Mathematical Introduction

to Quantum Field Theory

Jan Dereziński
Dept. of Math. Methods in Phys.,
Faculty of Physics, University of Warsaw
Pasteura 5, 02-093 Warszawa, Poland
email [email protected]
June 10, 2024

Lasciate ogni speranza, voi ch’ intrate. (Dante Aligheri, Commedia Divina)

Contents
1 Minkowski space 5
1.1 Coordinates in Minkowski space . . . . . . . . . . . . . . . . . . . 5
1.2 Causal structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Lorentz and Poincaré groups . . . . . . . . . . . . . . . . . . . . 6
1.4 Euclidean space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Joint spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6 Quantum mechanics . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7 Relativistic quantum mechanics . . . . . . . . . . . . . . . . . . . 10

2 Algebras and axioms 11

2.1 States and observables . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Superselection sectors . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Composite quantum systems . . . . . . . . . . . . . . . . . . . . 12
2.4 ∗-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 Commutant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 Observables – infinite dimension . . . . . . . . . . . . . . . . . . 15
2.7 Haag-Kastler axioms for observable algebras . . . . . . . . . . . . 16
2.8 Quantum fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.9 Wightman axioms for bosonic fields . . . . . . . . . . . . . . . . 17
2.10 Relationship between Haag-Kastler and Wightman axioms . . . . 19

1
3 The Laplace and Helmholtz equation 20
3.1 Tempered distributions . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Fourier transformation . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Green’s functions on the Euclidean space . . . . . . . . . . . . . 21
3.4 Bessel equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.5 Macdonald function . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.6 Bessel and Hankel functions . . . . . . . . . . . . . . . . . . . . . 25

4 Wave and Klein-Gordon equations 26

4.1 Propagators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.2 Invariant measure . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3 Propagators for the Klein-Gordon and wave equation . . . . . . . 28
4.4 Einstein causality of propagators . . . . . . . . . . . . . . . . . . 29
4.5 Propagators in position representation in dimension 1 + 3. . . . . 29
4.6 Classical propagators and solving the Klein-Gordon equation . . 31

5 Second quantization 32
5.1 Vector and Hilbert spaces . . . . . . . . . . . . . . . . . . . . . . 32
5.2 Direct sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.3 Tensor product . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.4 Fock spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.5 Creation/annihilation operators . . . . . . . . . . . . . . . . . . . 36
5.6 Integral kernel of an operator . . . . . . . . . . . . . . . . . . . . 37
5.7 Second quantization of operators . . . . . . . . . . . . . . . . . . 38
5.8 Symmetric/antisymmetric tensor product . . . . . . . . . . . . . 39
5.9 Exponential law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.10 Wick symbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.11 Wick symbol and coherent states . . . . . . . . . . . . . . . . . . 42
5.12 Particle number preserving operators . . . . . . . . . . . . . . . . 42
5.13 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

6 Formalism of classical mechanics 44

6.1 Dynamical systems . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.2 Hamiltonian dynamics . . . . . . . . . . . . . . . . . . . . . . . . 44
6.3 Symplectic form . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.4 Lagrangian formalism . . . . . . . . . . . . . . . . . . . . . . . . 47
6.5 Noether Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.6 Classical field theory . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.7 Hyperbolic classical field theory . . . . . . . . . . . . . . . . . . . 49

7 Canonical Commutation Relations 50

7.1 Symplectic vector spaces . . . . . . . . . . . . . . . . . . . . . . . 50
7.2 Quantization of linear and quadratic observables . . . . . . . . . 52
7.3 Quantization of symplectic transformations . . . . . . . . . . . . 53
7.4 Weyl operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
7.5 Stone-von Neumann Theorem . . . . . . . . . . . . . . . . . . . . 55

2
 7.6 Representations of the CCR . . . . . . . . . . . . . . . . . . . . . 55
7.7 Fock representations of the CCR . . . . . . . . . . . . . . . . . . 56
7.8 Equivalence of representations of CCR . . . . . . . . . . . . . . . 57
7.9 Two steps of quantization with an infinite number of degrees of
freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
7.10 Positive energy Fock quantization . . . . . . . . . . . . . . . . . . 60
7.11 Positive energy quantization for charged systems . . . . . . . . . 63

8 Free neutral scalar bosons 65

8.1 Classical fields off-shell and on-shell . . . . . . . . . . . . . . . . 65
8.2 Symplectic form and Poisson bracket . . . . . . . . . . . . . . . . 66
8.3 Stress-energy tensor . . . . . . . . . . . . . . . . . . . . . . . . . 68
8.4 Simultaneous diagonalization of the symplectic form, Hamilto-
nian and momentum . . . . . . . . . . . . . . . . . . . . . . . . . 69
8.5 Plane wave functionals . . . . . . . . . . . . . . . . . . . . . . . . 70
8.6 Positive frequency space . . . . . . . . . . . . . . . . . . . . . . . 71
8.7 Quantization of scalar fields . . . . . . . . . . . . . . . . . . . . . 73
8.8 Two-point functions . . . . . . . . . . . . . . . . . . . . . . . . . 75
8.9 Spacetime smeared fields . . . . . . . . . . . . . . . . . . . . . . . 76
8.10 Physical meaning of the 2-point functions and Feynman propagators 77

9 Free charged scalar bosons 78

9.1 Lagrangian formalism . . . . . . . . . . . . . . . . . . . . . . . . 78
9.2 Charged fields as a pair of neutral fields . . . . . . . . . . . . . . 79
9.3 Classical 4-current . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.4 Stress-energy tensor . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.5 Simultaneous diagonalization . . . . . . . . . . . . . . . . . . . . 81
9.6 Negative frequency space . . . . . . . . . . . . . . . . . . . . . . 82
9.7 Plane wave functionals . . . . . . . . . . . . . . . . . . . . . . . . 82
9.8 Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
9.9 Smeared fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

10 Some historical remarks on QFT 86

10.1 Physicist’s strategy in QFT . . . . . . . . . . . . . . . . . . . . . 86
10.2 Renormalizability . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
10.3 Counterterms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
10.4 Asymptotic freedom . . . . . . . . . . . . . . . . . . . . . . . . . 89
10.5 Axiomatic Quantum Field Theory . . . . . . . . . . . . . . . . . 90
10.6 Constructive Field Theory . . . . . . . . . . . . . . . . . . . . . . 91

11 Symplectic and metaplectic group 92

11.1 Classical and quantum mechanics over a symplectic vector space 92
11.2 Quadratic Hamiltonians . . . . . . . . . . . . . . . . . . . . . . . 93
11.3 Weyl-Wigner quantization for a symplectic vector space . . . . . 94
11.4 The Weyl-Wigner symbol of the exponential of a quadratic Hamil-
tonians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

3
12 Metaplectic group in the Schrödinger representation 96
12.1 Generating functions of symplectic transformations . . . . . . . . 96
12.2 Action integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
12.3 Composition of generating functions . . . . . . . . . . . . . . . . 98
12.4 Linear symplectic transformations . . . . . . . . . . . . . . . . . 99
12.5 The metaplectic group . . . . . . . . . . . . . . . . . . . . . . . . 100
12.6 The Weyl-Wigner quantization in the Schrödinger representation 102

13 Pseudounitary spaces 103

13.1 From complex to real spaces and back . . . . . . . . . . . . . . . 103
13.2 Bilinear forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
13.3 Sesquilinear forms . . . . . . . . . . . . . . . . . . . . . . . . . . 104
13.4 Involutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
13.5 Admissible involutions . . . . . . . . . . . . . . . . . . . . . . . . 105
13.6 Pseudo-unitary transformations as 2x2 matrices . . . . . . . . . . 105
13.7 Symplectic transformations as 2x2 matrices . . . . . . . . . . . . 107
13.8 Pseudo-unitary generators . . . . . . . . . . . . . . . . . . . . . . 108
13.9 Unitary operators on Krein spaces . . . . . . . . . . . . . . . . . 109
13.10Positive symplectic transformations . . . . . . . . . . . . . . . . . 109
13.11Pairs of admissible involutions . . . . . . . . . . . . . . . . . . . . 110

14 Fock representation in the real (or neutral) formalism 112

14.1 Canonical commutation relations . . . . . . . . . . . . . . . . . . 112
14.2 Fock representation . . . . . . . . . . . . . . . . . . . . . . . . . . 113
14.3 Squeezed vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
14.4 Metaplectic group in the Fock representation . . . . . . . . . . . 114
14.5 Implementation of symplectic transformations . . . . . . . . . . . 115
14.6 Comparison of two Fock representations . . . . . . . . . . . . . . 116

15 Fock representation in the complex (or charged) formalism 118

15.1 Charged canonical commutation relations . . . . . . . . . . . . . 118
15.2 Fock representations . . . . . . . . . . . . . . . . . . . . . . . . . 119
15.3 Gauge invariant squeezed vectors . . . . . . . . . . . . . . . . . . 119
15.4 Comparison of squeezed vectors in the real and complex formalism120
15.5 Implementation of pseudo-unitary transformations . . . . . . . . 121
15.6 Comparison of two Fock representations . . . . . . . . . . . . . . 122

16 Coherent states 123

16.1 General coherent states in the Schrödinger representation . . . . 123
16.2 From Schrödinger to Fock representation . . . . . . . . . . . . . . 124
16.3 Bargmann-Segal representation . . . . . . . . . . . . . . . . . . . 125
16.4 Bargmann kernel . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
16.5 Examples of Bargmann kernels . . . . . . . . . . . . . . . . . . . 126
16.6 Wick symbol of an operator . . . . . . . . . . . . . . . . . . . . . 128

4
17 Time-dependent Hamiltonians 129
17.1 Schrödinger and Heisenberg picture . . . . . . . . . . . . . . . . . 129
17.2 Time-ordered exponential . . . . . . . . . . . . . . . . . . . . . . 129
17.3 Schrödinger and Heisenberg picture for time-dependent Hamilto-
nians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
17.4 Classical dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 131
17.5 Time-dependent perturbations . . . . . . . . . . . . . . . . . . . 133

18 Euclidean fields and spectral shift function 134

18.1 Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
18.2 Neutral Euclidean fields . . . . . . . . . . . . . . . . . . . . . . . 135
18.3 Charged Euclidean fields . . . . . . . . . . . . . . . . . . . . . . . 139
18.4 Spectral shift function . . . . . . . . . . . . . . . . . . . . . . . . 142

19 Scalar field with a masslike perturbation 144

19.1 Lagrangian and Hamiltonian formalism . . . . . . . . . . . . . . 144
19.2 Dynamics in the interaction picture . . . . . . . . . . . . . . . . . 145
19.3 Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
19.4 Quantum Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . 148
19.5 The vacuum energy . . . . . . . . . . . . . . . . . . . . . . . . . . 149
19.6 Renormalized scattering operator and Hamiltonian . . . . . . . . 151

1 Minkowski space
1.1 Coordinates in Minkowski space
By definition, the Minkowski space, denoted R1,n , is the vector space R1+n
equipped with the canonical pseudo-Euclidean form of signature (− + · · · +).
Its coordinates will be typically denoted by xµ , µ = 0, 1, . . . , n. The pseudo-
Euclidean form is then given by
n
X
gµν xµ xν = −(x0 )2 + (xi )2 . (1.1)
i=1

(Throughout these notes the velocity of light has the value 1 and we use the
Einstein summation convention). We use the metric tensor [gµν ] to lower indices
and its inverse [g µν ] to raise indices:

xµ = gµν xν , xµ = g µν xν .

For a function R1,n 3 x 7→ f (x), we will sometimes use various kind of

notation for partial derivatives:
∂f (x)
= ∂xµ f (x) = ∂µ f (x) = f,µ (x).
∂xµ
Writing Rn we will typically denote the spatial part of the Minkowski space
obtained by setting x0 = 0. If x ∈ R1,n , then ~x will denote the projection of

5
x onto Rn . Latin letters i, j, k will sometimes denote the spatial indices of a
vector. Note that xi = xi .
On R1,n we have the standard Lebesgue measure denoted dx. The notation
d~x will be used for the Lebesgue measure on Rn ⊂ R1,n .
We will often write t for x0 = −x0 . The time derivative will be often denoted
by a dot:
∂f (t) ∂f (x0 )
f˙(t) = = ∂t f (t) = = ∂0 f (x0 ) = f,0 (x0 ).
∂t ∂x0

1.2 Causal structure

A nonzero vector x ∈ R1,n is called
timelike if xµ xµ < 0,
causal if xµ xµ ≤ 0,
lightlike if xµ xµ = 0,
spacelike if xµ xµ > 0.
A causal vector x is called
future oriented if x0 > 0,
past oriented if x0 < 0.
The set of future/past oriented causal vectors is called the future/past light
cone and denoted J ± . We set J := J + ∪ J − .
If O ⊂ R1,n , its causal shadow is defined as J(O) := O + J. We also define
its future/past shadow J ± (O) := O + J ± .
Lemma 1.1. Let Oi ⊂ R1,n , i = 1, 2. Then
J + (O1 ) ∩ O2 =∅ ⇔ O1 ∩ J − (O2 ) = ∅, (1.2)
J(O1 ) ∩ O2 =∅ ⇔ O1 ∩ J(O2 ) = ∅. (1.3)
We will write O1 × O2 iff (1.3) holds. We then say that O1 and O2 are
spatially separated.

1.3 Lorentz and Poincaré groups

The pseudo-Euclidean group O(1, n) is called the full Lorentz group. Its con-
nected component of unity is denoted SO0 (1, n) and called the connected or
proper Lorentz group.
Λ ∈ O(1, n) if for any vector [xµ ] ∈ R1,n
gµν xµ xν = gαβ Λα µ β ν
µ x Λν x . (1.4)
The full Lorentz group contains special elements: the time reversal T and
the space inversion (the parity) P and the space-time inversion X := PT:
T(x0 , ~x) = (−x0 , ~x), P(x0 , ~x) = (x0 , −~x), Xx = −x.

6
It consists of four connected components

SO0 (1, n), T·SO0 (1, n), P·SO0 (1, n), X·SO0 (1, n).

O(1, n) has three subgroups of index two:

SO0 (1, n) ∪ X·SO0 (1, n), (1.5)

SO0 (1, n) ∪ P·SO0 (1, n), (1.6)
SO0 (1, n) ∪ T·SO0 (1, n). (1.7)

(1.5) coincides with SO(1, n) for even 1+n and (1.6) coincides with SO(1, n)
for odd 1 + n.
The affine extension of the full Lorentz group R1,n o O(1, n) is called the full
Poincaré group. Its elements will be typically written as (y, Λ). On x ∈ R1,n it
acts by
(y, Λ)x := y + Λx.
Here is the multiplication:

(y1 , Λ1 )(y2 + Λ) = (y1 + Λ1 y2 , Λ1 Λ2 ). (1.8)

We will often write y instead of (y, 1l) and Λ instead of (0, Λ). It is the full
symmetry group of the Minkowski space.
Example 1.2. Let us determine O(1, 1). We set
1 1
x+ := x + t, x− := x − t; x= (x+ + x− ), t= (x+ − x− ).
2 2
 
a b
Now, let A = .
c d

x2 − t2 = x− x+ = (ax+ + bx− )(cx+ + dx− )

is solved by
ad + bc = 1, ac = 0, bd = 0.
This has 4 types of solutions:

a > 0, d > 0, b = c = 0, (1.9)

a < 0, d < 0, b = c = 0, (1.10)
b > 0, c > 0, a = d = 0, (1.11)
b > 0, c > 0, a = d = 0. (1.12)

Finally, we set    
1 1 −1 a b 1 1
A= .
2 1 1 c d −1 1

7
1.4 Euclidean space
We will sometimes consider the Euclidean space Rd equipped with the form

|x|2 = (x1 )2 + · · · (xd )2 . (1.13)

The orthogonal group O(d) has only two connected components, one of them is
the group SO(d). We also have the Euclidean group Rd o O(d).
Note that if d = 1 + n and we set x0 = ±ixd , then (1.13) becomes the
Minkowski form (1.1). This trick is called the Wick rotation.

1.5 Joint spectrum

Assume first that H is a finite dimensional Hilbert space. The spectrum of an
operator A then is defined as the set of its eigenvalues.
We say that A is self-adjoint if A = A∗ . The spectral theorem says that if
A is self-adjoint, then X
A= a1la (A). (1.14)
a∈sp(A)

where 1la (A) is the orthogonal projection onto eigenvectors of A with eigenvalue
a. More generally, if Ω ⊂ sp(A), then we set
X
1lΩ (A) := 1la (A).
a∈Ω

Let self-adjoint operators A1 , . . . , An commute. Then so do their spectral

projections. Define the joint spectrum of A1 , . . . , An by

sp(A1 , . . . , An ) := {(a1 , . . . , an ) ∈ Rn : 1l{a1 } (A1 ) · · · 1l{an } (An ) 6= 0.}.

For any subset Ω ⊂ sp(A1 , . . . , An ) we define the spectral projection of A1 , . . . , An

onto Ω: X
1lΩ (A1 , . . . An ) := 1l{a1 } (A1 ) · · · 1l{an } (An ).
(a1 ,...,an )∈Ω

Assume now that H is a Hilbert space of any dimension. The spectrum of

an operator A is defined as

sp(A) := {z ∈ C | (z − A)−1 does not exist}.

The set of eigenvalues of A is called the point spectrum of A and is contained

in sp(A)
The spectral theorem says that if A is self-adjoint, then we can define for
any Borel set Ω ⊂ sp(A) the corresponding spectral projection, denoted 1lΩ (A).
They are orthogonal projections. They satisfy

1lΩ1 (A)1lΩ2 (A) = 1lΩ1 ∩Ω2 (A). (1.15)

8
 Let A1 , . . . , An be self-adjoint operators that strongly commute, that is their
spectral projections commute. We say that (a1 , . . . , an ) ∈ sp(A1 , . . . , An ) if for
any  > 0
1l[a1 −,a1 +] (A1 ) · · · 1l[an −,an +] (An ) 6= 0. (1.16)
Clearly

sp(A1 , . . . , An ) ⊂ sp(A1 ) × · · · × sp(An ), (1.17)

−1 −1
sp(U A1 U , . . . , U An U ) = sp(A1 , . . . , An ) (1.18)

Example 1.3. Consider the commuting self-adjoint operators L2 = L2x +L2y +L2z
and Lz .

sp(L2 ) = {`(` + 1) | ` = 0, 1, . . . } (1.19)

sp(Lz ) = Z, (1.20)
sp(L2 , Lz ) = `(` + 1), m | ` = 0, 1, . . . ; m = −`, −` + 1, . . . , ` .


(1.21)
Pd
Example 1.4. Consider pj = −i∂xj , −∆ = j=1 p2j . They are commuting
self-adjoint operators.

sp(pj ) = R; (1.22)
sp(−∆) = [0, ∞[ (1.23)
sp(p1 , p2 , p3 , −∆) = {(k1 , k2 , k3 , k12 + k22 + k32 | k1 , k2 , k3 ∈ R}. (1.24)

This follows from

(F −1 pj Ff )(k) = kj f (k). (1.25)

1.6 Quantum mechanics

Pure quantum states are described by normalized vectors in a Hilbert space
H. The dynamics is usually described by considering a strongly continuous 1-
parameter unitary group on H, that is, a strongly continuous function R 3 t 7→
U (t) ∈ U (H) such that

U (t1 )U (t2 ) = U (t1 + t2 ), t1 , t2 ∈ R,

The Stone Theorem says that U (t) := e−itH for a uniquely defined self-adjoint
operator H, called a Hamiltonian.
In typical situations the Hamiltonian is bounded from below, which means
that there exists E ∈ R such that

(f |Hf ) ≥ E(f |f ), f ∈ H. (1.26)

Equivalently, sp(H) ⊂ [E, ∞[. It does not affect any physical predictions if we
subtract from the Hamiltonian the infimum of its spectrum.
The Hamiltonian has often a ground state, that means inf sp(H) is an eigen-
value. The ground state is often nondegenerate.
It will be convenient to formalize these properties.

9
Definition 1.5. We will say that H, H, Ω satisfy the standard requirements of
quantum mechanics (QM) if
(1) H is a Hilbert space;
(2) H is a positive self-adjoint operator on H (called the Hamiltonian);
(3) Ω is a normalized eigenvector of H with eigenvalue 0;
(4) Ω is nondegenerate as an eigenvector of H.

1.7 Relativistic quantum mechanics

Let us assume that there are no fermions. Relativistic covariance of a quantum
system described by a Hilbert space H is expressed by choosing a strongly
continuous unitary representation of the connected Poincaré group

R1,3 o SO0 (1, 3) 3 (y, Λ) 7→ U (y, Λ) ∈ U (H). (1.27)

We will denote the self-adjoint generator of space-time translations by P =

(P 0 , P~ ). P 0 = H is the Hamiltonian. P~ is called the momentum. Thus
~
U ((t, ~y ), 1l) = e−itH+i~yP .

(We assume that the Planck constant ~ equals 1).

Proposition 1.6. sp(P ) is invariant wrt SO0 (1, n).
Proof. We compute for Λ ∈ SO0 (1, n):

U (Λ)eixP U (Λ)−1 = U (0, Λ)U (x, 1)U (0, Λ−1 ) (1.28)

T
= U (Λx, 1) = ei(Λx)P = eixΛ P
. (1.29)

Differentiating wrt x we obtain

U (Λ)P U (Λ)−1 = ΛT P. (1.30)

Hence ΛT sp(P ) = sp(P ).

Let us show that

SO0 (1, n) = {ΛT | Λ ∈ SO0 (1, n)}. (1.31)

Note that g = g −1 (which is valid in the standard coordinates). Using this we

check that
ΛT gΛ = g ⇒ ΛgΛT = g. (1.32)
Besides, both sides of (1.31) contain 1l and Λ 7→ ΛT is continuous. 2

Definition 1.7. Suppose a representation of the proper Poincaré group is given.

The following conditions will be called the basic requirements of relativistic quan-
tum mechanics (RQM):

10
(1) Existence of a Poincaré invariant vacuum: There exists a (normalized)
vector Ω invariant with respect to R1,3 o SO0 (1, 3).
(2) Spectral condition: The joint spectrum of the energy-momentum operator
is contained in the forward light cone, that is, sp(P ) ⊂ J + .
(3) Uniqueness of the vacuum: The vector Ω is unique up to a phase factor.

Note that conditions (1)-(3) imply the standard requirements of QM.

More precisely, (2) implies H ≥ 0. (1) and (2) imply that Ω is the ground
state of H. (3) implies that this ground state is unique.
Conversely, the Poincaré invariance of sp(P ) and the boundedness from be-
low of H in any system of coordinates imply (2).

2 Algebras and axioms

2.1 States and observables
Let us describe basic framework of quantum mechanics. To avoid technical
complications, in the first part of this section we will assume that the Hilbert
space H describing a quantum system is finite dimensional, so that it can be
identified with CN , for some N .
In basic courses on Quantum Mechanics we learn that a quantum state
is described by a density matrix ρ and a yes/no experiment by an orthogonal
projection P . The probability of the affirmative outcome of such an experiment
equals
Tr(ρP ).
Two orthogonal projections P1 and P2 are simultaneously measurable iff they
commute.
We say that a family of orthogonal projections P1 , . . . , Pn is an orthogonal
partition of unity on H iff
n
X
Pi = 1l, Pi Pj = δij Pj , i, j = 1, . . . n.
i=1

Clearly, all elements of an orthogonal partition of unity commute with one

another. Therefore, in principle, one can design an experiment that measures
simultaneously all of them.
If P1 , . . . , Pn is an orthogonal partition of unity, then setting Hi := RanPi ,
n
i = 1, . . . , n, we obtain an orthogonal direct sum decomposition H = ⊕ Hi .
i=1
Thus specifying an ortogonal partition of unity is equivalnt to speciifying an
orthogonal direct sum decomposition.
To any self-adjoint operator A we can associate an orthogonal partition of
unity given by the spectral projections of A onto its eigenvalues:

1l{a} (A), a ∈ sp(A). (2.1)

11
By measuring thePobservable A we mean measuring the partition of unity (2.1).
Clearly, A = a1la (A). Hence, the average eigenvalue of A in such an
a∈sp(A)
experiment equals X
TrρA = aTrρ1la (A). (2.2)
a∈sp(A)

We call (2.2) the expectation value of the observable A in the state ρ.

2.2 Superselection sectors

So far we assumed that all orthogonal projections on H, hence all self-adjoint
operators on H, correspond to possible experiments. We say that all self-adjoint
elements of B(H) are observable.
Sometimes this is not the case. We are going to describe several situations
where only a part of self-adjoint operators are observable.
It may happen that the Hilbert space H has a distinguished direct sum
decomposition
n
H = ⊕ Hn (2.3)
i=1

such that only self-adjoint operators that preserve each subspace Hi are mea-
surable. We say then that Hi , i = 1, . . . , n, are superselection sectors.
Let Qi denote the orthogonal projection onto Hi . Then linear combinations
of Qi can be measured simultaneously with all other observables. We say that
they are classical observables.
If we choose an o.n. basis of H compatible with (2.3), then only block
diagonal self-adjoint matrices are observable. States are also described by block
diagonal matrices.
Superselection sectors arise typically when we have a strictly conserved quan-
tity, this means a self-adjoint operator Q that commutes with all possible dy-
namics. For instance, the total charge of the system usually determines a su-
perselection sector. Another example of a superselection sector is the fermionic
parity: states of an even and odd number of fermions form two superselection
sectors.

2.3 Composite quantum systems

Suppose that two quantum systems are described by Hilbert spaces H1 , H2 .
Then the composite system is described by the tensor product H1 ⊗ H2 . Ob-
servables of the first system are described by self-adjoint elemens of B(H1 )⊗1lH2 ,
whereas observables of the second system are described by self-adjoint elements
of 1lH1 ⊗ B(H2 ). Note that they commute, so that one can simultaneously
measure them. From the point of view of the first system only self-adjoint ele-
ments of B(H1 ) ⊗ 1lH2 are observable. Again, we have a situation where not all
self-adjoint elements of B(H) are observable.

12
 Let H1 = Cp with an o.n. basis e1 , . . . , ep and H2 = Cq with an o.n. basis
f1 , . . . , fq . Then ei ⊗ fj i = 1, . . . , p, j = 1, . . . , q is an o.n. basis of H1 ⊗ H2 .
Matrices in B(Cp ) ⊗ 1lCq have the form
 
A 0
 0 A
 , A ∈ B(Cp ),


 
A
and matrices in 1lH1 ⊗ B(H2 ) have the form
 
b11 1l b12 1l
 b21 1l b22 1l
[bij ] ∈ B(Cq ),

 ,
 
bqq 1l

2.4 ∗-algebras
n
Consider the Hilbert space H = CN , N =
P
pi qi ,
i=1
n
H = ⊕ Cpi ⊗ Cqi ,
i=1

and the set n

A := ⊕ B(Cpi ) ⊗ 1lqi .
i=1
Note that A is a vector space closed wrt the multiplication and the Hermitian
conjugation. It is an example of what mathematicians call a ∗-algebra, which
we recall below.
As discussed before, in the finite dimensional case, observables of a quantum
system are described by the self-adjoint part of a certain ∗-subalgebra of B(H).
Let A be a vector space over C. We say that A is an algebra if it is equipped
with an operation
A × A 3 (A, B) 7→ AB ∈ A
satisfying
A(B + C) = AB + AC, (B + C)A = BA + CA,
(αβ)(AB) = (αA)(βB).
If in addition
A(BC) = (AB)C,
we say that it is an associative algebra. (In practice by an algebra we will usually
mean an associative algebra).
The center of an algebra A equals
Z(A) = {A ∈ A : AB = BA, B ∈ A}.
Let A, B be algebras. A map φ : A → B is called a homomorphism if it is
linear and preserves the multiplication, ie.

13
(1) φ(λA) = λφ(A);
(2) φ(A + B) = φ(A) + φ(B);
(3) φ(AB) = φ(A)φ(B).
We say that an algebra A is a ∗-algebra if it is equipped with an antilinear
map A 3 A 7→ A∗ ∈ A such that (AB)∗ = B ∗ A∗ , A∗∗ = A and A 6= 0 implies
A∗ A 6= 0.
If H is a Hilbert space, then B(H) equipped with the hermitian conjugation
is a ∗-algebra
If A, B are ∗-algebras, then a homomorphism π : A → B satisfying π(A∗ ) =
π(A)∗ is called a ∗-homomorphism.
Theorem 2.1. (1) Every finite dimensional ∗-algebra A is ∗-isomorphic to
n
⊕ B(Cpi ),
i=1

for some p1 , . . . , pn
(2) If in addition A is a subalgebra of B(CN ) and contains the identity on CN ,
n
pi qi , and a basis of CN such that
P
then there exist q1 , . . . , qn with N =
i=1

n
A = ⊕ B(Cpi ) ⊗ 1lqi . (2.4)
i=1

2.5 Commutant
If B ⊂ B(H), then the commutant of B is defined as

B0 := {A ∈ B(H) : AB = BA, B ∈ B}.

Theorem 2.2. 1. A commutant is always an algebra containing 1lH .

2. If B is ∗-invariant, then so is B0 .
3. B0 = B000 = . . . .
4. B ⊂ B00 = B0000 = . . . .
Proof. 1. and 2. are immediate. The following inclusions are easy and
they imply 3. and 4.:

B1 ⊂ B2 ⇒ B01 ⊃ B02 , (2.5)

00
B⊂B . (2.6)

2
We say that A ⊂ B(H) is a von Neumann algebra if A = A00 . Clearly, von
Neumann algebras are ∗-algebras.

14
 It is easy to see that all ∗-subalgebras of B(CN ) containing 1lN are von
Neumann algebras. Indeed, if A is given by (2.4), then A is obviously ∗-invariant
and n
A0 = ⊕ 1lpi ⊗ B(Cqi ).
i=1
00
So, A = A.

2.6 Observables – infinite dimension

In infinite dimensions we have several technical complications of the formalism
developed in the previous section.
It is still reasonable to assume that observables are described by self-adjoint
elements of a ∗-algebra. However, the theory of ∗-algebras is much richer in
infinite dimension. Here are a few examples of algebras acting on an infinite
dimensional H:
1. Finite rank operators on H.
2. Compact operators on H.

3. Bounded operators on H, that is, B(H).

4. Bounded multiplication operators on H = l2 (N). This algebra is isomor-
phic to l∞ (N).
5. Bounded multiplication operators on H = L2 (R). This algebra is isomor-
phic to L∞ (R).
The definition of a von Neumann algebra is still valid in any dimension. But
Theorem 2.1 does not extend to infinite dimension. Besides, there are other
kinds of ∗-algebras that are interesting candidates for a description of quantum
systems, such as C ∗ -algebras. We will however stick to von Neumann algebras.
Note that in the list above only 3,4,5 are von Neumann algebras.
If B is a ∗-invariant subset of B(H), then B00 is the smallest von Neumann
algebra containing B. We will say that B00 is generated by B. For instance,
the von Neumann algebra generated by finite rank or compact operators is the
whole B(H).
Physically, if we know that self-adjoint operators A1 , . . . , An are observables,
then as the observable algebra it is natural to take

A = {A1 , . . . , An }00 .

Observables are often described by unbounded self-adjoint operators. This is

not a serious problem. What is relevant for quantum measurements are spectral
projections, which are bounded. Thus by saying that an algebra A ⊂ B(H) is
generated by A1 , . . . , An we will mean that it is generated by spectral projections
of these operators (or, equivalently, by their bounded Borel function).

15
 1. Consider the operators φ̂i , i = 1, 2, 3 on L2 (R3 ). They are self-adjoint and
commute. They have simple joint spectrum. The von Neumann algebra
generated by φ̂i , i = 1, 2, 3 is equal to the operators of multiplication by
functions in L∞ (R3 ).
2. Consider in addition the operators π̂i := i−1 ∂xi , i = 1, 2, 3 on L2 (R3 ).
The von Neumann algebra generated by φ̂i , π̂i , i = 1, 2, 3, coincides with
B(L2 (R3 )).

2.7 Haag-Kastler axioms for observable algebras

Let us assume that there are no fermions. Recall that the relativistic covariance
of a quantum system described by a Hilbert space H is expressed by choosing
a strongly continuous unitary representation of the connected Poincaré group
R1,3 o SO0 (1, 3) 3 (y, Λ) 7→ U (y, Λ) ∈ U (H), (2.7)
which defines the 4-momentum P such that U (x, 1l) = eixP . We assume the
basic requirements of relativistic quantum mechanics, which we recall:
(1) Existence of a Poincaré invariant vacuum: There exists a (normalized)
vector Ω invariant with respect to R1,3 o SO0 (1, 3).
(2) Spectral condition: The joint spectrum of the energy-momentum operator
is contained in the forward light cone, that is, sp(P ) ⊂ J + .
(3) Uniqueness of the vacuum: The vector Ω is unique up to a phase factor.
We still need some postulates that express the idea of causality. In the
mathematical physics literature one can find two kinds of axioms that try to
formalize this concept: the Haag-Kastler and the Wightman axioms. Even
though the Wightman axioms were formulated earlier, it is more natural to
start with the Haag-Kastler axioms.
Definition 2.3. In addition to the basic requirements of relativistic quantum
mechanics, suppose to each open bounded set O ⊂ R1,3 we associate a von
Neumann algebra A(O) ⊂ B(H). We will say that the family {A(O)}O is a
net of observable algebras satisfying the Haag-Kastler axioms if the following
conditions hold:
(1) Isotony: O1 ⊂ O2 implies A(O1 ) ⊂ A(O2 ).
(2) Poincaré covariance: for (y, Λ) ∈ R1,3 o SO0 (1, 3), we have
∗


U(y,Λ) A(O)U(y,Λ) = A (y, Λ)O .

(3) Einstein causality: Let O1 × O2 . Then

Ai ∈ A(Oi ), i = 1, 2, implies A1 A2 = A2 A1 .

Self-adjoint elements of the algebras A(O) are supposed to describe observ-

ables in O. This means that in principle an observer contained in O can measure
a self-adjoint operator from A(O), and only from A(O).

16
Remark 2.4. One can ask why von Neumann algebras are used in the Haag-
Kastler axioms to describe sets of observables. We would like to argue that it is
a natural choice.
Suppose we weaken the Haag-Kastler axioms as follows: We replace the
family of von Neumann algebras A(O) by arbitrary sets B(O) of self-adjoint
elements of B(H), and otherwise we keep the axioms unchanged. Then, if we
set A(O) := B(O)00 (which obviously contain B(O)), we obtain a family of von
Neumann algebras satisfying the usual Haag-Kastler axioms. In particular, to
see that the Einstein causality still holds, we use the following easy fact:
Let B1 , B2 , be two ∗-invariant subsets of B(H) such that

A1 ∈ B1 , A2 ∈ B2 implies A1 A2 = A2 A1 .

Set A1 := B001 , A2 := B002 . Then

A1 ∈ A1 , A2 ∈ A2 implies A1 A2 = A2 A1 .

In fact, B1 ⊂ B02 ⇒ B001 ⊂ B000

2 .

2.8 Quantum fields

In practical computations of quantum field theory the information is encoded
in quantum fields. In practice, fields are divided into neutral and charged fields,
which are described in somewhat different formalisms. However, since charged
fields can be decomposed into neutral fields, we will restrict ourselves to neutral
fields.
Again, we will restrict ourselves to bosonic fields. They are typically denoted
by R1,3 3 x 7→ φ̂a (x), where a = 1, . . . , n enumerates the “internal degrees of
freedom”, eg. the species of particles and the value of their spin projected on
a distinguished axis. They commute for spatially separated points, which is
expressed by the commutation relations

[φ̂a (x), φ̂b (y)] = 0, (x − y)2 > 0.

One can try to interpret quantum fields as “operator valued tempered dis-
tributions”, which become (possibly unbounded) self-adjoint operators when
smeared out with real Schwartz test functions. We can organize the internal de-
grees of freedom of neutral fields into a finite dimensional vector space V = Rn .
Thus for any f = (fa ) ∈ S(R1,3 , Rn ) we obtain a smeared-out quantum field,
which is the operator
XZ
φ̂[f ] := fa (x)φ̂a (x)dx. (2.8)
a

2.9 Wightman axioms for bosonic fields

Let us now formulate the Wightman axioms for neutral fields.

17
Definition 2.5. We assume that the basic requirements of Relativistic Quantum
Mechanics are satisfied and V is a finite dimensional real vector space equipped
with a representation

SO0 (1, 3) 3 Λ 7→ σ(Λ) ∈ L(V). (2.9)

We suppose that D is a dense subspace of H containing Ω and we have a

map
S(R1,3 , V) 3 f 7→ φ̂[f ] ∈ L(D) (2.10)
satisfying the following conditions:
(1) Continuity: For any Φ, Ψ ∈ D,

S(R1,3 , V) 3 f 7→ (Φ|φ̂[f ]Ψ) (2.11)

is continuous.
(2) Poincaré covariance: for (y, Λ) ∈ R1,3 o SO0 (1, 3) we have
∗
= φ̂ σ(Λ)f ◦ (y, Λ)−1 .
 
U(y,Λ) φ̂[f ]U(y,Λ)

(3) Einstein causality: Let suppf1 × suppf2 . Then

φ̂[f1 ]φ̂[f2 ] = φ̂[f2 ]φ̂[f1 ].

(4) Cyclicity of the vacuum: Let Falg denote the algebra of polynomials gener-
ated by φ̂[f ]. Then Falg Ω is dense in H.
(5) Hermiticity: For any Φ, Ψ ∈ D,

(Φ|φ̂[f ]Ψ) = (φ̂[f ]Φ|Ψ).

In what follows a map (2.10) satisfying Axiom (1) will be called an operator
valued distribution. By saying that it is cyclic we will mean that it satisfies
Axiom (4).
Note that free bosons satisfy both the Haag-Kastler and Wightman axioms.
In particular, the joint spectrum of the Hamiltonian and momentum is contained
in the forward light cone:

sp(P ) = {0} ∪ {p ∈ R1,3 | p2 = −m2 or p2 ≤ −4m2 ; p0 ≥ 0}. (2.12)

This follows from the following inequality on the relativistic energy:

p q q
p~ + m + ~k 2 + m2 ≥ (~
2 2 p + ~k)2 + 4m2 , (2.13)

which is saturated for p~ = ~k.

It is easy to extend the Haag-Kastler and Wightman axioms to include
fermions. The Poincaré group R1,3 o SO0 (1, 3) has to be replaced by its 2-fold
covering R1,3 o Spin0 (1, 3), and (in the case of Wightman axioms) we should
allow for anticommutation between fields.

18
2.10 Relationship between Haag-Kastler and Wightman
axioms
“Morally”, Wightman axioms are stronger than the Haag-Kastler axioms. In
fact, let Aalg (O) be the algebra of polynomials in φ̂[f ] with suppf ⊂ O, which
can be treated as a ∗-subalgebra of L(D). Then the family O 7→ Aalg (O) is
almost a net of field algebras. Unfortunately, elements of Aalg (O) are defined
only on D and not on the whole H, and often do not extend to bounded operators
on H.
We know that the fields φ̂[f ] are Hermitian (symmetric) on D. Suppose
they are essentially self-adjoint. Then their closures are self-adjoint operators
on H. We could consider the von Neumann algebra A(O) generated by bounded
functions of φ̂[f ], suppf ⊂ O. Then there is still no guarantee that the net
O 7→ A0 (O) satisfies the Haag-Kastler axioms: we are not sure whether the
Einstein causality holds.
To understand this, we recall that there are serious problems with commu-
tation of unbounded operators [Reed-Simon-I]. One says that two self-adjoint
operators commute (or strongly commute) if all their spectral projections com-
mute. There exist however examples of pairs of two self-adjoint operators A, B
and a subspace D ⊂ DomA ∩ DomB with the following property:
(1) A and B preserve D and are essentially self-adjoint on D.
(2) A and B commute on D.
(3) A and B do not commute strongly.
(4) D is dense.
One of the most important topics in QFT is that of gauge invariance. In the
older literature one distinguishes between gauge invariance of the first kind–wrt
a global symmetry–and of the second kind–wrt a local symmetry. In modern
physics literature the first meaning seems to have disappeared, although it is
still used in some parts of mathematical literature. Thus in the modern physics
usage by gauge invariance one one means local gauge invariance.
Global symmetries are well understood in the framework of Haag-Kastler
axioms, thanks to the work of Doplicher-Haag-Roberts. Unfortunately, to my
understanding, we do not know how to accommodate (local) gauge invariance
in axioms of QFT.
The Haag-Kastler axioms are so abstract, general and have so little structure
that we do not know how to see the gauge invariance. The Wightman axioms do
not apply to gauge fields, because apparently for them one needs an indefinite
product Hilbert space or nonlocal fields like Wilson loops. I am not aware of a
successful adaptation of Wightman axioms that accommodates gauge fields.

19
3 The Laplace and Helmholtz equation
3.1 Tempered distributions
The space of Schwartz functions on Rn is defined as
S(Rn ) := Ψ ∈ C ∞ (Rn ) : |xα ∇βx Ψ(x)|2 dx < ∞,

α, β ∈ Nn .
R
(3.1)
Remark 3.1. (3.1) is equivalent to the definition
S(Rn ) = Ψ ∈ C ∞ (Rn ) : |xα ∇βx Ψ(x)| ≤ cα,β ,

α, β ∈ Nn . (3.2)
more common in the literature.
S 0 (Rn ) denotes the space of continuous linear functionals on S(Rn ). This
means. a linear functional S(Rn ) 3 Ψ 7→ hT |Ψi ∈ C belongs to S 0 iff there
exists N such that
 X Z  21
|hT |Ψi| ≤ C |xα ∇βx Ψ(x)|2 dx .
|α|+|β|<N

If f is a function which is locally integrable (L1 on all bounded intervals),

and satisfies some mild growth conditions, then it defines a distribution in S 0
by the formula Z
hTf |Ψi = f (x)Ψ(x)dx. (3.3)

Motivated by (3.3), we will often use the integral notation

Z
hT |Ψi = T (x)Ψ(x)dx

also for distributions that are not given by such integrals

Here are some examples of elements of S 0 (R):
Z
δ(t)Φ(t)dt := Φ(0), (3.4)
Z Z  Z +∞ !
Φ(x) Φ(x)
P dx := lim + dx, (3.5)
x &0 −∞  x
Z Z
(t ± i0) Φ(t)dt := lim (t ± i)λ Φ(t)dt.
λ
(3.6)
&0

(3.6) is simply given by the locally integrable function tλ , however for λ < −1
it is not.
Here some examples of functions that do not correspond to elements of S 0 (R):
t 1 1
e , |t| , t . The first blows up at infinity too fast. The last two can be regularized
to make a distribution in S 0 (R). Note the Sochocki formula and its consequence:
1 1
= P ± iπδ(t), (3.7)
t ∓ i0 t
1 1
− = 2πiδ(t). (3.8)
t − i0 t + i0
Note that for λ > −1

20
3.2 Fourier transformation
The definition of the Fourier transform of Rd 3 ~x 7→ f (~x) on a Euclidean space
will be standard: Z
~
fˆ(~k) := e−ik·~x f (~x)d~x.

It is also common normalize the Fourier transformation as follows

1
Ff (k) = p fˆ(k).
(2π)d

F is unitary on L2 (Rd ).
Often, we will drop the hat – the name of the variable will indicate whether
we use the position or momentum representation:
Z Z
~ 1 ~
f (~k) = e−ik·~x f (~x)d~x, f (~x) = eik·~x f (~k)d~k.
(2π)d

On a Minkowski space, for the time variable (typically t) we reverse the sign
in the Fourier transform:
Z Z
1
f (ε) = eiεt f (t)dt, f (t) = e−iεt f (ε)dε.
2π

The Fourier transformation is a continuous map from S 0 into itself. We have

continuous inclusions
S(Rn ) ⊂ L2 (Rn ) ⊂ S 0 (Rn ).

3.3 Green’s functions on the Euclidean space

Let Rd be the Euclidean space and ∆ the Laplacian. The following equations
are invariant wrt Rd o O(d):

the Laplace equation −∆ζ =0, (3.9)

2
the Helmholtz equation (−∆ ± m )ζ =0. (3.10)

We also have their inhomogeneous versions:

the Poisson equation −∆ζ =f, (3.11)

2
the inhomogeneous Helmholtz equation (−∆ ± m )ζ =f. (3.12)

We will say that G is Green’s function for −∆ ± m2 if

(−∆ ± m2 )G(x) = δ(x). (3.13)

If we are given f and G is Green’s function, then

Z
ζ(x) = G(x − y)f (y)dy (3.14)

21
solves (3.12)
Assume that ζ, f ∈ S 0 (Rd ) and apply the Fourier transformation
Z Z
ζ(x) = (2π)−d eipx ζ(p)dp, f (x) = (2π)−d eipx f (p)dp (3.15)

to (3.12). Then

(p2 + m2 )ζ(p) = f (p), ζ(p) = (p2 + m2 )−1 f (p) (3.16)

[
Using G ? f = Ĝfˆ we obtain

eixp
Z Z
ζ(x) = Gd,m (x − y)f (y)dy, Gd,m (x) = (2π)−d dp. (3.17)
(p2 + m2 )

For m > 0 or m = 0 and d 6= 1, 2, the function (p2 + m2 )−1 is locally integrable

and belongs to S 0 (Rd ). Hence its Fourier transform is well defined and is the
unique Green’s function of −∆+m2 (at least if we are interested only in solutions
in S 0 (Rd ), which is usually the case).
The case m > 0 can be reduced to m = 1:
eixp
Z
d−2 −d
Gd,m (x) = m Gd (mx), Gd (x) = (2π) dp. (3.18)
(p2 + 1)

The Euclidean invariance suggests to look for Green’s functions that depend
only on r = |x|, so that one can write G(x) = G(r). Such Green’s functions
away from r = 0 satisfy the radial part of the Helmholtz equation:
 d−1 
− ∂r2 − ∂r ± m2 G(r) = 0 (3.19)
r
The massless Green’s function can be expressed in terms of elementary func-
tions.
Γ( d
2 −1) 2−d
Theorem 3.2. For d ≥ 3, Gd,0 (r) = d r .
4π 2
1 1
E.g. G3,0 (r) = 4πr , G4,0 (r) = 4π 2 r 2 .

Proof. It is elementary to check that G(r) = cd r2−d satisfies the Laplace

equation away from the origin. Clearly, it belongs to S 0 (Rd ). We need to find
cd . We take a radial test function Ψ(r). We need to check that

−∆Gd,0 (x) = δ(x), (3.20)

Z
or ∆G(x)Φ(x)dx = −Φ(0). (3.21)

22
But the lhs of (3.21) equals
Z ∞
d−1 
Z 
G(x)∆Φ(x)dx = G(r)rd−1 |Sd−1 | ∂r2 + ∂r Φ(r)dr (3.22)
0 r
Z ∞  d − 1
= cd |Sd−1 | Φ(r) ∂r2 − ∂r rdr (3.23)
0 r
!
0

∞
+ rΦ (0) − Φ(0) + (d − 1)Φ(0) (3.24)
0

= cd |Sd−1 |(d − 2)Φ(0), (3.25)

d
2π
where |Sd−1 | = Γ( d
is the surface of the d − 1-dimensional sphere.
2)
Γ( d −1)
In the massive case, For d ≥ 3, Gd,m (r) ∼ 2 d r2−d . near zero, but for
4π 2
large r we have exponential decay, more precisely
1 1−d
Gd,m (r) ∼ d−1 r 2 e−mr . (3.26)
2(2π) 2

This will follow from the analysis below.

3.4 Bessel equations

There are two basic forms of the Bessel equation:
 1 µ2 
the modified Bessel equation ∂r2 + ∂r − 2 − 1 v = 0, (3.27)
r r
2
 1 µ 
the (standard) Bessel equation ∂r2 + ∂r − 2 + 1 v = 0. (3.28)
r r
One can pass from one to the other by substituting ±ir for r.
We have
d
 d − 1  − d +1 1  d 2 1
r−1+ 2 ∂r2 + ∂r r 2 = ∂r2 + ∂r − 1 − , (3.29)
r r 2 r2
Set µ = d2 − 1 and m = 1. We see that if F satisfies satisfies the modified
d
Bessel equation, then r1− 2 F (r) satisfies (3.19) for +m2 , and if F satisfies the
d
standard Bessel equation is, then r1− 2 F (r) satisfies (3.19) for −m2 .

3.5 Macdonald function

One of standard solutions of the modified Bessel equation is the Macdonald
function:
1 ∞
Z  r 
Kµ (r) := exp − (s + s−1 ) s±µ−1 ds. (3.30)
2 0 2
The integral (3.30) is absolutely convergent. Substitution s = t−1 shows that µ
can be replaced by −µ (and thus Kµ = K−µ ).

23
Theorem 3.3. For | arg z| < π − ,

Kµ (z)
lim
−z
√ = 1.
|z|→∞ e √ π
2z

Proof. We use the steepest descent method. Set φ(t) := − 12 (t + t−1 ). We

compute
1
φ0 (t) = − (1 − t−2 ), φ00 (t) = −t−3 .
2
Hence φ has a critical point at t0 = 1 with φ(t0 ) = −1 and φ00 (t0 ) = −1. Thus

1 ∞ −µ−1
Z
Kµ (z) = t exp(zφ(t))dt
2 0
1 ∞ φ00 (t0 )
Z  
' exp zφ(t0 ) + z (t − t0 )2 dt
2 −∞ 2
Z ∞ √
1 −z  z 2
 1 −z 2π
= e exp (t − 1) dt = e √ .
2 −∞ 2 2 z
2

Theorem 3.4. For µ > 0,

1  r −µ
lim mµ Kµ (mr) = Γ(µ) . (3.31)
m&0 2 2
2t
Proof. We set s = mr :

mµ ∞
Z  mr 
µ
m Kµ (mr) = exp − (s + s−1 ) sµ−1 ds (3.32)
2 0 2
Z ∞
m2 r2 µ−1
 
1 2
  µ
= exp −t − t dt (3.33)
2 r 0 4t
1 r −µ
 
→ Γ(µ). (3.34)
2 2
2
For half-integer µ the Macdonald function can be expressed in terms of
elementary functions, e.g.
 π  12
K± 12 (r) = e−r . (3.35)
2r
Theorem 3.5.
1 d
Gd (r) = d r1− 2 K d −1 (r). (3.36)
(2π) 2 2

e−r e−r
E.g. G1 (r) = 2 , G3 (r) = 4πr .

24
 Proof. We will use the following identities:
Z ∞
1
= e−sA ds, (3.37)
A 0
Z
sp2
 2π  d2 x2
dpe− 2 eipx = e− 2s . (3.38)
s
Now
eipx dp
Z
(2π)−d
(1 + p2 )
|x| ∞
Z Z
2 |x|s
= (2π)−d ds dpe−(1+p ) 2 eipx
2 0
 |x| 1− d2 d Z ∞ d 1 |x|
= (2π)−d π2 dss− 2 e−(s+ s ) 2
2 0
d
d
 |x|  1− 2
= (2π)−d 2π 2 K1− d (|x|).
2 2

2
Using
Gd,0 (x) = lim md−2 Gd (mx), (3.39)
m&0

Γ( d
2 −1) 2−d
we obtain an alternative proof of the zero mass formulas Gd,0 (r) = d r ,
4π 2
d ≥ 3.

3.6 Bessel and Hankel functions

We will also need the following solutions of the standard Bessel equation:
2 ∓i π (µ+1)
the Hankel functions: Hµ(1)/(2) (r) = Hµ± (r) := e 2 Kµ (∓ir), (3.40)
π
1
Jµ (r) := Hµ+ (r) + Hµ− (r)


the Bessel function: (3.41)
2
For large r > 0 the Hankel functions are oscillating:
Hµ± (r)
lim
12 iµπ
= 1,
r→∞ 2 iπ
πr e±ir e∓ 2 ∓ 4

The Bessel function can be expanded in a power series near zero:

∞
2n+µ
X (−1)n 2r
Jµ (r) = .
n=0
n!Γ(µ + n + 1)

Let’s go back to the Helmholtz equation with the negative sign at m2 :

(−∆ − m2 )ζ = f. (3.42)

25
 The Green’s function Gd,m is well defined not only for m ≥ 0, but also for
Re(m) > 0, which guarantees m2 ∈ C\] − ∞, 0]. Taking the limit at imaginary
line, that is setting Gd,±im , we obtain two Green’s functions of (3.42):

e−ixp
Z
Gd,∓im (r) = (2π)−d dp (3.43)
(p2 − m2 ∓ i0)
i  m  d2 −1 ±
=± H d −1 (mr). (3.44)
4 2πr 2

Thus in the case −m2 we have many Green’s functions that belong to S 0 .
We also have many solutions in S 0 of the (homogeneous) Helmholtz equation,
e.g.

1  m  d2 −1
i Gd,im (r) − Gd,−im (r) = J d −1 (mr). (3.45)
2 2πr 2

4 Wave and Klein-Gordon equations

4.1 Propagators
Let d = 1 + n. Let 2 be the d’Alembertian on the Minkowski space R1,n :
n
X
2 := −∂02 + ∂i2 . (4.1)
i=1

The following equations are invariant wrt R1+n o O(1, n):

the wave equation −2ζ =0, (4.2)

2
the Klein-Gordon equation (−2 + m )ζ =0. (4.3)

We are interested also in their inhomogeneous versions:

the inhomogeneous wave equation −2ζ =f, (4.4)

2
the inhomogeneous Klein-Gordon equation (−2 + m )ζ =f. (4.5)

We will say that G• is Green’s function of the Klein-Gordon equation if

(−2 + m2 )G• (x) = δ(x). (4.6)

Thus, for any f , Z

ζ(x) = G• (x − y)f (y)dy (4.7)

satisfies (4.5).
We say that G• is a solution the Klein-Gordon equation if

(−2 + m2 )G• (x) = 0. (4.8)

26
The ansatz analogous to (3.17)
eixp
Z
• −d
G (x) = (2π) dp (4.9)
(p2 + m2 )
is incomplete and needs to be precised to make it well defined.
Every bisolution of G• can be written as
Z
dp
G• (x) = eipx g(p)δ(p2 + m2 )
(2π)3
where g is a function on the two-sheeted hyperboloid p2 + m2 = 0 (see below
for the meaning of δ(p2 + m2 )).

4.2 Invariant measure

Let f be a differentiable function on R. The following fact (under appropriate
assumptions) is easy:
Lemma 4.1. Let δ be an approximate delta function, that is
Z
lim δ (t)φ(t)dt = φ(0).
&0

Then Z X φ(si )
lim δ (f (s))φ(s)ds = . (4.10)
&0 |f 0 (si )|
f (si )=0

This suggests the following notation:

Z X φ(si )
δ(f (s))φ(s)ds = . (4.11)
|f 0 (si )|
f (si )=0

Now consider R1,n and apply (4.11) to s = p0 for fixed p~ and

f (p) = p2 + m2 = −(p0 )2 + p~2 + m2 .
We have
d(p2 + m2 )
= 2p0 ,
dp0
p
and p2 + m2 = 0 iff p0 = ± p~2 + m2 . Hence we can write
 p   p 
δ p0 − p~2 + m2 δ p0 + p~2 + m2
δ(p2 + m2 )dp = p d~
p+ p d~
p. (4.12)
2 p~2 + m2 2 p~2 + m2
(4.12) is a measure on R1,3 invariant wrt Lorentz transformations. In fact,
 1 1 
2πiδ(p2 + m2 ) = lim 2 − , (4.13)
&0 p + m2 − i p2 + m2 + i
where the rhs is obviously Lorentz invariant.

27
4.3 Propagators for the Klein-Gordon and wave equation
Introduce

• the forward/backward or retarded/advanced propagator

eix·p
Z
1
G∨/∧ (x) := d
dp, (4.14a)
(2π) p + m ∓ i0sgn(p0 )
2 2

• the Feynman/anti-Feynman(-Stueckelberg) propagator

eix·p
Z
1
GF/F (x) := d
dp, (4.14b)
(2π) p + m2 ∓ i0
2

• the Pauli–Jordan propagator or the commutator function

Z
i
GPJ (x) := eix·p sgn(p0 )δ(p2 + m2 )dp (4.14c)
(2π)d
Z
1 d~
p i~
xp~
 p
0 2 + m2

= e sin x p
~ (4.14d)
(2π)d−1
p
p~2 + m2

• the positive/negative frequency, or particle/antiparticle, or Wightman-anti-

Wightman solution (two-point function)
Z
1
G(±) (x) := eix·p θ(±p0 )δ(p2 + m2 )dp (4.14e)
(2π)d
1
Z
d~
p √ 2 2
∓ix0 p
~ +m +i~
xp~
= e . (4.14f)
(2π)d−1
p
2 p~2 + m2

(4.14a), (4.14b) are distinguished Green’s functions (inverses) and (4.14c), (4.14e)
are distinguished solutions of the Klein–Gordon equation −2 + m2 . We will call
them jointly “propagators”.
Note the identities satisfied by the propagators:

G∨ − G∧ = GPJ (4.15a)
(+) (−)
= iG − iG , (4.15b)
GF − GF = iG(+) + iG(−) , (4.15c)
F F ∨ ∧
G +G =G +G , (4.15d)
GF = iG(+) + G∧ = iG(−) + G∨ , (4.15e)
F (+) ∨ (−) ∧
G = −iG + G = −iG +G . (4.15f)

To prove these identities we use repeatedly

 1 1 
θ(±p0 )2πiδ(p2 + m2 ) = θ(±p0 ) 2 − , (4.16)
p + m2 − i0 p2 + m2 + i0

28
4.4 Einstein causality of propagators
Proposition 4.2. We have suppG∨/∧ ⊂ J ∨/∧ and suppGPJ ⊂ J.
Proof. Let us prove that suppG∨ ⊂ J ∨ . By the Lorentz invariance it
suffices to prove that G∨ is zero on the lower half-plane. We write

eipx
Z
dp
G∨ (x) =
(p + m − i0sgnp ) (2π)4
2 2 0
0 0
e−ip x +i~p~x dp0 d~
Z
p
=
.
p~2 + m2 − (p0 + i0)2 (2π)4

Next we continuously deform the contour of integration, replacing p0 by p0 +iR,

where R ∈ [0, ∞[. We do not cross any singularities of the integrand and note
0 0
that e−ix (p +iR) goes to zero (remember that x0 < 0).
Analogously one proves suppG∧ ⊂ J ∧ . By (4.15a) we obtain suppGPJ ⊂ J.
2
Note that
(
(+) −iGF (x) on R1,3 \J ∧ ,
G (x) = (4.17)
iGF (x) on R1,3 \J ∨ ;
(
−iGF (x) on R1,3 \J ∨ ,
G(+) (x) = G(−) (x) = (4.18)
iGF (x) on R1,3 \J ∧ .

Hence

−iGF (x − y) =θ(x0 − y 0 )G(+) (x − y) + θ(y 0 − x0 )G(−) (x − y), (4.19)

F 0 0 (−) 0 0 (+)
iG (x − y) =θ(x − y )G (x − y) + θ(y − x )G (x − y). (4.20)

4.5 Propagators in position representation in dimension

1 + 3.
Below we give the formulas for the massive propagators in the position repre-
sentation in dimension 1 + 3.

• The forward/backward or retarded/advanced propagator:

1 mθ(−x2 )θ(±x0 ) p
G∨/∧ (x) = θ(±x0 )δ(x2 ) − √ J1 (m −x2 ).
2π 4π −x2

• The (anti-)Feynman(-Stueckelberg) Green’s function:

1 mθ(−x2 ) ∓ p 2
GF/F (x) = δ(x2 ) − √ H1 (m −x )
4π 8π −x2
miθ(x2 ) √
± √ K1 (m x2 ). (4.21)
4π 2 x2

29
 • The Pauli-Jordan
1 msgnx0 θ(−x2 ) p
GPJ (x) = sgnx0 δ(x2 ) − √ J1 (m −x2 ).
2π 4π −x2

• The positive/negative frequency solution:

i
G(±) (x) = ∓ sgnx0 δ(x2 )
4π
imθ(−x2 ) ∓sgnx0 p 2 mθ(x2 ) √
− √ H1 (m −x ) + √ K1 (m x2 ).
8π −x 2 2
4π x 2

The above formulas are somewhat sloppily written in the neighborhood of

the surface of the light cone, where they describe irregular distribution.
To obtain these formulas, first we find the Euclidean Green’s function in 4
dimensions:
m
GE (x) = K1 (m|x|). (4.22)
4π 2 |x|
Using the formula for the function K1 we can write
∞
1 m2 X  |x|m Hk + Hk+1  1  m2 |x|2 k
GE (x) = + ln + γ +
4π 2 |x|2 8π 2 2 2 k!(k + 1)! 4
k=0
1 |x|m
= 2 2
+ m2 ln u(m2 |x|2 ) + m2 v(m2 |x|2 ), (4.23)
4π |x| 2

where u, v are analytic functions and Hk = 11 + · · · + k1 . Then we apply the

Wick rotation. This means that we replace x4 with ±ix0 in the argument. We
also have to replace the measure of integration dx1 · · · dx4 with ±idx0 dx1 · · · .
This leads to the Feynman/anti-Feynman propagator

GF/F (x0 , ~x) = ±iGE (±ix0 , ~x). (4.24)

We obtain (4.21). This is obvious for spacelike x0 , ~x, where the Wick rotation
does not change the sign of x2 and therefore we still obtain an expression in-
volving the Macdonald function. Inside the light cones the sign of x2 changes.
More precisely, we can interpret the Wick rotation as e±iφ x0 with φ ∈ [0, π2 ]
and we obtain expressions involving Hankel functions. For φ ∈]0, π2 [ we have
±iφ 0 2


±Im (e x ) + (~x) > 0. Therefore, the Eclidean square |x|2 has to be re-
2

placed with the Lorentzian square x2 ± i0. On the surface of the light cone the
terms involving u, v are sufficiently regular and cause no problem. However the
function |x|1 2 becomes

1 1 1 iπ
= 2 ∓ iπδ(x2 ) = 2 ∓ δ(x0 + |~x|) + δ(x0 − |~x|) ,


(4.25)
x2 ± i0 x x 2|~x|
1
where we used the Sochocki formula and x2 is meant in terms of the principal
value.

30
 Then we compute

G∨ (x) + G∧ (x) = GF (x) + GF (x) (4.26)

2
1 mθ(−x ) p
= δ(x2 ) − √ J1 (m −x2 ). (4.27)
2π 4π −x2
Now we can obtain the forward/backward propagators from

G∨/∧ = θ(±x0 ) GF (x) + GF (x) .

(4.28)

Finally, we can read off the formulas for positive/negative frequency solutions
from (4.17).
Below we give the formulas for the massless propagators in the position
representation in dimension 1 + 3.

• The forward/backward propagator:

1 δ(x0 ∓ |~x|)
G∨/∧ (x) = θ(±x0 )δ(x2 ) = .
2π 4π|~x|

• The (anti-)Feynman propagator:

±i
GF/F (x) = .
4π 2 (x2 ± i0)

• The Pauli-Jordan or the commutator function:

1 δ(x0 − |~x|) δ(x0 + |~x|)
GPJ (x) = sgnx0 δ(x2 ) = − .
2π 4π|~x| 4π|~x|

• The positive frequency, resp. negative frequency:

1
G(±) (x) = .
4π(x2 ± i0sgnx0 )

4.6 Classical propagators and solving the Klein-Gordon

equation
A function on R1,n is called space compact if there exists a compact K ⊂ R1,n
such that suppf ⊂ J(K). It is called future/past space-compact if there exists a
compact K ⊂ R1,n such that suppf ⊂ J ± (K).
∞
The set of space compact smooth functions will be denoted Csc (R1,n ).
Proposition 4.3. Let f ∈ Cc∞ (R1,n ).
(1) ζ ∨/∧ (x) := G∨/∧ (x − y)f (y)dy is the unique solution of
R

(−2 + m2 )ζ =f. (4.29)

future/past space compact.

31
 GPJ (x − y)f (y)dy is a solution of
R
(2) ζ(x) :=

(−2 + m2 )ζ =0. (4.30)

Every smooth space-compact solution of (4.30) is of this form.

GPJ (x) is the unique solution of the Klein-Gordon equation satisfying

GPJ (0, ~x) = 0, ĠPJ (0, ~x) = δ(~x). (4.31)

Proposition 4.4. Let α, β ∈ Cc∞ (Rn ). Then there exists a unique ζ ∈ Csc
∞
(R1,n )
that solves
(−2 + m2 )ζ = 0 (4.32)
with initial conditions ζ(0, ~x) = α(~x), ζ̇(0, ~x) = β(~x). It satisfies suppζ ⊂
J(suppα ∪ suppβ) and is given by
Z Z
ζ(t, ~x) = ĠPJ (t, ~x − ~y )α(~y )d~y + GPJ (t, ~x − ~y )β(~y )d~y . (4.33)
Rn Rn

Proof. Clearly, (4.33) satisfies (4.32). Using (4.31) we check that ζ(0, ~x) =
α(~x). We have
G̈PJ (0, ~x) = (∆ − m2 )GPJ (0, ~x) = 0. (4.34)
Now we can verify that ζ̇(0, ~x) = β(~x). 2

5 Second quantization
In this chapter we describe the terminology and notation of multilinear algebra.
We will concentrate on the infinite dimensional case, where it is often natural to
use the structure of Hilbert spaces. We will introduce Fock spaces and various
classes of operators acting on them. In quantum physics the passage from a
dynamics on one-particle spaces to a dynamics on Fock spaces is often called
second quantization – hence the name of the chapter.
We will consider two setups: that of vector spaces and that of Hilbert spaces.
If X , Y are vector spaces, then L(X , Y) will denote the set of linear operators
from X to Y. If X , Y are Hilbert spaces, then B(X , Y) will denote the set of
bounded operators fro X to Y.

5.1 Vector and Hilbert spaces

Let V be a vector space. A set {ei : i ∈ I} ⊂ V is called linearly independent
if for any finite subset {ei1 , . . . , ein } ⊂ {ei : i ∈ I}

c1 ei1 + · · · + cn ein = 0 ⇒ c1 = · · · = cn = 0. (5.1)

{ei : i ∈ I} is a Hamel basis (or simply a basis) of V if it is a maximal linearly

independent set. It means that it is linearly independent and if we add any

32
v ∈ V to {ei : i ∈ I} ⊂ V then it is not linearly independent anyP more. Note
that every v ∈ V can be written as a finite linear combination v = i∈I λi ei in
a unique way.
Let V be a vector space over C or R equipped with a scalar product (v|w)
(positive, nondegenerate, sesquilinear form). It defines a metric on V by
p
kv − wk := (v − w|v − w). (5.2)

We say that V, (·|·) is a Hilbert space if V is complete.

If V, (·|·) is not necessarily complete, then we can always complete it, that
is find a larger complete space V cpl , (·|·) in which V is embedded as a dense
subspace. V cpl is uniquely defined and is called the completion of V.
For instance, if we take Cc (R), Cc∞ (R) or S(R) with the usual scalar product
(f |g) = f (x)g(x)dx, then its completion is L2 (R).
R

If V is a Hilbert space, then {ei : i ∈ I} is called an orthonormal basis

(o.n.b.) if it is a maximal orthonormal Pthat every v ∈ V can be written
set. Note
as a linear combination v = i∈I λi ei , where i∈I |λi |2 < ∞, in a unique way
P
Note that in a finite dimensional Hilbert space every orthonormal basis is a
basis. This is not true in infinite dimensional Hilbert spaces.

5.2 Direct sum

Let (Vi )i∈I be a family of vector spaces. The algebraic direct sum of Vi will be
denoted
al
⊕ Vi , (5.3)
i∈I

It consists of sequences (vi )i∈I , which are zero for all but a finite number of
elements.
al
If (Vi )i∈I is a family of Hilbert spaces, then ⊕ Vi has a natural scalar prod-
i∈I
uct.   X
(yi )i∈I (vi )i∈I = (yi |vi ). (5.4)
i∈I

The direct sum of Vi in the sense of Hilbert spaces is defined as

 cpl
al
⊕ Vi := ⊕ Vi .
i∈I i∈I

al
If I is finite, then ⊕ Vi = ⊕ Vi
i∈I i∈I
Let (Vi ), (Wi ), i ∈ I, be families of vector spaces. If 
ai ∈ L(Vi , Wi ), i ∈ I,
al al
then their direct sum is denoted ⊕ ai and belongs to L ⊕ Vi , ⊕ Wi . It is
i∈I i∈I i∈I
defined as  
⊕ ai (vi )i∈I = (ai vi )i∈I (5.5)
i∈I

33
 Let Vi , Wi , i ∈ I be families of Hilbert spaces, and ai ∈ B(Vi , W
i ) with 
supi∈I kai k < ∞. Then the operator ⊕ ai is bounded. Its extension in B ⊕ Vi , ⊕ Wi
i∈I i∈I i∈I
will be denoted by the same symbol.

5.3 Tensor product

Let V, W be vector spaces. The algebraic tensor product of V and W will be
al
denoted V ⊗ W. Here is one of its definitions
Let Z be the space of finite linear combinations of vectors (v, w), v ∈ V,
w ∈ W. In Z we define the subspace Z0 spanned by

(λv, w) − λ(v, w), (v, λw) − λ(v, w),

(v1 + v2 , w) − (v1 , w) − (v2 , w), (v, w1 + w2 ) − (v, w1 ) − (v, w2 ).
al
We set V ⊗ W := Z/Z0 . If v ∈ V, w ∈ W, we define v ⊗ w := (v, w) + Z0 .
Remark 5.1. Note that (v, w) above is just a symbol and not an element of
V ⊕ W. Elements of the space Z have the form
n
X
λn (vn , wn ). (5.6)
j=1

In particular, in general

(v1 , w1 ) + (v2 , w2 ) 6∼ (v1 + v2 , w1 + w2 ), (5.7)

λ(v, w) 6∼ (λv, λw). (5.8)
al
V ⊗ W is a vector space and ⊗ is an operation satisfying

(λv) ⊗ w = λv ⊗ w, v ⊗ (λw) = λv ⊗ w,
(v1 + v2 ) ⊗ w = v1 ⊗ w + v2 ⊗ w, v ⊗ (w1 + w2 ) = v ⊗ w1 + v ⊗ w2 .

Vectors of the form v ⊗ w are called simple tensors. Not all elements of V ⊗ W
al
are simple tensors, but they span V ⊗ W.
If {ei }i∈I and {fj }j∈J are bases of V, resp. W, then {ei ⊗ fj }(i,j)∈I×J is a
al
basis of V ⊗ W,
al
If V, W are Hilbert spaces, then V ⊗ W has a unique scalar product such
that

(v1 ⊗ w1 |v2 ⊗ w2 ) := (v1 |v2 )(w1 |w2 ), v1 , v2 ∈ V, w1 , w2 ∈ W.

To see this it is enough to choose o.n.b’s {ei }i∈I and {fj }j∈J in V, resp. W.
al
Then every element of V ⊗ W can be written as an (infinite) linear combination
of ei ⊗fj and we can use them as an orthonormal set defining this scalar product.
We set
al
V ⊗ W := (V ⊗ W)cpl ,

34
and call it the tensor product of V and W in the sense of Hilbert spaces. If
{ei }i∈I and {fj }j∈J are o.n.b’s of V, resp. W, then {ei ⊗ fj }(i,j)∈I×J is an
o.n.b. of V ⊗ W,
al
If one of the spaces V or W is finite dimensional, then V ⊗ W = V ⊗ W.
Let V1 , V2 , W1 , W2 be vector spaces. If a ∈ L(V1 , V2 ) and b ∈ L(W1 , W2 ),
al al
then there exists a unique operator a ⊗ b ∈ L(V1 ⊗ W1 , V2 ⊗ W2 ) such that on
simple tensors we have

(a ⊗ b)(y ⊗ w) = (ay) ⊗ (bw). (5.9)

To see this it is enough to choose bases (ei )i∈I in V1 and (fj )j∈J in W1 and to
define a ⊗ b on the basis (ei ⊗ fj )(i,j)∈I×J by

(a ⊗ b)ei ⊗ fj := (aei ) ⊗ (bfj ). (5.10)

Then we check that thus defined operator satisfies (5.9) and is unique. It is
called the tensor product of a and b.
If V1 , V2 , W1 , W2 are Hilbert spaces and a ∈ B(V1 , V2 ), b ∈ B(W1 , W2 ), then
a ⊗ b is bounded. It extends uniquely to an operator in B(V1 ⊗ W1 , V2 ⊗ W2 ),
denoted by the same symbol.
To prove the boundedness of a ⊗ b = a ⊗ 1l 1l ⊗ b, it is sufficient to consider
al al
the operator a ⊗ 1l from V1 ⊗ W to V2 ⊗ W. Let e1P , e2 , . . . and f1 , f2 . . . be
orthonormal bases in V1 , W resp. Consider a vector cij ei ⊗ fj .
X 2 X X 2
a ⊗ 1l cij ei ⊗ fj = cij aei
i j i
X X 2 X X
≤ kak2 cij ei ≤ kak2 |cij |2
j i j i
X 2
= kak2 cij ei ⊗ fj .
ij

5.4 Fock spaces

Let Y be a vector space. Let Sn denote the permutation group of n elements
al n
and σ ∈ Sn . Θ(σ) is defined as the unique operator in L(⊗ Y) such that

Θ(σ)y1 ⊗ · · · ⊗ yn = yσ−1 (1) ⊗ · · · ⊗ yσ−1 (n) . (5.11)

To see that Θ(σ) is well defined we first choose a basis {ei }i∈I of Y. Then
al n
we define Θ(σ) on the corresponding basis of ⊗ Y:

Θ(σ)ei1 ⊗ · · · ⊗ ein = eiσ−1 (1) ⊗ · · · ⊗ eiσ−1 (n) .

al n
Then we extend by linearity Θ(σ) to the whole ⊗ Y. It is easy to see that the
operator defined in this way satisfies (5.11).

35
 We can check that
al n
Sn 3 σ 7→ Θ(σ) ∈ L(⊗ Y) (5.12)

is a group representation.
al n
We say that a tensor Ψ ∈ ⊗ Y is symmetric, resp. antisymmetric if

Θ(σ)Ψ = Ψ, (5.13)
resp. Θ(σ)Ψ = sgn(σ)Ψ. (5.14)

We define the symmetrization/antisymmetrization projections

1 X 1 X
Θns := Θ(σ), Θna := sgnσΘ(σ).
n! n!
σ∈Sn σ∈Sn

They project onto symmetric/antisymmetric tensors.

We will often write s/a to denote either s or a.
If Y is a Hilbert space, then Θ(σ) is unitary and Θns/a are orthogonal pro-
jections.
Let Y be a vector space. The algebraic n-particle bosonic/fermionic space is
defined as
al n al n
⊗s/a Y := Θns/a⊗ Y.
The algebraic bosonic/fermionic Fock space or the symmetric/antisymmetric
tensor algebra is
∞
al al al n
Γs/a (Y) := ⊕ ⊗s/a Y.
n=0

The vacuum vector is Ω := 1 ∈ ⊗0s/a Y = C.

If Y is a Hilbert space, then the n-particle bosonic/fermionic space is defined
as
⊗ns/a Y := Θns/a ⊗n Y.
The bosonic/fermionic Fock space is
∞
Γs/a (Y) := ⊕ ⊗ns/a Y.
n=0

5.5 Creation/annihilation operators

For z ∈ Y we define the creation operator
√
â∗ (z)Ψ := Θn+1
s/a n + 1z ⊗ Ψ, Ψ ∈ ⊗ns/a Y,
∗
and the annihilation operator â(z) := (â∗ (z)) . (We often omit the hat).
We will sometimes write (z| and |z) for the following operators

V 3 v 7→ (z|v := (z|v) ∈ C, (5.15)

C 3 λ 7→ λ|z) := λz ∈ V. (5.16)

36
Then on ⊗ns/a Y we have
√
a∗ (z) = Θn+1
s/a n + 1|z) ⊗ 1ln⊗ , (5.17)
√
a(z) = n(z| ⊗ 1l(n−1)⊗ . (5.18)

Above we used the compact notation for creation/annihilation operators pop-

ular among mathematicians. Physicists commonly prefer the traditional nota-
tion, which is longer and less canonical.
One version of the traditional notation Puses a fixed basis {ei }i∈I of Z and
set a∗i := a∗ (ei ), ai := a(ei ). Then if z = i zi ei , we have
X X
a∗ (z) = zi a∗i , a(z) = z i ai , (5.19)
i i
[ai , a∗j ]∓ = δij , [ai , aj ]∓ = 0. (5.20)

Alternatively, one often identifies Z with, say, L2 (Rd , dξ). If z equals a

function Ξ 3 ξ 7→ z(ξ), then
Z Z
a∗ (z) = z(ξ)a∗ξ dξ, a(z) = z(ξ)aξ dξ.

Note that formally

[a(ξ), a∗ (ξ 0 )]∓ = δ(ξ − ξ 0 ), [a(ξ), a(ξ 0 )]∓ = 0. (5.21)

The space ⊗ns/a Z can then be identified with the space of symmetric/antisymmetric
square integrable functions L2 (Rnd ), and then
√
a(ξ)Φ (ξ10 , . . . , ξn−1
0
) = nΦ(ξ, ξ10 , . . . , ξn−1
0


). (5.22)

5.6 Integral kernel of an operator

Every linear operator A on Cn can be represented by a matrix [Aji ].
One would like to generalize this concept to infinite dimensional spaces (say,
Hilbert spaces) and continuous variables instead of a discrete variables i, j. Sup-
pose that a given vector space is represented, say, as L2 (Rd ), or more generally,
L2 (X) where X is a certain space with a measure. One often uses the represen-
tation of an operator A in terms of its integral kernel Rd ×Rd 3 (x, y) 7→ A(x, y),
so that Z
AΨ(x) = A(x, y)Ψ(y)dy.

Note that strictly speaking A(·, ·) does not have to be a function. E.g. in the
case X = Rd it could be a distribution, hence one often says the distributional
kernel instead of the integral kernel. Sometimes A(·, ·) is ill-defined anyway. At
least formally, we have
Z
AB(x, y) = A(x, z)B(z, y)dz,

37
 A∗ (x, y) = A(y, x).
Here is a situation where there is a good mathematical theory of inte-
gral/distributional kernels:
Theorem 5.2 (The Schwartz kernel theorem). B is a continuous linear trans-
formation from S(Rd ) to S 0 (Rd ) iff there exists a distribution B(·, ·) ∈ S 0 (Rd ⊕
Rd ) such that
Z
(Ψ|BΦ) = Ψ(x)B(x, y)Φ(y)dxdy, Ψ, Φ ∈ S(Rd ).

Note that ⇐ is obvious. The distribution B(·, ·) ∈ S 0 (Rd ⊕ Rd ) is called the

distributional kernel of the transformation B. All bounded operators on L2 (Rd )
satisfy the Schwartz kernel theorem.
Examples:
(1) e−ixy is the kernel of the Fourier transformation
(2) δ(x − y) is the kernel of identity.
(3) ∂x δ(x − y) is the kernel of ∂x .

5.7 Second quantization of operators

For a contraction q on Z the operator q ⊗n commutes with Θ(σ), σ ∈ Sn .
Therefore, it preserves ⊗ns/a Z. We define the operator Γ(q) on Γs/a (Z) by

Γ(q) = q ⊗ ··· ⊗ q .
⊗n
s/a
Z ⊗n
s/a
Z

Γ(q) is called the second quantization of q.

Similarly, for an operator h on Z the operator h ⊗ 1(n−1)⊗ + · · · + 1(n−1)⊗ ⊗ h
preserves ⊗ns/a Z. We define the operator dΓ(h) by

dΓ(h) = h ⊗ 1(n−1)⊗ + · · · + 1(n−1)⊗ ⊗ h .

⊗n
s/a
Z ⊗n
s/a
Z

dΓ(h) is called the (infinitesimal) second quantization of h.

Note the identities

Γ(eith ) = eitdΓ(h) , Γ(q)Γ(r) = Γ(qr), [dΓ(h), dΓ(k)] = dΓ([h, k]),

−1 −1
Γ(q)dΓ(h)Γ(q = dΓ(qhq ). (5.23)

Let {ei | i ∈ I} be an orthonormal basis of Z. Write âi := â(ei ). Let h be

an operator on Z given by the matrix [hij ]. Then
X
dΓ(h) = hij â∗i âj . (5.24)
ij

38
Let us prove it in the bosonic case. Let Φ ∈ Γns (Z).
â∗i âj Φ = nΘns |ei ) ⊗ 1l(n−1)⊗ (ej | ⊗ 1l(n−1)⊗ Φ (5.25)
(n−1)⊗
= nΘns |ei )(ej |
⊗ 1l Φ (5.26)
1 X
= Θ(σ)|ei )(ej | ⊗ 1l(n−1)⊗ Θ(σ)−1 Φ (5.27)
(n − 1)!
σ∈Sn
n
X
= 1l(k−1)⊗ |ei )(ej | ⊗ 1l(n−k)⊗ Φ. (5.28)
k=1

More generally, if the integral kernel of an operator h is h(x, y), then

Z
dΓ(h) = h(x, y)â∗x ây dxdy. (5.29)

h(ξ)â∗ξ âξ dξ.

R
For instance, if h is the multiplication operator by h(ξ), then dΓ(h) =

5.8 Symmetric/antisymmetric tensor product

Let Ψ ∈ ⊗ps/a Z, Φ ∈ ⊗qs/a Z. We set
p+q
Ψ ⊗s/a Φ := Θs/a Ψ ⊗ Φ. (5.30)
Note that
z ⊗ · · · ⊗ z = z ⊗s · · · ⊗s z. (5.31)
n⊗
If there are n terms, it is often written as z . In the antisymmetric case one
usually prefers
(p + q)!
Ψ ∧ Φ := Ψ ⊗a Φ. (5.32)
p!q!
The operations ⊗s , ⊗a , ∧ are associative. We have
X
y1 ∧ · · · ∧ yn = sgn(σ)yσ(1) ⊗ · · · ⊗ yσ(n) , (5.33)
σ∈Sn
1 X
y1 ⊗a · · · ⊗a yn = sgn(σ)yσ(1) ⊗ · · · ⊗ yσ(n) . (5.34)
n!
σ∈Sn

Let {ei }i∈I be a linearly ordered orthonormal basis in Z. Then

√
n!ei1 ⊗a · · · ⊗a ein , i1 < · · · < in , (5.35)
forms an o.n.b of ⊗na (Z).
√
n!
√ e⊗k1 ⊗s · · · ⊗s e⊗k
im ,
m
k1 + · · · + km = n, (5.36)
k1 ! · · · kn ! i1
forms an o.n.b of ⊗m
s (Z).
If dim Z = d, then
(d + n − 1)! d!
dim ⊗ns Z = , dim ⊗na Z = . (5.37)
(d − 1)!n! n!(d − n)!

39
5.9 Exponential law
Let Z, W be Hilbert spaces. We can treat them as subspaces of Z ⊕ W. Let
Φ ∈ ⊗ns/a Z, Ψ ∈ ⊗m
s/a W. We can identify Φ ⊗ Ψ with
r
(n + m)!
U Φ ⊗ Ψ := Φ ⊗s/a Ψ ∈ ⊗n+m
s/a (Z ⊕ W). (5.38)
n!m!
Theorem 5.3. The map (5.38) extends to a unitary map

U : Γs/a (Z) ⊗ Γs/a (W) → Γs/a (Z ⊕ W). (5.39)

It satisfies

U Ω ⊗ Ω = Ω, (5.40)


dΓ(h ⊕ g)U = U dΓ(h) ⊗ 1l + 1l ⊗ dΓ(g) , (5.41)
Γ(p ⊕ q)U = U Γ(p) ⊗ U Γ(q), (5.42)
∗ ∗ ∗


a (z ⊕ w)U = U a (z) ⊗ 1l + 1l ⊗ a (w) , (5.43)


a(z ⊕ w)U = U a(z) ⊗ 1l + 1l ⊗ a(w) , in the bosonic case, (5.44)
∗
= U a∗ (z) ⊗ 1l + (−1)N ⊗ a∗ (z) ,


a (z ⊕ w)U (5.45)
= U a(z) ⊗ 1l + (−1)N ⊗ a(z) , in the fermionic case.


a(z ⊕ w)U (5.46)

Proof. Let us prove the unitarity of this map in the symmetric case:
1 X
Φ ⊗s Ψ = Θ(σ)Φ ⊗ Ψ (5.47)
(n + m)!
σ∈Sn+m
n!m! X
= Θ(σ)Φ ⊗ Ψ. (5.48)
(n + m)!
[σ]∈Sn+m /Sn ×Sm

The terms on the right are mutually orthogonal. The maps Θ(σ) are unitary.
The number of cosets in Sn+m /Sn × Sm is (n+m)!
n!m! . Therefore the square norm
of (5.47) is
n!m!
kΦ ⊗ Ψk2 . (5.49)
(n + m)!
2

5.10 Wick symbol

Let Z = L2 (Rd ) or Z = Cn . The variable ξ will be interpreted R as an elemenrt
of Rd in the former case, as ξ = 1, 2, . . . , nP
in the latter case. dξ will have the
n
usual meaning in the first case, it will be ξ=1 in the second.

(ξ1 , · · · ξm , ξk0 , · · · , ξ10 ) 7→ b(ξ1 , · · · ξm , ξk0 , · · · , ξ10 ) (5.50)

40
be a complex function. Note that (5.50) can be also interpreted as the integral
kernel of an operator b from ⊗k Z to ⊗m Z:
Z Z
(Φ|bΨ) = · · · Φ(ξ1 , · · · ξm )b(ξ1 , · · · ξm , ξk0 , · · · , ξ10 )

Ψ(ξk0 , · · · , ξ10 )dξ1 · · · dξm dξk0 · · · dξ10 . (5.51)

We can restrict (5.51) to Φ ∈ ⊗ks/a Z to Ψ ∈ ⊗m
s/a Z. Then (5.51) will depend
only on the symmetrization/antisymmetrization of b, that is
bs/a := Θm k
s/a bΘs/a . (5.52)
Thus to describe integral kernels of operators from ⊗ks/a Z to ⊗m s/a Z it is enough
to consider functions symmetric/antisymmetric separately wrt the first m and
the last k arguments.
In this subsection we will put “hats” on the creation/annihillation operators.
The symbols a∗ (ξ), a(ξ) without hats will be reserved for classical variables,
which in the bosonic case commute and in the fermionic anticommute.
By a polynomial on Z ⊕Z we will mean a linear combination of the following
expressions
Z Z
b(a∗ , a) = · · · b(ξ1 , · · · ξm , ξk0 , · · · , ξ10 ) (5.53)

a∗ (ξ1 ) · · · a∗ (ξm )a(ξk0 ) · · · a(ξ10 )dξ1 · · · dξm dξk0 · · · dξ10 ,

where b are symmetric/antisymmetric separately wrt the first m and the last k
arguments. In the symmetric case this can be interpreted as a usual polynomial
on Z ⊕ Z, but it is common to use this term also in the antisymmetric case.
The Wick quantization of b(a∗ , a) is defined as
Z
∗
b(â , â) = b(ξ1 , · · · ξm , ξk0 , · · · , ξ10 ) (5.54)

â∗ (ξ1 ) · · · â∗ (ξm )â(ξk0 ) · · · â(ξ10 )dξ1 , · · · dξk dξ10 · · · dξm
0
.
(Actually, by (5.52), in (5.53) and (5.54) we can consider b which is not
symmetric/antisymmetric.)
Here is an equivalent definition of b(â, â): Its only nonzero matrix elements
are between Φ ∈ ⊗p+m p+k
s/a Z, Ψ ∈ ⊗s/a Z, and equal
p
(m + p)!(k + p)!
∗
(Φ|b(â , â)Ψ) = (Φ|b ⊗ 1⊗p
Z Ψ). (5.55)
p!
To see this it is enough to use the formal identity (5.22) several times:
Φ|â∗ (ξ1 ) · · · â∗ (ξm )â(ξk0 ) · · · â(ξ10 )Ψ


(5.56)
= â(ξm ) · · · â(ξ1 )Φ|â(ξk0 ) · · · â(ξ10 )Ψ


(5.57)
p
= (m + p) · · · (p + 1)(k + p) · · · (p + 1) (5.58)
Z
0
× Φ(ξm , . . . , ξ1 , ηp , . . . , η1 )Ψ(ξm , . . . , ξ10 , ηp , . . . , η1 )dηp · · · dη1 . (5.59)

41
 Essentially every operator on a Fock space can be written as a linear com-
bination of (5.54).

5.11 Wick symbol and coherent states

In the bosonic case, we have the identities
∗ ∗
e−â (b)+â(b)
â(v)eâ (b)−â(b)
= â(v) + (v|b), (5.60)
−â∗ (b)+â(b) ∗ â∗ (b)−â(b)
e â (v)e = â(v) + (v|b). (5.61)
We also introduce the coherent state corresponding to b ∈ Z:
∗
Ωb := eâ (b)−â(b)
Ω. (5.62)
Note that â(v)Ωb = (v|b)Ωb . We have the identity
(Ωb |c(â∗ , â)Ωb ) =c(b∗ , b). (5.63)

5.12 Particle number preserving operators

If m = k, then the operator b(â∗ , â) preserves the number of particles and (5.55).
For Φ ∈ ⊗ns/a Z, Ψ ∈ ⊗ns/a Z it can be rewritten as
n! ⊗(n−m)
(Φ|b(â∗ , â)Ψ) = (Φ|b ⊗ 1Z Ψ). (5.64)
(n − m)!
n!
But (n−m)!m! is the number of m-element subsets of {1, 2, . . . , n}. Therefore in
the obvious notation, we can rewrite (5.64) as
1 X
b(â∗ , â) = bi1 ,...,im . (5.65)
m!
1≤i1 <···<im ≤n

In particular, for m = 2 we can write

1 ∗ X
b(â , â) = bij . (5.66)
2
1≤i<j≤n

Finally, for m = 1, we have

X
b(â∗ , â) = bi = dΓ(b). (5.67)
1≤i≤n

5.13 Examples
Consider the Schrödinger Hamiltonian of n identical particles on L2 (RdN )
n
X X
Hn = − ∆i + V (xi − xj ), (5.68)
i=1 1≤i<j≤n
n
X 1
Pn = ∂xi , (5.69)
i=1
i

42
In the momentum representation
n
X
Hn = p2i
i=1
X
+(2π)−d δ(p0i + p0j − pj − pi )V̂ (p0i − pi ).
1≤i<j≤N
n
X
Pn = pi .
i=1

Consider the 2nd quantization of L2 (Rd ). We have the position representa-

tion, with the generic variables x, y and the momentum representation with the
generic variables k, k 0 . We can pass from one representation to the other by
Z Z
∗ −d ∗ −ikx ∗ −d
a (k) = (2π) 2 a (x)e dx, a (x) = (2π) 2 a∗ (k)eikx dk, (5.70)
Z Z
d d
a(k) = (2π)− 2 a(x)eikx dx, a(x) = (2π)− 2 a(k)e−ikx dk. (5.71)

In the 2nd quantized notation we can rewrite all this as

Z
∞
H := ⊕ Hn = − a∗x ∆x ax dx (5.72)
n=0
Z Z
+ dxdyV (x − y)a∗x a∗y ay ax
Z
= p2 a∗p ap dp (5.73)
Z Z Z
+ (2π)−d dpdqdk V̂ (k)a∗p+k a∗q−k aq ap
Z
∞ 1
P := ⊕ Pn = a∗x ∂x ax dx (5.74)
n=0 i
Z
= pa∗p ap dp. (5.75)

 
2π d
Consider L2 ([0, L]d ) ' L2 L Z and its 2nd quantization. Again we use
x, y in the position representation with periodic boundary conditions and k, k 0
in the momentum representation. We can pass from one representation to the
other by
Z
d d
X
a∗ (k) = L− 2 a(x)e−ikx dx, a∗ (x) = L− 2 a(k)eikx , (5.76)
k
Z
−d −d
X
a(k) = L 2 a(x)eikx dx, a(x) = L 2 a(k)e−ikx . (5.77)
k

43
 Here are the analogs of (5.73) and (5.75):
X
H= p2 a∗p ap
p
XXX
+ L−d V̂ (k)a∗p+k a∗q−k aq ap ,
p q k
X
P = pa∗p ap .
p

6 Formalism of classical mechanics

6.1 Dynamical systems
Suppose that a system is described by a manifold Y called a “phase space”. The
space C ∞ (Y) of smooth functions on Y describes possible observables.
To define a dynamics one needs to fix a vector field V on Y and one has the
equations of motion
d

ζ(t) = V ζ(t) , ζ(0) = ζ0 . (6.1)
dt
The evolution of an observable F ∈ C ∞ (Y) is then given by

d
F (ζ) = hdF (ζ)|V (ζ)i. (6.2)
dt
If we fix coordinates xi on Y, then for each y ∈ Y, the tangent space Ty Y
∂
is spanned by the vectors ∂x i and its dual, called the cotangent space by the

1-forms dxi statisfying

∂
hdxi | j i = δij . (6.3)
∂x
∂
The vector field V can be written as V = V i (x) ∂x i and the 1-form dF as
∂F (x) i
dF = ∂xi dx , and the equations (6.1) and (6.2) become

d i
ζ (t) = V i ζ(t) ,


(6.4)
dt
d ∂F (ζ)
F (ζ) = · V i (ζ). (6.5)
dt ∂xi

6.2 Hamiltonian dynamics

Let us first recall basic facts about Hamiltonian dynamics. Let us begin with
the phase space Y = R2n with coordinates φi , πj , i = 1, . . . , n. Let us consider a
Hamiltonian H, that is a function on the phase space, so that it can be expressed
in terms of φ and π. The Hamiltoni equations generated by H are
d i

(φ , πi ) = ∂πi H(φ, π), −∂φi H(φ, π) . (6.6)
dt

44
Another form:     
d φ 0 1 ∂φ H
= . (6.7)
dt π −1 0 ∂π H
A vector field of the form
∂ ∂
∂πi H(φ, π) i
− ∂φi H(φ, π) (6.8)
∂φ ∂πi
for some function H is called a Hamiltonian vector field. Thus the Hamilton
equations are given by a Hamiltonian vector field.
Given two functions F, G on Y one introduces the Poisson bracket which is
a bilinear antisymmetric map C ∞ (Y) × C ∞ (Y) → C ∞ (Y), as

{F, G} = ∂φi F ∂πi G − ∂πi F ∂φi G. (6.9)

The Poisson bracket satisfies the Leibniz and Jacobi identity

{F, GH} = {F, G}H + G{F, H}, (6.10)

{{F, G}, H} + {{G, H}, F } + {{H, F }, G} = 0. (6.11)

The coordinates φ, π satisfy

{φi , φj } = {πi , πj } = 0, {φi , πj } = δji . (6.12)

The Hamilton equations can be rewritten as

d i
φ = {φi , H}, (6.13)
dt
d
πi = {πi , H}. (6.14)
dt
More generally, the evolution of every observable F is given by
d
F = {F, H}. (6.15)
dt
Note that the Poisson bracket is preserved by a Hamiltonian flow:
d
{F, G} = {{F, H}, G} + {F, {G, H}} = {{F, G}, H}. (6.16)
dt

6.3 Symplectic form

Let us continue with the phase space Y = R2n with the coordinates φi and πi .
∂ ∂
For any y ∈ Y, the tangent space Ty R2n is spanned by the vectors ∂φ i , ∂π
i
∗ 2n i
and the cotangent space Ty R by the 1-forms dφ , dπi . One can introduce the
2-form on Y:
ω = dπ i ∧ dφi , (6.17)
called the symplectic form.

45
 Clearly,
D ∂ ∂ E D ∂ ∂ E
ω| , = ω| , = 0, (6.18)
∂φi ∂φj ∂πi ∂πj
D ∂ ∂ E D ∂ ∂ E
ω| , j = − ω| i , = δji . (6.19)
∂πi ∂φ ∂φ ∂πj

One can introduce a linear map ω : Ty R2n → T∗y R2n such that

hω|v, zi = hv|ωzi. (6.20)

Thus
 ∂ 
ω =dπi , (6.21)
∂φi
 ∂ 
ω = − dφi . (6.22)
∂πi
   
0 −1 −1 0 1
Thus ω is given by the matrix ω = . Clearly, ω = . A
1 0 −1 0
vector field on a symplectic manifold is called Hamiltonian if it has the form

V (y) = −ω −1 dH(y). (6.23)

In the previous subsection the phase space was Y = R2n and φi , πi were
functions on Y satisfying the Poisson bracket relations (6.12).
As a side remark, let us note the following properties of the symplectic form:
1. ω(y) is nondegenerate at every point y ∈ Y;

2. dω = 0.
Nondegenerate means: Let z ∈ Ty Y be a tangent vector. If for any vector
v ∈ Ty Y we have ω(z, v) = 0, then z = 0.
One can be more general: Let Y be a manifold equipped with a 2-form
satisfying 1. and 2. Then we say that Y, ω is a symplectic manifold. The
Darboux Theorem says that on any symplectic manifold locally we can always
choose coordinates, say φi , πj , i = 1, . . . , n, such that (6.17) holds.
Specifying a symplectic form is equivalent to specifying a Poisson bracket.
Indeed, dF = (∂φ F, ∂π F ), dG = (∂φ G, ∂π G), we can write the Poisson bracket
as
D 
{F, G} = hdF |ω −1 dGi = − ω|ω −1 (dF ), ω −1 (dG) , F, G ∈ C ∞ (Y). (6.24)

In most our applications, the phase space wil have the structure of a vector
space and φi , πj , i = 1, . . . , n, can be chosen to be the coordinates in a basis.
Then the tangent space to Ty Y at any point y ∈ Y can be identified with Y
itself and the form ω is simply a nondegenerate antisymmetric bilinear form on

46
Y. The Darboux Theorem says that we can identify a symplectic manifold with
a symplectic vector space at least locally.
The space Y T of linear functionals on Y obviously is contained in C ∞ (Y).
For a linear functional on Y, its derivative is the original functional itself. There-
fore, (6.17) can be simplified and written as
ω = π i ∧ φi , (6.25)
If dim Y is finite, then Y T = ωY (because ω is nondegenerate) and (6.24)
determines the Poisson bracket on the whole C ∞ (Y), consistently with (6.17).

6.4 Lagrangian formalism

Suppose that Rn is described by the coordinates φi . Let L(t, φ, φ̇) be a function
on R × Rn × Rn called the Lagrangian. The Euler-Lagrange equations are
∂L d ∂L
= . (6.26)
∂φi dt ∂ φ̇i
∂L
Define the canonical momentum conjugate to φi , that is πi := ∂ φ̇i
.
The initial consitions for (6.26) can be expressed in terms φ, φ̇. Suppose that
we can express φ̇ in terms of φ, π. Then initial conditions can be described in
terms of φ, π and we can pass from the Lagrangian to the Hamiltonian formalism.
More precisely, we can introduce
the tautological 1-form, θ := πi dφi , (6.27)
i
the symplectic 2-form ω := dπi ∧ dφ = dθ, (6.28)
H(t, φ, π) := πi φ̇i − L t, φ, φ̇(φ, π) .


and the Hamiltonian (6.29)
We check that ∂φ H = −∂φ L, ∂π H = φ̇. Hence the Euler-Lagrange equations
imply the Hamilton equations.

6.5 Noether Theorem

Suppose that φ depend on a parameter α. Define
∂L
P := ∂α φi . (6.30)
∂ φ̇i


Suppose that L t, φ(α), φ̇(α) is independent of α. The Noether Theorem says


that along the evolution P t, φ(t), φ̇(t) is constant. Indeed,
d  d ∂L  ∂L d i
P = φi,α + φ,α (6.31)
dt dt ∂ φ̇ i ∂ φ̇i dt
∂L ∂L i
= i φi,α + φ̇,α = 0. (6.32)
∂φ ∂ φ̇i
If α = ~x, so that the Lagrangian is invariant wrt translations, then P is
called the (total) momentum. (Note a confusing collision of terminology with
canonical momenta).

47
6.6 Classical field theory
Consider the space Rd (the “spacetime”, where however the metric or Lorentz
structure is for the moment irrelevant) and a space Rn (the “internal degrees
of freedom”, whose indices will be as a rule omitted). A field configuration is a
function Rd 3 x 7→ ζ(x) = [ζ i (x)] ∈ Rn . The classical field φi (x) is the “value
of the ith coordinate at x of the field configuration”, that is

hφi (x)|ζi = ζ i (x). (6.33)

Thus the classical field is a linear functional on field configurations.

Suppose that L(x) = L(x, φ(x), φ,µ (x)) is a function called the Lagrangian
density. Z
I := L(x)dx (6.34)

is called the action. Thus the action is a (typically nonlinear) functional on field
configurations.
Let us compute the the derivative in the direction of ε ∈ Cc∞ (Rd ) of the
action at the configuration ζ:

hI|ζ + εi − hI|ζi
Z 



= L x, ζ(x) + ε(x), ζ,µ (x) + ε,µ (x) − L x, ζ(x), ζ,µ (x) dx
Z Z
∂L ∂L
≈ (x, ζ(x), ζ,µ (x))ε(x)dx + (x, ζ(x), ζ,µ (x))ε,µ (x)dx
∂φ(x) ∂φ,µ (x)
Z 
∂L ∂L 
= (x, ζ(x), ζ,µ (x)) − ∂µ (x, ζ(x), ζ,µ (x)) ε(x)dx (6.35)
∂φ(x) ∂φ,µ (x)
Z  ∂L 
+ ∂µ (x, ζ(x), ζ,µ (x))ε(x) dx
∂φ,µ (x)

The last term vanishes by the Stokes Theorem. Hence the derivative in the
direction of ε is given by (6.35). If we require that ζ is stationary, that is this
derivative vanishes, we obtain the Euler-Lagrange equations:

∂L(x)
∂φ(x) L(x) − ∂µ =0 (6.36)
∂φ,µ (x)

Let π µ (x) denote the canonical momentum conjugate to φ(x) in the direction
of xµ :
∂L(x)
π µ (x) = . (6.37)
∂φ,µ (x)
Consider the space of solutions of the Euler-Lagrange equations. The opera-
tion d will denote the exterior derivative on this (infinite dimensional) space.
Similarly, on this space we use ∧. If something holds on solutions of the Euler-
Lagrange equation, we will say that it is true on shell.

48
 We introduce

the tautological current θµ (x) := π µ (x)dφ(x), (6.38)

the symplectic current j µ (x) := dπ µ (x) ∧ dφ(x) = dθµ . (6.39)

The divergence of the tautological current is the differential of the Lagrangian

and that of the symplectic current vanishes:

∂µ θµ (x) = dL, (6.40)

µ
∂µ j (x) = 0. (6.41)

Indeed,

∂L(x) ∂L(x)
∂µ θµ (x) = ∂µ ∧ dφ(x) + ∧ dφ,µ (x) (6.42)
∂φ,µ (x) ∂φ,µ (x)
∂L(x) ∂L(x)
= ∧ dφ(x) + ∧ dφ,µ (x) = dL. (6.43)
∂φ(x) ∂φ,µ (x)

Introduce the Noetherian stress-energy tensor

∂L(x)
Tνµ (x) := − φ,ν (x) + δνµ L(x). (6.44)
∂φ,µ (x)

Here is the Noether Theorem in the context of classical field theory: on solutions
of the Euler-Lagrange equation we have

∂µ Tνµ (x) = ∂ν L(x) (6.45)

In particular, if L is invariant wrt translations, then the stress-energy tensor is

conserved (its divergence vanishes).
Indeed,
 ∂L(x)  ∂L(x)
∂µ Tνµ (x) = −∂µ φ,ν (x) − φ,µν (x) (6.46)
∂φ,µ (x) ∂φ,µ (x)
∂L(x) ∂L(x)
+ φ,ν (x) + φ,αν (x) + ∂ν L(x). (6.47)
∂φ(x) ∂φ,α (x)

Then we use the Euler-Lagrange equations (6.36).

6.7 Hyperbolic classical field theory

Suppose now that the Euler-Lagrange equations (6.36) are characterized by a
finite speed of propagation. A typical example of such a situation is when Rd
is the Minkowski space, the Lagrangian is
1
L(x) = − g µν ∂µ (x)φ∂ν φ(x) − P (φ(x)), (6.48)
2

49
which leads to the Klein-Gordon equation with nonlinear terms:

(−2 + m2 (x))φ(x) = P 0 (φ(x)). (6.49)

One can then introduce an identification M ' R × Σ, where R describes “time”

and Σ is a “spatial cross-section”, in the case of the Minkowski space Σ = Rd−1 .
The Lagrangian is given by the integral of the Lagrangian density over a constant
time surface {t} × Σ Z
L(t) := L(t, ~x)d~x. (6.50)

The momentum conjugate to the field φ(t, ~x) coincides with the temporal coor-
dinate of (6.37):
∂L(t)
π(t, ~x) := = π 0 (t, ~x). (6.51)
∂ φ̇(t, ~x)
The Hamiltonian obtained from the Lagrangian by the Legendre transformation
coincides with the integral of the 00-component of the stress-energy tensor over
the corresponding constant time surface:
Z Z
H(t) := φ̇(t, ~x)π(t, ~x)d~x − L(t) = T00 (t, ~x)d~x. (6.52)

It is natural to use the set of space-compact solutions of the Euler-Lagrange

equations as the phase space. It is equipped with a symplectic form
Z Z
ω := j 0 (t, ~x)d~x = dπ 0 (t, ~x) ∧ dφ(t, ~x)d~x. (6.53)

Using (6.41) we can show that (6.53) does not depend on t. Actually, instead
of integrating over {t} × Σ, we can integrate the symplectic current over any
“Cauchy surface” obtaining the same ω.
The symplectic form ω corresponds to the equal-time Poisson brackets

{φ(t, ~x), φ(t, ~y )} = {π(t, ~x), π(t, ~y )} = 0,

{φ(t, ~x), π(t, ~y )} = δ(~x − ~y ). (6.54)

7 Canonical Commutation Relations

7.1 Symplectic vector spaces
Let Y be a vector space. Recall that it is called symplectic if it is equipped with
a nondegenerate antisymmetric 2-form. Thus we have a bilinear form

Y × Y 3 (y, z) 7→ ω(y, z) (7.1)

such that ω(y, z) = −ω(z, y) and for every y 6= 0 we can find z such that
ω(y, z) 6= 0.

50
 Every finite dimensional symplectic vector space has an even dimension. If
we choose a basis, we can write
X
ω(y, z) = ωij y i z j , (7.2)
ij

where [ωij ] is an antisymmetric invertible matrix.

Let φi , i = 1, . . . , 2n, denote the coordinate functionals, that is for y ∈ Y we
have φi (y) = y i . Then using
φi ∧ φj (y, z) = φi (y)φj (z) − φi (z)φj (y). (7.3)
we see that (7.2) can be written as ω = 21 ωij φi ∧ φj .
Let [ω ij ] be the inverse of [ωij ]. Then it is also an invertible antisymmetric
matrix. We can equip functions on Y with a Poisson bracket: for F, G ∈ C ∞ (Y)
we set X
{F, G} = − ω ij ∂φi F ∂φj G. (7.4)
In particular,
{φi , φj } = −ω ij . (7.5)
2n
We say that a symplectic space is R is equipped with a symplectic basis if
the symplectic form is given by the matrix
 
0 1l
ω= . (7.6)
−1l 0
In every finite dimensional symplectic vector space we can find a symplectic
basis. It is then natural to separate the variables into two families. writing
xi = φi , pi := φn+i i = 1, . . . , n. Then ω = xi ∧ pi and
{F, G} = ∂xi F ∂pi G − ∂pi F ∂xi G. (7.7)
In particular,
{xi , xj } = {pi , pj } = 0, {xi , pj } = δji . (7.8)
Recall that by an affine transformation on a vector space Y we mean a
transformation of the form
Y 3 y 7→ T y + z ∈ Y, (7.9)
where T is linear and z ∈ Y.
A linear transformation T on a symplectic space Y, ω) is called symplectic
if it preserves the symplectic form, that is
ω(T y, T z) = ω(y, z). (7.10)
Equivalently, it preserves the Poisson product: that means
{F ◦ T, G ◦ T } = {F, G} ◦ T, (7.11)
where ◦ denotes the composition. We say that (7.9) is affine symplectic, if T is
symplectic.

51
Remark 7.1. There exists a useful generalization of a symplectic vector space:
a symplectic manifold. More precisely, let Y be a manifold equipped with a
2-form ω such that dω = 0 is ω is nondegenerate on the whole Y. Then we
say that (Y, ω) is a symplectic manifold. The Darboux Theorem says that on
any symplectic manifold locally we can always choose coordinates, say xi , pj ,
i = 1, . . . , n, such that (7.8) holds. One can also define the Poisson bracket on
Y. However, we will avoid using this concept.

7.2 Quantization of linear and quadratic observables

Suppose for a moment that Y is a symplectic manifold. Informally one can say
that quantization is a map that to a function F on Y associates an operator F̂
on a certain Hilbert space H, which satisfies the following conditions:

1̂ = 1l; (7.12)
1
(F̂ Ĝ + ĜF̂ ) ≈ Fd G; (7.13)
2
\
[F̂ , Ĝ] ≈ i~{F, G}. (7.14)

Here ~ is a small positive parameter and ≈ means some kind of equality modulo
terms small for small ~.
One can prove that one cannot replace ≈ with =.
Almost always we will assume that Y is a symplectic vector space. As we
discussed above, if Y is finite dimensional, we can always find coordinates xi , pi ,
i = 1, . . . , n satisfying (7.8). In other words, Y = R2n is a symplectic vector
space equipped with a symplectic basis.
With the above classical system we associate a quantum system as follows.
Let ~ be a real parameter. We consider the Hilbert space H := L2 (Rn ) equipped
with the operators
∂
x̂i Ψ(x) = xi Ψ(x), p̂i Ψ(x) := ~ Ψ(x). (7.15)
i∂xi
They satisfy the Heisenberg commutation relations:

[x̂i , x̂j ] = [p̂i , p̂j ] = 0, [x̂i , p̂j ] = i~δji . (7.16)

Note that 1st degree polynomials, that is,

V = ξi xi + η i pi + c, (7.17)
i i
V̂ = ξi x̂ + η p̂i + c1l. (7.18)

satisfy (7.14) exactly:

[V̂ , V̂ 0 ] = i~{V,
\ V 0} (7.19)

One would like to extend the quantization to more general functions, not
only 1st order polynomials. There are many possibilities in the literature (e.g.

52
the Weyl quantization, Wick quantization, etc.). We will not discuss them here.
Actually, for our present purposes we will only need quantization of second order
polynomials, which we define as follows:
1 1
H= Aij xi xj + B ij pi pj + Cij xi pj , (7.20)
2 2
1 1 1
Ĥ = Aij x̂i x̂j + B ij p̂i p̂j + Cij (x̂i p̂j + p̂j x̂i ). (7.21)
2 2 2
Note that with this definition we have exact versions of (7.13) and (7.14)

[Ĥ, Ĥ 0 ] = i~{H,
\ H 0 }, (7.22)
1
(V̂ V̂ 0 + V̂ 0 V̂ ) = Vd
V0 (7.23)
2
for polynomials V, V 0 of degree ≤ 1 and for polynomials H, H 0 of degree ≤ 2.

7.3 Quantization of symplectic transformations

Let H be a Hamiltonian given by a quadratic polynomial on a symplectic vector
space Y with the coordinates φi , i = 1, . . . , 2n.
1X X
H= hij φi φj + di φi . (7.24)
2 i

The Hamilton equations

d i
φ (t) = {φi (t), H}, (7.25)
dt
define an affine symplectic transformation
φ(0) 7→ φ(t). (7.26)
Note that if the variables are divided into “positions” and “momenta”, then
in general they are mixed by the flow generated by a quadratic Hamiltonian.
What is preserved is the “symplectic structure”.
The variables φ(t) are 1st order polynomials and we can quantize them
obtaining φ̂(t). Likewise, we can quantize H(t) obtaining Ĥ(t).
Theorem 7.2.
Ĥ Ĥ
φ̂i (t) =eit ~ φ̂i (0)e−it ~ . (7.27)

Proof. Clearly, (7.27) is satisfied for t = 0. We check that φ̂(t) satisfies the
Heisenberg equations:
d i
i~ \
φ̂ (t) = i~{φ(t), H} = [φ̂i (t), Ĥ], (7.28)
dt
This implies (7.27) for all t. 2
Thus we can first solve the classical Hamilton equations obtaining φ(t) and
then put the hat, or first put the hat and then solve the quantum Heisenberg
equation—we obtain the same φ̂(t).

53
7.4 Weyl operators
Proposition 7.3 (Baker-Campbell-Hausdorff formula). Suppose that
   
[A, B], A = [A, B], B = 0.
Then
1
eA+B = eA eB e− 2 [A,B] .
Proof. We will show that for any t ∈ R
1 2
et(A+B) = etA etB e− 2 t [A,B]
. (7.29)
First, using the Lie formula, we obtain
∞ n
X t
etA Be−tA = adnA (B)
n=0
n!
= B + t[A, B].
Now
d tA tB − 1 t2 [A,B] 1 2
e e e 2 = AetA etB e− 2 t [A,B]
dt
1 2
+etA BetB e− 2 t [A,B]
1 2
−etA etB t[A, B]e− 2 t [A,B]

1 2
= (A + B)etA etB e− 2 t [A,B]
.
Besides, (7.29) is true for t = 1. 2
Let ξ = (ξ1 , . . . , ξd ), η = (η 1 , . . . , η d ) ∈ Rd . We will write
x̂(ξ) := ξi x̂i , p̂(η) := η j p̂j .
Clearly,
[x̂(ξ), p̂(η)] = i~ξ · η.
Therefore,
i~
eix̂(ξ) eip̂(η) = e− 2 ξη ei(x̂(ξ)+p̂(η)) (7.30)
= e−i~ξ eip̂(η) eix̂(ξ) . (7.31)
The operators ei(x̂(ξ)+p̂(η)) are sometimes called Weyl operators. They satisfy
the Weyl commutation relations:


0 0 i~ 0 0 0 0
ei(x̂(ξ)+p̂(η)) ei(x̂(ξ )+p̂(η )) = e− 2 (ξη −ηξ ) ei x̂(ξ+ξ )+p̂(η+η ) . (7.32)
The Weyl commutation relations, at least formally, imply the Heisenberg com-
mutation relations.
Weyl operators translate the position and momentum:
i i
e ~ (−p̂(y)+x̂(w)) x̂e ~ (p̂(y)−x̂(w)) = x̂ − y,
i i
e ~ (−p̂(y)+x̂(w)) p̂e ~ (p̂(y)−x̂(w)) = p̂ − w.

54
7.5 Stone-von Neumann Theorem
Operators x̂i , p̂i are unbounded. Hence the Heisenberg commutation relations
(7.16) are problematic–without specifying the domain it is not clear what they
precisely mean. The Weyl commutation relations (7.32) involve only bounded
operators, hence their meaning is clear.
The following theorem is one of mathematical foundations of Quantum Me-
chanics:
Theorem 7.4. Suppose that

R2n 3 (ξ, η) 7→ Ŵ (ξ, η) ∈ U (H) (7.33)

is a family of operators satisfying

1. the Weyl commutation relations:
i~ 0 0
Ŵ (ξ, η)Ŵ (ξ 0 , η 0 ) = e− 2 (ξη −ηξ ) Ŵ (ξ + ξ 0 , η + η 0 ),

2. strong continuity: R2n 3 (ξ, η) 7→ Ŵ (ξ, η)Ψ is continuous,

3. irreducibility: there are no nontrivial subspaces in H invariant wrt Ŵ (ξ, η).
Then there exists a unitary U : H → L2 (Rn ) such that

U Ŵ (ξ, η)U ∗ = ei(x̂(ξ)+p̂(η)) . (7.34)

If we drop the irreducibility condition, then there exists a Hilbert space K and a
unitary operator U : H → L2 (Rn ) ⊗ K such that

U Ŵ (ξ, η)U ∗ = ei(x̂(ξ)+p̂(η)) ⊗ K. (7.35)

7.6 Representations of the CCR

Let (Y, ω) be a symplectic space, possibly infinite dimensional. Let H be a
Hilbert space. We say that a map

Y 3 y 7→ Ŵ (y) ∈ U (H) (7.36)

is a representation of the CCR or a CCR representation if

i 0
Ŵ (y)Ŵ (y 0 ) = e− 2 ω(y,y ) Ŵ (y + y 0 ). (7.37)

We say that it is regular if

t 7→ Ŵ (ty)
is strongly continuous for any y ∈ Y.
One of the main examples of regular CCR representations is the Schrödinger
representation

Rn ⊕ Rn 3 (ξ, η) 7→ Ŵ (ξ, η) := ei(x̂(ξ)+p̂(η)) ∈ U L2 (Rn ) .

(7.38)

55
In this case Y = Rn ⊕ Rn , y = (ξ, η), y 0 = (ξ 0 , η 0 ),

ω(y, y 0 ) = ξη 0 − ηξ 0 . (7.39)

If Y is a finite dimensional symplectic space, then its dimension is necessarily

even and we can find a basis so that the symplectic form is (7.39). The Stone-
von Neumann theorem says that every irreducible strongly continuous CCR
representation over a finite dimensional symplectic space is unitarily equivalent
to a multiple of the Schrödinger representation.
Note that if we have a regular CCR representation over Y and y ∈ Y, then

R 3 t 7→ Ŵ (ty) ∈ U (H) (7.40)

is a 1-parameter unitary group. Hence it has a generator, which will be denoted

φ̂(y), so that
Ŵ (ty) = eitφ̂(y) . (7.41)
It is easy to show that on an appropriate domain

φ̂(ty) = tφ̂(y), (7.42)

φ̂(y + y 0 ) = φ̂(y) + φ̂(y 0 ), (7.43)
0 0
[φ̂(y), φ̂(y )] = iω(y, y ). (7.44)

We can also extend these “fields” to the complexification of Y, denoted

CY := Y + iY, that is consisting of y = yR + iyI , yR , yI ∈ Y. Note that we have
the complex conjugation y := yR − iyI . We set

φ̂(y) := φ̂(yR ) + iφ̂(yI ). (7.45)

Clearly, φ̂(y)∗ = φ̂(y).

There are many irregular CCR representations. For instance, we can con-
sider l2 (Y) with basis {ey | y ∈ Y} and
i 0
Ŵ (y)ey0 = e− 2 y·ωy ey+y0 . (7.46)

Usually, they are only a non-physical mathematical curiosity.

7.7 Fock representations of the CCR

Let Z be a Hilbert space. It is then a symplectic space with the form

ω(z, z 0 ) := 2Im(z|z 0 ). (7.47)

Set ∗
Ŵ (z) := eâ (z)−â(z)
. (7.48)
Then, using the Baker-Campbell-Hausdorff formula we see that


Z 3 z 7→ Ŵ (z) ∈ U Γs (Z) (7.49)

56
is a regular irreducible CCR representation over the symplectic space Z. Of
course, if Z is finite dimensional, it is unitarily equivalent to the Schrödinger
representation, which follows from the Stone-von Neumann Theorem, but is also
an obvious consequence of the well known theory of the harmonic oscillator.
We have
(z|z) ∗
Ŵ (z) = e− 2 eâ (z) −â(z)
e , (7.50)
â(z)Ω = 0, z ∈ Z. (7.51)

Hence,
(z|z)
(Ω|Ŵ (z)Ω) = e− 2 . (7.52)
Therefore, if we know the vacuum state we can recover the real part of the scalar
product on Z.
The symplectic form ω fixes the imaginary part of the scalar product, see
(7.47). Then there are many ways you can complete it to a full scalar product.
Each of them leads to a Fock representation of CCR satisfying (7.52). If Z
has an infinite dimension, the resulting representations in general will not be
unitarily equivalent, as we will illustrate in the next section.

7.8 Equivalence of representations of CCR

Suppose that (Y, ω) is a symplectic space and Y 3 y 7→ Wi (y) ∈ U (Hi ), i = 1, 2
are two representations of CCR. We say that they are unitarily equivalent if
there exists a unitary U : H1 → H2 such that

U W1 (y) = W2 (y)U, y ∈ Y. (7.53)

The Stone-von Neumann Theorem says that all CCR representations over a
finite dimensional symplectic space are unitarily equivalent up to a multiplicity.
This is not the case for an infinite number of degrees of freedom.
In fact, for instance, there are many inequivalent Fock representations over
the same infinite dimensional symplectic space.
Let us do a computation that illustrates this. Let us start with the symplectic
space R2 with the Schrödinger representation on L2 (R) given by x̂, p̂. For any
ω > 0 we introduce the creation/annihilation operators

1 √ ip̂  1 √ ip̂ 
a∗ω := √ ωx̂ − √ , aω := √ ωx̂ + √ . (7.54)
2 ω 2 ω

The vacuum annihilated by aω is

√
4
ω − ωx2
Ωω (x) = √ e 2 . (7.55)
4
π

The vectors
1
eω,n := √ a∗n
ω Ωω (7.56)
n!

57
form an orthonormal basis of eigenvectors of the harmonic ascillator a∗ a. Thus
∞
X
Uω Φ := |n)(eω,n |Φ) (7.57)
n=0

is a unitary operator Uω : L2 (R) → Γs (C) ∼ l2 (0, 1, 2, . . . ). We have

√
1 ω
x̂ = √ (a∗ω + aω ), p̂ = i √ (a∗ω − aω ). (7.58)
2ω 2
Obviously, these representation are unitarily equivalent. In fact, setting
√ √
Vω Φ(x) := 4 ωΦ( ωx), (7.59)

we obtain
√ 1
Vω x̂Vω−1 = ωx̂, Vω p̂Vω−1 = √ p̂, (7.60)
ω
hence
Vω â∗1 Vω−1 = â∗ω , Vω â1 Vω−1 = âω , Vω e1,n = eω,n , (7.61)
and Vω = Uω−1 U1 . We can also compute
s
2
(Ωω |Ω1 ) = 1 1 . (7.62)
ω + ω− 2
2

So far we treated L2 (R) as the underlying space. Let us now treat the
Fock space Γs (C) as the main space. Then for any ω > 0 we have a Fock
representation
 iξ √
2 ∗ iη ω ∗ 
R 3 (ξ, η) 7→ Wω (ξ, η) := exp √ (a + a) + √ (a − a) . (7.63)
2ω 2
They are different, but unitarily equivalent.
It is straightforward to generalize this construction to a finite number of
degrees of freedom. Thus, for any finite sequence of positive numbers we obtain
the equivalence of two CCR representations of the symplectic space R2n : the
Schrödinger representation on L2 (Rn ) and the Fock representation on Γs (Cn ).
The vacua corresponding to ω = (ω1 , . . . , ωn ) and 1 = (1, . . . , 1) have a positive
scalar product:
n s
Y 2
(Ωω |Ω1 ) = 1
−1
> 0. (7.64)
j=1 ωj2 + ωj 2
The Schrödinger representation does not work for n = ∞, since there is
no generalization of the Lebesgue measure to R∞ . However, the Fock space
Γs (l2 ) is well defined. Thus we can define creation/annihilation operators a∗i , aj ,
i = 1, 2, . . . .

58
 For any infinite sequence ω = (ω1 , ω2 , . . . ) we can represent the commutation
relations
[x̂i , p̂j ] = iδij (7.65)
by setting √
1 ωi
x̂ω,i = √ (a∗ + ai ), p̂ω,i = i √ (a∗i − ai ). (7.66)
2ωi i 2
However, if 1 = (1, 1 . . . , . . . ), the infinite product
∞ s
Y 2
(Ωω |Ω1 ) = 1
− 12
. (7.67)
j=1 ωj + ωj
2

is usually zero. One can show that if (7.67) is zero then the representations
given by ω and 1 are inequivalent.

7.9 Two steps of quantization with an infinite number of

degrees of freedom
Description of quantum systems with an infinite number of degrees of freedom
(especially free systems) usually proceeds in two steps.
1. First one describes the symplectic space of classical fields. This automat-
ically fixes commutation relations satisfied by quantum fields.

2. Then one chooses a Hilbert space where the quantum fields are repre-
sented. (Typically, this is a Fock space)
In the mathematically oriented literature there exist several equvalent ways
of presenting CCR relations. Probably, the most economical way involves the
notion of a CCR representation, which we introduced above. An alternative
way is to use a ∗ algebra of CCR. Then the two steps described above can be
described as foillows:
1. Introduce an abstract ∗-algebra of CCR over the space (Y, ω).
2. Find a representation of this algebra.

This approach has a minor problem: there are several, essentially equivalent
but mathematically different ∗-algebras that can be used to describe Canonical
Commutation Relations. Let us describe two of them (there are others).
The “Weyl CCR C ∗ -algebra” is the C ∗ algebra generated by {Ŵ (y) | y ∈ Y}
satisfying

Ŵ (y)∗ = Ŵ (−y), Ŵ (0) = 1l, (7.68)

0 − 2i y·ωy 0 0
Ŵ (y)Ŵ (y ) = e Ŵ (y + y ). (7.69)

59
 The “field CCR ∗-algebra” is the ∗-algebra generated by {φ̂(y) | y ∈ Y}
satisfying

φ̂(αy + α0 y 0 ) = αφ̂(y) + α0 φ̂(y 0 ), φ̂(y)∗ = φ̂(y), (7.70)

0 0
[φ̂(y), φ̂(y )] = iy · ωy 1l. (7.71)

The advantage of the Weyl CCR C ∗ -algebra is that it is a true C ∗ -algebra,

hence it is prefered by numerous “C ∗ -algebras lovers”. It also allows us to con-
sider irregular representations. On the other hand, it consists of “almost periodic
functions on the phase space”, whose physical significance can be doubted.
The “field CCR ∗-algebra” seems “closer to the physics”. However, it is not
a C ∗ -algebra, and its representations involve unbounded operators, which could
be problematic.
The second step of quantization, that is fixing a representation, can be per-
formed by fixing a state, and then using the GNS representation. For instance,
the Fock state σ in the setting of Subsection 7.7 is given by
(z|z)
σ(Ŵ (z)) = e− 2 . (7.72)

The GNS representation wrt σ acts naturally in the bosonic Fock space, as
described in Subsection 7.7. The state σ corresponds to the Fock vacuum:
σ = (Ω| · Ω).

7.10 Positive energy Fock quantization

Consider a quadratic Hamiltonian on a phase space Y. If we fix coordinate
functions φi , it can be written as
1X
H= hij φi φj (7.73)
2 i,j

for some symmetric matrix [hij ]. As we discussed above, if Y is finite dimen-

sional, we could quantize H as
1X
Ĥ sym = hij φ̂i φ̂j . (7.74)
2 i,j

Because of a finite number of degrees of freedom this is well defined and essen-
tially unique, since all irreducible representations are equivalent, so that this
quantization is essentially unique.
If the number of degrees of freedom is infinite, usually only the phase space
with its symplectic structure is given beforehand. In the quantum theory one
also needs a CCR representation on a Hilbert space, which is less canonical and
more tricky to choose. How to select a physically motivated CCR representation,
if we are given a symplectic space?

60
 Suppose that the symplectic form is given by the matrix [ωij ], with its in-
verse denoted [ω ij ]. This is expressed by two equivalent identities, one for the
symplectic form, the other for the Poisson bracket:
X 1
ω= ωij φi ⊗ φj = ωij φi ∧ φj ; (7.75)
ij
2

{φi , φj } = ω ij . (7.76)

Introduce the operator k := ω −1 h, or

kji := ω ik hkj . (7.77)

For ζ ∈ R2n , note that kζ = ω −1 dH(ζ) is the Hamiltonian vector field generated
by H. Thus the dynamics generated by H is rt = etk . Thus on the classical
level we have the dynamics t 7→ φi (t) with φi (0) = φi is given by the linear
transformation
φi (t) = rt,j
i
φj . (7.78)
We would like to have a quantization, such that the analogous identity is true
on the quantum level. In other words, we would like to have a Hilbert space H
equipped with operators φ̂i and Ĥ such that

eitĤ φ̂i e−itĤ = rt,j

i
φ̂j . (7.79)

If the number of degrees of freedom is finite this is achieved by (7.74). We

can also subtract from (7.74) any real number and (7.79) will still hold.
Let us now assume that H is strictly positive. We will describe the whole
procedure as if the number of degrees of freedom were finite, so that Y = R2n ,
but it is easy to generalize it to the general case.
The matrix [hij ] defining the Hamiltonian can be used to fix a sesquilinear
scalar product on the complexification of the phase space, that is on C2n :
i
ζ hij ξ j =: hζ|hξi, (7.80)

Note that −ik is self-adjoint in the scalar product (7.80). Therefore, we can
diagonalize −ik in a basis orthonormal in (7.80). Note that all eigenvalues of −ik
are nonzero and real. Besides, if v i is an eigenvector of −ik with eigenvalue εi ,
then vi is an eigenvector with eigenvalue −εi , because −ik is purely imaginary.
Thus
X

hζ|hξi = hζ|hv i ihvi |hξi + hζ|hvi ihv i |hξi , (7.81)
i
X
−1
ε−1


ihζ|ωξi = hζ|h(−ik) ξi = i hζ i |hv i ihvi |hξi − hζ i |hvi ihv i |hξi ,
i

61
where we can assume that all εi are positive. Introduce the following functionals
acting on real vectors ζ:
1
hai |ζi := √ hvi |hζi, (7.82)
εi
1 1
ha∗i |ζi := √ hv i |hζi = √ hvi |hζi. (7.83)
εi εi

Then for real ζ, ξ (7.81) yields

X
hH|ζi = εi ha∗i |ζihai |ζi, (7.84)
i
X
ihζ|ωξi = hζ|h(−ik)−1 ξi = ha∗i |ζihai |ξi − hai |ζiha∗i |ξi ,


i

which can be rewritten as

X
H= εi a∗i ai , (7.85)
X
iω = a∗i ∧ ai , (7.86)

where the last line is equivalent to

{ai , a∗j } = −iδij , {ai , aj } = {a∗i , a∗j } = 0. (7.87)

Now we quantize the fields on a Hilbert space with the vector Ω, so that

[âi , â∗j } = δij , [âi , âj ] = [â∗i , â∗j ] = 0, âi Ω = 0. (7.88)

Clearly, what we obtain is the bosonic space with the 1-particle space spanned
by â∗i Ω. We choose the Hamiltonian
X
Ĥ = εi â∗i âi . (7.89)

sym
Note that Ĥ is positive, ĤΩ = 0 and
P (7.79) holds. If Ĥ is well defined, then
1
it differs from Ĥ by a constant 2 λi .
Let us now describe the above procedure in a basis independent way. Sup-
pose that H is a positive quadratic form on a real vector space Y equipped with
a symplectic form ω. We first extend H to a bilinear form on Y by polarization
identity. This form is generated by an operator from Y to the dual of Y, which
we denote by h

hζ|hξi := H(ζ + ξ) − H(ζ) − H(ξ), ζ, ξ ∈ Y. (7.90)

Equivalently, hζ = dH(ζ). Thus H(ζ) = 21 hζ|hζi and for k := ω −1 h, the

symplectic dynamics is rt = etk .
We extend h and iω as maps from CY to the dual of CY. We consider the
sesquilinear forms
hζ|hζi (7.91)

62
as a scalar product on CY called the energy scalar product. We easily check
that −ik is a self-adjoint operator on CY. We can diagonalize −ik. Because of
nondegeneracy of ω, this operator has a zero nullspace. Hence we have also a
nondegenerate sesquilinear form

ihζ|ωζi = hζ|h(−ik)−1 ζi

Let W (+) and W (−) be the positive and negative subspace of CY of the
self-adjoint operator −ik. We have

W (+) = W (−) , (7.92)

iω endows Z := W (+) with the positive scalar product

(z|z 0 ) = ihz|ωz 0 i (7.93)

and −ik is self-adjoint wrt (7.93). Let us set

 
Ĥ := dΓ − ik Z
. (7.94)

We have
eitĤ W (z)e−itĤ = W (etk z), z ∈ Z. (7.95)
The scalar product (7.93) will be treated as the basic one on Z and called
the dynamical scalar product. We have

hz|hz 0 i = (z| − ikz 0 ). (7.96)

Thus using (·|·) we can identify −ik restricted to Z with h restricted to Z.

The above construction guarantees that
(1) There exists a unitary group on the Hilbert space Γs (Z) that implements
the dynamics, see (7.95).
(2) The Hamiltonian Ĥ generating the dynamics is positive, because h is
Z
positive.
(3) The Fock vacuum Ω is a nondegenerate ground state of the Hamitonian.

7.11 Positive energy quantization for charged systems

Suppose that the symplectic space Y is equipped with a U (1) symmetry. More
precisely, we assume that it is Y = YR ⊕ YI and its coordinates are spanned by
φiR , φiI satisfying

{φiR , φjR } = ω ij , {φiI , φjI } = ω ij , {φiR , φjI } = 0, (7.97)

for some symplectic matrix ω on YR ' YI . The element eiθ ∈ U (1) acts on the
coordinates as

φiR 7→ cos θφiR − sin θφiI , φiR 7→ sin θφiR + cos θφiI . (7.98)

63
 In such a case it is customary to treat the space Y as a complex space
Y = CYR , setting
1 1
ψ i = √ (φiR + iφiI ), ψ i∗ = √ (φiR − iφiI ), (7.99)
2 2
so that
{ψ i , ψ j } = {ψ i∗ , ψ j∗ } = 0, {ψ i , ψ j∗ } = −ω ij , (7.100)
and the action of the group is ψ i 7→ eiθ ψ i , ψ i∗ 7→ e−iθ ψ i∗ .
Note that the space Y is equipped with complex conjugation c. It satisfies
cψ i c = ψ i∗ , c2 = 1l. It can be interpreted as the charge conjugation.
Suppose that H is a (real) quadratic Hamiltonian that is invariant wrt U (1).
Then it can be written as X
H= hij ψ i∗ ψ j (7.101)
ij

for some Hermitian matrix [hij ]. The symplectic form leads to the following
sesquilinear form, which will be called the charge:
X
Q=i ωij ψ i∗ ψ j (7.102)
ij

We diagonalize simultaneously H and Q obtaining I = I+ t I−

X
H= εi a∗i ai , (7.103)
i∈I
X X
Q= a∗i ai − a∗i ai , (7.104)
i∈I+ i∈I−

{ai , a∗j } = 0, i 6= j, {ai , a∗i } = ∓i, i ∈ I± . (7.105)

(Note that (7.104) and (7.105) are equivalent). We rename ai = b∗i , a∗i = bi for
i ∈ I− . Thus
X X
H= εi a∗i ai + εi b∗i bi , (7.106)
i∈I+ i∈I−
X X
Q= a∗i ai − b∗i bi , (7.107)
i∈I+ i∈I−

{ai , a∗j } = {bi , b∗j } = −iδij . (7.108)

Note that the space Y = Y+ ⊕ Y− .

The quantization of this system yields Γs (Y+ ⊕ cY− ). The Hamiltonian and
charge are
   
Ĥ = dΓ h ⊕ chc , Q̂ = dΓ 1l ⊕ −1l .
Y+ cY− Y+ cY−

The quantized charge conjugation is Ĉ := dΓ(c)

64
 If in addition the matrix [hij ] is real then we have

1X
hij φiR φjR + φiI φjI


H= (7.109)
2 ij

Clearly, then H = H ◦ c, Q = −Q ◦ c Then then we can choose the diagonalizing

functionals so that cai c = bi , ca∗i c = b∗i .
After quantization we have Ĉâi Ĉ = b̂i , Ĉâ∗i Ĉ = b̂∗i , Ĉ Ĥ Ĉ = Ĥ, Ĉ Q̂Ĉ = −Q̂.

8 Free neutral scalar bosons

8.1 Classical fields off-shell and on-shell
In the Lagrangian formalism of neutral field theory one starts from a field φ(x),
where x ∈ R1,3 , which can be treated as a linear functional on, say, real smooth
functions on R1,3 , such that

hφ(x)|f i := f (x), f ∈ C ∞ (R1,3 , R). (8.1)

(The choice of smooth functions does not matter much, since the Lagrangian
formalism serves mainly to obtain formal identities). One also chooses a local
Lagrangian density L(x), which is a function of the field φ(x), ∂µ φ(x) =: φ,µ (x)
and of x ∈ R1,3 . The Euler-Lagrange equation reads then

∂L(x)
∂φ(x) L(x) − ∂µ =0 (8.2)
∂φ,µ (x)

To obtain the Klein-Gordon equation, we use the Lagrangian density

L(x) = − 21 ∂µ φ(x)∂ µ φ(x) − 21 m2 φ(x)2 . (8.3)

The Euler-Lagrange equation then yields

(−2 + m2 )φ(x) = 0. (8.4)

Let ζ ∈ C ∞ (R1,3 ) solve the Klein-Gordon equation

(−2 + m2 )ζ(x) = 0. (8.5)

Recall the identity (4.33):

Z Z
PJ
ζ(t, ~x) = Ġ (t, ~x − ~y )ζ(0, ~y )d~y + GPJ (t, ~x − ~y )ζ̇(0, ~y )d~y . (8.6)
R3 R3

Recall that a function on R1,3 is called space compact if there exists a com-
pact K ⊂ R1,n such that suppf ⊂ J(K). The set of space compact smooth
∞
functions will be denoted Csc (R1,3 ). Therefore all solutions of (8.5) with com-
pactly supported Cauchy data are space compact.

65
 Let YKG denote the space of real, resp. space-compact solutions of the Klein-
Gordon equation We can endow the space YKG with the standard topology of
Cc∞ (R3 ) ⊕ Cc∞ (R3 ) given by the initial conditions. The space of real continuous
T T
functionals on YKG will be denoted by YKG . The action of T ∈ CYKG on
ζ ∈ YKG will be denoted by hT |ζi, and sometimes simply by T ζ.
The field φ(x) understood as (8.1) will be called an off-shell field. It is used
in the Lagrangian formalism. When we go from the Lagrangian to Hamiltonian
formalism, we enforce the on-shell condition, that is, we restrict ourselves to
solutions of the E-L equation. We are also more careful in the choice of the
space on which the fields act. We restrict ourselves to YKG . Thus, for x ∈ R1,3 ,
the on-shell field φ(x) acting on ζ ∈ YKG gives

hφ(x)|ζi := ζ(x),

We will not distinguish the notation for on-shell and off-shell fields.
Clearly, for any ζ ∈ YKG we have

(−2 + m2 )hφ(x)|ζi = 0.

Thus the equation (8.4) for on-shell fields is a tautology.

In the on-shell formalism we also introduce the variable conjugate to φ(x):

∂L(x)
π(x) := = φ,0 (x) = φ̇(x). (8.7)
∂φ,0 (x)

Clearly, φ̇(x) = π(x) and by (8.6)

Z Z
φ(t, ~x) = ĠPJ (t, ~x − ~y )φ(0, ~y )d~y + GPJ (t, ~x − ~y )π(0, ~y )d~y . (8.8)

The Poincaré group R1,3 o O(1, 3) acts on YKG by

r(y,Λ) ζ(x) := ζ (y, Λ)−1 x .

Equivalently,
T−1
r(y,Λ) φ(x) = φ(Λx + y). (8.9)

8.2 Symplectic form and Poisson bracket

∞
For ζ1 , ζ2 ∈ Csc (R1,3 ) we define

j µ (x, ζ1 , ζ2 ) := ∂ µ ζ1 (x)ζ2 (x) − ζ1 (x)∂ µ ζ2 (x). (8.10)

Writing j µ (x) for brevity, we easily check that

∂µ j µ (x) = (2 − m2 )ζ1 (x)ζ2 (x) − ζ1 (x)(2 − m2 )ζ2 (x),

66
This implies Green’s identity
Z Z
j (t+ , ~x)d~x − j 0 (t− , ~x)d~x
0
(8.11)
Z  
= (−2 + m2 )ζ1 (x)ζ2 (x) − ζ1 (x)(−2 + m2 )ζ2 (x) dx. (8.12)
t− <x0 <t+

Thus if ζ1 , ζ2 ∈ YKG , then

∂µ j µ (x) = 0.
One says that j µ (x) is a conserved 4-current.
A space-like subspace of codimension 1 will be called a Cauchy subspace.
The flux of j µ across any Cauchy subspace S does not depend on its choice. It
defines a symplectic form on YKG
Z
ω(ζ1 , ζ2 ) = j µ (x, ζ1 , ζ2 )dsµ (x)
S
Z  
= −ζ̇1 (t, ~x)ζ2 (t, ~x) + ζ1 (t, ~x)ζ̇2 (t, ~x) d~x. (8.13)

Thus (YKG , ω) is an (infinite dimensional) symplectic space.

Clearly, the form (8.13) is well defined also if only ζ2 ∈ YKG , and ζ1 is a
distributional solution of the Klein-Gordon equation.
r(y,Λ) are symplectic (preserve the symplectic form) for Λ ∈ O↑ (1, 3), other-
wise they are antisymplectic (change the sign in front of the symplectic form).
By (8.13), the symplectic form can be written as
Z


ω(ζ1 , ζ2 ) = − hπ(t, ~x)|ζ1 ihφ(t, ~x)|ζ2 i + hφ(t, ~x)|ζ1 ihπ(t, ~x)|ζ2 i d~x,

or more simply, Z
ω= φ(t, ~x) ∧ π(t, ~x)d~x. (8.14)

The conserved 4-current can be written as

jµ (x) = φ(x) ∧ ∂µ φ(x).

By (8.14), the symplectic structure on the space YKG leads to the Poisson
bracket

{φ(t, ~x), φ(t, ~y )} = {π(t, ~x), π(t, ~y )} = 0,

{φ(t, ~x), π(t, ~y )} = δ(~x − ~y ). (8.15)

Using (8.8) we obtain

{φ(x), φ(y)} = −GPJ (x − y). (8.16)

Therefore, the Pauli-Jordan solution is often called the commutator function.

67
 The relations (8.15) can be viewed as mnemotechnic identities that yield the
correct Poisson bracket for more regular functions, eg. the smeared out fields
Z
φ[f ] := f (x)φ(x)dx. (8.17)

We have Z Z
{φ[f ], φ[g]} = − f (x)g(x)GPJ (x − y)dxdy. (8.18)

Note that formally φ(t, ~x) and π(t, ~x) generate the algebra of all functions
on YKG .

8.3 Stress-energy tensor

The Noether Theorem suggests to introduce the stress-energy tensor

∂L(x) ,ν
T µν (x) := − φ (x) + g µν L(x) (8.19)
∂φ,µ (x)
1
∂ µ φ(x)∂ ν φ(x) − g µν ∂α φ(x)∂ α φ(x) + m2 φ(x)2 .


=
2
It is easy to check that the stress-energy tensor is conserved on solutions of
the Klein-Gordon equation (on shell):

∂µ T µν (x) = 0.

We express the stress-energy tensor in terms of φ(x) and π(x). Its compo-
nents with the first temporal coordinate are called the Hamiltonian density and
momentum density:
1 ~

2 
H(x) := T 00 (x) = π(x)2 + ∂φ(x) + m2 φ(x)2 ,
2
P i (x) := T 0i (x) = −π(x)∂ i φ(x).

They are examples of quadratic functionals on YKG : One easily checks

{φ(t, ~x), H(t, ~y )} = φ̇(t, ~x)δ(~x − ~y ), (8.20)

{π(t, ~x), H(t, ~y )} = π̇(t, ~x)δ(~x − ~y ), (8.21)
i i
{φ(t, ~x), P (t, ~y )} = −∂ φ(t, ~x)δ(~x − ~y ), (8.22)
i i
{π(t, ~x), P (t, ~y )} = −∂ π(t, ~x)δ(~x − ~y ). (8.23)

We introduce the (total) Hamiltonian and momentum:

Z Z
H := T µ0 (x)dsµ (x) = H(t, ~x)d~x, (8.24)
ZS Z
i
P := T µi (x)dsµ (x) = P i (t, ~x)d~x. (8.25)
S

68
where S is any Cauchy subspace. They are examples of quadratic functionals:
Z
1 ~

2 
hH|ζi = ζ̇(t, ~x)2 + ∂ζ(t, ~x) + m2 ζ(t, ~x)2 ,
2
Z
hP i |ζi = − ζ̇(t, ~x)∂ i ζ(t, ~x).

H and P~ are the generators of the time and space translations:

φ̇(x) = {φ(x), H}, π̇(x) = {π(x), H},
~
∂φ(x) ~
= −{φ(x), P~ }, ∂π(x) = −{π(x), P~ }.
The observables H, P 1 , P 2 and P 3 are in involution. (This means that the
Poisson bracket of every pair among these observables vanishes).

8.4 Simultaneous diagonalization of the symplectic form,

Hamiltonian and momentum
Let us stress that the space YKG is real, which reflects the fact that in this
section we consider neutral fields. It is however useful to complexify the space
YKG , that is to consider the space of smooth space-compact complex solutions
of the Klein-Gordon equation. A possible notation for this space is CYKG , but
we will also use a different letter WKG := CYKG .
We multiply the current (8.10) by i and extend it by sesquilinearity to CYKG ,
obtaining the Hermitian form
ij µ (x, ζ 1 , ζ2 ) := i ∂ µ ζ 1 (x)ζ2 (x) − ζ 1 (x)∂ µ ζ2 (x) .


(8.26)
After integrating on a Cauchy surface we obtain the Hermitian form
Z


iω(ζ 1 , ζ2 ) = i − ζ̇1 (t, ~x)ζ2 (t, ~x) + ζ1 (t, ~x)ζ̇2 (t, ~x) d~x. (8.27)

We also extend by sesquilinearity to CYKG the Hamiltonian and the mo-

mentum:
Z
1 ~
|ζ̇(t, ~x)|2 + |∇ζ(t, ~x)|2 + m2 |ζ(t, ~x)|2 d~x,


hH|ζi = (8.28)
2
Z
1

hPj |ζi = − ζ̇(t, ~x)∇j ζ(t, ~x) + ∇j ζ(t, ~x)ζ̇(t, ~x) d~x. (8.29)
2
We are going to diagonalize simultaneously
p (8.27), (8.28) and (8.29).
~ 3 ~ ~ 2
For k ∈ R , set ε = ε(k) := k + m . 2

Every ζ ∈ CYKG can be written in a unique way as

ζ = ζ (+) + ζ (−) , (8.30)
where
Z
ζ (±)
(x) = ζ (±) (k)| ± k)d~k. (8.31)

69
where we smear out ζ (±) (k) and with “positive”, resp. “negative frequency
plane waves”
1 ~ 0 ~
|k) = p q ei(−ε(k)x +k~x) , (8.32)
(2π)3 2ε(~k)
1 ~ 0 ~
| − k) = |k) = p q e−i(−ε(k)x +k~x) . (8.33)
(2π)3 2ε(~k)
We obtain
Z Z
(+) (+) (−) (−)
iω(ζ1 , ζ2 ) = ζ1 (k)ζ2 (k)d~k − ζ1 (k)ζ2 (k)d~k (8.34)
Z
1
ε(~k) ζ (+) (k)ζ (+) (k) + ζ (−) (k)ζ (−) (k) d~k


hH|ζi = (8.35)
2
Z
1
i
k i ζ (+) (k)ζ (+) (k) + ζ (−) (k)ζ (−) (k) d~k.


hP |ζi = (8.36)
2

8.5 Plane wave functionals

T
If T ∈ CYKG , we define T ∗ ∈ CYKG
T
by
hT ∗ |ζi := hT |ζi, ζ ∈ YKG .
Note that in this context the star does not denote the Hermitian conjugation
(which in our text is the standard meaning of the star).
Let k := (ε(~k), ~k). k ∈ R1,3 of this form will be called on shell. Recall that
YKG is a real space and therefore if ζ ∈ YKG , then ζ (+) = ζ (−) . For k on shell,
we define plane wave functionals a(k), a∗ (k) as functionals on the real space
YKG by
ha(k)|ζi := ζ (+) (k), ha∗ (k)|ζi := ζ (−) (k) = ha(k)|ζi (8.37)
Clearly, from hφ(x)|ζi = ζ(x), we obtain
d~k
Z
eikx a(k) + e−ikx a∗ (k) ,


φ(x) = q
(2π)3 2ε(~k)
p
q
Z d~k ε(~k)
√ eikx a(k) − e−ikx a∗ (k) .


π(x) = p
i (2π)3 2
After setting x0 = 0, we can invert these relations:
s !
ε(~k)
Z
d~x ~
−ik~ x i
a(k) = p e φ(0, ~x) + q π(0, ~x) , (8.38)
(2π)3 2
2ε(~k)
s !
ε(~k)
Z
∗ d~x i~
k~x i
a (k) = p e φ(0, ~x) − q π(0, ~x) . (8.39)
(2π)3 2
2ε(~k)

70
 (8.34) can be rewritten with ζ1 , ζ2 ∈ YKG as
Z  
iω(ζ1 , ζ2 ) = ha(k)|ζ1 iha(k)|ζ2 i − ha(k)|ζ1 iha(k)|ζ2 i d~k. (8.40)

Rewriting it in a shorter form we see that a(k), a∗ (k) diagonalize the symplectic
form:
Z
iω = d~ka∗ (k) ∧ a(k). (8.41)

They also diagonalize simultaneously the Hamiltonian and the momentum:

Z
H = d~kε(~k)a∗ (k)a(k), (8.42)
Z
P~ = d~k~ka∗ (k)a(k). (8.43)

(8.41) is equivalent to

{a(k), a(k 0 )} ={a∗ (k), a∗ (k 0 )} = 0, (8.44)

{a(k), a∗ (k 0 )} = − iδ(~k − ~k 0 ). (8.45)

Hence,

{a(k), H} = −iε(~k)a(k), {a∗ (k), H} = iε(~k)a∗ (k), (8.46)

{a(k), P~ } = −i~ka(k), {a∗ (k), P~ } = i~ka∗ (k), (8.47)

8.6 Positive frequency space

(8.30) gives a decomposition of the space CYKG into two subspaces
(+) (−)
CYKG = WKG ⊕ WKG . (8.48)
(+) (+) (+)
(8.34) restricted to WKG is positive definite. For ζ1 , ζ2 ∈ W (+) we will
write
(+) (+) (+) (+)
(ζ1 |ζ2 ) := iζ1 ωζ2 . (8.49)
(+)
The Hilbert space of positive energy solutions is denoted ZKG , and is the com-
(+) (+)
pletion of WKG in this scalar product. ZKG can be identified with L2 (R3 ),
(8.49) rewritten as
Z
(+) (+)
(ζ1 |ζ2 ) = ζ (+) (k)ζ (+) (k)d~k. (8.50)

and ζ (+) (k) = (k|ζ (+) ).

71
 We can use f ∈ ZKG to smear out the functionals a(k) and a∗ (k):
Z Z
ha(f )|ζi = f (k)ζ (+) (k)d~k, a(f ) = f (k)a(k)d~k; (8.51)
Z Z
∗ (+)
ha (f )|ζi = f (k)ζ (k)dk, ~ a (f ) = f (k)a∗ (k)d~k;
∗
(8.52)

{a(f ), a(f 0 )} ={a∗ (f ), a∗ (f 0 )} = 0, {a(f ), a∗ (f 0 )} = −i(f |f 0 ). (8.53)

Using the “smeared notation” on the right we can write
a∗ (k) = a∗ |k) .


(8.54)
(−)
In this section we will not use WKG for quantization, however we will do
this when we consider charged fields. Anticipating our discussion of charged
(−)
fields of the next section, we introduce the space complex conjugate to WKG ,
(−)
denoted WKG and equipped with the scalar product
(−) (−) (−) (−)
(ζ 1 |ζ 2 ) := iζ1 ωζ2 . (8.55)
(−) (−)
We set ZKG to be the completion of WKG in this scalar product. (The bar is
(−)
again the complex conjugation). ZKG can be identified with L2 (R3 ) and (8.55)
rewritten as
Z
(−) (−) (−) (−)
(ζ 1 |ζ 2 ) = ζ1 (k)ζ2 (k)d~k

(−)
and ζ (−) (k) = (−k|ζ ).
(−) (+)
Note that WKG = WKG , where we use the usual (internal) complex con-
(−) (+)
jugation in WKG . Therefore in principle we could identify ZKG and ZKG . In
particular, with this identification
| − k) = |k). (8.56)
This identification is consistently applied in this section, however in the next
(−) (+)
section we treat ZKG and ZKG as two separate Hilbert spaces.
(+) (−)
R1,3 o O↑ (1, 3) acts on ZKG and ZKG in a natural way.
(+)
We have a natural identification of YKG with WKG . Indeed, ζ ∈ YKG can be
(+)
projected onto ζ (+) ∈ WKG , as in (8.30).This identification allows us to define
a real scalar product on YKG :
(+) (+)
hζ1 |ζ2 iY := Re(ζ1 |ζ2 ).
We can compute explicitly this scalar product:
Z Z
hζ1 |ζ2 iY = ζ̇1 (0, ~x)G(+) (0, ~x − ~y )ζ̇2 (0, ~y )d~xd~y (8.57)
Z Z
+ ζ1 (0, ~x)(−∆~x + m2 )G(+) (0, ~x − ~y )ζ2 (0, ~y )d~xd~y .

72
8.7 Quantization of scalar fields
There are several equivalent presentations of free scalar quantum fields.
The description in typical physics textbooks can be described more or less
as follows. We want to construct H, Ĥ, Ω such that H is a positive self-adjoint
operator on H, Ω is a normalized eigenvector of H with eigenvalue 0 and a
self-adjoint operator valued distribution
R1,3 3 x 7→ φ̂(x), (8.58)
˙
such that, with π̂(x) := φ̂(x),
(1) (−2 + m2 )φ̂(x) = 0,
(2) [φ̂(0, ~x), φ̂(0, ~y )] = [π̂(0, ~x), π̂(0, ~y )] = 0,
[φ̂(0, ~x), π̂(0, ~y )] = iδ(~x − ~y ).
(3) eitĤ φ̂(x0 , ~x)e−itĤ = φ̂(x0 + t, ~x).
(4) Ω is cyclic for φ̂(x).
Let us describe quantum scalar fields following the above strategy, as an
(essentially unique) solution of the above problem. Let R1,3 3 x 7→ φ̂(x), π̂(x)
satisfy (1). Then the Fourier transform of φ̂ has to be supported on the mass
hyperboloid. Therefore, by the same argument as in the classical case we can
intorduce â∗ (k) and â(k) such that

d~k
Z
eikx â(k) + e−ikx â∗ (k) ,


φ̂(x) = q (8.59)
(2π)3 2ε(~k)
p
q
Z d~k ε(~k)
√ eikx â(k) − e−ikx â∗ (k) ,


π̂(x) = p (8.60)
i (2π)3 2
with the inverse transformation
s !
ε(~k)
Z
d~x −i~
k~x i
â(k) = p e φ(0, ~x) + q π(0, ~x) , (8.61)
(2π)3 2 ~
2ε(k)
s !
ε(~k)
Z
d~x ~ i
â∗ (k) = p eik~x φ̂(0, ~x) − q π̂(0, ~x) . (8.62)
(2π)3 2
2ε(~k)

(These are identities (8.38) and (8.39) decorated with hats). Again, repeating
the classical arguments, (2) implies
[â(k), â(k 0 )] =[â∗ (k), â∗ (k 0 )] = 0, (8.63)
[â(k), â∗ (k 0 )] =δ(~k − ~k 0 ). (8.64)
We still need the Hilbert space and the Hamiltonian. Since we have an
infinite number of degrees of freedom we cannot use the symmetric quantization

73
to define Ĥ, because this would produce an infinite constant. Differentiating
(3) wrt time we obtain
i[Ĥ, φ̂(x)] = π̂(x). (8.65)
This is equivalent to

[â(k), Ĥ] = ε(~k)â(k), [â∗ (k), Ĥ] = −ε(~k)a∗ (k), (8.66)

ĤΩ = 0 implies Ĥâ(k)Ω = −ε(~k)â(k)Ω. But Ĥ ≥ 0. Thus we should

assume
â(k)Ω = 0. (8.67)
By (4), Ω is cyclic for â(k) and â∗ (k). Using the commutation relations and
(8.67) we see that Ω is cyclic just for â∗ (k). In other words, H is spanned by
vectors of the form
Z
Ψ = Ψ(~k1 , . . . , ~kn )â∗ (k1 ) · · · â∗ (kn )Ωd~k1 · · · d~kn .

From the commutation relations and (8.67) we obtain

Z
(Ψ|Ψ ) = n! Ψ(~k1 , . . . , ~kn )Ψ0 (~k1 , . . . , ~kn )d~k1 · · · d~kn .
0

This is exactly the scalar product for Γs (L2 (R3 )).

Besides the Hamiltonian Ĥ satisfying (8.66), we also want the momentum
~
operator P̂ , which satisfies
~ ~
[â(k), P̂ ] = ~kâ(k), [â∗ (k), P̂ ] = −~kâ∗ (k). (8.68)
~
This fixes Ĥ and P̂ up to a constant. We choose the normal ordered form for
these operators, which guarantees that they annihilate Ω:
Z
Ĥ := â∗ (k)â(k)ε(~k)d~k,
Z
~
P̂ := â∗ (k)â(k)~kd~k.

(+)
Recall that L2 (R3 ) coincides with ZKG , the

completion of WKG . Thus the
Hilbert space H can be identified with Γs ZKG , Ω with the Fock vacuum, â∗ (k)
with the creation operators in the “physicist’s notation”.
As usual, we can also introduce the smeared versions of (8.61), (8.62) for
f ∈ ZKG :
Z
â(f ) = f (k)â(k)d~k; (8.69)
Z
â∗ (f ) = f (k)â∗ (k)d~k; (8.70)

[â(f ), â(f 0 )] = [â∗ (f ), â∗ (f 0 )] = 0, [â(f ), â∗ (f 0 )] = (f |f 0 ). (8.71)

74
Using the “smeared notation” on the right we can write

â∗ (k) = â∗ |k) .

(8.72)
(+)
The group R1,3 o O↑ (1, 3) acts on WKG , and hence on ZKG by unitary
transformations r(y,Λ) . It acts unitarily also on Γs (ZKG ) by U (y, Λ) :=
  ZKG
Γ r(y,Λ) . On classical fields the Poincare group acts by φ(Λx + y), see
ZKG
(8.9). On quantum fields the analog of this action is unitarily implemented:

U (y, Λ)φ̂(x)U (y, Λ)∗ = φ̂ (y, Λ)x .

This is true even though we only required that time translations are imple-
mented.
One of possible alternative presentations of quantization of the free scalar
field in the mathematical style goes as follows. We have the symplectic space
(YKG , ω). This symplectic space is equipped with a symplectic dynamics rt
generated by a positive classical Hamiltonian H. We would like to find a CCR
representation
YKG 3 ζ 7→ W (ζ) ∈ U (H) (8.73)
We want the quantum Hamiltonian to be compatible with the classical Hamil-
tonian, so that
W (rt (ζ)) = eitĤ W (ζ)e−itĤ . (8.74)
We assume that the representation is Fock. Then we apply the method of a
positive energy representations, as described in Subsection 7.10.

8.8 Two-point functions

Introduce the time-ordering operations for space-time dependent operators:

T A(x)B(y) = θ(x0 − y 0 )A(x)B(y) + θ(y 0 − x0 )B(y)A(x),

(8.75)
0 0 0 0


T A(x)B(y) = θ(y − x )A(x)B(y) + θ(x − y )B(y)A(x). (8.76)

Note the identities

[φ̂(x), φ̂(y)] = −iGPJ (x − y)1l, (8.77a)

(+)
(Ω|φ̂(x)φ̂(y)Ω) = G (x − y), (8.77b)
F
(Ω|T(φ̂(x)φ̂(y))Ω) = −iG (x − y), (8.77c)
F
(Ω|T(φ̂(x)φ̂(y))Ω) = iG (x − y). (8.77d)

75
 In fact,

d~kd~k 0
Z Z
0
(Ω|φ̂(x)φ̂(y)Ω) = √ √ eikx−ik y (Ω|â(k)â∗ (k 0 )Ω)
(2π)3 2ε 2ε0
d~k
Z
= eik(x−y)
(2π)3 2ε(~k)
= G(+) (x − y);
(Ω|T(φ̂(x)φ̂(y))Ω) = θ(x0 − y 0 )(Ω|φ̂(x)φ̂(y)Ω) + θ(y 0 − x0 )(Ω|φ̂(y)φ̂(x)Ω)
= θ(x0 − y 0 )G(+) (x − y) + θ(y 0 − x0 )G(−) (x − y)
= −iGF (x − y),

where at the end we used (4.19).

Differentiating if needed (8.77b) with respect time we obtain the equal time
correlation functions expressed as real symmetric kernels:

(Ω|φ̂(0, ~x)φ̂(0, ~y )Ω) = G(+) (0, ~x − ~y ), (8.78)

(Ω|φ̂(0, ~x)π̂(0, ~y )Ω) = 0, (8.79)
(Ω|π̂(0, ~x)π̂(0, ~y )Ω) = −∂t2 G(+) (0, ~x − ~y )
= (−∆~x + m2 )G(+) (0, ~x − ~y ). (8.80)

8.9 Spacetime smeared fields

For f ∈ Cc∞ (R1,3 , R) set
Z
φ̂[f ] := f (x)φ̂(x)dx. (8.81)

The operators φ̂[f ] are essentially self-adjoint for f ∈ Cc∞ (O, R) on, say, smooth
vectors in the Fock space with compact supports. Therefore, we can define
W (f ) := eiφ̂[f ] .
We have
Z Z
[φ̂[f ], φ̂[g]] = −i f (x)g(x)GPJ (x − y)dydy. (8.82)

In particular, if supp(f ) × supp(g), then φ̂[f ] and φ̂[g] commute.

(8.77b) implies the following identities for spacetime smeared fields and Weyl
operators:
Z Z
2
(Ω|φ̂[f ] Ω) = f (x)G(+) (x − y)f (y)dxdy, (8.83)
 Z Z 
1
(Ω|eiφ̂[f ] Ω) = exp − f (x)G(+) (x − y)f (y)dxdy . (8.84)
2

(8.81) satisfy the Wightman axioms with D := Γfin

s (ZKG ).

76
 For an open set O ⊂ Rd we set
A(O) := {exp(iφ̂[f ]) : f ∈ Cc∞ (O, R)}00 .
The algebras A(O) satisfy the Haag-Kastler axioms.

8.10 Physical meaning of the 2-point functions and Feyn-

man propagators
Let us now describe Gedankenexperiments measuring the 2-point function and
the Feynman propagator.
First let us note a general fact. Suppose that we are able to create a state
Φ ∈ H, kΦk = 1. On the measurement side, let us select two non-parallel vectors
Ψi ∈ H, i = 1, 2. Suppose for any i, j we can measure
Ai := |Ψi )(Ψi | + |Ψi )(Ψi |, i = 1, 2, (8.85)
B := |Ψ1 )(Ψ2 | + |Ψ2 )(Ψ1 |, C := i|Ψ1 )(Ψ2 | − i|Ψ2 )(Ψ1 |. (8.86)
2
Measuring Ai we obtain (Φ|Ai Φ) = |(Φ|Ψi )| , i = 1, 2. Measuring B, C, we
can determine (Φ|Ψ 1)
(Φ|Ψ2 ) . Thus from these measurements we can determine the
amplitudes (Ψi |Φ), i = 1, 2 up to an overall phase factor.
Gedankenexperiment 1. Let us choose spacetime functions f, gi . Suppose
we prepare the state given by Φ := φ̂[f ]Ω. We can assume that it is normalized.
Set Ψi := φ̂[gi ]Ω. Then up to an overall phase factor we are able to measure
the amplitudes
Z
(Ψi |Φ) = (Ω|φ̂(y)φ̂(x)Ω)gi (y)f (x)dxdy, (8.87)

which can be expressed in terms of the 2-point function.

Gedankenexperiment 2. Suppose now that we can perturb the dynamics
by adding to the Lagrangian −λf (x)φ(x). Then the interaction Hamiltonian
becomes Z
ĤInt (t) = λ f (t, ~x)φ̂(t, ~x)d~x, (8.88)

where φ̂(x) are the free fields. Suppose we measure the vaccum–vacuum ampli-
tude. The resulting quantity is
  Z  
Ω|Texp − i ĤInt (t)dt Ω (8.89)
∞
X Z Z
= (−i)n ··· (Ω|ĤInt (tn ) · · · ĤInt (t1 )Ω)dtn · · · dt1 (8.90)
n=0 tn >···>t1

λ2 Z Z 
= exp − f (x)f (y)(Ω|T{φ̂(x)φ̂(y)}Ω)dxdy , (8.91)
2
which is expressed in terms of the Feynman propagator.
Note that the second scenario is probably more realistic. Thus one can argue
that the Feynman propagator is more physical than the 2-point function.

77
9 Free charged scalar bosons
The formalism used in physics to describe complex fields, and especially to
quantize them, is different from the real case, therefore we devote to it a separate
section.

9.1 Lagrangian formalism

Consider the space of complex smooth functions on the spacetime C ∞ (R1,3 ) =
C ∞ (R1,3 , C). Clearly, the space C ∞ (R1,3 ) is equipped with a complex conjuga-
tion f 7→ f and a U (1) symmetry f 7→ eiθ f , θ ∈ R/2πZ = U (1).
If T is a real linear functional on C ∞ (R1,3 ), then we have two kinds of natural
complex conjugations of T :

hT |ζi := hT |ζi, hT ∗ |ζi := hT |ζi. (9.1)

Both maps T 7→ T and T 7→ T ∗ are antilinear. When restricted to the real

subspace YKG ⊂ C ∞ (R1,3 ), the functionals T and T ∗ coincide. (Note that here
∗ does not denote the Hermitian conjugaton!)
A special role is played by complex linear functionals on C ∞ (R1,3 ). The
space of such functionals will be denoted C ∞ (R1,3 )T . If T ∈ C ∞ (R1,3 )T , then
T ∈ C ∞ (R1,3 )T , unlike T ∗ , which is antilinear.
Let ψ(x), ψ ∗ (x) be the linear functionals on C ∞ (R1,3 )

hψ(x)|f i := f (x), hψ ∗ (x)|f i := f (x).

If L(x) is a Lagrangian density, which is a function of x, ψ(x), ψ ∗ (x), ψ,µ (x)

∗
and ψ,µ (x), then the Euler-Lagrange equations read
∂L
∂ψ∗ L − ∂µ ∗
= 0, (9.2)
∂ψ,µ
∂L
∂ψ L − ∂µ = 0. (9.3)
∂ψ,µ
If L(x) is real, then (9.2) implies (9.3).
We consider the Lagrangian density

L(x) = −∂µ ψ ∗ (x)∂ µ ψ(x) − m2 ψ ∗ (x)ψ(x). (9.4)

Then the Euler-Lagrange equations are equivalent to the (complex) Klein-Gordon

equation:
(−2 + m2 )ψ(x) = 0. (9.5)
The variables conjugate to ψ(x) and ψ ∗ (x) are
∂L
η ∗ (x) := = ∂0 ψ ∗ (x),
∂ψ,0 (x)
∂L
η(x) := ∗ (x) = ∂0 ψ(x).
∂ψ,0

78
 WKG will denote the space of smooth space-compact complex solutions of the
Klein-Gordon equation
(−2 + m2 )ζ = 0. (9.6)
(In the context of neutral fields, it was denoted CYKG , because it was an aux-
iliary object, the complexification of the phase space YKG . Now it is the basic
object, the phase space itself). In the on-shell formalism we will consider ψ(x),
ψ ∗ (x), η ∗ (x) and η(x) as functionals on WKG .
(8.6) implies
Z Z
ψ(t, ~x) = ĠPJ (t, ~x − ~y )ψ(0, ~y )d~y + GPJ (t, ~x − ~y )η(0, ~y )d~y . (9.7)

9.2 Charged fields as a pair of neutral fields

Let us go back to the Lagrangian formalism. For x ∈ R1,3 let us introduce the
fields φR (x), φI (x) as the functionals on C ∞ (R1,3 ) given by
√ √
hφR (x)|f i := 2Ref (x), hφI (x)|f i := 2Imf (x). (9.8)

or equivalently
1 1
ψ ∗ (x) = √ φR (x) − iφI (x) .



ψ(x) = √ φR (x) + iφI (x) ,
2 2
Clearly, the Lagrangian density can be rewritten as
1 1
L(x) = − ∂µ φR (x)∂ µ φR (x) − m2 φR (x)2 (9.9)
2 2
1 1
− ∂µ φI (x)∂ µ φI (x) − m2 φI (x)2 . (9.10)
2 2
The usual real formalism yields a pair of neutral fields with the usual equal time
Poisson brackets (we write only the non-vanishing ones):

{φR (t, ~x), πR (t, ~y )} = {φI (t, ~x), πI (t, ~y )} = δ(~x − ~y ). (9.11)

The fields with an additional symmetry

φθR = cos θφR − sin θφI , (9.12)

φθI = sin θφR + cos θφI . (9.13)

The equal-time Poisson brackets, which follow from (9.11) are

{ψ(t, ~x), η ∗ (t, ~y )} = {ψ ∗ (t, ~x), η(t, ~y )} = δ(~x − ~y ). (9.14)

(We write only non-vanishing ones). Using (9.7) we obtain

{ψ(x), ψ(y)} = {ψ ∗ (x), ψ ∗ (y)} = 0,

∗
{ψ(x), ψ (y)} = −GPJ (x − y).

79
9.3 Classical 4-current
The Lagrangian is invariant w.r.t. the U (1) symmetry ψ 7→ e−iθ ψ. The Noether
4-current associated to this symmetry is called simply the 4-current. It is
 ∂L(x) ∂L(x) 
J µ (x) := i ψ ∗ (x) ∗
− ψ(x)
∂ψ,µ ∂ψ,µ
= i ∂ ψ (x)ψ(x) − ψ ∗ (x)∂ µ ψ(x) .
µ ∗

It is conserved on shell and real:

∂µ J µ (x) = 0,
µ ∗
J (x) = J µ (x).

Up to a coefficient, it coincides with the current (8.26) evaluated at ζ = ζ1 = ζ2 .

hJ µ (x)|ζi = ij µ (ζ, ζ, x)
i ∂ µ ζ(x)ζ(x) − ζ(x)∂ µ ζ(x) .


=

The 0th component of the 4-current is called the charge density

Q(x) := J 0 (x) = i −η ∗ (x)ψ(x) + ψ ∗ (x)η(x) .

We have the relations

{Q(t, ~x), ψ(t, ~y )} = iψ(t, ~y )δ(~x − ~y ),

{Q(t, ~x), η(t, ~y )} = iη(t, ~y )δ(~x − ~y ),
{Q(t, ~x), Q(t, ~y )} = 0. (9.15)

The (total) charge Z

Q := Q(t, ~x)d~x

is conserved (does not depend on time) and coincides with the quadratic form
obtained from (8.27):
hQ|ζi = iζωζ. (9.16)

9.4 Stress-energy tensor

The Lagrangian is invariant w.r.t. space-time translations. This leads to the
stress-energy tensor
∂L(x) ν ∂L(x)
T µν (x) := − ∂ ψ(x) − ∂ ν ψ ∗ (x) ∗ + g µν L(x)
∂ψ,µ (x) ∂ψ,µ (x)
= ∂ µ ψ ∗ (x)∂ ν ψ(x) + ∂ ν ψ ∗ (x)∂ µ ψ(x)
−g µν ∂α ψ ∗ (x)∂ α ψ(x) + m2 ψ ∗ (x)ψ(x) .

It is conserved on shell
∂µ T µν (x) = 0.

80
The components of the stress-energy tensor with the first temporal coordinate
are called the Hamiltonian density and momentum density. We express them
on-shell in terms of ψ(x), ψ ∗ (x), η(x) and η ∗ (x):

H(x) := T 00 (x) ~ ∗ (x)∂ψ(x)

= η ∗ (x)η(x) + ∂ψ ~ + m2 ψ ∗ (x)ψ(x),
i
P (x) := T (x)0i
= −η ∗ (x)∂ i ψ(x) − ∂ i ψ ∗ (x)η(x).
~
H(x) and P(x) acting on ζ ∈ WKG yield
~
hH(x)|ζi = |ζ̇(x)|2 + |∂ζ(x)| 2
+ m2 |ζ(x)|2 ,
~
hP(x)|ζi ~
= −ζ̇(x)∂ζ(x) ~
− ∂ζ(x)ζ̇(x).

We easily check

{ψ(t, ~x), H(t, ~y )} = ψ̇(t, ~x)δ(~x − ~y ), (9.17)

{η(t, ~x), H(t, ~y )} = η̇(t, ~x)δ(~x − ~y ), (9.18)
i i
{ψ(t, ~x), P (t, ~y )} = −∂ ψ(t, ~x)δ(~x − ~y ), (9.19)
i i
{η(t, ~x), P (t, ~y )} = −∂ η(t, ~x)δ(~x − ~y ). (9.20)

We introduce the (total) Hamiltonian and momentum:

Z Z
H := T µ0 (x)dsµ (x) = H(t, ~x)d~x, (9.21)
ZS Z
i
P := T µi (x)dsµ (x) = P i (t, ~x)d~x. (9.22)
S

where S is any Cauchy subspace.

H and P~ are the generators of the time and space translations:

ψ̇(x) = {ψ(x), H}, η̇(x) = {η(x), H},

~
∂ψ(x) ~
= −{ψ(x), P~ }, ∂η(x) = −{η(x), P~ }.

The observables H, P 1 , P 2 , P 3 and Q are in involution.

9.5 Simultaneous diagonalization

We have the following observables in involution:
Z


hQ|ζi = i − ζ̇(x)ζ(x) + ζ(x)ζ̇(x) d~x, (9.23)
Z
~
|ζ̇(x)|2 + |∇ζ(x)| 2
+ m2 |ζ(x)|2 d~x,


hH|ζi = (9.24)
Z


hPj |ζi = − ζ̇(x)∇j ζ(x) + ∇j ζ(x)ζ̇(x) d~x. (9.25)

Note that (8.28) and (8.29) from the neutral case differ from (9.24) and (9.25)
only by the prefactor 21 .

81
 We use the Fourier transformation and (8.30):
Z
ζ (+) (k)ζ (+) (k)d~k − ζ (−) (k)ζ (−) (k) d~k


hQ|ζi = (9.26)
Z
hH|ζi = ε(~k) ζ (+) (k)ζ (+) (k) + ζ (−) (k)ζ (−) (k) d~k


(9.27)
Z
hP i |ζi = k i ζ (+) (k)ζ (+) (k) + ζ (−) (k)ζ (−) (k) d~k.


(9.28)

9.6 Negative frequency space

p
Recall that for ~k ∈ R3 , set ε = ε(~k) := ~k 2 + m2 and every ζ ∈ WKG can be
written in a unique way as
ζ = ζ (+) + ζ (−) , (9.29)
where
Z
1 ~ 0 ~
ζ (±) (x) = ζ (±) (k) p q e±i(−ε(k)x +k~x) d~k. (9.30)
(2π)3 2ε(~k)

(9.29) gives a decomposition of the space CYKG into two subspaces

(+) (−)
CYKG = WKG ⊕ WKG . (9.31)
We have already introduced the positive energy space W (+) together with its
(−)
completion Z (+) . We will also need the negative frequency space WKG . Let
(−)
c denote the complex conjugation on WKG . We actually have two distinct
interpretations of c: as the usual complex conjugation inside W, so that c :
(−) (+)
WKG → WKG and In particular, c| − k) = |k), or as the identity on W (−) , such
that ci = −ic. We have the scalar product
(−) (−) (−) (−)
(cζ1 |cζ2 ) := iζ1 ωζ2 . (9.32)
(−) (−) (−)
We set ZKG to be the completion of cWKG in this scalar product. ZKG can be
identified with L2 (R3 ) and (9.32) rewritten as
Z
(−) (−) (−) (−)
(cζ1 |cζ2 ) = ζ1 (k)ζ2 (k)d~k

(+) (−)
R1,3 o O↑ (1, 3) acts on ZKG and ZKG in a natural way.

9.7 Plane wave functionals

Plane wave functionals are defined as linear or antilinear functionals on the
complex space WKG , for any ζ ∈ WKG given by

ha(k)|ζi = ζ (+) (k), ha∗ (k)|ζi = ζ (+) (k) (9.33)

hb(k)|ζi = ζ (−) (k) hb∗ (k)|ζi = ζ (−) (k). (9.34)

82
 Thus
Z r
ε(~
p) i 
~ d~x
a(k) = ψ(0, ~x) + q η(0, ~x) e−ik~x p ,
2 (2π)3
2ε(~k)
s
ε(~k) ∗
Z 
i 
~ d~x
a∗ (k) = ψ (0, ~x) − q η ∗ (0, ~x) eik~x p ,
2 (2π)3
2ε(~k)
s
ε(~k) ∗
Z 
i 
~ d~x
b(k) = ψ (0, ~x) + q η ∗ (0, ~x) e−ik~x p ,
2 (2π)3
2ε(~k)
s
ε(~k)
Z 
i 
~ d~x
b∗ (k) = ψ(0, ~x) − q η(0, ~x) eik~x p ,
2 (2π)3
2ε(~k)

The only non-vanishing Poisson bracket are

{a(k), a∗ (k 0 )} = {b(k), b∗ (k 0 )} = −iδ(~k − ~k 0 ).

We have the following expressions for the fields:

d~k
Z
eikx a(k) + e−ikx b∗ (k) ,


ψ(x) = q
(2π)3 2ε(~k)
p
q
Z d~k ε(~k)
√ eikx a(k) − e−ikx b∗ (k) .


η(x) = p
3
i (2π) 2

We have accomplished the diagonalization of the basic observables:

Z
d~kε(~k) a∗ (k)a(k) + b∗ (k)b(k) ,


H =
Z
P~ = d~k~k a∗ (k)a(k) + b∗ (k)b(k) ,

Z
d~k a∗ (k)a(k) − b∗ (k)b(k) .


Q =

(+) (−)
As usual, for f ∈ KKG , g ∈ KKG , we have smeared versions of the above

83
functionals:
Z Z
ha(f )|ζi = f (k)ζ (+) (k)d~k, a(f ) = f (k)a(k)d~k; (9.35)
Z Z
ha∗ (f )|ζi = f (k)ζ (+) (k)d~k, a (f ) = f (k)a∗ (k)d~k;
∗
(9.36)
Z Z
hb(g)|ζi = b(k)ζ (−) (k)d~k, b(g) = g(k)b(k)d~k; (9.37)
Z Z
hb∗ (g)|ζi = g(k)ζ (−) (k)d~k, b∗ (g) = g(k)b∗ (k)d~k; (9.38)

{a(f ), a∗ (f 0 )} = −i(f |f 0 ); {b(g), b∗ (g 0 )} = −i(g|g 0 ). (9.39)

Thus, for k on the mass shell, using physicist’s notation on the left and
mathematician’s on the right, we can write

a∗ (k) = a∗ |k) ,


(9.40)
∗ ∗


b (k) = b c| − k) . (9.41)

9.8 Quantization
In principle, we could quantize the complex Klein-Gordon equation as a pair of
real Klein-Gordon fields. However, we will use the formalism of quantization of
charged bosonic systems, see Subsect. 7.11.
We want to construct (H, Ĥ, Ω) satisfying the usual requirements of QM
(1)-(3) and an operator valued distribution

R1,3 3 x 7→ ψ̂(x) (9.42)

˙
satisfying, with η̂(x) := ψ̂(x),
(1) (−2 + m2 )ψ̂(x) = 0;
(2) the only non-vanishing 0-time commutators are

[ψ̂(0, ~x), η̂ ∗ (0, ~y )] = iδ(~x − ~y ), [ψ̂ ∗ (0, ~x), η̂(0, ~y )] = iδ(~x − ~y ); (9.43)

(3) eitĤ ψ̂(x0 , ~x)e−itĤ = ψ̂(x0 + t, ~x);

(4) Ω is cyclic for ψ̂(x), ψ̂ ∗ (x).
The above problem has an essentially unique solution, which we describe
below.
We set
(+) (−)
H := Γs (ZKG ⊕ ZKG ).
(+)
Creation/annihilation operators for the particle space ZKG ' L2 (R3 ) are de-
(−)
noted with the letter a and for the antiparticle space ZKG ' L2 (R3 ) with the

84
letter b. Thus we put hats and do all the obvious modifications to the classical
formulas. Ω is the Fock vacuum. The quantum field is

d~k
Z  
ψ̂(x) := q eipx â(k) + e−ikx b̂∗ (k) ,
(2π)3 2ε(~k)
p
q
Z d~k ε(~k)  
η̂(x) := p √ eikx â(k) − e−ikx b̂∗ (k) .
i (2π)3 2

The quantum Hamiltonian, momentum and charge are

Z  
Ĥ := â∗ (k)â(k) + b̂∗ (k)b̂(k) ε(~k)d~k, (9.44)
Z 
~

P̂ := â∗ (k)â(k) + b̂∗ (k)b̂(k) ~kd~k,
Z  
Q̂ := â∗ (k)â(k) − b̂∗ (k)b̂(k) d~k.

Equivalently, for any t

Z  
Ĥ = : η̂ ∗ (t, ~x)η̂(t, ~x) + ∂~ ψ̂ ∗ (t, ~x)∂~ ψ̂(t, ~x) + m2 ψ̂ ∗ (t, ~x)ψ̂(t, ~x) :d~x,
Z 
~

P̂ = : −η̂ ∗ (t, ~x)∂~ ψ̂(t, ~x) − ∂~ ψ̂ ∗ (t, ~x)η̂(t, ~x) :d~x,
Z  
Q̂ = i : −η̂ ∗ (t, ~x)ψ̂(t, ~x) + ψ̂ ∗ (t, ~x)η̂(t, ~x) :d~x.

Thus all these operators are expressed in terms of the Wick quantization of their
classical expressions.
 Note thatthe whole
 R1,3 o O↑ (1, 3) acts unitarily on H by U (y, Λ) :=
group 
Γ r(y,Λ) (+)
⊗ Γ r(y,Λ) (−)
, with
ZKG ZKG

U (y, Λ)ψ̂(x)U (y, Λ)∗ = ψ̂ (y, Λ)x .

Moreover,
[ψ̂(x), ψ̂ ∗ (y)] = −iGPJ (x − y), [ψ̂(x), ψ̂(y)] = 0.
Note the identities for the 2-point functions:

[ψ̂(x), ψ̂ ∗ (y)] = −iGPJ (x − y)1l, (9.45a)

∗ (+)
(Ω|ψ̂(x)ψ̂ (y)Ω) = G (x − y), (9.45b)
∗ (−)
(Ω|ψ̂ (x)ψ̂(y)Ω) = G (x − y), (9.45c)
(Ω|T(ψ̂(x)ψ̂ ∗ (y))Ω) = −iGF (x − y), (9.45d)
∗ F
(Ω|T(ψ̂(x)ψ̂ (y))Ω) = iG (x − y). (9.45e)

85
9.9 Smeared fields
For f ∈ Cc∞ (R1,3 , C) we set
Z Z
ψ̂[f ] := f (x)ψ̂(x)dx, ψ̂ ∗ [f ] := f (x)ψ̂ ∗ (x)dx. (9.46)

We obtain an operator valued distribution satisfying the Wightman axioms with

(+) (−)
D := Γfin
s (ZKG ⊕ ZKG ).
For an open set O ⊂ R1,3 the field algebra is defined as
n   o00
F(O) := exp iψ̂ ∗ [f ] + iψ̂[f ] : f ∈ Cc∞ (O, C) .

The observable algebra A(O) is the subalgebra of F(O) fixed by the automor-
phism
B 7→ eiθQ̂ Be−iθQ̂ .
The algebras F(O) and A(O) satisfy the Haag-Kastler axioms.

10 Some historical remarks on QFT

This section should be treated as a collection of gossips and loose statements.
I will discuss only quantum field theory in its traditional sense such as QED or
Standard Model. Various more modern constructions that grew out of Quan-
tum Field Theory, such as String Theory, 2-dimensional Conformal Theories,
topological field theories, Chern-Simons Theory, supersymmetric theories, etc,
are outside of the scope of these remarks.
In this section I will consider both bosons and fermions. Bosons will be
denoted φ and fermions ψ.

10.1 Physicist’s strategy in QFT

Here is an outline of the usual physicist’s strategy. It was essentially developed
by Feynman, Schwinger, Tomonaga and Dyson in the late 40’s.
1. Start with free fields. They are given by quantizing the following La-
grangians:

1 m2 2
L(x) = − ∂µ φ(x)∂ µ φ(x) − φ (x) for bosons, (10.1)
2 2
L(x) =iψ(x)γµ ∂ µ ψ(x) + mψ(x)ψ(x), for fermions. (10.2)

2. Add a perturbation local in fields. Typical perturbations include λφ(x)4 ,

Yukawa λφ(x)ψ(x)ψ(x). Often these perturbations are obtained by the
minimal coupling prescription and lead to a gauge theory. Usually one
assumes that the perturbation is Lorentz invariant, even in fermions, and
the resulting Hamiltonian is positive.

86
 3. Compute formally the perturbation expansion of the scattering operator
between in states Φ− and out states Φ+ :

lim (Φ+ |eitH0 e−i2tHλ eitH0 Φ− ) = (Φ+ |S(λ)Φ− ) (10.3)

t→∞
∞
X
= λn (Φ+ |Sn Φ− ), (10.4)
n=0

expressing it in Feynman diagrams.

4. The terms Sn given by the above prescriptions are ill defined in many
ways. Give them a meaning by renormalization, which leads to a well-
defined formal series expressing scattering amplitudes.

5. From scattering amplitudes compute scattering cross-sections of physical

processes.

10.2 Renormalizability
Consider the Lagrangian of the form (10.1) or (10.2) in d spacetime dimensions.
We want the action to be scalar and the kinetic term to have no dimensionful
coefficient. This implies that in the units of length the boson field has the
dimension [φ] = 1 − d2 and the fermion field the dimension [ψ] = 21 − d2 . Clearly
deg ∂ = −1.
The full Lagrangian is typically the sum of monomials in fields. The action
integral should be dimensionless. The integral includes the Lebesgue measure
dd x, which has dimension d. Therefore, if the dimension of a monomial is c,
then one has to put a coupling constant in front of dimension −d − c.
We will say that a monomial is
1. super-renormalizable if the corresponding coupling constant has a negative
dimension.

2. marginally or just renormalizable if it has a dimensionless coupling con-

stant.
3. non-renormalizable if it has a coupling constant of a positive dimension.
Super-renormalizable and marginally renormalizable terms are joiuntly called
renormalizable. One can show that in renormalizable theories one needs only
a finite number of parameters to fix renormalization conditions. In super-
renormalizable theories it is enough to renormalize Feynman diagrams up to
a certain finite order—in marginally renormalizable theories one has to do it in
any order (but still the number of parameters is finite).
In the old days, physicists used to believe that physical theories should be
renormalizable. Nowadays physicists view non-renormalizable Lagrangians as
useful tools for the description of the matter as well. The dominant view says
that all known quantum field theories that we use are only effective, and their

87
validity is limited to low energies. Therefore, one should not treat them as
ultimate theories.
Nevertheless, renormalizable theories are distinguished. In fact, they have
clearly a better predictive power than non-renormalizable theories. There are
also well-known arguments attributed to Kenneth Wilson, involving the change
of scale, often called the renormalization group, which explain a different role
of renormalizable and non-renormalizable terms in the Lagrangian.
Let us review the renormalizability of various monomials in the Lagrangian
in various dimensions.

• d=2. deg φ = 0, deg ψ = − 12 .

super-renormalizable P (φ), P (φ)∂φ P (φ)ψψ;

2
marginally renormalizable P (φ)(ψψ) , P (φ)ψ∂ψ, P (φ)(∂φ)2 .

• d=3. deg φ = − 12 , deg ψ = −1.

super-renormalizable φ3 , φ4 , φ5 , φ2 ∂φ, φψψ;

6 3 2
marginally renormalizable φ , φ ∂φ, φ ψψ.

• d=4. deg φ = −1, deg ψ = − 23 .

super-renormalizable φ3 ;
marginally renormalizable φ4 , φ2 ∂φ, φψψ.

• d=6. deg φ = −2, deg ψ = − 25 .

marginally renormalizable φ3 .

φ, φ2 , ∂φ, φ∂φ and ψψ are always super-renormalizable (this includes the

mass terms).
(∂φ)2 and ψ∂ψ are always marginally renormalizable (this includes the ki-
netic terms).

10.3 Counterterms
Let us consider for instance the λφ44 theory. Naive computations based on the
Lagrangian

1 m2 2
L(x) = − ∂µ φ(x)∂ µ φ(x) − φ (x) − λφ(x)4 (10.5)
2 2
lead to ill-defined quantities. In order to obtain sensible predictions one has to
consider the Lagrangian involving counterterms, which is traditionally written

88
as
1 m2 2 
L(x) = − Z ∂µ φ(x)∂ µ φ(x) + φ (x) − Z 2 gφ(x)4 (10.6)
2 2
∞
X
Z= Zn λ n , Z0 = 1; (10.7)
n=0
X∞
m= mn λn , m0 = m; (10.8)
n=0
X∞
g= gn λn , g1 = 1. (10.9)
n=1

Then one recursively computes the time-ordered N -point correlation function



G(xN , . . . , x1 ) := Ω| T φ(xN ) · · · φ(x1 ) Ω .

Usually one introduces some regularization depending on a parameter Λ

(a cut-off, Pauli-Villars, dimensional, etc.). Regularized quantities are finite,
and then one takes the limit Λ → ∞. It is also possible to avoid the use of
a regularization (this is the case of the BPHZ method, and also the Epstein-
Glaser method). All these schemes give the same answers, which depend on 3
parameters.
There are various ways to fix these parameters. Typically, one imposes the
“mass shell condition” on G(p), the Fourier transform of the 2-point function:
G(p) ≈ (p2 + m2 )−1 , p2 ≈ −m2 . (10.10)
One needs also a condition on the 4-point function.
From the time-ordered N -point functions the LSZ formulas lead to scattering
amplitudes (matrix elements of the scattering operator).
The situation is different in the massless case. The mass-shell condition
cannot be usually applied. Besides, scattering amplitudes are ill defined because
of the infra-red problem. Instead, one can compute inclusive cross-sections.
Let us replace φ4 with φn for n > 4. This perturbation is non-renormalizable.
There exist solutions of the recursive procedure indicated above, however in
order to fix them one needs an infinite number of parameters.

10.4 Asymptotic freedom

Some classes of Feynman diagrams can be summed up. For instance, when com-
puting the 2-point function for interacting photon in QED using the geometric
series, one can sum up diagrams that consist of repeated 1-particle irreducible
terms. One obtains
1
G(k) = G0 (k). (10.11)
1 − Σ(k)
The quantity Σ(k), usually called the self-energy, can be computed in the lowest
order. One obtains a function that grows for large k. Therefore, at least in

89
this approximation, there is a pole for large energies and it is clear that (10.11)
becomes worthless. Its discovery is attributed to Landau, and the corresponding
pole is called the Landau pole. The same phenomenon can be seen for most other
QFT’s, such as λφ4 .
There exists a version of the above argument based on the renormalization
group equation that also predicts the existence of a pole in these theories and
is even more convincing.
Landau, after discovering the pole named after him, announced the death of
Quantum Field Theory. In reality, if the coupling constant is very small, the pole
is very far in high energies and one can ignore it in perturbative calculations—
this is the case of QED. However, when the coupling constant is not so small,
problems related to the Landau pole may appear close to physical energies. This
is the case of the Standard Model, where there is a φ4 term in the Higgs field.
Quantization of the Yang-Mills theory is much more difficult than that of
fields with an Abelian gauge or no local gauge at all. Nevertheless, by the early
70’s it was well understood and it was proven by t’Hooft and Veltman that
the gauge invariance survives renormalization. In the Yang-Mills theory instead
of photons we have gluons. Gluons interact with themeselves, therefore in the
gluon self-energy beside fermion loops we have (bosonic) gluon loops. They
change the sign of the self-energy, and therefore (if there are not too many
fermions) there is no Landau pole in the ultraviolet and the gluon propagator
becomes suppressed for large energies. This property of the Yang-Mills theory
is called asymptotic freedom. It was discovered in the early 70’s. It implies that
the Yang-Mills theory can be applied in large energies. This lead in the 70’s to
an enthusiastic revival of QFT.
Apparently, t’Hooft did first the computation proving the asymptotic free-
dom of the Yang-Mills theory, but did not recognize its physical importance.
The asymptotic freedom of Yang-Mills was shown (simultaneously?) by Gross-
Wilczek and Politzer. For many years they were on the list of candidates for
the Nobel Prize, until eventually they got it.

10.5 Axiomatic Quantum Field Theory

Axiomatic Quantum Field Theory tries to derive theorems starting from axioms,
such as the Wightman and Haag-Kastler axioms. Here are some of its early
successes:
• The existence of the CPT transformation (in the framework of Wightman
axioms).
• The link between spin and statistics (in the framework of Wightman ax-
ioms).
• The Haag-Ruelle scattering theory—construction of the scattering matrix
starting from Haag-Kastler axioms with a discrete mass shell.
• The Doplicher-Haag-Roberts theory—description of superselection sectors
and their relationship to global gauge groups.

90
 Another current of mathematical research has been devoted to perturbative
Quantum Field Theory. There is good understanding of renormalization of
quantum fields, and also on a curved background.

10.6 Constructive Field Theory

Essentially all computations in Quantum Field Theory are perturbative around
one of free theories. Free theories are not very interesting, because they have
a trivial scattering operator. It is natural to ask whether there exist models of
Quantum Field Theory other than free theories. This question was posed by
Wightman, who formulated the set of axioms called nowadays after him. He
proposed to try to construct interacting models satisfying these axioms.
It is natural to expect that mathematically constructible theories should be
renormalizable. Wightman proposed to start constructions from the simplest,
theories, in 1+1 dimensions, going up the ladder of difficulty, so that eventually
one will be able to construct physically relevant models, expected to be difficult.
This program was initiated in the late 60’s. Probably the most famous team
in this program was that of James Glimm and Arthur Jaffe. They started
with constructing the least complicated of the models from the list in Subsect.
10.2, that is P (φ)2 . To define this model in a finite volume one only needs
to Wick order the interaction and to subtract a divergent constant from the
Hamiltonian. It took several papers and ingenious ideas before this model was
fully constructed and all axioms were verified. An especially successful method
turned out to be the Euclidean approach, which starts from a classical model on
a Euclidean space with a local interaction and then applies the Wick rotation.
A similar successful construction has been accomplished for the Yukawa
model in 2 dimensions, φψψ2 , and for λφ43 . These models are much more difficult
and they require more complicated renormalization. Still, they are quite far
from physical interest. In particular, they are super-renormalizable and live in
dimension < 4.
A major problem with more physical models such as QED4 and λφ44 was the
Landau pole, which essentially means that they are not likely to be constructed,
at least using the perturbative strategy. This problem seems to be absent in
the YM4 . A Polish mathematician Tadeusz Balaban wrote a series of extremely
difficult papers where he studied YM4 in a finite volume. He considered Eu-
clidean YM4 on a lattice and apparently proved that the partition function is
bounded away from zero uniformly in the lattice spacing. YM4 on continuous
Euclidean spacetime, also in a finite volume, was studied by Rivasseau, Magnen
and Seneor by different methods. Both Balaban and Rivasseau et al. claim
that the ultraviolet problem of in a finite volume can be controlled. Neither has
written a proof of this.
To my knowledge there exists one marginally renormalizable model that has
been constructed: the Gross-Neveu model in dimension 2 with the Lagrangian

g2
ψ a (iγ µ ∂µ − m)ψ a + (ψ ψ a )2 . (10.12)
2N a

91
(N is the number of fermion species. For N = 1 the model reduces to the com-
pletely integrable Thirring model). It was constructed by Gawedzki–Kupiainen,
and by Rivasseau et al.
In the early 90’s the interest in constructive field theory waned and essen-
tially this topic was abandoned by researchers. One reason for the collapse of
the topic was that it became prohibitively complicated. Another reason was
the philosophy proclaimed by Wilson saying that quantum field theories that
we know are probably only low-energy effective approximations and one should
not be surprised if they cannot be expressed in a mathematically satisfactory
way.
Nowadays it is even very difficult to determine what has been proven and
what are the proofs—the old literature is mostly unreadable. I know only one
recent result in some kind of constructive field theory. Unfotunately, it is nega-
tive: Aizenman and Duminil-Copin proved that the λφ44 theory is trivial. More
precisely, if one tries to approximate it on a lattice, then in the limit one obtains
a trivial theory.
At the turn of millenium the Clay Institute funded prizes of 1 milion dollars
each for proving 7 important mathematical conjectures. One of them was proven
(The Poincare Conjecturé by Grigorii Perelman, who declined the prize). Other
are still open, including two conjectures in mathematical physics.
One of them is the existence of solutions to the Navier–Stokes equation. The
problem is clearly formulated (by Charles Fefferman).
The other is the construction of the Yang-Mills Theory and the proof of
the existence of a positive mass gap, (formulated by Arthur Jaffe and Edward
Witten).
To my understanding, the formulation of the problem is vague. The descrip-
tion of the problem mentions Wightman axioms, however they seem not suitable
for gauge theories. What is worse, even if we construct something that the Prize
Committee will accept as the quantized Yang-Mills Theory, we have to prove
the positivity of the mass gap (which involves controlling not only ultraviolet
divergencies, but also the large volume limit). In other words, we need to show
that the lightest glueball is massive—which is supposed to be the expression of
the confinement.
.

11 Time-dependent Hamiltonians
11.1 Schrödinger and Heisenberg picture
Suppose that H is a (time-independent) Hamiltonian. It generates the dynamics
e−itH on the Hilbert space H. If we prepare a state ρ at time 0 and measure an
observable A at time t > 0, then the expectation value of the measurement is

TrρeitH Ae−itH . (11.1)

In quantum physics two equivalent ways of expressing (17.1) are used:

92
(1) The Schrödinger picture: We let the state evolve ρ(t) := e−itH ρeitH and
keep the observable constant. Then (17.1) equals Trρ(t)A.
(2) The Heisenberg picture: We let the observable evolve A(t) := eitH Ae−itH
and keep the state constant. Then (17.1) equals TrρA(t).
(By the Schrödinger picture one also means the unitary evolution Ψ(t) :=
e−itH Ψ on H.)

11.2 Time-ordered exponential

We will often use the formalism of time-dependent Hamiltonians. In this sub-
section we describe the main concepts of this formalism.
Let t 7→ Bn (t), . . . , B1 (t) be time dependent operators. Let tn , . . . , t1 be
pairwise distinct. We define the time-ordered product of Bn (tn ),..., B1 (t1 ) by

T (Bn (tn ) · · · B1 (t1 )) := Bσn (tσn ) · · · Bσ1 (tσ1 ),

where (σ1 , . . . , σn ) is the permutation such that tσn ≥ · · · ≥ tσ1 .

Consider a family of self-adjoint operators

t 7→ H(t). (11.2)

For t+ > t− , we define the time-ordered exponential

!
Z t+
Texp −i H(t)dt (11.3)
t−
∞
X Z Z
n
:= (−i) ··· H(tn ) · · · H(t1 )dtn · · · dt1
n=0 t+ ≥tn ≥···≥t1 ≥t−
∞ Z t+ Z t+
X 1
= (−i)n ··· T (H(tn ) · · · H(t1 )) dtn · · · dt1 .
n=0 t− t− n!

For brevity, we will write U (t+ , t− ) for (17.3) and call it the dynamics generated
by t 7→ H(t). Note that U (t+ , t− ) are unitary. (The above constructions can
be easily made rigorous if H(t) are bounded. If they are unbounded, the above
definition should be viewed only as a heuristic indication how to define the
family of unitary operators U (t+ , t− ). In most of this subsection we are not
very precise about the boundedness of operators, types of limits, etc.)
We also set U (t− , t+ ) := U (t+ , t− )−1 . Thus U (t+ , t− ) is the solution of the
following two equivalent equations:
d
U (t+ , t− ) = −iH(t+ )U (t+ , t− ), U (t, t) = 1l; (11.4)
dt+
d
equivalently, U (t+ , t− ) = U (t+ , t− )iH(t− ), U (t, t) = 1l. (11.5)
dt−

Clearly, if H(t) = H does not depend on time, then U (t+ , t− ) = e−i(t+ −t− )H .

93
 We also have
n  i(t − t )  jt + (n − j)t 
+ − + −
Y
U (t+ , t− ) = lim exp − H , (11.6)
n→∞
j=1
n n

where in the product the indices increase from the right to the left:
n
Y
Aj := An · · · A1 . (11.7)
j=1

11.3 Schrödinger and Heisenberg picture for time-dependent

Hamiltonians
The formalism of the Schrödinger and Heisenberg picture described for time-
independent Hamiltonians in Subsection 17.1 is somewhat more complicated if
the Hamiltonian is time-dependent. Then the Hamiltonian in the Schrödinger
picture and the Hamiltonian in the Heisenberg picture can be different.
Let us assume that the evolution U (t+ , t− ) defined as in (17.3) corresponds
to the Schrödinger picture, that is the evolution of vector states is

Ψt+ = U (t+ , t− )Ψt− . (11.8)

Thus the family of self-adjoint operators (17.2) can be called the Hamiltonian
in the Schrödinger picture. Hence the evolution of a density matrix ρ in the
Schrödinger picture from time 0 to time t is

ρ(t) = U (t, 0)ρU (0, t), (11.9)

and satisfies the equation

d
ρ(t) = −i [H(t), ρ(t)] , (11.10)
dt
ρ(0) = ρ. (11.11)

Let us introduce the evolution of an observable A in the Heisenberg picture:

A(t) := U (0, t)AU (t, 0), (11.12)

where we treat t = 0 as the reference time. We then have two ways to express
the time evolution of the expectation value:

Trρ(t)A = TrρA(t). (11.13)

Equivalently, A(t) is the solution of

d
A(t) = i H Hp (t), A(t) ,
 
(11.14)
dt
A(0) = A,

94
where the Hamiltonian in the Heisenberg picture is defined as

t 7→ H Hp (t) := U (0, t)H(t)U (t, 0). (11.15)

Thus a quantum dynamics is described by two time-dependent Hamiltonians:

t 7→ H(t) and t 7→ H Hp (t). If they do not depend on time, they coincide.
The dynamics can be obtained as a solution of equations similar to (17.4)
and (17.5), involving the Hamiltonian in the Heisenberg picture. However, one
of the times has to be the reference time (in our case 0), and the Hamiltonian
appears “on the wrong side”:
d
U (t, 0) = −U (t, 0)iH Hp (t); (11.16)
dt
d
U (0, t) = iH Hp (t)U (0, t). (11.17)
dt
Thus we can compare:
 Z t 
U (t, 0) = Texp −i H(s)ds , (11.18)
0
 Z t 
U (t, 0)−1 = U (0, t) = Texp i H Hp (s)ds . (11.19)
0

11.4 Classical dynamics

To define an evolution on a classical phase space Rd we need to fix a vector field

R × Rd 3 (t, x) 7→ X(t, x) ∈ Rd (11.20)

The equation
d

x(t) = X t, x(t) . (11.21)
dt
for any initial condition x(0) = x0 ∈ Rd , we obtain a solution R 3 t 7→ x(t, x0 ).
This defines a flow R(t, 0)on Rd such that R(t, 0)x0 = x(t, x0 ) and (17.22) can
be rewritten as
d

R(t, 0)x0 = X t, R(t, 0)x0 . (11.22)
dt
In classical mechanics the phase space is described by coordinates (φ, π) ∈
Rm × Rm with the Poisson bracket

{φi , φj } = {πi , πj } = 0,
i
{φ , πj } = δji .

The time evolution is described by a Hamiltonian

R × R2m 3 (t, φ, π) 7→ H(t, φ, π) ∈ R, (11.23)

95
and the Hamilton equations

φ̇(t) = {φ(t), H(t)}

π̇(t) = {π(t), H(t)}. (11.24)

Note that the classical evolution equations (17.22) and (17.24) are analogs
of the quantum equations in the Heisenberg picture (17.14). In particular, the
classical Hamiltonian (17.23) is the analog of the quantum Hamiltonian in the
Heisenberg picture (17.15).
There exists also a dual picture, which is the analog of the Schrödinger
picture of Quantum Mechanics. Consider the evolution given by the backward
flow. The analog of the equation (17.22) is
d
R(0, t)x0 = −X Sp t, R(0, t)x0 ,


(11.25)
dt
where
X Sp (t, y) = R0 (0, t)X t, R(t, 0)y .


(11.26)
Thus the vector field is minus the backward transport of the original field.
R0 (0, t) is the derivative of the flow.
In the Hamiltonian case we have the dynamics

φSp (t), π Sp (t) = R(0, t) φ, π .

(11.27)

The map R(0, t) is symplectic, therefore the Hamiltonian equation is transported

to a Hamiltonian equation, however for a different Hamiltonian:

H Sp (t, φ, π) = H t, R(t, 0)(φ, π) .

(11.28)

More precisely, there is also the change of the sign:

φ̇Sp (t) = −{φSp (t), H Sp (t)}

π̇ Sp (t) = −{π Sp (t), H Sp (t)}. (11.29)

Thus the classical and the quantum cases are analogous. However, whereas
in the quantum case the Schrödinger picture seems preferred, in the classical
case the analog of the Heisenberg picture seems to be more common. Therefore,
in the quantum case we put the superscript Hp but not Sp, and the other way
around in the classical case.

11.5 Time-dependent perturbations

Our time-dependent Hamiltonians will usually have the form

H(t) := Hfr + λV (t),

where Hfr is a self-adjoint operator and R 3 t 7→ V (t) is a family of self-adjoint

operators.

96
 We can introduce the so-called interaction picture or the Furry picture. The
evolution in the interaction picture is

UInt (t+ , t− ) := eit+ Hfr U (t+ , t− )e−it− Hfr .

Note that UInt (t, t) = 1l and

d
UInt (t+ , t− ) = −iHInt (t+ )UInt (t+ , t− ); (11.30)
dt+
d
UInt (t+ , t− ) = UInt (t+ , t− )iHInt (t− ), (11.31)
dt−

where the Hamiltonian for the interaction picture is

HInt (t) = eitHfr V (t)e−itHfr . (11.32)

Therefore, we can write

!
Z t+
UInt (t+ , t− ) = Texp −i HInt (t)dt .
t−

Thus if

ρInt (t) = UInt (t, 0)ρUInt (0, t)

= eitHfr U (t, 0)ρU (0, t)e−itHfr , (11.33)
itHfr −itHfr
Afr (t) = e Ae , (11.34)

then the expectation value (17.13) coincides with

TrAfr (t)ρInt (t). (11.35)

We define the scattering operator by

S := lim UInt (t+ , t− )

t+ ,−t− →∞
 Z ∞ 
= Texp −i HInt (t)dt . (11.36)
−∞

We also introduce the Møller operators

S− := lim U (0, −t)eitHfr = lim UInt (0, −t)

t→∞ t→∞
 Z 0 
= Texp −i HInt (t)dt , (11.37)
−∞
−itHfr
S+ := lim U (0, t)e = lim UInt (0, t)
t→∞ t→∞
 Z ∞ −1
= Texp −i HInt (t)dt . (11.38)
0

97
Clearly, S = S +(−1) S − .
Note that both Møller operators and the scattering operator trivially exist
if V (t) decays sufficiently fast as |t| → ∞. In fact, this is a typical situation in
QFT, where we usually impose a temporal “adiabatic cutoff”.
In quantum mechanics one often applies this formalism to time independent
potentials, but this is a different story.
The interaction picture described above has also its “Heisenberg picture
version”:
Hp
UInt (t+ , t− ) := U (0, t+ )e−i(t+ −t− )Hfr U (t− , 0) (11.39)
Z t+ !
Hp
= Texp − HInt (t)dt , (11.40)
t−
Hp
HInt (t) := UInt (0, t)HInt (t)UInt (t, 0) = U (0, t)V (t)U (t, 0). (11.41)

12 Euclidean fields and spectral shift function

12.1 Trace
Let A be a positive operator on a Hilbert space H. Let {ei , i ∈ I} be an
orthonormal basis in H. One can show that
X
TrA := (ei |Aei ) (12.1)
i∈I

does not depend on the choice of a basis and defines a number ∈ [0, ∞] called
the trace of A. We say that A is trace class if TrA < ∞.
If A is any operator, then it is trace class if it can be written as a linear
combination of positive trace class operators
n
X
Ai = c i Ai . (12.2)
i=1

Then one sets

n
X
TrA = ci TrAi . (12.3)
i=1

One can show that (18.3) does not depend on the decomposition (18.2). Note
that for any unitary U
TrA = TrU AU ∗ . (12.4)
If A is an operator on L2 (Rd ), and A(x, y) is its distributional kernel, then
under some conditions one can show that
Z
TrA = A(x, x)dx. (12.5)

98
For instance, if A = f (x̂)g(p̂), then
Z Z
dp dxdp
A(x, y) = f (x)g(p)ei(x−y)p , TrA = f (x)g(p) . (12.6)
(2π)d (2π)d

Suppose that A is an operator such that ]−∞, 0] is disjoint from its spectrum.
Then we can define ln(A). Suppose in addition that A − 1l is trace class. Then
it is easy to show that ln(A) is trace class. We then can define the so-called
Fredholm determinant of A:

det A := eTr ln(A) . (12.7)

12.2 Neutral Euclidean fields

Suppose that I(φ) is a function of (classical) fields describing the Hamiltonian
(or the “Euclidean Action for a field theory”). Then
Z
e−βI(φ) Dφ (12.8)

1
is called the partition function. The parameter β has the interpretation of kT ,
where T is the temperature. Anyway, we will assume β = 1.
In many situations, e.g. the thermodynamic limit, the partition function is
infinite and instead it is more natural to consider the ratio of partition functions
for two Hamiltonians, I and I0 . We can then introduce the parameter E equal
to the logarithm of its relative partition function:
R −I(φ)
−E e Dφ
e := R −I (φ) . (12.9)
e 0 Dφ

Depending on circumstances, E is called the free energy, pressure or effective

action.
Consider R4 with the Euclidean signature and the Euclidean neutral field
with the mass squared perturbed by κ ∈ S(R4 ):
Z
1
∂µ φ(x)∂ µ φ(x) + m2 φ2 (x) + κ(x)φ2 (x) dx


I(φ) = (12.10)
2
Z
1
∂µ φ(x)∂ µ φ(x) + m2 φ2 (x) dx.


I0 (φ) = (12.11)
2
We would like to compute the renormalized value of E. Formally we have
 det(p̂2 + m2 )  12  − 21
e−E = = det 1l + κ(x̂)(p̂ 2
+ m2 −1
) , (12.12)
det(p̂2 + m2 + κ(x̂))
1
E = Tr log(p̂2 + m2 + κ(x̂)) − log(p̂2 + m2 )


2
1
= Tr log 1l + κ(x̂)(p̂2 + m2 )−1 .


(12.13)
2

99
Unfortunately, (18.13) are divergent. We will show how to renormalize E. This
means, we will show how to modify the Hamiltonian by adding local countert-
erms so that we obtain a finite expression.
Let us try to analyze E, in spite of the fact that it is ill defined. We have
∞ ∞
X (−1)n+1
n X
E = Tr κ(x̂)(p̂2 + m2 )−1 =: En .
n=1
2n n=1

Let us compare this with the classical approximation of E:

Z
1
dxdp
E cl := log(p2 + m2 + κ(x) − log(p2 + m2 )


(12.14)
2 (2π)4
∞
X (−1)n+1 Z ∞
2 −1 n dxdp
X
2
Encl .


= κ(x)(p + m ) =: (12.15)
n=1
2n (2π)4 n=1

Note that Z Z
1 dp
E1 = E1cl = κ(x)dx (12.16)
2 (2π)4 (p2 + m2 )
is divergent. But the classical and quantum quantities coincide. We will simply
drop them.
E2 and E2cl are also divergent, but in general different:
1  
−E2 = Tr κ(x̂)(p̂2 + m2 )−1 κ(x̂)(p̂2 + m2 )−1
4Z
1 dp2 dp1
= κ(p1 − p2 ) κ(p2 − p1 ) (12.17)
4 (2π)4 (p22 + m2 ) (2π)4 (p21 + m2 )
|κ(k)|2
Z
1 dk dq
= (12.18)
4 (2π) (2π) ((q + 2 k) + m2 )((q − 12 k)2 + m2 )
4 4 1 2
Z
dk
= |κ(k)|2 π(k 2 ) , (12.19)
(2π)4
Z Z
1 dp
−E2cl = κ(x)2 dx (12.20)
4 (2π) (p2 + m2 )2
4
Z Z
1 dk dp
= |κ(k)|2 , (12.21)
4 (2π) (2π) (p2 + m2 )2
4 4
Z
dk
= |κ(k)|2 π(0) , (12.22)
(2π)4
where in (18.18) and (18.19) we used κ(−k) = κ(k) and we set p1 = q − 12 k,
p2 = q + 12 k, and we introduced
d4 q
Z
1 1
π(k 2 ) = , (12.23)
4 (2π) ((q + 2 k) + m )((q − 21 k)2 + m2 )
4 1 2 2

We will see that the following renormalized expressions are finite:

Z
dk
E2ren :=E2 − E2cl = |κ(k)|2 π ren (k 2 ) , (12.24)
(2π)4
π ren (k 2 ) :=π(k 2 ) − π(0). (12.25)

100
 All the terms En and Encl with n ≥ 3 are convergent.
To rigorously define the above expressions we need first to perform some
kind of regularization. For instance, we can use the the Pauli-Villars method
with m0 = m, C0 = 1, m1 = M , and C1 = −1. (See Subsection 18.4 for more
about this method). Let us set π(m, k 2 ) to be the formal expression (18.23).
Set
1
X
π reg (k 2 ) := π(m, k 2 ) − π(M, k 2 ) = Ci π(mi , k 2 ). (12.26)
i=0

(18.26) is given by a convergent integrals. The rigorous definition of π ren (k 2 ) is

π ren (k 2 ) := lim π reg (k 2 ) − π reg (0) .


(12.27)
M →∞

ren 2
We will compute π (k ).
In order not to clutter the formulas, in the following computations we use the
ill defined π(k 2 ). The following formulas become well defined when we replace
1
Ci , so that instead of π(k 2 ) we obtain π reg (k 2 ):
P
m with mi and insert
i=0
Z 4
d q 1
4π(k 2 ) = 4 1
(2π) ((q + 2 k) + m )((q − 21 k)2 + m2 )
2 2

∞ Z ∞
d4 q
Z Z  

2 1 2 2

= dα 1 dα2 exp −(α 1 + α 2 ) q + k + m − (α1 − α2 )qk
(2π)4 0 0 4
Z ∞ Z ∞  
1 1 α1 α2 2
= dα1 dα2 exp −(α1 + α2 )m2 − k
(4π)2 0 0 (α1 + α2 )2 α1 + α2
Z 1 Z ∞
(1 − v 2 )k 2
  
1 dρ 2
= dv exp −ρ m +
(4π)2 2 −1 0 ρ 4
Z 1 2 2 
1  k (1 − v )
= 2
dv log m2 +
(4π) 2 −1 4
Z 1
(1 − v 2 )k 2
   
1 2
= dv log 1 + + log m .
(4π)2 2 −1 4m2
We used the identities (18.28) and (18.29):
Z ∞
1
= dα exp(−αA), (12.28)
A 0
Z
dp 1
exp −(ap2 + bp) = exp b2 /4a .



(12.29)
(2π)4 (4π)2 a2
Then we changed the variables to
(1 − v) (1 + v)
α1 = ρ , α2 = ρ , (12.30)
2 2
1
so that α1 + α2 = ρ, − α1 + α2 = ρv, dα1 dα2 = ρdvdρ.
2

101
At the end we use the identity (18.31):
Z ∞X
dρ X
Ci e−ρAi = − Ci log(Ai ). (12.31)
0 i
ρ i
P
valid if Ci = 0.
Now the renormalized π is given by

π ren (k 2 ) π reg (k 2 ) − π reg (0)

:= lim
M →∞
Z 1
1  k 2 (1 − v 2 )   k 2 (1 − v 2 ) 
= lim log 1 + − log 1 + dv
M →∞ 4(4π)2 2 −1 4m2 4M 2
Z 1
1  k 2 (1 − v 2 ) 
= 2
log 1 + dv. (12.32)
4(4π) 2 −1 4m2

Note that
π ren (0) = 0, (12.33)
which is our renormalization condition. Finally, we evaluate π ren using
Z
(1 + w)
log(1 − w2 )dw = w log(1 − w2 ) − 2w + log , 0 < w < 1. (12.34)
(1 − w)

We obtain
r
ren 2 1 1 1+θ  k2
π (k ) = 2
log −2 , θ= . (12.35)
4(4π) θ 1−θ k2 + 4m2

Here are operator-theoretic formulas for the renormalized quantities:

1 
2 
E2ren = − Tr κ(x)(p̂2 + m2 )−1 − κ(x)2 (p̂2 + m2 )−2 ,
4
1 
E ren = Tr log 1l + κ(x)(p̂2 + m2 )−1 − κ(x)(p̂2 + m2 )−1


2
1 
+ κ2 (x)(p̂2 + m2 )−2 . (12.36)
2

12.3 Charged Euclidean fields

Consider Euclidean charged field on R4 in the presence of magnetic potential
[Aµ ]. Formally we have
Z  R  

∗
(∂µ + ieAµ (x))ψ(x) ∂ µ + ieAµ (x) ψ(x) + m2 ψ ∗ (x)ψ(x) dx Dψ ∗ Dψ


exp −
Z  R

exp − ∂µ ψ ∗ (x)∂ µ ψ(x) + m2 ψ ∗ (x)ψ(x) dx Dψ ∗ Dψ
det K0 

= = exp − Tr log(K) − log(K0 ) , (12.37)
det K

102
where
K0 := p̂2 + m2 , K := − ∂ µ + ieAµ (x) ∂µ + ieAµ (x) + m2 .



(12.38)
are operators on L2 (R4 ).
Our goal is to compute the renormalized value of


E := Tr log(K) − log(K0 ) . (12.39)
We have
 
E =Tr log 1l + − ieAµ ∂µ − ie∂µ Aµ + e2 Aµ Aµ (p̂2 + m2 )−1


(12.40)
∞
X (−1)n+1  n
Tr − ieAµ ∂µ − ie∂µ Aµ + e2 Aµ Aµ (p̂2 + m2 )−1 .


=
n=1
n

We have E(A) = E(−A). Besides, E(A) is real (because of the self-adjointness

of K and K0 ). Therefore, terms of odd order in e vanish. This is the content of
Furry’s theorem for charged bosons. Hence (18.40) can be written as
∞
X
E= e2n En .
n=1

The expressions for En are convergent for n ≥ 3. E2 is logarithmically divergent,

but its physically relevant gauge invariant part is convergent. E1 is quadratically
divergent and its gauge-invariant part is logarithmically divergent. It needs an
infinite renormalization, which will be described below.
The lowest nonzero terms are of the second order in e, and hence of the first
e2
order in α = 4π .
Z
2 2 dp2
−2e E1 = e (p1 + p2 )µ Aµ (p1 − p2 )
(2π)4 (p22 + m2 )
dp1
× (p2 + p1 )ν Aν (p2 − p1 ) (12.41)
(2π)4 (p21 + m2 )
Z
dk dp
− e2 2Aµ (−k)Aµ (k) (12.42)
(2π)4 (2π)4 (p2 + m2 )
Z
dk
=: Aµ (−k)Aν (k)2Πµν (k). (12.43)
(2π)4
(18.43) defines the vacuum energy tensor Πµν (k).
We will compute Πµν (k) using the Pauli-Villars regularization. The ultra-
violet problem is more severe now than it was for the mass-like perturbation,
where a single additional fictitious particle sufficed to make the expressions well
defined. Now we need two fictitious particles:
m20 := m2 , C0 := 1,
m21 2
:= m + 2Λ , 2
C1 := 1,
m22 2
:= m + Λ , 2
C2 := −2.

103
Using
2
X 2
X
Ci = Ci m2i = 0 (12.44)
i=0 i=0

we can check that with this choice the sums used in the following computations
are integrable.

d4 q 
Z
2 4qµ qν
2Πµν (k) = e
(2π)4 (q + 12 k)2 + m2i − i0 (q − 21 k)2 + m2

gµν gµν 
− −
(q + 12 k)2 + m2 (q − 12 k)2 + m2

d4 q 4qµ qν − 2gµν (q 2 + 14 k 2 + m2 )
Z
2
= e
(2π)4 (q + 12 k)2 + m2 (q − 12 k)2 + m2



Z ∞ Z ∞
d4 q
Z  
 1 2
= e2 dα 1 dα 2 4q q
µ ν − 2gµν q 2
+ k + m 2
(2π)4 0 0 4
 
 1 
× exp −(α1 + α2 ) q 2 + k 2 + m2 − (α1 − α2 )qk
4
Z 4 Z ∞ Z ∞  
d q  1 2
= e2 dα 1 dα 2 4∂ ∂
zµ zν − 2g µν ∂z
2
+ k + m 2
(2π)4 0 0 4
 
 1 
× exp −(α1 + α2 ) q 2 + k 2 + m2 − (α1 − α2 )qk + zq
4 z=0
2 Z ∞ Z ∞  
e 1 
2 1 2 2
= dα1 dα 2 4∂ ∂
zµ zν − 2g µν ∂z + k + m
(4π)2 0 0 (α1 + α2 )2 4
 1 2 
 1 
× exp −(α1 + α2 ) k 2 + m2 + (α1 − α2 )k − z
4 4(α1 + α2 ) z=0

∞ ∞
e2 (α1 − α2 )2
Z Z
= dα1 dα2 − (gµν k 2 − kµ kν )
(4π)2 0 0 (α1 + α2 )4
!
m2

α1 α2 2 1
−2gµν k + +
(α1 + α2 )4 (α1 + α2 )3 (α1 + α2 )2
 
α1 α2 2
× exp −(α1 + α2 )m2 − k
α1 + α2
=: (−gµν k 2 + kµ kν )2Πgi (k 2 ) + 2Πgd 2
µν (k ).

104
 Let us compute the gauge dependent part of the vacuum energy tensor:
Πgd 2
µν (k )
Z ∞ Z ∞
e2
 
2 α1 α2 2
= dα 1 dα 2 exp −(α 1 + α 2 )m − k
(4π)2 0 0 α1 + α2
2 2
 
α1 α2 k 1 m
× gµν + +
(α1 + α2 )4 (α1 + α2 )3 (α1 + α2 )2
1 ∞
e2 (1 − v 2 )ρ 2 (1 − v 2 ) 2 m2
Z Z    
2 1
= dv dρ exp −ρm − k gµν k + 2+
(4π)2 2 −1 0 4 4ρ ρ ρ
1 ∞
e2 (1 − v 2 )ρ 2
Z Z  
d 1
=− 2
dv dρ exp −ρm2 − k gµν
(4π) 2 −1 0 dρ ρ 4
Z 1
e2 (1 − v 2 )ρ 2
 
1 ρ=∞
2
=− 2
dv exp −ρm − k gµν .
(4π) 2 −1 ρ 4 ρ=0

It vanishes (remember to insert the sum including “fictitious particles”!). To

compute the gauge invariant part we proceed similarly as in Subsection 18.2,
and we obtain
Z ∞ Z ∞
e2 (α1 − α2 )2
Πgi (k 2 ) = 2
dα 1 dα2
2(4π) 0 0 (α1 + α2 )4
 
α1 α2 2
× exp −(α1 + α2 )m2 − k
α1 + α2
2 Z 1 Z ∞
(1 − v 2 )k 2
  
e dρ 2 2
= dv v exp −ρ m +
4(4π)2 −1 0 ρ 4
2 Z 1  2 2

e (1 − v )k
= dvv 2 log m2 + .
4(4π)2 −1 4
We define
Πren (k 2 ) Πgi (k 2 ) − Πgi (0)


:= lim (12.45)
Λ→∞
Z 1
e2 (1 − v 2 )k 2
 
2
= dvv log 1 + .
4(4π)2 −1 4m2
Using
w3 2w3
Z
2w
w2 log(1 − w2 )dw = log(1 − w2 ) − −
3 9 3
1 (1 + w)
+ log , 0 < w < 1, (12.46)
3 (1 − w)
we obtain
Πren (k 2 )
! r
e2 1 1+θ 2 2 k2
= log − − , θ= .
2 · 3(4π)2 θ3 1 − θ 3 θ2 k 2 + 4m2

105
 Note that the Fourier transform of the electromagnetic field is

Fµν (k) = kµ Aν (k) − kν Aµ (k). (12.47)

Hence
1
− Fµν (k)F µν (k) = −k 2 |A(k)|2 + |kA(k)|2 . (12.48)
2
Thus the renormalized 1st order contribution to the vacuum energy is
Z
dk
E1ren = − Πren (k 2 )Fµν (k)F µν (k). (12.49)
2(2π)4
with the renormalization condition

Πren (0) = 0. (12.50)

12.4 Spectral shift function

Suppose H and H0 be two self-adjoint operators. We say that the pair H,
H0 possesses the spectral shift function if for any f ∈ Cc∞ (R), the operator
f (H) − f (H0 ) is trace class. Clearly, the map

Cc∞ (R) 3 f 7→ f (H) − f (H0 ) ∈ R (12.51)

is a linear functional. Under some conditions on H, H0 there exists a function

ξ(s) = ξH,H0 (s) such that
Z
Tr f (H) − f (H0 ) = ξ(s)f 0 (s)ds.


(12.52)

ξ(s) is defined up to an additive constant. If H, H0 are bounded from be-

low, which will always be true in our applications, then one can assume that
lim ξ(s) = 0. ξH,H0 (s) is called the spectral shift function.
s→−∞
Suppose for the moment that H, H0 are trace class. Then they can be
diagonalized. Let En , E0,n be the eigenvalues of H, resp. H0 , in the increasing
order, counting with multiplicities. Then
∞ ∞ Z En

X
X
Tr f (H) − f (H0 ) = f (En ) − f (E0,n ) = f 0 (s)ds, (12.53)
n=1 n=1 E0,n

which explains the name “spectral shift function”.

Let us describe the basic method of computing the spectral shift. Let us
write H = H0 + λV . We introduce


DH,H0 (z) = D(z) : = Tr log(H − z) − log(H0 − z)
= Tr log (H − z)(H0 − z)−1 = Tr log(1l + λV (H0 − z)−1



∞
X λn (−1)n+1
n
= Tr V (H0 − z) .
n=1
n

106
Clearly, if the spectral shift function is well defined, then
Z
D(z) = ξ(s)(s − z)−1 ds, (12.54)

which can be inverted:

1

ξ(s) = D(s + i0) − D(s − i0) . (12.55)
2πi
This method can fail if the rhs of (18.54) is not integrable. Suppose that
n ∈ N, the spectral shift exists and satisfies
Z
|ξ(s)|(|s| + 1)−n−1 ds < ∞. (12.56)

Then there exists an improved method of computing the spectral shift, which
is essentially an adaptation of the Pauli-Villars method from QFT. Choose
c0 , , . . . cn and Λ0 , . . . , Λn such that

c0 + · · · + cn = 0,
c0 Λ0 + · · · + cn Λn = 0,
...
c0 Λn0 + · · · + cn Λnn = 0.

Now we easily check that for s → ∞

n
X
ci (s + Λi − z)−1 = O(s−1−n ), (12.57)
i=0

Hence
 n n 
Dreg (z) :=Tr log Π (H + Λi − z)ci − log Π (H0 + Λi − z)ci
i=0 i=0
Z Xn
= ξ(s) ci (s + Λi − z)−1 ds. (12.58)
i=0

is well defined. Now

n
X 1
Dreg (s + i0) − Dreg (s − i0) .


ci ξ(s + Λi ) = (12.59)
i=0
2πi

In practice, we set c0 = 1, Λ0 = 0, Λi = αi Λ with ci and αi > 0 fixed for

i = 1, . . . , n. Then
1
Dreg (s + i0) − Dreg (s − i0) .


ξ(s) = lim (12.60)
Λ→∞ 2πi

107
13 Scalar field with a masslike perturbation
13.1 Lagrangian and Hamiltonian formalism
Consider the Lagrangian density
1 1
L(x) = − ∂µ φ(x)∂ µ φ(x) − (m2 + κ(x))φ(x)2 , (13.1)
2 2
where R1,3 3 x 7→ κ(x) is a given function. In most of this subsection we will
assume that κ is Schwartz and m > 0. (19.1) leads to the equations

(−2 + m2 )φ(x) = −κ(x)φ(x), (13.2)

The variable conjugate to φ(x) is π(x) := φ̇(x), so that

{φ(t, ~x), φ(t, ~y )} = {π(t, ~x), π(t, ~y )} = 0,

{φ(t, ~x), π(t, ~y )} = δ(~x − ~y ). (13.3)

The theory of the Klein-Gordon equation with a variable mass is very similar
to the one with a constant mass. We can define the corresponding retarded and
advanced propagators as the unique distributional solutions of

− 2x + m2 + κ(x) G∨/∧ (x, y) = δ(x − y),

(13.4)

satisfying
suppG∨/∧ ⊂ {x, y : x ∈ J ∨/∧ y)}.
The Pauli-Jordan function:

GPJ (x, y) := G∨ (x, y) − G∧ (x, y)

satisfies
suppGPJ ⊂ {x, y : x ∈ J(y)}.
and can be used to solve the initial value problem of (19.2):
Z
φ(t, ~x) = − ∂s GPJ (t, ~x, s, ~y ) φ(0, ~y )d~y
s=0
Z
+ GPJ (t, ~x, 0, ~y )π(0, ~y )d~y . (13.5)

Using (19.5) we obtain

{φ(x), φ(y)} = −GPJ (x − y).

The free (constant mass) field will be denoted by φfr , πfr . We can use the
same phase space YKG for the free and perturbed scalar particle, identifying
them for the initial condition at t = 0:

φfr (0, ~x) = φ(0, ~x), πfr (0, ~x) = π(0, ~x). (13.6)

108
Thus we can introduce the plane wave functionals a(k), a∗ (k) as in the free case,
since they depend only on the Cauchy data at time 0.
We easily obtain the Hamiltonian density from the Lagrangian:
1 2 1 ~
2 1 2
H(x) = π (x) + ∂φ(x) + (m + κ(x))φ2 (x),
2 2 2
so that the full Hamiltonian generating the dynamics, first in the “Heisenberg
picture” and then in the “Schrödinger picture”, are
Z
H(t) = H(t, ~x)d~x
Z 
1 2 1 ~
2 1 
= π (t, ~x) + ∂φ(t, ~x) + (m2 + κ(t, ~x))φ2 (t, ~x) d~x, (13.7)
2 2 2
Z 
Sp 1 2 1 ~
2 1 2 
H (t) = π (~x) + ∂φ(~x) + (m + κ(t, ~x))φ2 (~x) d~x, (13.8)
2 2 2

where φ(~x) = φ(0, ~x), π(~x) = π(0, ~x).

13.2 Dynamics in the interaction picture

The classical interaction picture Hamiltonian can be expressed in terms of plane
wave functionals:
Z
1
HInt (t) = κ(t, ~x)φ2fr (t, ~x)d~x (13.9)
2
d~k1 d~k2 κ(t, ~k1 + ~k2 )  −itε(~k1 )−itε(~k2 )
Z
1
= q q e a(−k1 )a(−k2 )
2
(2π)3 2ε(~k1 ) 2ε(~k2 )

~ ~ ~ ~
+2eitε(k1 )−itε(k2 ) a∗ (k1 )a(−k2 ) + eitε(k1 )+itε(k2 ) a∗ (k1 )a∗ (k2 ) .

We can also introduce the creation operators in the interaction picture:

a∗t (k) := UInt (0, t)a∗ (k)UInt (t, 0). (13.10)

They satisfy the equations generated by the Heisenberg picture interaction

Hp
Hamiltonian HInt :
Hp
HInt (t) = UInt (0, t)HInt (t)UInt (t, 0) (13.11)
d~k1 d~k2 κ(t, ~k1 + ~k2 )
Z
1 
~ ~
= q q e−itε(k1 )−itε(k2 ) at (−k1 )at (−k2 )
2
(2π)3 2ε(~k1 ) 2ε(~k2 )

~ ~ ~ ~
+2eitε(k1 )−itε(k2 ) a∗t (k1 )at (−k2 ) + eitε(k1 )+itε(k2 ) a∗t (k1 )a∗t (k2 ) .

109
We obtain
Hp
ȧ∗t (k)
 ∗
= at (k), HInt (t)
dk1 κ(t, −~k + ~k1 )
~
Z
= i q q
(2π)3 2ε(~k) 2ε(~k1 )
 
~ ~ ~ ~
× e−itε(k)−itε(k1 ) at (−k1 ) + e−itε(k)+itε(k1 ) a∗t (k1 ) ,
a∗0 (k) = a∗ (k).

We obtain a symplectic evolution of the firm of the form

" #
pt+ ,t− qt+ ,t−
(13.12)
qt+ ,t− pt+ ,t−

more precisely,
" ∗
a∗t− (k1 )
# " #" #
at+ (k) Z pt+ ,t− (k, k1 ) qt+ ,t− (k, k1 )
= d~k1 .
at+ (k) qt+ ,t− (k, k1 ) pt+ ,t− (k, k1 ) at− (k1 )

(19.12) has a limit as t+ , −t− → ∞, which can be called the classical scattering
operator.
One can try to solve the equations of motion by iterations. The first iteration
is often (at least in the quantum context) called the Born approximation, and
it gives the following formula for the elements of (19.12):
t+
κ(s, −~k + ~k1 )
Z
~ ~
pBorn
t+ ,t− (k, k1 ) = δ(~k − ~k1 ) + i ds q q e−isε(k)+isε(k1 ) ,
t− (2π)3 2ε(~k) 2ε(~k1 )
t+
κ(s, −~k + ~k1 )
Z
~ ~
qtBorn
+ ,t−
(k, k1 ) = i ds q q e−isε(k)−isε(k1 ) .
t− (2π)3 2ε(~k) 2ε(~k1 )

13.3 Quantization
We are looking for quantum fields R1,3 7→ φ̂(x) satisfying

(−2 + m2 )φ̂(x) = −κ(x)φ̂(x), (13.13)

˙
with the conjugate field π̂(x) := φ̂(x) having the equal time commutators

[φ̂(t, ~x), φ̂(t, ~y )] = [π̂(t, ~x), π̂(t, ~y )] = 0,

[φ̂(t, ~x), π̂(t, ~y )] = iδ(~x − ~y ). (13.14)

110
and coinciding with the free field at time 0. The solution is given by putting
“hats” onto (19.5):
Z
φ̂(t, ~x) = − ∂s GPJ (t, ~x, s, ~y ) φ̂(0, ~y )d~y
s=0
Z
+ GPJ (t, ~x, 0, ~y )π̂(0, ~y )d~y . (13.15)

Clearly,
[φ̂(x), φ̂(y)} = −iGPJ (x − y).
We would like to check whether the classical scattering operator and the
classical dynamics are implementable in the Fock space for nonzero κ. By Thm
14.5, we need to check the Shale condition, that is, whether the off-diagonal
elements of (19.12) are square integrable. For simplicity, we will restrict our-
selves to the Born approximation; the higher order terms do not change the
conclusion.
The verification of the Shale condition is easier for the scattering operator.
Consider
Z ∞
κ(s, −~k + ~k1 ) ~ ~
Born
q∞,−∞ (k, k1 ) = i ds q q e−isε(k)−isε(k1 ) . (13.16)
−∞ (2π)3 2ε(~k) 2ε(~k1 )

Recall that κ is a Schwartz function. Therefore, we can integrate by parts as

many times as we want:
∞ ~ ~
∂sn κ(s, −~k + ~k1 ) e−isε(k)−isε(k1 )
Z
Born
q∞,−∞ (k, k1 ) = in+1 ds
. (13.17)
~ ~ n
q q
−∞ (2π)3 2ε(~k) 2ε(~k1 ) ε(k) + ε(k1 )

This decays in ~k and ~k1 as any inverse power, and hence is square integrable on
R3 × R3 . Therefore the classical scattering operator is implementable.
Next let us check the implementability of the dynamics, believing again that
it is sufficient to check the Born approximation. We integrate by parts once:

qtBorn
+ ,t−
(k, k1 )
~ ~ ~ ~
−κ(t+ , −~k + ~k1 )e−it+ ε(k)−it+ ε(k1 ) + κ(t− , −~k + ~k1 )e−it− ε(k)−it− ε(k1 )
= q q
(2π)3 2ε(~k) 2ε(~k1 ) ε(~k) + ε(~k1 )

~ ~
∂s κ(s, −~k + ~k1 )e−isε(k)−isε(k1 )
Z t+
+ ds q q
. (13.18)
t− (2π)3 2ε(~k) 2ε(~k1 ) ε(~k) + ε(~k1 )

Using that κ(s, ~k + ~k1 ) decays fast in the second variable, we see that (19.18)
can be estimated by
C
,
(ε(~k) + ε(~k1 ))2

111
which is square integrable. Therefore, the dynamics is implementable for any
t− , t+ .
By a similar computation we check that if we freeze t0 ∈ R, the dynamics
generated by the momentary Hamiltonian HInt (t0 ) is implementable.

13.4 Quantum Hamiltonian

We would like to find a time-dependent Hamiltonian Ĥ(t) such that
 Z t −1  Z t 
φ̂(t, ~x) = Texp −i Ĥ(s)ds φ̂(0, ~x)Texp −i Ĥ(s)ds . (13.19)
0 0

Formally, the quantum Hamiltonians, first in the Heisenberg, then in the Schrödinger
picture, are given by
Z 
1 2 1
2 1 
Ĥ Hp (t) := π̂ (t, ~x) + ∂~ φ̂(t, ~x) + (m2 + κ(t, ~x))φ̂2 (t, ~x) d~x, (13.20)
2 2 2
Z 
1 2 1 ~
2 1 2 
Ĥ(t) := π̂ (~x) + ∂ φ̂(~x) + (m + κ(t, ~x))φ̂2 (~x) d~x. (13.21)
2 2 2
We will treat the Schrödinger picture Hamiltonian, that is (19.21), as the stan-
dard one. It is expressed in terms of zero time fields. It is clear that it is
ill-defined. One could improve it by putting : · · · :, that is by the Wick order-
ing. We will see later on that even the Wick-ordered expression (19.21) does
not define an operator.
Formally (19.47) remains true if we add a time dependent constant C(t)
to (19.21). We will see that in order to define correct Hamiltonians Ĥ(t) this
constant has to be infinite, even after Wick ordering. We will obtain bounded
from below Hamiltonians Ĥren (t), however the vacuum will not be contained in
their form domain. Therefore, the condition (Ω|Ĥren (t)Ω) = 0 for all t, which
is equivalent to the Wick ordering, cannot be imposed.
The interaction Hamiltonian is formally given by
Z
1
ĤInt (t) = κ(t, ~x)φ̂2fr (t, ~x)d~x, (13.22)
2
d~k
Z  
~ ~ ~ ~
φ̂fr (x) = q e−itε(k)+ik~x â(k) + eitε(k)−ik~x â∗ (k) , (13.23)
(2π)3 2ε(~k)
p

where (19.23) is taken from (8.59). Here is the Wick ordered interaction Hamil-
tonian:
d~k1 d~k2 κ(t, ~k1 + ~k2 )  −itε(~k1 )−itε(~k2 )
Z
1
:ĤInt (t): = q q e â(−k1 )â(−k2 )
2
(2π)3 2ε(~k1 ) 2ε(~k2 )

~ ~ ~ ~
+2eitε(k1 )−itε(k2 ) â∗ (k1 )â(−k2 ) + eitε(k1 )+itε(k2 ) â∗ (k1 )â∗ (k2 ) .

where κ(t, ~k) is a partial Fourier transform of κ(t, ~x).

112
13.5 The vacuum energy
We would like to compute
!
 Z 
e−iE := (Ω|ŜΩ) = Ω|Texp − iĤInt (t)dt Ω . (13.24)

Not surprisingly, (19.24) will require a renormalization. So eventually we will

compute its renormalized version:
!
 Z 
−iE ren ren ren
e := (Ω|Ŝ Ω) = Ω|Texp − iĤInt (t)dt Ω . (13.25)

Anyway, we treat (19.24) as the starting point of our computations. The

path integrals approach leads to the following expression for (19.24) in terms of
Gaussian integrals:
R R
exp(i L(x, φ, φµ )dx)Dφ
e−iE = R R , (13.26)
exp(i L0 (x, φ, φµ )dx)Dφ
where the Lagrangians are
1 1 2
L(x) = − ∂µ φ(x)∂ µ φ(x) − (m + κ(x))φ(x)2 ,
2 2
1 1 2
L0 (x) = − ∂µ φ(x)∂ µ φ(x) − m φ(x)2 ,
2 2
and we use the Minkowski signature x = (x0 , . . . , x3 ). (19.26) can be evaluated
as
i  
E = − Tr ln(−2 + m2 + κ − i0) − ln(−2 + m2 − i0) . (13.27)
2
It is much more convenient to do computations in the Euclidean setting.
Various symbols from the Euclidean case will be decorated by the subscript E.
In particular, the Euclidean version of the path integrals becomes
exp(− LE (x, φ, φµ )dx)Dφ
R R
−E E
e =R R , (13.28)
exp(− LE 0 (x, φ, φµ )dx)Dφ

where the Euclidean Lagrangians are

1 1 2
LE (x) = ∂µ φ(x)∂ µ φ(x) + (m + κ(x))φ(x)2 ,
2 2
1 1 2
E
L0 (x) = ∂µ φ(x)∂ µ φ(x) + m φ(x)2 ,
2 2
and we use the Euclidean signature x = (x1 , . . . , x4 ). We have
E
 − 12
e−E = det(p̂2 + m2 + κ)(p̂2 + m2 )−1 , (13.29)
1  
E E = Tr ln(p̂2 + m2 + κ) − ln(p̂2 + m2 ) , (13.30)
2

113
where p̂2 is the 4-dimensional Laplacian.
As we computed in Subsection 18.2, E E needs to be renormalized and equals
(see (18.35)):
∞
X
E E,ren = E2E,ren + EnE , (13.31)
n=3
Z
dk E,ren 2
E2E,ren = − |κ(k)|2 π (k ), (13.32)
(2π)4
r
1 1 1+θ  k2
π E,ren (k 2 ) = log − 2 , θ= . (13.33)
4(4π)2 θ 1−θ k 2 + 4m2

We would like to find the Minkowskian analogs of E E and E2E :

∞
X
E = E2ren + En , (13.34)
n=2
Z
dk ren 2
E2ren = |κ(k)|2 π (k ). (13.35)
(2π)4

The Wick rotation consists in replacing x4 with ix0 everywhere except for
κ(x), where we do not change anything. More precisely, we replace x4 with zx0 ,
where z is a complex parameter which is continued as eiα with α ∈ [0, π2 ]. Thus
the Euclidean quantities are treated as functions F (z), where F (z) z=1 is the
Euclidean value and F (z) z=i the Wick rotated value.
This implies replacing p4 with −ip0 . Moreover, the Euclidean x2 and p2
are replaced with Minkowskian x2 + i0 and p2 − i0. The Euclidean Lebesgue
measures dx and dk are replaced in Rthe Minkowski space by idx, resp. −idk.
Consequently, the Euclidean action LE (x, φ, φ,µ )dx becomes after the Wick
E
rotation −i L(x, φ, φ,µ )dx. Thus e−E (z) z=i = e−iE , and consequently
R

E = −iE E (z) z=i

(13.36)
 Z (−i)dk E,ren 2 
E2 = −iE2E,ren (z) z=i = −i − |κ(k)|2 π (k − i0)
(2π)4
Z
dk E,ren 2
= |κ(k)|2 π (k − i0). (13.37)
(2π)4

Thus comparing with (19.35) we obtain

π ren (k 2 ) = π E,ren (k 2 − i0). (13.38)

114
 Taking into account (19.33), we obtain
√
ren 2 1 1 1+θ  k2
π (k ) = 2
log −2 , θ= √ , 0 < k 2 ; (13.39)
4(4π) θ 1−θ k + 4m2
2
√
1 2  −k 2
= 2
arctan θ − 2 , θ = √ , −4m2 < k 2 < 0; (13.40)
4(4π) θ k 2 + 4m2
√
1 1 θ+1   −k 2
= log − iπ − 2 , θ = √ , k 2 < −4m2 . (13.41)
4(4π)2 θ θ−1 −k 2 − 4m2
Here is the calculation. First we assume that k 2 > 0. Then we just take the
Euclidean value. Then we use analytic continuation, remembering
√
that k 2 may
2 k 2
have negative imaginary part. As k decreases, θ := √k2 +4m2 first varies from
1+θ
1 to 0, then from 0 to −i∞, finally, from ∞ to 1. Therefore, 1−θ first varies
from ∞ to 1, then goes over the lower semicircle, finally, from −1 to −∞. Next
1+iy
we use log 1−iy = 2i arctan y, and for y < 0, log(y − i0) = log |y| − iπ.
(19.39) corresponds to a spatial transfer of energy-momentum. In (19.40)
the transfer is time-like, but the energy is below the 2-particle threshold. In
(19.41) it is above this threshold, and π ren acquires a nonzero imaginary part
responsible for the decay of the vacuum.
Here are operator-theoretic formulas for the renormalized vacuum energies:
i 
2 
E2ren = − Tr κ(x)(−2 + m2 − i0)−1 − κ(x)2 (−2 + m2 − i0)−2 ,
4
i 
ren
= − Tr log 1l + κ(x)(−2 + m2 − i0)−1 − κ(x)(−2 + m2 − i0)−1


E
2
1 
+ κ2 (x)(−2 + m2 − i0)−2 . (13.42)
2

13.6 Renormalized scattering operator and Hamiltonian

The naive Hamiltonian Ĥ(t) and the naive scattering operator Ŝ are ill defined.
However, in Ŝ only the overall coefficient is ill defined. We can renormalize Ŝ
by multiplying it by a divergent phase:
κ(x)2 dx
R R
Ŝ ren = eiC κ(x)dx+iπ(0)
Ŝ, (13.43)

where C is the constant responsible for the Wick ordering and π(0) is the 2nd
order renormalization. Note that both infinite quantities are quite well behaved
– they depends locally on the interaction, and therefore the renormalization
preserves the Einstein causality. This manifests itself in the identity

Ŝ ren (κ2 )Ŝ ren (κ1 ) = Ŝ ren (κ2 + κ1 ), (13.44)

whenever suppκ2 is later than suppκ1 .

If we cut off the perturbation in time by setting κt+ ,t− (x) := 1l[t− ,t+ ] (x0 )κ(x),
then we can repeat the same constructions, obtaining the renormalized scatter-
ing operator Ŝ ren (t2 , t1 ). For t3 > t2 > t1 , as a special case of (19.44), we

115
have
Ŝ ren (t3 , t2 )Ŝ ren (t2 , t1 ) = Ŝ ren (t3 , t1 ). (13.45)
Thus we get a unitary evolution given by

U ren (t+ , t− ) := e−it+ H0 Ŝ ren (t+ , t− )eit− H0 . (13.46)

Its generator will be called Ĥ ren (t), so that

 Z t −1  Z t 
φ̂(t, ~x) = Texp −i Ĥ ren (s)ds φ̂(0, ~x)Texp −i Ĥ ren (s)ds , (13.47)
0 0

Formally, we can write for the Hamiltonian and Lagrangian density

Z Z
ren 2
Ĥ (t) = Ĥ(t) − π(0) κ(t, ~x) d~x − C κ(t, ~x)d~x, (13.48)
1 1
Lren (x) = L(x) + π(0)κ(x)2 + Cκ(x). (13.49)
2 2

116