

total angular momentum density third rank tensor

Jaµ (x) = J µρ σ (x) = − 12 M µρσ (x)

M µρ σ (x) = x ρ T µσ (x) − x σ T µρ (x)

δL
− i ( S ρ σ )AB uB (x)
δ ∂µ uA (x)
def
= L µρσ (x) + S µρσ (x) (2.100)

where the third rank tensors

L µρσ (x) = − L µσρ (x) = x ρ T µσ (x) − x σ T µρ (x)

δL
S µρσ (x) = − S µσρ (x) = ( − i ) ( S ρ σ )AB uB (x)
δ ∂µ uA (x)
are respectively the relativistic orbital angular momentum tensor and
the relativistic spin angular momentum tensor of the wave field. The
corresponding charge is the total angular momentum antisymmetric
tensor of the system
Z
M µ ν = d x M 0µ ν (t, x) M µν + M ν µ = 0 (2.101)

Notice however that, as we shall see further on, in the case of Lorentz
boost transformations the angular momentum density tensor does not
satisfy the condition (2.95) owing to its explicit time dependence so
that, consequently, the related charge is not conserved in time.

3. Internal symmetries

Xaν ≡ 0 a
∆ uA (x) ≡ − TAB uB (x) δ ω a (2.102)

where T a (a = 1, 2, . . . , s) are the generators of the internal symmetry

Lie group in some given representation. In this case the corresponding

76
 Noether current and charge yields the internal symmetry current and
charge multiplets
δL
Jµa (x) = T a uB (x) (a = 1, 2, . . . , s) (2.103)
δ ∂µ uA (x) AB
Z
a
Q = d x J0a (x) (a = 1, 2, . . . , s) (2.104)

References
1. N.N. Bogoliubov and D.V. Shirkov
Introduction to the Theory of Quantized Fields
Interscience Publishers, New York, 1959.
2. L.D. Landau, E.M. Lifšits
Teoria dei campi
Editori Riuniti/Edizioni Mir, Roma, 1976.
3. Claude Itzykson and Jean-Bernard Zuber
Quantum Field Theory
McGraw-Hill, New York, 1980.
4. Pierre Ramond
Field Theory: A Modern Primer
Benjamin, Reading, Massachusetts, 1981.

2.3.1 Problems
1. Construct the energy momentum the total angular momentum tensor
densities for classical electromagnetism with no sources, i.e. for the
classical radiation field. 2
Solution. The classical Maxwell’s lagrangian is given by
L(x) = − 41 F µν (x) F µν (x) F µν (x) = ∂ µ A ν (x) − ∂ ν A µ (x)
hence from Noether theorem we get the canonical energy momentum
tensor
δL
T µρ (x) = ∂ ρ A ν (x) − L(x) g µρ
δ ∂µ A ν (x)
= − F µν (x) ∂ ρ A ν (x) + 14 F λν (x) Fλν (x) g µρ
2
See e.g. L.D. Landau, E.M. Lifšits, Teoria dei campi, Editori Riuniti/Edizioni Mir,
Roma, 1976, § 33 p. 114.

77
which is not symmetric with respect to µ and ρ . To remedy this, in
accordance with the general rule, we shall introduce
 
µρ µρ ρ µν
Θ (x) = T (x) + ∂ ν A (x) F (x)

and thereby, using the field equations ∂ ν F µ ν (x) = 0 , we can write

Θ µρ (x) = T µρ (x) + F µ ν (x) ∂ ν A ρ (x)

= 41 F λν (x) Fλν (x) g µρ − F µν (x) F ρν (x) = Θ ρ µ (x)

Notice that the above obtained symmetric energy momentum tensor of

the electromagnetic field is traceless

g µρ Θ µρ (x) = 0

Let us express explicitly the well known components :

Θjk = 12 δ jk (E 2 + B 2 ) − Ej Ek − Bj Bk (Maxwell stress tensor)

0j j jkl k l
Θ =S =ε E B S=E×B (Poynting vector)
1 2 2
Θ00 = 2 (E + B ) (energy density)

From equations (2.23) and (2.100) we derive the canonical total angular
momentum density of the radiation field

M µρσ = x ρ T µσ − x σ T µρ + i F µν ( S ρσ ) νλ Aλ
   
= x ρ Θ µσ − F µν ∂ ν A σ − x σ Θ µρ − F µν ∂ ν A ρ
− F µρ A σ + F µσ A ρ
= x ρ Θ µσ − x σ Θ µρ − ∂ ν [ F µν ( xρ A σ − xσ A ρ ) ]

The very last term does not contribute at all to the continuity equation
by virtue of the antisymmetry with respect to the pair of indices µ ν .
On the one hand, it turns out to be manifest that the total angular
momentum tensor can always appear to be of a purely orbital form.
On the other hand, it is clear that the spin angular momentum second
rank tensor of the radiation field is not conserved in time.

78
Chapter 3

The Scalar Field

3.1 Normal Modes Expansion

The simplest though highly nontrivial example of a quantum field theory
involves a real scalar field φ : M → R . As we shall see later on, the
real scalar quantum field will describe neutral spinless massive or massless
particles. The most general Poincaré invariant lagrangian density that fulfill
all the criteria listed in section 2.2 takes the general form
1
L= ∂ µ φ (x) ∂ µ φ (x) − V [ φ(x) ] (3.1)
2
where V(φ) is assumed to be a real analytic functional of its relativistic wave
field argument, that is
1 2 2 g λ
V [ φ(x) ] = υ m 4 + κ m 3 φ (x) ± m φ (x) + m φ 3 (x) + φ 4 (x) + · · ·
2 3! 4!
with m > 0 and υ, κ, g, λ ∈ R . Notice that the structure of the kinetic term
is such to fix the canonical dimension of the scalar field : [ φ ] = cm −1 = eV
in natural units ~ = c = 1 . Moreover, the important case of the negative
quadratic mass term does generally provide the remarkable phenomenon of
the spontaneous symmetry breaking, which lies at the heart of the present
day Standard Model of the High Energy Particle Physics and Quantum Field
Theory of the Fundamental Interactions.
The Euler-Lagrange equations of motion read
g λ
 φ(x) + κ m 3 ± m 2 φ (x) + m φ 2 (x) + φ 3 (x) + · · · = 0 (3.2)
2 6
while the conserved energy momentum tensor (2.98) and vector (2.99) are
T µν (x) = ∂ µ φ (x) ∂ ν φ (x) − g µν L(x) = T νµ (x) (3.3)

79
 Z Z h i
Pµ = dx T µ0 (x) = d x ∂0 φ(x)∂ µ φ(x) − g 0µ L(x) (3.4)

It is very easy to check that thanks to the Euler–Lagrange equations of

motion the energy momentum current is conserved, i.e.

∂µ T µν (x) = 0

Moreover, the stability requirement that the total energy must be bounded
from below, to avoid a colapse of the mechanical system, entails that the
analytic potential has to be a bounded from below functional of the real
scalar field. Finally, as it will be better focussed later on, the constraining
criterion of power counting renormalizability for the corresponding quantum
field theory will forbid the presence of coupling parameters with canonical
dimensions equal to positive integer powers of length. In such a circumstance,
the Lagrange density for a selfinteracting, stable, renormalizable real scalar
field theory reduces to
1 µν
g ∂µ φ(x)∂ ν φ(x) − V [ φ(x) ]
L= (3.5)
2
1 mg 3 λ
V [ φ(x) ] = V0 + κ m 3 φ (x) ± m2 φ 2 (x) + φ (x) + φ 4 (x)
2 3! 4!
in which V0 is a finite classical zero point energy, while κ, g and λ are real
numerical constants, with the dimensionless positive quartic coupling λ > 0
to guarantee stability. Hence we obtain the canonical conjugate momentum
field
δL δS
Π(x) ≡ = = ∂0 φ(x) = φ̇(x) (3.6)
δ∂0 φ(x) δ∂0 φ(x)
Z Z tf
L(x0 ) ≡ d x L(x) S≡ d x 0 L(x0 ) (3.7)
ti

To proceed further on we shall restrict ourselves to the paradigmatic case of

the quartic even functional potential
1 2 2 λ
V [ φ(x) ] = m φ (x) + φ 4 (x) (3.8)
2 4!
which corresponds to the selfinteracting theory of a neutral spinless field
in four spacetime dimensions, the so called λφ44 field theory, with a null
minimum classical energy. The classical Hamilton functional then reads
Z
H ≡ Π φ̇ − L = P = d x T 00 (x0 , x) = H0 + HI ≥ 0
0

80
 Z
1 h
2 2 2 2
i
H0 = d x Π (x) − φ(x)∇ φ(x) + m φ (x) (3.9)
2
Z
λ
HI = d x φ 4 (x) (3.10)
4!
the total momentum
Z
P≡− d x Π (x) ∇φ(x) (3.11)

and the total orbital angular momentum

Z h i
L µν = d x x µ T 0 ν (t, x) − x ν T 0 µ (t, x) (3.12)

We observe en passant that the above expressions are indeed obtained by

making use of the asymptotic radial behaviour
lim |x| φ(x0 , x) x · ∇φ(x0 , x) = 0
|x|→∞

by virtue of which and of the Gauß theorem we can write

Z Z
d x ∇φ(x) · ∇φ(x) = − d x φ(x) ∇ 2 φ(x)

It is also worthwhile to remark that the orbital angular momentum tensor

M λµν (x) = x µ T λν (x) − x ν T λµ (x)
does fulfill the continuity equation thanks to the symmetry of the energy
momentum tensor T ρσ (x) .
The equations of motion can also be written in the hamiltonian canonical
form. From the fundamental canonical Poisson’s brackets
{φ(t, x) , Π(t, y)} = δ (x − y)
{φ(t, x) , φ(t, y)} = 0 = {Π(t, x) , Π(t, y)}
one can immediately get

φ̇ (x) = { φ (t, x) , H } = δ H/δ Π (x)
Π̇ (x) = { Π (t, x) , H } = − δ H/δ φ (x)
and thereby
φ̇(x) = Π (x) (3.13)
λ 3
Π̇ (x) = φ̈(x) = ∇ 2 φ(x) − m2 φ(x) − φ (x) (3.14)
3!
81
 Consider the free scalar field theory in which the action, the lagrangian
and the hamiltonian are quadratic functional of the scalar field function,
so that the Euler–Lagrange equations of motion becomes linear and exactly
solvable: namely,
1
L0 = g µν ∂ µ φ (x) ∂ ν φ (x) − 21 m2 φ 2 (x)
2
(3.15)
Z h i
H0 = 2 d x Π 2 (x) − φ (x)∇2 φ (x) + m2 φ 2 (x)
1
(3.16)

 + m2 φ(x) = 0


(3.17)

which is the Klein-Gordon relativistic wave equation. To solve it, let us

introduce the Fourier decomposition
Z
−3/2
φ(x) ≡ (2π) d k φ̃(k) exp {− i k · x} φ̃∗ (k) = φ̃(−k) (3.18)

so that
(k 2 − m2 ) φ̃(k) = 0
the most general solution of which reads

φ̃(k) = f (k) δ(k 2 − m2 )

f (k) being an arbitrary complex function which is regular on the momentum

space hyperbolic manifold k 2 = m2 (k ∈ M) and which fulfill the reality
condition f (k) = f (−k) in order the scalar field to be real. Then we have
Z
−3/2
φ(x) ≡ (2π) d k f (k) δ(k 2 − m2 ) exp {− i k · x} (3.19)

and from the well known tempered distribution identities

1
δ(ax) = δ(x) θ(x) + θ(−x) = 1 (3.20)
|a|

one can readily obtain the decomposition

1
δ(k 2 − m2 ) = [ θ(k0 ) δ(k0 − ω k ) + θ(− k0 ) δ(k0 + ω k ) ]
2ω k

1
ω k = ω(k) ≡ (k 2 + m2 ) 2 = ω −k (3.21)

82
Hence
Z
−3/2
φ(x) = (2π) d k f (k) exp {− i k · x}
1
× [ θ(k0 ) δ(k0 − ω k ) + θ(− k0 ) δ(k0 + ω k ) ]
2ω k Z
−3/2 dk
= (2π) f (ω k , k) exp {− i x0 ω k + ik · x}
2ω k
Z
−3/2 dk
+ (2π) f (−ω k , −k) exp {i x0 ω k − ik · x}
2ω k
Z
dk f (ω k , k)
= · √ exp {− i x0 ω k + i k · x} + c.c.
[ 2ω k (2π)3 ] 1/2 (2ω k )
where we used the reality condition. It is very convenient to set

u k (x) ≡ [ 2ω k (2π)3 ] −1/2 exp {− i x0 ω k + i k · x} (3.22)

so that
Xh i
φ(x) = fk u k (x) + fk∗ u∗k (x)
k
X h i
Π(x) = i ωk fk∗ u∗k (x) − fk u k (x) (3.23)
k

with the suitable notations

Z
f (ω k , k) X
fk ≡ √ dk ≡
(2ω k ) k

which endorse the fact that (3.23) is nothing but the normal mode expansion
of the real scalar field. The wave functions u k (x) (k ∈ R) do constitute a
complete and orthonormal set, i.e. the normal modes of the real scalar free
field : namely, they fulfill the orthonormality relations

Z ↔
d x u∗h (x) i ∂ 0 u k (x) = δ(h − k) (3.24)
Z ↔
Z ↔
d x u h (x) i ∂ 0 u k (x) = d x u∗h (x) i ∂ 0 u∗k (x) = 0 (3.25)

as well as the closure relation

X 1
u k (x) u ∗k (y) = D (−) (x − y) (3.26)
k
i

83
where I have set
Z
(−) def dk 2 2


D (x − y) = i θ (k 0 ) δ k − m exp {− i k · (x − y)}
(2π) 3
From these orthonormality relations it is easy to invert the normal mode
expansions that yields

Z ↔
d x u∗k (x) i ∂ 0 φ(x) = f k (3.27)

As it is well known the normal mode decomposition is what we need to

diagonalize the energy momentum vector of the mechanical system. As a
matter of fact, from the equalities
Z h i
1 2 2 2
P0 = d x 2 Π (x) + ∇φ(x) · ∇φ(x) + m φ (x)
Z n h io
= d x 21 φ̇(x) φ̇(x) − φ(x) ∇ 2 φ(x) − m2 φ(x)
Z n o
= d x 21 φ̇ 2 (x) − φ(x) φ̈(x) (3.28)
Z
P = − d x Π (x)∇φ(x) (3.29)

by substituting the normal mode expansions (3.23) we immediately obtain,

for instance,
Z X h i
P = − dx i ω k fk∗ u∗k (x) − fk u k (x)
k
Xh i
× i h fh u h (x) − i h fh∗ u∗h (x)
h

and from the orthonormality relations we readily get

X  
P = 1
2
k fk fk∗ + fk∗ fk (3.30)
k
X  
P0 = 1
2
ω k fk fk∗ + fk∗ fk (3.31)
k

The normal mode complex amplitudes actually share the fractional canonical
dimensions [ f k ] = cm 3/2 = eV −3/2 . The complex amplitudes of the normal

84
modes are also called the holomorphic coordinates of the real scalar free
field. It is also quite evident that by introducing the related real canonical
coordinates (
Qk ≡ (2ω k ) −1/2 (fk + fk∗ )

1/2
P k ≡ − i 21 ω k (fk − fk∗ )

1
−1
q  
1
fk = 2
ω k2 Q k + i ω k 2 P k (3.32)

we can write eventually

X X  
1 2 2 2
P0 = Hk = 2
P k + ω Q
k k
k k
X  
= 1
2
ω k fk fk∗ + fk∗ fk (3.33)
k

with [ P k ] = cm , [ Q k ] = cm 2 which explicitly shows that a real scalar

field is dynamically fully equivalent to an assembly of an infinite number of
decoupled harmonic oscillators of principal frequencies ω k = (k 2 + m2 ) 1/2 .
As a matter of fact, we can rewrite the field and conjugated momentum
expansions (3.23) in the suggestive form
Xh i
φ(x) = fk (t) u k (x) + fk∗ (t) u∗k (x)
k
X h i
Π(x) = i ω k fk∗ (t) u∗k (x) − fk (t) u k (x) (3.34)
k

where of course

fk (t) = fk exp {−i ω k t} fk∗ (t) = fk exp {i ω k t} (3.35)

u k (x) = [ 2ω k (2π)3 ] −1/2 exp {i k · x}

and from the standard canonical Poisson’s brackets 1 for the linear oscillator

{Q h , Q k } = 0 {P h , P k } = 0

{Q h , P k } = δ ( h − k)
1
We recall the definition of the Poisson’s brackets for two differentiable functions F
and G on the phase space of a mechanical system
X  ∂F ∂G ∂G ∂F 
{F , G} ≡ −
i
∂qi ∂pi ∂qi ∂pi

85
from (3.32) it is very simple to derive the canonical hamiltonian equations
for the time dependent holomorphic coordinates fk (t) : namely,

{fh , fk } = 0 = {f ∗h , f ∗k } {fh , fk∗ } = − i δ(h − k) (3.36)

f˙k (t) = {fk (t) , H} = − i ω k fk (t) (3.37)

the solution of which is just provided by (3.35).

86
3.2 Quantization of a Klein-Gordon Field
Once that the dynamical treatment of the free real scalar field has been
developed within the canonical hamiltonian formulation, the quantization of
the system will directly follow in accordance with the Dirac correspondence
principle – see any textbook of quantum mechanics e.g. [4]. According to
the quantization rules for a linear hamonic oscillator, we shall introduce for
each normal mode of the real scalar field the corresponding linear operators
acting on the related Hilbert space and the associated algebra, i.e.

Q k 7−→ Q b†
bk = Q P k 7−→ Pbk = Pbk†
k

−i~d
Pbk =
dQk
1 b
{Q h , P k } 7−→ [ Q h , Pbk ]
i~

b k = 1 Pbk2 + 1 ω k2 Q
H k 7−→ H b 2k (3.38)
2 2

and the Poisson’s brackets among the holomorphic coordinates turn into the
commutators among the creation distruction operators

fk 7−→ ak fk∗ 7−→ ak† (3.39)

[ ah , a k ] = 0 [ a†h , a†k ]=0 (3.40)
[ ah , a†k ] = δ(h − k) (3.41)

As a consequence the scalar field function φ(x) together with its conjugated
momentum Π(x) will turn, after quantization, into operator valued tempered
distributions, the normal mode expansions of which can be obtained in a
straightforward way from (3.23) and (3.39), that is
Xh i
φ(x) = ak u k (x) + a†k u∗k (x)
k
X h i
Π(x) = i ω k a†k u∗k (x) − ak u k (x) (3.42)
k
[ φ(t, x) , φ(t, y) ] = 0 = [ Π(t, x) , Π(t, y) ]
[ φ(t, x) , Π(t, y) ] = i δ (x − y) (3.43)

This means that the classical expressions (3.31) and (3.33) of the energy and
momentum of the free real scalar field will turn into the quantum operator

87
expressions
X  
P0 = 1
2
Pbk2 + ω k2 Q
b2
k
k
X  
= 1
2
ω k ak a†k + a†k ak
k
X X
= ω k a†k ak + δ(0) 1
2
ωk (3.44)
k k
X  
P = 1
2
k ak a†k + a†k ak
k
X
= k a†k ak (3.45)
k

It is then easy to check that the definition of the conjugate momentum field
Π(x) and the Klein-Gordon field equation can be recast into the canonical
Heisenberg form
1
φ̇(x) = [ φ(x) , P0 ] = Π(x)
i~
1
Π̇(x) = [ Π(x) , P0 ] = (∇2 + m2 ) φ(x)
i~
It is worthwhile to spend some few words concerning the divergent quantity

Z
X
1 dk 1
c U0 = δ(0) 2
~ω k ≡ V ~ω k
k
(2π)3 2
K
k2 p 2
Z
= V ~c dk k + m2 c2 /~2 (3.46)
0 4π 2
where V is the volume of a very large box and ~K  mc is a very large
wavenumber. The latter is called the vacuum energy or zero–point energy of
the real scalar field. Since we know that a free real scalar field is dynamically
equivalent to an infinite (continuous) set of linear oscillators, we can roughly
understand the divergent quantity U0 to be generated by summing up the
quantum fluctuations of the canonical pair of operators φ(t, x) and Π(t, x) ,
alias Qb k and Pbk (k ∈ R3 ) at each point x ∈ V of a very large box in the
three dimensional space.
It turns out to be more appropriate to discuss the vacuum energy density,
i.e. the regularized vacuum energy per unit volume of the quantum real scalar
field. The vacuum state vector or vacuum state amplitude of the actual
quantum mechanical system under investigation is nothing but the ground

88
state of the system and physically corresponds to the absence of field quanta,
i.e. spinless massive particles of rest mass m . It is defined by

ak | 0 i = 0 = h 0 | a†k ∀ k ∈ R3 (3.47)

so that, after setting

Z K
c ~c p
h 0 | P0 | 0 i ≡ h ρ i c 2 = 2 k2 d k k 2 + (mc/~)2
V 4π 0

together with ξ ≡ ~K/mc , then we have2

Z ξ √
m4 c 3
hρi = x2 1 + x2 d x
4π 2 ~3 0
m4 c 3
 
2 3/2 1 2 1/2 1  p
2
= ξ(1 + ξ ) − ξ(1 + ξ ) − ln ξ + 1 + ξ
16π 2 ~3 2 2
4 2 2 4 3
    mc 2 
~K m cK mc ~K 1
= + − ln − + ln 2 + O
16π 2 c 16π 2 ~ 32π 2 ~3 mc 4 ~K
4
 
K ~
≈ 2
(~K  mc) (3.48)
16π c
If we trust in general relativity and in quantum field theory up to the Planck
scale

~c/GN = 1.220 90(9) × 10 19 GeV/c2

p
MP =
= 2.176 45(16) × 10 −11 g

where GN is the newtonian gravitational constant

GN = 6.674 2(10) × 10 −8 cm3 g −1 s −2

then we might take K ' cMP /~ which eventually gives

MP
hρi ≈ 2 3
≈ 2 × 10 115 ( GeV/c 2 ) cm − 3 (3.49)
16π `P
p
in which `P = ~ GN /c 3 = 1.61624(8)×10−33 cm denotes the Planck length.
This is a truly enormously large value. So it is no surprise that Paul Adrien
Maurice Dirac soon suggested that this zero–point energy must be simply
discarded, as it turns out to be irrelevant for any laboratory experiment in
which solely energy differences are indeed observable. However, Wolfgang
2
See for example [23] eq. (2.272 2.) p.105

89
Pauli soon afterwards recognized that this vacuum energy surely couples to
Einstein’s gravity and it would then give rise to a large cosmological constant,
so large that the size of the universe could not even reach the earth–moon
distance. On the contrary, the present day observed value of the so called
dark energy density of the universe is
3H02
ρΛ = ΩΛ ρc = ΩΛ = 1.05368(11) × 10−5 ΩΛ3 ( GeV/c 2 ) cm − 3
8πGN
where ΩΛ = 0.73(3) is the dark energy density fraction, H0 = 100 ΩΛ Km
s−1 Mpc−1 is the present day Hubble expansion rate. This leads to the
cosmological constant value

Λ = 3H02 /c 2 = 3.50 × 10−52 ΩΛ2 m−2 (3.50)

which is extremely small but nonvanishing 3 . This eventually means that the
dark energy density and the vacuum energy density of all the fundamental
quantum fields in the universe differ by nearly 120 orders of magnitude : this
is the cosmological constant puzzle, the solution of which is still unknown.
Leaving aside this intringuing puzzle, we now turn back to the realm of
galileian laboratory experiments and endorse the Dirac’s point of view. To
this concern, we shall introduce the useful concept of an operator written
in normal form as well as the concept of the normal product of operators
[8]. The normal form of an operator involving products of creation and
annihilation operators is said to be the form in which in each term all the
creation operators are written to the left of all the annihilation operators.
We consider an example. We write down in normal form the product of the
two operators
Xn o
F (x) G(y) ≡ Fk∗ (x) a†k + Fk (x) a k
k
Xn †
o
∗
× Gh (y) a h + Gh (y) a h
h
XX
= Fk∗ (x) G∗h (y) a†k a†h + h.c.
k h
XX 
+ Fk∗ (x) Gh (y) + Fh (x) G∗k (y) a†k a h
k h
X
+ Fk (x) G∗k (y) (3.51)
k

3
This tiny but nonvanishing measured value of the cosmological constant does actually
constitute a striking phenomenological evidence against supersymmetry.

90
The sum of terms not involving any ordinary c-number functions is called
the normal product of the original operators. The normal product may also
be defined as the original product reduced to its normal form with all the
commutator functions being taken equal to zero in the process of reduction.
The normal product of the operators F (x) and G(y) is denoted by the symbol
def
X
: F (x) G(y) : = F (x) G(y) − Fk (x) G∗k (y)
k

We now agree by definition to express all dynamical variables which depend

quadratically upon operators with the same arguments, such as the energy,
momentum and angular momentum of the radiation fields, in the form of
normal products. For example, we shall write the energy momentum four
vector quantum operator
Z X
P0 = d x 21 : Π 2 (x) − φ(x) Π̇(x) : = ~ω k a†k a k (3.52)
k
Z X
P= d x : Π(x)∇φ(x) : = ~k a†k a k (3.53)
k

Now, if we keep the definition of the vacuum state to be given by (3.47)

it follows that the expectation values of all the dynamical variables vanish
for the vacuum state, e.g. h Pµ i ≡ 0 . By this method we exclude from
the theory at the outset the so called zero-point quantities of the type of
the zero-point energy, which usually arise in the process of the quantization
of the field theories and turn out to be, strictly speaking, mathematically
ill-defined divergent quantities.

91
3.3 The Fock Space
The quantum theory of the free real scalar field naturally gives rise to the
concept of spinless neutral massive particle. As a matter of fact, it appears to
be clear that the nonnegative operators a †k a k do possess integer eigenvalues
N k = 0, 1, 2, . . . which are interpreted as the numbers of particles of a given
wave number k and a given energy ω k . The hamiltonian operator turns out
to be positive definite and we can readily derive, from the canonical com-
mutators (3.43), the eigenvalues and the common eigenstates of the energy
momentum commuting operators Pµ : namely,
X
E ({N k }) = ωk Nk N k = 0, 1, 2, . . . (3.54)
k
X
P ({N k }) = k Nk N k = 0, 1, 2, . . . (3.55)
k

The setting up of the eigenstates leads to the well known construction of

the so called Fock space :
Vladimir Alexandrovich Fock
Sankt Petersburg 22.12.1898 – 27.12.1974
Konfigurationsraum und zweite quantelung
Zeitschrift der Physik A 75, 622–647 (1932).
Actually, in order to describe N −particle states, consider any set of particle
3
P
numbers {N k = 0, 1, 2, . . . | k ∈ R } such that k N k = N . Then the
generic N –particle normalized energy momentum eigenstate can be written
as Y
|{N k }iN ≡ ( N k ! ) −1/2 a †kN k | 0 i
k

which satisfies

P0 |{N k }iN = EN ({N k }) |{N k }iN

M hh | kiN = δM N δk1 h1 δk2 h2 . . . δkN hN

It is very important to gather that, owing to the commutation relation

[ a †h , a †k ] = 0 , by its very construction any N –particle state is completely
symmetric under the exchange of any wave numbers, i.e. the scalar spinless
particles obey the Bose-Einstein statistics.
In particular, the 1-particle energy momentum eigenstates are given by
| k i = a†k | 0 i and satisfy

Pµ | k i = kµ | k i k0 = ω k

92
The 1-particle wave functions in the coordinate representation, for a given
wave number, are defined in terms of the matrix elements of the field operator
(3.42) and read

u k (x) ≡ h 0 | φ(x) | k i
= [ 2ω k (2π)3 ] −1/2 exp{−i k · x} k0 = ω k (3.56)

Notice that they turn out to be normalized in such a way to satisfy the
orthonormality and closure relations
Z ↔
( u k , u h ) ≡ d x u ∗k (x) i ∂ 0 u h (x) = δ (k − h) (3.57)
X 1 (−)
u ∗k (y) u k (x) ≡ D (x − y) (3.58)
k
i
∓i
Z
(±)
d k δ k 2 − m 2 θ (k 0 ) exp {± i k · x}


D (x) = (3.59)
(2π) 3
The 1-particle wave functions (3.56) satisfy by construction the Klein-Gordon
wave equation
 + m2 u k (x) = 0 ∀ k ∈ R3

and do thereby represent a complete orthonormal base, with respect to the

scalar product, in the 1-particle Hilbert space

H1 = V1 V1 = { | k i = a †k | 0 i k ∈ R 3 }

More precisely, eq. (3.56) does explicitly realize the isomorphism between
the Fock space representation H1 of the space of 1-particle states and the
spacetime coordinate representation L2 (R3 ) of the 1-particle wave functions
with respect to (3.57).
It turns out that the previously introduced 1-particle energy momentum
eigenstates | k i = a†k | 0 i are improper and have to be normalized, in the
sense of tempered distribution according to

h h | k i = δ (h − k ) (3.60)

Moreover they satisfy the closure or completeness relation in the 1-particle

Hilbert space H1 that reads
X
| k ih k | = I1 (3.61)
k

It is important to remark that these orthonormality and closure relationships,

as well as the normal mode decompositions (3.42), are not Lorentz covariant.

93
As a consequence, the insofar developed quantization procedure for a real
scalar field is set up in a particular class of inertial reference frames connected
by spatial rotations belonging to the group SO(3) .

The N −particle wave functions in the coordinates representation can be

readily obtained in the completely symmetric form
N
Y
uN (x1 , x2 , . . . , xN ) = ( N kj ! ) −1/2 u kj (xj )
j=1
≡ h x1 x2 . . . xN | k1 k2 . . . kN i (3.62)

Furthermore, we can write the generic normalized element of the N –particle

completely symmetric Hilbert space – the closure of the symmetric product
of 1–particle Hilbert spaces
s s s s
HN ≡ VN VN = { H1 ⊗ H1 ⊗ . . . ⊗ H1 } = H1⊗ n
| {z }
N times

in the form
XX X
| ϕN i ≡ ... C(k1 , k2 , . . . , kN ) |{N k }iN
k1 k2 kN

with XX X
... | C(k1 , k2 , . . . , kN ) | 2 = 1
k1 k2 kN

To end up, we are now able to write the generic normalized element of the
Fock space of the spinless neutral scalar particle states
∞
M s
F ≡ C ⊕ H 1 ⊕ H2 ⊕ . . . ⊕ H N ⊕ . . . = H1⊗ n
n=1

in the form ∞ ∞
X X
|Φi = CN | ϕ N i | CN | 2 = 1
N =0 N =0

which summarizes the setting up of the Fock space of the particle states, the
structure of which is characterized by the canonical quantum algebra (3.43)
and the energy momentum operators (3.53). From the above construction
of the Fock space of the states of a free real scalar field, it appears quite
evident that all the quantum states can be generated by linear combinations
of repeated applications of the creation operators on the vacuum state. This
property is known as the cyclicity of the vacuum state.

94
 We end up this section by discussing the behaviour of the N -particle
states and of the field operators under Lorentz transformations. First of all,
we recall that the above introduced 1-particle states of the basis {| k i =
a†k | 0 i} in the Hilbert space H1 do not satisfy covariant orthonormality
and completeness relations. To remedy this, consider first the completeness
relation and the trivial identity
X Z
| k ih k | = d k a†k | 0 ih 0 | ak
k
Z
dk 1 1
= 3
[ (2π)3 2 ω k ] 2 a†k | 0 ih 0 | ak [ (2π)3 2 ω k ] 2
(2π) 2 ω k
Z Z
def †
= D k a (k) | 0 ih 0 | a(k) ≡ D k | k ih k | = I1

whence we get
Z Z Z
def dk 1
d k θ(k0 ) δ k 2 − m2


Dk = 3
= (3.63)
(2π) 2 ω k (2π)3
def 1
|ki = [ (2π)3 2 ω k ] 2 a†k | 0 i = a† (k) | 0 i (3.64)

The Lorentz invariant completeness relation for the 1-particle states can also
be written in the two equivalent forms
X Z
| k i h k | = I1 = D k | k ih k | (3.65)
k

Now it is clear that the 1-particle states of the new basis, which will be named
covariant 1-particle states,
1
{ | k i = [ (2π)3 2 ω k ] 2 a†k | 0 i | k ∈ R3 } (3.66)

fulfill manifestly Lorentz covariant orthonormality relations

h k 0 | k i = (2π)3 2 ω k δ (k − k 0 ) (3.67)

Furthermore we can write

Z h i
φ(x) ≡ D k a (k) e −i kx + a† (k) e i kx (3.68)
Z h i
Π(x) ≡ D k k0 − i a (k) e −i kx + i a† (k) e i kx (3.69)

so that the 1-particle wave functions in the coordinate representation, which

correspond to the 1-particle state | k i and are still defined in terms of the

95
matrix elements of the field operator (3.42), will coincide with the plane
waves

u k (x) ≡ h 0 | φ (x) | k i = exp{−i k · x} ( k0 = ω k ) (3.70)

which are normalized in such a way to satisfy

Z ↔
d x u ∗k (x) i ∂ 0 u h (x) = 2 ω k δ(k − h) (3.71)

Now it becomes clear that to each element of the restricted Lorentz group,
which is univocally specified by the six canonical coordinates ω µν = (α, η) ,
there will correspond a unitary operator U (ω) : H1 → H1 so that

U (ω) | k i = | Λ k i h k | U † (ω) = h Λ k | (3.72)

By virtue of the covariant orthonormality relation (3.67) we can write

h Λ h | Λ k i = h h | k i = h h | U † (ω) U (ω) k i = δ (h − k ) (2π)3 2ω k (3.73)

where we obviously understand e.g.

1
| Λ k i = | k 0 i = [ (2π)3 2 k 00 ] 2 a †k 0 | 0 i = a† (Λ k) | 0 i (3.74)
k µ0 = Λ µν k ν k0 = ω k
g µν k µ0 k ν0 = k 2 = m2

In turn, under a Lorentz transformation the creation annihilation operator

undergo the change

U (ω) a† (k) U † (ω) = a† (Λ k) U (ω) a (k) U † (ω) = a (Λ k) (3.75)

under the assumption that the vacuum is invariant i.e. U (ω) | 0 i = | 0 i .

As a consequence we immediately obtain the transformation law of the
quantized real scalar field under passive Lorentz transformations : namely,
def
φ 0 (x) = U (ω) φ (x) U † (ω)
Z h i
= D k a (Λ k) e −i kx + a† (Λ k) e i kx
Z
D k 0 a (k 0 ) exp −i (Λ−1 k 0 )µ x µ + h. c.

=
= φ (Λ x) (3.76)

96
where I have made use of the relation in matrix notation

(Λ−1 k 0 )µ x µ = (Λ−1 k 0 )> g x = k 0 > (Λ−1 )> g x

= k 0 > ( g Λ g) g x = k 0 > g Λ x = kµ0 Λµν x ν

Let us now collect the ten conserved dynamical quantities related to the
quantized real scalar field : namely,
Z
P0 = d x 21 : Π 2 (x) + ∇φ (x) · ∇φ (x) + m2 φ 2 (x) :
Z
Pk = d x : Π (x)∇k φ (x) :
Z
L jk = d x : xj Π (x)∇k φ (x) − xk Π (x)∇j φ (x) :
L 0k = Z
x 0 Pk
− dx 1
2
xk : Π 2 (x) + ∇φ (x) · ∇φ (x) + m2 φ 2 (x) :

Owing to the self-adjointness of the field operators

φ(x) = φ† (x) Π(x) = Π† (x)

all the above ten conserved dynamical quantities turn out to be self–adjoint
operators corresponding to physical observables. For example
Z
P = − d x : [ ∇ φ† (x) ] Π † (x) :
†

Z
= − d x : [ ∇ φ(x) ] Π(x) :
Z
= − d x : Π(x)∇ φ(x) : (3.77)

thanks to normal ordering. It appears thereby evident that the ten selfadjoint
operators (Pµ , Lρσ ) acting on the Fock space are the generators of a unitary
infinite dimensional representation of the Poincaré group on F . From the
canonical commutation relations (3.43) it is immediate to show that

[ φ (x) , Pµ ] = i ∂µ φ(x) (3.78)

[ φ (x) , Lµν ] = i xµ ∂ ν φ (x) − i xν ∂ µ φ (x) (3.79)

By direct inspection, it is straightforward to verify that, using the canonical

commutation relations (3.43) and the normal ordering prescription, the self-
adjoint operators (Pµ , Lρσ ) do actually fulfill the Poincaré Lie algebra (1.40)

97
so that, in any neighbourhood of the unit element ω ρσ = 0 = a µ , we can
safely write

U (ω, a) = exp i a µ Pµ − 12 i ω ρσ Lρσ


U : F → F

This is the way how to realize the unitary irreducible representation of the
Poincaré group with mass m and spin zero from the quantization of the real
scalar free field. In fact, for an infinitesimal Poincaré transformation we have

φ 0 (x) ≡ U (δ ω, δ a) φ (x) U † (δ ω, δ a)
= φ (x) + i  µ [ Pµ , φ (x) ] − 21 i  ρσ [ Lρσ , φ (x) ]
= φ (x) + (  µ +  µν xν ) ∂µ φ (x)
= φ (x + δ x) (3.80)

in which we have denoted by δa µ =  µ and δω ρσ =  ρσ the infinitesimal

parameters.

98
3.4 Special Distributions
We have already met the positive and negative frequency scalar distributions
Z
(±) 1 dk
exp {± i k · x} δ k 2 − m 2 θ (k 0 )


D (x) = ± 3
i (2π)
The latter ones are characterized by
1 (−)
h 0 | φ (x) φ (y) | 0 i = D (x − y)
i
D (+) (x − y) = − D (−) (y − x)
[ D (±) (x) ] ∗ = D (∓) (x)
From the normal mode expansion (3.42) and the canonical commutation
relations (3.43) we obtain the commutator between two real scalar free field
operator at arbitrary points, which is known as the Pauli-Jordan distribution
1
[ φ(x) , φ(y) ] ≡ D (x − y) (3.81)
i
where
Z
def dk 2 2


D (x) = i exp {− i k · x} δ k − m sgn (k0 )
(2π) 3
≡ D (−) (x) + D (+) (x) (3.82)

where the sign distribution is defined to be


+1 for k0 > 0
sgn (k0 ) = θ(k0 ) − θ(−k0 ) =
−1 for k0 < 0

The Pauli-Jordan distribution is a Poincaré invariant solution of the Klein-

Gordon wave equation

 x + m2 D (x − y) = 0

with the initial conditions

∂
lim D (x − y) = 0 lim D (x − y) = δ (x − y)
x0 → y0 x0 → y0 ∂x 0
in such a manner that
1
lim [ Π (x) , φ (y) ] = lim ∂ 0 D (x − y) = − i δ (x − y)
x0 → y0 x0 → y0 i

99
in accordance with the canonical equal-time commutation relations.
The Pauli-Jordan distribution is real

D ∗ (x) = D (x)

and enjoys as well the very important property of vanishing for spacelike
separations, that is

D (x − y) = 0 for (x0 − y0 )2 < (x − y)2 (3.83)

The above feature is known as the microcausality property.

A further very important distribution related to causality is the causal
Green function or Feynman propagator. It is defined as follows:

h 0 | φ(x) φ(y) | 0 i for x0 > y0
DF (x − y) =
h 0 | φ(y) φ(x) | 0 i for x0 < y0
= θ(x0 − y0 ) h 0 | φ(x) φ(y) | 0 i
+ θ(y0 − x0 ) h 0 | φ(y) φ(x) | 0 i
≡ h 0 | T φ(x) φ(y) | 0 i (3.84)

the last line just defining the chronological product of operators in terms of
the time ordering symbol T that prescribes the place for the operators that
follow in the order with the latest to the left. It is easy to check, by applying
the Klein-Gordon differential operator  x + m2 to the Feynman propagator
and taking (3.43) into account, that the causal Green function is a solution
of the inhomogeneous equation

( x + m2 ) DF (x − y) = − i δ(x − y)

so that its Fourier representation reads

exp {− i k · (x − y)}
Z
i
DF (x − y) = 4
dk (3.85)
(2π) k 2 − m2 + iε
In fact, from the integral representation of the Heaviside step distribution
Z ∞
0 i dp0
θ (± z ) = ± lim+ exp {− i p0 z 0 } (3.86)
ε→0 2π −∞ p0 ± i ε
and from the normal mode expansion of the free real scalar field
Xh † ∗
i
φ(x) = ak u k (x) + ak u k (x)
k

u k (x) ≡ [ 2ω k (2π)3 ] −1/2 exp {− i x0 ω k + i k · x}

100
by making use of the canonical commutation relations (3.43) we can write
Z ∞
1
h 0 | T φ(x) φ(y) | 0 i = d k0 exp {i (k0 − ω k ) (x0 − y0 )}
2πi −∞
X
× (2π)− 3 [ 2 ω k (k0 − i ε) ] −1
k
× exp {i k · (x − y)}
Z ∞
1
+ d k0 exp {− i (k0 − ω k ) (x0 − y0 )}
2πi −∞
X
× (2π)− 3 [ 2 ω k (k0 − i ε) ] −1
k
× exp {i k · (x − y)}

Changing the integration variable from k0 to k00 = − k0 in the first integral

of the right hand side of the previous equality we obtain
Z ∞
1
DF (x − y) = − d k0 exp {− i (k0 + ω k ) (x0 − y0 )}
2πi −∞
X
× (2π)− 3 [ 2 ω k (k0 + i ε) ] −1
k
× exp {i k · (x − y)}
Z ∞
1
+ d k0 exp {− i (k0 − ω k ) (x0 − y0 )}
2πi −∞
X
× (2π)− 3 [ 2 ω k (k0 − i ε) ] −1
k
× exp {i k · (x − y)}

and a translation with respect to the k0 integration variable yields

Z
i
h T φ(x) φ(y) i = 4
d k exp {− i k · (x − y)} [ 2 ω k ] −1
(2π)
 
1 1
× −
k0 − ω k + i ε k0 + ω k − i ε
Z
d k − i k·(x−y) i
= 4
e
(2π) k − m2 + iε
2

which proofs the Fourier representation (3.85).

The latter one just involves the (+iε) prescription in momentum space
that corresponds to causality in coordinate space. As a matter of fact, if
we define the creation φ (+) (x) and destruction φ (−) (x) parts of the free real

101
scalar field operator according to
X
φ (−) (x) = ak u k (x) (3.87)
k
X
φ (+) (x) = a†k u∗k (x) (3.88)
k

it turns out that we have

h T φ(x) φ(y) i0 = θ(x0 − y0 ) h φ (−) (x) φ (+) (y) i0
+ θ(y0 − x0 ) h φ (−) (y) φ (+) (x) i0
1 1
= θ(x0 − y0 ) D (−) (x − y) + θ(y0 − x0 ) D (−) (y − x)
i i
= i θ(y0 − x0 ) D (x − y) − i θ(x0 − y0 ) D (−) (x − y)
(+)

(3.89)
which shows that, first, a particle is created out of the vacuum by the creation
part of the free real scalar field operator and then, later, it is annihilated by
the destruction part of the free real scalar field operator : the opposite never
occurs, what precisely endorses the causality requirement in coordinate space.

3.4.1 Wick Rotation : Euclidean Formulation

The causal +iε prescription in momentum space enjoys a crucial feature,
the so called Wick rotation property. To see this, consider once again the
Fourier representation (3.85) and change the energy integration variable as
k0 = − i k4 so that
Z ∞ X exp {− k4 x0 + i k · x}
i
DF (x) = d (− i k 4 ) (3.90)
(2π)4 − ∞ k
−k42 − k2 − m2

where the +iε prescription has been dropped since the denominator is now
positive definite. If we further set i x0 ≡ x4 then we finally get
DE (xE ) = − DF (− ix4 , x)
Z
1 2 2 −1


= d kE exp {i k E µ x E µ } kE + m (3.91)
(2π)4
where we use the notation
xE = xE µ = (xk , x4 ) kE = kE µ = (kj , k4 ) ( j, k = 1, 2, 3 )
kE · xE = kE µ xE µ = k1 x1 + k2 x2 + k3 x3 + k4 x4
Z Z ∞ X
d kE = d k4
−∞ k

102
The Wick rotation is nothing but a counterclockwise rotation of a π/2 angle
in the complex energy plane that drives the real k0 axis over the imaginary
k4 axis. The location of the poles of the Feynman propagator in the complex
energy plane, that corresponds to the causal +iε prescription, is such that
no singularity crossing occurs owing to the Wick rotation. It is precisely this
crucial aspect that encodes the causality requirement in momentum space.
In the configuration space we turn to the so called euclidean formulation,
according to which the action and the lagrangian in the Minkowski spacetime
are transformed into the purely imaginary euclidean action and Lagrange
density. As a matter of fact, if we change the time integration variable
according to x0 = − i x4 , then we readily obtain

S [ φ ] 7→ SE [ φE ]
Z
i  
= d xE ∂µ φE (xE ) ∂µ φE (xE ) + m2 φE2 (xE ) (3.92)
2
with the euclidean indices always lower case so that
∂φE
∂µ φE ≡ ( µ = 1, 2, 3, 4 )
∂xE µ

If we assume the asymptotic behaviour for the euclidean scalar field

q
lim φE (xE ) xE2 = 0 (3.93)
xE → ∞

then we can also write

Z
i
SE [ φE ] = d xE φE (xE ) ( − ∂E2 + m2 ) φE (xE )
2
Z
i
= d kE φ̃E (kE )( kE2 + m2 ) φ̃E (− kE ) (3.94)
2
in which I have set by definition
Z
1
φE (xE ) ≡ d kE φ̃E (kE ) exp { i kE µ xE µ }
(2π)4

103
3.5 The Generating Functional
3.5.1 The Symanzik Functional Equation
Consider the vacuum expectation value
 Z 
Z0 [ J ] = T exp i d x φ(x) J(x)
0
∞
in
Z Z
def
X
= d x1 J(x1 ) · · · d xn J(xn )
n=0
n!
× h 0 | T φ (x1 ) · · · φ (xn ) | 0 i (3.95)
the suffix zero denoting the free field theory, where J(x) are the so called
classical external sources with the canonical dimensions [ J ] = eV 3 . The
vacuum expectation values of the chronological ordered products of n free
scalar field operators at different spacetime points are named the n−point
Green functions of the (free) field theory. By construction, the latter ones can
be expressed as functional derivatives of the generating functional : namely,
(n)
G 0 (x1 , · · · , xn ) ≡ h 0 | T φ (x1 ) · · · φ (xn ) | 0 i (3.96)
n (n)

= ( − i ) δ Z 0 [ J ] / δ J(x1 ) · · · δ J(xn ) J = 0
Taking one functional derivative of the generating functional (3.95) we find
  Z 
δ
(− i) Z 0 [ J ] = T φ(x) exp i d y φ(y) J(y) (3.97)
δ J(x) 0

In order to evaluate the above quantity it is convenient to introduce the

operator [17]
( Z 0 )
Z t
E(t 0 , t ) ≡ T exp i d y0 d y φ(y) J(y) (3.98)
t

so that we can write

δ
(− i) Z 0 [ J ] = h 0 | E(∞ , x0 ) φ(x) E(x0 , − ∞ ) | 0 i (3.99)
δ J(x)
Taking a derivative with respect to x0 we find
∂
h 0 | E(∞ , x0 ) φ(x) E(x0 , − ∞ ) | 0 i =
∂ x0
h 0 | E(∞ , x0 ) Π(x) E(x0 , − ∞ ) | 0 i −
Z
i h 0 | E(∞ , x0 ) d y [ φ(x0 , y) , φ(x0 , x) ] J(x0 , y) E(x0 , − ∞ ) | 0 i
= h 0 | E(∞ , x0 ) Π(x) E(x0 , − ∞ ) | 0 i (3.100)

104
owing to microcausality. One more derivative evidently yields
∂
h 0 | E(∞ , x0 ) Π(x) E(x0 , − ∞ ) | 0 i =
∂ x0
h 0 | E(∞ , x0 ) Π̇(x) E(x0 , − ∞ ) | 0 i + J(x) Z 0 [ J ]

whence we eventually obtain the functional differential equation for the free
scalar field generating functional, that is
h iδ i
(  x + m2 ) + J(x) Z 0 [ J ] = 0 (3.101)
δJ(x)

where we used the fact that the free scalar field operator valued distribution
does satisfy the Klein-Gordon wave equation. The above functional equation
has been first obtained by
Kurt Symanzik
Z. Naturforschung 9A, 809 (1954)
and will thereby named the Symanzik functional equation. This functional
differential equation (3.101) has a unique solution that fulfills causality
 Z Z 
1
Z 0 [ J ] = exp − d x d y 2 J(x) DF (x − y) J(y) (3.102)

as it can be seen by direct substitution.

The classical action for the real scalar field in the presence of an external
source can be rewritten as
Z h i
S0 [ φ , J ] = d x 21 g µν ∂µ φ(x)∂ ν φ(x) − 12 m2 φ 2 (x) + φ(x) J(x)
Z h i
d x − 21 φ(x)  + m2 φ(x) + φ(x) J(x)


=
˙
Z
≡ S0 [ φ ] + d x φ(x) J(x) (3.103)

where the symbol =

˙ indicates that the total four divergence term
Z  
1 µν
2
g d x ∂ µ φ(x)∂ ν φ(x)

has been dropped as it does not contribute to the equations of motion

− δS0 [ φ ] / δ φ (x) = (  + m2 ) φ(x) = J(x) (3.104)

105
3.5.2 The Functional Integral
To reconstruct the solution (3.101) via Fourier methods we formally write
Z 0 [ J ] as a functional Fourier transform
Z  Z 
Z 0 [ J ] = Dφ Z 0 [ φ ] exp i d x φ(x) J(x)
e (3.105)

where Dφ formally denotes integration over an infinite dimensional function

space of classical scalar fields φ : M → R . Heuristically, a preliminary
although quite suggestive way to understand the functional measure Dφ is
in terms of Z Y Z ∞
Dφ :=: dφx
x∈M −∞

Here :=: denotes a formal equality, the precise mathematical meaning of

which has to be further specified, while the Minkowski spacetime point is
treated as a discrete index, in such a way that the functional integral could
be imagined as a continuous infinite generalization of a multiple Lebesgue
integral. Taking the functional derivative operator i δ/δJ(x) through the
functional Fourier integral (3.105) equation (3.101) becomes
h
2iδ i
0 = (x + m ) + J(x) Z 0 [ J ]
δJ(x)
Z h δS i  Z 
0
= Dφ + J(x) Ze0 [ φ ] exp i d y φ(y) J(y)
δ φ (x)

which enables us to identify

Ze0 [ φ ] = N exp {i S 0 [ φ ]} (3.106)

N being any arbitrary classical external source independent quantity. As a

matter of fact we have

Z h δS i  Z 
0
Dφ + J(x) Z 0 [ φ ] exp i d y φ(y) J(y)
e
δ φ (x)
Z h δS i  Z 
0
= Dφ + J(x) exp i S 0 [ φ ] + i d y φ(y) J(y)
δ φ (x)
 
− iδ
Z Z
= Dφ exp i S 0 [ φ ] + i d y φ(y) J(y)
δ φ (x)

106
so that, if we assume the validity of the functional integration by parts, the
very last expression formally yields
  φ = +∞
exp − i 12 φ x  x + m2 − iε φ x − φ x J x  φ = −∞
 

Y Z ∞ k
d φ y exp 12 φ y  y + m2 − iε iφ y + i h φ y J y i


×
−∞ y 6= x
y ∈M
= 0

the convergence factor being provided by the causal +iε prescription, where
we have taken into account that the boundary values for the codomain of the
scalar field functional space are just

− ∞ < φx < ∞ ∀x ∈ M

while I have used the discrete index like notation

Z
h φy Jy i ≡ d y φ(y) J(y)

A comparison with equation (3.102) leads to the formal equalities

 Z Z 
1
Z 0 [ J ] = exp − d x d y 2 J(x) DF (x − y) J(y)
Z  Z 
:=: N Dφ exp i S 0 [ φ ] + i d x φ(x) J(x)
Y Z ∞
d φ x exp − 21 i φ x K x φ x + i φ x J x

:=: N
x∈M −∞

(3.107)

where as usual
Z
1
φ(x)  + m2 − iε φ(x) ≡ h − 12 φ x K x φ x i


S0 [ φ ] = − dx 2

The functional measure Dφ , which is a formal entity until now, can be

implemented by the requirement of invariance under field translations

φ(x) 7→ φ 0 (x) = φ(x) + f (x)

Once it is assumed, after the change of variable

φ(x) 7→ φ 0 (x) = φ(x) − (  + m2 − iε ) − 1 J(x)

R
= φ(x) − i d y DF (x − y) J(y)
≡ φ x − h i D xy J y i (3.108)

107
we find

Z 0 [ J ] :=: exp − d x d y 12 J(x) DF (x − y) J(y)

 R R

× N Dφ exp − i d x 21 φ(x) (  + m2 − iε ) φ(x)

R  R

= Z 0 [ 0 ] exp − d x d y 12 J(x) DF (x − y) J(y)

 R R

Proof : from the change of the integration variable

φ x 7→ φ 0x = φ x − h i D xy J y i d φ x = d φ 0x

equation (3.107) yields

R∞
d φ x exp − 21 i φ x K x φ x + i φ x J x

−∞
R∞
= −∞ d φ 0x exp − 21 i (φ 0x + h i D xy J y i) K x (φ 0x + h i D xz J z i)


× exp { i φ 0x J x − J x h D xy J y i}

Now, from the equality

K x h i D xz J z i = J x
and the further equality

h i D xy J y i K x φ 0x =
˙ φ 0x K x h i D xy J y i = J x φ 0x

which is true by negleting twice a boundary term, we can finally write

R∞
d φ x exp − 21 i φ x K x φ x + i φ x J x

−∞
R ∞
˙ exp − 21 J x h D xy J y i −∞ d φ 0x exp − 12 i φ 0x K x φ 0x
 
= (∀x ∈ M)

and thereby
Y R∞
d φ x exp − 21 i φ x K x φ x + i φ x J x

Z 0 [ J ] :=: N −∞
x∈M
Y Y R∞
1
d φ 0z exp − 12 i φ 0z K z φ 0z
 
=
˙ exp − J x h D xy J y i N
2 −∞
x∈M z ∈M
Z
= exp h − 21 J x D xy J y i N Dφ 0 exp {i S 0 [ φ 0 ]}


= exp − d x d y 12 J(x) DF (x − y) J(y) Z 0 [ 0 ]

 R R

Quod Erat Demonstrandum

Hence, self-consistency actually entails the formal identification

Z
Z 0 [ 0 ] :=: N Dφ exp {i S 0 [ φ ]} = 1

As a matter of fact, from the very definition (3.95) it appears quite evident
that Z 0 [ 0 ] is nothing but the vacuum to vacuum amplitude h 0 | 0 i that we

108
suppose to be normalized to one. Thus, at this point, the strategy should
be clear : if we were able to give a precise mathematical meaning to the
above formal quantity Z 0 [ 0 ] , then we will be able in turn to set up a
mathematically sound and precise definition of the functional integral (3.105).
To this aim, let us first consider the euclidean formulation. Then we have
to make the replacements
(0)
i S 0 [ φ ] 7−→ − S E [ φE ] xE µ = (x , x4 = i x0 ) (3.109)

Z  
(0)
S E [ φE ] = d xE 1
2
∂µ φE (xE ) ∂µ φE (xE ) + m2 φ2E (xE )
Z
=
˙d xE φE (xE ) 12 ( m2 − ∂µ ∂µ ) φE (xE )
Z n o
(0) (0)
Z E [ 0 ] :=: N DφE exp − S E [ φE ] (3.110)

The above quantity is, formally, an absolutely convergent gaussian integral.

3.5.3 The Zeta Function Regularisation

According to the previously suggested heuristic interpretation, we could now
understand the latter as
(0)
YZ
d φx exp − 12 φx KE φx KE ≡ m2 − ∂µ ∂µ

Z E [ 0 ] :=: N
x

and if we assume that the functional integration variable can be rescaled

φx 7→ φx0 = µ φx

where µ is an arbitrary mass scale which does not influence the relevant J(x)
dependence of the generating functional, then we come to the expression

(0)
YZ ∞
0
d φx0 exp − 12 φx0 µ −2 KE φx0

Z E [ 0 ] :=: N (3.111)
x −∞

in which the dimensionless, positive, second order and symmetric differential

operator µ −2 ( m2 − ∂µ ∂µ ) is involved, the spectrum of which is purely
continuous and given by the positive eigenvalues µ −2 ( m2 + kµ kµ ) with
kµ ∈ R ( µ = 1, 2, 3, 4 ) . Then, for any finite dimensional real and symmetric
n × n matrix A = A> , with n ∈ N , it is well known that a real orthogonal
matrix R ∈ SO(n) always exists such that R> A R = diag(λ1 , · · · , λn ) where

109
λi (i = 1, 2, . . . , n) are the real eigenvalues of the symmetric matrix. As a
consequence we immediately obtain as a result of the gaussian integral
Z ∞ Z ∞
− n/2
I = (2π) d x1 · · · d xn exp {− 12 x> A x}
Z−∞
∞ Z −∞
∞
= (2π)− n/2 d y1 · · · d yn exp {− 12 (Ry)> A Ry}
−∞ −∞
Z ∞ Z ∞ ( n
)
X
= (2π)− n/2 d y1 · · · d yn exp − 21 λi yi2
−∞ −∞ i=1
n
Y Z ∞
= (2π)− n/2 d yi exp − 12 λi yi2

i=1 −∞
n r
Y 2π
= (2π)− n/2 = ( det A ) − 1/2 (3.112)
i=1
λi

In view of this simple result we shall attempt to define the formal quantity
(0)
Z E [ 0 ] to be precisely given by

(0)
YZ ∞
0
d φx0 exp − 12 φx0 µ −2 KE φx0

Z E [ 0 ] :=: N
x −∞

≡ N 0 det k µ −2 ( m2 − ∂µ ∂µ ) k − 1/2 (3.113)

where the determinant of a positive symmetric differential operator can be

suitably defined by means of the so called Zeta function regularization :
Steven W. Hawking (1977)
Zeta function regularization of path integrals in curved space time
Communication of Mathematical Physics, Vol. 55, p. 133
The idea beyond this method is as simple as powerful and is based upon the
analytic continuation tool. Consider as an example a positive operator A > 0
with a purely discrete spectrum, so that its spectral decomposition reads
∞
X
A= λ k Pk λk > 0 tr Pk = dk < ∞ ∀ k ∈ N
k=1

The complex powers of the positive operator A > 0 can be easily obtained
in terms of its spectral resolution
∞ Z ∞
X 1
A−s
= λ−s
k Pk = d t t s−1 exp {−tλk } Pk <e s > 0
k=1
Γ(s) 0

110
Let us now further suppose the positive operator A to be compact and of the
trace class. Hence, from the spectral decomposition theorem we can write
the integral kernel, or Green function,
∞
X ∞
X
hx | A −s
| yi = λ−s
k hx | Pk | yi = λ−s ∗
k ψk (x) ψk (y) (3.114)
k=1 k=1
Z
A −s
ψk (x) = λ−s
k ψk (x) d x ψk (x) ψn∗ (x) = δ kn (3.115)

Z ∞
X
Tr A −s
= d x hx | A −s
| xi = λ−s
k dk
k=1
∞ Z ∞
X 1
= dk d t t s−1 exp {−tλk }
k=1
Γ(s) 0
Z ∞ ∞
1 s−1
X
= dt t dk exp {−tλk } < ∞ (3.116)
Γ(s) 0 k=1

always in the strip <e s > 0 of the complex plane. Now we have
∞ ∞
d X d X
Tr A−s = λ−s
k dk = (− ln λk ) λ−s
k dk (3.117)
ds k=1
ds k=1

and thereby we obtain the zeta function regularization of the determinant of

a positive and compact operator of the trace class : namely,
∞ 
X def d
ln det A = dk ln λk = − Tr A−s (3.118)
k=1
ds s=0

provided the analytic continuation is possible to include the imaginary axis

<e s = 0 , via some suitable deformation of the integration path in complex
plane, just like in the original case of the Riemann’s Zeta function 4
(0+)
Γ(1 − s) (−θ) s−1 e −q θ
Z
ζ(s, q) = − dθ (3.119)
2πi ∞ 1 − e −θ
in which it is assumed that the path of integration does not pass through the
points 2nπi where n is a natural number.
4
Gradshteyn and Ryzhik [23] § 9.5 pp.1100-1103 ; Bruno Pini (1979) Lezioni sulle
distribuzioni, 1. Distribuzioni temperate, § 4 Appendice, cap. 3. p. 276.

111
 In order to apply the above treatment to the case of interest, one is
faced with the problem that the euclidean Klein-Gordon operator is neither
compact nor of the trace class. To overcome this difficulty, it is expedient
to introduce a very large box, e.g. a symmetric hypercube of side 2L , to
impose periodic boundary conditions on its faces and to make eventually
the transition to the infinite volume continuum limit. In the presence of a
symmetric hypercube with periodic boundary conditions, the spectrum of
the euclidean Klein-Gordon operator is purely discrete and nondegenerate

π2
λn = m2 + nµ nµ nµ ∈ Z µ = 1, 2, 3, 4 (3.120)
L2
and in the limit of L → ∞ we can safely replace
∞ Z ∞
X d kµ
7−→ 2L ( µ = 1, 2, 3, 4 )
nµ = −∞ −∞ 2π

d4 k
X Z
7−→ V
n
(2π) 4
−2
Then we find for A = µ ( m2 − ∂µ ∂µ )
Z
dk
Tr A−s =
˙ Vµ 2s
( m2 + k 2 ) − s
(2π) 4
Z ∞
V µ 2s
= 2
d t t s−3 exp {−t m2 }
16π Γ(s) 0
V m4  µ  2s Γ(s − 2)
=
16π 2 m Γ(s)
4 
Vm µ  2s
= 2
( s2 − 3s + 2 ) −1 (3.121)
16π m
where =
˙ means that the transition to the continuum limit is understood.
Hence
d
Tr A−s = ( s 2 − 3s + 2 ) −1 Tr A−s
ds h
2 µ i
× 2( s − 3s + 2 ) ln − 2s + 3 (3.122)
m
and thereby

V m4
  
2

2 m 3
det k m − ∂µ ∂µ /µ k = exp ln − (3.123)
16π 2 µ 4

112
Turning back to equation (3.113) we see that

V m4
  
(0) 0 m 3
Z E [ 0 ] = 1 ⇐⇒ N ≡ exp ln −
32π 2 µ 4
and the transition to the Minkowski spacetime can be immediately done by
simply replacing Veuclidean ↔ i Vminkowskian
The conclusion of all the above formal reasoning is as follows : we are
enabled to define the functional integral for a free scalar field theory by the
equalities
 Z Z 
1
Z 0 [ J ] = exp − d x d y J(x) DF (x − y) J(y)
2
Z  Z 
def
= N Dφ exp i S 0 [ φ ] + i d x φ(x) J(x)
Z
S 0 [ φ ] = − d x 21 φ(x)  + m2 − iε φ(x)

  1/2
N = constant × det k  + m2 k
i V m4
  
def m 3
= exp ln − (Zeta regularization)
32π 2 µ 4
Z
Z0 [ 0 ] = N Dφ exp {i S 0 [ φ ]} = 1

Notice that the integral kernel DF (x − y) which appears in the exponent

of the right hand side of eq. (3.102) is just the opposite of the inverse for
the kinetic operator − i (  + m2 ) that specifies the exponent i S 0 [ φ ] of the
functional integrand.
To summarize the above long discussion concerning the meaning and the
construction of the functional integration, I can list a number of key features.
The functional integral does fulfil by construction the following properties :
1. linearity
Z Z Z
Dφ (α F [ φ ] + β G [ φ ]) = α Dφ F [ φ ] + β Dφ G [ φ ]

for any pair of complex functions α, β : M → C

2. translation invariance
Z Z
Dφ F [ φ + α ] = Dφ F [ φ ] ∀α : M → C

113
 3. rescaling
Z Z
−1
Dφ F [ ( A φ ) (x) ] = (det A) Dφ F [ φ ]

where A is any invertible integro-differential operator

4. integration by parts
Z Z
δF [φ] δ G[ φ ]
0 = Dφ G [ φ ] + Dφ F [ φ ]
δ φ(x) δ φ(x)

The above properties 1 .− 4. are valid for

R any functional
integrand F, G of the
gaussian kind P [ φ ] exp i S 0 [ φ ] + i d x φ(x) J(x) (n ∈ N) with P [ φ ]
any polynomial functional of the scalar field and its derivatives.

References

1. N.N. Bogoliubov and D.V. Shirkov (1959) Introduction to the Theory

of Quantized Fields, Interscience Publishers, New York.

2. C. Itzykson and J.-B. Zuber (1980) Quantum Field Theory, McGraw-

Hill, New York.

3. R. J. Rivers (1987) Path integral methods in quantum field theory,

Cambridge University Press, Cambridge (UK).

4. M.E. Peskin and D.V. Schroeder (1995) An Introduction to Quantum

Field Theory, Perseus Books, Reading, Massachusetts.

5. I.S. Gradshteyn, I.M. Ryzhik (1996) Table of Integrals, Series, and

Products, Fifth Edition, Alan Jeffrey Editor, Academic Press, San
Diego.

114
3.6 Problems
1. The complex scalar field. Consider the field theory of a complex
valued free scalar field with the Lagrange density

L (x) = ∂µ φ∗ (x) ∂ µ φ(x) − m2 φ∗ (x) φ(x)

It is easier to analyse the theory by considering φ(x) and φ∗ (x) as the

independent variables in configuration space rather than the real and
imaginary parts of the complex scalar field function.
(a) Find the hamiltonian and the canonical equations of motion
Solution. The action is given by
Z
S [ Φ ] = d x L [ Φ(x) ] Φ(x) = u(x) + i v(x)

which leads to the conjugated canonical momenta

δS δS
= ∂ 0 Φ∗ (x) ≡ Π(x) = ∂ 0 Φ(x) ≡ Π∗ (x)
δ Φ̇(x) ∗
δ Φ̇ (x)

and to the classical hamiltonian functional

Z  
H [ Π, Φ ] = d x Π(x) Φ̇(x) + Π∗ (x) Φ̇∗ (x) − L (x)
Z  
= d x | Π(x) | 2 + | ∇Φ(x) | 2 + m2 | Φ(x) | 2

The Poisson brackets are evidently given by

{Φ(t, x) , Π(t, y)} = δ (x − y) = {Φ∗ (t, x) , Π∗ (t, y)}

all the others being equal to zero, so that the Hamilton equations read

Φ̇(x) = {Φ(x) , H} = Π(x)
Π̇(x) = {H , Π(x)} = Φ̈(x)

Φ̇∗ (x) = {Φ(x)∗ , H} = Π∗ (x)



Π̇∗ (x) = {H, Π∗ (x)} = Φ̈∗ (x)

whence we immediately find the Klein-Gordon wave equations

Φ̈(x) = ∇ 2 Φ(x) − m2 Φ(x) (  + m2 ) Φ(x) = 0

115
(b) Diagonalize the hamiltonian operator introducing creation and
annihilation operators. Show that the complex scalar free field contains
two types of massive spinless particles of rest mass m .
Solution. The normal mode decomposition of the complex scalar free
field can be easily obtained by a straightforward generalization of the
treatment for the real scalar free field (3.42). The result is evidently
Xh i
Φ(x) = ak u k (x) + bk† u∗k (x)
k
X h i
Π(x) = iωk − bk u k (x) + a†k u∗k (x)
k

u k (x) ≡ [ 2ω k (2π)3 ] −1/2 exp {− i x0 ω k + i k · x}

where ω k = ( k 2 + m2 ) 1/2 together with the canonical commutation

relations

[ Φ(t, x) , Φ(t, y) ] = 0 = [ Π(t, x) , Π(t, y) ]

[ Φ(t, x) , Π(t, y) ] = i δ (x − y)

It is clear that the main difference with respect to the real case is the
appearence of two kinds of creation and destruction operators, as the
reality conditions no longer hold true, which satisfy the algebra

[ ak , ap ] = [ bk , bp ] = 0
[ ak , bp ] = [ ak† , bp ] = 0
[ ak , ap† ] = [ bk , bp† ] = δ (k − p)

Then the normal ordered hamiltonian and momentum operator takes

the diagonal form
Z
H [ Π, Φ ] = d x : | Π(x) | 2 + | ∇Φ(x) | 2 + m2 | Φ(x) | 2 :
X  † †

= ω k ak ak + bk bk = P0
k
Z
P = − d x : Π(x)∇Φ(x) + Π∗ (x)∇Φ∗ (x) :
X  † †

= k ak ak + b k b k
k

It follows therefrom that the complex scalar free field describe two kinds
of particles with the very same value m of the rest mass.

116
 (c) Rewrite the conserved Noether charge
Z h i
Q = i d x Φ∗ (x)Π∗ (x) − Φ(x)Π(x)

in terms of creation and annihilation operators and evaluate the charge

of the particles of each type.
Solution. From the invariance of the classical lagrangian under U (1)
phase transformations Φ(x) 7→ Φ0 (x) = e−iα Φ(x) we immediately get
the above written Noether charge. Moreover, from the normal mode
expansion and the normal ordering prescription we readily obtain
Z
Q = i d x : Φ∗ (x)Π∗ (x) − Φ(x)Π(x) :
X † †

= ak ak − b k b k
k

which is understood so that each particle normal mode carries one unit
of positive charge, whereas each antiparticle normal mode carries one
unit of negative charge, the sign of the charge being conventional.
2. Poincaré covariance
(a) The four energy momentum operators Pµ ( µ = 0, 1, 2, 3 ) and
the six orbital angular momentum operators Lµν = − L νµ are the
infinitesimal operators, or generators, of the Poincaré group, for the
infinite dimensional unitary representation acting on the Fock space of
a real scalar quantized field φ(x) .
(a) Show that [ φ(x) , Lµν ] = i (xµ ∂ν − xν ∂µ ) φ(x)
Solution. We have
T 0µ (x) = Π (x) ∂µ φ (x) − δµ0 L (x)
and the equal–time canonical commutation relations
[ φ (t, x) , L (t, y) ] = Π (t, y) [ φ (t, x) , Π (t, y) ] = i Π (t, x) δ (x − y)
 
0
[ φ (t, x) , Π (t, y) ∂ν φ (t, y) ] = i ∂ν φ (t, y) + δν Π (t, y) δ (x − y)

Then we obtain for x0 = y0 = t

Z
[ φ (x) , Lµν ] = d y yµ [ φ (t, x) , T 0ν (t, y) ] − µ ↔ ν
= i xµ ∂ν φ (t, x) − i xν ∂µ φ (t, x)

117
and thereby
i
[ φ (x) , Lµν ] δ ω µν = x ν ∂µ φ (x)  µν = δ x µ ∂µ φ (x)
2
It follows that the finite passive Lorentz transformations for the real
quantized scalar field read

φ 0 (x) = U (ω) φ (x) U † (ω) = φ (Λ x)

where  
i µν
U (ω) = exp − ω Lµν
2
(b) Show that [ Lµν , Pρ ] = − i g µρ Pν + i g νρ Pµ .
Solution.
Let us first calculate the commutator [ Lµν , P0 ] . To this aim, for any
analytic functional of the scalar field φ (x) and conjugated momentum
Π (x) operators we have

[ F (φ (x), Π (x)) , Pµ ] = i ∂µ F

In fact, for example,

[ φ2 (x) , Pµ ] = 2φ (x) [ φ (x) , Pµ ] = 2φ (x) i∂µ φ (x) = i∂µ φ2 (x)

and iterating we obviously get ∀ n ∈ N

[ φ n (x) , Pµ ] = i ∂µ φ n (x)

[ Π n (x) , Pµ ] = i ∂µ Π n (x)
so that the above statement holds true. Then we find
Z  
[ Lµν , P0 ] = d x xµ [ T 0ν (x) , P0 ] − xν [ T 0µ (x) , P0 ]
Z  
0 0
= d x xµ i ∂ 0 T ν (x) − xν i ∂ 0 T µ (x)
Z  
= d x − xµ i ∂  T ν (x) + xν i ∂  T µ (x)
Z  
˙ i d x g µ T ν (x) − g ν T µ (x)
=
Z  
0 0
= i d x Tµν (x) − g µ 0 T ν (x) − Tνµ (x) − g ν 0 T µ (x)
= − i g µ 0 Pν + i g ν 0 Pµ

118
 in which, as usual, a boundary term has been neglected. Furthermore
we find
Z  
0 0
[ Lµν , P ] = d x xµ [ T ν (x) , P ] − xν [ T µ (x) , P ]
Z  
= d x xµ i ∂  T 0ν (x) − xν i ∂  T 0µ (x)
Z  
˙ − i d x g µ T 0ν (x) − g ν T 0µ (x)
=
= − i g µ  Pν + i g ν  Pµ

up to a boundary term, which completes the proof.

(c) A fully detailed check of the canonical commutation relations

[ Lµν , L ρ σ ] = − i g µ ρ Lνσ + i g ν ρ Lµσ + i g µ σ Lνρ − i g ν σ Lµρ

is straightforward although somewhat tedious and can be found in [21],

7.8, pp. 144 - 147.

3. Special distributions in configuration space.

(a) Evaluate the scalar distribution

i h 0 | φ(x) φ(y) | 0 i = i [ φ (−) (x) , φ (+) (y) ] ≡ D (−) (x − y)

explicitly in terms of Bessel functions.

Solution. Let us consider the positive and negative parts of the Pauli-
Jordan distribution (3.82) in the four dimensional Minkowski space-
time: namely,
±1
Z
(±)
D (x) ≡ exp {± i k · x)} δ (k 2 − m2 ) θ(k0 ) d 4 k
(2π)3 i
±1
Z
= d k (k2 + m2 )−1/2
(2π)3 2i
× exp {± i x0 (k2 + m2 )1/2 ∓ i k · x}
Z ∞
±1 k sin(kr)
≡ 2
dk 2
4iπ r 0 (k + m2 )1/2
× exp ± i x0 (k 2 + m2 )1/2

Z ∞
±i d cos(kr)
= 2
dk 2
4π r dr 0 (k + m2 )1/2
× exp ± i x0 (k 2 + m2 )1/2


119
where r ≡ | x | , k = | k | , whence it is clear that the positive and
negative parts of the Pauli-Jordan commutator are complex conjugate
quantities
[ D (±) (x) ] ∗ = D (∓) (x)
Then we can write
Z ∞
(+) i d cos(kr)
D (x) = 2
· dk 2
8π r d r −∞ (k + m2 )1/2
exp i x (k + m2 )1/2
 0 2
×
Z ∞
i d
= 2
· d k (k 2 + m2 )− 1/2
8π r d r −∞
exp i x0 (k 2 + m2 )1/2 + i kr

×

Consider now the integral

Z ∞
0
I(x , r) = d k (k 2 + m2 )− 1/2
−∞
× exp {i x0 (k 2 + m2 )1/2 + i kr} (x0 > 0)

and perform the change of variable k = m sinh η , so that (k 2 +m2 )1/2 =

m cosh η . Then we obtain
Z ∞
0
I(x , r) = dη exp{im (x0 cosh η + r sinh η)}
−∞

Here x0 > 0 so that two cases should be distinguished, i.e. 0 < x0 < r
and x0 > r . By setting λ ≡ (x0 )2 − x2 it is convenient to carry out
respectively the substitutions
 0 √ √
x = √ −λ sinh ξ , r = √−λ cosh ξ , 0 < x0 < r
x0 = λ cosh ξ , r = λ sinh ξ , x0 > r

in such a way that we can write

√
Z ∞
0
I(x , r) = θ(−λ) dη exp{im −λ sinh(ξ + η)}
Z ∞−∞ √
+ θ(λ) dη exp{im λ cosh(ξ + η)}
−∞
√
Z ∞
= θ(−λ) dη exp {im −λ sinh η}
Z ∞−∞ √
+ θ(λ) dη exp {im λ cosh η} (x0 > 0)
−∞

120
Now we can use the integral representations of the cylindrical Bessel
functions of real and imaginary arguments [23] eq.s 8.4211. p. 965
and 8.4324. p. 969 that yield
h √ (1)
√ i
I(x0 , r) = θ(x0 ) 2θ(−λ) K0 (m −λ) + θ(λ) πi H0 (m λ)
h √ (2)
√ i
+ θ(−x0 ) 2θ(−λ) K0 (m −λ) − θ(λ) π i H0 (m λ)
√
= 2θ(−λ) K0 (m −λ)
h √ √ i
+ π θ(λ) i sgn(x0 ) J0 (m λ) − N0 (m λ)

and finally
i d 1 d
D (+) (x) = 2
I(x0 , r) = 2
I(x0 , λ)
8π r dr 4 i π dλ
We note that in the neighborhood of the origin the cylindrical Bessel
functions of real and imaginary arguments may be represented in the
form
 z 2
J0 (z) = 1 − + O(z 4 )
 2
2  z 2  z 2
N0 (z) = 1− ln + C + O(z 2 )
π 2 2 π
  z 2  z
K0 (z) = − 1 + ln − C + O(z 2 )
2 2
where C is the Mascheroni’s constant. By replacing the differentiation
with respect to x with differentiation with respect to λ and taking
into account the discontinuity of the function I(x0 , x) on the light-cone
manifold λ = 0 we obtain the following expression for the positive part
of the Pauli-Jordan commutator
1 m √
D (+) (x) = sgn(x0 ) δ(λ) − i θ(−λ) 2 √ K1 (m −λ)
4π 4π −λ
m h √ √ i
− i θ(λ) √ N1 (m λ) − i sgn(x0 ) J1 (m λ)
8π λ
so that the Pauli-Jordan commutator and the Feynman propagator are
respectively expressed by

D(x) = D (+) (x) + D (−) (x)

1 m √
= sgn(x0 ) δ(λ) − √ θ(λ) sgn(x0 ) J1 (m λ)
2π 4π λ

121
 DF (x) = i θ(−x0 ) D (+) (x) − i θ(x0 ) D (−) (x)
1 m θ(−λ) √
= δ(λ) + 2 √ K1 (m −λ)
4πi 4π −λ
m h √ √ i
+ θ(λ) √ N1 (m λ) + i J1 (m λ)
8π λ
(b) Evaluate the scalar causal 2-point Green function of order n in
the D−dimensional Minkowski spacetime, which is defined to be
exp {− i k · z} D
Z
(D) i
Gn (z) = D
d k
(2π) (k 2 − m2 + iε) n
explicitly in terms of Bessel functions.
Solution. It is very instructive to first compute the integral
exp {− i k · z} D
Z
D i
In (z) ≡ D
d k
(2π) (k 2 − m2 + iε)n
i (2m)1−n
 n−1 Z 
d exp {− i k · z} D
= d k
(2π)D (n − 1)! dm n−1 k 2 − m2 + iε
where z = (z 0 , z 1 , . . . , z D−1 ) and k = (k 0 , k 1 , . . . , k D−1 ) are coordinate
and conjugate momentum in a D-dimensional Minkowski spacetime,
so that k · z = k 0 z 0 − k 1 z 1 − · · · − k D−1 z D−1 = k 0 z 0 − k · z , while
n is a sufficiently large natural number that will be better specified
further on. Turning to a D-dimensional euclidean space, after setting
z 0 = izD , k 0 = ikD we immediately obtain
(−1)n exp {i kE · zE } D
Z
D
In (z) ≡ d kE
(2π) D (kE2 + m2 )n
with kE = (k, kD ) , zE = (z, zD ) . The spherical polar coordinates of
kE are k, φ, θ1 , θ2 , . . . , θD−2 and we have


 k1 = k cos θ1
k2 = k sin θ1 cos θ2




k3 = k sin θ1 sin θ2 cos θ3


 ·········
k = k sin θ1 sin θ2 · · · sin θD−2 cos φ

 D−1



kD = k sin θ1 sin θ2 · · · sin θD−2 sin φ
with 0 ≤ θi ≤ π for i = 1, 2, . . . , D − 2 and 0 ≤ φ ≤ 2π while k =
|kE | = (k2 + kD
2 1/2
) ≥ 0 . It turns out that
∂ (k1 , k2 , · · · , kD )
= k D−1 (sin θ1 )D−2 (sin θ2 )D−3 · · · (sin θD−2 )
∂ (k, φ, θ1 , · · · , θD−2 )

122
If we choose the euclidean momentum Ok1 axis along zE we evidently
obtain kE · xE = kzE cos θ1 ≡ kzE cos θ and thereby we immediately
obtain
Z ∞
−D
D n
In (z) = (−1) (2π) d k k D−1 (k 2 + m2 )− n
0
Z π
× dθ (sin θ)D−2 exp {i kzE cos θ}
0
D−2
YZ π
× (2π) dθj (sin θj ) D−j−1
j=2 0

Now we have
Z π Z 1
D−j−1
dθj (sin θj ) = 2 d tj (1 − t2j ) (D−j−2)/2
0
Z 10
= d y y −1/2 (1 − y) (D−j)/2−1
0
= B(1/2, D/2 − j/2)
√ Γ(D/2 − j/2)
= π
Γ(D/2 − j/2 + 1/2)

so that
D−2
YZ π
dθj (sin θj ) D−j−1
j=2 0

π (D−3)/2 Γ(1)Γ(3/2)Γ(2) · · · Γ(D/2 − 1)

=
Γ(3/2)Γ(2) · · · Γ(D/2 − 1)Γ(D/2 − 1/2)
π (D−3)/2
=
Γ(D/2 − 1/2)

and thereby

2 (−1)n (4π)− D/2 ∞

Z
InD (z) = √ d k k D−1 (k 2 + m2 )− n
Γ(D/2 − 1/2) π 0
Z π
× dθ (sin θ)D−2 exp {i kzE cos θ}
0

Next we find
Z π
dθ (sin θ)D−2 exp {i kzE cos θ}
0

123
 Z π/2
= 2 dθ (sin θ)D−2 cos(kzE cos θ)
Z0 1
= 2 (1 − t2 )(D−3)/2 cos(tkzE ) d t
0

The value of the latter integral is reported in [23] eq. 3.7717. p. 464
and turns out to be
D/2−1 
√
 
2 D−1
π Γ JD/2−1 (kzE ) (<e D > 1)
kzE 2
so that we further obtain
−D/2+1
InD (z) = (−1)n (4π)− D/2 2D/2 zE
Z ∞
× k D/2 (k 2 + m2 )− n JD/2−1 (kzE ) d k
0

and from [23] eq. 6.5654. p. 710 we come to the expression

−D/2+1
InD (z) = (−1)n (4π)− D/2 2 D/2 zE
m D/2−n zEn−1
× KD/2−n (mzE )
2 n−1 Γ(n)
D/2−n
2 (−1)n

2m
= KD/2−n (mzE )
(4π) D/2 Γ(n) zE
with 0 < <e D < 4n − 1 .

Another much more quick method to get the same result is in terms of
the Mellin’s transform
(−1)n
Z
D
In (z) = dd kE exp { i kEµ zEµ }
Γ(n) (2π)D
Z ∞
× d t t n−1 exp {−t kE2 − t m2 }
0
(−1)n ∞
Z
= d t t n−1 exp{− t m2 }
Γ(n) 0
zE 2 zE2
Z   
1 d
× d kE exp − t kE − i −
(2π)D 2t 4t
n Z ∞
(−1)
= d t t n−1−D/2 exp {− t m2 − zE2 /4t}
(4π)D/2 Γ(n) 0
D/2−n  √
2 (−1) n

2m 
= √ K D/2−n m − λ
(4π)D/2 Γ(n) −λ

124
where zE = (z2 + z42 )1/2 = (z2 − z02 )1/2 = (− λ)1/2 , z 2 < 0 . In the
case n = 1 , D = 4 we recover the Feynman propagator outside the
light-cone
m  √ 
DF (z) = − I14 (zE ) = 2 √ K1 m − λ (z 2 < 0)
4π −λ
and from the series representation of the Basset function of order one
 
1 z z 1 1
K1 (z) = + ln + C − ψ(2) + O(z 3 ln z)
z 2 2 2 2

we obtain the leading behaviour of the causal Green’s function in the

four dimensional Minkowski spacetime near the outer surface of the
light-cone: namely,

−1 m2
DF (z) ≈ + ln (m | λ | 1/2 ) (λ = z 2 < 0)
4π 2 λ 8π 2

On the other side of the light-cone, i.e. for z 2 > 0 , we have to use the
integral representation
n
(− i)n ∞
 Z
1
2 2
= d t t n−1 exp {i t (k 2 − m2 + iε)}
k − m + iε Γ(n) 0

so that
Z
1
InD (z) = dd k exp { − i kµ z µ }
(2π)D
(− i)n ∞
Z
× d t t n−1 exp {i t (k 2 − m2 + i0)}
Γ(n) 0
(− i)n ∞ z2
Z   
n−1 2
= dt t exp − i m t +
Γ(n) 0 4m2 t
Z   
1 d z 2
× d k exp i t k −
(2π)D 2t
n Z ∞   
(− i ) n−D/2−1 2 λ
= dt t exp − i m t +
(4π)D/2 Γ(n) 0 4m2 t

with λ = z 2 > 0 . Now we have [23] formula 3.47111. p. 384

Z ∞
β2
    
ν −1 1 1 (1)
dt t exp iµ t + = π i exp − π iν β ν H−ν (βµ)
0 2 t 2

125
with =m µ > 0 , =m (β 2 µ) ≥ 0 . Hence we obtain
√
D/2−n
π (− 1) n+1
 
D 1 2m (2)
In (z) = D/2
exp πiD √ HD/2−n (m λ)
(4π) Γ(n) 4 λ
For n = 1 , D = 4 we recover the Feynman propagator with a time-like
argument
m h √ √ i
DF (z) = √ N1 (m λ) + iJ1 (m λ) (z 2 = λ > 0)
8π λ
in agreement with the very last expression of Problem 2. From the series
representations of the Bessel functions we obtain the leading behaviours
z2
 
z 4
J1 (z) = 1+ + O(z )
2 8
2 z z 
N1 (z) = − + ln + C + O(z) (3.124)
πz π 2
whence
1 m2
DF (z) ≈ − + ln (m | λ | 1/2 ) (λ = z 2 > 0)
4π 2 λ 8π 2

Finally, consider the causal Green’s function in the four dimensional

Minkowski spacetime, that means n = 1 , D = 4 . To this concern it is
convenient to set

z ≡ |z| k = |k| k · z = kz cos θ

and thereby
∞ ∞
exp{− iz 0 k0 }
Z Z
0 i
DF (z , z) = d k k sin(kz) d k0
z (2π)3 −∞ −∞ k02 − k 2 − m2 + iε
The last integral has two simple poles in the complex energy plane
√
k0 = ω(k) − iε k0 = − ω(k) + iε ω(k) = k 2 + m2

For z 0 > 0 we have to close the contour in the lower half-plane of the
complex energy, that yields
Z ∞
0 θ(z 0 ) exp{− i(z 0 − i0)ω(k)}
DF (z , z) = d k k exp{ikz}
iz (2π)2 −∞ 2ω(k)
θ(z 0 )
Z ∞
dk k n √ o
= √ 2
exp ikz − i(z0 − i0) k + m 2
8izπ 2 −∞ k 2 + m2

126
Conversely, for z 0 < 0 we have to close the contour in the upper half-
plane =m (k0 ) > 0 that gives

θ(−z 0 ) ∞ d k k
Z
0
DF (z , z) = √
8izπ 2 −∞ k 2 + m2
n √ o
× exp −ikz + i(z0 + i0) k 2 + m2

As a consequence, for z 0 = 0 we obtain

Z ∞
1 d k k sin(kz)
DF (0, z) = 2
√
8zπ −∞ k 2 + m2
 Z ∞
1 d d k cos(kz)
= 2
− √
4zπ dz 0 k 2 + m2
m
= K1 (mz)
4zπ 2
and thanks to Lorentz covariance
m  √ 
DF (z) = √ K1 m −z 2 (z 2 < 0)
4π 2 −z 2
in accordance with the previously obtained result. Finally, when m = 0
we find
1
lim DF (z 0 , z) = ×
m→0 8izπ 2
Z ∞
d k θ(z 0 ) exp{ik(z − z 0 )} + θ(−z 0 ) exp{−ik(z − z 0 )}
 
−∞
1 1
= δ(z02 − z 2 ) = δ(λ)
4πi 4πi
and consequently we eventually come to the singular behaviour in the
neighbourhood of the light-cone

1 1 m2
DF (z) ≈ δ(λ) − + ln (m | λ | 1/2 ) (z 2 = λ ∼ 0)
4πi 4π 2 λ 8π 2

127
Chapter 4

The Spinor Field

4.1 The Dirac Equation

We have already obtained the Poincaré invariant and parity-even kinetic term
(2.63) for the Dirac wave field as well as the other parity-even local invariant
(2.62) quadratic in the Dirac spinor fields. Then, it is easy to set up the most
general Lagrange density for the free Dirac field, which satisfies the general
requirements listed in Sect. 2.2 : namely,
↔
LD = 21 ψ̄ (x) γ µ i ∂ µ ψ (x) − M ψ̄ (x) ψ (x) (4.1)
Beside this form of the Dirac lagrangian, in which the kinetic term contains
↔
the left-right derivative operator ∂ , we can also use the equivalent form, up
to a four divergence term,
−
LD = ψ̄ (x) γ µ i ∂ µ ψ (x) − M ψ̄ (x) ψ (x)
˙ LD − 12 i ∂ µ ψ̄ (x) γ µ ψ (x)


= (4.2)
Notice that the spinor fields in the four dimensional Minkowski spacetime
have canonical dimensions in natural units [ ψ ] = cm−3/2 = eV 3/2 . The free
spinor wave equation can be obtained as the Euler-Lagrange field equation
from the above lagrangian by treating ψ(x) and ψ̄ (x) as independent fields.
This actually corresponds to take independent variations with respect to
<e ψα (x) and =m ψβ (x) , in which the spinor component indices run over the
values α, β = 1L, 2L, 1R, 2R . Then we obtain the celebrated Dirac equation
(i ∂/ − M ) ψ(x) = 0 (4.3)
where I have employed the customary notation
i ∂/ ≡ γ µ i∂µ

128
 Paul Adrien Maurice Dirac (1928)
The Quantum Theory of the Electron
Proceedings of the Royal Society, Vol. A117, p. 610
Taking the hermitean conjugate of the Dirac equation

0 = i ∂ µ ψ † (x) γ µ † + M ψ † (x) = i ∂ µ ψ † (x) γ 0 γ µ γ 0 + M ψ † (x) (4.4)

and after multiplication by γ 0 from the right we come to the adjoint Dirac
equation
←
i ∂ µ ψ̄ (x) γ µ + M ψ̄ (x) ≡ ψ̄ (x) ( i ∂/ +M ) = 0 (4.5)

The Dirac equation (4.3) can also be written à la Schrödinger in the form

∂ψ
i~ = Hψ H = αk pk + βM (4.6)
∂t
where H denotes the 1-particle hamiltonian selfadjoint operator with αk =
γ 0 γ k , pk = i∂k . Owing to the transformation rule (2.82) it is immediate to
verify the covariance of the Dirac equation, that means

(i ∂/ 0 − M ) ψ 0 (x 0 ) = i γ µ Λµν ∂ν − M Λ 1 (ω) ψ(x)

2

= Λ (ω) (i ∂/ − M ) ψ(x) = 0
1 (4.7)
2

To solve the Dirac equation, let us first consider the plane wave stationary
solutions

ψ p (x) = Γ(p) exp {− i p · x} (4.8)

where the spinor Γ(p) fulfills the algebraic equation

(p/ − M ) Γ(p) = 0 ⇔ (H − p0 ) Γ(p) = 0 (4.9)

H being the hermitean matrix (4.6), which admits nontrivial solutions iff

det k p/ − M k= 0 (4.10)

This determinant, which is obviously independent of the γ−matrices specific

representation, can be most easily computed in the so called ordinary or
standard or even Dirac representation, that is
   
0 1 0 0 σk
γD ≡β= γ Dk =  (4.11)
 
  

0 −1 −σk 0
   

129
If we choose the momentum rest frame p = 0 we find solutions iff
det k p/ − M k= (p0 − M )2 (p0 + M )2 = 0 (4.12)
and using Lorentz covariance
det k p/ − M k= (p2 − M 2 )2 = 0
which drives to the two pairs of degenerate solutions with frequencies

1
p0± = ± p 2 + M 2 2 ≡ ± ω p (4.13)
As a consequence, it follows that we have two couples of plane wave stationary
solutions (4.8) with two possible polarization states (r = 1, 2) :

Γ − , r ( p) e− i p x

p0 = ω p
ψ p , r (x) = ( r = 1, 2 ) (4.14)
Γ + , r ( − p) e i p x p0 = − ω p
with
( p/ − M ) Γ − , r ( p) = 0 (−p/ − M ) Γ + , r ( − p) = 0 ( p0 = ω p )
Actually, it is a well established convention to set
1 1
Γ − , r ( p) ≡ [ (2π)3 2ω p ] − 2 u r ( p) = [ (2π)3 2ω p ] − 2 u r (p)
1 1
Γ + , r ( − p) ≡ [ (2π)3 2ω p ] − 2 v r ( p) = [ (2π)3 2ω p ] − 2 v r (p)
together with
u p , r (x) = [ (2π)3 2ω p ] −1/2 u r (p) exp {− i t ω p + i p · x)}
v p , r (x) = [ (2π)3 2ω p ] −1/2 v r (p) exp {+ i t ω p − i p · x)}
The so called spin-amplitudes or spin-states do fulfill
(ω p γ 0 − γ k pk − M ) u r (p) = 0 ( r = 1, 2 ) (4.15)
(ω p γ 0 − γ k pk + M ) v r (p) = 0 ( r = 1, 2 ) (4.16)
which is nothing but that the degenerate solution of the eigenvalue problem
H u r (p) = ω p u r (p) H v r (−p) = − ωp v r (−p) ( r = 1, 2 ) (4.17)
together with the orthonormality and closure relations
u r† (p) us (p) = 2ω p δrs = v r† (p) vs (p) (4.18)
X 
u r (p) ⊗ ur† (p) + v r (−p) ⊗ vr† (−p) = 2ω p

(4.19)
r=1,2

130
The conventional normalization (4.18) is fixed in order to recover
Z Z
d x u q , s (x) u p , r (x) = δ rs δ (p − q) = d x v †q , s (x) v p , r (x)
†
(4.20)

while the closure relation (4.19) indeed ensures the completeness relation
X  
u p , r (x) ⊗ u †p , r (y) + v p , r (x) ⊗ v †p , r (y) = δ (x − y)
x0 = y0
p,r

in which the notation has been introduced for brevity

X def X Z
= dp
p,r r=1,2

It is worthwhile to notice that, from the covariance (4.7) of the Dirac

equation, the transformation property of the spin-states readily follows. For
instance, from eq. (2.82) we find

 
( p/ 0 − M ) u r ( p 0 ) = Λ µν p ν γ µ − M u r ( Λ p)
 
= Λ µν p ν Λ µρ Λ 1 γ ρ Λ−1 1 − M u r ( Λ p)
2 2
−1
= Λ 1 (p/ − M ) Λ 1 u r ( Λ p)
2 2

= Λ 1 (p/ − M ) u r ( p) = 0 (4.21)
2

Hence, the Lorentz covariance of the Dirac equation actually occurs, provided
the following relationships hold true for the spin-states

u r ( p 0 ) = u r ( Λ p) = Λ 1 u r ( p) ⇔ u r ( p ) = Λ −1
1 ur ( Λ p ) (4.22)
2 2

v r ( p 0 ) = v r ( Λ p) = Λ 1 v r ( p) ⇔ v r ( p ) = Λ −1
1 vr ( Λ p ) (4.23)
2 2

It is not difficult to prove the further equalities : namely,

ū r (p) us (p) = 2M δrs = − v̄ r (p) vs (p) (4.24)

u r† (p) vs (−p) = 0 ū r (p) vs (p) = 0
†
v r (p) us (−p) = 0 v̄ r (p) us (p) = 0 (r, s = 1, 2) (4.25)

Proof. From eq. (4.15) we obviously get

us† (p) (ω p γ 0 − γ k pk − M ) u r (p) = 0 ( ∀ r, s = 1, 2)

131
and taking the adjoint equation

us† (p) (ω p γ 0 + γ k pk − M ) u r (p) = 0 ( ∀ r, s = 1, 2)

so that adding together

M †
ūs (p) u r (p) = u (p) u r (p) = 2M δrs
ωp s

In a quite analogous way one can readily prove that

M †
v̄s (p) v r (p) = − v (p) v r (p) = − 2M δrs
ωp s

Moreover, from eq.s (4.15) and (4.16) we obtain

v̄s (−p) (ω p γ 0 − γ k pk − M ) u r (p) = 0 ( ∀r, s = 1, 2)

ūs (p) (ω p γ 0 + γ k pk + M ) v r (−p) = 0 ( ∀r, s = 1, 2)

and taking the hermitean conjugated of the very last equality

v̄s (−p) (ω p γ 0 + γ k pk + M ) u r (p) = 0 ( ∀r, s = 1, 2)

so that, by summing up the first equality, we get

vs† (−p) u r (p) = 0 ( ∀r, s = 1, 2)

and analogously
u†s (p) v r (−p) = 0 ( ∀r, s = 1, 2)

which completes the proof. Quod Erat Demonstrandum

Notice that from the above orthogonality properties of the spin-states the
following orthogonality relations hold true between the positive and negative
frequency eigenspinor wave functions
Z Z
d x u q , s (x) v p , r (x) = 0 = d x v †q , s (x) u p , r (x)
†
(4.26)

It follows therefrom that the most general solution of the Dirac equation
can be written in the form
X h i
∗
ψ(x) = c p , r u p , r (x) + d p , r v p , r (x) (4.27)
p,r
X h i
ψ̄(x) = c ∗p , r ū p , r (x) + d p , r v̄ p , r (x) (4.28)
p,r

which is nothing but the normal mode expansion of the free Dirac spinor
classical wave field, where c p , r and d p , r are arbitrary complex coefficients.

132
 In the chiral representation (2.68) for the gamma matrices we can build
up a very convenient set of spin-states as follows. Consider the matrices
±M ω p − pz −px + i py 
 
 0
±M −px − i py ω p + pz 
 
 0
p/ ± M = 
 
  (4.29)
− ±M
 

 ω p + p z p x i p y 0 


px + i py ω p − pz ±M
 
0

where we have set p = (p1 , p2 , p2 ) = (px , py , pz ) . Notice that we can define the
two projectors on the bidimensional spaces spanned by the positive energy
and negative energy spin-states respectively : namely,

E ± ( p) ≡ (M ± p/) / 2M ( p0 = ω p ) (4.30)

which satisfy by definition

E ±2 = E ± E+ E− = 0 = E− E+
tr E ± = 2 E+ + E− = I

Moreover we have

E ±† ( p) = E ± ( p̃ ) p̃ µ = p µ (4.31)

Now, if we introduce the constant bispinors

−1
       
 1   0   0   
       
 0   1   1   0 
ξ1 ≡  ≡ ≡ η2 ≡ 
       
 
 ξ 2

 
 η 1

 
  



 1 




 0 




 0 




 1 


−1
       
0 1 0

which are the common eigenvectors of the γ 0 matrix

γ0 ξ r = ξ r γ0 η r = − η r ( r = 1, 2 )
1
and of the spin matrix 2
Σ3 = 4i [ γ 1 , γ 2 ]

(Σ3 − 1) ξ1 = (Σ3 − 1) η2 = 0 (Σ3 + 1) ξ2 = (Σ3 + 1) η1 = 0

and do indeed satisfy by direct inspection

ξ r> γ k ξ s = 0 = η r> γ k η s ∀ r, s = 1, 2 ∨ k = 1, 2, 3

then we can suitably define

u r (p) ≡ 2M (2ω p + 2M ) −1/2 E + ξ r

( r = 1, 2 ) (4.32)
v r (p) ≡ 2M (2ω p + 2M ) −1/2 E − η r

133
the explicit form of which is given by
M + ω p − pz 
 

 −px − i py 

 
u1 (p) = (2ω p + 2M ) −1/2 


 


 M + ω p + p 
z 

 
px + i py
−px + i py 
 


 M +ω +p  
−1/2  p z 
u2 (p) = (2ω p + 2M ) 
 p − ip
 



 x y 

M + ω p − pz
 

−px + i py
 

 
 M +ω +p  
−1/2  p z 
v1 (p) = (2ω p + 2M )  
−p
 


 x + i p y



−M + pz − ω p
 

−M − ω p + pz 
 

 

−1/2  px + i py 

v2 (p) = (2ω p + 2M ) 
 
 (4.33)


 ω p + M + pz  

 
px + i py
their orthonormality and completeness relations being in full accordance with
formulæ (4.18), (4.19) and (4.25). In fact we have for instance
v r† ( p) v s ( p) = (2ω p + 2M ) −1 η r> (M − p̃/) (M − p/)η s
= (2ω p + 2M ) −1 η r> M 2 − 2M γ 0 ω p + ω 2p + p 2 η s

= (2ω p + 2M ) −1 η r> 2ω 2p + 2M ω p η s

= 2ω p 1
2
η r> η s = 2ω p δ rs
in which I have made use of the property
η r> γ k γ 0 η s = − η r> γ k η s = 0 ∀ r, s = 1, 2 ∨ k = 1, 2, 3
Finally, taking the normalization (4.24) into account together with
E + u r ( p) = u r ( p) E − v r ( p) = v r ( p) ( r = 1, 2 )
it is immediate to obtain the so called sums over the spin-states, that is
X  u r (p) ⊗ ū r (p) = 2M E + (p)
(4.34)
v r (p) ⊗ v̄ r (p) = − 2M E − (p)
r=1, 2

or even for α, β = 1L, 2L, 1R, 2R ,

X  u r , α (p) ū r , β (p) = ( p/ + M ) αβ
( p0 = ω p ) (4.35)
v r , α (p) v̄ r , β (p) = ( p/ − M ) αβ
r=1, 2

134
Hence, from the orthonormality relations (4.24) we can immediately verify
that we have
X
u r (p) ⊗ ū r (p) u s (p) = 2M u s (p) = 2M E + u s (p)
r=1, 2
X
v r (p) ⊗ v̄ r (p) v s (p) = − 2M v s (p) = − 2M E − v s (p)
r=1, 2

135
4.2 Noether Currents
From Noether theorem and from the Lagrange density (4.1) we obtain the
canonical energy momentum tensor of the free Dirac spinor wave field which
turns out to be real though not symmetric

T µν (x) ≡ (δ LD / δ ∂µ ψ) ∂ν ψ + ∂ν ψ̄ (δ LD / δ ∂µ ψ̄) − δ νµ LD
i µ µ

= ψ̄ (x) γ ∂ν ψ (x) − ∂ν ψ̄ (x) γ ψ (x) (4.36)
2
T µν (x) 6= T νµ (x)
where we have taken into account that the Dirac lagrangian vanishes if the
equations of motion hold true as it occurs in the Noether theorem. The
corresponding canonical total angular momentum density tensor for the Dirac
field can be obtained from the general expression (2.100) and reads
def
M µκλ (x) = x κ T µλ (x) − x λ T µκ (x) + S µκλ (x)
= x κ T µλ (x) − x λ T µκ (x) + 12 ψ̄ (x) {σ λκ , γ µ } ψ (x)

As a matter of fact we have

δ L / δ ∂ µ ψ (x) = 12 ψ̄ (x) i γ µ δ L / δ ∂ µ ψ̄ (x) = − 12 i γ µ ψ (x)

(− i )( S λ κ ) ψ (x) = − i σ λ κ ψ (x) i ψ̄ (x) ( S λ κ ) = i ψ̄ (x) σ λ κ

where
i  λ κ
†
σ λκ ≡ γ ,γ σ λ κ = γ0 σ λ κ γ0
4
is the spin tensor for the Dirac field. Hence from the general expression
(2.100) we get

def δL
S µ λ κ (x) = (− i ) S λ κ AB uB (x)


δ ∂ µ uA (x)
δL/δ ∂ µ ψ (x) (− i ) S λ κ ψ (x)


=
i ψ̄ (x) S λ κ δL/δ ∂ µ ψ̄ (x)


+
= 1
2
ψ̄ (x) { γ µ , σ λκ } ψ (x)

Notice that

M 0 jk (x) = x j T 0 k (x) − x k T 0 j (x) + ψ † (x) σ j k ψ (x) (4.37)

M 0 k 0 (x) = x k T 0 0 (x) − x 0 T 0 k (x) (4.38)

136
 It is rather easy to check, using the Dirac equation, that the continuity
equations actually hold true

∂ µ T µν = 0 ∂µ M µλκ = 0 (4.39)

which lead to the four conserved energy momentum Noether charge

Z Z ↔
Pµ = d x T µ (x , x) = 2 d x ψ † (x) i ∂ µ ψ (x)
0 0 1

Z
= d x ψ † (x) i ∂µ ψ (x) (4.40)

while from the spatial integration of eq. (4.37) we obtain the three conserved
Noether charges corresponding to the spatial components of the relativistic
total angular momentum
Z
M jk = d x M 0jk (t, x) =˙
Z
d x x j ψ † (x) i ∂ k ψ (x) − { j ↔ k } + ψ † (x) σ jk ψ (x) (4.41)
 

in which we have discarded, as customary, the boundary term

Z  
1 †
2
i d x ∂ k x j ψ (t, x) ψ(t, x) − { j ↔ k } = 0

Furthermore, from the spatial integration of eq. (4.38) we find

Z Z
0k 0k0
M = dx M (t, x) = d x x k T 0 0 (t, x) − x 0 P k (4.42)

in such a way that the constancy in time of these latter spacetime components
of the relativistic total angular momentum leads to the definition of the
velocity for the center of the energy, viz.,
Z
k def −1
Xt = M d x x k T 0 0 (t, x)

which corresponds to the relativistic generalization of the center of mass,

that means
Z
Ṁ = 0 ⇔ P = M Ẋ t = d x x k Ṫ 0 0 (t, x)
0k k k
(4.43)

It follows that the so called center of momentum frame P = 0 just coincides

with the inertial reference frame in which the center of the energy is at rest.

137
 Notice however that owing to the lack of symmetry for the canonical
energy momentum tensor of the Dirac field 1 its spin angular momentum
tensor is not constant in time. Actually we find, for instance,

∂ µ S µjk (x) = T jk − T kj 6= 0 (4.44)

and consequently the corresponding Noether charges are not conserved in

time so that
Z Z
S ij x = d x S ij (t, x) = 2 ε ijk d x ψ † (x) Σ k ψ (x)
0 0 1


(4.45)

where
 
 σk 0
Σk ≡  (4.46)


0 σk
 

However, in the case that the spinor wave field ψ does not depend on some
of the spatial coordinates (x1 , x2 , x3 ) = (x, y, z) it is possible to achieve that
the continuity equation should hold for some of the components of the spin
angular momentum density tensor and, consequently, that the corresponding
Noether charges – i.e. its space integrals – remain constant in time. Thus,
for example, if ∂ 1 ψ = 0 = ∂ 2 ψ so that ψ(t, z) = ψ (t, 0, 0, z) and cosequently
we obtain
↔ ↔
2i T 12 = ψ̄ (t, z) γ 1 ∂ y ψ (t, z) = 0 = ψ̄ (t, z) γ 2 ∂ x ψ (t, z) = 2i T 21

and thereby
µ µ
∂ µ M 12 = ∂ µ S 12 =0 (4.47)

Hence it follows that the component of the spin vector along the direction of
propagation, which is named the helicity, is conserved in time : namely,
Z ∞
dh def
= 0 h = d z 12 ψ † (t, z) Σ 3 ψ (t, z) (4.48)
dt −∞

After insertion of the normal modes expansions (4.28) one gets

Z ∞ h i
1 ∗ ∗ ∗ ∗
h= 2 dp cp,1 cp,1 − cp,2 cp,2 − dp,1 dp,1 + dp,2 dp,2 (4.49)
−∞
1
The lack of symmetry in the canonical energy momentum tensor of the Dirac field can
be removed using a trick due to Frederik J. Belifante, On the spin angular momentum of
mesons, Physica 6 (1939) 887-898, and L. Rosenfeld, Sur le tenseur d’impulsion-énergie,
Mémoires de l’Academie Roy. Belgique 18 (1940) 1-30, see Problem 1.

138
Proof. By passing in (4.46) to the momentum representations (4.28) and carrying out the
integration over the three dimensional configuration space we obtain
Z ∞
h = dz 12 ψ † (t, z) Σ 3 ψ (t, z)
−∞
Z ∞ X h i
= dz c ∗p , r u †p , r (t, z) + d p , r v p† , r (t, z)
−∞ p,r
X h i
× 1
2 Σ3 c q , s u q , s (t, z) + d ∗q , s v q , s (t, z)
q,s
√
in which we have set p = (0, 0, p) , q = (0, 0, q) , ω p = (p2 + M 2 ) together with

u p , r (t, z) = [ 4πω p ] −1/2 u r (p) exp{− i t ω p + i pz} (r = 1, 2)

−1/2
v q , s (t, z) = [ 4πω q ] v s (q) exp{+ i t ω q − i qz} (s = 1, 2)
the normalization being now consistent with the plane waves independent of the transverse
x> = (x1 , x2 ) spatial coordinates. From (4.25) and the commutation relation
[ ω pγ0 − p γ3 , Σ 3 ] = 0 (4.50)
together with the definition (4.32)
u r (p) ≡ (2ω p + 2M ) −1/2 M + ω p γ 0 − p γ 3
ξ r


(r = 1, 2) (4.51)
v r (p) ≡ (2ω p + 2M ) −1/2 M − ω p γ 0 + p γ 3 η r
it can be readily derived that
( Σ 3 − 1 ) u1 (p) = ( Σ 3 − 1 ) v2 (p) = 0
( Σ 3 + 1 ) u2 (p) = ( Σ 3 + 1 ) v1 (p) = 0
u r† (p) Σ 3 vs (− q) = 0 v r† (− p) Σ 3 us (q) = 0 (r, s = 1, 2) (4.52)
Hence the spin component along the direction of propagation, which is named helicity of
the Dirac spinor wave field, turns out to be time independent and takes the form
Z ∞
dp X h ∗ i
h = c p , r c p , r u r† (p) Σ 3 u r (p) + d p , r d ∗p , r v r† (p) Σ 3 v r (p)
−∞ 4 ω p r=1 , 2
Z ∞ h i
= 21 d p c ∗p , 1 c p , 1 − c ∗p , 2 c p , 2 − d p , 1 d ∗p , 1 + d p , 2 d ∗p , 2
−∞

which proves equation (4.49) and shows that particles and antiparticles exhibit opposite
helicities. Quod Erat Demonstrandum

Finally, the Dirac lagrangian is manifestly invariant under the internal

symmetry group U (1) of the phase transformations
ψ 0 (x) = ei q α ψ(x) ψ̄ 0 (x) = ψ̄(x) e− i q α 0 ≤ α < 2π
where q denotes the particle electric charge, e.g. q = − e (e > 0) for the
electron. Then from eq. (2.103) the corresponding Noether current becomes
J µ (x) = q ψ̄ (x) γ µ ψ (x) (4.53)

139
and will be identified with the electric current carried on by the spinor field,
which transforms as a true four vector under the space inversion (2.59) or,
more generally, under the improper orthochronus Lorentz transformations of
L↑− , that means

J00 (x0 , − x) = J0 (x) J 0 (x0 , − x) = − J (x)

If instead we consider the internal symmetry group U (1) of the chiral

phase transformations

ψ 0 (x) = exp{− i α γ5 } ψ(x) ψ̄ 0 (x) = ψ̄(x) exp{− i α γ5 } 0 ≤ α < 2π

then the free Dirac Lagrange density is not invariant, owing to the presence
of the mass term, so that the Noether theorem (2.93) yields in this case
 
µ µ
∂µ J5 (x) = ∂µ ψ̄ (x) γ γ5 ψ (x) = 2 i M ψ̄ (x) γ5 ψ (x) (4.54)

Notice that the axial vector current J5µ (x) is a pseudovector since we have
the transformation law under the space inversion (2.59)

J50 0 (x0 , − x) = − J 50 (x) J5k 0 (x0 , − x) = J 5k (x)

It follows therefrom that the chiral phase transformations are a symmetry

group of the classical free Dirac theory only in the massless limit.

140
4.3 Quantization of a Dirac Field
The above discussion eventually leads to the energy momentum vector of the
free Dirac wave field that reads
Z Z
P µ = d x T µ (x , x) = d x ψ † (x) i∂µ ψ (x)
0 0
(4.55)

and inserting the normal mode expansions (4.28) we obtain

Z Xh i
Pµ = dx c ∗q , s u †q , s (x) + d q , s v †q , s (x)
q,s
X h i
× p µ c p , r u p , r (x) − d ∗p , r v p , r (x) (4.56)
p,r

where we have set p µ ≡ (ω p , − p) . Taking the orthonormality relations

(4.20) and (4.26) into account we come to the expression
X  
Pµ = p µ c ∗p , r c p , r − d p , r d ∗p , r (4.57)
p,r

Now the key point: as we shall see in a while, in order to quantize the
relativistic spinor wave field in such a manner to obtain a positive semidefinite
energy operator, then we must impose canonical anticommutation relations.
As a matter of fact, once the normal mode expansion coefficients turn into
creation and destruction operators, that means

c p , r , c ∗p , r 7→ c p , r , c †p , r d p , r , d ∗p , r 7→ d p , r , d †p , r

had we assumed the canonical commutation relations

[ d p , r , d †q , s ] = δ rs δ(p − q)

as in the scalar field case, then we would find

X  
Pµ = p µ c †p , r c p , r − d †p , r d p , r − 2 U0 g µ 0
p,r

where U0 is again the divergent zero-point energy (3.46)

Z
X
1 ω p dp
c U0 = δ(0) 2
~ ωp = V ~
p
2(2π)3
V ~c K
Z p
2
= dp p p2 + M 2 c2 /~2 (4.58)
4π 2 0

141
whereas V is the volume of a very large box and ~K  M c is a very large
wavenumber, the factor two being due to spin. This means, however, that
even assuming normal ordering prescription to discard U0 , still the spinor
energy operator P0 is no longer positive semidefinite.

Turning back to the classical spinor wave field, it turns out that it is not
convenient to understand the normal mode expansion coefficients

d p , r r = 1, 2 , p ∈ R3

as ordinary complex numbers. On the contrary, we can assume all those

coefficients to be anticommuting numbers, also named Graßmann numbers 2

{d p , r , d q , s } = 0 = {d p , r , d ∗q , s } ( ∀ r, s = 1, 2 p, q ∈ R3 )

in which {a, b} = ab + ba . Notice en passant that in the particular case r = s

and p = q we find d 2q , s = 0 = d ∗q 2, s . Internal consistency then requires that
also the normal mode expansion coefficients c p , r ( r = 1, 2 , p ∈ R3 ) must
be taken Graßmann numbers, in such a manner that the whole classical Dirac
spinor relativistic wave field becomes Graßmann valued so that

{ψ (x) , ψ (y)} = {ψ (x) , ψ̄ (y)} = {ψ̄ (x) , ψ̄ (y)} = 0 (4.59)

Under this assumption, the canonical quantization of such a system is

then achieved by replacing the Graßmann numbers valued coefficients of the
normal mode expansion by creation annihilation operators acting on a Fock
space and postulating the canonical anticommutation relations, that means

{c p , r , c †q , s } = δ rs δ (p − q) = {d p , r , d †q , s }
( ∀ r, s = 1, 2 p, q ∈ R3 )
all the other anticommutators vanishing (4.60)

As a consequence, the quantized Dirac spinor wave field becomes an operator

valued tempered distribution acting on a Fock space and reads
X h i
ψ (x) = c p , r u p , r (x) + d †p , r v p , r (x) (4.61)
p,r
X h i
ψ † (x) = c †p , r u > ∗ >∗
p , r (x) + d p , r v p , r (x) (4.62)
p,r

2
Hermann Günther Graßmann (Stettino, 15.04.1809 – 26.09.1877) has introduced the
modern linear algebra in his masterpiece : Die Lineale Ausdehnungslehre, ein neuer Zweig
del Mathematik (Linear Extension Theory, a New Branch of Mathematics (1844).

142
where it has been suitably stressed that the symbol † is indeed referred to the
hermitean conjugation of operators acting on a Fock space. The canonical
anticommutation relations (4.60) actually imply

{ψα (t, x) , ψβ (t, y)} = 0 = {ψα† (t, x) , ψβ† (t, y)} (4.63)
{ψα (t, x) , ψβ† (t, y)} = δ (x − y) δ αβ (4.64)

( α, β = 1L, 2L, 1R, 2R )

Then, if we adopt once again the normal product to remove the divergent
and negative zero-point energy contribution to the cosmological constant,
we come to the operator expression for the energy momentum of the Dirac
spinor quantum free field: namely,
Z
P0 = d x : ψ † (x) i∂0 ψ (x) :
Z
= d x : ψ † (x) H ψ (x) :
X h i
= ω p c †p , r c p , r + d †p , r d p , r (4.65)
p,r
Z
P = d x : ψ † (x) (− i ∇ ) ψ (x) :
X h i
= p c †p , r c p , r + d †p , r d p , r (4.66)
p,r

It is important to remark that the canonical anticommutation relations

{d p , r , d †q , s } = δ rs δ (p − q) {d p , r , d q , s } = 0

( ∀ r, s = 1, 2 p, q ∈ R3 )
do guarantee the positive semidefiniteness of the energy operator P0 , while
the remaining anticommutators (4.60) can be derived from the requirement
that the energy momentum operators P µ do realize the self-adjoint generators
of the spacetime translations of a unitary representation of the Poincaré
group. As a matter of fact

[ P µ , ψ (x) ] =
X h  X i
p µ c †p , r c p , r + d †p , r d p , r , c q , s u q , s (x) +
p,r q,s
X h  X i
† † †
pµ cp,r cp,r + dp,r dp,r , d q , s v q , s (x) =
p,r q,s

143
 X X  
pµ u q , s (x) c †p , r {c p , r , c q , s } − {c †p , r , c q , s } c p , r +
p,r q,s
X X  
pµ u q , s (x) d †p , r {d p , r , c q , s } − {d †p , r , c q , s} d p , r +
p,r q,s
X X  
pµ v q , s (x) c †p , r {c p , r , d †q , s } − {c †p , r , d †q , s } c p , r +
p,r q,s
X X  
pµ v q , s (x) d †p , r {d p , r , d †q , s } − {d †p , r , d †q , s } d p , r
p,r q,s
X h i
= − i ∂µ ψ (x) = pµ − c p , r u p , r (x) + d †p , r v p , r (x) (4.67)
p,r

if and only if the canonical anticommutation relations (4.60) hold true.

From Noether theorem and canonical anticommutation relations (4.60) it
follows that the classical vector current (4.53) is turned into the quantum
operator

J µ (x) ≡ : q ψ̄ (x) γ µ ψ (x) : (4.68)

which satisfies the continuity operator equation

1
∂µ J µ (x) = [ P µ , J µ (x) ] = 0 (4.69)
i~
As a consequence we have the conserved charge operator
Z X  
† † †
Q ≡ d x : q ψ (x) ψ (x) : = q cp,r cp,r − dp,r dp,r (4.70)
p,r

whence it is manifest that the two types of quanta of the Dirac field do
carry opposite charges. According to the customary convention for the Dirac
spinor describing the electron positron field, we shall associate to particles the
creation annihilation operators of the c-type and the negative electric charge
q = − e (e > 0), whilst the creation annihilation operators of the d-type and
the positive electric charge +e will be associated to the antiparticles, so that
the electric charge operator becomes
X  
† †
Qe = (− e) c p , r c p , r − d p , r d p , r (4.71)
p,r

Finally, it turns out that also the helicity (4.49) of the Dirac field, which
is a constant of motion, will be turned by the quantization procedure into

144
the normal ordered operator expression
1 ∞
Z
h = d z : ψ † (t, z) Σ 3 ψ (t, z) :
2 −∞
1 ∞
Z X h
= dp c †p , r c p , r u r† (p) Σ 3 u r (p)
2 −∞ r=1 , 2
i
− d †p , r d p , r v r† (p) Σ 3 v r (p) (2ω p ) −1 (4.72)

where
{c p , r , c †q , s } = δ rs δ (p − q) = {d p , r , d †q , s }
( ∀ r, s = 1, 2 p, q ∈ R )
and all other anticommutators vanish.
It is convenient to choose our standard spin-states, i.e. the orthogonal
and normalized solutions of eq.s (4.15) and (4.16), in which we have to put
p = (0, 0, p) . Then the helicity operator eventually becomes

1 ∞
Z h i
† † † †
h = dp cp,1 cp,1 − cp,2 cp,2 + dp,1 dp,1 − dp,2 dp,2
2 −∞
(4.73)

The above expression does actually clarify the meaning of the polarization
indices r, s, . . . = 1, 2 . In conclusion, from the expressions (4.65), (4.66),
(4.71) and (4.73), it follows that the operators c †p , r and c p , r do correspond
respectively to the creation and annihilation operators for the particles of
momentum p , mass M with p2 = M 2 , electric charge − e , and positive
helicity equal to 12 (r = 1) or negative helicity equal to − 12 (r = 2) .
Conversely, the operators d p† , r and d p , r will correspond respectively to the
creation and annihilation operators for the antiparticles of momentum p ,
mass M with p2 = M 2 , electric charge + e , and positive helicity equal to
1
2
(r = 1) or negative helicity equal to − 12 (r = 2) .

The cyclic Fock vacuum is defined by

cp,r | 0 i = 0 dp,r | 0 i = 0 ( ∀ r = 1, 2 p ∈ R3 ) (4.74)

while the 1-particle energy momentum, helicity and charge eigenstates will
correspond to

| p r − i ≡ c †p , r | 0 i ( r = 1, 2 p ∈ R3 ) (4.75)

145
whereas the 1-antiparticle energy momentum, helicity and charge eigenstates
will be

| p r + i ≡ d †p , r | 0 i ( r = 1, 2 p ∈ R3 ) (4.76)

Owing to the canonical anticommutation relations (4.60), it is impossible

to accomodate two particles or two antiparticles in the very same quantum
state as e.g.
c †p , r c †p , r | 0 i = − c †p , r c †p , r | 0 i = 0
d †p , r d †p , r | 0 i = − d †p , r d †p , r | 0 i = 0
As a consequence the multiparticle states do obey Fermi-Dirac statistics and
the occupation numbers solely take the two possible values

N p , r , ± = 0, 1 ( r = 1, 2 p ∈ R3 )

which drives to the Pauli exclusion principle valid for all identical particles
with half integer spin. The generic multiparticle state, that corresponds to
an element of the basis of the Fock space, will be written in the form
A Y
Y B
c † (pa , ra ) d † (pb , rb ) | 0 i (4.77)
a=1 b=1

By virtue of (4.60) those states are completely antisymmetric with regard

to the exchange of any pairs (pa , ra ) and (pb , rb ) and correspond to the
presence of A particles and B antiparticles.

4.4 Covariance of the Quantized Spinor Field

Consider the general structure of the Poincaré transformations generators as
normal ordered bilinear operator expressions, which can be readily obtained
from the classical expressions (4.40), (4.41) and (4.42) : namely,
Z
Pµ = d x : ψ † (x) i ∂ µ ψ (x) :
Z
M jk
= d x : x j ψ † (x) i ∂ k ψ (x) − x k ψ † (x) i ∂ j ψ (x) :
Z
+ d x : ψ † (x) σ jk ψ (x) :
Z
M 0k
= x P − d x x k : ψ † (x) i ∂0 ψ (x) :
0 k
(4.78)

146
It can be verified by direct inspection that, owing to the anticommutation
relations (4.60) and (4.64), those operator expressions indeed generate the
infinitesimal Poincaré transformations for the quantized Dirac spinor field :
namely,

δψ(x) = i [ P µ , ψ (x) ]  µ − 21 i [ Mρσ , ψ (x) ]  ρσ

=  ∂ µ + 12  µν ( x ν ∂ µ − x µ ∂ ν ) + 21 i σ jk  jk ψ (x) (4.79)
 µ 

Moreover it is straightforward albeit quite heavy and tedious to verify that,

by virtue of the canonical anticommutation relations, the operators (4.78) do
indeed fulfill the Lie algebra (1.37) of the Poincaré group – for the explicit
check see [21], 8.6, pp. 163 - 165. The covariant 1-particle states and the
corresponding creation annihilation operators can be defined in analogy with
the construction (3.66) and take the form
1
| p r − i = [ (2π)3 2 ω p ] 2 c †p , r | 0 i ≡ c †r ( p) | 0 i (4.80)
∀ p∈R ∀ r = 1, 2
1
| q s + i = [ (2π)3 2 ω q ] 2 d †q , s | 0 i ≡ d †s ( p) | 0 i (4.81)
∀ q∈R ∀ s = 1, 2

which satisfy

h ± s q | p r ± i = δ rs (2π) 3 2ω p δ (p − q) (4.82)

The normal mode decomposition of the Dirac field with respect to the new
set of covariant creation annihilation operators becomes
X Z h i
− ipx † ipx
ψ (x) = Dp c r ( p) u r ( p) e + d r ( p) v r ( p) e (4.83)
r=1,2

where p µ = (ω p , − p) together with

(p/ − M ) u r ( p) = 0 (p/ + M ) v r ( p) = 0 r = 1, 2 (4.84)

It is clear that the transformation law for the covariant 1-particle states
and corresponding creation annihilation operators will be determined by the
unitary operators associated to the Lorentz matrices. Actually, if we denote
as usual the Lorentz matrices by Λ(ω) = Λ(α, η) and by U (ω) = U (α, η) the
related unitary operators acting on the Fock space F , where (α, η) are the
canonical angular and rapidity coordinates of the restricted Lorentz group

147
L↑+ = O(1, 3)+
+ , then we can write

U (ω) | p r ± i = | Λ p r ± i (4.85)
U (ω) c r ( p) U † (ω) = c r (Λ p) (4.86)
U (ω) c †r ( p) U † (ω) = c †r (Λ p) (4.87)
U (ω) d r ( p) U † (ω) = d r (Λ p) (4.88)
U (ω) d r† ( p) U † (ω) = d r† (Λ p) (4.89)
∀ p∈R ∀ r = 1, 2

so that

h± s Λq|Λp r ±i =
δ rs (2π) 3 2ω p δ (p − q) = h ± s q | p r ± i (4.90)

As a consequence we can eventually write

ψ 0 (x) ≡ U (ω) ψ (x) U † (ω)

X Z h
= Dp U (ω) c r ( p) U † (ω) u r ( p) e− i p x
r=1,2
i
+ U (ω) d r† ( p) U † (ω) v r ( p) e i p x
X Z h i
= D p c r ( Λ p) u r ( p) e− i p x + d r† (Λ p) v r ( p) e i p x
r=1,2
X Z h 0x0
= Dp0 c r ( p 0 ) u r (Λ−1 p 0 ) e− i p
r=1,2
0 0
i
+ d r† ( p 0 ) v r (Λ−1 p 0 ) e i p x
X Z h
− ip 0 x 0
= D p 0 c r ( p 0 ) Λ−1 0
1 ur ( p ) e
2
r=1,2
i
ip 0 x 0
+ d r† ( p 0 ) Λ−1
1
0
vr ( p ) e = Λ−1 0
1 ψ (x ) (4.91)
2 2

where the transformation law (4.23) of the spin-states has been used. In
addition, taking into account formulæ (4.67) and (4.78), from the canonical
anticommutation relations (4.64) we eventually obtain for a quite general
Poincaré transformation

ψ 0 (x) = U (a, ω) ψ (x) U † (a, ω)

= Λ−1
1 ( ω) ψ (Λ x + a) (4.92)
2

U (a, ω) ≡ exp i a µ Pµ − 21 i ω ρσ Mρσ


(4.93)

148
which shows that the general transformation law of the quantized spinor field
consists in an infinite dimensional irreducible unitary representation of the
Poincaré group acting on the Fock space F . In particular, for an infinitesimal
Poincaré transformation we find
ψ (x) + δ ψ (x) = ψ (x) + i [ P µ , ψ (x) ]  µ − 12 i [ Mρσ , ψ (x) ]  ρσ
= ψ (x) +  µ ∂ µ ψ (x) +  ρσ x σ ∂ρ ψ (x)
+ 12 i  jk σ jk ψ (x)
where δ a µ =  µ and δ ω ρσ =  ρσ are the infinitesimal parameters. In
particular, from the finite transformation rule (2.80), it is simple to check,
for instance, that the mass operator is Lorentz invariant, i.e.
ψ̄ 0 (x) ψ 0 (x) = ψ̄(x 0 ) ψ(x 0 )
The reader should gather the difference between the transformation law
(4.92) for the quantized Dirac spinor field and the transformation law (2.77)
of the classical relativistic Dirac spinor wave field under restricted Lorentz
group, i.e.
ψ 0 (x 0 ) = Λ 1 (ω) ψ (x) = exp − 12 i σ µν ω µν ψ (x) x0 = Λ x

2

which is unitary with respect to the inner product

Z
( ψ2 , ψ1 ) = dx ψ2† (t, x) ψ1 (t, x) (4.94)

Notice that the density

% (x) = ψ † (t, x) ψ(t, x)
does represent the positive semidefinite probability density in the old Dirac
theory, i.e. the relativistic quantum mechanics of the electron. Conversely,
the corresponding operator valued local density
%b(x) = q : ψ † (t, x) ψ(t, x) :
has not a definite sign for it represents the charge density of the quantized
spinor field, see equation (4.70).
Beside the transformation laws of the quantized Dirac field under the
continuous Poincarè group, a very important role in the Standard Model
of the fundamental interactions for Particle Physics is played by the charge
conjugation, parity and time reversal discrete transformations, the so called
CPT symmetries, on the quantized spinor matter fields. We shall analyze in
details the latter ones at the end of the present chapter.

149
4.5 Special Distributions
From the canonical anticommutation relations (4.60) and the normal mode
expansion (4.28) of the Dirac field, the so called canonical anticommutator at
arbitrary points between two free Dirac spinor field operators can be readily
shown to be equal to zero
{ψα (x) , ψβ (y)} = 0 = {ψ̄α (x) , ψ̄β (y)} ( α, β = 1L, 2L, 1R, 2R )
On the contrary, the canonical anticommutator at arbitrary points between
the free Dirac field and its adjoint does not vanish: it can be easily calculated
to be
{ψα (x) , ψ̄β (y)} ≡ Sαβ (x − y) = − i ( i ∂/x + M )αβ D (x − y) (4.95)
where D (x − y) is the Pauli-Jordan distribution of the real scalar field.
Proof. From the normal modes expansions (4.62) of the spinor fields
X − 21
ψα (x) = (2π)3 2ω p
p,r
h i
× c p , r u α , r (p) e− i p x + d †p , r v α , r (p) e i p x
p0 = ω p
X
3
− 21
ψ̄β (y) = (2π) 2ω q
q,s
h i
× c †q , s ū β , s (q) e i q y + d q , s v̄ β , s (q) e− i q y
q0 = ω q

and the canonical anticommutation relations (4.60) one finds

Z X h
{ψα (x) , ψ̄β (y)} = Dp u α , r (p) ū β , r (p) e− i p (x−y)
r = 1,2
i
+ v α , r (p) v̄ β , r (p) e i p (x−y)
p0 = ω p

Taking into account the sums over the spin states (4.35)
X  u α , r (p) ū β , r (p) = ( p/ + M ) αβ
( p0 = ω p )
v α , r (p) v̄ β , r (p) = ( p/ − M ) αβ
r = 1,2

we readily come to the expression, by omitting the spinorial indices for the sake of brevity,
{ψ (x) , ψ̄ (y)} ≡ S (x − y)
 
i p (x−y) M − p/
Z
dp p/ + M − i p (x−y)
= e −e
(2π)3 2ωp 2ωp p0 = ω p
Z  
= (i ∂/x + M ) D p e− i p (x−y) − e i p (x−y)
p0 = ω p

d 4 p − i p (x−y)
Z
= (i ∂/x + M ) e δ (p2 − M 2 ) sgn (p0 )
(2π)3
= − i (i ∂/x + M ) D (x − y) (4.96)

150
where use has been made of the formulæ (4.34). Quod Erat Demonstrandum

The canonical anticommutator at arbitrary points (4.96) is a solution of

the Dirac equation which does not vanish when (x − y) is a spacelike interval
with x0 6= y 0 . In fact, at variance with the real scalar field case in which

D (x − y) ≡ 0 ∀ (x0 − y0 )2 < (x − y)2 (x0 6= y 0 ) (4.97)

we find instead the nonvanishing equal time anticommutator

S (0, x − y) = γ 0 δ (x − y) (4.98)

in agreement with (4.64).

The causal Green’s function or Feynman propagator for the Dirac field is

h 0 | ψα (x) ψ̄β (y) | 0 i for x0 > y 0
h 0 | T ψα (x) ψ̄β (y) | 0 i =
− h 0 | ψ̄β (y) ψα (x) | 0 i for x0 < y 0
c F
≡ i Sαβ (x − y) = S αβ (x − y)

Actually we have

S F (x − y) = (i ∂/x + M ) DF (x − y) (4.99)

where DF (x − y) is the Feynman propagator of the real scalar field.

Proof. From the very definition as the vacuum expectation value of the chronological
product of spinor fields we can write

S F (x − y) = θ (x0 − y 0 ) ψ (x) ψ̄ (y) 0 − θ (y 0 − x0 ) ψ̄ (y) ψ (x) 0



X
= θ (x0 − y 0 ) [ (2π) 3 2ω p ] −1 u r (p) ⊗ ū r (p)
p,r

× exp {− i ω p (x0 − y 0 ) + i p · (x − y)}

X
− θ (y 0 − x0 ) [ (2π) 3 2ω p ] −1 v r (p) ⊗ v̄ r (p)
p,r

× exp {+ i ω p (x0 − y 0 ) − i p · (x − y)} (4.100)

Hence, from the sums over the spin-states (4.34) we obtain

Z h i
S (x − y) = θ (x − y ) Dp (M + p/) e− i p (x−y)
F 0 0
p0 = ω p
Z h i
+ θ (y 0 − x0 ) Dp (M − p/) e i p (x−y)
p0 = ω p
Z k
= θ (x0 − y 0 ) (i ∂/x + M ) Dp e− i p (x−y)
p0 = ω p
Z k
0 0 i p (x−y)
+ θ (y − x ) (i ∂/x + M ) Dp e
p0 = ω p

151
and if we recall the definitions
Z k
± i D(±) (x − y) = Dp e± i p (x−y)
p0 = ω p
Z
dp
= exp{±i ω p (x0 − y 0 ) ∓ i p · (x − y)}
(2π)3 2ω p
we readily come to the expessions
1 (−)
S F (x − y) = θ (x0 − y 0 ) (i ∂/x + M ) D (x − y)
i
+ i θ (y 0 − x0 ) (i ∂/x + M ) D (+) (x − y)
= (i ∂/x + M ) i θ y 0 − x0 D (+) (x − y) − (i ∂/x + M ) i θ x0 − y 0 D (−) (x − y)




h (−) i
− γ 0 δ x0 − y 0 D (x − y) + D (+) (x − y) = (i ∂/x + M ) DF (x − y)

where I did make use of the Pauli-Jordan distribution property

D (x) = D (+) (x) + D (−) (x) , D (0, x) = 0

and of the relation (3.89)

DF (x − y) = i θ y 0 − x0 D (+) (x − y) − i θ x0 − y 0 D (−) (x − y)

Quod Erat Demonstrandum

The Fourier representation of the spinorial Feynman propagator reads

S F (x − y) = (i ∂/x + M ) D F (x − y)
Z
i p/ + M
= 4
dp 2 exp{−ip · (x − y)}
(2π) p − M 2 + iε
Z
1 i
= 4
dp exp{−ip · (x − y)} (4.101)
(2π) p/ − M + iε
and consequently

F
(i ∂/x − M ) S αβ (x − y) = i δ(x − y) (4.102)

 
i ( p/ + M ) αβ def  i
=  (4.103)


2 2
p − M + iε p/ − M αβ
 

It is also convenient to write the adjoint form of the inhomogeneous equation

for the Feynman propagator of the spinor field. To this concern, let us first
obtain the hermitean conjugate of equation (4.102) viz.,

i δ (x − y) = i (∂/∂x µ ) S F† (x − y) γ µ † + M S F† (x − y)
= i (∂/∂x µ ) S F† (x − y) γ 0 γ µ γ 0 + M S F† (x − y) (4.104)

152
Multiplication by γ 0 from left and right yields

i δ (x − y) = i γ 0 (∂/∂x µ ) S F† (x − y) γ 0 γ µ + γ 0 M S F† (x − y) γ 0
←
def
= S̄ F ( y − x) ( i ∂/x + M ) (4.105)

where

S̄ F ( y − x) = γ 0 S F† (x − y) γ 0
−i
Z
p/ + M
= 4
dp 2 exp{− i p · (y − x)} (4.106)
(2π) p − M 2 − iε

4.5.1 The Euclidean Fermions

Also for the present fermion propagator we can safely perform the Wick
rotation i p0 = p4 , ix0 = x4 that yields
Z Z ∞
F −4
S αβ ( −i x4 , x) = i (2π) dp d p4
−∞
exp {i p4 x4 + i p · x}
( γ 0 p4 − i γ k pk + i M ) αβ
p24 + p2 + M 2
It follows that if we define the hermitean euclidean gamma matrices
−
− −  − −
γ µ= γ k , γ 4 γ4 ≡ γ0 γ k ≡ − i γ k ( k = 1, 2, 3 ) (4.107)

−†
−
n− − o
γµ = γµ γ µ , γ ν = 2 δ µν (4.108)

or even more explicitly

0 0 0 −i 
 
  
−i −i
 
− 0 σ 1

 0 0 0 

γ1 =   =  (4.109)
 
  

i σ1 0 0 i 0 0
 
 


 

i 0 0 0
0 0 0 −1 
 
  
0 −i σ2 
 
−

 0 0 1 0  
γ2 =   =  (4.110)

  
 

i σ2 0 
 0 1 0 0  

−1 0 0 0
 

0 0 −i 0 
 
  
0 −i σ3 
 
−

 0 0 0 i  
γ3 =   =  (4.111)

   

i σ3 0 

 i 0 0 0 


0 −i 0 0
 

153
  
   0 0 1 0 
 
−  0 1   0 0 0
 1 

γ4 =   = 
  
 (4.112)
1 0 1 0 0 0
 
 


 

0 1 0 0

Then we can write

exp {i pE · xE }
Z
F i −
S αβ ( −i x4 , x) = d pE ( γ µ pE µ + i M ) αβ
(2π) 4 p2E + M 2
Z  
i 1
= d pE exp {i pE · xE }

 

(2π)4 p/E − i M αβ
 

= ( ∂/E − M ) αβ DE (xE )
E
≡ − S αβ ( xE ) (4.113)

where we understand
− − i∂
γ µ pE µ ≡ p/E i ∂/E ≡ γ µ
∂ xE µ

This suggests that the proper classical variables for the setting up of an
euclidean formulation of the Dirac spinor field theory, in analogy with what
we have already seen in the scalar field case, should be two euclidean bispinors
ψE and ψ̄E obeying

{ψE (x) , ψE (y)} = {ψ̄E (x) , ψ̄E (y)} = {ψE (x) , ψ̄E (y)} = 0 (4.114)

for all points x and y of the four dimensional euclidean space R4 .

The last of these relations is crucial, for it implies that ψ̄E does not
necessarily coincide with the adjoint of ψ times some matrix γ4 . Thus, if
we want to set up a meaningful euclidean formulation for the Dirac spinor
field theory, then we can treat ψE and ψ̄E as totally independent classical
Grassmann valued variables. This independence is the main novelty of the
euclidean fermion field theory; the rest of the construction is straightforward.
For instance, we use the definition of the hermitean euclidean gamma
matrices to derive the O(4) transformation law for ψE in the usual way3 ,
−
while define ψ̄E to transform like the transposed of ψE . Next, we define γ 5 ,
a hermitean matrix, by
− − − − − −†
γ 5 = γ 1 γ 2 γ 3 γ 4 = γ 5 = − γ5
3
Remember that the orthogonal group O(4) of the rotations in the euclidean space R4
is a semisimple Lie group O(4) = O(3)L × O(3)R .

154
 − −
Thus, ψ̄E ψE is a scalar, ψ̄E γ 5 ψE a pseudoscalar, ψ̄E γ µ ψE a vector etc.

The euclidean action for the free Dirac field is given by

Z
SE [ ψE , ψ̄E ] = d 4 xE ψ̄E (xE ) ( ∂/E + M ) ψE (xE ) (4.115)

Here the overall sign is purely conventional : we could always absorb it into
ψE if we wanted to (remember that we are free to change ψE without touching
ψ̄E ). Conversely, the lack of the factor i in front of the derivative term is
not at all conventional : it is there just to ensure that the euclidean fermion
propagator, which is named 2-point Schwinger’s function, is proportional to
(i p/E − M )/(p2E + M 2 ) ; if it were not for this i , then we would have tachyon
poles after a Wick rotation back to the Minkowski momentum space.
It is worthwhile to notice that the above Dirac euclidean action can be
obtained from the corresponding one in the Minkowski spacetime, after the
customary standard replacements
− −
x4 = i x 0 γk ≡ − i γk ( k = 1, 2, 3 ) γ4 ≡ γ0
ψ (x) 7→ ψE (xE ) ψ̄ (x) 7→ ψ̄E (xE )
As a matter of fact we readily obtain

S [ ψ, ψ̄ ] 7→ i SE [ ψE , ψ̄E ]
Z
= d 4 xE ψ̄E (xE ) ( i∂/E + iM ) ψE (xE ) (4.116)

Furthermore, the euclidean Dirac operator ( ∂/E + M ) is precisely that

one which gives, according to the definition (4.113), the 2-point Schwinger
function inversion formula4

( ∂/E + M ) α β S βEη ( xE ) = δ (xE ) δ α η

Finally, it is worthwhile to remark that the euclidean action for the free
Dirac field is not a real quantity. As we shall see, this fact will not cause any
troubles in the analytic continuation to the Minkowski spacetime.
4
The n−point Green functions in the euclidean space are usually named the n−point
Schwinger functions.

155
 To end up we have
Z
SE0 [ φE ∂ φ ∂ φ + 12 m 2 φE2
1 
] = d xE 2 µ E µ E
Z
1
φE (xE ) − ∂ 2 + m 2 φE (xE )


=
˙ d xE 2
(4.117)
Z
SE0 [ ψ̄ , ψE ] = d xE ψ̄ (xE ) ( ∂/E + M ) ψE (xE ) (4.118)

in such a manner that we can summarize the useful relationships

DF (− i x4 , x) → − DE (xE ) (4.119)
exp { i kE · xE }
Z
1
DE (xE ) = d kE (4.120)
(2π) 4 kE2 + m 2
− ∂E2 + m 2 DE (xE ) = δ (xE )


(4.121)
F
S αβ ( −i x4 , x) → − S Eαβ (xE ) (4.122)
Z  
d pE i
S Eαβ (xE ) = exp { i pE · xE } (4.123)

 

(2π) 4 − p/E + iM
 
αβ
( ∂/E + M ) αβ S Eβη (xE ) = δ (xE ) δ αη (4.124)

156
4.6 The Fermion Generating Functional
4.6.1 Symanzik Equations for Fermions
Consider the vacuum expectation value
 Z h i
Z 0 [ ζ , ζ̄ ] ≡ T exp i d x ζ̄(x) ψ (x) + ψ̄ (x) ζ(x) (4.125)
0
∞
in
X Z Z Z Z Z Z
= d x1 d x2 · · · d xn d y 1 d y 2 · · · d y n
n=0
n!
D    E
T ζ̄(x1 ) ψ (x1 ) + ψ̄ (y1 ) ζ(y1 ) · · · ζ̄(xn ) ψ (xn ) + ψ̄ (yn ) ζ(yn )
0

the suffix zero denoting the free field theory, where ζ(x) and ζ̄(x) are the
so called classical fermion external sources, which turn out to be Grassmann
valued functions with the canonical dimensions [ ζ ] = cm − 5/2 and which
satisfy

{ζ(x) , ζ(y)} = {ζ̄(x) , ζ̄(y)} = 0 (4.126)

{ζ(x) , ζ̄(y)} = {ζ̄(x) , ζ(y)} = 0 (4.127)
{ζ(x) , ψ (y)} = {ζ̄(x) , ψ̄ (y)} = 0 (4.128)
{ζ(x) , ψ̄ (y)} = {ζ̄(x) , ψ (y)} = 0 (4.129)

Notice that, just owing to (4.129), the products ζ̄(x) ψ (x) and ψ̄ (y) ζ(y) do
not take up signs under time ordering, i.e.
   ζ̄(x) ψ (x) ψ̄ (y) ζ(y) if x0 > y 0
T ζ̄(x) ψ (x) ψ̄ (y) ζ(y) =
ψ̄ (y) ζ(y) ζ̄(x) ψ (x) if x0 < y 0

In fact

ζ̄(x) ψ (x) ψ̄ (y) ζ(y) = ζ̄(x) ζ(y) ψ (x) ψ̄ (y) (x0 > y 0 )
(x0 < y 0 )


ψ̄ (y) ζ(y) ζ̄(x) ψ (x) = ζ̄(x) ζ(y) − ψ̄ (y) ψ (x)

so that
D  E

T ζ̄(x) ψ (x) ψ̄ (y) ζ(y) = ζ̄(x) ζ(y) T ψ (x) ψ̄ (y) 0
0
= ζ̄(x) ζ(y) SF (x − y)
D  E
= T ψ̄ (y) ζ(y) ζ̄(x) ψ (x)
0

157
 The functional differentiation with respect to the classical Grassmann
valued sources is defined by
{δ/δ ζ̄(x) , ζ̄(y)} = δ(x − y) = {δ/δζ(x) , ζ(y)}
{δ/δ ζ̄(x) , ζ(y)} = 0 = {δ/δζ(x) , ζ̄(y)}
{δ/δ ζ̄(x) , δ/δ ζ̄(y)} = 0 = {δ/δζ(x) , δ/δζ(y)}
{δ/δ ζ̄(x) , δ/δζ(y)} = 0 = {δ/δζ(x) , δ/δ ζ̄(y)} (4.130)
where all operators act on their right. It follows that
− i δZ 0 [ ζ , ζ̄ ]/δ ζ̄(x) =
  Z h i
T ψ (x) exp i d y ζ̄(y) ψ (y) + ψ̄ (y) ζ(y)
0
i δZ 0 [ ζ , ζ̄ ]/δζ(x) =
 Z h i
T ψ̄ (x) exp i d y ζ̄(y) ψ (y) + ψ̄ (y) ζ(y) (4.131)
0
where the plus sign in the second equality is because
Z Z
δ/δζ(x) d y ψ̄ (y) ζ(y) = d y δ/δζ(x)[ ψ̄ (y) ζ(y) ]
Z
= d y ψ̄ (y)[ − δζ(y)/δζ(x) ]
Z
= − d y ψ̄ (y) δ(x − y)

= − ψ̄ (x) (4.132)
Taking one more functional derivatives of the generating functional (4.125)
we find
(δ/δ ζ̄(x)) δZ 0 [ ζ , ζ̄ ] / δζ (y) = (4.133)
 Z h i
T ψ (x) ψ̄ (y) exp i d y ζ̄(y) ψ (y) + ψ̄ (y) ζ(y)
0
The vacuum expectation values of the chronological ordered products of n
pairs of free spinor field and its adjoint operators at different spacetime points
are named the n−point fermion Green’s functions of the (free) Dirac spinor
quantum field theory. By construction, the latter ones can be expressed as
functional derivatives of the generating functional : namely,

( n) def
S 0 ( x1 , · · · , xn ; y1 , · · · , yn ) =
h 0 | T ψ (x1 ) ψ̄ (y1 ) · · · ψ (xn ) ψ̄ (yn ) | 0 i
= δ (2n) Z 0 [ ζ , ζ̄ ] / δ ζ̄(x1 ) δζ(y1 ) · · · δ ζ̄(xn ) δζ(yn ) ζ = ζ̄ = 0


158
 Now it should be evident that the very same steps, which have led to
establish the functional equation (3.101) for the free scalar field generating
functional, can be repeated in a straightforward manner. To this purpose,
let me introduce the finite times chronologically ordered exponential operator
for the spinor fields that reads
( Z 0 )
Z t
E(t 0 , t ) ≡ T exp i d y0 d y [ ψ̄(y) ζ(y) + ζ̄(y) ψ(y) ] (4.134)
t

in such a manner that we can write

 
( i ∂/x − M ) − i δZ 0 [ ζ , ζ̄ ] / δ ζ̄ (x) =
Z h
0
iγ d y h 0 | E (∞, x0 ) ψ (x) ψ̄ (x0 , y) ζ (x0 , y)
i
− ψ̄ (x0 , y) ζ (x0 , y) ψ (x) E (x0 , −∞) | 0 i
Z
0
= iγ d y h 0 | E (∞, x0 ) {ψ (x) , ψ̄ (x0 , y)} ζ (x0 , y) E (x0 , −∞) | 0 i

= − ζ (x) Z 0 [ ζ , ζ̄ ]

hence
h i
( i ∂/x − M ) i δ/ δ ζ̄ (x) − ζ (x) Z 0 [ ζ , ζ̄ ] = 0 (4.135)

In a complete analogous way we find the functional differential formula

h  ←  i
i δ/ δ ζ (x) i ∂/x + M + ζ̄ (x) Z 0 [ ζ , ζ̄ ] = 0 (4.136)

Proof. First we obtain

− i δZ 0 [ ζ , ζ̄ ] / δ ζ (x) =
h 0 | E (∞, x0 ) ψ̄(x0 , x) E (x0 , −∞) | 0 i

and taking left time derivative

←
∂
h 0 | E (∞, x0 ) ψ̄(x0 , x) E (x0 , −∞) | 0 i 0
Z ∂x
= i d y h 0 | E (∞, x0 ) ψ̄ (x) [ ψ̄ (x0 , y) ζ (x0 , y) + ζ̄ (x0 , y) ψ (x0 , y) ]E (x0 , −∞) | 0 i
Z
− i d y h 0 | E (∞, x0 )[ ψ̄ (x0 , y) ζ (x0 , y) + ζ̄ (x0 , y) ψ (x0 , y) ] ψ̄ (x) E (x0 , −∞) | 0 i
 Z h i
+ T ∂0 ψ̄ (x) exp i d y ζ̄(y) ψ (y) + ψ̄ (y) ζ(y)
0

159
Now, using the anticommutation relations (4.128), (4.129) and the equal time canonical
anticommutation relations (4.60) we find
←
h 0 | E (∞, x0 ) ψ̄(x0 , x) E (x0 , −∞) | 0 i ( i ∂/x + M )
Z
= γ 0 d y h 0 | E (∞, x0 ) ζ̄ (x0 , y){ψ̄ (x0 , x) , ψ (x0 , y)}E (x0 , −∞) | 0 i
Z
− γ 0 d y h 0 | E (∞, x0 ){ψ̄ (x0 , x) , ψ̄ (x0 , y)} ζ (x0 , y) E (x0 , −∞) | 0 i
←
 Z h i
+ T ψ̄ (x)( i ∂/x + M ) exp i d y ζ̄(y) ψ (y) + ψ̄ (y) ζ(y)
0
Z
= d y h 0 | E (∞, x0 ) ζ̄ (x0 , y) δ(x − y) E (x0 , −∞) | 0 i = ζ̄ (x) Z 0 [ ζ , ζ̄ ]

←
where use has been made of the adjoint Dirac equation ψ̄(x)( i ∂/x + M ) = 0 . This proves
the Symanzik equation (4.136). Quod Erat Demonstrandum

The solution of the above couple of functional differential equations that

satisfies causality is
 Z Z 
Z 0 [ ζ , ζ̄ ] = exp − d x d y ζ̄ (x) SF (x − y) ζ (y) (4.137)

It is worthwhile to remark that, in close analogy with the case of the free
scalar field generating functional, the integral kernel − SF (x − y) which
appears in the exponent of the right hand side of eq. (4.137) does exactly
coincide with the inverse of the Dirac operator (i ∂/ − M ) that specify the
classical spinor action , since (M − i ∂/x ) SF (x−y) = δ (x − y) . The inversion
of the classical kinetic Dirac operator is precisely provided by the Feynman
propagator or causal 2-point Green’s function, which indeed encodes :

1. the relativistic covariance properties under the action of the Poincaré

group

2. the canonical anticommutation relations (4.64), the related property

(4.97) of the microcausality and the Fermi-Dirac statistics for the multi-
particle states

3. the causality requirement, that means the possibility to perform the

Wick rotation and turn smoothly to the euclidean formulation.

Hence the generating functional (4.137) does truly contain all the mutually
tied up key features of the relativistic quantum field theory.

160
4.6.2 The Integration over Grassmann Variables
In order to find a functional integral representation for Z 0 [ ζ , ζ̄ ] I need to
primary define the concept of integration with respect to Grassmann valued
functions on the Minkowski spacetime. This latter tool has been long ago
constructed by
F.A. Berezin, The Method of Second Quantization
Academic Press, New York (1966)
Hereafter I shall follow
Sidney R. Coleman
The Uses of Instantons
Proceedings of the 1977 International School of Subnuclear Physics
Erice, Antonino Zichichi Editor, Academic Press, New York (1979)
Consider first a real function Rf : G → R of a Grassmann variable a with
a2 = 0 : we want to define da f (a) . We require this to have the usual
linearity property :
Z
da [ αf (a) + β g (a) ]
Z Z
= α da f (a) + β da g (a) ( ∀ α, β ∈ R ) (4.138)

In addition, we would like the integral to be translation invariant

Z Z
da f (a + b) = da f (a) (∀ b ∈ G) (4.139)

It is easy to show that those conditions determine the integral, up to a

normalization constant.
The reason is that there are only two linearly independent functions of
a ∈ G , that is 1 and a , any higher powers being vanishing. As a matter of
fact we have ∀ a, b ∈ G with {a, a} = {a, b} = {b, b} = 0
f (a) = f (0) + f 0 (0) a f (a + b) = f (0) + f 0 (0) (a + b) (4.140)
whence from (4.138)
Z Z Z
da f (a) = da f (0) + da f 0 (0) a
Z Z
0
= f (0) da 1 + f (0) da a
Z Z Z
da f (a + b) = da f (0) + da f 0 (0) (a + b)
Z Z Z
0 0
= f (0) da 1 + f (0) da a + b f (0) da 1

161
and from the translation invariance requirement (4.139)
Z Z
0
da a = N b f (0) da 1 ≡ 0 (∀b ∈ G) (4.141)

One can always choose the normalization constant N such that

Z
da a = 1

But then, translation invariance just requires

Z
da 1 = 0

Hence Z
da f (a) = f 0 (0) ∀f : G → R

As a straightforward generalization for any many Grassmann variables

function, it is natural to define multiple integrals by iteration. Thus, e.g. , a
complete integration table for the four linearly independent functions of two
anticommuting variables a and ā is provided by

Z Z 
 ā a = 1
ā = 0

da d ā (4.142)

 a = 0
1 = 0


with {a, a} = {ā, ā} = {a, ā} = 0 . As an application of this table we can
calculate
Z Z Z Z
da d ā exp { λ ā a} = da d ā (1 + λ ā a) = λ (4.143)

The generalization to a multi-dimensional space is straightforward. Consider

some n × n hermitean matrix A = A† and two sets of n Grassmann variables

{θi , θj } = {θi , θ̄j } = {θ̄i , θ̄j } = 0 ( i, j = 1, 2, . . . , n )

Then a unitary n × n matrix always exists such that

U † AU = diag (λ1 , λ2 , . . . , λn ) λj ∈ R ( j = 1, 2, . . . , n )

As a consequence, if we set

a=Uθ ā = θ̄ U †

162
so that
{ai , aj } = {ai , āj } = {āi , āj } = 0 ( i, j = 1, 2, . . . , n ) (4.144)
then we can write
Z Z
I = d ā exp {ā A a}
da
Z Z
= det (U U ) dθ dθ̄ exp {θ̄ U † A U θ}
†

Yn Z Z
= dθj d θ̄j exp { λj θ̄j θj }
j=1
Yn Z Z


= dθj d θ̄j 1 + λj θ̄j θj
j=1
Yn
= λj = det A (4.145)
j=1

As a further application, let us consider the Taylor expansion for a real

function of the Grassmann variables (ā, a) satisfying (4.144): namely,
n   n  
X ∂f X ∂f
f (ā, a) = f (0, 0) + ā + aı
=1
∂ā  0 ı=1
∂aı 0
n X n  2 
1
X ∂ f
+ 2 ā aı
ı=1 =1
∂aı ∂ā 0

where   
∂f ∂
≡ f (ā, a) et cetera
∂ā 0 ∂ā ā=a=0
Then we have e.g.
Z Z
∂
da d ā f (ā, a) =
∂ā
  Z Z  2  Z Z
∂f 1 ∂ f
da d ā − da a d ā ≡ 0
∂ā 0 2 ∂a ∂ā 0
and in general
Z Z
da d ā [ ∂f (ā, a)/∂ā ] = 0 ( ∀  = 1, 2, . . . , n ) (4.146)
Z Z
da d ā [ ∂f (ā, a)/∂a ı ] = 0 ( ∀ ı = 1, 2, . . . , n ) (4.147)

163
4.6.3 The Functional Integrals for Fermions
Consider now the straightforward generalization of the functional integral
representation (3.105) to the fermion case : namely,

Z Z
Z 0 [ ζ , ζ̄ ] = Dψ Dψ̄ Ze0 [ ψ̄ , ψ ]
Z h i
× exp i d x ζ̄(x) ψ (x) + ψ̄ (x) ζ(x) (4.148)

where we understand once again formally

Z Z Y Z Z
Dψ Dψ̄ :=: d ψ x d ψ̄ x
x∈M

Inserting into (4.135) and (4.136) and making use of (4.146) and (4.147) we
are enabled to identify

Ze0 [ ψ̄ , ψ ] = N exp i S 0 [ ψ̄ , ψ ] (4.149)
and by comparison with (4.137) we find
 Z Z 
Z 0 [ ζ , ζ̄ ] = exp − d x d y ζ̄ (x) SF (x − y) ζ (y)
Z Z

:=: N Dψ Dψ̄ exp i S 0 [ ψ̄ , ψ ]
Z h i
× exp i d x ζ̄(x) ψ (x) + ψ̄ (x) ζ(x) (4.150)
Z
S 0 [ ψ̄ , ψ ] = d x ψ̄ (x) (i ∂/ − M ) ψ (x)

As I did in the case of the scalar field, the functional measure Dψ Dψ̄ ,
which is a formal entity until now, can be implemented by the requirement
of invariance under field translations ψ(x) 7→ ψ(x) + θ(x) , ψ̄(x) 7→ ψ̄(x) +
θ̄(x) . Once it is assumed, after the change of variables
ψ (x) 7→ ψ 0 (x) = ψ (x) + (i ∂/ − M )−1 ζ (x)
R
= ψ (x) − i dy SF (x − y) ζ (y)
= ψ x − h i S xy ζ y i (4.151)
←
ψ̄ (x) 7→ ψ̄ 0 (x) = ψ̄ (x) + ζ̄ (x)( i ∂/ +M )−1
R
= ψ̄ (x) − i dy ζ̄ (y) S̄F (y − x)
= ψ̄ x − h i ζ̄ y S̄ yx i (4.152)

164
where use has been made of the equation (4.105). Then we readily obtain
 Z Z 
−1
Z 0 [ ζ , ζ̄ ] :=: Z 0 [ 0 , 0 ] exp − i d x d y ζ̄ (x) (i ∂/ − M ) ζ (y)
 Z Z 
= Z 0 [ 0 , 0 ] exp − d x d y ζ̄ (x) SF (x − y) ζ (y)

so that consistency requires

Z Z

Z 0 [ 0 , 0 ] = h 0 | 0 i :=: N Dψ Dψ̄ exp i S 0 [ ψ̄ , ψ ] = 1

Proof : the formal expression (4.150) can be suitably rewritten as

Y R R 
Z 0 [ ζ , ζ̄ ] :=: N d ψ x d ψ̄ x exp i ψ̄ x (i ∂/ x − M ) ψ x + i ψ̄ x ζ x + i ζ̄ x ψ x
x∈M

Then, once again, it is very convenient to perform the change of variables

ψ x = ψ x0 + h i S xy ζ y i d ψ x = d ψ x0

ψ̄ x = ψ̄ x0 + h i ζ̄ y S̄ yx i d ψ̄ x = d ψ̄ x0
in such a manner that we have
R R 
d ψ x d ψ̄ x exp i ψ̄ x (i ∂/ x − M ) ψ x + i ψ̄ x ζ x + i ζ̄ x ψ x
d ψ x d ψ̄ x exp i ψ̄ x0 + h i ζ̄ y S̄ yx i (i ∂/ x − M ) (ψ x0 + h i S xz ζ z i)
R R 

=
× exp i ψ̄ x0 ζ x + i ζ̄ x ψ x0 − h ζ̄ y S̄ yx i ζ x − ζ̄ x h S xy ζ y i


and from the first equality

ψ̄ x0 (i ∂/ x − M ) h i S xz ζ z i = − ψ̄ x0 ζ x

together with the second conditioned equality

h i ζ̄ y S̄ yx i (i ∂/ x − M ) ψ x0
←
= − ( ∂/∂x µ ) h ζ̄ y S̄ yx i γ µ ψ x0 − h i ζ̄ y S̄ yx i ( i ∂/ x +M ) ψ x0
 
←
=
˙ − h i ζ̄ y S̄ yx i ( i ∂/ x +M ) ψ x0 = − ζ̄ x ψ x0

which is true by neglecting the four divergence term, we can eventually write

exp i ψ̄ x0 + h i ζ̄ y S̄ yx i (i ∂/ x − M ) (ψ x0 + h i S xz ζ z i)


˙ exp i ψ̄ x0 ( i ∂/ x − M ) ψ x0 − i ψ̄ x0 ζ x − i ζ̄ x ψ x0 + h ζ̄ y S̄ yx i ζ x

=

Hence, collecting altogether we come to the double integral

R R 
d ψ x d ψ̄ x exp i ψ̄ x (i ∂/ x − M ) ψ x + i ψ̄ x ζ x + i ζ̄ x ψ x
d ψ x0 d ψ̄ x0 exp i ψ̄ x0 (i ∂/ x − M ) ψ x0
 R R 
=˙ exp − ζ̄ x h S xy ζ y i ( ∀x ∈ M )

165
and thereby
Y R R 
N dψx d ψ̄ x exp i ψ̄ x (i ∂/ x − M ) ψ x + i ψ̄ x ζ x + i ζ̄ x ψ x
x∈M
Y Y R
d ψ z0 d ψ̄ z0 exp i ψ̄ z0 (i ∂/ z − M ) ψ z0
 R 
=
˙ exp − ζ̄ x h S xy ζ y i N
x∈M z ∈M
Y Z Z
Dψ 0 Dψ̄ 0 exp i S 0 [ ψ̄ 0 , ψ 0 ]
 
= exp − ζ̄ x h S xy ζ y i N
x∈M
 R R
= exp − d x d y ζ̄(x) SF (x − y) ζ(y) Z 0 [ 0 , 0 ] = Z 0 [ ζ , ζ̄ ]

Quod Erat Demonstrandum

To the aim of giving some precise mathematical meaning to the above

expression, it is convenient to turn to the euclidean formulation just like I
did in the case of the real scalar field – see formula (3.113). Then, taking the
Dirac euclidean action (4.115) into account, we can consider
Z Z
(0)
ZE [ 0 , 0 ] = N DψE Dψ̄E
 Z 
× exp i d xE ψ̄E (i ∂/E + iM ) ψE

After rescaling
ψE0 , x = i µ ψE , x
where µ is some arbitrary mass scale, the we can write
(0)
ZE [ 0 , 0 ] =
Y Z Z
0 0
d ψE x d ψ̄E0 x exp ψ̄E0 x (i ∂/E x + iM ) µ−1 ψE0 x

N
xE ∈ R 4
def
= N 0 det k (i ∂/E + iM )/µ k (4.153)

which can be understood as a formal continuous generalization of the formula

(4.145) that represents a determinant. Nevertheless, it should be noticed
that the euclidean free Dirac operator (i ∂/E + iM ) is not hermitean but only
normal, i.e. , it commutes with its adjoint [ i ∂/E − iM , i ∂/E + iM ] = 0 .
This fact implies that the euclidean free Dirac operator i ∂/E + iM also
commutes with the positive definite diagonal operator

(i ∂/E − iM )(i ∂/E + iM ) = M 2 − ∂E µ ∂E µ I = (i ∂/E + iM )(i ∂/E − iM )

that means
[ i ∂/E ± iM , M 2 − ∂E µ ∂E µ ] = 0

166
where the eigenfunctions and the eigenvalues of the euclidean Klein-Gordon
operator are respectively given by

φ E , p (xE ) = (2π)−2 exp { i pE µ xE µ } λ (pE ) = pE2 + M 2 > 0

Notice that the normal operator i ∂/E + iM is also invertible, since we have

−1
(i ∂/E + iM ) −1 = (i ∂/E − iM ) M 2 − ∂E µ ∂E µ

From the explicit form

i∂4 + ∂3 ∂1 − i∂2
 
 iM 0 
∂1 + i∂2 i∂4 − ∂3
 

 0 iM 

i ∂/E + iM =  
i∂4 − ∂3 −∂1 + i∂2
 


 iM 0 


−∂1 − i∂2 i∂4 + ∂3
 
0 iM

it can be readily checked by direct inspection that

 2
det k (i ∂/E + iM )/µ k= det k M 2 − ∂E2 /µ2 k

so that from the Zeta function definition (3.123) we eventually obtain

V M4
  
M 3
det k (i ∂/E + iM ) /µ k = exp ln − (4.154)
8π 2 µ 4

Turning back to (4.153) we have

V M4
  
(0) 0 M 3
ZE [ 0 , 0] = 1 ⇐⇒ N ≡ exp − ln −
8π 2 µ 4

and again the transition to the Minkowski spacetime can be immediately

done by simply replacing Veuclidean ↔ i Vminkowskian
The conclusion of all the above formal reasoning is as follows : we are
allowed to define the functional integral for a free Dirac spinor quantum field
theory by the equalities
 Z Z 
Z0 [ ζ , ζ̄ ] = exp − d x d y ζ̄ (x) SF (x − y) ζ (y)
Z Z
def 
= N Dψ Dψ̄ exp i S 0 [ ψ̄ , ψ ]
 Z h i
× exp i d x ζ̄(x) ψ (x) + ψ̄ (x) ζ(x) (4.155)

167
where Z
S 0 [ ψ̄ , ψ ] = d x ψ̄ (x) (i ∂/ − M ) ψ (x)

N = constant × det k i ∂/ − M k
V M4
  
def M 3
= exp − i ln − (Zeta regularization)
8π 2 µ 4
Z Z

Z0 [ 0 , 0 ] = N Dψ Dψ̄ exp i S 0 [ ψ̄ , ψ ] = 1 = h 0 | 0 i

It is clear that, by construction, the functional integral for a free Dirac spinor
quantum field theory does satisfy all the requirements of linearity, translation
invariance, rescaling and integration by parts which I have discussed in the
case of the real scalar field.

168
4.7 The C, P and T Transformations
The charge conjugation C is the discrete internal symmetry transformation
under which the particles and antiparticles are interchanged. The parity
transformation P or spatial inversion is the discrete spacetime symmetry
transformation such that
x µ = (t, r) 7→ x 0 µ = (t, − r) = P x µ
Under P the handedness of the particles motion is reversed so that, for
example, a left handed electron e−
L is transformed into a right handed positron
+
eR under the combined CP symmetry transformation. Thus, if CP were an
exact symmetry, the laws of Nature would be the same for matter and for
antimatter.
The experimental evidence actually shows that most phenomena in the
particles Physics are C and P symmetric and thereby also CP symmetric.
In particular, these symmetries are respected by the electromagnetic and
strong interactions as well as by classical gravity. On the other hand, the
weak interactions violate C and P in the strongest possible way. Hence, while
weak interactions do violate C and P symmetries separately, the combined
CP symmetry is still preserved. The CP symmetry is, however, violated in
certain rare processes, as discovered long ago in neutral K mesons decays
and recently observed in neutral B decays. Thus, only the combined discrete
CPT symmetry transformation, where T denotes the time inversion
x µ = (t, r) 7→ x 0 µ = (− t, r) = T x µ
is an exact symmetry for all laws of Nature, just like for the invariance under
the restricted Poincaré continuous group. In the sequel we will examine in
some detail the C , P and T transformations for the quantized Dirac field.

4.7.1 The Charge Conjugation

We have seen before that the quantization of the free Dirac wave field leads
to the appearance of particle and antiparticles, which are characterized by
the very same mass and spin although opposite charge. The latter one may
have a different interpretation depending upon the physical context: namely,
it could be electric, barionic, leptonic et cetera. In any case, the existence of
the charge conjugation invariance just implies
1. the existence of the antiparticles
2. the equality of all the quantum numbers, but charge, for a particle
antiparticle pair

169
We are looking for a unitary charge conjugation operator in the theory of the
quantized Dirac field, the action of which will be given by
ψ c (x) = C ψ (x) C † (4.156)
Charge conjugation is conventionally defined as the operation in which
particles and antiparticles are interchanged. It follows thereby that if we set
C cp,r C = dp,s C dp,r C = cp,s (4.157)
∀ r, s = 1, 2 p ∈ R3
with
C2 = I C = C † = C −1 (4.158)
It turns out that the standard spin-states (4.33) do fulfill the remarkable
relationship
u r (p) = − i γ 2 v ∗r (p)
v r (p) = − i γ 2 u∗r (p)
Hence, the transformation law (4.156) can be rewritten as
X h i
c †
ψ (x) = d p , r u p , r (x) + c p , r v p , r (x)
p,r
1 X h 2 ∗ † 2 ∗
i
= d p , r γ v p , r (x) + c p , r γ u p , r (x)
i p,r

>
>
= − i γ 2 ψ † (x) = − i ψ̄ (x) γ 0 γ 2 (4.159)
which corresponds to the quantum mechanical counterpart of the classical
transformation (2.83). Moreover we readily find

>
>
ψ̄ c (x) = ψ c † (x) γ 0 = − i γ 2 ψ (x) γ 0 = − i γ 0 γ 2 ψ (x) (4.160)
Working out the transformations of bilinears is a little bit tricky and it helps
to write the spinor indices explicitly. For instance, the mass scalar operator
becomes

>
>
: ψ̄ c (x) ψ c (x) : = : − i γ 0 γ 2 ψ (x) − i ψ̄ (x) γ 0 γ 2 :
= − : γ α0 β γ β2 δ ψ δ (x) ψ̄ η (x) γ η0 ω γ ω2 α :
= + : ψ̄ η (x) γ η0 ω γ ω2 α γ α0 β γ β2 δ ψ δ (x) :
= + : ψ̄ (x) γ 0 γ 2 γ 0 γ 2 ψ (x) :
= − : ψ̄ (x) ( γ 0 )2 ( γ 2 )2 ψ (x) :
= + : ψ̄ (x) ψ (x) : (4.161)

170
where the change of sign in the third line is just owing to the spinor field
canonical anticommuation relations. Hence the femion mass scalar operator
is invariant under charge conjugation, i.e.

: ψ̄ c (x) ψ c (x) : = : ψ̄ (x) ψ (x) :

while, in a quite analogous way, one can show that the electric current density
changes its sign : namely,

: ψ̄ c (x) γ µ ψ c (x) : = − : ψ̄ (x) γ µ ψ (x) :

4.7.2 The Parity Transformation

Suppose it is possible, with a suitable modification of some experimental
apparatus, to realize the space inversion and to obtain the parity transformed
state of a Dirac particle or antiparticle. From the quantized Dirac field point
of view we look for a unitary operator P satisfying

ψ 0 (x 0 ) = ψ 0 (x0 , − x) = P ψ (x) P †
= e i η γ 0 ψ (x0 , − x) (0 ≤ η < 2π) (4.162)

where the transformation law (2.59) of the classical Dirac wave field under
space inversion has been taken into account. Inserting the normal mode
expansion (4.28) we come to the relations
X h i
† † † †
P ψ (x) P = P c p , r P u p , r (x) + P d p , r P v p , r (x)
p,r
X h i
= e iη c p , r γ 0 u p , r (t, −x) + d †p , r γ 0 v p , r (t, −x)
p,r

Now, it turns out that the previously introduced standard spin-states (4.33)
do satisfy
γ 0 u r (− p) = u r (p) γ 0 v r (− p) = − v r (p)
so that we can write
X h i
† †
P ψ (x) P = P c p , r P u p , r (x) + P d †p , r †
P v p , r (x)
p,r
X h i
= e iη c −p , r u p , r (x) − d †−p , r v p , r (x) (4.163)
p,r

Hence, if we require

P c p , r P † = e iη c − p , r P d p , r P † = − e − iη d − p , r (4.164)

171
then the transormation law (4.162) is fulfilled. Furthermore, it can be readily
verified from the expressions (4.65) and (4.66) that we have
P P µ P † = Pµ (4.165)
If we choose η = 2πk (k ∈ Z) then the relative parity of the particle-
antiparticle system is equal to minus one and, contextually, the square of
the parity operator (or space inversion operator) P is equal to the identity
operator, that is
P2 = I P † = P = P −1 (η = 2πk , k ∈ Z)
As a consequence we can write for instance
P ψ (t, x) P = γ 0 ψ (t, − x) P ψ † (t, x) P = ψ̄ (t, − x) (4.166)
† †
P ψ̄ (t, x) P = P ψ (t, x) γ P = P ψ (t, x) P γ = ψ̄ (t, − x) γ 0
0 0

P (ψ̄ ψ) (t, x) P = + (ψ̄ ψ) (t, − x) (4.167)

P (ψ̄ γ5 ψ) (t, x) P = − (ψ̄ γ5 ψ) (t, − x) et cetera (4.168)

4.7.3 The Time Reversal

Let us turn now to the implementation of time reversal transformation. It
is known5 that the time reversal transformation in quantum mechanics is
achieved by means of antilinear and antiunitary operators T : H → H ,
where H is a Hilbert space, which satisfy
T (a | α i + b | β i) = a∗ T | α i + b∗ T | β i a, b ∈ C α, β ∈ H
h α | T †T β i = h T α | T β i = h β | α i ∀ α, β ∈ H
so that invariance under time reversal requires that
[ T , P0 ] = [ T , H ] = 0
where P0 = H denotes the infinitesimal generator of the time translations
of the system. The physical significance of T as the time reversal operator
requires that, while spatial relations must be unchanged, all momenta and
angular momenta must be reversed. Hence, we shall postulate the conditions
†
T PT = −P {T , P} = 0 (4.169)
†
T L ρσ T = − L ρσ {T , L ρσ } = 0 (4.170)
†
T S ρσ T = − S ρσ {T , S ρσ } = 0 (4.171)
5
See E.P. Wigner, Group Theory and Its Application to Quantum Mechanics of Atomic
Spectra, translated by J.J. Griffin, Academic Press, New York, 1959, Appendix to Chapter
20, p. 233. See also V. Bargmann, J. Math. Phys. 5 (1964) 862.

172
 Since the antiunitary time reversal operator T is defined to reverse the
sign of all momenta and spins we therefore require
†
T cp,r T = exp { i η p , r } c −p , − r (4.172)
†
T dp,r T = exp { i ζ p , r } d −p , − r (4.173)
T c †p , r T †
= exp {− i η p , r } c †−p , − r (4.174)
T d †p , r T †
= exp {− i ζ p , r } d †−p , − r (4.175)

where the notation − r refers to the change of sign of the spin eigenvalue,
that is

c −p , − 1 = c −p , 2 c −p , − 2 = c −p , 1 et cetera
Although the phase factors exp { i η p , r } and exp { i ζ p , r } are arbitrary, it
is always possible to choose them in such a manner that the (Dirac spinor)
fields undergo utmost simple transformation laws under time reversal.
This means that, in order to obtain the time reversed spinor field operator
T ψ(x) T † , we have to perform on the complex conjugated c-number part
(i.e. not operator part) of the spinor some unitary operation in such a way
that eventually we come to the transformation rule

ψ 0 (x 0 ) = ψ 0 (− t, x) = T ψ (t, x) T †
(4.176)

By inserting once again the the normal mode expansion (4.28) we get, up to
some trivial substitutions in the integrands,
X
T ψ (t, x) T † = ∗
c p , r exp { i η −p , − r } u − r (−p)
p,r

× [ (2π) 3 2ω p ] −1/2 exp {− i (−t) ω p + i p · x }

X
+ d †p , r exp {− i ζ −p , − r } v −∗ r (−p)
p,r

× [ (2π) 3 2ω p ] −1/2 exp {i (−t) ω p − i p · x }

In arriving at this equation, the antiunitarity nature of T has been exploited

and has resulted in complex conjugation. It is easily seen that the right-hand
side of this equation becomes a local expression for the field at time − t if a
4 × 4 matrix Θ can be found such that

∗
u− r (−p) exp { i η −p , − r } = Θ u r (p) (4.177)
∗
v − r (−p) exp {− i ζ −p , − r } = Θ v r (p) (4.178)

173
The closure relation (4.19) and the orthonormality relation (4.18) imply that
Θ must be unitary. Furthermore, the above relations (4.177) and (4.178) are
consistent with the spin-states eigenvalue equations (4.17) only if Θ satisfies
the conditions

[Θ, H ] = 0
k ∗ k
β∗ Θ = Θβ


α Θ = −Θα (k = 1, 2, 3)
As a matter of fact we have, for example,

H u r (p) = α k p k + β M u r (p) = ω p u r (p)

− α k ∗ p k + β ∗ M u∗r (− p) = ω p u∗r (− p)

− α k ∗ p k + β ∗ M Θ u r (p) = ω p Θ u r (p)

Θ α k p k + β M u r (p) = Θ H u r (p) = ω p Θ u r (p)

The unitary solution to these equations is unique except for an arbitrary

phase factor that we can suitably choose to be equal to one, that means

η p , r = ζ p , r = 2k π i ( k ∈ Z , ∀ p ∈ R 3 , r = 1, 2 )
In the spinorial-chiral-Weyl representation (2.68), the matrix
 
1 3 σ2 0 
Θ = −γ γ = −i 

 
0 σ2


is a solution. It has the important property

Θ Θ∗ = − I Θ Θ† = I
which can be proved to hold true independently of the representation of the
Dirac matrices. It follows that the Dirac particle-antiparticle spinor quantum
wave field is transformed under time reversal according to

†
T ψ (t, x) T = Θ ψ (− t, x)
T ψ † (t, x) T †
= ψ † (− t, x) Θ † (4.179)
It is now rather easy to derive the action of the time reversal operator on
various bilinears. First of all we have

†
T ψ̄ (t, x) T = T ψ † (t, x) T † γ 0 ∗
= ψ † (− t, x) Θ † γ 0 ∗
= ψ̄ (− t, x) γ 1 γ 3 (4.180)

174
Then the transformation law for the scalar mass bilinear under time reversal
becomes

†
= ψ̄ (− t, x) γ 1 γ 3 − γ 1 γ 3 ψ (− t, x)


T ψ̄ (t, x) ψ (t, x) T
= + ψ̄ (− t, x) ψ (− t, x) (4.181)

A quite analogous calculation gives for instance

ψ † (− t, x) ψ (− t, x) for µ = 0

µ †
T ψ̄ (t, x) γ ψ (t, x) T =
− ψ̄ (− t, x)γ k ψ (− t, x) for µ = 1, 2, 3

References

1. N.N. Bogoliubov and D.V. Shirkov (1959) Introduction to the Theory

of Quantized Fields, Interscience Publishers, New York.

2. E. Merzbacher, Quantum Mechanics. Second edition, John Wiley &

Sons, New York, 1970.

3. C. Itzykson and J.-B. Zuber (1980) Quantum Field Theory, McGraw-

Hill, New York.

4. Pierre Ramond, Field Theory: A Modern Primer, Benjamin, Reading,

Massachussets, 1981.

5. R. J. Rivers (1987) Path integral methods in quantum field theory,

Cambridge University Press, Cambridge (UK).

6. M.E. Peskin and D.V. Schroeder (1995) An Introduction to Quantum

Field Theory, Perseus Books, Reading, Massachusetts.

175
4.8 Problems
1. The Belifante tensor. In general, for a relativistic wave field with
non zero spin, the canonical energy momentum tensor needs not to be
symmetric. Show that one can always find a term B λµν antisymmetric
under λ → µ or ν such that the Belifante tensor
Θ µν (x) = T µν (x) − ∂ λ B λ µν (x)
is symmetric and the corresponding conserved Noether current for the
Lorentz transformations is written in the form
M µλκ (x) = x λ Θ µκ (x) − x κ Θ µλ (x)

Solution. One can always decompose the canonical energy momentum

tensor of the Dirac wave field into its symmetric and antisymmetric
parts

i ↔ i ↔
T µν (x) = ψ̄ (x) γ µ ∂ ν ψ (x) + ψ̄ (x) γ ν ∂ µ ψ (x)
4 4
i ↔ i ↔
+ ψ̄ (x) γ µ ∂ ν ψ (x) − ψ̄ (x) γ ν ∂ µ ψ (x)
4 4
in which

i ↔ i ↔
T µν (x) − T ν µ (x) = ψ̄ (x) γ µ ∂ ν ψ (x) − ψ̄ (x) γ ν ∂ µ ψ (x)
2 2
λµν
= ∂λ S (x)
in accordance with eq. (4.44), where the third rank tensor of the spin
angular momentum density of the Dirac field is defined by the Noether
theorem and reads

1
S λ µν (x) ≡ ψ̄ (x) {γ λ , σ µν } ψ (x)
2
which enjoys by construction

S λ µ ν (x) + S λ ν µ (x) = 0 S λ µν (x) = S µνλ (x)

If we introduce the auxiliary quantity
λµν 1  λµν µλν νλµ

B (x) ≡ S (x) − S (x) − S (x)
2
176
which is evidently related to a non-vanishing spin angular momentum
density tensor and which fulfills by definition

1
B λµν + B µλν = 0 ∂ λ B λ µ ν (x) = ∂ λ S λ µ ν (x)
2
In fact we have

2∂ λ B λ µ ν (x) = ∂ λ S λ µ ν (x) − ∂ λ S µλν (x) − ∂ λ S νλµ (x)

= ∂ λ S λ µ ν (x) + ∂ λ S µνλ (x) + ∂ λ S νµλ (x)
= ∂ λ S λ µ ν (x) + ∂ λ S λµν (x) + ∂ λ S λνµ (x)
= 2∂ λ S λ µ ν (x) + ∂ λ S λνµ (x) = ∂ λ S λ µ ν (x)

Then the symmetric energy momentum tensor is obtained by setting

µν def 1  µν νµ

Θ (x) = T (x) + T (x)
2
i ↔ i ↔
= ψ̄ (x) γ µ ∂ ν ψ (x) + ψ̄ (x) γ ν ∂ µ ψ (x)
4 4
µν 1 λµν
= T (x) − ∂ λ S (x)
2
= T µν (x) − ∂ λ B λ µ ν (x)

which apparently satisfies the continuity equation since

∂ µ Θ µν (x) = ∂ µ T µν (x) − ∂ λ ∂ µ B λ µν (x) = ∂ µ T µν (x) = 0

Consequently the total angular momentum density tensor for the Dirac
field can be written in the purely orbital form. In fact we have

M µ κ λ (x) = x κ Θ µ λ (x) − x λ Θ µ κ (x)

1
= x κ T µλ (x) + x κ ∂ ρ S ρ µ λ (x)
2
1
− x λ T µκ (x) − x λ ∂ ρ S ρ µ κ (x)
2
= x T (x) − x λ T µκ (x)
κ µλ

1 h κµλ λµκ
i
− S (x) − S (x)
2
1 h
λ ρµκ κ ρµλ
i
− ∂ρ x S (x) − x S (x)
2

177
 and taking the symmetry properties of the spin angular momentum
density tensor suitably into account

M µ κ λ (x) = x κ Θ µλ (x) − x λ Θ µκ (x)

= x κ T µλ (x) − x λ T µκ (x) + S µ λ κ (x)
1 h κ λρµ i
+ ∂ρ x S (x) − x λ S κ ρ µ (x)
2
in such a manner that the continuity equations hold true : namely,

 
∂µ M µ κ λ (x) = ∂ µ x κ Θ µλ (x) − x λ Θ µκ (x)
 
= ∂ µ x κ T µλ (x) − x λ T µκ (x) + S µλκ (x)
1  
− ∂ ρ ∂µ x λ S κρµ (x) − x κ S λρµ (x)
2  
κ µλ λ µκ µλκ
= ∂ µ x T (x) − x T (x) + S (x) = 0

2. The massless neutrino − antineutrino field. The free quantum

theory of a massless left–handed or right-handed spinor field leads to
the description of the neutrino particles and antineutrino antiparticles
in the limit of a negligible neutrino mass. Discuss the corresponding
canonical quantization.

Solution. The Weyl lagrangians for the left–handed and right–handed

massless spinors are very simple: namely,
LL = ψL† (x) σ µ i ∂ µ ψL (x) LR = ψR† (x) σ̄ µ i ∂ µ ψR (x) (4.182)
where
σµ = ( I , σ ) = σµ† σ̄µ = ( I , − σ ) = σ̄µ†
whereas I denotes the 2 × 2 identity matrix. The Euler–Lagrange field
equations drive immediately to the Weyl equation for left–handed and
right–handed massless spinors
σ µ i∂µ ψL (x) = 0 i∂µ ψL† (x) σ µ = 0 (4.183)
µ
σ̄ i∂µ ψR (x) = 0 i∂µ ψR† (x) σ̄ µ =0 (4.184)
After setting
Z
−3/2
ψL (x) = (2π) d4 p ψ̃L (p) e−i p ·x

178
 Z
−3/2
ψR (x) = (2π) d4 p ψ̃R (p) e−i p ·x

we come to the decoupled spinorial equations

( p0 + σ · p ) ψ̃L (p) = 0 ( p0 − σ · p ) ψ̃R (p) = 0

These homogeneous algebraic equations admit nontrivial solutions iff

det ( p0 ± σ · p ) = p20 − p2 = 0 ⇔ p0 = ± |p| = ± p

whence we readily obtain the algebraic equations for positive energy

particles of momentum p for the left and right spinors respectively
 
p + p z p x − ip y
 ψ̃L (p, p) = 0 (4.185)

 

px + ipy p − pz

 
p − p z −p x + ip y
 ψ̃R (p, p) = 0 (4.186)

 

−px − ipy p + pz


which clearly drive to the relations

ψ̃R (p, p) = ψ̃L (p, − p) = ψ̃L (− p, p) = ψ̃R (− p, − p)

ψ̃L (p, p) = ψ̃R (p, − p) = ψ̃R (− p, p) = ψ̃L (− p, − p)

For example, one can easily write two linearly dependent left–handed
spinor solutions with the positive energy for the Weyl equation, e.g. ,
   
− p x + ip y 0 − p + p z
ψ̃L (p, p) =  ψ̃L (p, p) =   (4.187)
 
  

p + pz px + ipy
  

which are proportional because

px + ipy
ψ̃L0 (p, p) = ψ̃L (p, p)
p + pz
The linearly dependent right–handed spinor solutions for the positive
energy Weyl equation are obtained after sending p into − p , that yields
 
px − ipy 
ψ̃R (p, p) = 

 
p − pz

 
0 p + pz 
ψ̃R (p, p) = (−1)  (4.188)

 
px + ipy


Furthermore, it is easy to check that the following relations hold true

ψ̃R0 (p, p) = − iσ2 ψ̃L∗ (p, p) ψ̃L0 (p, p) = − iσ2 ψ̃R∗ (p, p) (4.189)

179
Of course the negative energy solutions of the left Weyl equation (4.183)
do involve right–handed spin states and correspond to the antiparticles,
the converse holding true for the right Weyl equation (4.184). As a
matter of fact, if we send p into − p in eq. (4.187), i.e. if we turn to the
antiparticle spin states, we get for instance
ψ̃L (p, p) 7→ { p ↔ − p} 7→ ψ̃L (− p, p) ∝ ψ̃R0 (p, p)
Thus, to our purposes, it is utmost convenient to take by definition the
following properly normalized spin states
 
− 12  − p x + ip y
uL (p) = ( p + pz )   ≡ u( p)


p + pz

 
− 12  − px + ipy 
vL (p) = ( p − pz )    ≡ u(− p)

− p + pz
 
− 21  p + pz 
uR (p) = (−1)( p + pz )    = −iσ2 u∗L (p) ≡ v( p)

px + ipy
 
− 21  p z − p  = −iσ2 vL∗ (p) ≡ v(− p)
vR (p) = (−1)( p − pz )   
px + ipy


which satisfy the orthonormality relations

u†L (p) uL (p) = vL† (p) vL (p) = 2p u†L (p) vL (p) = 0
u†L (p) uR (p) = 0 vL† (p) vR (p) = 0 (4.190)
† † †
uR (p) uR (p) = vR (p) vR (p) = 2p uR (p) vR (p) = 0
Then we can suitably define the massless Weyl spinor wave functions
1
u ± p (x) ≡ [ (2π)3 2 | p | ] − 2 u(± p) e ∓ i p · x ( p0 = | p | )
− 12
v ± p (x) ≡ [ (2π)3 2 | p | ] v(± p) e ∓ i p · x ( p0 = | p | )
which are solutions of the Weyl equation
σ µ i∂µ u ± p (x) = 0 ( ∀ p ∈ R3 , p 0 = | p | )
σ̄ µ i∂µ v ± p (x) = 0 ( ∀ p ∈ R3 , p 0 = | p | )
and do satisfy from (4.190) the orthonormality relations
Z
dx (u ± p (t, x))† u ± q (t, x) = δ(p − q)
Z
dx u†p (t, x) u − q (t, x) = 0
Z
dx (v ± p (t, x))† v ± q (t, x) = δ(p − q)
Z
dx v †p (t, x) v − q (t, x)

180
Hence we can write the normal modes decomposition of the quantized
left–handed Weyl spinor field in the form
X
ψL (x) = [ c p u p (x) + d†p u − p (x) ] (4.191)
p
X
ψR (x) = [ d p v p (x) + c†p v − p (x) ] (4.192)
p

the creation and destruction operators satisfying the usual canonical

anticommutation relations

{c p , c†q } = {d p , d†q } = δ(p − q) (4.193)

all the remaining anticommutators being equal to zero.

Since the generators of the rotations group for both the irreducible
Weyl spinor representations τ 1 0 and τ 0 1 are S jk = 12 ε jk` σ` it follows
2 2
that the corresponding Pauli-Lubanski operator reads

W 0 = 21 ε 0 ijk pi S jk = 12 ε 0 ijk pi 12 ε jk` σ` = 41 pi σ` (2 δ`i ) = 12 p · σ

Wj = 21 pσj − 12 iεjk` pk σ`
which entails

g µν Wµ Wν = 41 pj pk σj σk − 14 ( pσj − iεjk` pk σ` ) ( pσj − iεjrs pr σs )

= − 21 p2 + 41 ( δ kr δ `s − δ ks δ `r ) pk pr σ` σs = 0

where the 2 × 2 identity matrix has been understood. Then, thanks

to the light–like nature of the Pauli-Lubanski operator, we can identify
the helicity operator to be
W0 |W|
h≡ = = 12 ~ n · σ n≡p
b = p/p
|p| p
together with the Weyl energy operator

H = 2W0 = c p · σ

in such a manner that we find

HuL (p) = pc uL (p) huL (p) = − 21 ~ uL (p)

HuR (p) = pc uR (p) huR (p) = 12 ~ uR (p)
HvL (p) = − pc vL (p) hvL (p) = 12 ~ vL (p)
HvR (p) = − pc vR (p) hvR (p) = − 21 ~ vR (p)

181
 It follows thereby that, mandatorily, the positive energy solutions of
the massless left Weyl spinor have negative helicity h = − 12 , while the
negative energy solutions exhibit positive helicity h = 21 , the situation
being reversed for the massless right Weyl spinors. Hence, it turns out
that the particles with negative helicity correspond to the left–handed
massless spinors, while the antiparticles to the right–handed massless
spinors. Thus, the spin of left–handed massless Weyl spinor is always
opposite to the direction of motion, while the spin of the right-handed
massless Weyl spinor is always towards the direction of motion.
In Nature, whenever the masses of neutrinos and antineutrinos can be
neglected 6 , neutrinos are left–handed, i.e. with negative helicity, while
antineutrinos are right–handed with positive helicity. Moreover, the
massless Weyl fields are charged fields. In fact, the Weyl lagrangians
(4.182) are invariant with respect to the phase transformations

ψL (x) 7→ ψL0 (x) = e iLθ ψL (x)

ψR (x) 7→ ψR0 (x) = e −iLθ ψR (x)
†
ψL† (x) 7→ (ψL0 (x)) = e −iLθ ψL† (x)
†
ψR† (x) 7→ (ψR0 (x)) = e iLθ ψR† (x)

where L is the so called lepton number of the Weyl spinor field while
0 ≤ θ ≤ 2π . Hence, from Noether’s theorem, it follows that the left–
handed and right–handed vector currents

JLµ (x) = L ψL† (x) σ µ ψL (x) JRµ (x) = (− L) ψR† (x) σ̄ µ ψR (x)

do satisfy the continuity equation

∂µ JLµ (x) = ∂µ JRµ (x) = 0

As a consequence, the lepton number operator can be expressed in the

form
Z
QL = L dx : ψL† (x)ψL (x) :
X
c†p c p − d†p d p = − QR

= L
p

6
According to C. Amsler et al. (Particle Data Group), Physics Letters B667, 1 (2008)
data from tritium beta decay experiments lead to the upper bound m(νe ) = m(ν̄e ) < 2
eV, while the present limits for heavier flavours are m(νµ ) = m(ν̄µ ) < 170 KeV and
m(ντ ) = m(ν̄τ ) < 18.2 MeV.

182
In Nature there are three leptonic numbers Le , Lµ , Lτ , one for each
flavour. The assignements are Lı = +1 ( ı = e, µ, τ ) for particles,
e.g. electrons e− and left–handed neutrinos νe , while − Lı = −1 for
antiparticles like, for instance, the antimuons µ+ and the right–handed
antineutrinos ν̄µ .
Notice that the charge conjugation transformation for the quantized
fields is nothing but the exchange between the particle and antiparticle
creation and destruction operators. For example, in the case of the
quantized massless right-handed Weyl spinor we have
X
ψRc (x) = [ c p v p (x) + d†p v − p (x) ]
p

and from the relations (4.189) we immediately obtain

v p (x) = −iσ2 u∗p (x) v − p (x) = −iσ2 u∗− p (x)

u p (x) = iσ2 v ∗p (x) u − p (x) = iσ2 v ∗− p (x)

that yields
X  >
ψRc (x) = −iσ2 [ c p u∗p (x) + d†p u∗− p (x) ] = −iσ2 ψL† (x) (4.194)
p
X  >
ψLc (x) = iσ2 [ d p v ∗p (x) + c†p v ∗− p (x) ] = iσ2 ψR† (x) (4.195)
p

The massless Weyl fields can also be represented by complex bispinors.

To this concern, it appears that if Ψ(x) is a four components complex
bispinor solution of the massless Dirac equation

i ∂/ ψ(x) = 0

then ψ 0 (x) = γ5 ψ(x) is also a solution of the same equation owing to

{γ5 , γ µ } = 0 . Thus complex bispinor solutions of the massless Dirac
equation are characterized by their chirality, i.e. , they are eigenfuctions
of the chirality matrix γ5 . It follows thereby that in the Weyl, spinorial,
chiral representation of the Clifford algebra we find

ψ± (x) ≡ 12 ( I ± γ5 ) ψ(x) γ5 ψ± (x) = ± ψ± (x)

Z   Z  
u p (x) † u − p (x)
ψ− (x) = dp c p   + dp d p 
 
  

0 0
  
Z   Z  
0 0
ψ+ (x) = dp d p   + dp c†p 
 
  

v p (x) v − p (x)
  

183
 in such a manner that charge conjugation acts as follows. From the
general relation (4.159) valid for any operator valued Dirac bispinor
quantum field, viz.,

>
ψ c (x) = − iγ 2 ψ † (x)

if we make use of the real and symmetric chirality or helicity projectors

P± ≡ 12 ( I ± γ5 ) = (P± )> = (P± )†

we can write

>
>
(ψ± (x)) c = − iγ 2 ψ † (x)P± = − iγ 2 P± ψ † (x)
h
> i
= P∓ − iγ 2 ψ † (x) = P∓ ψ c (x)
≡ ψ∓c (x) (4.196)

in accordance with our previous eqs. (4.194,4.195). Thus the charge

conjugated of the left–handed component of a Dirac bispinor is the
right component of the charge conjugated Dirac bispinor

(ψ− (x)) c = (P− ψ(x)) c = ψ+c (x) = P+ ψ c (x)

the converse holding true for the right–handed component, the opposite
charges of the charge conjugated bispinors being the lepton numbers.
In conclusion, the chiral bispinor quantized fields ψ± (x) do describe
massless neutrinos νe , νµ , ντ of negative chirality/helicity but positive
lepton number, together with the massless antineutrinos ν̄e , ν̄µ , ν̄τ of
positive chirality/helicity but negative lepton charge.

3. Majorana fermions. One can write a relativistically invariant field

equation for a massive 2-component left Weyl spinor wave field ψL
that transforms according to (2.41). Call such a 2-component field
χa (x) (a = 1, 2) . Let us consider the Weyl spinor wave field as a classical
anticommuting field, i.e. a Grassmann valued left Weyl spinor field
function over the Minkowski spacetime which satisfies

{ χa (x) , χb (y) } = 0 ( x, y ∈ M a, b = 1, 2 )

together with the complex conjugation rule

(χ1 χ2 )∗ = χ∗2 χ∗1 = − χ∗1 χ∗2

so to imitate the hermitean conjugation of quantum fields.

184
(a) Show that the classical lagrangian
1 † ↔ m  >
χ (x) σ µ i ∂ µ χ(x) + χ (x) σ2 χ(x) + χ† (x) σ2 χ∗ (x)

LL =
2 2
is real with χ† = (χ∗ )> and yields the Majorana wave field equation
i σ µ ∂ µ χ (x) + m σ2 χ ∗ (x) = 0
where σ µ = (1 , − σk ) . That is, show that this equation, named the
Majorana field equation, is relativistically invariant and that it implies
the Klein-Gordon equation (+m2 ) χ(x) = 0 . This form of the fermion
mass is called a Majorana mass term.
(b) Show that all we have obtained before can be formulated in terms
of a self-conjugated left handed real bispinor called the free Majorana
spinor ψL = ψL∗ = ψLc and derive the symmetries of the corresponding
Action.
(c) Perform the quantum theory of the Majorana massive field.
Solution
(a) Consider the transformation rule (2.55)

Λ L† σ µ Λ L = Λ µν σ ν
Then we immediately get
↔ ↔
[ χ 0 (x 0 ) ] † σ µ i ∂ µ0 χ 0 (x 0 ) = χ † (x) Λ †L σ µ Λ L i ∂ ρ χ (x) Λ µρ
↔
= χ † (x) Λ µν σ ν i ∂ ρ χ (x) Λ µρ
↔
= χ † (x) σ µ i ∂ µ χ (x)
which vindicates once more the Poincaré invariance of the left Weyl
kinetic lagrangian
1 † ↔
TL [ χ ] = χ (x) σ µ i ∂ µ χ (x)
2
Furthermore we have
χ 0 > (x 0 ) σ2 χ 0 (x 0 ) = χ > (x) Λ >
L σ2 Λ L χ (x)
= χ (x) σ Λ L (σ2 )2 Λ L χ (x)
> 2 −1

= χ> (x) σ2 χ(x)

= − i χ1 (x) χ2 (x) + i χ2 (x) χ1 (x)
= − 2 i χ1 (x) χ2 (x)

185
for anticommuting Grassmann valued Weyl spinor fields. It follows
therefrom that the mass term is the real Lorentz scalar
h i
m ∗ ∗
L L = − i m χ1 (x) χ2 (x) + χ1 (x) χ2 (x)
h i
∗
= i m χ∗2 (x) χ∗1 (x) + χ2 (x) χ1 (x) = (L m
L)

The Lagrange density can be rewritten as

1 † ↔ m  >
χ (x) σ µ i ∂ µ χ(x) + χ (x) σ2 χ(x) + χ † (x) σ2 χ∗ (x)

LL =
2 2
† m  >
µ
χ (x) σ2 χ(x) + χ † (x) σ2 χ∗ (x)

= χ (x) σ i ∂ µ χ(x) +
2
i 
† µ

− ∂ µ χ (x) σ χ(x)
2
m  >
˙ χ † (x) σ µ i ∂ µ χ(x) + χ (x) σ2 χ(x) + χ † (x) σ2 χ∗ (x)

=
2
where = ˙ means that the 4-divergence term can be disregarded as it does
not contribute to the equations of motion. If we treat χa (x) (a = 1, 2)
and χ ∗a (x) (a = 1, 2) as independent field variables, then the Euler-
Lagrange field equations yield

∂µ δ L L / ∂µ δ χ (x) = i ∂ µ χ† (x) σ µ = δ L L / δ χ (x) = m χ> (x) σ 2

hence, taking the transposed and complex conjugate equation

i σ µ ∂ µ χ (x) + m σ 2 χ∗ (x) = 0 (I )

which is the Majorana field equation. Multiplication to the left by σ2

and taking complex conjugation gives

i σ2 ∂ 0 χ∗ (x) − i σ2 σk∗ ∂ k χ∗ (x) + m χ (x) = 0

Remembering that σ2 σk∗ σ2 = − σk and that σ̄ µ = ( 1 , σk ) we come to

the equivalent form of the Majorana left equation, i.e.

i σ̄ µ ∂ µ χ ∗ (x) + m σ2 χ (x) = 0 ( II )

Now, if act from the left with the operator i σ̄ ν ∂ ν to equation ( I ) and
use equation ( II ) we obtain

σ̄ ν σ µ ∂ ν ∂ µ χ (x) + m2 χ (x) = (  + m2 ) χ (x) = 0

so that the left-handed Weyl massive spinor satisfies the Klein-Gordon

wave equation.

186
(b) According to (2.84) we can introduce the Majorana left handed
self-conjugated bispinor
 
χ (x)
χL (x) =   = χLc (x) ( III )
 

− σ2 χ ∗ (x)


with the Lagrange density

1 ↔ m
LL = χ̄ L (x) i ∂/ χL (x) − χ̄ L (x) χL (x)
4 2
in such a manner that the Majorana mass term can be written in the
two equivalent forms
m  > m
Lm χ (x) σ2 χ(x) + χ † (x) σ2 χ∗ (x) = −

L = χ̄ L (x) χL (x)
2 2
R
so that it is clear that the Majorana’s action d x L L is no longer
invariant under the phase transformation

χL (x) 7−→ χL0 (x) = χL (x) e i α

Taking the Grassmann valued Majorana left handed spinor wave field
χL (x) to be defined by the self-conjugation constraint ( III ) , it is
easy to see that the pair of coupled Weyl equations ( I ) and ( II ) is
equivalent to the single bispinor equation

(α µ i ∂ µ − β m) χL (x) = 0 ( IV )

in which
 µ   
µ σ 0 0 1 
0
α = β=γ = γ µ = β αµ

 
 
0 σ̄ µ 1 0
  

while the Majorana lagrangian then becomes

1 † ↔ m †
LL = χ L (x) α µ i ∂ µ χL (x) − χ (x) β χL (x)
4 2 L
It is immediate to verify that the bispinor form ( IV ) of the field
equations does coincide with the two equivalent forms ( I ) and ( II ) of
the Majorana wave field equation : namely,

i σ µ ∂ µ χ (x) + m σ 2 χ∗ (x) = 0


i σ̄ µ σ2 ∂ µ χ ∗ (x) + m χ (x) = 0

187
Since the Majorana spinor wave field χL (x) has the constraint ( III )
which relates the lower two components to the complex conjugate of
the upper two components, a representation must exist which makes
the Majorana spinor wave field real, with the previous two independent
complex variables χa ∈ C (a = 1, 2) replaced by the four real variables
ψM, α ∈ R (α = 1, 2, 3, 4) . To obtain this real representation, we note
that    
χ ∗ 0 − σ2 
χL =  χL =   χL

 
 
 
− σ2 χ∗ σ2 0

∗
A transformation to left handed real bispinor fields ψM = ψM can be
made by writing

χL = S ψM χ∗L = S ∗ ψM = S ∗ S −1 χL

whence  
 0 − σ2
∗
S = S


σ2 0


Now, if we set  
0 i σ2 
 = i γ 2 ≡ ρ2

 
− i σ2 0


ρ 2 = ρ †2 ρ 22 = I
so that
exp {i ρ 2 θ} = I cos θ + i ρ 2 sin θ
then the solution for the above relation is the unitary matrix
√ 
2 
S = exp {− π i ρ 2 / 4} = I − i ρ2
2
which fulfills
√   √
2 1 σ2  2
I + γ2


S=  =

 
2 − σ2 1 2


or even more explicitly

0 −i 
 
√  1 0
2
 

 0 1 i 0 
S= 
 

2 

 0 i 1 0 

−i
 
0 0 1

It follows that we can suitably make use of the so called Majorana

representation for the gamma matrices which is given by the similarity

188
and unitary transformation acting on the gamma matrices in the Weyl
representation, viz.,
µ
γM ≡ S† γµ S
 
    0 i 0 0 
− −

0 σ 2 0 
 i 0 0 0 

γM =  =

 
  
0 0 0 −i 
 
0 σ2 
 

 
0 0 i 0
−i 0 0 0 
 
  
−
 
1 i σ 3 0 
 0 i 0 0 

γM =  =
 
  
− i σ3 −

0 0 0 i 0
 
 


 

0 0 0 i
0 0 0 −i 
 
    
2 0 σ2  
 0 0 i 0  
γM =  =

   
− σ2 0
 

 0 i 0 0  

−i 0 0 0
 
 
   0 i 0 0 
 
3 i σ1 0  
 i 0 0 0  
γM =  =

   

0 i σ1 

 0 0 0 i 


 
0 0 i 0
0 0 0 −i 
 
    
5 0 σ2  
 0 0 i 1  
γM =  =

   
−
 
σ2 0 
 0 i 0 0 


 
i 0 i 0
which satisfy by direct inspection
µ ν
{γM , γM } = 2 g µν ν
{γM 5
, γM }=0
0 0† k k† 5 5†
γM = γM γM = − γM γM = γM
ν ν∗ 5 5∗
γM = − γM γM = − γM
The result is that, at the place of a complex self-conjugated bispinor,
which corresponds to a left handed Weyl spinor, one can employ a real
Majorana bispinor: namely,
χL (x) = χLc (x) ↔ ψM (x) = S † χL (x) = ψM
∗
(x)
A quite analogous construction can obviously be made, had we started
from a right handed Weyl spinor. In so doing, the Majorana lagrangian
and the ensuing Majorana wave field equation take the form
↔
1 > ν >
LM = 4
ψM (x) αM i ∂ ν ψM (x) − 12 m ψ M (x) β M ψM (x)

189
 ∗
(i ∂/M − m) ψM (x) = 0 ψM (x) = ψM (x)
ν 0 ν 0 0
αM = γM γM αM =I β M ≡ γM
The only relic internal symmetry of the Majorana action is the discrete
Z2 symmetry, i.e. ψM (x) 7−→ − ψM (x) . The Majorana hamiltonian
reads
k
HM = αM p̂ k + m β M ( p̂ k = − i ∇k )

To solve the Majorana wave equation we set

d4 p e
Z
ψM (x) = ψM (p) exp {− i p · x}
(2π)3/2

with the reality condition

∗
ψeM (p) = ψeM (− p)

so that
ν
( p/M − m ) ψeM (p) = 0 p/M ≡ p ν γM
which implies

ψeM, α (p) = ( p/M + m ) αβ φeβ (p)

p2 − m2 φeα (p) = 0


(α = 1, 2, 3, 4)
φeα (p) = δ p2 − m2 fα (p)

Furthermore, from the reality condition on the Majorana spinor field

∗
ψeM, α (p) = ψM, α (− p)
e

we find
∗ ∗ e∗
ψeM, α (p) = ( p/M + m ) αβ φ β (p)

= ( − p/M + m ) αβ φe∗ (p)

β

= ψeM, α (− p) ⇐⇒ f β∗ (p) = f β (− p)

thanks to the circumstance that the gamma matrices in the Majorana

representation are purely imaginary. Then we can write
Z ∞ Z
 − 1/2
dp0 dp (2π) 3 2ω p

ψM (x) = θ (p0 )
−∞
× ( p/M + m ) αβ fβ (p) δ (p0 − ω p ) exp {− i p · x}

190
 Z ∞ Z
 − 1/2
(2π) 3 2ω p

+ dp0 dp θ (− p0 )
−∞
× ( p/M + m ) αβ fβ (p) δ (p0 + ω p ) exp {− i p · x}
Z
 − 1/2 +
= 2m dp (2π) 3 2ω p EM ( p ) f (p) e− i p x


Z
 − 1/2 −
+ 2m dp (2π) 3 2ω p EM ( p ) f ∗ (p) e i p x


def
Xh i
+ − ipx − ∗ ipx
= EM ( p ) a p e + EM ( p ) a p e
p
∗
= ψM (x)
p
where p0 = ω p , whereas a p = 2mf (p)/ (2π)3 2ω p . The projectors
onto the spin states are
±
EM ( p ) = ( m ± p/M ) / 2m ( p0 = ω p )

with
± ∓ ± ±
[ EM ( p ) ] ∗ = EM (p) [ EM ( p ) ] † = EM ( p̃ ) ( p̃ µ = p µ )
± ± ± ∓
[ EM ( p ) ] 2 = EM (p) EM ( p ) EM (p) = 0
± + −
tr EM (p) = 2 EM ( p ) + EM (p) = I
Now, in order to set up the spin states of the Majorana real spinor field,
let me start from the spin matrices in the Majorana representation
 
2 3 0 i σ3
ΣM, 1 = i γM γM =
 

− i σ3 0
 
 
3 1 σ2 0 
ΣM, 2 = i γM γM = 

 
0 σ2

 
1 2 0 − i σ1
ΣM, 3 = i γM γM = 

 

i σ1 0


and from the common eigenvectors of the matrix βM and of the diagonal
spin matrix ΣM, 2 that are e.g.
   
 0   i 
   
 0   1  0
ξ+ ≡  ξ− ≡ 
   

 
 
 
 γM ξ± = ξ±

 1 
  0 

    
 
i 0

191
or equivalently
   
 1   0 
   
 i   0  0
η+ ≡  η− ≡  η± = − η±
   
 
  
 γM


 0 
 
 i 

   
  
0 1

which do indeed satisfy by direct inspection

0
γM ξ± = ξ± ξ ∓† ξ ± = 0 ( ΣM, 2 ∓ 1 ) ξ ± = 0

in such a manner that we have by construction

ξ r† ξ s = 2 δ rs ξ r† γM
k
ξs = 0 ∀ r, s = ± ∨ k = 1, 2, 3

Notice that, for example, the spin states ξ r ( r = ± ) are in fact the
two degenerate eigenstates of the Majorana hamiltonian in the massive
neutral spinor particle rest frame p = 0 with positive eigenvalue p0 = m
and with opposite spin projections on the OY axis, viz.,
31
σM ξ ± ≡ 4i [ γM
3 1
, γM ] ξ ± = 12 ΣM, 2 ξ ± = ± 12 ξ ±

Then we define the Majorana spin states to be

1 +
u r ( p ) ≡ 2m (2ω p + 2m)− 2 EM

(p) ξr
−21 − ( r = ± ∨ p0 = ω p )
u r ( p ) ≡ 2m (2ω p + 2m) EM ( p ) ξ r∗
∗

which are the two eigenstates of the positive energy projector

+
EM ( p ) ur ( p ) = ur ( p ) ( r = +, − ∨ p0 = ω p )

with
u †r ( p ) u s ( p ) = 2 ω p δ rs
In fact we have for instance

u †r ( p) u s ( p) = (2ω p + 2m) −1 ξ r† (m + p̃/M ) (m + p/M ) ξ s

= (2ω p + 2m) −1 ξ r† 2ω 2p + 2m ω p ξ s

= 2ωp 1
2
ξ r† ξ s = 2 ω p δ rs

in which I have made use of the property

ξ r† γM
k 0
γM ξ s = ξ r† γM
k
ξs = 0 ∀ r, s = +, − ∨ k = 1, 2, 3

192
Of course, a completely equivalent construction can be made, hade we
started from the further degenerate pair of the constant eigenspinors
ηr (r = +, −) of the matrix βM .
In conclusion, the general normal mode decomposition of the Majorana
real spinor wave field becomes
Xh i
ψM (x) = a p , r u p , r (x) + a ∗p , r u ∗p , r (x) = ψM
∗
(x)
p,r

with
1
u p , r (x) ≡ [ (2π)3 2 ω p ] − 2 u r ( p) exp{− i ω p t + i p · x}

{a p , r , a q , s } = {a p , r , a ∗q , s } = {a ∗p , r , a ∗q , s } = 0
∀ p , q ∈ R3 ∀ r , s = +, −

It is important to realize that the real Majorana massive spinor wave

field does exhibit two opposite helicity states.

(c) The transition to the quantum theory is performed as usual by the

introduction of the creation annihilation operators a p , r and a †p , s which
satisfy the canonical anticommutation relations

{a p , r , a q , s } = 0 = {a †p , r , a †q , s }

{a p , r , a †q , s } = δ rs δ (p − q)
∀ p , q ∈ R3 ∀ r , s = +, −
so that
Xh i
†
ψM (x) = a p , r u p , r (x) + a †p , r u ∗p , r (x) = ψM (x)
p,r

where the hermitean conjugation refers only to the creation destruction

operators acting on the Fock space. Instead we have the expansion of
the left Majorana adjoint spinor
Xh i
†
† ∗
ψ̄M (x) = a p , r ū p , r (x) + a p , r ū p , r (x) = ψ̄M (x)
p,r

in which
1
ū p , r (x) ≡ [ (2π)3 2 ω p ] − 2 [ u > ∗ 0
r ( p) ] γM exp{ i ω p t − i p · x}

193
Notice that from the normalization
Z
0
( u p , r , u q , s ) = d x ū p , r (x) γM u q , s (x) = δ rs δ(p − q)

we can readily obtain the normal modes expansions of the observables

involving the Majorana massive spinor field. For example the energy
momentum four vector takes the form
Z ↔
i 0
Pµ = d x : ψ̄M (x) γM ∂ µ ψM (x) :
2
X
= p µ a †p , r a p , r ( p0 = ω p )
p,r

Moreover, for example, if ∂ 1 ψM = ∂ 3 ψM = 0 that implies ψM (t, y) =

ψM (t, 0, y, 0) , then we obtain for the energy momentum tensor
↔
2i T 13 (t, y) = ψ̄M (t, y) γM
1
∂ z ψM (t, y) = 0
↔
2i T 31 (t, z) = ψ̄M (t, y) γM
3
∂ x ψM (t, y) = 0
and thereby
µ µ
∂ µ M 13 = ∂ µ S 13 =0 (4.197)
whence it follows that the helicity is conserved in time. After insertion
of the normal modes expansion one gets
Z ∞
h = dy : 21 ψM (t, y) ΣM, 2 ψM (t, y) :
−∞
Z ∞ X h i
† ∗
= dy : a p , r u p , r (t, y) + a p , r u p , r (t, y)
−∞ p,r
X h i
× 1
2
ΣM, 2 a †q , s u ∗q , s (t, y) + a q , s u q , s (t, y) :
q,s

in which we have set

p
p = (0, p, 0) q = (0, q, 0) ωp = p2 + m2
u p , r (t, y) = [ 4πω p ] −1/2 u r (p) exp{− i t ω p + i py} ( r = +, − )
the normalization being now consistent with the occurrence that the
spinor plane waves are independent of the transverse spatial coordinates
x⊥ = (x1 , x3 ) . From the commutation relation
0 2
[ ω p γM − p γM , ΣM, 2 ] = 0

194
 together with the definition

u ± (p) ≡ (2ω p + 2m) −1/2 m + ω p γM

0 2


− p γM ξ±

it can be readily derived that

( ΣM, 2 ∓ 1 ) ξ ± = 0 ⇒ ( ΣM, 2 ∓ 1 ) u ± (p) = 0

which yields in turn

Z ∞ h i
h= 1
2
d p a †p , + a p , + − a †p , − a p , −
−∞

Thus, in general, the 1-particle states a †p , ± | 0 i do represent neutral

Majorana massive particles with energy momentum p µ = (ω p , p) and
positive/negative helicities.
4. Neutrino mass terms of Dirac and Majorana type. Write the
most general mass term involving left and right complex Weyl bispinors
and real self-conjugated Majorana bispinors. Discuss its structure.

Solution. Any Dirac bispinor

 
 ψL (x)
ψ(x) = 


ψR (x)
 

can always be written as the sum of its left–handed and right-handed

Weyl components

ψ(x) = ψ− (x) + ψ+ (x) = (P− + P+ ) ψ(x)

P± ≡ 12 (I ± γ5 )
   
ψL (x)   0
ψ− (x) =  ψ+ (x) = 
  

0 ψR (x)
   

A Dirac–type real mass term in the Lagrangian does indeed connect

the independent left and right components of the same Dirac bispinor

LM
D = − M ψ̄(x) ψ(x)

∗
= − M ψ̄L (x) ψR (x) + ψ̄R (x) ψL (x) = L M


D (4.198)

From a Dirac bispinor ψ(x) one can always build up a pair of self-
conjugated Majorana bispinors ϕ(x) and χ(x) according to

ϕ(x) = ψ− (x) + (ψ− (x)) c = ψ− (x) + ψ+c (x) (4.199)

χ(x) = ψ+ (x) + (ψ+ (x)) c = ψ+ (x) + ψ−c (x) (4.200)

195
where use has been made of the charge conjugation properties (4.196)
connecting left–handed and right–handed Weyl spinor fields. Of course,
by the very construction we have

ϕ(x) = ϕ c (x) χ(x) = χ c (x)

Moreover, the inversion formulæ can be easily obtained through the

application of the chirality or helicity projectors, that yield
 
ϕL (x) 
ψ− (x) = P− ϕ(x) ≡ 

 
0

 
0
ψ+ (x) = P+ χ(x) ≡ 
 

χR (x)
 
 
c 0
ψ+ (x) = P+ ϕ(x) ≡ 
 

− iσ2 ϕL (x)
 
 
c iσ 2 χ R (x)
ψ− (x) = P− χ(x) ≡ 
 

0
 

It follows therefrom that we can build up two Majorana–type mass

terms of the kind
∗
Lm
L = − m− ϕ̄(x) ϕ(x) = (L mL)
− m− ψ̄− (x) ψ−c (x) + ψ̄−c (x) ψ− (x)


= (4.201)
m ∗
LR = − m+ χ̄(x) χ(x) = (L mR)
− m+ ψ̄+ (x) ψ+c (x) + ψ̄+c (x) ψ+ (x)


= (4.202)

which connect the left ϕL (x) and right χR (x) components of the self–
conjugated Majorana spinors. This means that the Majorana–type
mass term does mix the massless chirality and helicity eigenstates
ψ± (x) , in such a manner that the lepton number conservation breaks
down. Notice that the application of the chirality matrix γ5 yields

γ5 ψ(x) = ψ+ (x) − ψ− (x) = ψ 0 (x)

γ5 ϕ(x) = ψ+c (x) − ψ− (x) = ϕ 0 (x)

γ5 χ(x) = ψ+ (x) − ψ−c (x) = χ 0 (x)
so that the the action of the chirality matrix γ5 clearly changes the signs
of all the mass terms (4.198),(4.201) and (4.202). As a consequence,

196
the bispinor fields ψ 0 (x), ϕ 0 (x), χ 0 (x) can be interpreted as the correct
mass eigenstates for the minus values of the masses M, m± because e.g.
0 0
Lm
R = − m+ χ̄(x) χ(x) = m+ χ̄ (x) χ (x)

Whenever both the Dirac–type and Majorana–type mass terms are

simultaneously present we can write
LDM = − M ψ̄L ψR − m− ψ̄− ψ−c − m+ ψ̄+ ψ+c + h.c.
= − 12 M (ϕ̄χ + χ̄ϕ) − m− ϕ̄ ϕ − m+ χ̄ χ
 1
 
m M  ϕ 

− 2
=  ϕ̄ χ̄   (4.203)
 

 1   
M m χ

2 +

the 2 × 2 mass matrix being readily diagonalized to yield the two mass
eigenstates
n 1
o
M± = 12 (m+ + m− ) ± [ (m− − m+ )2 + M 2 ] 2 (I )

corresponding to the pair of Majorana bispinors mass eigenstates


η(x) = cos θ ϕ(x) − sin θ χ(x) = η c (x)
(4.204)
ξ(x) = sin θ ϕ(x) + cos θ χ(x) = ξ c (x)
with
tan 2θ = M/(m− − m+ ) ( II )
One can easily invert the above relations ( I ) and ( II ) to obtain
M = ( M+ − M− ) sin 2θ
m− = M+ cos2 θ + M− sin2 θ
m+ = M+ sin2 θ + M− cos2 θ
Hence the most general mass term (4.203) for a four component fermion
field actually describes two real Majorana bispinors with distinctive
masses. Since the Majorana–type mass terms (4.201) and (4.202) do
violate the conservation of any additive quantum number carried by
the fermion field, such as electric charge, lepton number and so on, it
follows that all elementary charged fermions must have m− = m+ = 0 .
For neutrinos and antineutrinos, instead, a Majorana–type mass term
violates the lepton number by two units because L(ϕ̄ ϕ) = 2 . The
presence of such a kind of Majorana masses leads, for example, to
neutrinoless double beta decays (Z − 1) → (Z + 1) e− e− , or Kaon
decays such as K − → π + e− e− in which ∆ L = − 2 .

197
Chapter 5

The Vector Field

5.1 General Covariant Gauges

The quantum theory of a relativistic massive vector wave field has been first
developed at the very early stage of the quantum field theory by the rumanian
theoretical and mathematical great physicist
Alexandru Proca (Bucarest, 16.10.1897 – Paris, 13.12.1955)
Sur les equations fondamentales des particules elémentaires
Comptes Rendu Acad. Sci. Paris 202 (1936) 1490
Nearly twenty years after, a remarkable generalization of the quantization
procedure for a massive vector relativistic wave field was discovered by the
swiss theoretician
E.C.G. Stueckelberg,
Théorie de la radiation de photons de masse arbitrairement petite
Helv. Phys. Acta 30 (1957) 209-215
whose quantization method is nowadays known as the Stueckelberg trick.
The covariant quantization of the radiation gauge field has been pionereed
long time ago by Gupta and Bleuler

1. Sen N. Gupta, Proc. Phys. Soc. A63 (1950) 681

2. K. Bleuler, Helv. Phys. Acta 23 (1950) 567

and further fully developed by Nakanishi and Lautrup

1. Noboru Nakanishi, Prog. Theor. Phys. 35 (1966) 1111; ibid. 49 (1973)

640; ibid. 52 (1974) 1929; Prog. Theor. Phys. Suppl. No.51 (1972) 1

198
 2. B. Lautrup, Kgl. Danske Videnskab. Selskab. Mat.-fys. Medd. 35
(1967) No. 11, 1

who completely clarified the subject. As we shall see below, there are many
common features shared by the quantum dynamics of the massive and the
massless relativistic vector fields, besides some crucial differences. Needless
to say, the most important property of the massless vector field theory is its
invariance under the so called gauge transformation of the first kind

A µ (x) 7→ A 0µ (x) = A µ (x) + ∂ µ f (x)

where f (x) is an arbitrary real function, its consequence being the exact
masslessness1 of the photon as well as the transversality of its polarizations.
Conversely, this local symmetry is not an invariance of the massive vector
field theory, so that a third longitudinal polarization indeed appears for the
massive vector particles.
In what follow, on the one hand I will attempt to treat contextually the
massive and massless cases though, on the other hand, I will likely to focuss
the key departures between the two items. The main novelty, with respect
to the previously studied scalar and spinor relativistic wave fields, is the
appearance of auxiliary, unphysical ghost field to setup a general covariant
and consistent quantization procedure, as well as the unavoidable presence
of a space of the quantum states – the Fock space – with an indefinite metric.
We start from the classical Lagrange density
1 µν 1
LA,B = − F (x) F µν (x) + m2 A µ (x) A µ (x)
4 2
1
+ A (x) ∂µ B (x) + ξ B 2 (x)
µ
(5.1)
2
where B (x) is an auxiliary unphysical scalar field of canonical engeneering
dimension [ B ] = eV 2 , while the dimensionless parameter ξ ∈ R is named the
gauge fixing parameter, the abelian field strength being as usual F µν (x) =
∂ µ A ν (x)−∂ ν A µ (x) , in such a manner that the action results to be Poincaré
invariant. The variations with respect to the scalar field B and the vector
potential A µ drive to the Euler-Lagrange equations of motion


∂ µ F µν (x) + m2 A ν (x) + ∂ ν B (x) = 0
(5.2)
∂ µ A µ (x) = ξ B (x)
1
The present day experimental limit on the photon mass is m γ < 6 × 10 −17 eV – see
The Review of Particle Physics C. Amsler et al., Physics Letters B667, 1 (2008) and 2009
partial update for 2010.

199
Taking the four divergence of the first equation and using the second equation
we obtain

m2 ∂ · A (x) = −  B (x) = m2 ξ B (x)

which shows that the auxiliary field is a free real scalar field that satisfies
the Klein-Gordon wave equation with a square mass ξ m 2 , which is positive
only for ξ > 0 . This latter feature makes it apparent the unphysical nature
of the auxiliary B-field, for it becomes tachyon-like for negative values of
the gauge fixing parameter. Nonetheless, it will become soon clear later on
why the introduction of the auxiliary and unphysical B-field turns out to
bet very convenient and eventually unavoidable, to the aim of building up
a covariant quantization of the real vector field, especially in the massless
gauge invariant limit m 2 → 0 .
If we write the above equations of motion for the vector potential and the
auxiliary scalar field we have

 + m 2 A µ (x) + ( 1 − ξ ) ∂ µ B (x) = 0


(5.3)
∂ · A (x) = ξ B (x) (5.4)

or even more explicitly

   
2 1
∂ µ ∂ ν A ν (x) = 0


g µν  + m − 1− (5.5)
ξ
2


 + m ξ B (x) = 0 (5.6)
∂ · A (x) = ξ B (x) (5.7)

the very last relation being usually named the subsidiary condition.
The above system (5.2) of field equations, which includes the subsidiary
condition, does nicely simplify for the particular value of the gauge fixing
parameter ξ = 1 : namely,


(  + m2 ) A µ (x) = 0 
∂ · A (x) = B (x) (ξ = 1) (5.8)
(  + m2 ) B (x) = 0


and in the massless limit we come to the d’Alembert wave equation

 A µ (x) = 0 =  B (x) ∂ · A (x) = B (x) (5.9)

200
This especially simple and convenient choice of the gauge fixing parameter is
named the Feynman gauge. If, instead, we put ξ = 0 in the Euler-Lagrange
equations (5.2) we get

 + m2 A µ (x) + ∂ µ B (x) = 0


∂ · A (x) = 0 =  B (x) (5.10)

and in the massless limit

 A µ (x) + ∂ µ B (x) = 0 ∂ · A (x) = 0 =  B (x) (5.11)

This latter choice is known as the Lorentz gauge in classical electrodynamics

or the Landau gauge in quantum electrodynamics, or even the renormalizable
gauge in the massive case. The general case of a finite ξ 6= 0, 1 is called the
general covariant gauge.

The canonical energy momentum tensor is obtained in accordance with

Noether theorem

T µν = A µ ∂ ν B − F µλ ∂ ν A λ − g µν L A , B
= A µ ∂ ν B − F µλ F ν λ − F µλ ∂ λ A ν − g µν L A , B
= − F µλ F ν κ g λ κ − g µν L A , B
− ∂ λ ( F µλ A ν ) + A ν ∂ λ F µλ + A µ ∂ ν B

and using the equation of motion ∂ λ F µ λ = ∂ µ B + m 2 A µ we eventually

obtain
˙ Θ µ ν − ∂ λ ( F µλ A ν )
T µν =
in which the symmetric energy momentum tensor reads
def
Θ µν = Aµ ∂ν B + Aν ∂µB
− g λ ρ F µλ F νρ + m 2 A ν A µ − g µν L A , B (5.12)

It follows therefrom that the conserved Hamilton functional becomes

Z Z n
P0 = d x T 0 0 (t, x) = d x A 0 (t, x) Ḃ (t, x) + F 0k (t, x) Ȧ k (t, x)
1 1
+ 4
F ρσ (t, x) F ρσ (t, x) −m2 A λ (t, x) A λ (t, x)
2
o
− A µ (t, x) ∂ µ B (t, x) − 21 ξ B 2 (t, x)

201
The form of the canonical conjugated momenta can be derived from the
Lagrange density (5.1)


µ 0 for µ = 0
Π µ (x) = δ L / δ Ȧ (x) = (5.13)
F k0 ≡ E k for µ = k = 1, 2, 3
Π (x) = δ L / δ Ḃ (x) = A 0 (x) (5.14)

Now, if we take into account that

F 0k (t, x) Ȧ k (t, x) = F 0k (t, x) F 0k (t, x) + F 0k (t, x) ∂ k A 0 (t, x)

= E k (t, x) E k (t, x) − A 0 (t, x) ∂ k E k (t, x)
h i
+ ∂ k F 0k (t, x) A 0 (t, x)

we can recast the energy, up to an irrelevant spatial divergence, in the form

Z n
P0 =
˙ dx − 1
2
m 2 Π 2 (t, x) − Π (t, x) ∂ k E k (t, x)

+ E k (t, x) E k (t, x) + 14 F jk (t, x) F jk (t, x) − 12 ξ B 2 (t, x)

1
2
o
1 2
+ 2 m A k (t, x) A k (t, x) + A k (t, x) ∂ k B (t, x)

Now we can rewrite the Euler-Lagrange equations (5.2) in the hamiltonian

canonical form, that is

Ḃ (x) = { B (x) , H } = − m 2 Π (x) − ∂ k E k (x)

= − m 2 A 0 (x) − ∂ k F 0k (x) (5.15)
∂ 0 A k (x) = { A k (x) , H } = F 0k (x) + ∂ k Π (x)
= F 0k (x) + ∂ k A 0 (x) (5.16)
Π̇ (x) = { Π (x) , H } = ∂ k A k (x) + ξ B (x) (5.17)
Ḟ 0k (x) = { E k (x) , H }
= ∂ j F jk (x) − m 2 A k (x) − ∂ k B (x) (5.18)

where H ≡ P 0 and I used the canonical Poisson brackets among all the
independent pairs of canonical variables ( A , B ; E , Π ) : namely,

{A  (t, x) , E k (t, y)} = g k δ (x − y)

{ B (t, x) , Π (t, y)} = { B (t, x) , A0 (t, y)} = δ (x − y) (5.19)

202
all the other Poisson brackets being equal to zero. It is very important
to realize that, in the massive case, the hamiltonian functional contains an
unusual negative kinetic term − 12 m 2 Π 2 (t, x) for the auxiliary field.

The canonical total angular momentum density follows from the Noether
theorem and yields

M µρσ ≡ x ρ T µσ − x σ T µρ + S µρσ
def
S µρσ = ( δ L / δ ∂ µ A ν ) (−i ) ( S ρ σ ) ν λ A λ
= − F µρ Aσ + F µσ Aρ (5.20)

Hence we find



M µρσ = x ρ Θ µσ + ∂ λ F λ µ Aσ


− x σ Θ µρ + ∂ λ F λ µ Aρ
+ F µσ Aρ − F µρ Aσ
= xρΘ µσ − x σ Θ µρ 
+ ∂ λ F λ µ ( x ρ Aσ − x σ A ρ ) (5.21)

Since the very last term does not contribute to the continuity equation
∂ µ M µ ρ σ = 0 , we see that the total angular momentum tensor can be always
written in the purely orbital form
Z
M = d x x ρ Θ 0 σ (x) − x σ Θ 0 ρ (x)
ρσ
 

203
5.2 Normal Modes Decomposition
To solve the Euler-Lagrange system of equations (5.2) in the general case, it is
very convenient to decompose the vector potential according to the definition

m −2 ∂ µ B (x)

µ def µ m 6= 0
A (x) = V (x) − (5.22)
− ξ ∂ µ D ∗ B (x) m = 0
where the integro-differential operator D defined by
def 1
−1
D = ∇2 ( x0 ∂ 0 − c ) (5.23)
2
for arbitrary constant c works as an inverse of the d’Alembert wave operator
 in front of any solution of the wave equation : namely,
 D ∗ f (x) = f (x) if  f (x) = 0
as it can be readily checked by direct inspection. Hence, the subsidiary
condition (5.7) entails the transversality condition, i.e.

∂µ Aµ = ξ B ⇔ ∂µ V µ
= 0
so that we eventually obtain from the equations of motion (5.4)


 (  + m2 ) V µ (x) = 0
∂ µ V µ (x) = 0 ( m 6= 0 ) (5.24)
(  + ξ m2 ) B (x) = 0



  V µ (x) + ∂ µ B (x) = 0
∂ µ V µ (x) = 0 (m = 0) (5.25)
 B (x) = 0


It is worthwhile to realize that the field strength does not depend upon the
unphysical auxiliary field B (x), albeit merely on the transverse vector field
V µ (x) since we have F µν (x) = ∂ µ V ν (x) − ∂ ν V µ (x) . This means that we can
also write

 + m2 F µν (x) = 0


or  F µν (x) = 0
The transverse real vector field V µ (x) is also named the Proca vector field
in the massive case, while it is called the tranverse vector potential in the
massless case. We shall now find the general solution of above system of the
equations of motion in both cases, i.e. , the massive and the massless cases.

204
5.2.1 Normal Modes of the Massive Vector Field
Let me first discuss the normal mode decomposition in the massive case. To
this purpose, if we set

Z
−3/2
V µ (x) = (2π) d k Veµ (k) exp {− i k · x} (5.26)

Veµ∗ (k) = Veµ (− k)

then we find

Veµ (k) = f µ (k) δ k 2 − m2

k · f (k) = 0 (5.27)

where fµ (k) are regular functions on the hyperboloid k 2 = m2 , which are

any arbitrary functions, but for the transversality condition k · f (k) = 0 and
the reality condition f µ∗ (k) = f µ (− k).
Next, it is convenient to introduce the three linear polarization real unit
vectors e µr (k) ( r = 1, 2, 3 ) which are dimensionless and determined by the
properties

1/2
k µ e µr (k) = 0 ( r = 1, 2, 3 ) k 0 ≡ ω k = k 2 + m2

− g µν e µr (k) e νs (k) = δ rs ( orthonormality relation ) (5.28)

3
X
e µr (k) e νr (k)
r=1
= − g µν + k µ k ν / k 2
= − g µν + k µ k ν / m 2 ( closure relation ) (5.29)

A suitable explicit choice is provided by


e 0r (k) = 0 
k · e r (k) = 0 for r, s = 1, 2
e ∗r (k) · e s (k) = δ rs


|k| k
b
e 03 (k) = e 3 (k) = ωk
m m

205
in such a manner that we can write the normal mode decomposition of the
classical Proca real vector field in the form

X 
V ν (x) = f k , r u νk , r (x) + f k∗ , r u νk ∗, r (x)

(5.30)
k,r

u νk , r (x) = [ (2π)3 2ω k ] −1/2 e νr (k) exp {− i ω k x0 + i k · x} (5.31)

where we have denoted as usual

Z 3
def
X X
= dk
k,r r=1

Notice that the set of the vector wave functions u νk , r (x) does satisfy the
orthonormality and closure relations

Z ↔
− g λσ d x u σk ∗, s (y) i ∂ 0 u λh , r (x) = δ (h − k) δ rs (5.32)
X
u λk , r (x) u νk ∗, r (y) = i g λ ν + m −2 ∂ xλ ∂ xν D (−) (x − y)


(5.33)
k,r

Next we have

X h i
B (x) = m b k u k (x) + b ∗k u ∗k (x)
k
0 −1/2
u k (x) = [ (2π) 3
2ωk ] exp {− i ω k0 x0 + i k · x}

1/2
ω k0 ≡ k 2 + ξ m2 (5.34)

so that from eq. (5.22) we eventually come to the normal mode decomposition
of the classical real vector potential

X 
A µ (x) = f k , r u µk , r (x) + f k∗ , r u µk ∗, r (x)

k,r
i X 0 µh i
+ k b k u k (x) − b ∗k u ∗k (x)
m k
k 0 µ = ( ω k0 , k ) (5.35)

As a final remark it is very instructive, in the massive case, to rewrite

the classical Lagrange density and the canonical energy momentum tensor as

206
functionals of the transverse Proca vector field V µ (x) and of the unphysical
auxiliary scalar field B (x) . Actually, making use of the decomposition (5.22)
we obtain

LA , B = − 14 F µν (x) F µν (x) + 21 m2 V µ (x) V µ (x)

− ( 1/2m 2 ) ∂ µ B (x) ∂ µ B (x) + 12 ξ B 2 (x)
≡ L V + LB (5.36)
µ
Notice that the lagrangian of the transverse vector field V (x) entails the
Euler-Lagrange equations

∂ µ F µν (x) + m2 V ν (x) = 0
so that the transversality condition

∂ µ V µ (x) = 0
does indeed follow from the equations of motion. The canonical conjugate
momenta are given by


µ 0 for µ = 0
Π (x) = δ LV / δ V̇ µ (x) = k (5.37)
F 0k = E for µ = k = 1, 2, 3
−2
Π (x) = δ LB / δ Ḃ (x) = − m Ḃ (x) (5.38)

and consequently we get the Poisson brackets

{Vk (t, x) , E ` (t, y)} = δk` δ(x − y) (5.39)

{B(t, x) , Π(t, y)} = δ(x − y) (5.40)

all the remaining Poisson brackets being equal to zero. Also the canonical
energy momentum second rank tensor can be conveniently expressed
in terms
of the transverse vector field, up to the irrelevant term ∂ λ A ν F µλ viz.,

Θ µν = 14 δ µν F ρσ F ρσ − 2m2 V λ V λ + m 2 V ν V µ − F µλ F νλ


 
1 µ 1
− 2
∂ B ∂ν B + 2
λ
∂ B ∂ λ B − 2 ξ B δ µν
1 2
(5.41)
m 2m

which appears to be symmetric and conserved. As a consequence, when the

total angular momentum density is expressed as a functional of the transverse
vector field V µ (x) and of the unphysical auxiliary field B (x) , i.e.

207
 M µ ρσ = x ρ Θ µσ − x σ Θ µρ − F µρ V σ
+ F µσ V ρ
(5.42)

we see that the spin angular momentum density third rank tensor

S µρσ = F µσ V ρ − F µρ V σ

does satisfy, thanks to the symmetry of Θ µν , the continuity equation

∂ µ S µ ρσ = 0
so that the spin angular momentum second rank tensor

Z
jk
V j Ek −Ej V k


S = dx (5.43)
Z
S k0 = d x E k (t, x) V 0 (t, x) (5.44)

is indeed conserved. As a final comment, I would like to remark that the

symmetric form (5.41) of the energy momentum tensor does not admit a well
defined massless limit. Thus, a conserved spin angular momentum tensor can
no longer be properly defined in the massless case.

5.2.2 Normal Modes of the Gauge Potential

Let me now turn to the discussion of the massless case. Here, it is very
important to gather that the transversality condition
Z
µ −3/2
0 = ∂ V µ (x) = (2π) d k (− i ) k µ Veµ (k) exp {− i k · x} (5.45)

Veµ∗ (k) = Veµ (− k)

on the light-cone k 2 = 0 is fulfilled by three independent linear polarization
unit real vectors ε µA (k) ( A = 1, 2, L ) which are again dimensionless and
defined by the properties

k µ ε µA (k) = 0 ( A = 1, 2, L ) k0 ≡ ω k = | k |

ε 0A (k) = 0 
k · ε A (k) = 0 for A, B = 1, 2
ε A (k) · ε B (k) = δ AB


208
  
ε µL (k) ≡ k µ / | k | = 1 , k
b

Now, if we introduce a further lightlike polarization vector

p
ε λS (k) = 12 ( | k | , − k ) / | k | ≡ k ∗λ / k · k ∗
 
k ∗λ = 1, −k b εL · εS = 1 (5.46)

then we can write

− g µν ε µA (k) ε νB (k) = η AB ( orthonormality relation ) (5.47)

η AB ε µA (k) ε νB (k) = − g µν ( closure relation ) (5.48)

where  
 1 0 0 0 
 

 0 1 0 0 

η AB =
  ( A, B = 1, 2, L, S )
−1



 0 0 0 


0 0 −1 0
 

According to the conventional wisdom, the labels L and S do correspond

respectively to the longitudinal polarization and the scalar polarization of the
transverse massless vector particles. Moreover, it is convenient to introduce
the transverse projector and the light-cone projector

k λ k ∗ν + k ν k ∗λ
Π λν
⊥ (k) = g − λν
k · k∗
X
= − ε A (k) ε νA (k)
λ
(5.49)
A=1,2

k k ∗ν + k ν k ∗λ
λ
Π λν
∨ (k) =
k · k∗
= ε L (k) ε νS (k) + ε λS (k) ε νL (k)
λ
(5.50)

which satisfy by construction

∗
Π λ⊥ν (k) = Π ν⊥λ (k) g µ ρ Π µ⊥ν (k) Π ρ⊥σ (k) = Π ν⊥σ (k)

∗
Π λ∨ν (k) = Π ν∨λ (k) g µ ρ Π µ∨ ν (k) Π ρ∨σ (k) = Π ν∨σ (k)
tr Π ⊥ (k) = g λ ν Π λ⊥ν (k) = 2 = tr Π ∨ (k) = g λ ν Π λ∨ν (k)
Π µν µν
⊥ (k) + Π ∨ (k) = g
µν

209
as it can be readily verified by direct inspection. As a consequence, for any
given light-like momentum k µ ( k 2 = 0 ) the physical photon polarization
density matrix

def
X
ρ λ⊥ν (k) = − 1
2
ε λA (k) ε νA (k) (5.51)
A=1,2

∗
ρ λ⊥ν (k) = νλ
ρ ⊥ (k) tr ρ ⊥ = 1 (5.52)

will represent a mixed state corresponding to unpolarized monochromatic

photons. In conclusion, we can finally write the normal mode decomposition
of the classical real transverse massless vector field in the form

X
V λ (x) = g k , A u λk , A (x) + g k∗ , A u λk ∗, A (x)

k,A

− ∂ λ D ∗ B (x) (5.53)

u λk , A (x) = [ (2π)3 2 | k | ] −1/2 ε λA (k) exp {− i k · x}

A = 1, 2, L, S k0 = | k |
together with

X 
g k , A u λk , A (x) + g k∗ , A u λk ∗, A (x)

B (x) = ∂ λ
k,A
1 X h i
= k · u k , S (x) g k , S − k · u ∗k , S (x) g k∗ , S (5.54)
i k

The real transverse massless vector field V λ (x) is also named the vector
potential in the Landau gauge ξ = 0 . Then, from eq. (5.22) we eventually
come to the normal mode decomposition of the classical real massless vector
potential

A λ (x) = V λ (x) + ξ ∂ λ D ∗ B (x)

X
g k , A u λk , A (x) + g k∗ , A u λk ∗, A (x)

=
k,A

− ( 1 − ξ ) ∂ λ D ∗ B (x) (5.55)

210
Notice that the complete and orthonormal system of the positive frequency
plane wave solutions u λk , A (x) for the massless gauge vector potential does
satisfy
Z ↔
− g λ ν d x u λh ∗, A (x) i ∂ 0 u νk , B (x) = δ (h − k) η AB (5.56)
X 1 λ ν (−)
η AB u λk , A (x) u νk ∗, B (y) = g 4 (x − y) (5.57)
k
i
X 1 λ ν
u λk , L (x) u νk ∗, S (y) = ∂ ∂ 4 (−) (x − y) (5.58)
k
i x ∗y

where the massless scalar positive frequency distribution is given by

Z
(−) i
d k δ k 2 θ (k 0 ) exp {− i k · x}


4 (x) = 3
(2π)

whereas I have set

k λ k ∗ν
Z
i
∂ xλ ∂ ∗ν y (−)
δ k 2 θ (k 0 ) exp {− i k · x}


4 (x − y) ≡ dk
(2π) 3 k · k∗

We are now ready to perform the general covariant canonical quantization of

the free vector fields, both in the massive and massless cases.

211
5.3 Covariant Canonical Quantization
The manifestly covariant canonical quantization of the massive and massless
real vector free fields can be done in a close analogy with the canonical
quantization of the real scalar free field, just like I have done in Section 3.2.

5.3.1 The Massive Vector Field

Let us first consider the tranverse vector free fields. In the massive case,
the classical Proca vector free field (5.31) is turned into an operator valued
tempered distribution : namely

X h i
V ν (x) = f k , r u νk , r (x) + f k† , r u νk ∗, r (x) (5.59)
k,r

u νk , r (x) = [ (2π)3 2ω k ] −1/2 e µr (k) exp {− i ω k x0 + i k · x} (5.60)

where the creation annihilation operators satisfy the canonical commutation

relations

[ f h , r , f k , s ] = 0 = [ f h† , r , f k† , s ]
[ f h , r , f k† , s ] = δ rs δ (h − k)
As a consequence, the normal mode expansion of the field strength becomes

X n
f k , r ω k u k , r (x) − k u 0k , r (x)
 
E (x) = i
k,r
o
− f k† , r ω k u ∗k , r (x) − k u 0k ∗, r (x)

(5.61)
X
B (x) = i k × u k , r (x) f k , r + c.c. (5.62)
k,r

where the massive electric and massive magnetic fields are defined by

E k = F k0 = − ∂ 0 A k + ∂ k A 0 B k = 21 ε j`k F j`

It is in fact a straightforward exercise to show that, by inserting the normal

mode expansion (5.60) and by making use of the orthonormality relations
among the polarization vectors e µr (k) , the energy momentum operator takes
the expected diagonal form, which corresponds to the sum over an infinite
set of independent linear harmonic oscillators, one for each component of the

212
wave vector k and for each one of the three independent physical polarization
e µr (k) ( r = 1, 2, 3 ) . Actually, if we recast the equations of motion (5.24) for
the Proca wave field in the Maxwell-like form


 ∂ j F jk = Ḟ 0k + m 2 V k ( displacement current )
∂ k F 0k = − m 2 V 0 ( Gauss law ) (5.63)
V̇ 0 = ∂ k V k ( subsidiary condition )


then we get the quite simple expression for the energy operator

Z
P0 = H = d x : Θ 0 0 (t, x) :

= 21 d x : E k (x) E k (x) + 12 F jk (x) F jk (x) :

R

+ 12 m2 d x : V02 (x) + V k (x) V k (x) :

R
↔
˙ 12 d x : F 0k (x) ∂ 0 V k (x) :
R
=
Z ↔
=
˙ d x : 12 V µ (x) ∂ 0 V̇ µ (x) :

where =˙ means, as usual, that I have dropped some spatial divergence term
and I have repeatedly made use of the equations of motion. Now, from the
orthonormality relation (5.28) we can easily recognize that

Z ↔
g λν d x u λk ∗, r (x) ∂ 0 u̇ νh , s (x) = ω k δ rs δ (h − k)
Z ↔
g λν d x u λk , r (x) ∂ 0 u̇ νh , s (x) ≡ 0

and consequently
X
P0 = ω k f k† , r f k , r
k,r
Z
P = d x : E (x) × B (x) + m2 V 0 (x) V (x) :
Z ↔
=
˙ d x : 21 V µ (x) ∂ k V̇ µ (x) :
X
= k f k† , r f k , r (5.64)
k,r

On the other hand, the energy momentum tensor of the auxiliary field is
provided by the B-dependent part of the general expression (5.41) : namely,

213
 Θ µν
B = m
−2
: 1
2
g µν ∂ λ B ∂ λ B − 12 ξ m 2 g µν B 2 − ∂ µ B ∂ ν B :
which drives to the conserved energy momentum operators
Z
1
P0 = d x : B (x) B̈ (x) − Ḃ 2 (x) : (5.65)
2m 2
Z
1
P= 2 d x : Ḃ (x) ∇ B (x) : (5.66)
m
It is important to gather that from the Lagrange density (5.36) it follows
that the conjugate momentum of the auxiliary field gets the wrong sign, that
means

Π (x) = − Ḃ (x)/m 2
From the normal mode decomposition of the auxiliary unphysical field and
its conjugate momentum
X h i
B (x) = m b h u h (x) + b †h u ∗h (x)
h
1 X 0
h
† ∗
i
Π (y) = i ω k b k u k (y) − b k u k (y)
m k
uk (x) = [ (2π)3 2 ω k0 ] −1/2 exp {− i ω k0 x 0 + i k · x}

1/2
ω k0 ≡ k 2 + ξ m2
it is evident that in order to recover the canonical commutation relation
[ B (t, x) , Π (t, y) ] = i δ (x − y)
that corresponds to the classical Poisson bracket (5.40) we have to require
[ b †h , b k ] = δ (h − k) (5.67)
all the other commutators vanishing.
Proof. We find
[ B (t, x) , Π (t, y) ] =
h i
i ω k0 b h u h (t, x) + b †h u ∗h (t, x) , b k u k (t, y) − b †k u ∗k (t, y)
XX
=
h k
n o
i ω k0 [ b h , b k ] u h (t, x) u k (t, y) − [ b h , b †k ] u h (t, x) u ∗k (t, y)
XX
=
h k
n o
i ω k0 [ b †h , b k ] u ∗h (t, x) u k (t, y) − [ b †h , b †k ] u ∗h (t, x) u ∗k (t, y)
XX
+
h k

214
so that, if we assume

[ b h , b †k ] = − δ (h − k) [bh , bk ] = 0 ( ∀ h, k ∈ R3 )

then we obtain
X n o
[ B (t, x) , Π (t, y) ] = i ω k0 u ∗k (t, x) u k (t, y) + u k (t, x) u ∗k (t, y)
k
Z
dk  
=i 3
exp{i k · (x − y)} + exp{i k · (y − x)} = i δ(x − y)
2(2π)

Thus we can eventually write

X
1/2 †
P0 = − k 2 + ξ m2 b k b k ≡ HB (5.68)
k
X
P = k b †k b k
k

From the above expression (5.68) for the conserved hamiltonian of the
auxiliary B−field, as well as from the unconventional nature of the canonical
commutation relations (5.67), it turns out that no physical meaning can
be assigned to the hamiltonian operator of the auxiliary scalar field. As a
matter of fact, for ξ ≥ 0 the hamiltonian operator becomes negative definite
and unbounded from below, while for ξ < 0 we se that at low momenta
k 2 < | ξ | m2 the energy becomes imaginary. In all cases, any physical
interpretation of the hamiltonian operator breaks down.
Moreover, from the conventional definition of the Fock vacuum

fk,r | 0 i = 0 bk | 0 i = 0 ∀ k ∈ R3 r = 1, 2, 3

and the unconventional canonical commutation relations (5.67), it follows

that e.g. the proper 1-particle states of the auxiliary field
Z Z
def †
|bi = d k b̃ (k) b k | 0 i d k | b̃ (k) | 2 = 1

do exhibit negative norm, that is

hb|bi = − 1

Hence the Fock space F of the quantum states for the free massive vector
field in the general covariant gauge is equipped with an indefinite metric,
which means that it contains normalizable states with positive, negative and
null norm. The auxiliary B−field is named a ghost field : its presence entails

215
a hamiltonian operator which is unbounded from below, leading thereby to
the instability, or even imaginary energy eigenvalues, that means metastable
states when exp {− i ω k0 x 0 } is less than one, or even runaway solutions when
exp {− i ω k0 x 0 } becomes very large.

Thus, in order to ensure some meaningful and sound quantum mechanical

interpretation of the free field theory of a massive vector field in the general
covariant gauge, we are necessarily led to select a physical subspace Hphys of
the large Fock space F , in which no quanta of the auxiliary field are allowed.
This can be achieved by imposing the subsidiary condition

B (−) (x) | phys i = 0 ∀ | phys i ∈ Hphys (5.69)

where B (−) (x) denotes, as usual, the positive frequency destruction part of
the auxiliary scalar Klein-Gordon ghost field i.e.
X
B (−) (x) = m b k u k (x)
k

This entails that the vacuum state is physical and cyclic, in such a manner
that all the physical states are generated by the repeted action of the massive
Proca field creation operators f k† , r on the vacuum.
From the normal mode decompositions of the massive Proca real vector
field and of the ghost field, taking the canonical commutation relations into
account, we can readily check that we have

[ V µ (x) , V ν (y) ] = i g µν + m −2 ∂ µ ∂ ν D (x − y ; m)


(5.70)
[ V µ (x) , B (y) ] = 0 (5.71)
[ B (x) , B (y) ] = i m 2 D (x − y ; ξ m) (5.72)

the first commutator being due to the closure relation (5.29). In a quite
similar way it is very easy to check that the following Feynman propagators
actually occur : namely,

h 0 | T (V µ (x) V ν (y)) | 0 i = − g µν + m −2 ∂ µ ∂ ν DF (x − y ; m)

h 0 | T (B (x) B (y)) | 0 i = − m 2 DF (x − y ; ξ m)

Now it is a very simple exercise to obtain the canonical commutator

and the Feynman propagator for the vector potential A µ , taking the basic

216
definition (5.22) into account. Actually, for the Feynman propagator, e.g. ,
we find the Fourier representation

F
D µν ( x ; m , ξ ) = h 0 | T A µ (x) A ν (0) | 0 i (5.73)
Z
i
= d k exp {− i k · x}
(2π) 4
− g µν + m −2 k µ k ν m −2 k µ k ν
 
× − 2
k 2 − m2 + i ε k − ξ m2 + i ε 0
 
exp {− i k · x} (1 − ξ ) kµ kν
Z
i
= dk 2 − g µν + 2
(2π) 4 k − m2 + i ε k − ξ m2 + i ε 0

which is the celebrated Stueckelberg propagator, together with

h 0 | T A µ (x) B (y)| 0 i = ∂ µ DF (x − y ; ξ m)
h 0 | T B (x) B (y) | 0 i = − m 2 DF (x − y ; ξ m)

An important comment is now in order. In the general covariant gauge, for

any finite value ξ ∈ R of the gauge fixing parameter, the leading asymptotic
behaviour for large momenta of the momentum space Feynman propagator
is provided by

 
e F (k ; ξ) = 1 (1 − ξ ) kµ kν
D µν − g µν + 2
k 2 − m2 + i ε k − ξ m2 + i ε 0
∼ k −2 d µν (|kµ | → ∞) (5.74)

where d µν is a constant 4×4 matrix, that is a momentum space isotropic and

scale homogeneous quadratically decreasing law. On the one hand, this naı̈ve
power counting property will become one of the crucial necessary though not
sufficient hypotesis in all the up to date available proofs of the perturbative
order by order renormalizability of any interacting quantum field theory. On
the other hand, the price to be paid is the unavoidable introduction of an
auxiliary unphysical B−field, that must be eventually excluded from the
physical sector of the theory, in such a manner to guarantee the standard
orthodox quantum mechanical interpretation. In the free field theory, the
subsidiary condition (5.69) is what we need to remove the unphysical quanta.
However, it turns out that an extension of the subsidiary condition to the
interacting theories appears to be, in general, highly nontrivial.
Hence, the crucial issue of the decoupling of the auxiliary field from the
physical sector of an interacting theory involving massive vector field will

217
result to be one of the the most selective and severe model building criterion
for a perturbatively renormalizable interacting quantum field theory. In other
words, this means that if we consider the scattering operator S of the theory,
then, for any pair of physical states | phys i and | phys 0 i , the unitarity
relation

X
h phys | S † S | phys 0 i = h phys | S † | ı i h ı | S | phys 0 i = h phys | phys 0 i
ı

must be saturated by a complete orthonormal set of physical intermediate

states or, equivalently, the contribution of the unphysical states must cancel
in the sum over intermediate states. This unitarity criterion will guarantee
the existence of a well-defined unitary restriction of the scattering operator
to the physical subspace H phys ⊂ F of the whole Fock space, thus allowing
a consistent physical interpretation of the theory.
To this concern, it is important to remark that in the limit ξ → ∞ , the
auxiliary B−field just disappears, so that we are left with the free Proca field,
the quanta of which do carry just the three physical polarizations. However,
the ultraviolet leading behaviour of the corresponding Feynman propagator
becomes

F 1  −2

lim D
e µν (k ; ξ) = − g µν + m k µ k ν (5.75)
ξ→∞ k 2 − m2 + i ε
The lack of scale homogeneity and naı̈ve power counting property of this
expression is the evident obstacle that makes the proof of perturbative order
by order renormalizability beyond the present day capabilities. In turn, this
is the ultimate reason why the spontaneous gauge symmetry breaking and the
Higgs mechanism, to provide the masses for the vector fields which mediate
the weak interaction2 , still nowadays appear to be the best buy solution of
the above mentioned renormalizabilty versus unitarity issue.

5.3.2 The Gauge Vector Potential

The quantization of the massless gauge vector potential (5.55) is obtained
from the canonical commutation relations
h i
†
g h , A , g k , B = δ ( h − k ) η AB (5.76)
2
The weak interaction is mediated by two charged complex vector fields W ± with
mass M ± = 80.425 ± 0.038 GeV and a neutral real vector field Z 0 with a mass M 0 =
91.1876 ± 0.0021 GeV.

218
all other commutators vanishing. As a matter of fact, from the normal mode
expansion (5.55) and the orthonormality and closure relations (5.56–5.58),
we readily obtain the canonical commutation relations

A λ (x) , A ν (y) = i g λ ν 4 (x − y) + i ( 1 − ξ ) ∂ xλ ∂ yν E (x − y)
 
(5.77)
[ B (x) , A ν (y) ] = i ∂ xν 4 (x − y) (5.78)
[ B (x) , B (y) ] = 0 (5.79)

where the massless Pauli-Jordan real and odd distribution is as usual

4 (x) = 4 (−) (x) + 4 (+) (x) = lim D (x ; m)

m→0
Z
1 d k
4 (±) (x) ≡ ± δ k 2 θ (k 0 ) exp {± i k · x}


i (2π) 3
lim 4 (x) = 0 lim ∂ 0 4 (x) = δ (x)
x0 →0 x0 →0
∗
4 (x) = 4 (x) = − 4 (− x)

whereas E (x) is named the massless dipole ghost invariant distribution and
is defined by the property

 E (x) = 4 (x) (5.80)

An explicit representation is provided by

1
−1
E (x) = ∇2 (x 0 ∂ 0 − 1) 4 (x)
2
∂
= − lim D (x ; m) (5.81)
m→0 ∂ m2
and it is an easy task to prove the following useful formula

∂ xµ ∂ xν E (x − y) = ∂ xµ ∂ ∗ν x + ∂ yν ∂ ∗µy 4 (x − y)

It is a nice exercise to verify the compatibility between the canonical

commutation reations (5.79) and the equations of motion

   
1
g µν  − 1 − ∂ µ ∂ ν A ν (x) = 0 (5.82)
ξ
 B (x) = 0 (5.83)
∂ · A (x) = ξ B (x) (5.84)

Moreover we find

219
 F λρ (x) , A ν (y) = g ν ρ i ∂ xλ − g λ ν i ∂ xρ 4 (x − y)
 

(5.85)
that implies
Ak (t, x) , E ` (t, y) = i g k ` δ (x − y)
 

[ B (t, x) , Π (t, y) ] = i δ (x − y) = [ B (t, x) , A 0 (t, y) ]

all the remaining equal-time commutation relations being equal to zero, in full
agreement with the Bohr, Sommerfeld and Dirac correspondence principle
and the classical Poisson brackets (5.19).
A further very important canonical commutation relation is

F ρ λ (x) , B (y) = 0
 
(5.86)
which tells us that a physical local operator such as the electromagnetic field
strength tensor indeed commutes with the unphysical auxiliary field for any
spacetime separation. Actually, a weaker condition will specify the concept
of gauge invariance in the quantum field theory of the electromagnetism as I
will show in the sequel.

Let us now set up the Hilbert space of the physical states. To this concern,
I will first define the Fock space F in the conventional way starting from the
cyclic vacuum state
g k , A | 0 i = 0 = h 0 | g †k , A ∀ k ∈ R 3 , A = 1, 2, L, S (5.87)
a generic polarized N −photon energy momentum eigenstate being given by
N
def
Y
| k1 A1 k2 A2 . . . kN AN i = g k†  , A | 0 i
=1

It is very important to realize that the Fock space F for the massless gauge
vector particles is of an indefinite metric. As a matter of fact, the inner
product 4 × 4 real symmetric matrix η ≡ k η kAB ( A, B = 1, 2, L, S ) does
satisfy
η2 = I tr η = 2
which means that it admits three positive eigenvalues equal to +1 and one
negative eigenvalue equal to −1 . Hence, negative norm states do indeed exist,
for example
1  
√ g †k , L + g †k , S | 0 i
2

220
as well as null norm states just like

g †k , L | 0 i g †k , S | 0 i

Actually we readily find

1
2
h0| ( g h , L + g h , S ) ( g †k , L + g †k , S ) | 0 i = − δ (h − k) (5.88)

Then, an arbitrary physical state | phys i ∈ H phys ⊂ F will be defined by the

auxiliary condition

B (−) (x) | phys i = 0 (5.89)

where the positive frequency part of the auxiliary B−field is given by the
normal mode expansion (5.54)
X
B (−) (x) = i k · u k , S (x) g k , S ( k0 = | k | ) (5.90)
k

To understand the meaning of the auxiliary condition (5.89) consider the

polarized 1-photon energy momentum eigenstates

| k A i = g †k , A | 0 i h B h | k A i = η AB δ ( h − k )

From the canonical commutation relations (5.76) it follows that the 1-photon
states with transverse polarizations are physical

B (−) (x) | k A i = 0 ∀ k ∈ R3 ∀ A = 1, 2

as well as the scalar photon 1-particle states

B (−) (x) | k S i = 0 ∀ k ∈ R3

Notice, however, that for any given wave packet ϕ (k) normalized to one, i.e.
Z
d k | ϕ (k) | 2 = 1

we find for A, B = 1, 2
Z Z
h ϕB |ϕA i = dk d h ϕ (h) ϕ ∗ (k) h B k | h A i = δ AB
Z Z
h ϕS |ϕS i = dk d h ϕ (h) ϕ ∗ (k) h S k | h S i = 0
Z Z
h ϕS |ϕA i = dk d h ϕ (h) ϕ ∗ (k) h S k | h A i = 0

221
From the above table of scalar products it follows that the 1-photon states
with transverse polarizations are physical states with positive norm and are
always orthogonal to the 1-particle scalar states, which are the physical states
with zero norm. Hence, the 1-particle physical Hilbert space
def
| k A i | k ∈ R 3 A = 1, 2, S

H 1 , phys = V 1 V1 ≡

is a Hilbert space with a positive semidefinite metric. It is clear that the very
same construction can be generalized in a straightforward way to define the
N -particle completely symmetric physical Hilbert space – the closure of the
symmetric product of 1–particle physical Hilbert spaces
s s s s
H N , phys ≡ VN VN = { H 1 , phys ⊗ H 1 , phys ⊗ . . . ⊗ H 1 , phys } = H 1⊗,nphys
| {z }
N times

so that
∞
M s
H phys ≡ C ⊕ H 1 , phys ⊕ H 2 , phys ⊕ . . . ⊕ H N , phys ⊕ . . . = H 1⊗,nphys
n=1

By their very construction, we see that the covariant physical photon states
are equivalence classes of positive norm photon states with only transverse
polarizations, up to the addition of any number of zero norm scalar photons.
This fact represents the quantum mechanical counterpart of the classical
gauge transformations of the second kind. As a matter of fact, in classical
electrodynamics the invariant Lorentz gauge condition ∂ · A (x) = 0 does
not fix univocally the gauge potential, for a gauge transformation A 0µ (x) =
A µ (x) + ∂ µ f (x) with f (x) satisfying the d’Alembert wave equation, is still
compatible with the Lorentz condition. Hence, an equivalence class of gauge
potentials obeying the invariant Lorentz condition indeed exists, what is
known as classical gauge invariance of the second kind. Notice that such an
invariance is no longer there for a non-covariant gauge condition, e.g. the
Coulomb gauge ∇ · A = 0 .

As a final remark, I’d like to stress that the notion of gauge invariant
local observable in the covariant quantum theory of the free radiation field is
as follows : a gauge invariant local observable O (x) is a self-adjoint operator
that maps the physical Hilbert space onto itself, i.e.

O (x) | phys i ∈ H phys ∀ | phys i ∈ H phys O (x) = O † (x) (5.91)

222
which implies

| phys i ∈ H phys ⇔ B (−) (x) | phys i = 0

B (−) (x) O (y) | phys i = B (−) (x) , O (y) | phys i

 

∝ B (−) (x) | phys i = 0

It follows therefrom that the Maxwell field equation as well as the usual
form of the energy momentum tensor for the radiation field hold true solely
in a weak sense, i.e. as matrix elements between physical states : namely

h phys 0 | ∂ µ F µ ν (x) + ∂ ν B (x) | phys i =

h phys 0 | ∂ µ F µ ν (x) | phys i = 0
h phys 0 | Θ µ ν (x) | phys i =
h phys 0 | 14 g µν F ρ σ (x) F ρ σ (x) − g ρ σ F µ ρ (x) F ν σ (x) | phys i
∀ | phys i , | phys 0 i ∈ H phys

In respect to the above definition, the canonical energy momentum tensor

of the massless vector gauge field theory is neither symmetric nor observable,
as it appears to be evident from its expression

T µ ν = : A µ ∂ ν B − F µλ ∂ ν A λ (5.92)
+ g µν 14 F ρσ F ρσ − A λ ∂ λ B − 12 ξ B 2 :

because, from the canonical commutation relations (5.79), we immediately

get

[ B (x) , T µ ν (y) ] = − i ∂ µ , x 4 (x − y) ∂ ν B (y)

+ i g µν ∂ xλ 4 (x − y) ∂ λ B (y)
− i F µλ (y) ∂ ν , x ∂ xλ 4 (x − y)

which does not fulfill the criterion (5.91) owing to the presence of the very
last term. Conversely, the symmetric energy momentum tensor

def
Θ µν = : A µ ∂ ν B + A ν ∂ µ B − g λ ρ F µλ F νρ − g µν L A , B :
= : A µ ∂ ν B + A ν ∂ µ B − g λ ρ F µλ F νρ
g µν 14 F ρσ F ρσ − A λ ∂ λ B − 12 ξ B 2 :


+

223
as well as, a fortiori, the energy momentum four vector do indeed satisfy
the requirements (5.91), i.e. they are observable in the quantum mechanical
sense, since we have

[ B (x) , Θ µ ν (y) ] = − i ∂ µ , x 4 (x − y) ∂ ν B (y)

− i ∂ ν , x 4 (x − y) ∂ µ B (y)
+ i g µν ∂ xλ 4 (x − y) ∂ λ B (y)

Θ µ ν (y) = Θ ν µ (y) Θ µ ν (y) = Θ †µ ν (y)

224
References

1. N.N. Bogoliubov and D.V. Shirkov (1959)

Introduction to the Theory of Quantized Fields
Interscience Publishers, New York.

2. Taichiro Kugo and Izumi Ojima (1979)

Local Covariant Operator Formalism of Non-Abelian Gauge Theories
and Quark Confinement Problem
Supplement of the Progress of Theoretical Physics, No. 66, pp. 1-130.

3. Claude Itzykson and Jean-Bernard Zuber (1980)

Quantum Field Theory
McGraw-Hill, New York.

225
5.4 Problems
1. Covariance of the vector field. Find the transformation laws of the
quantized vector wave field under the Poincaré group.

Solution. We shall treat in detail the massless case, the generalization

to the massive case being straightforward. We have to recall some
standard definitions : namely,
Z Z Z
def dk 1 2


4k = = d k θ(k0 ) δ k (5.93)
(2π) 3 2 | k | (2π) 3
def 1
|kAi = [ (2π) 3 2 | k | ] 2 g †k , A | 0 i = g †A (k) | 0 i (5.94)

i) covariant 1-particle states for the massless vector field

n 1
o
3 † 3
| k A i = [ (2π) 2 | k | ] g k , A | 0 i | k ∈ R
2 (5.95)

ii) orthonormality and closure relations

h h A | k B i = (2π) 3 2 | k | δ (h − k) η AB
XZ
4 k | k A i h k A | = I1
A

iii) gauge potential covariant normal mode expansion

XZ h
†
i
λ λ − i kx λ∗ i kx
A (x) = 4 k εA (k) g A (k) e + εA (k) g A (k) e
A

iv) wave functions

u λ (x) ≡ h 0 | A λ (x) | k A i = εAλ (k) exp {− i k · x} ( k0 = | k | )

Zk ↔
d x u νh ∗ (x) i ∂ 0 u λk (x) = 2 | k | δ ( h − k )

For each element of the restricted Lorentz group, which is univocally

specified by the six canonical coordinates ω µν = (α, β) , there will
correspond a unitary operator U (ω) : F 1 → F 1 so that

U (ω) | k A i = | Λ k A i
h A k | U † (ω) = h A Λ k |
hA Λh|Λk B i = hAh|kB i
= h A h | U † (ω) U (ω) k B i
= η AB δ ( h − k ) (2π) 3 2 | k |

226
 where we obviously understand e.g.
1
| A Λ k i = | A k 0 i = [ (2π) 3 2 k 00 ] 2 g k† 0 , A | 0 i = g †A (Λ k) | 0 i
k µ0 = Λ µν k ν k 00 = | k 0 |
g µν k µ0 k ν0 = k 2 = 0

Under a Lorentz transformation, the creation annihilation operators do

transform according to the law

U (ω) g †A (k) U † (ω) = g †A (Λ k)

U (ω) g A (k) U † (ω) = g A (Λ k)

under the assumption that the vacuum is invariant i.e. U (ω) | 0 i = | 0 i .

As a consequence
def
A 0 λ (x) = U (ω) A λ (x) U † (ω)
XZ h i
= 4 k εAλ (k) g A (Λ k) e − i k x + εAλ ∗ (k) g †A (Λ k) e i k x
A
XZ h
0
= Λ λ
σ 4k εA0 σ (k 0 ) g A (k 0 ) exp {− i k 0 · x 0 }
A
i
+ εA0 σ ∗ (k 0 ) g A† (k 0 ) exp { i k 0 · x 0 }
= Λ λσ A σ ( Λ x )

in which I have used the standard four vector transformation rule (2.21)

εAλ (k) = Λ λ σ εA0 σ (k 0 )

the new equivalent basis of the polarization vectors { ε 0Aµ (k 0 ) } still

obeying the orthonormality and closure relations

− g µν ε 0Aµ ∗ (k 0 ) ε 0Bν (k 0 ) = η AB

η AB ε 0Aµ ∗ (k 0 ) ε 0Bν (k 0 ) = − g µν

2. The generating functional. Construct the generating functional for

the massive and massless real vector free field theories.

Solution. This can be done by a straightforward generalization of the

real scalar field case. In particular, we can conveniently define
def
u A (x) = ( A µ (x) , B (x) ) A = (µ , •)

227
 def
j A (x) = ( J µ (x) , K (x) )
Z Z h i
d x u (x) j A (x) = d x A µ (x) J µ (x) + B (x) K (x)
A

so that we have a 5-dimensional diagonal metric tensor

g AB = g AB = diag (+1, −1, −1, −1, +1)
The classical action can be written in the form
Z
1 (x)
S0 [ u ] = − d x u A (x) K AB u B (x)
2
with
K µ( xν ) = − g µ ν  + m 2 + ∂ µ ∂ ν K µ( x•) = ∂ µ K •( x• ) = ξ

(x)
The kinetic operator K AB can be univocally inverted by means of the
causal (+ iε)−prescription leading to the causal Green’s function
Z
(c ) d k e (c )
D AB ( x − y ; m , ξ ) = D ( k ; m , ξ ) exp {− i k · (x − y)}
(2π) 4 AB
 
F i (1 − ξ ) kµ kν
D µν ( k ; m , ξ ) = 2
e − g µν + 2
k − m2 + i ε k − ξ m2 + i ε 0
e F (k ; m , ξ ) = kν
D ν•
k 2 − ξ m2 + i ε 0
e F (k ; m , ξ ) = − i m2
D ••
k 2 − ξ m2 + i ε 0
In fact
g ν ρ K µ( xν ) D ρFσ (x) + K µ( x•) D σF • (x) = − i g µ σ δ (x)
g µ ν K µ( x•) D νF • (x) + ξ D •F• (x) = − i δ (x)
This means that we can write
(x) (c)
g BC K AB D CD (x) = − i g AD δ (x)
and thereby
  Z 
A
Z 0[ j ] = T exp i d x u A (x) j (x)
0
n o
1
R R (c) A B
= exp − 2 d x d y D AB (x − y) j (x) j (y)
Z  Z 
A
= N D u A exp i S 0 [ u ] + i d x u (x) j A (x)

228
with Z
Z 0[ 0 ] = N D u A exp {i S 0 [ u ]} = 1

Turning to the euclidean formulation in the usual way

x4 = i x 0 A 4 (xE ) = i A 0 (− i x4 , x)
− g µ ν x µ x ν = xE2 − g µ ν A µ (x) A ν (x) = A E µ (xE ) A E µ (xE )
et cetera, we find
Z Z n o
(0) (0)
ZE [ 0 ] =N D AE µ D BE exp − SE [ A E µ , BE ] = 1

where
(0)
SE [ A E µ , BE ] = 12 A E µ [ ( m 2 − ∂ E2 ) δ µ ν + ∂ E µ ∂ E ν ] A E ν
+ A E µ ∂ E µ BE − 12 ξ BE2
After rescaling with an arbitrary mass scale µ
A E µ 7→ A E0 µ = µ A E µ BE 7→ BE0 = µ 2 BE
we come to
(0)
ZE [ 0 ] = N 0 det k K E k
 
( m2 − ∂λ ∂λ ) δ µν + ∂µ ∂ν ∂ρ
K E = µ −2

 

∂ρ − ξ µ2
 

Hence, from the ζ−function regularization technique we obtain for

<e ξ µ 2 < 4m2
Z
−s 2s dk 2 2 2 −s


Tr K E = V µ 4m + 3k − ξ µ
(2π) 4
Z ∞
V µ 2s s−3
 2 2


= d t t exp −t 4m − ξ µ
144π 2 Γ(s) 0
2  2 −s
V m2 ξ µ2

4m Γ(s − 2)
= 1 − − ξ
9π 2 4m2 µ2 Γ(s)
2 −s
V m2 ξ µ2 4m2
   
= 1 − − ξ ( s2 − 3s + 2 ) −1
9π 2 4m2 µ2
Hence
d
Tr K E− s = ( s 2 − 3s + 2 ) −1 Tr K E− s
ds
h  4m2 i
× 3 − 2s − ( s 2 − 3s + 2 ) ln − ξ
µ2

229
and thereby
2 
V m2 ξ µ2 4m2
   
1 3
det k K E k= 1− ln −ξ −
9π 2 4m2 2 µ2 4
(0)
so that the normalization condition Z E [ 0 ] = 1 yields
( 2  )
2 2
  2 
V m ξ µ 1 4m 3
N 0 ≡ exp 1− ln −ξ −
18π 2 4m2 2 µ2 4

The transition to the Minkowski spacetime can be immediately done

by simply replacing Veuclidean ↔ i Vminkowskian which leads to the
final result
  Z 
A
Z 0[ j ] = T exp i d x u A (x) j (x)
0
n o
1
R R (c)
= exp − 2 d x d y D AB (x − y) j A (x) j B (y)
Z  Z 
A
= N D u A exp i S 0 [ u ] + i d x u (x) j A (x)
 1 / 2
(x)
N = constant × det K AB
( 2  )
i V m2 ξ µ2
  2 
def 1 4m 3
= exp 1− ln −ξ −
18π 2 4m2 2 µ2 4

230
Bibliography

[1] John David Jackson, Classical electrodynamics, John Wiley & Sons,
New York, 1962.

[2] L.D. Landau, E.M. Lifšits, Teoria dei campi, Editori Riuniti/Edizioni
Mir, Roma, 1976.

[3] Roberto Soldati, Note on line del corso di meccanica statistica,

http://www.robertosoldati.com, 2007.

[4] E. Merzbacher, Quantum Mechanics. Second edition, John Wiley &

Sons, New York, 1970 ;
L.D. Landau and E.M. Lifšits, Meccanica Quantistica. Teoria non
relativistica, Editori Riuniti/Edizioni Mir, Roma, 1976 ;
C. Cohen–Tannoudji, B. Diu and F. Laloë, Mécanique quantique I,
Hermann, Paris, 1998.

[5] G.Ya. Lyubarskii, The Application of Group Theory in Physics,

Pergamon Press, Oxford, 1960 ;
Mark Naı̈mark and A. Stern, Théorie des représentations des groupes,
Editions Mir, Moscou, 1979.

[6] Eugene Paul Wigner, Gruppentheorie und ihre Anwendung auf die
Quantenmechanik der Atomspektrum, Fredrick Vieweg und Sohn,
Braunschweig, Deutschland, 1931, pp. 251–254 ; Group Theory and
its Application to the Quantum Theory of Atomic Spectra, Academic
Press Inc., New York, 1959, pp. 233–236.

[7] Valentine Bargmann and E.P. Wigner, Group theoretical discussion of

relativistic wave equations, Proceedings of the National Academy of
Sciences, Vol. 34, No. 5, 211–223 (May 1948); Relativistic Invariance
and Quantum Phenomena, Review of Modern Physics 29, 255–268
(1957).

231
 [8] N.N. Bogoliubov and D.V. Shirkov, Introduction to the Theory of
Quantized Fields, Interscience Publishers, New York, 1959.

[9] V.B. Berestetskij, E.M. Lifšits and L.P. Pitaevskij, Teoria quantistica
relativistica, Editori Riuniti/Edizioni Mir, Roma, 1978.

[10] Sidney R. Coleman, The Uses of Instantons, Proceedings of the 1977

International School of Subnuclear Physics, Erice, Antonino Zichichi
Editor, Academic Press, New York, 1979.

[11] C. Itzykson and J.-B. Zuber, Quantum Field Theory, McGraw-Hill,

New York, 1980.

[12] Pierre Ramond, Field Theory: A Modern Primer, Benjamin, Reading,

Massachusetts, 1981.

[13] Ian J.R. Aitchison and Anthony J.G. Hey, Gauge theories in particle
physics. A practical introduction, Adam Hilger, Bristol, 1982.

[14] Ta-Pei Cheng and Ling-Fong Li, Gauge theory of elementary particle
physics, Oxford University Press, New York, 1984.

[15] B. de Wit, J. Smith, Field Theory in Particle Physics – Volume 1,

Elsevier Science Publisher, Amsterdam, 1986.

[16] Stefan Pokorski, Gauge field theories, Cambridge University Press,

Cambridge (UK), 1987.

[17] R.J. Rivers, Path integral methods in quantum field theory, Cambridge
University Press, Cambridge (UK), 1987.

[18] Michel Le Bellac, Des phénomènes critiques aux champs de jauge. Une
introduction aux méthodes et aux applications de la théorie quantique
des champs, InterEditions/Editions du CNRS, 1988.

[19] Lowell S. Brown, Quantum Field Theory, Cambridge University

Press, Cambridge, United Kingdom, 1992.

[20] M.E. Peskin and D.V. Schroeder, An Introduction to Quantum Field

Theory, Perseus Books, Reading, Massachusetts, 1995.

[21] Voja Radovanović, Problem Book in Quantum Field Theory, Second

Edition, Springer Verlag, Berlin-Heidelberg, 2008.

232
[22] M. Abramowitz, I.A. Stegun, Handbook of Mathematical Functions
with Formulas, Graphs and Mathematical Tables, Dover, New York,
1978.

[23] I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series, and Products,
Fifth Edition, Alan Jeffrey Editor, Academic Press, San Diego, 1996.

[24] Hendrik Brugt Gerhard Casimir, Zur Intensität der Streustrahlung

gebundener Elektronen, Helvetica Physica Acta 6, 287–304 (1933).

[25] I.E. Tamm, On Free Electron Interaction with Radiation in the Dirac
Theory of the Electron and in Quantum electrodynamics, Zeitschrift
der Physik 62, 545 (1930).

[26] Vladimir A. Fock, Konfigurationsraum und zweite quantelung,

Zeitschrift der Physik A 75, 622–647 (1932).

233