Lectures on mathematical foundations of QFT

Wojciech Dybalski
February 7, 2018

Introduction
The central question of these mathematical lectures is the following:
• Is QFT logically consistent?
Although it may not seem so, this question is quite relevant for physics. For
example, if QFT contained a contradiction and, say, the magnetic moment of the
electron could be computed in two different ways giving two completely different
results, which of them should be compared with experiments? It turns out that
such a situation is not completely ruled out in QFT, since we don’t have enough
control over the convergence of the perturbative series. If we take first few terms
of this series, we often obtain excellent agreement with experiments. But if we
managed to compute all of them and sum them up, most likely the result would
be infinity.
For this and other reasons, the problem of logical consistency of QFT fascinated
generations of mathematical physicists. They managed to solve it only in toy
models, but built impressive mathematical structures some of which I will try to
explain in these lectures.
The strategy to study the logical consistency of QFT can be summarized as
follows: Take QFT as presented in the physics part of this course. Take the
whole mathematics with its various sub-disciplines. Now try to ‘embed’ QFT into
mathematics, where the problem of logical consistency is under good control. The
‘image’ of this embedding will be some subset of mathematics which can be called
Mathematical QFT. It intersects with many different sub-disciplines including al-
gebra, analysis, group theory, measure theory and many others. It differs from
the original QFT in several respects: First, some familiar concepts from the phys-
ical theory will not reappear on the mathematics side, as tractable mathematical
counterparts are missing. Second, many concepts from mathematics will enter
the game, some of them without direct physical meaning (e.g. different notions
of continuity and convergence). Their role is to control logical consistency within
mathematics.
It should also be said that the efforts to ‘embed’ QFT into mathematics trig-
gered a lot of new mathematical developments. Thus advancing mathematics is
another important source of motivation to study Mathematical QFT.
1
1 Wightman quantum field theory
The main references for this section are [1, Section VIII], [2, Section IX.8,X.7].

1.1 Relativistic Quantum Mechanics

p H (a space with a scalar
We consider a quantum theory given by a Hilbert space
product h | i, which is complete in the norm k · k = h · |· i ) and:

(a) Observables {Oi }i∈I . Hermitian / self-adjoint operators.

(b) Symmetry transformations {Uj }j∈J . Unitary/anti-unitary operators.

1.1.1 Observables
1. Consider an operator O : D(O) → H i.e. a linear map from a dense domain
D(O) ⊂ H to H. D(O) = H only possible for bounded operators O (i.e.
with bounded spectrum). In other words, O ∈ L(D(O), H), which is the
space of linear maps between the two spaces.

2. D(O† ) := { Ψ ∈ H | |hΨ|OΨ0 i| ≤ cΨ kΨ0 k for all Ψ0 ∈ D(O) }. Thereby, O† Ψ

is well defined for any Ψ ∈ D(O† ) via the Riesz theorem.

3. We say that O is Hermitian, if D(O) ⊂ D(O† ) and O† Ψ = OΨ for all

Ψ ∈ D(O).

4. We say that O is self-adjoint, if it is Hermitian and D(O) = D(O† ). Ad-

vantage: we can define eitO and then also a large
R classitOof other functions via
the Fourier transform. E.g. f (O) = (2π) −1/2 ˇ
dt e f (t) for f ∈ C0 (R)∞

(smooth, compactly supported, complex-valued).

5. We say that operators O1 , O2 weakly commute on some dense domain D ⊂

D(O1 ) ∩ D(O2 ) ∩ D(O1† ) ∩ D(O2† ) if for all Ψ1 , Ψ2 ∈ D

0 = hΨ1 |[O1 , O2 ]Ψ2 i = hΨ1 |O1 O2 Ψ2 i − hΨ1 |O2 O1 Ψ2 i

= hO1† Ψ1 |O2 Ψ2 i − hO2† Ψ1 |O1 Ψ2 i. (1)

6. We say that two self-adjoint operators O1 , O2 strongly commute, if

[eit1 O1 , eit2 O2 ] = 0 for all t1 , t2 ∈ R. (2)

No domain problems here, since eitO is always bounded, hence D(eitO ) = H.

7. Let O1 , . . . , On be a family of self-adjoint operators which mutually strongly

commute. For any f ∈ C0∞ (Rn ) we define
Z
f (O1 , . . . , On ) = (2π)−n/2
dt1 . . . dtn eit1 O1 . . . eitn On fˇ(t1 , . . . , tn ). (3)

2
 Definition 1.1 [4] The joint spectrum Sp(O1 , . . . , On ) of such a family of
operators is defined as follows: p ∈ Sp(O1 , . . . , On ) if for any open neigh-
bourhood Vp of this point there is a function f ∈ C0∞ (Rn ) s.t. suppf ⊂ Vp
and f (O1 , . . . , On ) 6= 0.

It is easy to see that for one operator O with purely point spectrum (e.g. the
Hamiltonian of the harmonic oscillator) Sp(O) is the set of all the eigenvalues.
But the above definition captures also the continuous spectrum without using
‘generalized eigenvectors’.

1.1.2 Symmetry transformations

We treat today only symmetries implemented by unitaries.
1. A linear bijection U : H → H is called a unitary if hU Ψ1 |U Ψ2 i = hΨ1 |Ψ2 i for
all Ψ1 , Ψ2 ∈ H. We denote by U(H) the group of all unitaries on H. Unitaries
are suitable to describe symmetries as they preserve transition amplitudes
of physical processes.

2. The Minkowski spacetime is invariant under Poincaré transformations x 7→

Λx + a, where (Λ, a) ∈ P+↑ (the proper ortochronous Poincaré group). We
consider a unitary representation of this group on H, i.e. a map P+↑ 3
(Λ, a) 7→ U (Λ, a) ∈ U(H) with the property

U (Λ1 , a1 )U (Λ2 , a2 ) = U ((Λ1 , a1 )(Λ2 , a2 )), (4)

i.e. a group homomorphism.

3. We say that such a representation is continuous, if (Λ, a) 7→ hΨ1 |U (Λ, a)Ψ2 i ∈

C is a continuous function for any Ψ1 , Ψ2 ∈ H.

1.1.3 Energy-momentum operators and the spectrum condition

The following fact is an immediate consequence of the Stone theorem and conti-
nuity is crucial here:
Theorem 1.2 Given a continuous unitary representation of translations R4 3
a 7→ U (a) := U (I, a) ∈ U(H), there exist four strongly commuting self-adjoint
operators Pµ , µ = 0, 1, 2, 3, s.t.
µ
U (a) = eiPµ a . (5)

We call P = {P0 , P1 , P2 , P3 } the energy-momentum operators.

In physical theories Pµ are unbounded operators (since values of the energy-
momentum can be arbitrarily large), defined on some domains D(Pµ ) ⊂ H. How-
ever, to guarantee stability of physical systems, the energy should be bounded
from below in any inertial system. The mathematical formulation is the spectrum
condition:
3
Definition 1.3 We say that a Poincaré covariant quantum theory satisfies the
spectrum condition if

Sp P := Sp(P0 , P1 , P2 , P3 ) ⊂ V + , (6)

where V + := { (p0 , p~) ∈ R4 | p0 ≥ |~p| } is the closed future lightcone.

1.1.4 Vacuum state

1. A unit vector Ω ∈ H is called the vacuum state if U (Λ, a)Ω = Ω for all
(Λ, a) ∈ R4 . This implies Pµ Ω = 0 for µ = 0, 1, 2, 3.

2. By the spectrum condition, Ω is the ground state of the theory.

3. We say that the vacuum is unique, if Ω is the only such vector in H up to

multiplication by a phase.

1.1.5 Relativistic Quantum Mechanics: Summary

Definition 1.4 A relativistic quantum mechanics is given by

1. Hilbert space H.

2. A continuous unitary representation P+↑ 3 (Λ, a) 7→ U (Λ, a) ∈ U(H) satisfy-

ing the spectrum condition.

3. Observables {Oi }i∈I , including Pµ .

Furthermore, H may contain a vacuum vector Ω (unique or not).

So far in our collection of observables {Oi }i∈I we have identified only global quanti-
ties like Pµ . (For example, to measure P0 , we would have to add up the energies of
all the particles in the universe of our theory). But actual measurements are per-
formed locally, i.e. in bounded regions of spacetime and we would like to include
the corresponding observables. We have to do it in a way which is consistent with
Poincaré symmetry, spectrum condition and locality (Einstein causality). This is
the role of quantum fields.

1.2 Quantum fields as operator-valued distributions

1.2.1 Tempered distributions
We recall some definitions:

1. The Schwartz-class functions:

S = { f ∈ C ∞ (R4 ) | sup |xα ∂ β f (x)| < ∞, α, β ∈ N40 }, (7)

x∈R4

∂ |β|
where xα := xα0 0 . . . xα3 3 and ∂ β = (∂x0 )β0 ...(∂x3 )β3
, |β| = β0 + · · · + β3 .
4
 2. The semi-norms kf kα,β := supx∈R4 |xα ∂ β f (x)| give a notion of convergence
in S: fn → f in S if kfn − f kα,β → 0 for all α, β.

3. We say that a linear functional ϕ : S → C is continuous, if for any finite set

F of multiindices there is a constant cF s.t.
X
|ϕ(f )| ≤ cF kf kα,β . (8)
α,β∈F

(Note that if fn → f in S then ϕ(fn ) → ϕ(f )). Such continuous func-

tionals are called tempered distributions and form the space S 0 which is the
topological dual of S.
Any measurable, polynomially growing function x 7→ ϕ(x) defines a tempered
distribution via
Z
ϕ(f ) = d4 x ϕ(x)f (x). (9)

The
R notation (9) is often used also if there is no underlying function, e.g. δ(f ) =:
δ(x)f (x)d4 x = f (0).

Definition 1.5 We consider:

1. A map S 3 f 7→ φ(f ) ∈ L(D(φ(f )), H).

2. A dense domain D ⊂ H s.t. for all f ∈ S

• D ⊂ D(φ(f )) ∩ D(φ(f )† ),
• φ(f ) : D → D,
• φ(f )† : D → D.

We say that (φ, D) is an operator valued distribution if for all Ψ1 , Ψ2 ∈ D the map

S 3 f 7→ hΨ1 |φ(f )Ψ2 i ∈ C (10)

is a tempered distribution. We say that (φ, D) is Hermitian, if φ(f ) is a Hermitian

operator for any real valued f ∈ S.

Note that a posteriori S 3 f 7→ φ(f ) ∈ L(D, D).

1.2.2 Wightman QFT

Definition 1.6 A theory of one scalar Hermitian Wightman field is given by:
1. A relativistic quantum theory (H, U ) with a unique vacuum state Ω ∈ H.

2. A Hermitian operator-valued distribution (φ, D) s.t. Ω ∈ D and U (Λ, a)D ⊂

D for all (Λ, a) ∈ P+↑ satisfying:
5
 (a) (Locality) [φ(f1 ), φ(f2 )] = 0 if supp f1 and supp f2 spacelike separated.
(In the sense of weak commutativity on D).

(b) (Covariance) U (Λ, a)φ(f )U (Λ, a)† = φ(f(Λ,a) ), for all (Λ, a) ∈ P+↑ and
f ∈ S. Here f(Λ,a) (x) = f (Λ−1 (x − a)).

(c) (Cyclicity of the vacuum) D = Span{ φ(f1 ) . . . φ(fm )Ω | f1 , . . . fm ∈ S, m ∈

N0 } is a dense subspace of H.

The distribution (φ, D) is called the Wightman quantum field.

Remarks:

1. Operator valued functions satisfying the Wightman axioms do not exist (we
really need distributions). The physical reason is the uncertainty relation:
Measuring φ strictly at a point x causes very large fluctuations of energy and
momentum, which prevent φ(x) from being a well defined operator. Such
observations were made already in [7], before the theory of distributions was
developed.

2. It is possible to choose D = D.

Example: Let H be the symmetric Fock space, then the energy-momentum op-
erators
d3 p d3 p
Z Z
0
P = p 0 †
a (p)a(p), ~ =
P p~ a† (p)a(p), (11)
(2π)3 2p0 (2π)3 2p0
p
where p0 = p2 + m2 , satisfy the spectrum condition and generate a unitary
µ
representation of translations U (a) = eiPµ a . Clearly, Ω = |0i is the unique vacuum
state of this relativistic QM. The Hermitian operator-valued distribution, given in
the function notation by

d3 p
Z
φ0 (x) = (eipx a† (p) + e−ipx a(p)). (12)
(2π)3 2p0

is a scalar Hermitian Wightman field.

6
2 Path integrals
The main references for this section are [5, Chapter 6] [6, Chapter 1].

2.1 Wightman and Schwinger functions

Consider a theory (H, U, Ω, φ, D) of one scalar Wightman field.

• Wightman functions are defined as

Wn (x1 , . . . , xn ) = hΩ|φ(x1 ) . . . φ(xn )Ωi. (13)

They are tempered distributions.

• Green functions are defined as

Gn (x1 , . . . , xn ) = hΩ|T (φ(x1 ) . . . φ(xn ))Ωi (14)

Recall that T φ(x1 )φ(x2 ) = θ(x01 −x02 )φ(x1 )φ(x2 )+θ(x02 −x01 )φ(x2 )φ(x1 ). This
multiplication of distributions by a discontinuous function may be ill-defined
in the Wightman setting. Approximation of θ by smooth functions may be
necessary. Then we obtain tempered distributions.

• Euclidean Green functions (Schwinger functions) are defined as

GE,n (x1 , . . . , xn ) = Wn ((ix01 , ~x1 ), . . . , (ix0n , ~xn )). (15)

The analytic continuation is justified in the Wightman setting. We obtain

real-analytic functions on R4n
6= = {(x1 , . . . , xn ) | xi 6= xj ∀i 6= j }, symmetric
under the exchange of variables.

The Schwinger functions are central objects of mathematical QFT based on

path-integrals. The idea is to express GE,n as moment functions of a measure µ
on the space SR0 of real-valued tempered distributions
Z
GE,n (x1 , . . . , xn ) = ϕ(x1 ) . . . ϕ(xn )dµ(ϕ). (16)
SR0

Today’s lecture:

• Measure theory on topological spaces.

• Conditions on dµ which guarantee that formula (16) really gives Schwinger

functions of some Wightman QFT. (Osterwalder-Schrader axioms).

• Remarks on construction of interacting functional measures dµ

7
2.2 Elements of measure theory
1. Def. We say that X is a topological space, if it comes with a family of subsets
T = {Oi }i∈I of X satisfying the following axioms:

• ∅, X ∈ T ,
S
• j∈J Oj ∈ T ,
• N
T
j=1 Oj ∈ T .

Oi are called the open sets.

2. Example: SR0 is a topological space. In fact, given ϕ0 ∈ SR0 , a finite family

J1 , . . . JN ∈ SR and ε1 , . . . , εN > 0 we can define a neighbourhood of ϕ0 as
follows:

B(ϕ0 ; J1 , . . . , JN ; ε1 , . . . εN )
:= { ϕ ∈ SR0 | |ϕ(J1 ) − ϕ0 (J1 )| < ε1 , . . . , |ϕ(JN ) − ϕ0 (JN )| < εN }. (17)

All open sets in SR0 can be obtained as unions of such neighbourhoods.

3. Def. Let X be a topological space. A family M of subsets of X is a σ-algebra

in X if it has the following properties:

• X ∈ M,
• A ∈ M ⇒ X\A ∈ M,
S∞
• An ∈ M, n ∈ N, ⇒ A := n=1 An ∈ M.

If M is a σ-algebra in X then X is called a measurable space and elements

of M are called measurable sets.

4. Def. The Borel σ-algebra is the smallest σ-algebra containing all open sets
of X. Its elements are called Borel sets.

5. Def. Let X be a measurable space and Y a topological space. Then a map

f : X → Y is called measurable if for any open V ⊂ Y the inverse image
f −1 (V ) is a measurable set.

6. Def. A measure is a function µ : M → [0, ∞] s.t. for any countable family

of disjoint sets Ai ∈ M we have
∞
[ ∞
X
µ( Ai ) = µ(Ai ). (18)
i=1 i=1

Also, we assume that µ(A) < ∞ for at least one A ∈ M.

• If µ(X) = 1, we say that µ is a probability measure.

• If µ is defined on the Borel σ-algebra, we call it a Borel measure.
8
7. We denote by Lp (X, dµ), 1 ≤ p < ∞ the space of measurable functions
f : X → C s.t.
Z 1/p
p
kf kp := |f (x)| dµ(x) < ∞. (19)
X

We denote by Lp (X, dµ) the space of equivalence classes of functions from

Lp (X, dµ) which are equal except at sets of measure zero. The following
statements are known as the Riesz-Fisher theorem:
• Lp (X, dµ) is a Banach space with the norm (19).
f¯1 (x)f2 (x)dµ(x).
R
• L2 (X, dµ) is even a Hilbert space w.r.t. hf1 |f2 i =
8. The following theorem allows us to construct measures on SR0 :

Theorem 2.1 (Bochner-Minlos) Let ZE : SR → C be a map satisfying

(a) (Continuity) ZE [Jn ] → ZE [J] if Jn → J in SR
(b) (Positive definiteness) For any J1 , . . . , JN ∈PSR0 , the matrix Ai,j :=
ZE [Ji − Jj ] is positive. This means z † Az := i,j z̄i Ai,j zj ≥ 0 for any
z ∈ CN .
(c) (Normalisation) ZE [0] = 1.
Then there exists a unique Borel probability measure µ on SR0 s.t.
Z
ZE [J] = eiϕ(J) dµ(ϕ) (20)
SR0

ZE [f ] is called the characteristic function of µ or the (Euclidean) generating

functional of the moments of µ. Indeed, formally we have:
Z
n δ δ
(−i) ... ZE [J]|J=0 = ϕ(x1 ) . . . ϕ(xn )dµ(ϕ), (21)
δJ(x1 ) δJ(xn ) SR0

so the generating functional carries information about all the moments of

the measure (cf. (16) above).
1 ∂ 2 ∂ 2
9. Example: Let C = −∆+m 2 , where ∆ = (∂x0 )2 + · · · + (∂x3 )2 is the Laplace

operator on R4 . For We consider the expectation value of C on f ∈ SR :

Z
¯ˆ 1 ˆ
hJ|CJi := d4 p J(p) J(p). (22)
p + m2
2

1
and set ZE,C [J] := e− 2 hJ|CJi . This map satisfies the assumptions of the
Bochner-Minlos theorem and gives a measure dµC on SR0 called the Gaussian
measure with covariance (propagator) C. In the physics notation:
Z Z
1 − 1 R d4 x ϕ(x)(−∆+m2 )ϕ(x)
F (ϕ)dµC (ϕ) = F (ϕ) e 2 D[ϕ]
NC
Z
1 − 1 R d4 x (∂µ ϕ(x)∂ µ ϕ(x)+m2 ϕ2 (x))
= F (ϕ) e 2 D[ϕ], (23)
NC
9
 for any F ∈ L1 (SR0 , dµC ). Since we chose imaginary time, we have a Gaussian
damping factor and not an oscillating factor above. This is the main reason
to work in the Euclidean setting.

2.3 Osterwalder-Schrader axioms

Now we formulate conditions, which guarantee that a given measure µ on SR0 gives
rise to a Wightman theory:

Definition 2.2 We say that a Borel probability measure µ on SR0 defines an Osterwalder-
Schrader QFT if this measure, resp. its generating functional ZE : SR → C,
satisfies:

1. (Analyticity) The function CN 3 (z1 , . . . , zN ) → ZE [ N

P
i=1 zj Jj ] ∈ C is entire
analytic for any J1 , . . . Jn ∈ SR .

Gives existence of Schwinger functions.

2. (Regularity) For some 1 ≤ p ≤ 2, a constant c and all J ∈ SR , we have

p
|ZE [J]| ≤ ec(kJk1 +kJkp ) . (24)

Gives temperedness of the Wightman field.

3. (Euclidean invariance) ZE [J] = ZE [J(R,a) ] for all J ∈ SR , where J(R,a) (x) =

f (R−1 (x − a)), R ∈ SO(4), a ∈ R4 .

Gives Poincaré covariance of the Wightman theory.

4. (Reflection positivity) Define:

• θ(x0 , ~x) = (−x0 , ~x) the Euclidian time reflection.

• Jθ (x) := J(θ−1 x) = J(θx) for J ∈ SR .
• R4+ = {(x0 , ~x) |x0 > 0}

Reflection positivity requires that for functions J1 , . . . , JN ∈ SR , supported in

R4+ , the matrix Mi,j := ZE [Ji − (Jj )θ ] is positive.

Gives positivity of the scalar product in the Hilbert space H (i.e.

hΨ|Ψi ≥ 0 for all Ψ ≥ 0). Also locality and spectrum condition.

5. (Ergodicity) Define:

• Js (x) = J(x0 − s, ~x) for J ∈ SR .

• (T (s)ϕ)(J) = ϕ(Js ) for ϕ ∈ SR0 .

10
 Ergodicity requires that for any function A ∈ L1 (SR0 , dµ) and ϕ1 ∈ SR0
1 t
Z Z
lim A(Ts ϕ1 )ds = A(ϕ)dµ(ϕ). (25)
t→∞ t 0 SR0

Gives the uniqueness of the vacuum.

Theorem 2.3 Let µ be a measure on SR0 satisfying the Osterwalder-Schrader
axioms. Then the moment functions
Z
GE,n (x1 , . . . , xn ) = ϕ(x1 ) . . . ϕ(xn )dµ(ϕ) (26)
SR0

exist and are Schwinger functions of a Wightman QFT.

Remark 2.4 The Gaussian measure dµC from the example above satisfies the
Osterwalder-Schrader axioms and gives the (scalar, Hermitian) free field.
Some ideas of the proof: The Hilbert space and the Hamiltonian of the Wightman
theory is constructed as follows:
• Def: E := L2 (SR0 , dµ).
• Def: AJ (ϕ) := eiϕ(J) for any J ∈ SR and (θAJ )(ϕ) := eiϕ(Jθ ) .
• Fact: E = Span{AJ |J ∈ SR }

• Def: E+ = Span{AJ |J ∈ S(R4+ )R } where S(R4+ )R are real Schwartz-class

functions supported in R4+ .
R
• Fact: hA1 |A2 i := (θA1 )(ϕ)A2 (ϕ)dµ(ϕ) is a bilinear form on E+ , which is
positive (i.e. hA|Ai ≥ 0) by reflection positivity. Due to the presence of θ it
differs from the the scalar product in E.
• Def: N = { A ∈ E+ | hA|Ai = 0} and set H = (E+ /N )cpl , where cpl denotes
completion. This H is the Hilbert space of the Wightman theory.
• T (t) : E+ → E+ for t ≥ 0. It gives rise to a semigroup e−tP0 : H → H with
a self-adjoint, positive generator P0 - the Hamiltonian. Thus eitP0 : H → H
gives unitary time-evolution.

2.4 Interacting measure

Interacting measures are usually constructed by perturbing the Gaussian measure
dµC . Reflection positivity severely restricts possible perturbations. Essentially,
one has to write:
1 R 4
dµI (ϕ) = e− LE,I (ϕ(x))d x dµC (ϕ), (27)
N
where N is the normalisation constant and LE,I : R → R some function (the
Euclidean interaction Lagrangian). For example LE,I (ϕ(x)) = 4!λ ϕ(x)4 . But this
leads to problems:
11
 • ϕ is a distribution so ϕ(x)4 in general does not makes sense.
This ultraviolet problem can sometimes be solved by renormalization.

• Integral over whole spacetime ill-defined. (But enforced by the translation

symmetry).

For φ4 theory in two-dimensional spacetime these problems were overcome and dµI
satisfying the Osterwalder-Schrader axioms was constructed. It was also shown
that the resulting theory is interacting, i.e. has non-trivial S-matrix. In the next
lecture we will discuss the S-matrix is in the Wightman setting.

12
3 Scattering theory
The main reference for this section is [3, Chapter 16].

3.1 Setting
We consider a Wightman theory (H, U, Ω, φ, D). Recall the key properties
1. Covariance: U (Λ, a)φ(x)U (Λ, a)−1 = φ(Λx + a),

2. Locality: [φ(x), φ(y)] = 0 for (x − y)2 < 0,

3. Cyclicity of Ω: Vectors of the form φ(f1 ) . . . φ(fm )Ω span a dense subspace

in H,
where smearing with test-functions from S in variables x, y is understood in prop-
µ
erties 1. and 2. Furthermore U (a) = eiPµ a and Sp P ⊂ V̄+ . Today we will impose
stronger assumptions on the spectrum:
A.1. The spectrum contains an isolated mass hyperboloid Hm i.e.

Hm ⊂ Sp P ⊂ {0} ∪ Hm ∪ Gm̃ , (28)

p p
where Hm = { p ∈ R4 | p0 = p~2 + m2 }, Gm̃ = { pp ∈ R4 | p0 ≥ p~2 + m̃2 }
for m̃ > m. (In other words, the mass-operator Pµ P µ has an isolated
eigenvalue m. Embedded eigenvalues can also be treated [13], but then
scattering theory is more difficult).

A.2. Define the single-particle subspace H(Hm ) as the spectral subspace of Hm .

That is, H(Hm ) = χ(P )H, where χ(P ) is the characteristic function of Hm
evaluated at P = (P0 , P1 , P2 , P3 ). We assume that U restricted to H(Hm ) is
an irreducible representation of P+↑ . (One type of particles).

Theorem 3.1 (Källen-Lehmann representation). For a Wightman field φ with

hΩ, φ(x)Ωi = 0 we have
Z
hΩ|φ(x)φ(y)Ωi = dρ(M 2 )∆+ (x − y; M 2 ), (29)
d3 p
Z
2 (M ) (M )
∆+ (x − y; M ) := h0|φ0 (x)φ0 (y)|0i = eip(y−x) , (30)
(2π)3 2p0
(M )
p
where dρ(M 2 ) is a measure on R+ , p0 = p~2 + M 2 , φ0 is the free scalar field
of mass M and |0i is the vacuum vector in the Fock space of this free field theory,
whereas Ω is the vacuum of the (possibly interacting) Wightman theory. Further-
more, given the structure of the spectrum (28), we have

dρ(M 2 ) = Zδ(M 2 − m2 )d(M 2 ) + dρ̃(M 2 ) (31)

where Z ≥ 0 and dρ̃ is supported in [m̃2 , ∞)

13
We assume in the following that:

A.3. hΩ, φ(x)Ωi = 0. This is not a restriction, since a shift by a constant φ(x) 7→
φ(x) + c gives a new Wightman field.

A.4. Z 6= 0 to ensure that hΨ1 |φ(x)Ωi = 6 0 for some single-particle vector Ψ1

(i.e. a vector living on Hm ). This means that the particle is ‘elementary’
(as opposed to composite) and we do not need polynomials in the field to
create it from the vacuum. This assumption can be avoided at a cost of
complications.

3.2 Problem and strategy

Take two single-particle states Ψ1 , Ψ2 ∈ H(Hm ). We would like to construct
vectors Ψout , Ψin describing outgoing/ incoming configuration of these two single-
particle states Ψ1 , Ψ2 . Mathematically this problem consists in finding two ‘mul-
tiplications’
out
Ψout = Ψ1 × Ψ2 , (32)
in
Ψin = Ψ1 × Ψ2 , (33)

which have all the properties of the (symmetrised) tensor product but take values
in H (and not in H ⊗ H). After all, we know from quantum mechanics, that sym-
metrised tensor products describe configurations of two undistinguishable bosons.
The strategy is suggested by the standard Fock space theory: With the help of
the field φ we will construct certain ‘time-dependent creation operators’ t 7→ A†1,t ,
t 7→ A†2,t s.t.

Ψ1 = lim A†1,t Ω, Ψ2 = lim A†2,t Ω. (34)

t→±∞ t→±∞

Then we can try to construct

Ψout = lim A†1,t A†2,t Ω, Ψin = lim A†1,t A†2,t Ω. (35)

t→∞ t→−∞

Of course analogous consideration applies to n-particle scattering states.

Plan of the remaining part of the lecture:

• Construction of A†t .

• Existence of limits in definitions of Ψout , Ψin .

• Wave-operators, S-matrix and the LSZ reduction formula.

14
3.3 Definition of A†t
The operators A†t are defined in (41) below. In order to motivate this definition,
we state several facts about the free field. It should be kept in mind that we are
interested in the interacting field, and the following discussion of the free field is
merely a motivating digression.
Recall the definition of the free scalar field:
d3 p
Z
φ0 (x) = (eipx a† (p) + e−ipx a(p)). (36)
(2π)3 2p0

(Here and in the following we reserve

p the letter p for momenta restricted to the
0
mass-shell i.e. p = (p , p~) = ( p~ + m2 , p~). For other momenta I will use q).
2

There are two ways to extract a† out of φ0 :

1. Use the formula from the lecture:

Z ↔
a (p) = i d3 x φ0 (x) ∂ 0 e−ipx
†
(37)

Since a† (p) is not a well-defined operator (only an operator valued distribu-

tion) we will smear both sides of this equality with a test-function. For this
purpose we define for any f ∈ C0∞ (R4 )

d3 p d3 p
Z Z
† †
a (f ) := 3 0
a (p)f (p), fm (x) = 3 0
e−ipx f (p), (38)
(2π) 2p (2π) 2p

where the latter is a positive-energy solution of the KG equation, that is

( + m2 )fm (x) = 0. We get
Z ↔
†
a (f ) = i d3 x φ0 (x) ∂ 0 fm (x). (39)

2. Pick a function h ∈ S s.t. supp b

h is compact and supp bh ∩ Sp P ⊂ Hm . Then
Z
2 † b 1
φ0 (h) = (2π) a (h), where bh(q) = eiqx h(x)d4 x. (40)
(2π)2

Now we come back to our (possibly) interacting Wightman field φ and perform
both operations discussed above to obtain the ‘time dependent creation operator’
Z ↔
†
At := i d3 x φ(h)(t, ~x) ∂ 0 fm (t, ~x), (41)

where φ(h)(t, ~x) := U (t, ~x)φ(h)U (t, ~x)† = φ(h(t,~x) ).

15
3.4 Construction of scattering states
Theorem 3.2 (Haag-Ruelle) For f1 , . . . , fn with disjoint supports, the following
limits exist
† †
Ψout
n = lim A1,t . . . An,t Ω, (42)
t→∞
† †
Ψin
n = lim A1,t . . . An,t Ω (43)
t→−∞

and define outgoing/incoming scattering states.

Proof. For n = 1 the expression

Z
†

↔
A1,t Ω = i d3 x φ(h)(t, ~x)Ω ∂ 0 fm (t, ~x) (44)

is independent of t and thus limt±∞ A†1,t (f1 )Ω (trivially) exist. Moreover, it is a

single-particle state. Justification:

• x 7→ φ(h)(x)Ω is a solution of the KG equation. This can be shown using the

Källen-Lehmann representation and the support property of b h to eliminate
the contribution from dρ̃. Assumption A.1. enters here. (Howework).
R ↔
• For any two solutions g1 , g2 of the KG equation d3 x g1 (t, ~x) ∂ 0 g2 (t, ~x) is
independent of t.

• We have i[Pµ , φ(h)(x)] = ( ∂x∂ µ )φ(h)(x). Since Pµ Ω = 0, we can write

P 2 φ(h)(x)Ω = Pµ P µ φ(h)(x)Ω = −i[Pµ , i[P µ , φ(h)(x)]]Ω

= −x φ(h)(x)Ω = m2 φ(h)(x)Ω, (45)

where in the last step we used the first item above. Hence φ(h)(x)Ω are
single-particle states of mass m.

For n = 2 we set Ψt := A†1,t A†2,t Ω and try to verify the Cauchy criterion:
Z t2 Z t2
kΨt2 − Ψt1 k = k ∂τ Ψτ dτ k ≤ k∂τ Ψτ kdτ. (46)
t1 t1

If we manage to show that k∂τ Ψτ k ≤ c/τ 1+η for some η > 0 then the Cauchy
criterion will be satisfied as we will have
1 1
kΨt2 − Ψt1 k ≤ c η − η . (47)
t1 t2

(Note that we use the completeness of H here i.e. the property that any Cauchy
sequence converges).

16
 Thus we study ∂τ Ψτ . The Leibniz rule gives

∂τ Ψτ = (∂τ A†1,τ )A†2,τ Ω + A†1,τ (∂τ A†2,τ )Ω

= [(∂τ A†1,τ ), A†2,τ ]Ω + A†2,τ (∂τ A†1,τ )Ω + A†1,τ (∂τ A†2,τ )Ω. (48)

Since (∂τ A†i,τ )Ω = 0 by the first part of the proof, only the term with the com-
mutator above is non-zero. To analyze it, we need some information about KG
wave-packets:
• Def. For the KG wave-packet fi,m we define the velocity support as
 
p~
Vi = | p ∈ suppfi (49)
p0
and let Viδ be slightly larger sets.
• Fact. For any N ∈ N we can find a cN s.t.
cN ~x
|fi,m (τ, ~x)| ≤ N
for ∈ / Viδ . (50)
τ τ

Due to (50), the contributions to k[(∂τ A†1,τ ), A†2,τ ]Ωk coming from the part of the
integration region in (41) where ~xτ ∈
/ Viδ , are rapidly vanishing with τ . So we only
have to worry about the dominant parts:
Z ↔
†(D)
Ai,t := i d3 x φ(h)(t, ~x) ∂ 0 fi,m (t, ~x). (51)
~
x
t
∈Viδ

Since V1δ , V2δ are disjoint, the Wightman axiom of locality gives for sufficiently
large τ .
†(D) †(D) cN
k[(∂τ A1,τ ), A2,τ ]Ωk ≤ . (52)
τN
This concludes the proof. 

3.5 Wave operators, scattering matrix, LSZ reduction

h(p)f (p) = (2π)−2 Z −1/2 f (p). This can be done,
In the following we choose h s.t. b
since f has compact support. After this fine-tuning, exploiting assumptions A.2,
A.3, A.4 one obtains the following simple formula for scalar products of scattering
states:
0 out
Theorem 3.3 (Haag-Ruelle) Let Ψout n , (Ψn0 ) be as in the previous theorem.
Then their scalar products can be computed as if these were vectors on the Fock
space:
0 out
hΨout
n |(Ψn0 ) i = h0|a(fn ) . . . a(f1 )a† (f10 ) . . . a† (fn0 0 )|0i (53)

and analogously for incoming states.

17
Let F be the symmetric Fock space. (This is not the Hilbert space of our Wightman
theory, but merely an auxiliary object needed to define the wave-operators). We
define the outgoing wave-operator W out : F → H as

W out (a† (f1 ) . . . a† (fn )|0i) = lim A†1,t . . . A†n,t Ω. (54)

t→∞

By Theorem 3.3 it is an isometry i.e. (W out )† W out = I. If it is also a unitary

i.e. Ran W out = H then we say that the theory is asymptotically complete that
is every vector in H can be interpreted as a collection of particles from H(Hm ).
This property does not follow from Wightman axioms (there are counterexamples)
and it is actually not always expected on physical grounds. For a more thorough
discussion of asymptotic completeness we refer to [12].
The incoming wave-operator W in : F → H is defined by taking the limit
t → −∞ in (54). The scattering matrix Ŝ : F → F is given by1

Ŝ = (W out )† W in . (55)

If Ŝ 6= I we say that a theory is interacting. If Ran W out = Ran W in , then Ŝ is a

unitary (even without asymptotic completeness).

Corollary 3.4 (LSZ reduction) [8] For f1 , . . . , f` , g1 , . . . gn ∈ S with mutually dis-

joint supports, we have

d3 k1 d3 p n
Z
† †
h0|a(f1 ) . . . a(f` )Ŝa (g1 ) . . . a (gn )|0i = . . . f1 (k1 ) . . . gn (pn )×
(2π)3 2k10 (2π)3 2p0n
` n
(−i)n+` Y 2 Y
× √ (ki − m2 ) (p2j − m2 )×
( Z)n+` i=1 j=1
Z P` Pn
× d4 x1 . . . d4 x` d4 y1 . . . d4 yn ei i=1 ki xi −i j=1 pj yj ×

×hΩ|T (φ(x` ) . . . φ(x1 )φ(y1 ) . . . φ(yn ))Ωi,

where T is the time-ordered product (which needs to be regularized in the Wightman

setting).

By analytic continuation one can relate the Green functions to Schwinger functions.
The latter can be studied using path integrals as explained in the previous lecture.
This led to a proof that for φ4 in 2-dimensional spacetime Ŝ 6= I [9]. It is a
big open problem if there is a Wightman theory in 4-dimensional spacetime with
Ŝ 6= I.

1
The notation Ŝ is used to avoid confusion with the Schwartz class S.

18
4 Renormalization
Main references for this section are [10, 11].

4.1 Introductory remarks

Consider a Wightman theory (H, U, Ω, φB , D) as in the previous lecture and sup-
pose we want to describe a collision of several particles: The LSZ formula gives:
` n
† † (−i)n+` Y 2 Y
h0|a(k1 ) . . . a(k` )Ŝa (p1 ) . . . a (pn )|0i = √ (ki − m ) (p2j − m2 )×
2
( Z) n+`
i=1 j=1
Z P` Pn
× d4 x1 . . . d4 x` d4 y1 . . . d4 yn ei i=1 ki xi −i j=1 pj yj ×

×hΩ|T (φB (x` ) . . . φB (x1 )φB (y1 ) . . . φB (yn ))Ωi,

The ‘renormalized’ field φ := Z −1/2 φB is again a Wightman field. To compute the

S-matrix of (H, U, Ω, φ, D) we drop Z and the index B on the r.h.s. of the formula
above.
Possibly after regularizing the time-ordered product, we can express the Green
functions of φ above by the Wightman functions and then analytically continue to
Schwinger functions GE,n . If the theory satisfies also Osterwalder-Schrader axioms,
we have
Z
δ δ
GE,n (x1 , . . . , xn ) = ... ZE [J]|J=0 , ZE [J] = eϕ(J) dµ(ϕ), (56)
δJ(x1 ) δJ(xn ) 0
SR

for some Borel measure dµ and J ∈ SR . (We changed eiϕ(J) to eϕ(J) , where J ∈ SR ,
making use of the Osterwalder-Schrader axiom of analyticity).
Today we will attempt to construct this measure in λφ4 theory. This endeavor
will not be completely successful. In the end we will obtain GE,n as power series
in the coupling constant λ whose convergence we will not control. Thereby we will
abandon the Wightman/Osterwalder-Schrader setting and delve into perturbative
QFT. Strictly speaking, we will also abandon the realm of quantum theories, as
there will be no underlying Hilbert space and thus no control that transition prob-
abilities of physical processes take values between zero and one. On the positive
side, the finiteness of individual terms in this expansion will be an interesting
problem in the theory of differential equations.

4.2 From action to generating functional

Consider the Euclidean action written in terms of the bare (unphysical) quantities:
Z
1 1 λB 4

SE [ϕB ] = d4 x ∂µ ϕB ∂ µ ϕB + m2B ϕ2B + ϕ . (57)
2 2 4! B

19
 √
Introduce the physical parameters via ϕB = Zϕ, mB = Zm m, λB = Zλ λ. This
gives
Z
1
SE [ϕ] = d4 x ∂µ ϕ∂ µ ϕ + m2 ϕ2


2
Zλ Z 2 4
Z
1 1 2
+ d4 x (Z − 1)∂µ ϕ∂ µ ϕ + (Zm Z − 1)m2 ϕ2 +

λϕ
2 2 4!
= SE,0 [ϕ] + SE,I [ϕ]. (58)

Let us start with the free action SE,0 . The corresponding generating functional is
Z
1
R 4
d p J(p)C(p)J(p)
ZE,0 [J] = e = eϕ(J) dµC (ϕ), (59)
b b
2

SR0
1 −SE,0 [ϕ]
dµC (ϕ) = ” e D[ϕ]”, (60)
NC
1
where C(p) = p2 +m2 and dµC is the corresponding Gaussian measure given formally

by (60). The candidate generating functional for the interacting theory is

Z
1
cand
ZE [J] = eϕ(J) e−SE,I [ϕ] dµC (ϕ), (61)
N SR0

where N is chosen so that ZEcand [0] = 1. Problems:

• dµC is supported on distributions so ϕ(x)4 appearing in SE,I [ϕ] is ill-defined

(UV problem).

• Integral over spacetime volume in SE,I [ϕ] ill-defined (IR problem).

4.3 Regularized generating functional

To make sense out of (61) we need regularization: We set
2 +m2
1 −p p2 +m2
CΛΛ0 (p) := 2 (e Λ20 − e− Λ2 ). (62)
p + m2
Here Λ0 is the actual UV cut-off. 0 ≤ Λ ≤ Λ0 is an auxiliary cut-off which will be
needed for technical reasons. Thus CΛΛ0 is essentially supported in Λ ≤ p2 ≤ Λ0 .

Lemma 4.1 dµC Λ0 is supported on smooth functions.

Λ

Thus the following regularized generating functional is meaningful

Z
1 Λ0
Λ,Λ0
ZE [J] = eϕ(J) e−SE,I,(V ) [ϕ] dµC Λ0 (ϕ), (63)
N Λ

where we also introduced a finite volume V . Of course, when we try to take the
limit Λ0 → ∞, we will get back divergent expressions. The idea of renormalization
20
 Λ0
is to absorb these divergencies into the parameters in SE,I,(V ) (therefore we added
a superscript Λ0 ). For simplicity, we rename the coefficients from (58) as follows:
Z
Λ0
d4 x aΛ0 ∂µ ϕ∂ µ ϕ + bΛ0 ϕ2 + cΛ0 λϕ4 .


SE,I,(V ) [ϕ] = (64)
V

We list the following facts:

• ZEΛ,Λ0 generates (regularized) Schwinger functions GΛ,Λ0
E,n .

• − log (N ZEΛ,Λ0 ) generates (regularized) connected Schwinger functions GΛ,Λ 0

E,c,n .

• ΣΛ,Λ0 [J] := − log (N ZEΛ,Λ0 [(CΛΛ0 )−1 J]) + 21 hJ|(CΛΛ0 )−1 Ji generates (regular-
ized) connected amputated Schwinger functions with subtracted zero-order
contribution GΛ,Λ 0
E,r≥1,a,c,n .

Given energy-momentum conservation, it is convenient to set

SnΛ,Λ0 (p1 , . . . , pn−1 )(2π)4 δ(p1 + · · · + pn )
Z
Λ,Λ0
:= ei(p1 x1 +···+pn xn ) GE,r≥1,a,c,n (x1 , . . . , xn )d4 x1 . . . d4 xn .(65)

4.4 The problem of perturbative renormalizability

The renormalization of φ4 in 4 dimensions is only understood perturbatively. This
means we treat SnΛ,Λ0 as a formal power series
X
SnΛ,Λ0 = λr Sr,n
Λ,Λ0
, (66)
r≥1

i.e. a series whose convergence we do not control. (Order by order the limit
V → R4 can be taken, so we will not discuss it anymore). Similarly, we treat the
coefficients from the interaction Lagrangian as formal power series:
X X X
aΛ 0 = λr aΛr 0 , bΛ0 = λr bΛr 0 , cΛ0 = λr cΛr 0 . (67)
r≥1 r≥1 r≥1

Furthermore, we impose the following BPHZ renormalization conditions (RC)

0,Λ0 0,Λ0 0,Λ0
Sr,4 (0) = δr,1 , Sr,2 (0) = 0, ∂p2 Sr,2 (0) = 0, (68)
which fix the physical values of the parameters2 .
Theorem 4.2 (Perturbative renormalizability) There are such {aΛr 0 , bΛr 0 , cΛr 0 }Λ0 ≥0
that the limits
0,Λ0
Sr,n (p) := lim Sr,n (p), p = (p1 , . . . , pn−1 ) (69)
Λ0 →∞

exist and are finite, and the renormalisation conditions (68) hold.
2
The physical meaning of our RC is not so direct, since we are in the Euclidean setting.
Before substituting our n-point functions to the LSZ formula for the S-matrix, they have to be
analytically continued to the real time and transferred to the on-shell renormalization scheme.
21
Example: Consider the leading contribution to the two-point and four-point func-
tion:
0,Λ0
S1,2 (p) = ......O...... + ......x......
d4 q Λ 0
Z
Λ0
= 12c1 C (q) + 2(aΛ1 0 p2 + bΛ1 0 ),
(2π)4 0
(q 2 +m2 )
d4 q
Z
Λ0 1 −
= 12c1 4 2 2
e Λ20 +2(aΛ1 0 p2 + bΛ1 0 ), (70)
(2π) q + m
| {z }
I Λ0
0,Λ0
S1,4 (p1 , p2 , p3 ) = 4!cΛ1 0 . (71)

For any finite Λ0 the integral above is convergent, but it diverges for Λ0 → ∞.
We have to choose the behavior of aΛ1 0 , bΛ1 0 , cΛ1 0 s.t. this divergence is compensated
and the RC are satisfied. We get from the three RC conditions, respectively,

4!c1Λ0 = 1, 12cΛ1 0 I Λ0 + 2bΛ1 0 = 0, 2aΛ1 0 = 0. (72)

Hence, cΛ1 0 = 1/4!, bΛ1 0 = −(1/4)I Λ0 , aΛ1 0 = 0 and thus bΛ1 0 absorbs the divergence
of the integral I Λ0 .

4.5 Flow equations

Properties of ΣΛ,Λ0 [J] := − log (N ZEΛ,Λ0 [(CΛΛ0 )−1 J]) + 12 hJ|(CΛΛ0 )−1 Ji:
Λ,Λ0 [J] Λ0
• e−Σ dµC Λ0 (ϕ)e−SE,I [J+ϕ] .
R
=
Λ

Λ0
• limΛ→Λ0 ΣΛ,Λ0 [J] = SE,I [J] since limΛ→Λ0 CΛΛ0 (p) = 0 and therefore, formally,
limΛ→Λ0 dµC Λ0 (ϕ) = δ(ϕ)D[ϕ].
Λ

Λ,Λ0 [J] Λ,Λ0 [J]

• ∂Λ (e−Σ δ
) = 21 h δJ |∂Λ CΛΛ0 δJ
δ
ie−Σ .
Λ,Λ0 Λ,Λ0
This gives the following equation for Sr,n (p) := Sr,n (p1 , . . . , pn−1 )

d4 p
Z
1 Λ,Λ0
Λ,Λ0
∂Λ Sr,n (p) = (∂Λ CΛΛ0 )(p)Sr,n+2 (p1 , . . . , pn−1 , p, −p)
2 (2π)4
1 X  n  Λ,Λ
− 0−1
[Sr0 ,n00 (p1 , . . . , pn0 −1 )(∂Λ CΛΛ0 )(q)SrΛ,Λ 0
00 ,n00 (−q, pn0 , . . . , pn−1 )]sym ,
2 0 00
n
r +r =r
n0 +n00 =n+2
(73)

where q = −p1 − · · · − pn0 −1 and sym denotes the symmetrisation in p1 , . . . , pn−1 .

Boundary conditions:
0,Λ0 0,Λ0 0,Λ0
(a) Λ = 0 : Sr,4 (0) = δr,1 , Sr,2 (0) = 0, ∂p2 Sr,2 (0) = 0 (RC).
22
 (b) Λ = Λ0 : ∂ w Sr,n
Λ0 ,Λ0
(p) = 0 for n + |w| ≥ 5. (By second bullet above).

With the help of the above equation one proves the following theorem:

Theorem 4.3 The following estimate holds

(
P1 (|p|) for 0 ≤ Λ ≤ 1,
|∂ w Sr,n
Λ,Λ0
(p)| ≤ 4−n−w |p|
,
Λ P2 (log Λ)P3 Λ for 1 ≤ Λ ≤ Λ0

where Pi are some polynomials independent of p, Λ, Λ0 , but depending on n, r, w.

Λ,Λ0
In particular this shows that Sr,n (p) stays bounded if Λ = 0 and Λ0 → ∞. With
more effort, one also shows convergence as Λ0 → ∞, required by the renormaliz-
ability property (69).

4.6 Outline of the proof of Theorem 4.3

Observation3 :
Λ,Λ0
Sr,n ≡ 0 for n > 2r + 2. (74)

Given this, the structure of the flow equation suggests the inductive scheme as in
the figure: Indeed, we can start the induction in the region where (74) holds and
therefore the estimate from Theorem 4.3 is satisfied. We suppose the estimate
holds for
3
Indeed, without assuming connectedness we clearly have vanishing of these functions for
n > 4r. Now (r − 2) verticies need to use at least two lines to keep the diagram connected. 2
verticies need to spend only one line. So altogether n > 4r − 2(r − 2) − 2.

23
 • (r, n1 ) for n1 ≥ n + 2.

• (r2 , n2 ) for r2 < r and any n2 .

Since only SrΛ,Λ 0

0 ,n0 as listed above appear on the r.h.s. of the flow equation (73),

we can apply the estimate to this r.h.s. Then (after quite some work) the flow
Λ,Λ0
equation gives the required estimate on the Sr,n , which appears on the l.h.s. of
the flow equation (73).

24
5 Symmetries I
Symmetries in physics are described by groups. We recall some definitions and
facts from the theory of groups and their representations following [14, Chapter
1], [15, Chapter 1], [16].

5.1 Groups
1. Def. A group is a set G with an operation · : G × G → G s.t.

• (g1 · g2 ) · g3 = g1 · (g2 · g3 ) for all g1 , g2 , g3 ∈ G,

• There exists e ∈ G s.t. g · e = e · g = g for all g ∈ G,
• For any g ∈ G there exists g −1 ∈ G s.t. g · g −1 = e.

2. Examples:

• Z2 = {1, −1} is the group of parity transformations.

• Let V be a vector space over the field K (R or C). Then GL(V ) denotes
the group of all invertible linear mappings V → V .
• For V = Kn we write GL(n, K) := GL(V ). This is the group of invert-
ible n × n matrices with entries in K.
• SO(3) = { R ∈ GL(3, R) | RT R = I, det R = 1} - the group of rota-
tions.
• SU (2) = {U ∈ GL(2, C) | U † U = I, det U = 1 } - special unitary group.

3. Let G, Ĝ be groups. Then H : G → Ĝ is a group homomorphism if

H(g1 g2 ) = H(g1 )H(g2 ) for any g1 , g2 ∈ G. (75)

If H is in addition a bijection then it is called an isomorphism.

(For Lie groups, which we discuss below, homomorphisms are required to be
smooth and isomorphisms should also have smooth inverse).

5.2 Lie groups

1. Def. G is a Lie group if it is a smooth real manifold and the group operation
and taking the inverse are smooth maps. The dimension of G is the dimension
of this manifold.

2. Def. A set M is an n-dimensional smooth manifold if the following hold:

• It is a Hausdorff topological space. (Distinct points have non-overlapping

neighbourhoods, unique limits).
• There
S is an open cover i.e. a family of open sets Uα ⊂ M, α ∈ I, s.t.
α∈I Uα = M .

25
 • There is an atlas A(M ) := { ηα : Uα → Oα | α ∈ I } given by some fam-
ily of open sets Oα ⊂ Rn and charts ηα , which are homeomorphisms.
(A homeomorphism is a continuous bijection whose inverse is also con-
tinuous).
• Let Oα,β := ηβ (Uα ∩ Uβ ) ⊂ Rn and let Oβ,α = ηα (Uα ∩ Uβ ) ⊂ Rn . Then
ηα ◦ ηβ−1 : Oα,β → Oβ,α is smooth.

3. Def. A map F : M → M̂ between two manifolds is smooth if all the maps

η̂β ◦ F ◦ ηα−1 : Oα → Ôβ are smooth whenever well-defined. It is called a
diffeomorphism if it is a bijection and the inverse is also smooth. We denote
by C ∞ (M ) the space of smooth maps M → R.

4. Def. A smooth map R × M 3 (t, x) 7→ γt (x) ∈ M , is called a flow (of a

vector field) if

γ0 = idM , γs ◦ γt = γs+t for t, s ∈ R. (76)

5. Def. A vector field X : C ∞ (M ) → C ∞ (M ) with the flow γ is given by

d
X(f ) = f ◦ γt |t=0 , f ∈ C ∞ (M ). (77)
dt

6. Fact. Given two vector fields X, Y , the commutator [X, Y ](f ) := X(Y (f )) −
Y (X(f )) is again a vector field.

7. Def. A tangent vector at point x ∈ M is a map Xx : C ∞ (M ) → R given by

Xx (f ) = dtd f ◦ γt (x)|t=0 . We denote by Tx M the space of all tangent vectors
at x (for different flows).
Example: Let M = Rn and γt = (γt1 , . . . , γtn ) be a flow. Then
n
X d i ∂f
Xx (f ) = γt (x)|t=0 i (x). (78)
i=1
dt ∂x

‘forgets’ f and thinks about ∂x∂ i as basis vectors then the expression
If one P
Xx = ni=1 dtd γti (x)|t=0 ∂x∂ i is clearly the tangent vector to t → γt (x) at x.

8. Def. (Transport of a vector field) Let F : M → M̂ be a diffeomorphism and

X a vector field on M given by a flow {γt }t∈R . Then F∗ X is a vector field
on M̂ given by F ◦ γt ◦ F −1 . That is
d
(F∗ X)(fˆ) = fˆ ◦ F ◦ γt ◦ F −1 |t=0 , fˆ ∈ C ∞ (M̂ ). (79)
dt

9. Now we can introduce the Lie algebra G0 of a Lie group G.

• Def. Let G be a Lie group and g ∈ G. Then, the left-multiplication

Lg : G → G, acting by Lg g̃ = gg̃ is a diffeomorphism.
26
 • Def. We say that a vector field X on G is left-invariant if ((Lg )∗ X) = X
for any g ∈ G.
• Fact. If X, Y are left-invariant, then also [X, Y ] is left-invariant.
• Def. The Lie algebra G0 of G is the vector space of left-invariant vector
fields on G with algebraic operation given by the commutator.

10. Def. For X ∈ G0 we set exp(X) := γ1 (e).

11. Fact. This exponential map is a diffeomorphism of a neighbourhood of zero

in G0 into a neighbourhood of e in G.

5.3 From multiplication law in G to algebraic operation

in G0
1. Fact. Left-invariance and (79) give Xg (f ) = Xe (f ◦ Lg ) for any g ∈ G.
Thus left-invariant vector fields are determined by their values at the neutral
element e. In this sense, G0 can be identified with Te G and has the same
dimension as G.

2. Let us choose a basis X 1 , . . . , X n in G0 . We have

n
X
A B
[X , X ]e = f CAB XeC , (80)
C=1

where f CAB are called the structure constants. In physics one usually defines
the infinitesimal generators4 tA := iX A and writes (80) as
n
X
A B
[t , t ] = if CAB tC , (81)
C=1

where evaluation at e is understood. We follow the mathematics convention

below, unless stated otherwise.

3. Let us determine f CAB from the multiplication law of the group G. We

work in some chart η : U → O ⊂ Rn whose domain U is a neighbourhood
of e, in which the group elements have the form g = η −1 (ε1 , . . . , εn ), e =
η −1 (0, . . . , 0). For ε1 , ε2 ∈ O, i.e. εi = (ε1i , . . . , εni ), i = 1, 2, we define the
multiplication function

m(ε1 , ε2 ) := η(η −1 (ε1 )η −1 (ε2 )) (82)

4
As we defined G0 as a real vector space, it may be unclear what the multiplication by ‘i’
means. We recall that is it always possible to ‘complexify’ a real vector space. Furthermore, in
the later part of these lectures we will represent Lie algebras on complex vector spaces. Then ‘i’
will be provided by the vector space.

27
which takes values in O i.e. m(ε1 , ε2 ) = (m1 (ε1 , ε2 ), . . . , mn (ε1 , ε2 )). Since
η −1 (0) = e, we have m(ε1 , 0) = ε1 and m(0, ε2 ) = ε2 . Consequently

∂mi ∂mi
(0, 0) = (0, 0) = δi,j . (83)
∂εj1 ∂εj2

Given a flow of a vector field γt in G we define the transported flow γ̃t (ε) =
η ◦ γt ◦ η −1 (ε) = (γ̃t1 (ε), . . . , γ̃tn (ε)).

Lemma 5.1 Let X 1 , . . . , X A . . . , X n be vector fields on G whose flows sat-

isfy

d i ∂mi
γ̃A,t (ε)|t=0 = A (ε, 0). (84)
dt ∂ε2

Then these fields are left-invariant. Furthermore, they are linearly indepen-
dent near e and thus span the Lie algebra.

Proof. We want to verify XgA (f ) = XeA (f ◦ Lg ) for f ∈ C ∞ (M ) and

g = η −1 (ε). Thus we set f˜ := f ◦ η −1 and compute

d d
XηA−1 (ε) (f ) = f (γA,t ◦ η −1 (ε))|t=0 = f˜(γ̃A,t (ε))|t=0
dt dt
n ˜ n
X d i ∂f X ∂mi ∂ f˜
= (γ̃A,t (ε))|t=0 i (ε) = A
(ε, 0) i
(ε). (85)
i=1
dt ∂ε i=1
∂ε 2 ∂ε

On the other hand

d
XηA−1 (0) (f ◦ Lη−1 (ε) ) = f (Lη−1 (ε) γA,t ◦ η −1 (0))|t=0
dt
d d
= f (η −1 (ε)η −1 (γ̃A,t (0)))|t=0 = f˜(m(ε, γ̃A,t (0)))|t=0
dt dt
X ∂ f˜ ∂mi d k
= i
(ε) k (ε, 0) γ̃A,t (0)|t=0 , (86)
i,k
∂ε ∂ε2 |dt {z }
δk,A by (83),(84).

which concludes the proof of left-invariance. It suffices to check linear inde-

pendence near e which follows from (85) and (83). 

Lemma 5.2 Under the assumptions of the previous lemma, we have

n
X
A B
[X , X ]e = f CAB XeC , (87)
C=1

∂ 2 mC ∂ 2 mC
with f CAB = ∂εA B (0, 0) − ∂εB A (0, 0). (Homework).
1 ∂ε2 1 ∂ε2

28
5.4 Abstract Lie algebras
1. Def. A Lie algebra is a vector space g over the field R together with a bilinear
form [ · , · ] : g × g → g which satisfies
• Antisymmetry: [X, Y ] = −[Y, X] for all X, Y ∈ g.
• Jacobi identity: [X, [Y, Z]] = [[X, Y ], Z] + [Y, [X, Z]] for all X, Y, Z ∈ g.
2. Examples:
• Let V be a vector space over a field K. Then gl(V ) denotes the Lie
algebra of all linear mappings V → V with [X, Y ] = X ◦ Y − Y ◦ X.
• For V = Kn we write gl(n, K) := gl(V ). This is the Lie algebra of all
n × n matrices with entries in K.
• so(3) = { X ∈ gl(3, R) | X T = −X } is the Lie algebra of rotations.
• su(2) = { X ∈ gl(2, C) | X † = −X, Tr(X) = 0 }.
3. Def. Let g, h be Lie algebras. A linear map h : g → h is a Lie algebra
homomorphism if h([X, Y ]) = [h(X), h(Y )]. If h is a bijection, it is called an
isomorphism.
4. For example, so(3) and su(2) are isomorphic Lie algebras.
5. Thm. If g is a finite-dimensional Lie algebra then there is a unique, up to
isomorphism, simply-connected Lie group G s.t. G0 = g.
(Recall that a topological space is simply connected if any loop can be con-
tinuously contracted to a point).
6. Fact: If H : G 7→ Ĝ is a Lie group homomorphism then
d
h(X) = H(exp(tX))|t=0 , X ∈ G0 (88)
dt
is a Lie algebra homomorphism. Furthermore H(exp(tX)) = exp(th(X)) for
all t ∈ R.
7. Thm. Let G, Ĝ be Lie groups and suppose G is simply-connected. Then
a Lie algebra homomorphism h : G0 → Ĝ0 can be lifted to a Lie group
homomorphism H.

5.5 Matrix Lie groups

1. Def. A closed subgroup of GL(n, K) is called a matrix Lie group. For example
SO(3) and SU (2).
2. Fact: Let G be a matrix Lie group. The Lie algebra G0 of this Lie group is
given by

G0 = { X ∈ gl(n, K) | etX ∈ G for all t ∈ R }. (89)

29
 We have GL(n, K)0 = gl(n, K), SO(3)0 = so(3), SU (2)0 = su(2). Further-
more, abstract exp : G0 → G coincide with the exponential function of a
matrix.

3. Recall that so(3) and su(2) are isomorphic Lie algebras. However SO(3)
and SU (2) are not isomorphic Lie groups. In particular, SU (2) is simply
connected, SO(3) is not.

30
6 Symmetries II
6.1 Representations
1. Def. A group homomorphism D : G → GL(V ) is called a representation.

2. Let D1 : G → GL(V1 ) and D2 : G → GL(V2 ) be two reps of G.

• Def: The direct sum of D1 , D2 , acting on V1 ⊕ V2 is defined by:

(D1 ⊕ D2 )(g)(v1 ⊕ v2 ) = (D1 (g)v1 ) ⊕ (D2 (g)v2 ). (90)

• Def: The tensor product of D1 , D2 acting on V1 ⊗ V2 is defined by

(D1 ⊗ D2 )(g)(v1 ⊗ v2 ) = (D1 (g)v1 ) ⊗ (D2 (g)v2 ). (91)

3. The property of irreducibility of a representation D can be explained as

follows:

• Def. We say that a subspace W ⊂ V is invariant, if D(g)w ∈ W for

any g ∈ G and w ∈ W ,
• Def. We say that a representation is irreducible if it has no invariant
subspaces except for {0} and V .
• Def. A representation is completely reducible if it is a direct sum of
irreducible representations.
• Fact. (Schur Lemma). Let D : G → GL(V ) be an irreducible represen-
tation and V complex. If A ∈ GL(V ) commutes with all D(g), g ∈ G,
then it has the form A = λI, λ ∈ C.

4. Def: A Lie algebra homomorphism d : g → gl(V ) is called a representation.

Irreducibility and complete reducibility of such representations are defined
analogously as for groups.

5. Thm. Let D : G → GL(V ) be a representation of a Lie group. Then

d
d(X) = D(exp(tX))|t=0 , for X ∈ G0 (92)
dt
defines a representation of G0 .

6. Thm. Let G be a simply-connected Lie group and G0 its Lie algebra. Let
d : G0 → gl(V ) be a representation. Then there exists a unique representa-
tion D : G → GL(V ) s.t. (92) holds.

7. Def. Let G be a matrix Lie group. We say that a bilinear form b : G0 ×G0 → R
is invariant, if b(gXg −1 , gY g −1 ) = b(X, Y ) for all g ∈ G, X, Y ∈ G0 .

31
 8. Fact. Let d : G0 → gl(V ) be a representation, b an invariant bilinear form
on G0 and {X 1 , . . . , X n } a basis in G0 . Then the Casimir operator
X
C := b(X A , X B )d(X A )d(X B ) (93)
A,B

is basis-independent and commutes with all d(X), X ∈ G0 . Thus, by the

Schur Lemma, it is a multiple of unity in any irreducible representation on
complex V .

6.2 Projective representations

1. Let V by a complex vector space.
• Def: U (1) = {eiϕ I | ϕ ∈ R } ⊂ GL(V ).
• Def: GL(V )/U (1) is the (Lie) group generated by the equivalence
classes A = {eiϕ A | ϕ ∈ R}.
• Def: A homomorphism D : G → GL(V )/U (1) is called a projective
representation (a representation up to a phase).
• For equivalence classes we have D(g1 )D(g2 ) = D(g1 g2 ). But for a given
choice of representatives D(g) ∈ D(g), (s.t. D(e) = I, g → D(g)
continuous)

D(g1 )D(g2 ) = eiϕ(g1 ,g2 ) D(g1 g2 ). (94)

for some function ϕ : G × G → R.

2. D gives rise to the homomorphism d : G0 → (GL(V )/U (1))0 given by
d
d(X) = D(exp(tX))|t=0 , X ∈ G0 . (95)
dt

3. We have (GL(V )/U (1))0 = (GL(V )0 /U (1)0 ) = gl(V )/(iR). The elements of
this Lie algebra are the equivalence classes Y = {Y + iz |z ∈ R}.
4. For equivalence classes we have [d(X A ), d(X B )] = C f CAB d(X C ). But for
P
any given choice of representatives d(X A ) ∈ d(X A )
X
[d(X A ), d(X B )] = f CAB d(X C ) − iz A,B I, (96)
C

where z A,B are called the central charges and the (-) sign in the last formula is
a matter of convention. In the physical notation one defines the infinitesimal
generators T A = id(X A ) so that (96) reads
X
[T A , T B ] = if CAB T C + iz A,B I. (97)
C

We follow the mathematical convection below unless stated otherwise.

32
 • In some cases it is possible to eliminate z A,B by passing to different
e A ) = d(X A ) + icA .
representatives d(X
• Then de becomes a Lie algebra representation G0 → gl(V ).
• Hence, by a theorem above, de can be lifted to a Lie group representation
D
e : Ge → GL(V ), where G e is the unique simply-connected Lie group
with the Lie algebra G0 . (The universal covering group).

5. Let us explain in more detail the concept of the covering space/group:

• Def. A topological space G is path connected, if for any g1 , g2 ∈ G there

is a continuous map γ : [0, 1] → G s.t. γ(0) = g1 , γ(1) = g2 .
• Def. A topological space G is simply connected, if it is path connected
and every loop in the space can be continuously contracted to a point.
• Def. The universal cover of a connected topological space G is a simply-
connected space G e together with a covering map Hc : G e → G. The
covering map is a local homeomorphism s.t. the cardinal number of
Hc−1 (g) is independent of g. The universal cover is unique.
• Fact: If G is a Lie group, Ge is also a Lie group and Hc : G e → G is a
homomorphism s.t. ker Hc is a discrete subgroup.

6. The situation above occurs in particular for projective unitary representa-

tions of SO(3).

• One can choose D s.t. eiϕ(g1 ,g2 ) ∈ {±1}. By continuity, eiϕ(g1 ,g2 ) = 1 for
g1 , g2 close to e, hence z A,B = 0.
• Thus D can be lifted to a unitary representation D e of SO(3)
f = SU (2).
More precisely, D(A) = D(Hc (A)), where A ∈ SU (2) and Hc : SU (2) →
e
SO(3) is the covering homomorphism.
• ker Hc := Hc−1 (e) = Z2 thus SU (2)/Z2 ' SO(3) and every element of
SO(3) corresponds to two elements in SU (2). That is SU (2) is a double
covering of SO(3).

6.3 Representations of rotations

1. Fact: so(3) = { X ∈ gl(3, R) | X T = −X } is the Lie algebra of SO(3) =
{ R ∈ GL(3, R) | RT R = I, det R = 1}. Indeed, let X ∈ so(3). Then
T
(etX )T etX = etX etX = e−tX etX = 1 (98)
det(etX ) = etTrX = 1, (99)

where we used that a real anti-symmetric metric has vanishing diagonal

elements and consequently is traceless.

33
2. We choose a basis in so(3) as follows
     
0 0 0 0 0 1 0 −1 0
L1 = 0 0 −1 , L2 =  0 0 0 , L3 = 1 0 0 , (100)
0 1 0 −1 0 0 0 0 0
~
so that eθ~n·L is the rotation around the axis ~n, k~nk = 1, by angle θ.

3. One verifies the commutation relations

[Li , Lj ] = εijk Lk . (101)

These generators are related to the angular momentum operators J i via

Li = −iJ i . They satisfy accordingly

[J i , J j ] = iεijk J k . (102)

4. Some facts about the irreducible representations of so(3):

• From quantum mechanics we know that there is only one Casimir opera-
tor J~2 = J12 +J22 +J32 , whose eigenvalues are j(j+1), for j = 0, 1/2, 1, . . ..
• The irreducible representations d(j) are labelled by j and are 2j + 1
dimensional.
• The basis vectors are denoted |j, mi, m = −j, −j + 1, . . . j, where
d(j) (J3 )|j, mi = m|j, mi.

5. Recall that Hc : SU (2) → SO(3) is the covering homomorphism. It gives

rise to the isomorphism hc : su(2) → so(3) which can be described as follows:
1
• let σ1 , σ2 , σ3 be the Pauli matrices. Then Y j = σ
2i j
is a basis of su(2),
since [Y i , Y j ] = εijk Y k .
• Then hc is given in this basis by hc (Y j ) = Lj .
• h−1
c = d
(1/2)
.

6. Since SU (2) is simply-connected, any representation d(j) gives rise to a repre-

−1 (j)
sentation D(j) of SU (2) according to D(j) (et(hc ) (X) ) = etd (X) . In particular
D(1/2) , is the defining representation of SU (2) as one can show using that
the Pauli matrices are traceless and hermitian.

7. Fact: Only for integer j the representation D(j) of SU (2) can be lifted to a
representation of SO(3).
But, as we know from the previous subsection, for any j it can be lifted to
a projective representation of SO(3). In order to accommodate half-integer
spins, we need to justify that the projective representations of SO(3) corre-
spond to rotation symmetry of physical quantum systems. This will be done
later today, using that quantum states are determined up to a phase.
34
 8. For any j = 0, 1/2, 1, . . . and a rotation R given by the axis ~n and angle θ
we define the Wigner functions
(j) ~(j)
Dmm0 (R) := hj, m|e−iθ~n·J |j, m0 i, (103)

where J~(j) denotes here the angular momentum in the representation d(j) .

9. Let us illustrate how the projective character of D(j) (R) for half-integer j
comes about: Consider a rotation R2π by 2π around the ~e3 -axis in the j = 1/2
representation. It can be computed in two ways: First, since rotation by 2π
is equal to identity, we obtain D(1/2) (R) = I. On the other hand, formula
(103) gives
     −iπ 
(1/2)
σ
−i2π 23 1 0 e 0
D (R2π ) = e = exp − iπ = = −I. (104)
0 −1 0 eiπ

Since we got two different results, D(1/2) is not a well defined homomorphism
SO(3) → GL(V ) (i.e. representation). But it can still be a well defined
homomorphism SO(3) → GL(V )/U (1) (i.e. projective representation) as
the two results differ only by a sign and thus belong to the same equivalence
class.

10. Any finite dimensional representation of SU (2) is completely reducible i.e.

can be represented as a direct sum of the irreducible representations D(j) .
In particular, we have

D(j1 ) ⊗ D(j2 ) = ⊕jj=|j

1 +j2
1 −j2 |
D(j) . (105)

This is what is called ”addition of angular momenta”.

6.4 Symmetries of quantum theories

When studying symmetries of a quantum theory, one has to take it seriously that
physical states are defined up to a phase. Thus we consider the following setting:

1. H - Hilbert space of physical states.

2. For Ψ ∈ H, kΨk = 1 define the ray Ψ̂ := { eiθ Ψ | θ ∈ R }.

3. Ĥ - set of rays with the ray product [Φ̂|Ψ̂] := |hΦ|Ψi|2 .

Definition 6.1 A symmetry transformation of a quantum system is an invertible

map Û : Ĥ → Ĥ s.t. [Û Φ̂|Û Ψ̂] = [Φ̂|Ψ̂]. Such transformations form a group.

Theorem 6.2 (Wigner) For any symmetry transformation Û : Ĥ → Ĥ we can

find a unitary or anti-unitary operator U : H → H s.t. Û Ψ̂ = U
d Ψ. U is unique
up to phase.
35
Anti-unitary operators are defined as follows:

• A map U : H → H is anti-linear if U (c1 Ψ1 + c2 Ψ2 ) = c1 U Ψ1 + c2 U Ψ2 .

• The adjoint of an anti-linear map is given by hΦ|U † Ψi = hU Φ|Ψi.

• An anti-linear map is called anti-unitary if U † U = U U † = I in which case

hU Φ|U Ψi = hΦ|U † U Ψi = hΨ|Φi.

Anti-unitary operators will be needed to implement discrete symmetries, e.g. time

reversal. For symmetries described by connected Lie groups anti-unitary operators
can be excluded, as we indicate below.

Application of the Wigner theorem:

1. Suppose a connected Lie group G is a symmetry of our theory i.e. there is a

group homomorphism G 3 g 7→ Û (g) into symmetry transformations.

2. The Wigner theorem gives corresponding unitary operators U (g). Since they
are determined up to a phase, they form only a projective representation:

U (g1 )U (g2 ) = eiθ1,2 U (g1 g2 ). (106)

(Since G is connected, we can exclude that some U (g) are anti-unitary.

Indeed for a connected group we have g = g02 for some g0 ∈ G. Now
U (g) = e−iθ U (g0 )U (g0 ) which is unitary no matter if U (g0 ) is unitary or
anti-unitary).

3. As discussed above5 , for a large class of connected Lie groups G (including

SO(3) and P+↑ ) a projective unitary representation of G corresponds to an
ordinary unitary representation of the covering group G e

e 3 g̃ 7→ Ũ (g̃) ∈ B(H).
G (107)

In particular, projective unitary representations of SO(3) correspond to ordi-

nary unitary representations of SU (2) and thus there is room for half-integer
spin!

5
Strictly speaking, in previous sections we tacitly assumed that the representations act on
finite-dimensional vector spaces, while here H can be infinite dimensional. Fortunately, the
relevant results generalize.

36
7 Mathematical framework for QED
Main reference for this section is [17, Chapter 7].

7.1 Classical field theory

Consider a Lagrangian L(φ, φ∗ , ∂φ, ∂φ∗ ; A, ∂A) where A is a vector field, φ a com-
plex scalar field. Suppose first that L is invariant under the transformations
φε (x) = eiε(x) φ(x), φ∗ε (x) = e−iε(x) φ∗ (x), Aµ,ε (x) = Aµ (x) + ∂µ ε(x), (108)
for some function ε. Thus the corresponding variation of the Lagrangian must
vanish. Exploiting the Euler-Lagrange equations we get
δL = (∂µ j µ )ε + (j ν − ∂µ F µν )∂ν ε − (F µν )∂µ ∂ν ε, (109)
where
∂L ∂L ∂L
j µ (x) := (x)iφ(x) − ∗
(x)iφ∗ (x), F µν := − (110)
∂(∂µ φ) ∂(∂µ φ ) ∂(∂µ Aν )
• From the ∂µ ∂ν ε-term of the (109) we obtain that the symmetric part of F µν
vanishes, i.e.
F µν = −F νµ (111)

• From the ∂ν ε-term of the (109) we get the local Gauss Law
∂µ F µν = j ν (112)

• From the ε-term of (109) we get ∂µ j µ = 0 and ∂t Q = 0, where

Z
Q = d3 y j 0 (0, ~y ). (113)

(Noether’s theorem)
• Furthermore, Q is the infinitesimal generator of the global U (1) symmetry,
i.e.
d
{Q, φ(0, ~x)} = − φε (0, ~x)|ε=0 = −iφ(0, ~x), (114)
dε
d
{Q, φ∗ (0, ~x)} = − φ∗ε (0, ~x)|ε=0 = iφ∗ (0, ~x). (115)
dε
Here the Poisson bracket is defined by
Z  
3 δF δG δF δG
{F, G} = d z − + ··· (116)
δφ(0, ~z) δπ(0, ~z) δπ(0, ~z) δφ(0, ~z)
where π(z) = ∂(∂∂L0 φ)(z)
is the canonical momentum and the omitted terms
∗
correspond to φ and A. (Note, however, that terms corresponding to A are
not relevant for (115)).
37
 • On the other hand using ∂µ F µν = j ν we can compute
Z
−iφ(0, ~x) = {Q, φ(0, ~x)} = d3 y {j 0 (0, ~y ), φ(0, ~x)}
Z Z
3 i,0
= d y {∂i F (0, ~y ), φ(0, ~x)} = lim ~ · E(0,
d3 y { ∇ ~ ~y ), φ(0, ~x)}
R→∞ B
R
Z
= lim ~ ~y ), φ(0, ~x)},
d~σ (~y ) · {E(0, (117)
R→∞ ∂BR

where E~ := (F 1,0 , F 2,0 , F 3,0 ), BR is a ball of radius R centered at zero,

∂BR its boundary (a sphere) and we used the Stokes theorem. Note that in
quantum theory, where { · , · } → −i[ · , · ] above6 , the last expression would
be zero by locality, giving a contradiction!

• One possible way out (which we will not follow) is to abandon locality of
charged fields but keep ∂µ F µν = j ν (Quantisation in the Coulomb gauge).

• We will follow instead the Gupta-Bleuler approach, where all fields are local,
but hΨ1 |(∂µ F µν − j ν )Ψ2 i = 0 only for Ψ1 , Ψ2 in some ‘physical subspace’
H0 ⊂ H. This will enforce hΨ|Ψi < 0 for some Ψ ∈ H thus we have to use
‘indefinite metric Hilbert spaces’ (Krein spaces)

• Incidentally, local, Poincaré covariant massless vector fields Aµ do exist on

Krein spaces (which is not the case on Hilbert spaces). Thus we will have
candidates for the electromagnetic potential.

7.2 Strocchi-Wightman framework [18, 19]

Definition 7.1 An ‘indefinite metric Hilbert space’ (Krein space) H is a vector
space equipped with a sesquilinear form h · | · i s.t.
• h · | · i is non-degenerate, i.e. for any Ψ 6= 0 there is some Φ ∈ H s.t.
hΨ|Φi = 6 0.

• H carries an auxiliary positive-definite scalar product ( · | ·) w.r.t. which it is

a Hilbert space.

• There is a bounded, invertible operator η on H, self-adjoint w.r.t. ( · | ·), s.t.

hΨ1 |Ψ2 i = (Ψ1 |ηΨ2 ).

Only the first property above is physically important. The role of the last two
properties is to provide a topology on H which is needed for technical reasons (e.g.
density of various domains).
Definition 7.2 A Strocchi-Wightman relativistic quantum mechanics is given by:
6
It should be mentioned that for gauge theories the standard quantisation prescription
{ · , · } → −i[ · , · ] may fail in general and a detour via ‘Dirac brackets’ is required. However, in
the above situation the simple analogy can be maintained.
38
 1. A Krein space H.
2. A physical subspace H0 ⊂ H s.t. hΨ|Ψi ≥ 0 for Ψ ∈ H0 .
3. The physical Hilbert space Hph := (H0 /H00 )cpl , where H00 := {Ψ ∈ H0 | hΨ|Ψi =
0}. Its elements are equivalence classes [Ψ] = {Ψ + Ψ0 | Ψ0 ∈ H00 }, where
Ψ ∈ H0 .
e+↑ 3 (Λ,
4. A h · | · i-unitary representation P e a) in H s.t. H0 is
e a) 7→ U (Λ,
invariant under U . Then U induces a unitary representation on Hph by
Uph (Λ,
e a)[Ψ] = [U (Λ, e a)Ψ]. We assume that Uph is continuous and satisfies
the spectrum condition.
5. A unique (up to phase) vacuum vector Ω ∈ H0 s.t. hΩ|Ωi = 1 and U (Λ,
e a)Ω =
Ω.
Definition 7.3 A Strocchi-Wightman QFT is given by:
1. A Strocchi-Wightman relativistic QM (H, H0 , U, Ω).
(κ)
2. A family of operator-valued distributions (φ` , D), κ ∈ I, ` = 1, 2, . . . rκ s.t.
• I is some finite or infinite collection of indices numbering the types
of the fields corresponding to finite-dimensional representations D(κ) of
Le↑+ = SL(2, C).
(κ)
• For a fixed κ ∈ I the field φ(κ) = (φ` )`=1,...rκ transform under D(κ) .
(κ) (κ̄)
• For any κ, ` there exists some κ, ` s.t. φ (f )† = φ ¯ (f¯).
` `

• Ω ∈ D and U (Λ,
e a)D ⊂ D for all (Λ,
e a) ∈ e+↑ .
P
satisfying:
(a) (Locality) If supp f1 and supp f2 are spacelike separated, then
(κ) (κ0 ) (κ) (κ0 )
[φ` (f1 ), φ`0 (f2 )]− = 0 or [φ` (f1 ), φ`0 (f2 )]+ = 0 (118)
in the sense of weak commutativity on D. (Here −/+ refers to commutator/anti-
commutator).

e a) ∈ P
(b) (Covariance) For all (Λ, e+↑ and f ∈ S
X (κ)
e a)φ(κ) (f )U (Λ,
U (Λ, e a)† = e −1 )φκ`0 (f(Λ,a) ).
D``0 (Λ (119)
`
`0

Here f(Λ,a) (x) = f (Λ−1 (x − a)).

(κ ) (κ )
(c) (Cyclicity of the vacuum) D = Span{ φ`1 1 (f1 ) . . . φ`mm (fm )Ω | f1 , . . . fm ∈
S, m ∈ N0 } is a dense subspace of H in the topology given by ( · | · ).
(κ)
The distributions (φ` , D) are called the Strocchi-Wightman quantum fields.
39
7.3 Free field examples
7.3.1 Free Wightman fields
1. Suppose first that H is a Hilbert space (i.e. h· | ·i is a positive-definite scalar
product and we can choose H0 = H). Then the above setting is called the
Wightman framework for fields with arbitrary (finite) spin.
2. Let us stay for a moment in this Hilbert space framework. Recall that
finite-dimensional irreducible representations D(κ) of Le↑+ are labelled by two
numbers (A, B). From the physics lecture you know the following free field
examples:
• (0, 0): scalar field φ
• 21 , 12 : massive vector field j µ

• 21 , 0 ⊕ 0, 21 : Dirac field ψ

• (1, 0) ⊕ (0, 1): Faraday tensor F µν .

3. It is however not possible to construct a massless free vector field Aµ on a
Hilbert space which is local and Poincaré covariant [25]. Such fields turn
out to exist on Krein spaces, which is usually given as the main reason to
introduce them.

7.3.2 Free massless vector field Aµ on Krein space

The Gupta-Bleuler electromagnetic potential has the form
3
d3 k X (λ)
Z
−ikx
Aµ (x) = [a (k)ε(λ)
µ (k)e + a(λ)† (k)ε(λ)∗
µ (k)e
ikx
], (120)
2k0 (2π)3 λ=0
(λ)
where k0 = |k| and εµ are polarisation vectors which satisfy the orthogonality
and completeness relations
0 )∗ 0
X
ε(λ),µ (k) · ε(λ
µ (k) = g λλ , (g λλ )−1 ε(λ) (λ)∗
µ (k) · εν (k) = gµν . (121)
λ

For the aλ , a(λ)† we have

0 0
[a(λ) (k), a(λ )† (k 0 )] = −g λλ 2k 0 (2π)3 δ(~k − ~k 0 ). (122)
Due to −g 00 = −1 we have ha(0)† (f )Ω|a(0)† (f )Ωi < 0 thus our ‘Fock space’ H turns
out to be a Krein space. Furthermore, the ‘photons’ above have four polarisations
and not two. These unphysical degrees of freedom will be eliminated by the Gupta-
Bleuler subsidiary condition (127) below.
Let us point out another peculiarity of this potential: We can form Fµν =
∂µ Aν − ∂ν Aµ so that εαβµν ∂β Fµν (x) = 0 is automatic. But the remaining free
Maxwell equations fail:
∂µ F µν (x) = −∂ ν (∂ρ Aρ )(x) 6= 0. (123)
This can be expected on general grounds:
40
Theorem 7.4 (Strocchi [20, 21]) Any Strocchi-Wightman vector field Aµ with
∂µ F µν (x) = 0 is trivial, i.e. hΩ|F µν (x)F αβ (y)Ωi = 0.
This highlights the necessity of a gauge-fixing term in the Lagrangian, which is
another point to which we will come below in the context of interacting QED.

7.4 Quantum Electrodynamics

Definition 7.5 QED is a Strocchi-Wightman QFT whose fields include F µν , j µ
and some ‘charged fields’ φ(κ) s.t. the physical subspace H0 satisfies:
(i) There is a dense domain D0 ⊂ H0 s.t. F µν (f )D0 ⊂ D0 , j µ (f )D0 ⊂ D0 and
e 0 ⊂ D0 .
U (a, Λ)D

(ii) For Ψ1 ∈ H0 and Ψ2 ∈ D0

hΨ1 |(∂µ F µν − j ν )(f )Ψ2 i = 0, hΨ1 |(εµνρσ ∂ ν F ρσ )(f )Ψ2 i = 0. (124)

For QED defined Ras above, one can define the electric charge operator by suitably
regularizing Q = d3 y j 0 (0, ~y ).
Theorem 7.6 (Strocchi-Picasso-Ferrari [22]) Suppose that Q is an infinitesimal
generator of the global U(1) symmetry, i.e. for some field φ

φ(x) = [Q, φ(x)] on H (125)

6 0 for some Ψ1 , Ψ2 ∈ H0 . Then

and that hΨ1 |φ(x)Ψ2 i =
1. There is Ψ ∈ D0 s.t. (∂µ F µν − j ν )(x)Ψ 6= 0.

2. There is 0 6= Ψ ∈ H0 s.t. hΨ|Ψi = 0, i.e. H00 6= {0}.

3. There is Ψ ∈ H s.t. hΨ|Ψi < 0.

The proof is simple and relies on a Stokes theorem computation analogous to (117).
This theorem shows that the Maxwell equations can hold on H0 at best in matrix
elements and that the Krein space framework is needed in (local) QED also in the
presence of interactions.
Definition 7.7 We say that QED is in the Gupta-Bleuler gauge if it contains (in
addition to other fields) a vector field Aµ s.t. Fµν = ∂µ Aν − ∂ν Aµ and

∂µ F µν − j ν = −∂ ν (∂ρ Aρ ) (126)

holds as an operator identity on H. Furthermore, the physical subspace is chosen

as

H0 := {Ψ ∈ H | (∂ρ Aρ )(+) (f )Ψ = 0 for all f ∈ S }, (127)

where (∂ρ Aρ )(+) is the positive frequency part of (∂ρ Aρ ).

41
We add several remarks on this definition:
1. Note that by applying ∂ν to (126) and using current conservation we obtain
(∂ρ Aρ ) = 0, thus the decomposition into positive and negative frequency
parts is meaningful. For this reason, (124) formally hold. But positivity of
the scalar product on H0 needs to be assumed. (Known only for the free
electromagnetic field).

2. The equation (126) comes from a classical Lagrangian with the gauge-fixing
term, e.g.
1 1
Lgf = (Dµ φ)∗ (Dµ φ) − Fµν F µν − (∂µ Aµ )2 . (128)
4 2
Lgf is still invariant under ‘residual’ local gauge transformations s.t. ε(x) =
0. Denote infinitesimal transformations of the fields as δε φ(x) = iε(x)φ(x),
δε φ(x)∗ = −iε(x)φ(x)∗ and δε Aµ (x) = ∂µ ε(x).

3. Def. Let A denote the algebra spanned by polynomials of quantum fields

Aµ , j µ , φ, φ∗ smeared with smooth, compactly supported functions. We
extend δε to A via the Leibniz rule and denote the subalgebra of (residual)
gauge-invariant elements by Agi .

4. It turns out, that a ‘local vector’ (i.e. Ψ = AΩ, where A ∈ A) belongs to H0

iff A ∈ Agi [17].

7.5 Electrically charged states of QED

Important problem in QED is a construction of physical electrically charged states.
Vectors of the form φ(f )Ω, where φ is a charged field, are not in H0 , because φ is
not invariant under residual gauge transformations. Moreover:
Proposition 7.8 [17] For any local vectors Ψ, Φ ∈ H0 in QED we have hΨ|QΦi =
0.
We face the problem of constructing a field φC which is invariant under (residual)
local gauge transformations (so that φC (f )Ω is ‘close’ to H0 ) and non-invariant
under global gauge transformations (so that φC (f )Ω is charged). Here is a candi-
date:
−1 ∂ i ](x) −1 ∂ i +∂ i ε)](x)
φC (x) := ei[(∆) iA
φ(x) = ei[(∆) i (A
eiε(x) φ(x), (129)

where the last equality holds for local gauge transformations but fails for global
(i.e. ε(x)=const). φC is simply the (non-local) charged field in the Coulomb gauge
expressed in terms of the Gupta-Bleuler fields. Indeed, we can see (129) as a gauge
transformation. Then the corresponding transformation applied to the potential
gives

Aµ,C (x) = Aµ (x) + ∂µ [(∆)−1 ∂i Ai ](x). (130)

42
which satisfies ∇~ ·A
~ C = 0. Then φC (f )Ω is a candidate for an electrically charged
states. Since φC (f ) is a very singular objects, this vector ‘escapes’ from H and
its control (in the perturbative or axiomatic setting) requires subtle mathematical
methods (cf. [17, 18]). These are outside of the scope of these lectures.

References
[1] M. Reed, B. Simon, Methods of modern mathematical physics I: Functional
Analysis. Academic Press, 1975.

[2] M. Reed, B. Simon, Methods of modern mathematical physics II: Fourier

analysis, self-adjointness. Academic Press, 1975.

[3] M. Reed, B. Simon, Methods of modern mathematical physics III. Scattering

theory. Academic Press,1979.

[4] W. Arveson, The harmonic analysis of automorphism groups. In Operator

algebras and applications, Part I (Kingston, Ont., 1980), Proc. Sympos. Pure
Math., 38, Amer. Math. Soc., Providence, R.I.,1982.D., pp. 199-269.

[5] J. Glimm, A. Jaffe, Quantum physics. A functional integral point of view.

Springer 1987.

[6] W. Rudin, Real and Complex Analysis. McGraw-Hill Book Company, 1987.

[7] N. Bohr and L. Rosenfeld, Kgl. Danske Vid. Sels., Math.-fys. Medd. 12 (1933).

[8] K. Hepp, On the connection between the LSZ and Wightman Quantum Field
Theory. Commun. Math. Phys. 1, 95–111 (1965).

[9] K. Osterwalder and R. Sénéor, The scattering matrix is nontrivial for weakly
coupled P (φ)2 models. Helvetica Physica Acta 49, (1976).

[10] G. Keller, C. Kopper, M. Salmhofer, Perturbative renormalization and effec-

tive Lagrangians in φ44 . Helv. Phys. Acta 65 32-52 (1992).

[11] J. Polchinski, Renormalization and effective Lagrangians. Nuclear Physics B

231, 269-295 (1984).

[12] W. Dybalski and C. Gérard, A criterion for asymptotic completeness in local

relativistic QFT. Commun. Math. Phys. 332, (2014) 1167–1202.

[13] W. Dybalski, Haag-Ruelle scattering theory in presence of massless particles.

Lett. Math. Phys. 72, 27–38 (2005).

[14] P.J. Olver, Applications of Lie groups to differential equations. Springer 1986.

[15] A.W. Knapp, Lie groups beyond an introduction. Birkhäuser 1996.

43
[16] A. Trautman, Grupy oraz ich reprezentacje z zastosowaniami w fizyce. Lecture
notes, 2011.

[17] F. Strocchi, An introduction to non-perturbative foundations of quantum field

theory. Oxford University Press, 2013.

[18] 0. Steinmann, Perturbative quantum electrodynamics and axiomatic field the-

ory. Springer, 2000.

[19] F. Strocchi and A. S. Wightman, Proof of the charge superselection rule in

local relativistic quantum field theory. J. Math. Phys. 15, 2198 (1974)

[20] F. Strocchi, Gauge problem in quantum field theory. Phys. Rev. 162 1429
(1967).

[21] F. Strocchi, Gauge problem in quantum field theory III. Phys. Rev. D 2 2334
(1970).

[22] R. Ferrari, L.E. Picasso and F. Strocchi, Some remarks on local operators in
quantum electrodynamics. Commun. Math. Phys. 35 25 (1974).

[23] D. Buchholz, Gauss’ Law and the infraparticle problem. Physics Letters B 174
331334 (1986).

[24] J. Fröhlich, G. Morchio and S. Strocchi, Infrared problem and spontaneous

breaking of the Lorentz group in QED. Physics Letters B 89 61-64 (1979).

[25] A.S. Wightman and L. Garding, Fields as operator-valued distributions in

relativistic quantum theory. Arkiv för Fysik 28 (1964) 129–184.

44