Chapter 1

Classical Field Theory

1.1 Trasformation Properties

In Poincaré invariant theories fields can be classified in the way they transform under the Lorentz
generators.
When performing a transformation we want the field to change in order to implement the
transformation. For example under a space-time shift x → x + ξ a field localized around some
point x0 , should transform into a field localized around x0 + ξ, i.e.

Φ(x) = cδ(x − x0 ) → cδ(x − x0 − ξ) = Φ(x − ξ) , (1.1)

analogously a field describing a plane wave propagating with momentum p should transform
into a field propagating a plane wave with momentum Λp under a Lorentz transformation, i.e.

Φp (x) = c sin(x · p) → ΦΛp = c sin(x · Λp) = Φ(Λ−1 x) . (1.2)

A field is called scalar if it transforms simply as

φ(x) → φ0 (x) = φ(Λ−1 x) . (1.3)

A vector field instead transforms as

V µ (x) → V 0µ (x) = Λµ ν V ν (Λ−1 x) . (1.4)

Analogously for other representation of the Lorentz group (spinorial and tensorial representation
fields).

Λp

x0 x0 + ξ p

1
2 CHAPTER 1. CLASSICAL FIELD THEORY

All fields transform the same way under translations:

φr (x) → φ0r (x) = φr (x − ξ) , (1.5)

where the index r can stand for any internal and/or space-time index.
It follows that the product of scalar fields computed at the same point is also a scalar field,
while a generic product of fields computed at the same point will be a generic tensor field
transforming as the product of the representation of the constituent fields. Notice that we are
only acting on the fields, without touching the coordinates. Therefore the product of a scalar
field with a function of the coordinates will not be a proper covariant field, e.g.

K(x) ≡ x2 φ(x) → x2 φ0 (x) = x2 φ(x − ξ) 6= K(x − ξ) . (1.6)

Also the derivative of a scalar field is a vector field, indeed:

∂
V µ (x) ≡ g µν ∂ν φ(x) → g µν ∂ν φ(Λ−1 x) =g µν Λ−1 ν ρ φ(Λ−1 x)
∂(Λ−1 x)ρ
=g µν Λ−1 ν ρ gρσ V σ (Λ−1 x) = Λµ ν V µ (Λ−1 x) . (1.7)

Analogously when applied to fields in other representations or when multiple derivatives are
applied to the same field. Hence a product of fields and derivatives where all the space time
indices are saturated is a scalar field.
Note that the four-volume integral of a scalar field is Poincaré invariant
Z Z Z Z
−1
d x φ(x) → d x φ(Λ x − ξ) = d x| det Λ|φ(x) = d4 x φ(x)
4 4 4
(1.8)

where we assumed that integral extends to infinity (both in space and time) and is well defined.

1.2 Field equations and Lagrangian

Analogously to any system of particles in classical mechanics fields are subject to equations that
determine their evolution. A field theory is said to be Poincaré invariant if the field equations
are covariant under the transformation, i.e. all the terms transform the same way. Covariance
guarantees that a Poincaré transformed solution is still solution of the same equations of motion,
i.e. physics is invariant under such transformations. For example if F µν (x) is solution of the
Maxwell equations then the Lorentz transformed solution F 0µν (x) will also be:

∂ ∂
∂ν F 0µν = Λµ ρ Λν σ ∂ν F ρσ (Λ−1 x) = Λµ ρ F ρν (Λ−1 x) = Λµ ρ F ρν (x0 ) = 0 . (1.9)
∂(Λ−1 x)ν ∂x0ν

Typical equations of motion for scalar field are, for example,

φ(x) = 0 , (1.10)
2
( + m )φ(x) = 0 , (1.11)
φ(x) = j(x) , (1.12)
φ(x) = V [φ(x)] , (1.13)
1.2. FIELD EQUATIONS AND LAGRANGIAN 3

describing respectively a free massless scalar field, a free massive scalar field, a free massless
scalar field coupled to a source, a self interacting scalar field with potential V (φ).
Note that when interactions are present (products of fields in the equations of motion) the
fields must be computed at the same point. If interactions between different space-time points are
present (e.g. φ(x)φ(y)) they would in general mediate instantaneous interactions with potential
catastrophic consequences for causality. Locality of interactions will thus always be assumed in
what follows.
For reasons that will become clear later we will be mainly interested to field equations
with up to 2 derivatives per fields. Such equations involve up to 2 time derivative. If boundary
conditions for the fields and their first time derivatives are given at some time t0 (or equivalently
on a space-like surface)
φr (t0 , ~x) = φ(0)
r (~x) , φ̇r (t0 , ~x) = φ̇(0)
r (~x) , (1.14)
the equations admit a unique solution.
The field on each point ~xi in space can be thought as an independent degree of freedom
qi (t) = φ(t, ~xi ), field equations thus correspond to the usual equations of motion for an infinite
system of degrees of freedom.
We can thus associate a Lagrangian and an action to the system:
Z Z X Z
S = dtL(qi (t), q̇i (t)) = dt ~
Li (qi (t), qi+1 (t), q̇i (t)) = dt d3 x L(φ(x), ∇φ(x), φ̇(x))
i
(1.15)
where the second step follows from locality1 and in the third we introduced the Lagrangian
density L defined as Z
L = d3 xL . (1.16)

Being the action S dimensionless in natural units, it follows that L has dimensions 4 in energy.
Field equations can now be derived by extremizing the action S[φ(x)]. Poincaré invariance
requires that if φ̄(x) is a solution of the equations of motion, the same should be true for the
transformed φ̄0 (x). It follows that if φ̄(x) extremizes the action, δS[φ̄(x)] = 0, so should the
transformed solution on the transformed action, δS 0 [φ̄0 (x)] = 0. A sufficient condition for this
to happen is that the action be invariant under Poincaré transformations S[φ(x)] = S[φ0 (x)].
Given that the action is the space-time integral of the Lagrangian density, this corresponds to
require L to be a scalar. In particular L must be a scalar function of
~
L(φ(x), ∇φ(x), φ̇(x)) = L(φ(x), ∂µ φ(x)) . (1.17)
We are now ready to derive the equation of motion from our action
Z
S[φr (x)] = d4 x L(φr (x), ∂µ φr (x)) (1.18)
Ω
where the index r represents for short any field or space-time index. To make the integral finite
we took a finite 4d box Ω with −L < xi < L and t1 < t < t2 , and impose some boundary
conditions on the space boundaries:
φr (xi = −L) = φr (xi = L) or φr (xi = ±L) = const . (1.19)
1
More precisely locality requires that the Lagrangians Li depend only on the variables (qj (t), q̇j (t)) with j in
a finite neighborhood of i.
4 CHAPTER 1. CLASSICAL FIELD THEORY

Because of locality the boundary effects due to the choice of boundary conditions should vanish
in the limit L → ∞.
In analogy to classical mechanics with a finite number of degrees of freedom we require
that the action is extremized around the solution of the equation of motion φ(x) for arbitrary
fluctuations δφ(x) that vanish at t = t1 and t = t2 :

δS[φr (x)] = S[φr (x) + δφr (x)] − S[φr (x)] = 0 , δφr (x)|t=t1,2 = 0 , (1.20)

the boundary conditions (1.19) give respectively the following extra constraints

δφr (xi = −L) = δφr (xi = L) or δφr (xi = ±L) = 0 . (1.21)

We thus get
Z  
4 δL δL
0 = δS = d x δφr + δ∂µ φr
Ω δφr δ∂µ φr
Z   Z  
4 δL δL 4 δL
= d x − ∂µ δφr + d x ∂µ δφr (1.22)
Ω δφr δ∂µ φr Ω δ∂µ φr

The last term is a total derivative and produce boundary terms

Z   Z t2 Z  
4 δL 3 δL ~ δL
d x ∂µ δφr = d x δφr + dtdΣ δφ = 0 (1.23)
Ω δ∂µ φr δ φ̇r t1 ∂Ω ~ r r
δ ∇φ

that vanish because of the boundary conditions (we used Stokes’ theorem in the last term).
Exploiting the arbitrariness of δφr inside the 4-volume Ω eq. (1.22) implies the equation of
motion
δL δL
∂µ − = 0. (1.24)
δ∂µ φr δφr

1.2.1 Examples
Free real scalar field
Consider the most generic local Lagrangian density of a single real scalar field, quadratic in the
field and with up to one derivative per field

L = A ∂µ φ(x)∂ µ φ(x) + Bφ(x)2 , (1.25)

We can always rescale the action and thus the Lagrangian by a constant without changing
the equation of motion. We choose A = 21 which is called the canonical normalization. The
coefficient in front of φ(x)2 now becomes B/2A, which we rename −m2 /2, so that
1 1
L = ∂µ φ(x)∂ µ φ(x) − m2 φ(x)2 , (1.26)
2 2
the choice of the sign in front of the second term will be clear in a moment. From the kinetic
term (the first one) we can read the energy dimensions of the field: the Lagrangian density
has dimension 4, each derivative has dimension 1, so a scalar field canonically normalized has
dimension 1. The constant m2 has thus dimension 2.
1.2. FIELD EQUATIONS AND LAGRANGIAN 5

The equation of motion can readily be derived:

( + m2 )φ(x) = 0 , (1.27)
which corresponds to is a free massive wave equation. Indeed φ(x) = c cos(px) = c cos(ωt − p~~x)
is solution provided p2 = ω 2 − p~2 = m2 . Should we have chosen the wrong sign in front of the
mass term in the Lagrangian, ω could be complex, the solution would not be associated to a
propagating wave but would show an exponential instability. As we will show later in such cases
the Hamiltonian would also be unbounded from below.

Free complex scalar field

A complex scalar field φ(x) is a scalar field that associates a complex number to each space-time
points, thus φ? (x) 6= φ(x). φ? (x) and φ(x) must be considered as two independent fields. We can
see them as two different linear combinations of √ two real scalar fields φ1 (x) and? φ2 (x) related to
the complex fields as φ(x) = (φ1 (x) + iφ2 (x))/ 2. We thus expect φ(x) and φ (x) to propagate
two independent degrees of freedom. Consider for example then the following Lagrangian density
L = ∂µ φ(x) ∂ µ φ? (x) − m2 |φ(x)|2 , (1.28)
We should take special care when dealing with complex field to make sure the Lagrangian is
real. This guarantees that the Hamiltonian will be real, meaning real energy in classical field
theory and unitary evolution in the quantum one.
Varying with respect to each field (φ(x) and φ? (x)) we get the two equations of motion:
φ + m2 φ = 0 ,
φ? + m2 φ? = 0 ,
the equations of two massive free scalar fields.

Generic quadratic Lagrangian

Consider now the generic quadratic Lagrangian
X 1 1

1 1
L= Kij ∂µ φi ∂ µ φj − Mij2 φi φj = ∂µ φT K∂ µ φ − φT M 2 φ , (1.29)
2 2 2 2
ij

where the matrices K and M can be chosen symmetric without loss of generality and we further
require that all eigenvalues of K are strictly positive. If O1 and O2 are two orthogonal matrices
and D is a diagonal invertible matrix than it follows that
O1T D−1 O2T O2 DO1 = 1 . (1.30)
If we define
φ̃ = O2 DO1 φ (1.31)
we choose O1 to diagonalize K into KD = O1 KO1T , we choose D such that D−1 K D D−1 =1
and we choose O2 to diagonalize D−1 O1 M 2 O1T D−1 into MD
2 we get
1 1 1X
L = ∂µ φ̃T ∂ µ φ̃ − φ̃T MD
2
φ̃ = (∂µ φ̃i )2 − MD
2 2
ii φ̃i . (1.32)
2 2 2
i
So every quadratic Lagrangian can be brought into a factorized canonical form by means of a
field redefinition.
6 CHAPTER 1. CLASSICAL FIELD THEORY

1.3 Hamiltonian
Exploiting the analogy of a field φ(t, ~x) with an infinite continuum of degrees of freedom in
classical mechanics qi (t) = φ(t, ~x) we can introduce also an Hamiltonian. To proceed we need
to introduce the analogue of the conjugate momenta of the variables qi (t), i.e. the conjugate
momentum density Πr
∂L δL(φ, ∂µ φ)
pi = −→ Πr (t, ~x) = (1.33)
∂ q̇i δ φ̇r (t, ~x)
which we use to get rid of the variables φ̇r and write an Hamiltonian only functions of the
fundamental variable φr and Πr :
Z  Z   Z
X
3

3 δL
H= pi q̇i − L = d x Πr φ̇r − L = d x φ̇r − L = d3 x H (1.34)
i
δ φ̇ r

where the Hamiltonian should be taken as a function of the fundamental variables only H =
H(φr , Πr ) and we introduced the Hamiltonian density H. Unlike the Lagrangian the Hamiltonian
density is not a scalar under Lorentz transformation, afterall the energy is not a scalar quantity
as well. In the Hamiltonian formalism Lorentz invariance is not manifest anymore, but we will
see that at the quantum level it makes unitarity manifest.
As an example consider a massive real scalar field, its conjugate momentum
δL
Π(x) = = φ̇(x)
δ φ̇
the Hamiltonian density will read
1 1 ~ 1
H = Π(x)2 − L = Π(x)2 + (∇φ(x))2
+ m2 φ(x)2 (1.35)
2 2 2
Note that in this case the Hamiltonian is the sum of three squares so it is positive definite and in
particular is bounded from below. The Hamiltonian is minimized by the classical configuration
φ(x) = 0 which has minimum energy and it’s called the classical vacuum. Notice that a wrong
sign in front of the mass term would result in a Hamiltonian unbounded from below, with obvious
implications regarding the stability of φ(x) = 0 configuration.
For a generic Lagrangian
1
L = (∂µ φ)2 − V (φ) (1.36)
2
where V (φ) is a generic function of φ, the Hamiltonian
1 1 ~
H = Π(x)2 − L = Π(x)2 + (∇φ(x))2
+ V (φ(x)) (1.37)
2 2
will still be bounded from below if the potential V (φ) is. The field configurations that minimize
the potential are called the vacuum configurations. They have Π(x) = φ̇(x) = 0 and ∇φ(x) = 0,
i.e. φ(x) = φ0 =const, and minimize the potential, i.e. V (φ) ≥ V (φ0 ).
Now consider again eq. 1.35 in the discretized limit φ(xi , t) → qi (t)
X1 1

2 2 2 2
H= p + (qi − qi−1 ) + m qi , (1.38)
2 i ∆x2
i
1.4. EXPLICIT SOLUTIONS OF FREE FIELD THEORY 7

where the second terms arises from the discretization of the spatial derivative. We can recognize
the Hamiltonian of an infinite number of coupled harmonic oscillators. The coupling is controlled
by the derivative term that couples the nearest neighbors oscillators. The system can easily be
solved exactly by diagonalizing the system of oscillators, we will see that this simply corresponds
to take the Fourier transform.

1.4 Explicit solutions of free field theory

Consider the free field equation of motion for a scalar field with mass m:

( + m2 )φ(x) = 0 .

Decomposing the solution in Fourier modes

d4 p −ipx
Z
φ(x) = e φ̃p
(2π)4
the equation of motion implies
(−p2 + m2 )φ̃p = 0
i.e. the only propagating waves with φ̃p 6= 0 are those satisfying the dispersion relation
p
p0 = ± p~ 2 + m2 ≡ ±ωp .

The constraint above can be implemented by parametrizing the solution as

h i
φ̃p = 2π δ(p2 − m2 ) ãp~ θ(p0 ) + b̃p~ θ(−p0 )

where the Dirac-δ imposes the constraint p2 = m2 and we used the step functions θ to separate
the two branches with positive and negative frequencies with coefficients ãp~ and b̃p~ respectively,
which are arbitrary functions of the spacial momentum p~ only.
Using the fact that
δ(p0 − ωp ) δ(p0 + ωp )
δ(p2 − m2 ) = δ((p0 )2 − ωp2 ) = + ,
2ωp 2ωp
the solution can be rewritten as
d3 p 1
Z h i
φ(x) = ãp~ e−i(ωp t−~p~x) + b̃p~ ei(ωp t+~p~x)
(2π)3 2ωp
d3 p 1
Z h i
= ãp~ e−i(ωp t−~p~x) + b̃−~p ei(ωp t−~p~x) (1.39)
(2π)3 2ωp

where we integrated over p0 and in the last step we changed p~ to −~ p in the integral.
For a real scalar field φ(x) = φ? (x) which implies that b̃−~p = ap?~ so we finally have

d3 p 1 h
Z i
−i(ωp t−~
p~
x) ? i(ωp t−~
p~
x)
φ(x) = ãp~ e + ãp~ e
(2π)3 2ωp
d3 p 1 
Z

= Re(ã p
~ ) cos(ω p t − p
~ ~
x ) + Im(ãp~ ) sin(ωp t − p
~~
x ) (1.40)
(2π)3 ωp
8 CHAPTER 1. CLASSICAL FIELD THEORY

Notice that the final form of the solution is just the linear combination of general solutions of
simple harmonic oscillators, one for each momentum p~ with frequency ωp . The Fourier transform
has thus diagonalized the Hamiltonian, so that in momentum space each mode p~ is associated
to an independent harmonic oscillator.
We now consider the same equation in the presence of a source j(x):

( + m2 )φ(x) = j(x) ,

which derives from the Lagrangian density

1 1
L = [∂µ φ(x)]2 − m2 φ(x)2 + j(x) φ(x) .
2 2
Again using Fourier decomposition

d4 p −ipx d4 p −ipx
Z Z
φ(x) = e φ̃p j(x) = e j̃p
(2π)4 (2π)4

the equation of motion implies

j̃p
[p2 − m2 ]φ̃p + j̃p = 0 , or φ̃p = −
p2 − m2

The general solution thus read

d4 p
Z
i
φ(x) = ij̃p e−ipx + free solution
(2π) p − m2
4 2

Focusing on the first term the solution can conveniently be written as the convolution
Z
φ(x) = i d4 y G(x − y)j(y) , (1.41)

where the Green function G(x) satisfies the equation

( + m2 )G(x) = −iδ (4) (x) ,

which is the equation for a massive scalar field with a localized source. The solution for G(x) is
then formally
d4 p
Z
i
G(x) = e−ipx .
(2π) p − m2
4 2

Notice that the integral over p0 has singularities over the path of integration so that the integral
is not well defined. The singularities are simple poles located at p0 = ±ωp and can be avoided
by deforming the path of integration onto the complex p0 plane. The ambiguity of the choice
of path is related to the freedom in the boundary conditions of the equation of motion, i.e. to
different choices of the coefficients ãp~ for the free part of the solution.
We discuss next some important cases.
1.4. EXPLICIT SOLUTIONS OF FREE FIELD THEORY 9

Retarded Green function

As first case consider the path which passes over both poles on the Im(p0 ) > 0 half plane. This
choice is called retarded Green function because the integral
d4 p
Z
i
GR (x) = 4 2 2
e−ipx . (1.42)
Im(p0 )>0 (2π) p − m

vanishes for t < 0. This is simple to check: for t < 0 we can close the contour of integration
on the complex p0 plane with a path extending at |p0 | → ∞ with Im(p0 )>0 so to close a loop
on the upper half plane. The loop does not contain any poles so the integral vanishes. The
contribution from the arch at |p0 | → ∞ also vanishes because the argument of the exponential
in (1.42) goes to −∞ for t < 0 and Im(p0 )>0. Therefore eq. (1.42) vanishes as well.
For t > 0 we can close the contour in the lower half plane. The contribution from the arch at
|p0 | → ∞ and Im(p0 )<0 still vanishes for the same reason as before, but now the loop encloses
the two singularities. Using Cauchy theorem the integral is thus
d3 p 1  −i(ωp t−~p~x)
Z 
GR (x) = e − e i(ωp t−~
p~
x)
θ(t) = θ(t)[∆+ (x) − ∆− (x)] .
(2π)3 2ωp
where we defined the functions
d3 p 1 ∓i(ωp t−~p~x)
Z
±
∆ (x) ≡ e
(2π)3 2ωp

which satisfy the relation ∆− (x) = ∆+ (−x) = ∆+ (x)? and correspond to the Green functions
computed on a loop around each single pole at p0 = ±ωp respectively. Such Green functions are
indeed the coefficients of the ãp~ and ãp?~ parameters of the free equation, so that the ambiguity
in the path of integration is a particular choice for ãp~ in the free part of the solution.
The retarded Green function corresponds to the field response to a point-like pulse source
localized at the origin with vanishing boundary condition for t < 0. From eq. (1.42) is clear that
GR (x) = GR (Λx) so that GR can only be a function of the Lorentz invariant combination x2 .
Since the response vanishes for t < 0 and at t = 0 for ~x 6= 0, using Lorentz invariance it follows
that the response vanishes for any xµ except inside the future light-cone, i.e. x2 ≥ 0 and x0 ≥ 0.
The convoluted solution in eq. (1.41) is also clear. The response to a generic source will
just be the linear superposition of the responses from each space-time point of the source, each
producing a signal inside their own future light cone.

Advanced Green function

A different choice would be to choose the path of integration passing below the two poles, in
the lower half plane Im(p0 ) < 0. This choice is called advanced Green function GA (x). For
t > 0 the contour can be closed in lower half plane without enclosing any singularities, so that
GA (t > 0, ~x) = 0. For t < 0 the contour can be closed in the upper half plane where instead
both poles contribute. The final result is similar to the retarded Greeen function, with inverted
time
GA (x) = −θ(−t)[∆+ (x) − ∆− (x)] ,
where the overall minus sign is due to the fact that the contour integral is computed in the
opposite sense.
10 CHAPTER 1. CLASSICAL FIELD THEORY

The physical meaning of this solution is the time reversal of the previous one: instead as a
source j(x) should be interpreted as a sink and the solution correspond to what field configuration
in the past could end up producing a sink j(x) localized in the origin.

Feynman or Time ordered Green function

A third possibility is to pass below the first pole at p0 = −ωp and above the one at p0 = ωp . This
case is called Feynman or time ordered Green function GF (x). The solution gets a contribution
from each pole at negative and positive time respectively and reads

GF (x) = θ(t)∆+ (x) + θ(−t)∆− (x) .

The solution is non-vanishing everywhere and is complex. We will compute the explicit form in
the following chapter where we will explain the physical meaning of the correspoding boundary
conditions.
Before concluding notice once again that the difference between the different choices, i.e. the
ambiguity in the choice of path of integration, correspond to a different choice of the coefficients
for the free part of the equation. For example

GR (x) − GA (x) = ∆+ (x) − ∆− (x)

corresponding to the choice ap = i in eq. (1.40).

1.5 Symmetries and Conserved Currents

1.5.1 Internal Symmetries
A theory is said to possess a symmetry if a transformation acting on the fields keeps the action
invariant2 , or equivalently if it maps solutions of the equations of motion into other solutions. If
the transformation is continuously connected to the identity by a continuous parameter Noether
theorem guarantees the presence of a conserved charge. We will derive this result in the case of
internal symmetries (associated to non space-time transformations) here, and discuss the case
of Poincaré symmetries separately in the next section.
Under an infinitesimal transformation

φr (x) → φ0r (x) = φr (x) + αδφr (x) (1.43)

where α is the parameter of the transformation and δφr (x) can be a generic local functions of
all fields.
By definition the action is invariant under such transformation, i.e. δS/δα = 0 identically
for any field configuration φr (x) even not satisfying the equation of motion. This implies that
the Lagrangian density must be invariant too, up to a total derivative, i.e.

δL
= ∂µ K µ . (1.44)
δα
2
This definition must be slightly modified at the quantum level...
1.5. SYMMETRIES AND CONSERVED CURRENTS 11

On the other hand we also have

δL δL δL
= δφr + δ(∂µ φr )
δα δφr δ(∂µ φr )
 
δL δL
= ∂µ δφr + ∂µ δφr
δ(∂µ φr ) δ(∂µ φr )
 
δL
= ∂µ δφr , (1.45)
δ(∂µ φr )

where in the second line we used the equations of motion. If we equate eqs. (1.44) and (1.45)
we get the identity  
δL
∂µ δφr = ∂µ K µ , (1.46)
δ(∂µ φr )
which means that on the equation of motion there exist a conserved current j µ defined as

δL
jµ ≡ δφr − K µ ,
δ∂µ φr
∂µ j µ = 0 . (1.47)

For trasformations that are also symmetries of the Lagrangian the last term vanishes.
As usual we can take a four-volume and integrate the relation above to get
Z Z t2 Z Z t2 Z
0= d4 x ∂µ j µ (x) = dt ∂t d3 x j 0 (x) + ~ · ~j(x)
dt d3 x∇
Ω t1 t1 Ω
=⇒
Z t2 Z
Q(t2 ) − Q(t1 ) = dt ~ · ~j(x)
dΣ
t1 ∂Ω
Z
Q(t) ≡ d3 x j 0 (x) (1.48)

the variation of the total charge Q(t) (the integral of the charge density j 0 (x)) is given by the
integrated charge flux exchanged along the spatial boundaries, if we take the spatial volume to
be infinite, or make sure the flux on the boundary vanishes, the charge is conserved.

Examples
• Consider a complex scalar field with Lagrangian

L = |∂µ φ|2 − V (|φ|2 ) , (1.49)

which is invariant under the U (1) transformation

φ → φ0 = eiα φ ' φ + iαφ . (1.50)

The expression for the conserved current is

j µ = i(φ∂ µ φ? − φ? ∂ µ φ) , (1.51)
12 CHAPTER 1. CLASSICAL FIELD THEORY

and the associated charge is

Z
Q=i d3 x (φφ̇? − φ̇φ? ) . (1.52)

It is easy to check that Q̇ = 0 using the equations of motion.

• Consider now a theory with N identical real scalar fields φi=1...N with Lagrangian density
1 1
L = (∂ µ φi )(∂µ φi ) − m2 φ2i , (1.53)
2 2
where repeated indices are summed over. The Lagrangian is invariant under a O(N )
transformation which rotates the scalar fields among themselves

φi → φ0i = Rij φj , RRT = 1 . (1.54)

Expanding into infinitesimal transformations

~
R = eα~ ·t ' 1 + α
~ · ~t , taij = −taji (1.55)

where ta with a = 1 . . . N (N − 1)/2 are the antisymmetric generators of the SO(N ) group
in the fundamental representation, the field transformation can be written as

φi → φi + αa taij φj . (1.56)

There are thus N (N − 1)/2 conserved currents

jµa = (∂µ φi )taij φj , (1.57)

and equal number of conserved charges

Z
a
Q = d3 x φ̇i taij φj . (1.58)

1.5.2 Poincaré Symmetries

Translations
Under infinitesimal space-time shifts all fields transform as

φr (x) → φr (x − ξ) ' φr (x) − ξ ν ∂ν φr (x) (1.59)

The Lagrangian density transforms as a scalar field

L(x) → L(x − ξ) ' L − ξ µ ∂µ L = L − ξ ν ∂µ (δνµ L) . (1.60)

We thus have 4 independent symmetries parametrized by the parameter ξ ν (corresponding to

α of the previous section), and for each of them there will be a conserved current (which will
therefore have an extra index, j µ ν ). We can now extract δφr and K µ from the equations above
and get

δφr = −∂ν φr ,
K µ = −δνµ L , (1.61)
1.5. SYMMETRIES AND CONSERVED CURRENTS 13

which we can use in eq. (1.47) to find our conserved current

δL
jµν = (−∂ν φr ) + δνµ L . (1.62)
δ(∂µ φr )
We can now define the conserved energy momentum tensor as
δL
T µν = −j µν = ∂ ν φr − g µν L , ∂µ T µν = 0 . (1.63)
δ(∂µ φr )
which provides 4 conserved charges one for each component of the second index:
Z Z  
µ 3 0µ 3 δL µ 0µ
P = d xT = d x ∂ φr − g L
δ(φ̇r )
Z   Z
0 3 δL
P = d x φ̇r − L = d3 x H = H
δ(φ̇r )
Z Z
~ 3 δL ~ ~
P = d x (−∇φr ) = d3 x P (1.64)
δ(φ̇r )
We can recognize the Hamiltonian for the zero component of P µ , while the other components
correspond to the total 3-momentum which is conserved as a result of the invariance under space
translations.
Note that indeed the Hamiltonian density is not a scalar but transforms as the time compo-
nent of a four-vector.

Lorentz transformations
The transformation of a field in a generic representation of the Lorentz group can be represented
as
φr (x) → φ0r (x) = Srs (Λ)φs (Λ−1 x) , (1.65)
where S(Λ) is the representation of the action of the Lorentz group on the field φr , e.g. the
identity matrix for scalar fields, Λ itself for vector fields or a more complicated expression for
spinors and tensors. Under infinitesimal transformations the following expressions hold

Λµ ν = δνµ + µ ν
Srs (Λ) = Srs (1 + ) = δrs + µν Σµν
rs
φr (x) → φr (x) + µν Σµν µ ν
rs φs (x) −  ν x ∂µ φr (x)
 
µν 1 µ ν ν µ


= φr (x) + µν Σrs φs (x) + x ∂ φr (x) − x ∂ φr (x) (1.66)
2
where we used the antisymmetric property of µν in the last expression.
As in the previous case we can use the fact that the Lagrangian density transforms as a
scalar field

L(x) → L(Λ−1 x) ' L(x) − µ ν xν ∂µ L

= L(x) − µν ∂ρ (g µρ xν L) ,
1
= L(x) + µν ∂ρ (xµ g νρ L − xν g µρ L) , (1.67)
2
14 CHAPTER 1. CLASSICAL FIELD THEORY

where we used the fact that

µν ∂ρ (g µρ xν ) = µν g µν = 0 , (1.68)
given the antisymmetry of µν .
We can now identify the fields and the Lagrangian density variations as
1 µ ν
δφr = Σµν ν µ
rs φs (x) + [x ∂ φr (x) − x ∂ φr (x)] ,
2
1
K ρ = (xµ g νρ L − xν g µρ L) , (1.69)
2
where the two free (antisymmetric) indices µν correspond to the six independent components
of µν , one for each generator of the Lorentz group. We will have six conserved currents j ρ,µν .
Their explicit expression is readily obtained by plugging eq. (1.69) in eq. (1.47)

δL µν δL 1 µ ν 1
j ρ,µν = Σ φs (x) + [x ∂ φr (x) − xν ∂ µ φr (x)] − (xµ g νρ L − xν g µρ L)
δ∂ρ φr rs δ∂ρ φr 2 2
δL µν 1 µ ρν
= Σ φs (x) + (x T − xν T ρµ ) . (1.70)
δ∂ρ φr rs 2

We can equivalently define the conserved tensor

δL µν
Mρµν ≡ 2j ρ,µν = 2 Σ φs (x) + xµ T ρν − xν T ρµ
δ∂ρ φr rs
∂ρ Mρµν = 0 . (1.71)

The corresponding conserved charges are

Z Z  
δL µν
J µν = d3 x M0µν = d3 x 2 Σrs φs (x) + xµ P ν − xν P µ (1.72)
δ φ̇r

The three spatial components correspond to the total angular momentum

Z  
ij 3 δL ij i j j i
J = d x 2 Σrs φs (x) + x P − x P (1.73)
δ φ̇r

which are the sum of the intrinsic spin associated to the Lorentz representation of the field and
the orbital momentum ~x × p~.
The other three charges associated to the boosts are
Z  
i 0i 3 δL 0i i i
K =J = d x 2 Σrs φs (x) + tP − x H (1.74)
δ φ̇r

For a scalar field the first term is absent, if we divide by H we get

P~
Z
1
d3 x H ~x = const + t (1.75)
H H

which corresponds to the conservation of the velocity of the center of mass.

1.5. SYMMETRIES AND CONSERVED CURRENTS 15

Notice that these latter charges have an explicit time dependence, therefore
d i ∂
K ={K i , H} + K i = 0 ,
dt ∂t
∂
{K i , H} = − K i 6= 0 (1.76)
∂t
when we will quantize the theory this will imply that boost charges do not commute with the
Hamiltonian—eigenstates of the boost operator will not have definite energy. Therefore they
will not be useful to classify Hamiltonian eigenstates, this is not surprising as energy is not
invariant under Lorentz boosts afterall.