Mathematical Aspects

of
Quantum Field Theory
Michael Keyl (FU Berlin)1

November 3, 2017

Block Course on Mathematical Aspects of Quantum Field Theory held at LMU

Physics. Pre-version done and proofread by Frederik vom Ende (TU Munich)2

1
Mail adress: michaelkeyl137@[Link]
2
Mail adress: [Link]-ende@[Link]
Contents
0 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 Wightman Quantum Field Theory . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Tempered Distributions . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Quantum Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Wightman Distributions . . . . . . . . . . . . . . . . . . . . . . . . 12
2 The Free Scalar Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1 Representation of the Poincaré Group . . . . . . . . . . . . . . . . 19
2.2 The Klein-Gordon Equation . . . . . . . . . . . . . . . . . . . . . . 21
2.3 The Free Quantum Field . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Time-Zero Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3 Interactive Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1 Naive Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 The Cuto Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . 46
3.3 Haag's Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4 Remove the Cuto . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
 0 Introduction

0 Introduction
In quantum mechanics, the standard approach is as follows. We have a Hilbert space
H (e.g. H = L2 (R3 ) for a particle in three dimensions), a Hamiltonian H describing
the dynamics (like a Schrödinder operator H = −∆ + V (x) for a particle moving in a
potential V ), and the standard observables position Qj and momentum Pk described by
self-adjoint operators on H. In Quantum Field Theory very loosely speaking, the main
dierence concerns these observables. While we still work on Hilbert spaces and with
Hamiltonians, the Qj and Pk are replaced by eld amplitudes R3 3 x 7→ φ(x) and eld
momenta R
3 3 x 7→ π(x) at a point (or space time event) x. Mathematically they are
usually not described by operators but quadratic forms in H.
There are many reasons to use elds rather than particles. Often both discriptions are
equivalent and the corresponding elds admit an interpretation in terms of particles. In
such a situation it is a matter of convenience which pictures should be preferred. But
in some cases the particle point of view (or more generally spoken: the point of view of
ordinary Quantum Mechanics) is too limited and elds are really needed for a consistent
description. The most prominent example where this happens is Relativistic Quantum
Mechanics which does not exist as a consistent theory. Problematic are in particular the
following issues:

• Locality. Consider particle described by a wave function Ψ ∈ L2 (R3 ) and located in

the region Σ, i.e. supp ψ ⊂ Σ. If we follow the usual rule and describe the probability
Σ ⊂ R3 by σ |Ψ(x)|2 dx the particle is located in Σ with
R
to detect a particle in
certainty. After evolving the wave function freely, and for an arbitrarily short time 
√
(e.g. with the relativistic Hamilton operator H = P 2 + m2 ) the new wave function
ψ has non compact support, i.e. the probability to nd the particle arbitrarily far
away from its original position is non-zero; cf. Fig. 1. Something like this can also
happen in non-relativistic models. In those cases, however, innite speed is not a
conceptual problem. In Special Relativity, on the other hand, it is.

Figure 1: Possible evolution of the amplitude of the wave function for some arbitrary
timestep ε > 0. This however is non-consistent with special relativity, in par-
ticular the speed of light.

• Particle creation. Even in simple setups involving relativistic particles in external

potential particle creation eects can occur. If you want to describe this within

1
 0 Introduction

Figure 2: Klein's result showed that if the potential is of the order of the electron mass,
the barrier is nearly transparent. Also for the dirac equation, more particles
can be reected than were sent in because of pair production. The number of
particles not being x cannot be described by quantum mechanics.

a model which is based on the assumption that the particle number is xed this
leads to contradictions. A typical example is Klein's paradox where an electron
beam hitting a potential barrier is described. Within relativistic Quantum Mechanics
we get strange behaviours of reection and transmission coecients which can be
explained if we involve pair creation; cf. Fig. 2. Note that dynamical changes of the
particle number can also occur in non-relativistic models (e.g. in solid state physics).
However, in the relativistic case particle creation and annihilation is the rule even
in simple cases involving only external potentials and no real interaction. It can be
avoided only if we restrict our attention to free particles.

Both problems can be resolved by elds: Firstly, in Quantum Field Theory we can localize
elds (observables) rather than wave functions (states), and secondly, elds are perfectly
capable to describe particle models where the particle number is not xed.

2
 1 Wightman Quantum Field Theory

1 Wightman Quantum Field Theory

1.1 Tempered Distributions
For this section we orient ourselves towards [RS80, Chapter V]. First we introduce the
following standard notation. For f : Rn → C, α, β ∈ Nn0 and Rn 3 x = (x1 , . . . , xn ) we
denote

n
∂ |α|
xβ = xβ1 1 . . . xβnn
X
|α| = αj Dα f = .
∂xα1 1 . . . ∂xαnn
j=1

Denition 1.1. The functions of rapid decrease, also called Schwartz functions, is the
set dened as
S (Rn ) := {f ∈ C ∞ (Rn , C) | kf kαβ < ∞ ∀α,β∈Nn0 }

where kf kαβ = supx∈Rn |xα Dβ f (x)|.

Remark 1.2 (Locally compact spaces). Let V be a complex vector space V and(pj )j∈I
a family of seminorms
3 which separates points, so pj (f ) = 0 for all j∈I implies f = 0.
Then (V, (pj )j∈I ) is said to be a locally convex space. A neighborhood base at 0 ∈ V is
given by the sets

N (j1 , . . . , jm ; ε) = {f ∈ V | pjk (f ) < ε ∀k∈{1,...,m} } ,

which can be regarded as an analogon of the family of ε-balls in Banach spaces. By

translating the N (j1 , . . . , jm ; ε) within V we can dene similar bases at each v ∈ V. In
this way V becomes a topological space, and addition and scalar multilpication become
continous maps. In other words, V equipped with topology is a topological vector space.
If J = N, then we can dene a metric d via

∞
X pn (f − g)
d(f, g) = 2−n
1 + pn (f − g)
n=1

which generates the same topology as before. If (V, (pn )n∈N ) is complete, then it is called
Fréchet space.

The following is an easy application of the neighbourhood denition given above.

Lemma 1.3. A linear functional φ : V → C is continous if and only if one can nd a
nite set j1 , . . . , jm ∈ I of seminorms and C ∈ R+0 such that
|φ(x)| ≤ C(pj1 (x) + . . . + pjm (x))

for each x ∈ V .
3
A seminorm is a norm which is not denite so for f, g ∈ V and λ ∈ C we have p(λf ) = |λ|p(f ) and
p(f + g) ≤ p(f ) + p(g), but p(f ) = 0 does not necessarily imply f = 0.

3
 1 Wightman Quantum Field Theory

Proposition 1.4 ([RS80], Theorem V.9). The vector space S (Rn ) with the natural topol-
ogy given by the seminorms k · kα,β is a Fréchet space.
Denition 1.5. The topological dual4 S 0 (Rn ) of S (Rn ) is called the space of tempered
distributions.
Example 1.6. 1. (δ -distribution). Consider δ x : S (Rn ) → C for some x ∈ Rn where
f 7→ δx (f ) := f (x). We then have

|δx (f )| ≤ sup |f (y)| = kf k0,0 ,

y∈Rn

so δx ∈ S 0 (Rn ). Sometimes it is useful to use the formal expression

Z
f (x) = δ(x − y)f (y) dy
Rn
which involves the  delta-function δ . When we do this it is important to keep in mind
that δ really does not exist as a function, i.e. we can not evaluate it at each x ∈ R, we
can only evaluate it under the integral.

2. (Measures). Consider a nite Borel measure

R µ, µ : S (Rn ) → C, f 7→
so we can dene

Rn f (x)µ(dx). Continuity of this map can be shown analogously to the δ -distribution

case, so µ ∈ S (R ).
0 n

3. Let us look at this special case of the second example. For g ∈ S (Rn ) we can dene
φg (f ) = Rn f (x)g(x) dx. Moreover, if g1 6= g2 as functions in S , then φg1 6= φg2 . This
R

embeds S naturally in S .
0

4. Similarly, for g ∈ Lp (Rn ) and p∈N we have

Z
φg (f ) = f (x)g(x) dx
Rn
which embeds Lp ,→ S 0 in a similar way.

Remark 1.7. S 0 (Rn ) with the weak-?-topology, which is generated by semi-

We equip
norms f 7→ |φ(f )| with φ ∈ S (R). Then, the linear subset S (R ) ⊂ S (R ) is dense
0 n 0 n

and the embedding ι : S (R ) → S (R ) is continous. This suggests extending continous

n 0 n

maps T : S → S to S as follows. If T : S → S is continous, then ι ◦ T : S → S is

0 0

continous as well by continuity of ι. Since S is dense is S , there is at most one continous

0

extension of ι ◦ T . To nd this extension we look for a continous S : S → S with the

adjoint S : S → S , φ 7→ S (φ) and ask S to satisfy S (φ) = φ ◦ S . Now S is well-
0 0 0 0 0 0 0
0
dened and continous in the weak-?-topology. Hence S is the extension we are looking
for and it can be expressed as T φ(f ) = φ(S(f )). In other words the general strategy is
to apply the adjoint of T to the test function (assuming that the double adjoint becomes
T again).
4
The topological dual is the space of continous linear functionals acting on the vector space.

4
 1 Wightman Quantum Field Theory

Example 1.8. 1. Consider a C ∞ function F : Rn → C which derivatives are polynomi-

exists C ∈ R0 and n ∈ N, such that
+
ally bounded, so there

|Dα F (x)| ≤ C(1 + kxk2 )n

for all α ∈ Nn0 . This means for any f ∈ S (Rn ) we have F f ∈ S (Rn ) where f 7→ F f
is continous. The extension in this case then is given by (F φ)f = φ(F f ). Then

Z Z
φg (F f ) = g(x)(F (x)f (x)) dx = (g(x)F (x))f (x) dx = φF g (f )
Rn Rn
2. Weak derivative. To extend Dα to S 0, partial integration implies that (Dα φ)(f ) =
(−1)|α| φ(Dα f ).
3. For the Fourier transform, we simply have φ̂(f ) = φ(fˆ).
4. f (·) → f (· − a) translation, f (·) → f (A·) with A ∈ GL(n, R)

Example 1.9 (Heaviside function). Dening

(
x x≥0
ν(x) = ,
0 x≤0

we get the Heaviside function via

d  Z ∞ Z ∞
φν (f ) = −φν (f 0 ) = − xf 0 (x) dx = f (x) dx,
dx 0 0

d
which implies
dx φν = φθ , where

(
x 1≥0
θ(x) = .
0 x≤0

Further

d  Z ∞
0
φθ (f ) = −φθ (f ) = − f 0 (x) dx = f (0)
dx 0

d
so
dx φθ = δ0 .

Theorem 1.10 (Regularity theorem for distributions, [RS80] Theorem V.10). Let φ ∈
S 0 (Rn ). Then φ = Dβ g for some polynomially bounded continous function g : Rn → C
and some β ∈ I+n , that is,
Z
φ(f ) = (−1)|β| g(x)(Dβ f )(x)dn x

for all f ∈ S (Rn ).

5
 1 Wightman Quantum Field Theory

Denition 1.11 (Support). We say φ ∈ S 0 (Rn ) vanishes on an open subset σ ⊂ Rn if

φ(f ) = 0 for all f with supp(f ) ⊂ σ . Then supp φ is the complement of the largest open
set on which φ vanishes.
Denition 1.12. Let φ ∈ S 0 (Rn ). We say x ∈ Rn is a regular point of φ if there exists
an open neighbourhood U of x and g ∈ C ∞ (U ), such that φ(f ) = φg (f ) for all f with
supp f ⊂ U . The complement of the set of regular points is called the singular support.

An easy example is the singular support of the δ -distribution, which obviously only is
the zero.

Theorem 1.13 (Kernel or nuclear theorem, [RS80] Theorem V.12). Let B(f, g) be a
separately continous bilinear functional on S (Rn ) × S (Rm ). Then there is a unique
tempered distribution T ∈ S 0 (Rn+m ) with B(f, g) = T (f ⊗ g) where
(f ⊗ g)(x1 , . . . , xn , xn+1 , . . . , xn+m ) = f (x1 , . . . , xn )g(xn+1 , . . . , xn+m ).

The result can easily be extended to more than two tensor factors.

1.2 Quantum Fields

Let us, as a reminder, quickly go over some notation and results regarding unbounded
operators. For this, we orient ourselves towards [RS80, Chapter VIII]. Here, H is any
separable Hilbert space.

• An operator is a linear map A : D(A) → H. The subspace D(A) ⊂ H is the domain

of A. If D(A) is dense, then the operator is called densely dened.

• The graph of A is dened to be Γ(A) = {(x, Ax) | x ∈ D(A)} ⊂ H × H. Then A is

called closed if Γ(A) ⊂ H × H is closed.
• An operator B is said to be an extension of A if Γ(A) ⊂ Γ(B), we then write A ⊂ B .
• An operator A is closable if it has a closed extension. If A is closable, it has a smallest
closed extension A characterized by Γ(A) = Γ(A). We then say A is the closure of
A.
• Let A be densely dened. We dene

D(A∗ ) = {x ∈ H | ∃y∈H ∀z∈D(A) : hAz, xi = hz, yi}

and for x ∈ D(A∗ ) we dene the adjoint via A∗ x := y for respective y. Note that
D(A∗ ) = {0} is possible.
• Let A be densely dened. Then A is closable if and only if D(A∗ ) is densely dened.
In
∗
that case, A is closed and the double adjoint satises A
∗∗ = A.

• A is said to be symmetric or hermitian if hAx, yi = hx, Ayi for all x, y ∈ D(A). This
is equivalent to D(A) ⊂ D(A∗ ) with Ax = A∗ x for all x ∈ D(A). Also, A is called
self-adjoint if A = A∗ . In particular, this means D(A) = D(A∗ ).

6
 1 Wightman Quantum Field Theory

• If A is symmetric, then A ⊂ A∗∗ ⊂ A∗ . If A is symmetric and closed, then A =

A∗∗ ⊂ A∗ . If A is self-adjoint, then A = A
∗∗ = A∗ .

• We call A essentially self-adjoint if A is symmetric and A is self-adjoint.

For the topic of quantum elds, we refer to [RS75, Chapter IX.8].

Denition 1.14. A quantum eld is a 4-tuple (H, D, Φ, Ω) consisting of a (separable)

Hilbert space H, a dense subspace D ⊂ H, a C-linear map
Φ : S (Rn ) → L(D, H),

where L(D, H) denotes the linear maps between D and H, and an element Ω ∈ D ⊂ H,
such that the following conditions hold.
(a) D is invariant, so Φ(f )D ⊂ D for all f ∈ S (Rn ).
(b) Ω is cyclic which means that the set
D0 := {Φ(f1 ) . . . Φ(fm )Ω | f1 , . . . , fm ∈ S (Rn ), m ∈ N} ⊂ D ⊂ H

is dense in H.
(c) Φ(f ) is closable for all f ∈ S (Rn ).
(d) For all x, y ∈ D, the map
S (Rn ) 3 f 7−→ hx, Φ(f )yi

is a tempered distribution.
Note that D0 can replace D since D0 is also dense in H.

Denition 1.15. A quantum eld (H, D, Φ, Ω) is called hermitian if 5

Φ(f ) = Φ(f )∗ |D

and if for f ∈ S (Rn , R) the operator Φ(f ) is essentially self-adjoint on D.

Remark 1.16 (Interpretation). A hermitian quatum eld describes a special observable,
or more precisely a whole family of observables. To understand this remark a bit better
let us pretend that we can write Φ(f ) as

Z
Φ(f ) = f (x)Φ(x) dx (1.1)
R
with self-adjoint operators Φ(x), x ∈ R. Note that this is usually not possible and we will
discuss in Ch. 2 how Eq. (1.1) can be interpreted in a mathematical rigorous way. For now
we just look at the operator-valued eld Φ(x) as a formal expression for an operator-
valued distribution in the same way as the delta-function δ(x) is a formal expression

5
f denotes the complex conjugate of f

7
 1 Wightman Quantum Field Theory

for the delta-distribution δ0 . In doing so we can look at Φ(x) as the observable which
measures the eld amplitude in x∈ Rn . The smeared out version Φ(f ) from Eq. (1.1)
can then be regarded as the averaged eld amplitude with averaging function f. This
interpretation makes it very clear that the Φ(f ) are local observables, i.e. they can be
measured in any region containing the support of f.
To keep this interpretation, self-adjointness of Φ(f ) for real-valued f is mandatory.
Mathematically, however, this self-adjointness condition is often annoying, since it is
dicult to prove. Most others drop it therefore and accept hermitian quantum elds
where the Φ(f ) are really only hermitian. We will see in the next Section why this can
be advantageous.

Remark 1.17. Up to now we can generalize everything to manifolds by replacing S (Rn )

with

D(M ) = {f : M → C | f is C∞ and compactly supported}

for some C ∞ -manifold M .

Remark 1.18 (Poincaré group) . The denitions presented so far are fairly general and
lack in particular any dynamical content. This will change now, when we study quantum
elds in Minkowski space. For that, let us have a short recap on some concpets and
notations.

• The Minkowski metric on R4 is given by

3
X
η(v, w) = v 0 w0 − v j wj .
j=1

• The Lorentz group is dened to be

O(3, 1) = {Λ ∈ GL(4, R) | η(Λv, Λw) = η(v, w)}.

The restricted Lorentz group then is

SO↑ (1, 3) = {Λ ∈ O(3, 1) | det Λ = 1, he0 , Λe0 i > 0}.

is the semi-direct product R o SO (1, 3). Hence

↑ 4 ↑
• The restricted Poincaré group P+
P+ is the set of pairs (a, Λ) where a ∈ R4 and Λ ∈ SO↑ (1, 3), and the group operation
↑

is

(a, Λ1 )(b, Λ2 ) = (a + Λ1 b, Λ1 Λ2 ).

The Poincaré group describes transformations from one inertial system into another
by the coordinate transformation v 7→ a + Λv .
↑
• Now assume that P+ 3 (a, Λ) 7→ U (a, Λ) ∈ U(H), with the unitary group U(H) on

8
 1 Wightman Quantum Field Theory

H, is a strongly continuous6 , unitary representation of P+

↑
on H. By strong continuity
we can dene generators of the translations

d
Pj ξ = −i U (λej , 1)ξ
dλ λ=0

for j = 0, . . . , 3 with standard basis (ej )j of R4 . Here ξ ∈ H is chosen such that the
limit λ → 0 exists. Note that U (λej , 1)ξ is a time translation for j = 0 and a space
translation if j = 1, 2, 3. Hence P0 is the Hamiltonian and P1 , . . . , P3 are momentum
operators. With an arbitrary a ∈ R we get
4

 X3 
U (a, 1) = exp i aj Pj . (1.2)
j=0

• If a = Λe0 with a Lorentz transfomation Λ, the generator Pa of the one parameter

group R 3 t 7→ U (ta, 1) ∈ U(H) is given by
3
X
∗
Pa = U (0, Λ)P0 U (0, Λ) = aj Pj
j=0

In other words, Pa is the Hamiltonian the inertial observer with four-velocity a is

seeing. The Lorentz transformation Λ describes the transition from our intertial
system (i.e. where we are at rest) into the one of observer a. Please note that ac-
cording to this reasoning the representation U (a, Λ) does not only describes the
transition between inertial systems, but also contains the complete description of
the dynamical structure of our theory.
• Since the translations form an abelian subgroup, the unitaries U (a, 1) mutually
commute and can therefore be jointly diagonalized. More precisely, by the spectral
theorem ([RS80], Ch. VII and Sec. VIII.3), there is a projection valued measure
7

E : B(R4 ) → B(H) Σ 7−→ E(Σ)

on R4 such that
Z  X 
hξ, U (a, 1)ξi = exp i aj λj hξ, E(dλ)ξi. (1.3)
R4 j

• Readers which are unfamiliar with measure theory should think of hξ, E(dλ)ξi as
Eξ (λ) dλ with singular (i.e. distributional) density function Eξ (λ). The latter can
map E from R into the
be written as Eξ (λ) = hξ, E(λ)ξi with a likewise singular 4

set of projection operators in H. Note, however, that this is a very handwaving point
of view. A mathematical rigorous interpretation requires measure theory.

6
This means that P+↑ 3 (α, Λ) 7→ U (α, Λ)ξ ∈ H is continous for all ξ ∈ H.
7
Here, B(R4 ) denotes the σ-algebra of Borel subsets of R4 and B(H) are the bounded operators on H.

9
 1 Wightman Quantum Field Theory

• Finally we dene two regions σ1 , σ2 to be spacelike separated if we can not reach σ1

8
from σ2 with a causal curve and vice versa. This is best described in a picture; cf.
Figure 3.

Figure 3: Two open regions σ1 , σ2 ⊂ R4 are

called spacelike separated if they
are separated by the light cone so
there is no physically possible in-
formation exchange between them.
Formally this means that for any
x ∈ σ1 , y ∈ σ2 we have η(x − y, x −
y) < 0.

Denition 1.19. A 5-tuple (H, D, Φ, Ω, U ) consisting of a hermitian quantum eld

(H, D, Φ, Ω) and a strongly continous representation of the Poincaré group U is called
Wightman quantum eld if the following conditions are fullled.
(a) (Local commutativity or microscopic causality). If f and g in S (R4 ) have supports
which are spacelike separated, then [Φ(f ), Φ(g)] = 0.
(b) (Special covariance). For each (a, Λ) ∈ P+↑ and all f ∈ S (R4 ) we have
Φ((a, Λ), f ) = U (a, Λ)Φ(f )U (a, Λ)∗ ,

where ((a, Λ), f )(x) = f (Λ−1 (x − a)).

(c) (Uniqueness and invariance of the vacuum). There exists a unique vector Ω ∈ H
such that for all a ∈ R4 we have
U (a, 1)Ω = Ω.

(d) (Spectral condition). The support of the spectral measure E from Eq. (1.3) is
contained in the forward light cone V + = {v ∈ R4 | V0 > 0, η(v, v) > 0}.
Remark 1.20. The local commutativity condition mathematically expresses the quantum-
mechanical statement that measurements in spacelike separated regions should be jointly
measurable; cf. the discussion in Remark 1.16. Note that commutation in the given form
does not guarantee jointly measurability, but it is a necessary condition; cf. in this context
[RS80], Sec. VIII.5.

8
This means by travelling with at most the speed of light.

10
 1 Wightman Quantum Field Theory

Remark 1.21 (Spectrum condition). The spectral condition means that the joint spec-
trum of Pj is contained in the forward light cone, so σ(P0 ) ⊂ R+
0 and the energy is positive
for all inertial systems. This can be seen very easily and without advanced knowledge
if we assume that all the Pj have purely discrete spectrum. Note that this is not very
realistic from the physical point of view since the four momenta usually have continuous
spectrum and the three-momentum typically has no eigenvalues at all. It is, however,
a very simple case where measure theory is not required and therefore it can help to
understand the role of the spectral measure E and the spectrum condition.
Since the Pj are mutually commuting, there exists a complete orthonormal system
(φn )n∈N in H with Pj φn = λjn φn for j = 0, . . . , 3. Dening λn = (λ0n , . . . , λ3n ) we get

 X3   X3  X∞
exp i aj Pj φn = exp(ia · λn )φn =⇒ exp i aj Pj = eia·λn |φn ihφn |.
j=0 j=0 n=0

9
where the sum is strongly convergent . With this and (1.2), we see that the matrix
elements of the unitaries are given by

∞
hx, U (a, 1)xi =
X
eia·λn |hx, φn i|2 .
n=0

Written as an integral, we get

Z ∞
hx, U (a, 1)xi =
X
ia·λ
e δ(λ − λn )|hx, λn i|2 dλ.
R4 n=0
| {z }
Ex (λ)
P∞
Another option is to use Eλ = n=0 δ(λ − λn )|φn ihφn | which yields

Z
hx, U (a, 1)xi = eia·λ dhx, Eλ xi.
R 4

This expression also makes sense in the non-discrete case. The support of the spectral
measure E obviously is given by

σ(E) = {λn | n ∈ N} ⊂ R4 .

The spectral condition now demands

σ(E) ⊂ V+ = {v ∈ R4 | η(v, v) > 0, v 0 > 0},

where V+ is the forward light cone. Now λn ∈ V+ by denition means λ0n > 0, which
implies positive energy.

9
This means that eia·λn |φn ihφn , ξi converges for all ξ ∈ H
P∞
n=0

11
 1 Wightman Quantum Field Theory

Remark 1.22 (Interpretation) . A Wightman quantum eld describes a physical sys-

tem which transforms covariantly under Poincaré tranformations (i.e. change of inertial
systems). It combines the unitary representation U (a, Λ) of the Poincaré group, and all
objects and concepts derived from it, like the four-momentum operators Pj , j = 0, . . . , 3
(cf. Remark 1.18), with a Hermitian quantum eld Φ(f ). The Pj contain in particular
the dynamical description of the model. As observables they are not that important from
a practical point of view, since they are global, i.e. they measure the four-momentum of
the whole universe. The eld operators Φ(f ) on the other hand are of local nature, as
pointed out in Remark 1.16. Hence they are more realisitically linked to quantities which
actually can be measured in an experiment. A typical model involving quantum elds
usually contains more than one eld, describing dierent physical quantities or observ-
ables. For example in addition to the eld Φ describing eld amplitudes we might want to
look at components of the energy momentum tensor, which are described by additional
elds within the same model.
The previous remark might create the impression that the elds are completely kine-
matical objects, while the dynamics is exclusively contained in the representation U (a, Λ).
This point of view, however, is wrong. Since a Wightman eld is a spacetime eld, it does
contain dynamical information. The axioms in Denition 1.19, in particular the Poincaré
covariance and the invariance of the vacuum, link the eld Φ and the representation
U (a, Λ) very closely together, in other words they are not independent, and the elds
contain informations about the representation U (a, Λ).

Remark 1.23 (Wightman axioms) . We haven't explicitly talked about the Wightman
axioms for a scalar eld, because we have distributed them over Denitions 1.14, 1.15
and 1.19. In Def. 1.14 and 1.15 we nd (numerations taken from [RS75], Sec. IX.8) :
Invariant domain for the elds (Axiom 4), regularity of the elds (Axiom 5) and the
cyclicity of the vaccuum (Axiom 8). In Denition 1.19 we have the existence of the
representation U (relativistic invariance of states; Axiom 1), the spectral condition
(Axiom 2), the invariance of the vacuum (existence and uniqueness of the vaccum;
Axiom 3), the Poincaré invariance of the elds (Axiom 6) and the local commutativity
(Axiom 7).

1.3 Wightman Distributions

Consider a hermitian quantum eld (H, D, Φ, Ω) and dene

W̃ (m) : S (Rn ) × . . . × S (Rn ) → C (f1 , . . . , fm ) 7−→ hΩ, Φ(f1 ) . . . Φ(fm )Ωi

| {z }
m factors

for any m ∈ N. By regularity of the eld, W̃ (m) is separately continous in all arguments
because

f1 7−→ h|{z}
Ω , Φ(f1 ) . . . Φ(fm )Ωi
| {z }
∈D ∈D

12
 1 Wightman Quantum Field Theory

is continous, the same is true for

f2 7−→ hΦ(f1 )∗ Ω, Φ(f2 ) . . . Φ(fm )Ωi

| {z } | {z }
∈D ∈D

and so forth. By the nuclear theorem (see Theorem 1.13), there exists a unique distribu-
tion W
(m) ∈ S 0 (Rn × ... × Rn ) = S 0 (Rn·m ) such that

W (m) (f1 ⊗ . . . ⊗ fm ) = W̃ (m) (f1 , . . . , fm ).

In particular, W̃ (m) is jointly continous.

Denition 1.24. The W (m) , m ∈ N are called Wightman distributions of the quantum
eld (H, D, Φ, Ω).
The tasks now are to reconstruct a quantum eld from W (m) and to translate the
Wightman axioms into conditions on W
(m) . The solution to both tasks is known as the

Wightman reconstruction theorem

Denition 1.25. A complex vector space A is called unital ∗-algebra if A is equipped

with a bilinear, associative product,
A × A 3 (A, B) 7−→ AB ∈ A

and an antilinear10 involution11 (∗-operation)

A 3 A 7−→ A∗ ∈ A

which satises (AB)∗ = B ∗ A∗ and there exists a unit 1 ∈ A with A1 = 1A = A for all
A ∈ A.

Denition 1.26. A functional ω : A → C is called a state of A if

(a) ω is linear and continous.
(b) (Positivity). ω(A∗ A) ≥ 0 for all A ∈ A.
(c) (Normalization). ω(1) = 1.
Lemma 1.27. A state ω of a *-algebra A has the following properties
(a) (Symmetry). ω(A∗ B) = ω(B ∗ A).
(b) (Cauchy-Schwarz). |ω(A∗ B)|2 ≤ ω(A∗ A)ω(B ∗ B).
Proof idea. Look at ω((λA + B)∗ (λA + B)) ≥ 0. Otherwise cf. [BR02] Lemma 2.3.10.

10
This means (A + λB)∗ = A∗ + λB ∗ .
11
This means A∗∗ = A.

13
 1 Wightman Quantum Field Theory

Denition 1.28. A cyclic representation of a ∗-Algebra is a 4-tuple (H, D, π, Ω) con-

sisting of a Hilbert space H, a dense subspace D ⊂ H, a vector Ω ∈ D and a complex
linear map
π : A → L(D, D) ⊂ L(D, H),

such that the following holds for all A, B ∈ A.

(a) π(AB) = π(A)π(B).
(b) π(A∗ ) = π(A)∗ |D .
(c) (Cyclicity). {π(A)Ω | A ∈ A} = D0 ⊂ D ⊂ H is dense.
Again, D0 is automatically an allowed domain and actually the smallest one.

Theorem 1.29 (GelfandNaimarkSegal (GNS)-representation). Let A be ∗-algebra and

ω:A→C be a state. Then there exists a cyclic representation (Hω , Dω , πω , Ωω ), such
that
ω(A) = hΩω , πω (A)Ωω i (1.4)

holds for all A ∈ A. If Dω = D0 , this representation is unique up to unitary equivalence.

Proof. On A we dene

hA, Bi = ω(A∗ B)

which is sesquilinear and positive semi-denite by Lemma 1.27, but ω(A∗ A) = 0 may
happen for A 6= 0. Therefore we dene

I ω = {A ∈ A | ω(A∗ A) = 0}.

We want to show that Iω is a left ideal in A. First, I ω is linear for the following reason.

(a) For A ∈ I ω, λ ∈ C now

ω((λA)∗ (λA)) = |λ|2 ω(A∗ A)

implies λA ∈ I ω .
(b) For A, B ∈ I ω we have

ω((A + B)∗ (A + B)) = ω(A∗ A) + ω(B ∗ B) + 2 Re ω(A∗ B) = 0.

=0 =0

It is easy to see that 2 Re ω(A∗ B) vanishes by Cauchy-Schwarz.

Now A ∈ Iω implies |hA, Bi|2 ≤ ω(A∗ A)ω(B ∗ B) so hA, Bi = 0 for all B∈A and thus

I ω = hA ∈ A | ω(AB) = 0 ∀B∈A }.

14
 1 Wightman Quantum Field Theory

Finally, Iω is a left ideal because

ω((BA)∗ C) = ω(A∗ (B ∗ C)) = 0 =⇒ BA ∈ I ω

for any A ∈ I ω , B, C ∈ A. Now we dene Dω = A\I ω and a scalar product h[A], [B]i =
hA, Bi which is well-dened as can be veried readily. Our Hilbert space H then is given
by the completion of Dω , which by construction is dense in H. For A, B ∈ A, I ∈ I ω
we have

AI =⇒ A(B + I) ∈ [AB].
A(B + I) = AB + |{z}
∈I ω

Thus we can dene πω by

πω (A)[B] = [AB],

as well as Ωω = [1]. Then πω (A)Ω = [A] and {πω (A)Ω | A ∈ A} = Dω ⊂ H is dense.

Finally, we have to check (1.4).

hΩω , πω (A), Ωω i = h[π], πω (A)[π]i = h[π], [Aπ]i = h[1], [A]i = ω(1A) = ω(A)

Now for the uniqueness. Let (H̃, D̃, π̃, Ω̃) be another cyclic representation which satises
(1.4) with D̃ = D̃0 = {π̃(A)Ω̃ | A ∈ A}. Dene

U : Dω → D̃ U πω (A)Ωω = π̃(A)Ω̃

After showing that U is well-dened and unitary, one extends it to H and sees that
U Dω = D̃ and U πω (A)U ∗ = π̃(A).

Proposition 1.30. Consider nite direct sums

A = C ⊕ S (Rn ) ⊕ S (Rn × Rn ) ⊕ . . .

with seminorms kf k(m)

αβ = kf
(m) k , where
αβ

f = f (1) ⊕ . . . ⊕ f (m) ⊕ . . . ⊕ f (n)

and f (n) ∈ S (Rn·m ), and consider the product

f g = [f (0) ⊕ f (1) ⊕ . . . ⊕ f (n) ] ⊗ [g (0) ⊕ . . . ⊕ g (k) ]
= f (0) ⊗ g (0) ⊕ (f (0) ⊗ g (1) + f (1) ⊗ g (0) ) ⊕ (f (0) ⊗ g (2) + f (1) ⊗ g (1) + f (2) ⊗ g (0) ) ⊕ . . . ,

as well as the ∗-operation

(f ∗ )(m) (x1 , . . . , xm ) = f (m) (xm , . . . , x1 ).

15
 1 Wightman Quantum Field Theory

Then A is locally convex space and ∗-algebra. Further, the operations are continous so A
is a topological ∗-algebra with unit 1 = 1 ⊕ 0 ⊕ 0 ⊕ . . . ⊕ 0.
Denition 1.31. The A just dened is called Borchers-Uhlmann-algebra (BU-Algebra).
The following proposition is a straightforward consequence of the denitions.

Proposition 1.32. Let (H, D, Φ, Ω) be a hermitian quantum eld with Wightman dis-
tributions W (m) . The functional
m
X
ω(f (0) ⊕ f (1) ⊕ . . . ⊕ f (m) ) = W (α) f (α) (1.5)
α=0

is a state of the BU-Algebra.

Proposition 1.33. Let (H, D, Φ, Ω) be a hermitian quantum eld. There is a unique
cyclic representation (H, D, π, Ω) such that
π(f1 ⊗ . . . ⊗ fm ) = Φ(f1 ) . . . Φ(fm ).

Further, π is the GNS representation of ω from (1.5).

Proof idea. First, we dene A0 ⊂ A generated by tensor products f1 ⊗ . . . ⊗ fm which
again is a ∗-algebra. The representation π is the GNS representation of ω restricted to
A0 by the denition of the Wightman distributions. Consider the GNS representation
(Hω , Dω , πω , Ωω ) of A with respect to the state ω . If we restrict πω to A0 we get a new
representation π̃ . We show that Ωω is cyclic for π̃ by using the following facts:

(a) The span of tensor products f1 ⊗ . . . ⊗ fm is dense in S (Rn × . . . × Rn ) by the N

representation theorem; cf. [RS80] Theorem V.13. Hence A0 is dense in A.

(b) The representation is continous in the sense that

S (Rn × . . . × Rn ) 3 f 7−→ hξ, πω (f )ψi

is in S 0 (Rn·m ) (by the nuclear theorem) for all ξ, ψ ∈ Dω .

Assume that Ωω π̃ . Then there is a 0 6= ξ ∈ Hω such that hπ̃(f )Ωω , ξi = 0
is not cyclic for
for all f ∈ A0 . But since πω is the GNS representation of A with respect to ω , the vector
Ωω is cyclic for πω , i.e. there is a f ∈ A such that hπω (f )Ωω , ξi 6= 0. By the denseness
of A0 in A there is a sequence of fk ∈ A0 k ∈ N converging to f ∈ A. Hence by the
continuity just stated we get

lim hπ̃(fk )Ωω , ξi = lim hπω (fk )Ωω , ξi = hπω (f )Ωω , ξi

k→0 k→0

in contradiction to hπ̃(fk )Ωω , ξi = 0 and hπω (f )Ωω , ξi =

6 0. Hence Ωω is a cyclic vector
for π̃ and the latter is therefore unitarily equivalent to the GNS representation of A0
with respect to ω. Hence there exists a unitary U : Hω → H such that

U πω (f1 ) . . . πω (fm )U ∗ = Φ(f1 ) . . . Φ(fm ).

16
 1 Wightman Quantum Field Theory

For arbitrary f ∈ S (Rn·m ) we dene Φ(f ) := U πω (f )U ∗ , which is the extension we are

looking for. The uniqueness follows from the fact that the matrix elements hξ, Φ(f )ψi for
ξ, ψ ∈ D0 are uniquely determined by ω.

Remark 1.34 (Fields and representations) . The previous results show that a quan-
tum eld (H, D0 , Φ, Ω) can be recovered (up to unitary equivalence) from its Wightman
distributions W
(m) . The W (m) dene a state ω on the BU-algebra A, this state has a

GNS representation (Hω , Dω , πω , Ωω ) which by Proposition 1.32 is unitarily equivalent

to the representation dened by Φ, i.e. there is a unitary U with U Φ(f1 ) · · · Φ(fm )U =
∗

πω (f1 ⊗ · · · ⊗ fm ) for all f1 , . . . , fm ∈ S (Rn ) and all m ∈ N. Hence we recover a unitarily

equivalent copy of Φ by

S (Rn ) 3 f 7→ Φω (f ) = πω (f ) ∈ L(Dω , Dω ). (1.6)

The last equation shows that we even have a little bit more. If ω is any continuous state
of A, it denes its GNS representation (Hω , Dω , πω , Ωω ) and via Eq. (1.6) a map Φω from
S (Rn ) to L(Dω , Dω ). Hence the 4-tuple (Hω , Dω , Φω , Ωω ) is a quantum eld, and would
be a hermitian quantum eld if the Φω (f ) would be essentially self-adjoint for real-valued
f . This is unfortunately something we can not get for free. Hence it is  at least in the
present context  advantageous to drop the self-adjointness condition from Def. 1.15. In
that case we get two one-to-one correspondences:

Hermitian quantum eld ←→ Representations of A

and

Sequences of Wightman distributions ←→ States of A.

Let us come back to Wightman elds. The next proposition tells us how the additional
properties translate into properties of the Wightman distributions.

Proposition 1.35 ([SW64], Chapter 3.3). Let (H, D, Φ, Ω, U ) a Wightman quantum

eld. Then the following statements hold.
(a) All the W (m) are Poincaré-invariant, so
W (m) ((a, Λ)f ) = W (m) (f )

where ((a, Λ)f ) = f (Λ−1 (x1 − a), . . .).

(b) (Spectrum condition). The Fourier transform Ŵ (m) of WP(m) has support in the

set of p = (p1 , . . . , pm ) ∈ R with j=1 pj = 0 and m + for any

n·m
Pm
j=k pj ∈ V
k = 2, . . . , m.
(c) (Locality). Whenever xk and xk+1 are spacelike separated, we have
W (m) (x1 , . . . , xk , xk+1 , . . . , xm ) = W (m) (x1 , . . . , xk+1 , xk , . . . , xm ).

17
 1 Wightman Quantum Field Theory

(d) (Cluster property).

lim W (m) (x1 , . . . , xk ,xk+1 + a, . . . , xm + a)
kak→∞

= W (k) (x1 , . . . , xk )W (m−k) (xk+1 , . . . , xm )

Remark 1.36 (Interpretation). The W (m) can be regarded as correlation functions. E.g.
W (2)  which is often called the 3-point function  describes the correlations between
two eld operators Φ(f ), Φ(g) in the vacuum. If W
(2) (f ⊗ g) factorizes, measurements

of Φ(f ) and Φ(g) in the vaccuum are uncorrelated. By the cluster property this happens
if the supports of f and g are very far apart. Spacial separation is, on the other hand,
not sucient. We only get invariance under permutations of f and g. A closer analysis
shows that the correlations of the elds between spacelike separated events in the vaccum
representation can be arbitrarily high, if the distance between the events is small enough.
Poincaré invariance shows that these correlations look always the same in all inertial
frames. In particular translations are interesting since it allows us to rewrite W (2) in
terms of a distribution W2 ∈ S (R4 ) such that we formally get
Z Z Z Z
(2) (2)
W (f ⊗ g) = W (x, y)f (x)g(y) dx dy = W2 (x − y)f (x)g(y) dx dy.
R4 R4 R 4 R4
Similarly we can rewrite W (m) in terms of a distribution Wm in S (R(m−1)·4 ). The Wm
have interesting analyticity properties which can be used to prove structural results like
the PCT theorem The spectrum condition translates to support properties in momentum
space.

The last result in this chapter combines the construction of a quantum eld from
a sequence of Wightman distributions discussed above with discussion from the last
proposition. In a nutshell it says that if the W (m) have all the properties from Proposition
1.35, the reconstructed eld is a Wightman QF.

Theorem 1.37 ([SW64], Theorem 3.7). Consider Rn = R4 . Let a state ω : A →

C of the BU-Algebra A be given and P consider the corresponding GNS representation
(Hω , Dω , πω , Ωω ). Further let ω(f ) = m
α=0 W
(α) (f (α) ) for W (α) ∈ S 0 (R4α ) and let

W (m) satisfy the four conditions from Proposition 1.35. Then there exists a strongly con-
tinous unitary representation Uω : P+↑ → U(Hω ) such that (Hω , Dω , πω , Ωω , Uω ) is a
Wightman quantum eld, although without the self-adjointness of the eld operators.

18
 2 The Free Scalar Field

2 The Free Scalar Field

2.1 Representation of the Poincaré Group
Denition 2.1. Let a group G, a C-vector space V and ξ : G × G → R be given. A map
U : G → GL(V ) is called projective representation of G if

U (f )U (g) = eiξ(f,g) U (f g).

The exponents ξ can not be freely chosen, but has to satisfy some contraints, which
can be easily deduced (e.g. look at U (f )U (g)U (h)). If all of them are satised, ξ is called
a cocycle.
By Wigner, a free, relativistic elementary particle is described by a projective uni-
↑
tary representation of P+ which actually has to be irreducible emphasizing on the term

elementary. So Wigners goal was to classify all the projective representations of

↑
P+ .
For more on this topic, we refer to [Bog+90, Chapter 7.2] and [RS75, Chapter IX.8].

The representations of
↑
P+ can be divided into classes via the mass, given by m2 ∈ R.
Here the physical case obviously is m
2 ≥ 0. All representations with m2 > 0 are fur-
ther characterized, up to unitary equivalence, by the spin s ∈ N0 = {0, , 1, , . . .} of
1 1 3
2 2 2
↑
the representation. There exists a unique unitary projective representation of P+ with
m2 > 0, s ∈ 21 N0 .
2
We will proceed with the easiest case m > 0, s = 0. The rst step is to dene the
mass shell

Hm = {p ∈ R4 | η(p, p) = m2 , p0 > 0} . (2.1)

Note, that Hm is an orbit of SO↑ (1, 3) since for Λ ∈ SO↑ (1, 3), p ∈ Hm we have Λp ∈ Hm .

Figure 4: The mass shell

from equation
(2.1) as part of
the forward light
cone.

We parametrize Hm by

R3 3 p 7−→ j(p) = (ω(p), p) ∈ Hm (2.2)

p
with ω(p) = kpk2 + m2 . This j can be regarded as a coordinate system such that
Hm becomes a smooth manifold. An only slightly more detailed analysis shows that it
actually is a smooth submanifold of R4 . This shows in paticular that Hm is a locally

19
 2 The Free Scalar Field

compact topological sapce which is homeomorphic to R3 . Hence we can consider Borel

subsets of Hm and dene the following Lorentz invariant measure.

Proposition 2.2 ([RS75], Thm IX.37) . For all Borel subsets Ω ⊂ Hm we write
Z
dp
Ωm (Ω) = .
j −1 (Ω) ω(p)

This denes a Borel measure on Hm which is Lorentz invariant: Ωm (ΛΩ) = Ωm (Ω) for
all Λ ∈ SO↑ (1, 3) and all Borel sets Ω ⊂ Hm .
With Ωm L2 (Hm , Ωm ) of square-integrable functions
we can dene the Hilbert space
on Hm . Since Ωm is Lorentz invariant, a transformation f ( · ) 7→ f (Λ
−1 · ) with a Lorenz
2
transformation Λ leads to a unitary operator U (Λ) on L (Hm , Ωm ), and therefore to a
↑
unitary representation of SO (1, 3) which is easily shown to be strongly continuous, i.e.
the map
SO↑ (1, 3) 3 Λ 7→ U (Λ)ψ ∈ L2 (Hm , Ωm ) (2.3)

ψ ∈ L2 (Hm , Ωm ). Combining this with a representation of the trans-

is continuous for all
↑
lation group leads to a representation of P+ .

Proposition 2.3. The map P+↑ 3 (b, Λ) 7→ U (b, Λ) ∈ U L2 (Hm , Ωm ) with

(U (b, Λ)ψ)(p) = eiη(b,p) φ(Λ−1 p).

is a strongly continuous, irreducible, unitary representation of the restricted Poincaré

group.
Proof. The proof is easy and therefore left as an exercise. Note that strong coninuity can
be dened as in Eq. (2.3).

Remark 2.4 (Interpretation) . The representation just constructed contains (almost)

everything we need for the quantum mechanics of one relativistic particle of mass m
and spin 0. Due to strong continuity we can dene the generators of the translations as
self-adjoint operators Pj , j = 0, . . . , 3

d iη(λej ,p)
Pj ψ = −i e ψ(p)
dλ λ=0

where the domains D(Pj ) are consisting of exactly those ψ for which the given limit
exists; cf. Remark 1.18. Explicitly we get

(P0 ψ)(ω(k), k) = ω(k)ψ(ω(k), k), (Pj ψ)(ω(k), k) = −kj ψ(ω(k), k) j = 1, 2, 3.

The Pj for j = 1, 2, 3 describe the three-momentum of the particle, P0 is its Hamiltonian.

P
With the reasoning already pointed out in Remark 1.18 the operator Pb = j bj Pj
describes the Hamiltoian in the inertial frame moving with four-velocity b ∈ H1 . Hence,

20
 2 The Free Scalar Field

we have almost everything we need for a quantum mechanical description. The only
missing component are the position operators. They do not come out of the representation
U directly, but they have to satisy a number of conditions. The corresponding analysis
was carried out by Newton and Wigner in 1949. The resulting theory has serious locality
problems as already mentioned in the introduction. We omit the discussion at that point,
but refer the reader to original work.

We can use the parametrization map j to pull everything back to the Hilbert space
L2 (R3 ). This is sometimes useful  in particular if we want to compare results with
expressions known from the physics literature.

Proposition 2.5. The map J : L2 (Hm , Ωm ) → L(R3 ) given by

(ψ ◦ j)(k) ψ(ω(k), k)
(Jψ)(k) = p = p
ω(k) ω(k)

is a unitary operator.
Proof. This is easily checked and therefore left as an exercise.

2.2 The Klein-Gordon Equation

Our goal is to quantize the Klein-Gordon equation. Therefore, we have to have a closer
look on the classical solutions and their relations to the discussion of the previous section.
Hence, let us start with

∂2Ψ
( + m2 )Ψ := − ∆Ψ + m2 Ψ = 0 (2.4)
∂t2
and
Ψ ∈ KC := {Ψ ∈ C 2 (R4 , C) | Ψt ( · ) := Ψ(t, · ) ∈ S (R3 )}. (2.5)

This somewhat strange function space arises from the desire to use the Fourier transform
with respect to the position variable x ∈ R3 . We are interested in solutions of the Cauchy
problem with initial data in S (R ):
3

Ψ(0, x) = f (x) and ∂t Ψ(0, x) = p(x) with f, p ∈ S (R3 ).

Using the shortcut notation Ψt (x) := Ψ(t, x) as in Eq. (2.5), now Ψ̂t is the Fourier
transform with respect to x∈ R3 . Fourier transforming all of (2.4) (in R3 ) gives

∂ 2 Ψ̂t (k)
+ kkk2 Ψ̂t (k) + m2 Ψ̂t (k) = 0.
∂t2
So for any k, we get an ordinary dierential equation of second order in t which yields
the solution

Ψ̂t (k) = b(k)eiω(k)t + c(k)e−iω(k)t

21
 2 The Free Scalar Field

p
with ω(k) = kkk2 + m2 > 0 for all k since m > 0. Using the initial conditions, we
easily get

fˆ(k) = b(k) + c(k) p̂(k) = iω(k)(b(k) − c(k))

and thus

1 ˆ i  1 ˆ i 
b(k) = f (k) − p̂(k) c(k) = f (k) + p̂(k) . (2.6)
2 ω(k) 2 ω(k)

Proposition 2.6. For all initial data f, p ∈ S (R3 ) there exists a unique solution Ψ ∈ KC
to the Klein-Gordon equation which is given by
Z 
1 i(k·x+ω(k)t) i(k·x−ω(k)t)

Ψ(t, x) = b(k)e + c(k)e d3 k (2.7)
(2π)3/2 R3

for
p all t ∈ R. The functions b(k), c(k) are derived from f, p as in (2.6) and ω(k) =
kkk2 + m2 .

Proof. We haven't shown yet that the presented solution really is in the set KC . To this
end rst note that ω and 1/ω are smooth functions. Furthermore they are polynomially
bounded, and the same is true for all its powers. Hence, for all f ∈ S (R3 ) the products
ω n f and ω −n f are Schwartz functions again. The same is true if we multiply for xed
t∈R with a phase factor e
±itω(k) . With this knowledge we conclude from Eq. (2.6) that

b and c are in S (R3 ). Hence the expression in Eq. (2.7) is for xed t the sum of the
inverse Fourier transform of two Schwartz functions, which is again a Schwartz function
in variable x for xed t.

Remark 2.7. For later use let us recall that we can multiply and divide Schwartz
functions by ω and still get Schwartz functions. This follows since ω(k) is bounded from
below by m > 0 (hence we avoid the singularities at zero) and polynomially bounded
from above (the same is obviously true for all powers of ω ). We have already used such
an argument in Example 1.8 to dene the product of a distribution and a polynomially
bounded function.

Now we want to look for a real-valued solution Ψ, f ∈ S (R3 , R). For the Fourier
transform this means
Z
1
fˆ(k) = f (x)eik·x dx,
(2π)3/2 R3

so fˆ(k) = fˆ(−k) which implies b(−k) = c(k). Putting this into the solution we have by
Proposition 2.6 gives us
Z Z
1 i(k·x+ω(k)t) 1
Ψ(t, x) = c(−k)e dk + c(k)ei(k·x−ω(k)t) dk.
(2π)3/2 R3 (2π)3/2 R3

22
 2 The Free Scalar Field

In the rst integral, we substitute

p k → −k , which produces no sign, so putting in a(k) =
2ω(k)c(k) gives the real-valued solution

Z 
1 −i(k·x−ω(k)t) i(k·x−ω(k)t)
 dk
Ψ(t, x) = a(k)e + a(k)e . (2.8)
(2π)3/2 R3
p
2ω(k)

Obviously the argument of the integral is real as sum of something and its complex
conjugate. Now let us introduce A ∈ L2 (Hm , Ωm ) by A = J −1 a, where J is unitary from
Proposition 2.5. We can express a in terms of A by

A(ω(k), k)
a(k) = (JA)(k) = p .
ω(k)

Inserting this into Eq. (2.8) and using the abbreviation ξ = (t, x), we get an integral over
the mass shell.
Z
1   dΩ(λ)
Ψ(ξ) = A(λ)eiη(ξ,λ) + A(λ)e−iη(ξ,λ) √ .
(2π)3/2 Hm 2
We summarize this discussion in the following two propositions.

Proposition 2.8. Consider the sets M = S (R3 )×S (R3 ) and S (Hm ) = J −1 S (R3 ) ⊂

L2 (Hm , Ωm ). For each (f, p) ∈ M the function Af,p : Hm → C with

1  
Af,p (ω(k), k) = √ ω(k)fˆ(k) + ip̂(k) (2.9)
2
is in S (Hm ). The corresponding map M 3 (f, p) 7→ Af,p ∈ S (Hm ) is real linear and
invertible. The inverse is given by S (Hm ) 3 A 7→ (fA , pA ) ∈ M with
Z 
1 ik·x −ik·x
 dk
fA (x) = a(k)e + a(k)e (2.10)
(2π)3/2 R3
p
2ω(k)
Z  p
i  ω(k)
pA (x) = a(k)eik·x − a(k)e−ik·x √ dk (2.11)
(2π) 3/2
R3 2
with a = JA and the unitary J from Proposition 2.5.
Proof. We have to show that JAf,p with (f, p) ∈ M and

!
1 i
ω(k)fˆ(k) + p
p
JAf,p (k) = √ p̂(k)
2 ω(k)

is in S (R3 ) . But this follows from the fact that f, p are Schwartz functions and Remark
2.7 above. Linearity of the map is obvious, and invertibility folows from the existence of
an inverse. That the given map is really the inverse is left as an exercise to the reader
(this can be easily done with slight modications of the calculations above).

23
 2 The Free Scalar Field

Proposition 2.9. For each A ∈ S (Hm ) = J −1 S (R3 ) ⊂ L2 (Hm , Ωm ) the expresssion

√ 1
Z
Ψ= 2 Re FA with (FA)(ξ) = eiη(ξ,λ) A(λ)dΩ(λ) (2.12)
(2π)3/2 Hm

hence Z
1   dΩ(λ)
Ψ(ξ) = A(λ)eiη(ξ,λ) + A(λ)e−iη(ξ,λ) √ . (2.13)
(2π)3/2 Hm 2
is a solution of the Klein-Gordon equation (2.4), and an element of the space K = {f ∈
Kc | f = f } with KC from Eq. (2.5). The corresponding initial data are given as (f, p) =
(fA , pA ) with the expressions from Proposition 2.8. Similarly, we get the solution with
initial data (f, p) ∈ M if A = Af,p holds.
Proof. That FA is a solution follows easily from

∂2
Z
1 dk
2
FA(t, x) = − ei(ω(k)t−k·x) ω(k)2 a(k) p
∂t (2π) 3/2
R3 ω(k)
Z
1 dk
(∆FA)(t, x) = ei(ω(k)t−k·x) kkk2 a(k) p .
(2π) 3/2
R3 ω(k)

With a = JA and the unitary J from Proposition 2.5. Hence Re FA is a real-valued

solution. The initial data for the solution FA are easily calculated as
Z
1 a(k)
(FA)(0, x) = e−ix·k p dk, (2.14)
(2π) 3/2
R3 ω(k)
∂ Z
 i −ix·k
p
FA (0, x) = e a(k) ω(k) dk (2.15)
∂t (2π)3/2 R3

and are obviously Schwartz functions (cf. Remark 2.7). Hence by Proposition 2.6 the
solution FA is in KC , its real part, obviously, is in K. The remaining statements follow
immediately from Eqs. (2.14), (2.15) and Proposition 2.8.

Remark 2.10 (Fourier transform). In the last proposition we have implicitly introduced
the map
Z
F : S (Hm ) → C (R ), A 7→ FA
∞ 4
with FA(ξ) = eiη(ξ,λ) g(λ) dΩ(λ)
Hm

and the domain

S (Hm ) = J −1 S (R3 ) ⊂ L2 (Hm , Ωm ).

Since J is unitary, S is an isomorphic copy of S (R3 ). Hence we can equip it with the
same topology such that it becomes a Fréchet space. This implies in particular that we
can dene the topological dual S 0 (Hm ) and embed S (Hm ) into S 0 (Hm ) via a map

24
 2 The Free Scalar Field

S (Hm ) 3 A 7→ φA ∈ S 0 (Hm ) with

Z
φA (f ) = A(λ)f (λ) dΩm (λ).
Hm

All our knowledge about tempered distributions can be translated to the mass shell
supported versions in S (Hm ) via the map J.
The map F can be regarded as some form of Fourier transform and as such it can be
extended to the distribution space S 0 (Hm ). This can be done in terms of

√ 1
Z
E : S (R ) 3 f 7→
4
2π f˜ Hm with ˜
f (λ) = eiη(ξ,λ) f (ξ) dξ. (2.16)
2π R4

Note that we have dened the Fourier transform f˜ for this particular purpose with the
Minkowski metric η rather than the usual scalar product. With the map E we write

√ Z
1
Z
φA (Ef ) = 2π 2
eiη(ξ,λ) f (ξ) dξ A(λ) dΩ(λ)
Hm (2π) R 4
Z  Z 
1 iη(ξ,λ)
= e A(λ) dΩm (λ) f (ξ) dξ
ZR 4 (2π)3/2 Hm
= FA(ξ)f (ξ) dξ = φF A (f )
R4
with φF A ∈ S 0 (R4 ) the regular distribution belonging to FA ∈ C ∞ (R4 )  please check
p
that this is well-dened although FA is not in L , cf. Example 1.6. Summarizing this
calculation we can conclude that S (Hm ) 3 φ 7→ φ ◦ E ∈ S (R ) is the (unique) weakly
0 0 4

contiunous extension of F to S (Hm ); cf. the discussion in Remark 1.7. The following
0

proposition shows how to recreate the real solution Re F in terms of the map E.

Proposition 2.11. For each A ∈ S (Hm ) the complex linear extension of the map
√ √ Z
S (R , R) 3 f 7→
4
2 Re φA (Ef ) = 2 Re Ef (λ)A(λ)dΩ(λ)
Hm

coincides with the regular distribution φ√2 Re F A

Z √
φ√2 Re F A (f ) = 2 Re FA(ξ)f (ξ)dξ
R4
√
associated to the solution 2 Re FA of the Klein-Gordon equation.

25
 2 The Free Scalar Field

Proof. We calculate φA (Ef ) with real-valued test function f:

√ 1  
2 Re φA (Ef ) = √ φA (Ef ) + φA (Ef )
Z2 Z
1
= A(λ) eiη(λ,ξ) f (ξ) dξ dΩ(λ)+
ZR
(2π) 3/2 4
Hm
Z
1
A(λ) e−iη(λ,ξ) f (ξ) dξ dΩ(λ)
Hm (2π)3/2 R4

Hence
√ Z √
2 Re φA (Ef ) = f (ξ) 2 Re(FA)(ξ) dξ,
R4
as claimed.

It is left as an Exercise to the reader that we can use this strategy to generate weak
(i.e. distributional) solutions of the Klein-Gordon equation. For us it is important to
remember this proposition when we introduce the free quantum eld in the next section.

Proposition 2.12. Consider (b, Λ) ∈ P+↑ and the action of P+↑ on functions f : R4 → C
given by f 7→ (b, Λ)f with [(b, Λ)f ](ξ) = f Λ−1 (ξ − b) . The map S (Hm ) 3 A 7→ Re FA

intertwines this action with the representation U (b, Λ) introduced in Proposition 2.3. In
other words
(b, Λ) Re FA = Re FU (b, Λ)A
holds for all (b, Λ) ∈ P+↑ and all A ∈ S(Hm ).
Proof. Ψ(ξ) = Re(FA)(ξ) and calculate
We write

Z 
−1 1 iη(Λ−1 (ξ−b),λ) −iη(Λ−1 (ξ−b),λ)

Ψ(Λ (ξ − b)) = A(λ)e + A(λ)e dΩm (λ)
(2π)3/2 Hm
Z 
1 iη(ξ−b,Λλ) −iη(ξ−b,Λλ)

= A(λ)e + A(λ)e dΩm (λ).
(2π)3/2 Hm

By substituting Λλ → λ̃, due to Lorentz invariance of the measure Ωm we get

Z
1  
Ψ(Λ−1 (ξ − b)) = A(Λ −1 λ̃)eiη(ξ−b,λ̃) + A(Λ−1 λ̃)e−iη(ξ−b,λ̃) dΩ (λ̃)
m
(2π)3/2 Hm
Z 
1 iη(x,λ̃) iη(b,λ̃)

−1 λ̃) + . . . dΩ (λ̃)
= e e A(Λ m
(2π)3/2 Hm | {z }
=(U (b,Λ)A)(λ̃)
 
= Re FU (b, Λ)A (ξ)

which concludes the proof.

Remark 2.13 (Canonical formalism). We can regard the space M = S (R, R)×S (R, R)
of initial data as the classical phase space of the system. The classical Hamilton function

26
 2 The Free Scalar Field

is given by
1 1
M 3 (f, p) 7→ h(f, p) = hp, pi + h(m2 − ∆)f, f i ∈ R
2 2
h · , · i denotes the real S (R3 , R), i.e. hf, gi = R3 f (x)g(x)dx.
R
where standard scalar product on
Using Gateaux dierentials we can easily dene partial derivatives of H as

 
∂ ∂ d
h(f, p) ∈ S (R ) with
3
h(f, p), v = h(f + λv, p) λ=0
∂f ∂f dλ
 
∂ ∂ d
h(f, p) ∈ S (R3 ) with h(f, p), w = h(f, p + λw) λ=0
∂p ∂p dλ

with v, w ∈ S (R3 ). It is easy to see that these partial dierentials exist in the given sense
and have the values

∂ ∂
h(f, p) = (∆2 − m2 )f with h(f, p) = p.
∂f ∂p

Now consider a curve R 3 t 7→ (ft , pt ) ∈ M which is dierentiable in the sense that

(t, x) 7→ ft (x) and (t, x) 7→ pt (x) are C 1 -functions on R4 . Then we can dene the time
derivative (f˙t , ṗt ) ∈ M in terms of partial derivatives

∂ ∂
f˙t (x) = ft (x), ṗt (x) = pt (x),
∂t ∂t
such that the Hamilton equations become

∂ ∂
f˙t = h(ft , pt ) = pt ṗt = − h(ft , pt ) = (m2 − ∆2 )ft . (2.17)
∂p ∂f
Dierentiating a second time and inserting the second equation into the rst shows that
(t, x) 7→ f (t, x) has to satisfy the Klein-Gordon equation. By Proposition 2.9 we can
therefore conclude that the system of equations (2.17) has a unique solution for all initial
data f0 = f , p0 = p. ψ ∈ K of the Klein-Gordon
We just take the unique solution
equation with initial data f, p and dene ft (x) = ψ(t, x) and pt (x) = ∂t ψ(t, x). Hence
Hamiltons equation with the Hamilton function h represent a reinterpretation of the
Cauchy problem for the Klein-Gordon equation as an innite-dimensional Hamiltonian
system; i.e. something like innite-dimensional classical mechanics.
We are discussing this topic since we want to justify the interpretation of M as the
classical phase space of our system, and f, p as the canonical variables. With substantially
more eort we could also introduce (in a mathematically rigorous way) Poisson brackets
and see that f, p satisfy some kind of canonical Poisson relations (maybe I will add this
in a future version of these notes). This reinterpretation in a canonical (i.e. Hamiltonian)
way is useful in the context of quantization. A possible strategy to quantize the eld
system we replace the classical elds f (x) and p(x) by operator-valued elds ϕ(x), π(x)
satisfying a version of canonical commutation relations, and generate the Hamiltonian
H of the quantum theory from h by replacing f and p with ϕ and π . The spacetime eld

27
 2 The Free Scalar Field

Φ(t, x) then is generated by evolving ϕ(x) with time evolution exp(itH).

While this procedure basically works (although with a substantial reinterpretation,
which is nesscessary in order to make the math work rigorously) this is not the most
adequate way to proceed. At that point the mass shell function Af,p ∈ S (Hm ) ⊂
L2 (Hm , Ωm ) comes into play. Since the map (f, p) 7→ Af,p is invertible, we can regard
the space S (Hm ) as an alternative version of the phase space, which is parametrized in
terms of complex variables A, A rather then f, p. Everything we have expressed in terms
of f, p can be reexpressed with A, A. E.g. the Hamilton function h can be written as a
function of a = JA and a as follows.
Z
1

h(a, a) = ω(k) a(k)a(k) + a(k)a(k) dk (2.18)
2 R3

It is left as an exercise to the reader to check this equation and to translate it into an
integral over Hm involving A, A rather than a, a.
The advantage of the variables A, A over f, p is the Poincaré covariance. The denition
of the initial data f, p requires a split of spacetime into space and time, and this xes an
inertial frame up to spatial rotations and translations. Hence, there is not one canonical
formalism, but there is (roughly speaking) a dierent one for each inertial frame. To
understand why this is dierent for the mass shell functions A, A, consider two inertial
frames Σ1 , Σ2 (b, Λ) transforming from Σ1 into Σ2 . If a
and a Poincaré transformation
eld is described in Σ1 by a spacetime function ψ it is described in Σ2 by (b, Λ)ψ . Hence
√
if ψ is given by A ∈ S via ψ = 2 Re FA the transformed eld is given by U (b, Λ)A; cf.
Proposition 2.12. Therefore the space S (Hm ) does not only contain one but all canonical
descriptions for all inertial frames and the unitaries U (b, Λ) represent the transformation
from ψ to (b, Λ)ψ . This intrinsic covariance is a great advantage if we are aiming at a
quantum theory which has a similar covariance  like a Wightman quantum eld theory.
Our strategy to quantize the Klein-Gordon equation is therefore to replace the func-
tions A, A in all expressions developed so far by appropriately chosen operator-valued
elds (satisfying some commutation relations which are motivated by the Poisson rela-
tions I have skipped). Applying this strategy in particular to the expression in (2.13)
leads to an operator-valued spacetime eld which (with an appropriate mathematical
reinterpretation of the steps just outlined) will become our Wightman eld.

2.3 The Free Quantum Field

First we need to make some general remarks on Fock spaces
12 . Let H be a separable
Hilbert space. The belonging Fock space then is given by

∞
M 
F(H) = C ⊕ H⊗n .
n=1

12
For more on the topic of Fock spaces of relativistic particles, we refer to [Bog+90, Chapter 7.3]

28
 2 The Free Scalar Field

Now for σ ∈ Sn , where Sn is the permutation group on n, we dene

Vσ : H⊗n → H⊗n σ(Ψ1 ⊗ . . . ⊗ Ψn ) = Ψσ−1 (1) ⊗ . . . ⊗ Ψσ−1 (n)

and based on this, the symmetrization operator

⊗n 1 X
Sn : H⊗n → H+ ⊂ H⊗n Sn = Vσ .
n!
σ∈Sn

⊗n
Note, that Sn is a projection. Now the range H+ of Sn is called symmetric or Bose
subspace. With this, we further dene the symmetric Fock space as follows.

∞
M 
F+ (H) = C ⊕ ⊗n
H+
n=1

With this reminder, we can return to the free quantum eld. We orient ourselves
towards [RS75, Chapter X.7]. For some f ∈H we dene

b− (f ) : H⊗n → H⊗(n−1) Ψ1 ⊗ . . . ⊗ Ψn 7−→ hf, Ψ1 iΨ2 ⊗ . . . ⊗ Ψn

A short calculation shows kb− (f )k = kf k so b− is bounded and it can be linearly extended

to all of H
⊗n for all n > 0. For n = 0 we dene b− (f ) = 0. It is easy to check that
b (f ) = (b (f ))∗ on product vectors acts like
+ −

b+ (f ) = (Ψ1 ⊗ . . . ⊗ Ψn ) = f ⊗ Ψ1 ⊗ . . . ⊗ Ψn .

With this, we want to dene creation and annihilation operators as already known from
the quantum harmonic oscillator.
For a densely dened self-adjoint operator H : D(H) → H on H we explain

F0 = {Ψ ∈ F+ (H) | ∃n∈N ∀m>n Ψ(m) = 0}

D̃H = {Ψ ∈ F0 | Ψ(n) ∈ D(H) ⊗ . . . ⊗ D(H) ∀n }

where F0 is the set of nite particle vectors. For Ψ ∈ D̃H we dene

(dΓ(H)Ψ)(n) = (H ⊗ 1 ⊗ . . . ⊗ 1 + 1 ⊗ H ⊗ 1 ⊗ . . . ⊗ 1 + . . .)Ψ(n) .

This is called second quantization. A similar construction can be done for unitaries U :
H→H where

(Γ(U )Ψ)(n) = (U ⊗ . . . ⊗ U )Ψ(n)

is said second quantization with

Γ(eitH ) = eit dΓ(H) .

29
 2 The Free Scalar Field

Now in order to dene creation and annihilation operators, we need the number operator
which is given by N = dΓ(1). For Ψ ∈ H+ ⊗n
we obviously have N Ψ = nΨ. Note that N
is essentially self-adjoint on F0 . Sadly, N is a global observable whereas the lab setting
obviously is local so on its own it is not as useful. This is why we dene the annihilation
operator on F+ (H) with domain F0 to be

√
A(f ) = N + 1b− (f ).

For each Ψ, Φ ∈ F0 we have

√ √ √
hΨ, A(f )Φi = hΨ, N + 1b− (f )Φi = h N + 1Ψ, b− (f )Φi = hS(b− (f ))∗ N + 1Ψ, Φi

which implies that creation operator is given by

√ √
A∗ (f ) = (A(f ))∗ = S(b− (f ))∗ N + 1 = N Sb+ (f ).

It is worth noting that both A(f ) and A∗ (f )|F0 are closable. We denote their closures by
A(f ) ∗
and A (f ).

Example 2.14. Consider a topological space with Borel measure (M, µ) and the asso-
ciated Hilbert space H = L2 (M, µ). By [RS80, Chapter II.4], we then have

⊗n
H⊗n = L2 (M
| × .{z
. . × M}, µ ⊗ . . . ⊗ µ) H+ = L2S (M × . . . × M, µ ⊗ . . . ⊗ µ)
n arguments

where L2S is the set of symmetric functions

13 . The operators A and A∗ are given by

√
Z
(A(f )Ψ)(n) (m1 , . . . , mn ) = n+1 f (m)Ψ(n+1) (m, m1 , . . . , mn ) dµ(m)
M
n
1 X
(A∗ (f )Ψ)(n) (m1 , . . . , mn ) = √ f (mj ) · Ψ(n−1) (m1 , . . . , m̂j , . . . , mn ).
n skip
j=1

Denition 2.15. The Segal quantization over H on F0 is dened via

1
H 3 f 7−→ ΦS (f ) = √ (A(f ) + A∗ (f )) ∈ F+ (H).
2

Note that the Segal quantization is R-, but not C-linear since f 7→ b− (f ) is an antilinear
map.

Theorem 2.16 ([RS75], Theorem X.41). The following statements hold.

(a) (Self-adjointness). For all f ∈ H, ΦS (f ) is essentially self-adjoint on F0 .

13
This means that L2S is the set of functions in L2 which are invariant under permutations of the
coordinates.

30
 2 The Free Scalar Field

(b) (Cyclicity of the vacuum). For the vacuum Ω = 1 ⊕ 0 ⊕ 0 ⊕ . . .

span{ΦS (f1 ) · · · ΦS (fn )Ω | f1 , . . . , fn ∈ H, n ∈ N} ⊂ F+ (H)

is dense.
(c) (Commutation relations). For each Ψ ∈ F0 , f, g ∈ H we have
[ΦS (f ), ΦS (g)]Ψ = i Imhf, giH Ψ.

Further, for the unitary operator w(f ) = exp(iΦs (f )) (Weyl operator) we have
W (f + g) = e−i Imhf,gi W (f )W (g).

(d) (Continuity). If fn → f in H, then

W (fn )Ψ → W (f )Ψ for all Ψ ∈ FS (H)
ΦS (fn )Ψ → Φs (f )Ψ for all Ψ ∈ F0
(e) (Covariance condition). For all unitary operators U on H,
Γ(U ) : D(ΦS (f )) → D(ΦS (U f ))

and for Ψ ∈ D(ΦS (U f )) we have

Γ(U )(ΦS (f ))Γ(U )∗ Ψ = ΦS (U f )Ψ

for all f ∈ H.
We can now use the Segal quantization to dene the free Hermitian scalar eld of
√
mass m. To this end note that we can write ΦS = 2 Re A∗ which is (almost) the same
expression we have already used in Proposition 2.11 to rewrite solutions to the Klein-
Gordon equation in a distributional sense. The only dierence is the quantity A which
is now an operator rather than a function (and complex conjugation becomes taking
adjoint). Hence with the map E from Equation (2.16), which was given by

√
E : S (R4 ) → L2 (Hm ) f 7−→ 2π f˜|Hm

using the 4-dimensional Fourier transform

Z
1
f˜(k) = f (x)eiη(x,k) dx,
(2π)2 R4

the free Hermitian scalar eld of mass m now is dened to be

√
Φm (f ) = ΦS (Ef ) = 2 Re A∗ (Ef )

31
 2 The Free Scalar Field

on f ∈ S (R4 , R) and

Φm (f ) = ΦS (Re(f )) + iΦS (Im(f ))

for arbitrary f ∈ S (R4 ). Applying the reasoning from Proposition 2.11 we see that
formally this Φm (f ) is an operator-valued solution of the Klein-Gordon equation. Exact
reasoning leads to the following theorem.

Theorem 2.17 ([RS75], Theorem X.42). The 5-tuple



F+ L2 (Hm ) , F0 , Φm , Ω, Γ U (·, ·)

is a Wightman quantum eld where

U (b, Λ)Ψ (p) = eiη(Λ,b) Ψ(Λ−1 p)

is the unitary representation of the restricted Poincaré group on L2 (Hm , Ωm ). Further

Φm ( + m2 )f = 0

holds for each f ∈ S (R4 ).

Proof. Most of the statement can be derived easily from earlier results, like the discus-
sion of the Klein-Gordon equation in Sec. 2.2 or properties of the Segal quantization in
Theorem 2.16. Only a few properties (like self-adjointness) require more work. In those
cases we refer the reader to the corresponding proof in [RS75].

Remark 2.18 (Free Hamiltonian) . Following the reasoning from Remark 1.18 we can
dene the free Hamiltonian of the theory (in the inertial system in which we are at rest)
as the generator of the time translations, i.e. as the self-adjoint operator H0 satisfying

Γ U (te0 , 1) = exp(itH0 ), ∀t ∈ R.

Using the properties of second quantization Γ(U ) this can be rewritten as

H0 = dΓ(P0 ) with (P ψ)(ω(k), k) = ω(k)ψ(k, ω(k)),

where ψ ∈ L2 (Hm , ωm ) has to be chosen such that P ψ is square-integrable again. This

gives the domain of P0 . We come back to H0 in Remark 2.24, where we derive a dierent
expression for it which is more familiar from the physics literature.

In Sec. 2.2 we have used the freedom to rewrite functions on the mass shell as functions
on R3 by using the unitary operator J : L2 (Hm , Ωm ) → L2 (R3 ) from Proposition 2.5. We
can do the same with the eld operators by applying the second quantization of J to
Φm (f ). For real-valued f this leads to

1
Φ̃m (f ) = Γ(J)Φm (f )Γ(J)∗ = √ (a(Ef ) + a∗ (Ef ))
2

32
 2 The Free Scalar Field

a∗ , a F+ L2 (R3 )


with the creation and annihilation operators on . We have used lower
case letters here, in order to make it easier to distinguish them from the corresponding
F+ L2 (Hm , Ω)


operators on . The relations to the latter are

f ◦ j  f ◦ j 
Γ(J)A(f )Γ(J)∗ = a √ Γ(J)(A(f ))∗ Γ(J)∗ = a∗ √ .
ω ω

F+ L2 (R3 )


Note that these are operators on . As before the nite particle vectors F0 ⊂
L2 (R3 )


F+ form the domain of these operators. In slight abuse of notation we here
have kept the same symbol (F0 ) we have already used for the nite particle vectors in
F+ L2 (Hm , Ωm )


, although strictly speaking both sets are dierent.

We now turn to the question whether we can evaluate Φ̃m at spacetime events rather
than test functions. The answer, basically, is yes, but the result of such an evaluation is
not an operator. To explain the details we need some preparations. (Also note that we
only treat the case Φ̃m . The mass shell based eld Φm can be treated similarly. To work
out the corresponding details is left as an exercise to the reader).
On the domain

DS = {Ψ ∈ F0 | Ψ(n) ∈ S ((R3 )n ) for all n}

we can dene the annihilation operator ak at momentum k via

√
(a(k)Ψ)(n) (k1 , . . . , kn ) = n + 1Ψ(n+1) (k, k1 , . . . , kn ), (2.19)

which is a well dened operator and related to a(f ) with test function f ∈ L2 (R3 ) by the
following Proposition.

Proposition 2.19. For all f ∈ L2 (R3 ) the annihilation operator a(f ) can be written as
Z
a(f ) = f (k)ak dk
R3
with ak from Eq. (2.19).
Proof. This is straightforward and again left as an exercise.

Basically, we would like to do something similar, but if we try to calculate the adjoint
a∗k of ak formally we get

n
1 X
(a∗ (k)Ψ)(n) (k1 , . . . , kn ) = √ δ(k − kj )Ψ(n−1) (k1 , . . . , k̂j . . . , kn ) (2.20)
n
j=1

with the obvious problem that δ occurs, so a∗ (·) is not a densely dened operator. We
can rectify this by using quadratic forms

Remark 2.20 (Quadratic forms. [RS80], Sec. VIII.6) . A quadratic form q on a Hilbert
space H is a map q : Q(q) × Q(q) → C, where Q(q) ⊂ H is a dense, linear subspace (the

33
 2 The Free Scalar Field

form domain ), and q is linear in the second and conjugate linear in the rst argument. If
X is an operator with dense domain D(X) we can immediately dene a quadratic form
X[φ, ψ] = hφ, Xψi with domain D(X). The converse is not true. There are quadratic
forms which do not belong to an operator (we will see an example very soon). Hence,
quadratic forms are more singular objects than operators. Nevertheless the notation
hψ, qφi is frequently used for q[ψ, φ] even if q is not an operator. In the following we will
use the phrase  X and q coincide in the sense of quadratic forms for an operator X and
a quadratic form q if q[ψ, φ] = hψ, Xφi holds for all ψ, φ in an appropriate domain.

Now assume a∗k would exist as an operator. Then we could assign it to the quadratic
form
a∗k [ψ, φ] = hψ, a∗k φi = hak ψ, φi ψ, φ ∈ DS
Hence, although a∗k is not an operator the quantity on the right hand side of this equation
only uses ak and is therefore well dened. We use this expression as the denition of a∗k
as a quadratic form. Now in analogy to Proposition 2.19 we get the following.
Proposition 2.21. For all f ∈ L2 (R3 ) we have
Z
a∗ (f ) = f (k)a∗k dk
R 3

in the sense of quadratic forms.

Proof. Again, this is straightforward.

Now, we can apply this to the eld Φ̃m and get the following result.

Theorem 2.22. The quantity

1
Z   d3 k
i(ω(k)t−kx) ∗ −i(ω(k)t−kx)
Φm (t, x) = e a k + e a k p (2.21)
(2π)3/2 R3 2ω(k)

is a well dened quadratic form on F+ L2 (R2 ) with domain DS . For a test function

f ∈ S (R4 ) it is related to the eld Φ̃m (f ) by

Z
Φ̃m (f ) = f (x, t)Φm (x, t)dxdt.
R4
Proof. Φm (t, x) is a well dened quadratic form is obvious from the deni-
The fact that
tions. The relation to Φm (f ) can be cheked with the arguments already used in the proof
of Proposition 2.11. [In a future version of these notes I might expand this proof.]

Remark 2.23 (Wick ordering). For this remark, cf. [Haa96, Chapter I.5.2]. The idea to
use quadratic forms in order to evaluate quantities at spacetime events can be extended
to powers of the eld like Φn (t, x) or more generally expressions involving polynomials

34
 2 The Free Scalar Field

in the ak , a∗k . (a∗k )n (ak )m we can easily use the same idea as
If we look in particular at
∗
above and just move all ak to the left under the scalar product:

hφ, (ak ∗)n (ak )m ψi = h(ak )n φ, (ak )m ψi =: (a∗k )n (ak )m [φ, ψ]. (2.22)

If we change the ordering of the ak , a∗k , however this procedure fails. In the product ak a∗k
we can not move a∗k to the left since ak and a∗k are not commuting. Hence we have to
move ak rst which would produce an (undened) operator a∗k in the left argument of
the scalar product. To understand the problem we are facing let us see what happens if
we calculate the vaccum expectation value hΩ, ak a∗k Ωi of ak a∗k with the formal expression
from Eq. (2.20).
Z
hφ, ak a∗k ψi = ha∗k Ω, a∗k Ωi = δ 2 (k − p)dp.
R3
The integral on the right hand side involves the square of the delta function which can
not be dened within distribution theory. Hence, the given vacuum expectation value
is just innite. Usually, physical quantities like energy should have vacuum expectation
value zero (naively speaking, the vacuum should just mean no physical particles). In that
sense the innities we see are just artifacts arising from wrong operator ordering. To
get the real physical quantities we just have to subtract these innities, and this can be
done by choosing the correct operator ordering. The only ordering where no problems
in terms of ininities arise is the one in Eq. (2.22). Hence the simple rule is: Whenever
we encounter a monomial in ak , a∗k is: move all creation operators to the left. This is
known as Wick or normal ordering and usually indicated by colons written to the left
and the right of an expression. E.g. to calculate : Φ(t, x)n : we formally expand Φ(t, x)n
∗
into a polynomial of ak , ak and in any monomial we get that way, we move all a∗k to the
left.

Remark 2.24 (Operator ordering and quantization). Operator ordering problems as the
ones just described are not uncommon in quantum theory and already happen within
the quantization of one non-relativistic particle. The (too) simple rule which tries to map
classical observables (functions on phase space) to quantum observables (operators on
a Hilbert space) is to replace the canonical phase space coordinates qj , pk by position
and momentum operators Qj , Pk qj , pk in a phase space
and to replace each occurence of
function F by these operators. But even if F is just a polynomial in qj , pk this procedure
is ambiguous since the qk , pk mutually commute such that qj pj = pj qj while Qj Pj and
Pj Qj are dierent operators. Hence, we need additional (physical) arguments in order to
make the operator ordering unambiguous. In the given example we might want to use
1
2 (Qj Pj + P j Qj ) since this combination is at least hermitian.
Based on these considerations we can describe the basic quantization rule for scalar
elds as follows:

1. Rewrite a classical eld observable (i.e. a function F :M→R on classical phase

space M; cf. the disicussion in Remark 2.13) as a polynomial in the complex valued
phase space variables a(k), a(k).

35
 2 The Free Scalar Field

2. Replace these functions by the creation and annihilation operators (or more pre-
cisely quadratic forms) ãk , ã∗k . (Note that we are adding a tilde to the operators
here in order to make the distinction from the functions a, a easier. We will drop
this tilde after this remark is nished).

3. Apply Wick ordering, in order to get rid of operator ordering ambiguities and the
corresponding innities. The choice of ãk , ã∗k as the proper replacement of a(k), a(k)
is motivated by a comparison of the commutation relations between ãk , ã∗k on the
one hand and the Poisson relations between a(k) and a(k) on the other.

This receipt provides a clear rule to turn any polynomial in a(k), a(k) into a quadratic
F+ L2 (R3 )


form on the Hilbert space . If we apply it to the classical solution of the
Klein-Gordon Equation from (2.8) we get the quantum eld Φ(t, x) as written in Eq.
(2.21). Another possible application is the classical Hamilton function h(a, a) from Eq.
(2.18).

Z Z
1 ∗
: h(ãk , ã∗k ) : = ∗
ω(p) : (ãk ãk + ãk ãk ) : dp = ω(k)ã∗k ãk dp = H0
R 3 2
| {z } R 3

harmonic oscillator

Please check yourself that (as a quadratic form) this really coincides with the free Hamil-
tonian H0 from Remark 2.18. Also note in this context that the number operator N can
be given by a similar expression as
Z
N= ã∗p ãp dp.
R 3

Finally note that exactly the same discussion can be given in terms of the mass shell
A(λ), A(λ) and the corresponding quadratic forms Ãλ , Ãλ on F+ L2 (Hm , Ωm ) .


functions
Working this out is left as an exercise.

For the last point in this section we have to talk about are the Wightman distributions
of Φm (t, x). They are calculated in the next theorem.

Theorem 2.25. For m, n ∈ N0 with m > 1 the Wightman distributions of the free eld
Φm from Theorem 2.17 are given by W (2n+1) = 0 and
X
W (2m) (f1 ⊗ . . . ⊗ f2m ) = W (2) (fσ(1) ⊗ fσ(2) ) . . . W (2) (fσ(2n−1) ⊗ fσ(2m) ) (2.23)
σ∈pair

where pair ⊂ S2n is the set of permutations which satisfy σ(1) < σ(3) < . . . < σ(2n−1)
and σ(2k+1) < σ(2k+2) for k > 0. The two-point function W (2) is given by
Z Z Z
1
W (2) (f ⊗ g) = eiη(y−x,p) f (x)g(y) dx dy dΩm (p),
2(2π)3 Hm R 4 R 4

36
 2 The Free Scalar Field

which can formally be rewritten as

Z Z
(2) 1
W (f ⊗ g) = ∆+ (y − x; m2 )f (x)g(y) dx dy
i R4 R4

where
d3 p
Z
i 2
∆+ (x; m ) = eiη(x,p) .
2(2π)3 ω(p)

Proof. The combinatorial structure of the W (n) given here is a consequence of the struc-
ture of the Segal quantization and therefore not special to the free eld. A corresponding
proof can be given using Wick's theorem. Typically, this is left as an exercise. In lack of
a good reference I am following this tradition. Please check validity of (2.23) yourself. To
√
calculate the two-point function we use Ef = 2π f˜ and further get

W (2) (f ⊗ g) = hΩm , Φm (f )Φm (g)Ωm i = hΦm (f )Ωm , Φm (g)Ωm i

2π ∗ ˜ ∗ 2π ˜
= ha (f ), a (ĝ)i = hf , giL2 (Hm )
2 Z Z 2 Z
2π 1 −iη(x,p) 1
= e f (x) dx eiη(y,p) g(y) dy dΩm (p)
2 Hm (2π)2 R4 (2π)2 R4
Z Z Z
1
= eiη(y−x,p) f (x)g(y) dx dy dΩm (p)
2(2π)3 Hm R4 R4

Written formally, this yields

Z Z
1
W (2) (f ⊗ g) = ∆+ (y − x; m2 )f (x)g(y) dx dy
i R4 R4

where

d3 p
Z
i 2
∆+ (x; m ) = eiη(x,p) ,
2(2π)3 ω(p)

as claimed.

Fields with this structure of the n-point functions are called generalized free elds or
quasi-free. The Wightman axioms imply that W (2) has to be of the following form
Z Z Z ∞
(2)
W (f ⊗ g) = ∆+ (x − y; m2 ) dρ(m) f (x)g(y) dx dy
R 4 R 4 0

where dρ(m) is a polynomial bounded measure. This is also called the Källen-Lehmann-
representation, see [RS75, Theorem IX.34].

37
 2 The Free Scalar Field

2.4 Time-Zero Fields

Before we can dene the time-zero eld, we need the following short result.

Proposition 2.26. For f ∈ S (R3 ) we dene the dual space element δf ∈ S 0 (R4 ) via
Z Z
(δf )(g) := δ(t)g(t, x)f (x) dt dx = g(0, x)f (x) dx.
R4 R3
Then the map E from (2.16) can be extended to distributions of the form δf .
Proof. Using the denition of δf we get
Z
1
(E(δf ))(ω(k), k) = δ(t)f (x)eiω(k)t e−ik·x dt dx
(2π)3/2 R4
Z
1
= f (x)e−ik·x dx = fˆ(k),
(2π)3/2 R3

so E(δf )(j(k)) = fˆ(k) with j from (2.2).

Denition 2.27. We dene the time-zero elds to be

ϕm (f ) = ΦS (E(δf )) πm (f ) = ΦS (iωE(δf ))

for f ∈ S (R3 , R) and its C-linear extension onto all of S (R3 ).

Φ
As in the spacetime eld

m we can use the unitary Γ(j) to transform ϕm and πm
into operators on F+ L (R ) . We dene
2 3

ϕ̃m (f ) = Γ(J)ϕm (f )Γ(J)∗ and π̃m (f ) = Γ(J)πm (f )Γ(J)∗ .

In analogy to Theorem 2.22 we can express ϕ̃m (f ) and π̃m (f ) as smeared out versions
of appropriately chosen quadratic forms.

Proposition 2.28. The time-zero elds ϕ̃m , π̃m can be written in terms of ak and a∗k
as quadratic forms via
Z
ϕ̃m (f ) = f (x)ϕm (x) dx

where
Z 
1 −ik·x ∗ ik·x
 dk
ϕm (x) = e ak + e ak p
(2π)3 R3 2ω(k)

and
Z
π̃m (g) = g(x)πm (x) dx
R3

38
 2 The Free Scalar Field

with
Z r
i 
−ik·x ∗ ik·x
 ω(k)
πm (x) = ω(k) e ak − e ak d3 k.
(2π)3/2 R3 2

Proof. In the case of ϕ̃m this follows immediately from Theorem 2.22 by setting t = 0.
In the case of π̃m we use (2.21) and take a time derivative to get

i
Z   d3 k
i(ω(k)t−kx) ∗ −i(ω(k)t−kx)
∂t Φm (t, x) = ω(k)e a k − ω(k)e ak
(2π)3/2 R3
p
2ω(k)

and thus

(∂t Φ̃m )(f ) = Φ̃m (iωf ) = Γ(J)ΦS (ωE(if ))Γ(J)+ .

Hence, setting t=0 again leads to the statement about π̃m .

Remark 2.29 (Time evolution) . We can look at the elds ϕm (x) and πm (x) as local
observables which we can evolve in time (i.e. we are in the Heisenberg picture). With the
free Hamiltonian H0 from Remark 2.18 we can dene

e−itH0 ϕm (f )eitH0 = ϕm,t (f ) e−itH0 πm (g)eitH0 = πm,t (g). (2.24)

This again yields

Z 
−itH0 itH0 1 −ik·x −itH0 ∗ itH0 ik·x −itH0 itH0
 dk
e ϕm (x)e = e e a e +e e ak e
(2π)3 R3 {zk
p
| } 2ω(k)
=eitω(k) a∗k

= Φm (t, x).

In other words we can reconstruct the spacetime eld Φm (x, t) from the

time-zero eld
and time evolution, which in turn is part of the representation Γ U (b, Λ) of the Poincaré
group. This observation leads to the question whether we can turn this into a strategy
↑
to costruct new models from a set of time-zero elds and a representation of P+ , or at
least a Hamiltonian. The latter depends on physical input like the type of interactions
we want to describe. For the elds, however, we'd like to have a universal construction,
which is completely independent from any dynamical input. We might even hope that
the time-zero elds developed for the free eld are already sucient. To explain up to
which degree this idea works or does not work is the main task of the rest of this section.

Remark 2.30 (Canonical commutation relations) . To understand the last remark a

bit better, let us have a short look at the quantum mechanical description of N non-
relativistic particles. The Hilbert space of this system is L (R ) with
2 n n = 3N and the most
important observables are position and momentum described by self adjoint operators

39
 2 The Free Scalar Field

Qj , Pk which are dened for ψ ∈ S (Rn ) and x ∈ Rn by

1 ∂ψ
Qj ψ (x) = xj ψ(x), Pj ψ (x) = (x). (2.25)
i ∂xj

The Qj and Pk should be reagarded as a the nite degree of freedom analog of the time-
zero elds ϕm (x), πm (x). The most important structural property of the Qj , Pk are the
canonical commutation relations, which (in their most elementary form) are given as

[Qj Qk ]ψ = [Pj , Pk ]ψ = 0, [Qj , Pk ]ψ = iδjk ψ (2.26)

for any ψ ∈ S (Rn ). These relations are the main reason why the Qj and Pk are chosen
in the given way: We are following Dirac's quantization rule which says: Quantization
means to replace Poisson brackets by operator commutators. The Qj and Pk should
be the quantum analogs of position and momentum coordinates qj , pk of the classical
phase space Rn × Rn and the relations in (2.26) exactly resemble the Poisson relations
between qj and pk . Large parts of the physics literature even tell the legend that up to
unitary equivalence the Qj , Pk from Eq. (2.25) are the only possible choice for operators
satisfying the commutation relations in (2.26). If we assume for the moment that this
is true (we will come to that in a minute), and if we accept that the CCRs in (2.26)
are mandatory requirements, we can conclude that the operators in (2.25) are the only
possible choice  completely independent from the dynamics our model should obey. The
latter is completely contained in the Hamiltonian of the system.
What can we learn from this for eld theory? First of all note that the expressions
for ϕm (x) and πm (x) in Proposition 2.28 resemble Eqs. (2.10) where we have expressed
the intial data (f, p) ∈ M for the classical solution in terms of the complex variables
a, a. According to the discussion in Remark 2.24 the elds ϕm (x) and πm (x) are the
quantizations of the canonical phase space variables f (x) and p(x). Since the latter are
the eld thoeretic replacement of phase space coordinates qj , pk (cf. our discussion in
Remark 2.13) this observation justies the claim that ϕm (x) and πm (x) are the eld
theoretic analogs of the operators Qj , Pk . Hence let us have a look at the commutation
relations they satisfy:

[ϕ̃m,t (f ), ϕ̃m,t (g)]ψ = [π̃m,t (f ), π̃m,t (g)]ψ = 0 [ϕ̃m,t (f ), π̃m,t (g)] = ihf, gi, (2.27)

where on the right hand side of the second equation we have used the scalar product in
L2 (R3 ), and ψ ∈ F0 is chosen arbitrarily. Rewriting this in a formal way we get the more
often used form

[ϕm,t (x), ϕm,t (y)] = [πm,t (x), πm,t (y)] = 0 [ϕm,t (x), πm,t (y)] = iδ(x − y). (2.28)

The latter version can be regarded as the natural continuous variable version of the
relations in (2.26). Therefore we proceed in analogy to the nite degree of freedom case
as follows. We declare the CCRs in Eq. (2.28), or maybe better the mathematically more
rigorous version in (2.27), as the fundamental quantization condition each choice of time-

40
 2 The Free Scalar Field

zero elds (for scalar eld theories) should satisfy. If there would be again a uniqueness
result, the time-zero elds constructed for the free eld would be the only choice and we
could proceed along the lines outlined in Remark 2.29. The big amounts of subjunctives
in the last sentence already indicate that there is a problem with this procedure to explain
why we have to clarify rst, in which sense the Qj and Pk are really unique. This step is
prepared by the following denition.

Denition 2.31. Consider a locally convex, real vector space V which also carries a real
scalar product h · , · i, and strongly continuous maps f 7→ U (f ), g 7→ V (g) from V into
the unitary group U(H) of a separable Hilbert space H. If for all f1 , f2 ∈ V the operators
U, V satisfy the Weyl relations

V (f1 + f2 ) = V (f1 )V (f2 )

U (f1 + f2 ) = U (f1 )U (f2 )
V (f )U (g) = U (g)V (f ) exp(hf, gi),

then {U, V } is called a representation of the Weyl relations over V .

Example 2.32 (Schrödinger representation). The central reference for all statements
concerning the Schrödinger representation is the book of Folland [Fol16]. Consider V =
Rn and the operators Qj , Pk from Eq. (2.25). Real linear combinations of the Qj and
of the Pk are self-adjoint operators (on appropriate domains containing S (Rn ) as a
subspace). Hence with f, g ∈ Rn we can dene

n n
! !
X X
U (f ) = exp i fj Qj , and V (g) = exp i gk Pk . (2.29)
j=1 k=1

By Stone's theorem [RS80, Sec. VIII.4] these maps are strongly continuous. It is also
easy to check that

n
!

X

U (f )ψ (x) = exp xj fj ψ(x), and V (g)ψ (x) = ψ(x + g)
j=1

hold. From the last equation it is easy to see that the U (f ), V (g) satisfy the Weyl
relations. Hence the maps Rn 3 f 7→ U (f ) and Rn 3 g 7→ V (g) form a representation of
the Weyl relation, which is called the Schrödinger representation. A deeper analysis also
shows that this representation is irreducible, i.e. the only bounded operators commuting
with all U (f ) and V (g) are multiples of the identity.

Theorem 2.33 (Stone-von Neumann). The Schrödinger representation is the only irre-
ducible representation of the Weyl relations over Rn up to unitary equivalence.
Proof. See [RS80, Thm VII.I.14].
Remark 2.34 (Canonical commutation relations). Assume self-adjoint operators Qj , Pk
are given such that

41
 2 The Free Scalar Field

1. we can dene U (f ), V (g) as in Eq. (2.29),

2. these U (f ), V (f ) are representation of the Weyl relations and

3. they are irreducible, then we can conclude according to the Stone-von Neumann
theorem that these Qj , Pk are unitarily equivalent to the choice in Eq. (2.25).

In that case the relations in (2.26) are an easy consequence of the Weyl relations. We will
rephrase this situation in the following by saying the Qj , Pk satisfy the Weyl form of the
CCR. If, however, we only know that the Qj , Pk are self-adjoint and satisfy the relations
in (2.26) we can not deduce the Weyl relations. All corresponding calculations you might
nd are formal and can not be made rigorous without additional assumptions. In general,
(2.26) is not even sucient to gurantee that U (f ) and V (g) can be dened according
to (2.29). As a consequence, there actually are examples for self-adjoint operators Q, P
satisfying (2.26), which are not unitarily equivalent to the version in (2.25); cf. [RS80,
Sec. VIII.5].
With this remark we can return to the time-zero elds. We dene a corresponding
representation of the Weyl relations  this time over V = S (R3 ) rather than V = Rn
and ask for uniqueness. The depressing answer is given in the next theorem.

Theorem 2.35. Let ϕm , πm be the time-zero eld and conjugate momentum of the free
scalar eld of mass m. Then
Um (f ) = exp(iϕm (f )) Vm (g) = exp(iπm (g))

is an irreducible representation of the Weyl relations over S (R4 , R). Further, (Um1 , Vm1 )
and (Um2 , Vm2 ) are unitarily inequivalent if the masses are dierent, so if m1 6= m2 .
Proof. [RS75, Thm X.46].

Remark 2.36 (Interpretation) . The theorem shows that we actually can consider the
Weyl form of the CCR as a mandatory requirement for time-zero elds, but this does not
lead to a unique choice. Even worse, the example for inequivalent representations given
depend on the mass. But mass is a dynamical parameter since it labels dierent versions
of the eld equations. Hence the representations of the Weyl realtions we have to deal
with in eld theory does contain dynamical information.
Remark 2.37 (Time evolution again) . The failure of unitary equivalence has another
striking consequence concerning time evolution. To explain what we've just lost, let us
have another look at the nite degree of freedom case. Here we can look at the observables
Qj , Pk in the Heisenberg picture and time dependent operators

Qj (t) = exp(−itH)Qk exp(itH), Pk (t) = exp(−itH)Pk exp(itH),

where H is a self-adjoint operator and the Hamiltonian of the model. If we assume the
other way round that time dependent, self-adjoint operators Qj (t), Pk (t) are given which
satisfy the Weyl form of the CCR at each instance of time, we can conclude from the
Stone-von Neumann theorem that there is a unitary T (t) satisfying T (t)Qj (0)T (t)∗ =

42
 2 The Free Scalar Field

Qj (t) and T (t)Pk (0)T (t)∗ = Pk (t). It is still unclear (without further knowledge) whether
these U (t) are given as exp(−itH) in terms of a Hamiltonian H , but at least we know
that the time evolution is given by unitaries on the same Hilbert space.
The lack of uniqueness in the eld theoretical case indicates therefore that we can not
expect that time evolution is unitary. This leads to the urgent question whether we can
preserve uniqueness of the Weyl relations at least on a more abstract level. The rest of
this section is devoted to an answer of this question. As a preparation we need some
material about C*-algebras.

Denition 2.38. Consider the space B(H) of bounded operators on a (not necessarily
separable) Hilbert space H. A linear subspace A which is closed under products14 and
adjoints15 is called a C*-algebra.
Denition 2.39. An invertible linear map α : A1 → A2 between two C*-algebras A1 ,
A2 is called *-isomorphism if α(AB) = α(A)α(B) and α(A∗ ) = α(A)∗ holds for all
A, B ∈ A1 . If A1 = A2 = A, the map α is called *-automorphism.

Remark 2.40 (C*-algebras). The central references for all statements about C*-algebras
are the books of Bratteli and Robinson [BR12; BR02]. The following is a short list of
additional remarks.

1. C*-algebras can be dened alternatively in an abstract way. We have chosen the

explicit form as algebras of operators since it leads us more directly to the desired
goal.

2. C*-algebras are in particular *-algebras. Hence all the corresponding material from
Sec. 1.3 applies. This in particular concerns states and representations.

3. Consider the GNS representation (Hω , Dω , πω , Ωω ) of a state ω of a C*-algebra A.

4. In the denition of *-isomorphisms we haven't added any boundedness require-
ments, because this is not necessary. All *-isomorphisms α satisfy kα(A)k = kAk
for all A ∈ A1 . Hence they are automatically bounded (k ·k here denotes the
operator norm).

Denition 2.41. Consider a real pre-Hilbert space (K, h · , · i) which is also a locally
convex space, and a representation {U, V } of the Weyl relations. The smallest C*-algebra
A containing the operators U (f ), V (g) for all f, g ∈ K is called the Weyl-algebra of
{U, V }.

Theorem 2.42. The Weyl algebras A1 , A2 of two representations {U1 , V1 } and {U2 , V2 }
over the same space (K, h · ,
· i) are *-isomorphic. More

precisely, there is a *-isomorphism
α : A1 → A2 with α U1 (f ) = U2 (f ) and α V1 (g) = V2 (g) for all f, g ∈ K.

Proof. [BR02, Theorem 5.2.8]

14
A, B ∈ A ⇒ AB ∈ A
15
A ∈ A ⇒ A∗ ∈ A

43
 2 The Free Scalar Field

Remark 2.43 (Interpretation) . Assume that by some construction we got time-zero

elds ϕt and πt which by all instances of time satisfy the Weyl form of the CCR, then


by the previous theorem we get a family of *-isomorphisms α(t) αt exp(iϕ0 (f )) =
with
exp(iϕt (f )) and similarly for πt . These αt take the role of the unitaries T (t) from Remark
2.37, and therefore we can hope for a dynamical description of eld theories in terms of
automorphisms of C*-algebras, rather than unitaries on Hilbert spaces. Or to formulate
this in another way: The Hilbert space structure which served us well in quantum me-
chanics is too rigid for eld theory and has to be replaced by C*-algebras. This point
of view is quite successful at least on the conceptual side; cf. the book of Haag [Haa96]
for an in-depth discussion of this point of view. For constructive purposes, however, our
problems are not completely solved. It might (and does) in particular happen that time-
zero elds do not exist at all, since their existence is not a consequence of the Wightman
axioms. For the most non-trivial example of an interacting model  quartic self interac-
tions in 1+1 dimensions  the material developed so far is sucient. We consider this
case in the next chapter.

44
 3 Interactive Fields

3 Interactive Fields
3.1 Naive Approach
For the rst part of this chapter, we orient ourselves towards [RS75, p.233 .]. An  in-
teracting eld theory is a eld theory satisfying the Wightman axioms which has a
nontrivial scattering theory. A natural way to construct such elds (cf. the discussion in
the last section; in particular Remark 2.29) is to try to make a pertubation of some free
theory, so

ϕt (x) = e−itH ϕ(x)eitH πt (x) = e−itH π(x)eitH

for some Hamiltonian H . This works if H is self-adjoint but otherwise most likely breaks
the Poincaré invariance.
In classical Lagrangian eld theory the simplest Hamiltonians are of the form
Z
H = H0 + λ F (ϕ(x)) dx
| R {z
3
}
=HI

where F is some function, say a polynomial with some coupling constant λ. Since we want
the Hamiltonian to be bounded below, we expect that the polynomial is of even order
with positive highest coecient. The most simple non-trivial case then is F (x) = x4 , so

Z
H = H0 + λ ϕ(x)4 dx.
R 3

This expression does not make sense unless we consider the Wick-ordered case (see Chap-
ter 2.3)
Z
H = H0 + λ : ϕ(x)4 : dx,
R3
which is at least a quadratic form. Unfortunately, this quadratic form does not arise from
an operator since formally computing HΩ “ = “ Ψ for Ψ(n) = 0 if n 6= 4 gives
  P 
exp − ix 4i=1 ki 4
P
Z δ i=1 ki
Ψ(4) (k1 , . . . , kn ) = Q4 dx = Q4
R3 i=1 (2π) (2ω(ki ))
3/2 1/2 (2π) 9/2
i=1 (2ω(ki ))
1/2

which certainly is not in L2 , alone because of the δ -function. This we can try to x by
smearing with a test function, so

  P 
g(x) exp − ix 4i=1 ki 4
P
Z ĝ i=1 k i
Q4 dx = Q4 ,
R3 i=1 (2π)
3/2 (2ω(ki ))1/2 (2π) 9/2
i=1 (2ω(k))
1/2

but we still do not get an L2 function because ω(ki ) grows too slowly at innity.

45
 3 Interactive Fields

3.2 The Cuto Hamiltonian

In order to obtain an operator, we restrict ourselves to one space dimension. The dis-
cussion of Wightman quantum elds and the construction of the free eld has to be
adopted accordingly. This is, however, staightforward and (again) left as an exercise for
the reader. We will try to sketch the idea of the construction of the (ϕ4 )2 model from
[GJ68; GJ70a; GJ70b] and accompanying papers.
4 4
R R
We replace the quadratic form
R : ϕ(x) : dx by
R g(x) : ϕ(x) : dx where g(x) is a
real-valued function in L2 (R). So we have

Z Z
Hg = H0 + HI (g) = ω(k)a∗k ak dk + g(x) : ϕ(x)4 : dx − Eg
R R
with domain

D(Hg ) = DS = {Ψ ∈ F0 | Ψ(m) ∈ S (Rn·m )}.

Here the constant Eg , also called self energy of the vacuum, is chosen so that

0 = inf{spectrum Hg }.

Further Eg is nite because of the spatial cuto and because of the limitation to only
one space dimension.
For the sake of locality, for g one often chooses some smooth function of compact
support which equals one on a very large interval. Either way, the eect of g is to turn
o the interaction for large values of x. Therefore, g is called the space cut-o and H(g)
is called the spatially cut-o Hamiltonian for the (ϕ4 )2 eld theory.
Theorem 3.1 ([RS75], Theorem X.62). The spatially cut-o Hamiltonian Hg is essen-
tially self-adjoint for any g ∈ S (R, R).
Now the solution to the cuto eld equation

∂2Φ ∂2Φ
− + m2 Φ + 4λgΦ3 = 0, (3.1)
∂t2 ∂x2
also compare (2.4), is given by the eld at time t

ϕt,g (x) = exp(−itHg )ϕ(x) exp(itHg )

with corresponding momentum at time t

πt,g (x) = exp(−itHg )π(x) exp(itHg ).

Here ϕ(x) and π(x) are chosen to be the time-zero elds of the free eld. Please keep
this in mind, since this choice will cause considerable trouble below. Since translations
aect the cuto and therefore the eld, this does not belong to a Wightman eld because
it violates Lorentz invariance. The simplest idea would be to take the limit g → 1.

46
 3 Interactive Fields

Unfortunately, this means Eg → ∞ so our Hamiltonian does not converge in a proper

sense.

Figure 5: The eld ϕt,g (x) in two-

dimensional space time. If
g(y) = 1 for |y| ≤ |x| + |t|,
then the eld is indepen-
dent of g. This means for
a time t1 , we get an asso-
ciated interval (a, b) such
that ϕt,g (x) = ϕt (x) on
the subset of the rhombus
(a, b) × (−t1 , t1 ).

However ϕt,g (x) is independent of g provided g(y) = 1 for |y| ≤ |x| + |t| as is illustrated
in Figure 5. This is a consequence of the fact that the eld equation (3.1) propagates
disturbances only with the speed of light (here set to 1), and therefore any changes to
g made far outside can not inuence anything in the rhombical region shown in the
gure. Using this idea we can prove thr following

Theorem 3.2. Consider t1 ∈ R+ and f1 , f2 ∈ S (R), with support in a bounded interval

(a, b). Furthermore assume that g ∈ S (R, R) satises g(x) = 1 for all x ∈ (a−t1 , b+t1 ). If
we choose the time-zero elds ϕ(f1 ) and π(f2 ) of the free eld we can dene the quantities
ϕt (f1 ) = exp(−itHg )ϕ(f1 ) exp(itHg )
(3.2)
πt (f2 ) = exp(−itHg )π(f2 ) exp(itHg )

for all t ∈ [−t1 , t1 ]. They are self-adjoint operators (on an appropriate domain) and are
independent of the cuto g, provided it satises the given condition.
Proof. [GJ68, Sec. V] together with Lemma 3.2.2 and Theorem 3.2.6 of [GJ70a].

Hence, by choosing the interval, on which g(x) = 1 holds, large enough, we can dene
ϕt (f1 ) and πt (f2 ) for all t and all f1 , f2 ∈ D(R)16 . These elds have the cuto removed.
From here on we can introduce spacetime elds (i.e. elds which are smeared over
space and time rather than over space at a xed time) by rst dening

Z Z
Φg (f ) = ϕt,g (ft )dt = f (t, x)ϕg (t, x)dxdt with f ∈ D(R2 ), ft ( · ) = f (t, · ),
R R
then proving that this leads to a closable operator admitting an invariant dense domain
[GJ70a, Theorem 3.2.3]. Using Theorem 3.2 we can show that this operator is independent
of g provided the support of f is contained in the rhombical region from Figure 5; cf.
[GJ70a, Thm 3.2.6]. In that way we get an operator-valued distribution on the Hilbert

16
D(R) = {f ∈ C ∞ (R) | supp f compact}

47
 3 Interactive Fields

F+ L2 (R)


space with an invariant dense domain we haven't specied (cf. [GJ70a, Sect
3.2] for details; cf also [GJ70a, Thm 3.2.7] for the continuity requirements a distribution
has to satisfy).
Z
D(R ) 3 f 7→ Φ(f ) =
2
f (t, x)ϕ(t, x)dxdt, (3.3)
R2
which nally can be extended to test functions from S (R2 ), i.e. it is a tempered operator-
valued distribution [GJ70a, Sec. 3.4]. We can recover the time-zero elds from Theorem
3.2 by restricting Φ and Φt to test functions δf ; i.e.
Z Z
φt (f1 ) = f1 (x)Φ(t, x)dx and πt (f2 ) = f2 (x)∂t Φ(t, x)dx.
R R
For the second equation cf. [GJ70a, Theorem 3.2.5]. Note that this shows in particular
that the free eld and the eld Φ just constructed share the time-zero elds at t = 0.
Furthermore it can be shown that this Φ(f ) satises the correct eld equations (Theorem
3.2.5 and the following remark of [GJ70a]), is self-adjoint if f is a real-valued function
[GJ70a, Sec. 3.3], and is covariant with respect to spacetime translations [GJ70a, Sec.
3.6]. Furthermore, two operators Φ(f1 ) and Φ(f2 ) commute if the supports of f1 and f2
are spacelike separated [GJ70a, Sec. 3.5]. Hence we have almost constructed a Wightman
eld. The most important property we haven't shown yet (apart from covariance with
respect to Lorentz boosts) is the existence of a vacuum, and actually, a vacuum vector
Ω ∈ F+ L2 (R) does not exist.

3.3 Haag's Theorem

The fact that the above approach did not work out is not surprising but this was predicted
by Haag's theorem which we will state now.

Theorem 3.3 ([Bog+90], Theorem 9.28). Let Φ, Φ0 be two scalar Wightman elds acting
on Hilbert spaces H, H0 . Assume for some instance of time t we have time-zero elds
Φ(t, x), ∂t Φ(t, x) Φ0 (t, x), ∂Φ0 (t, x)

which are irreducible and there exists unitary T such that

T Φ(t, x)T ∗ = Φ0 (t, x)
(3.4)
T ∂t Φ(t, x)T ∗ = ∂t Φ0 (t, x).

Then all Wightman functions up to n = 4 are identical. If Φ0 is a free eld, then Φ is

also free eld in the sense that all Wightman functions are identical.
We can choose for Φ the eld constructed in (3.3) and the free eld for Φ0 . The method
outlined in the last section used the time-zero elds for the free eld to construct the
interacting eld Φ. Φ(0, x) = Φ0 (0, x) and ∂Φ(0, x) = ∂Φ0 (0, x); cf. the
Hence we have
denition of ϕt and πt in Eq. (3.2) and check that πt (x) = ∂t Φ(t, x) really holds (in this
context also cf. [GJ70a, Theorem 3.2.5]). Hence the condition in (3.4) is satised for t = 0

48
 3 Interactive Fields

and T = 1. Since Φ0 is the free eld and Φ is not, the latter can not be a Wightman
eld. We are seeing here consequences of the non-uniqueness of representations of the
Weyl relations, as discussed in Sec. 3.3.

3.4 Remove the Cuto

The previous section clearly tells us that we have to come up with another construction
to remove the cut-o. Although there is no vacuum vector for the eld ϕt from (3.4),
F+ L2 (R)


there is a vacuum vector Ωg in for Hg :

Theorem
3.4. For each cut-o g there is (up to scalar multiples) a unique vector Ωg ∈
F+ L2 (R3 ) which is normalized (kΩg k = 1) and satises Hg Ωg = 0.

Proof. [GJ70a, Theorems 2.2.1 and 2.3.1]

g → 1 in the Fock space F+ L2 (R3 )

As partially stated before, simply taking the limit
yields the following problems.

• Eg → ∞ so the Hamlitonian does not converge

• Ωg → 0 in the weak sense

The way out of this conundrum is to follow the idea outlined in Remark 2.43 and to
dene the eld dynamics algebraically, rather than in terms of unitaries. The rst step
is to dene a C*-algebra A in terms of the elds Φ(f ) from Eq. (3.3).

Denition 3.5. With H = F+ L2 (R2 ) we dene A ⊂ B(H) as the smallest C*-algebra17

containing all unitaries exp(iΦ(f )) for arbitrary f ∈ D(R2 , R); i.e. smooth real-valued
functions with compact support.
Proposition 3.6. There is a unique one parameter group αt of automorphisms αt : A →
A satisfying



αt exp iΦ(f ) = exp(itHg ) exp iΦ(f ) exp(−itHg ),
where g(x) = 1 holds on a suciently large interval.
Proof. [GJ70a, Sec. 4].

This automorphism group describes the time evolution of the interacting elds alge-
braically. In a similar way we can proceed with the vacuum.

Denition 3.7. For any A ∈ A, we dene the expectation values

ωg (A) = hΩg , AΩg i.

17
We are glancing over some technicalities here, because the original construction in [GJ70b] is a bit
more involved. However, the simplication we are applying here shouldn't cause big dierences.

49
 3 Interactive Fields

Theorem 3.8. The following statements hold.

(a) ωg is a state of A.
(b) There is a sequence (gn )n∈N of cut-os such that g→∞ and there is a state ω
such that limn→∞ ωgn (A) = ω(A) for all A ∈ A.
Proof. [GJ70b, Theorem 2.1].

Having this ω the idea is to get back a eld which now should be a Wightman eld. The
idea is to use the GNS representation of the state ω just constructed. Hence we have a
Hilbert space Hω , a *-morphism πω : A → B(Hω ), and a cyclic vector Ωω ∈ Hω such
that ω(A) = hΩω , πΩ (A)Ωω i.

Theorem 3.9. There is a positive self-adjoint operator Hω on Hω such that

and Hω Ωω = 0


exp(itHω )πω (A) exp(−itHω ) = π αt (A) (3.5)

Proof. [GJ70b, Theorem 2.1].

Remark 3.10. (a) The rst statement in (3.5) says that the automorphism group αt
is unitarily implemented in the representation πω , and the corresponding unitary
group is generated by the operator Hω which is the renormalized Hamiltonian of the
interacting theory. It is the limit of Hg in the sense that

hπw (B)Ωw , eiHw t πw (A)Ωi

is obtained through a limit as g→1 of

hBΩg , eiHg t AΩg i.

(b) The GNS vacuum Ωω is, again by Eq. (3.5) an eigenvalue of H with eigenvalue 0.
Since H is positive, Ωω is eigenvector with the lowest energy.

(c) We call Ωω the physical vacuum and πω the physical representation. The vectors in
Hω are called physical vectors.

Now let us come to the nal step, which is the reconstruction of the elds in the new
representation. This is done in terms of the following lemma.

Lemma 3.11. Consider f ∈ S (R2 , R). The one parameter unitary group
R 3 λ 7→ Vλ = πω exp iλΦ(f )



∈ U(Hω )

is strongly continuous.
Proof. This is a consequence of the locally Fock property proved in [GJ70b, Thm 2.2];
cf. also the corresponding remarks in Sec. 1 of [GJ70b].

50
 3 Interactive Fields

Now we can introduce the renomalized elds as

1 d
Φren (f )ψ = Vλ ψ ,
i dλ λ=0

with the unitaries Vλ from the preceding lemma. The set of ψ ∈ Hω for which the limit
exists forms the domain of Φren (f ). Now all properties of the eld Φ(f ) constructed in
Sec. 3.2 can be carried over. To see this we again have to use the locally Fock property
of πω from [GJ70b, Thm 2.2]. This basically nishes the construction. Let us summarize
the result.

Theorem 3.12. The elds Φren (f ) admit an invariant dense domain D ⊂ Hω , and a
strongly continuous repesentation P+↑ 3 (b, Λ) 7→ U (b, Λ) ∈ U(Hω ) such that the time
translations coincide with exp(itHω ). The ve-tuple (Hw , D, Φω , Ωω , U ) is a Wightman
quantum eld.
Proof. This is not explicitly shown in [GJ70b] but implicitly discussed [GJ70b, Sec. 1].
Basically, all properties can either be carried over from the elds Φ(f ) by using [GJ70b,
Thm 2.2], or they are consequences of related properties of the GNS representation. The
only missing point is the representation U . The space translations are treated in [GJ70b]
along the same lines as the time evolution exp(itHω ). The whole Poincaré group is treated
in [CJ70]. Together with the positivity of Hω (which generates the time translations) this
also shows that the spectrum condition holds.

Remark 3.13. We again can use Haag's theorem to analyze this construction. Since
the Wightman eld we got is not the free eld, by Theorem 3.3 said eld has to be
unitarily inequivalent to the free eld for all times. Hence, the renormalization process
just outlined, resulted in a change of representations of the CCR. The original one was
the vacuum representation of the free eld, which does not contain a Hamiltonian for the
interacting model. The new representation is the vacuum representation of the interacting
eld and it does not contain the free Hamiltonian. A similar statement holds for the
vacuum. The renormalized vacuum Ωω can not be described in terms of density operators
in the vacuum representation of the free eld. Roughly speaking, this is caused by the
fact that we can say that this new vacuum contains innitely many particles of the free
theory, and therefore it is not dened in our original Fock space (which only contains
states with nitely many particles).

51
Bibliography
[Bog+90] N.N. Bogolubov et al. General Principles of Quantum Field Theory. Mathe-
matical physics and applied mathematics. Kluwer, 1990.

[BR02] O. Bratteli and D.W. Robinson. Operator algebras and quantum statistical
mechanics. Vol. 2: Equilibrium states. Models in quantum statistical mechan-
ics. Springer, 2002.
[BR12] O. Bratteli and D.W. Robinson. Operator algebras and quantum statistical
mechanics. Vol. 1: C*-and W*-Algebras. Symmetry Groups. Decomposition
of States. Springer Science & Business Media, 2012.
[CJ70] J.T. Cannon and A. Jae.  Lorentz covariance of the λ(φ4 )2 quantum eld
theory. In: Communications in Mathematical Physics 17.4 (1970), pp. 261
321.

[Fol16] G.B. Folland. Harmonic Analysis in Phase Space.(AM-122). Vol. 122. Prince-
ton university press, 2016.

[GJ68] J. Glimm and A. Jae.  A λ(φ4 )2 Quantum Field Theory without Cutos.
I. In: Physical Review 176.5 (1968).

[GJ70a] J. Glimm and A. Jae.  The λ(φ4 )2 Quantum Field Theory Without Cut-
os: II. The Field Operators and the Approximate Vacuum. In: Annals of
Mathematics. Second Series 91 (1970).

[GJ70b] J. Glimm and A. Jae.  The λ(φ4 )2 Quantum Field Theory Without Cutos:
III. The Physical Vacuum. In: Acta Mathematica 125 (1970).
[Haa96] R. Haag. Local Quantum Physics. Theoretical and Mathematical Physics.
Springer-Verlag Berlin Heidelberg, 1996.

[RS75] M. Reed and B. Simon. Fourier Analysis and Self-Adjointness. Vol. 2. Meth-
ods of Modern Mathematical Physics. Academic Press, 1975.

[RS80] M. Reed and B. Simon. Functional Analysis. Vol. 1. Methods of Modern

Mathematical Physics. Academic Press, 1980.

[SW64] R.F. Streater and A.S. Wightman. PCT, Spin and Statistics, and all that.
Mathematical physics monograph series. W.A. Benjamin, 1964.

52