See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/235426843

The Theory of Open Quantum Systems

Book · January 2006

DOI: 10.1093/acprof:oso/9780199213900.001.0001

CITATIONS READS
6,253 19,639

2 authors, including:

Francesco Petruccione
University of KwaZulu-Natal
398 PUBLICATIONS   17,342 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Quantum Tomography View project

Open quantum systems View project

All content following this page was uploaded by Francesco Petruccione on 14 July 2014.

The user has requested enhancement of the downloaded file.

 CERN-OPEN-2005-033
Quantum Theory of Open Systems

15/08/2002
Thesis at ETH Zuerich

Manuel Lorenzo Sentı́s

Supervision: Prof. Dr. Jürg Fröhlich and Dr. Marco Merkli

August 15, 2002

Table of Contents

Table of Contents 1

Acknowledgements 3

Abstract 4

1 Introduction 5
1.1 Description of Physical Systems . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.1 Finite dimensional quantum system coupled to a radiation field . . . 6
1.1.2 Open System under Local Perturbations . . . . . . . . . . . . . . . . 9
1.2 Description of Problems to be Solved . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Summary of Methods to be Used . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Mathematical Theory of Quantum Open Systems 13

2.1 Liouvillian as a Time Evolution Operator . . . . . . . . . . . . . . . . . . . 13
2.2 KMS States and GNS Construction . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.1 Modular Operator and Modular Conjugation . . . . . . . . . . . . . 19
2.2.2 KMS States for the System Atom + Reservoir . . . . . . . . . . . . 22
2.3 Return to Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3 Hilbert Spaces. Positive Commutator Method 26

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 The Araki-Woods Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Glued Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Positive commutator Method . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.5 Positive Commutator Method for physical cases . . . . . . . . . . . . . . . . 36
3.6 Uniqueness of KMS states . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4 Complex Deformations and Feshbach Map 43

4.1 Complex Dilatation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.2 Complex Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3 Feshbach Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

1
 2

5 Renormalization Group Method 56

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2 General Procedure for T > 0 . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2.1 Step 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2.2 Step 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.3 Step 0 of the RG in Second Order in g for T > 0 . . . . . . . . . . . . . . . 64
5.4 RG for T=0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.5 Smooth Feshbach Map. Cuspidal Domains of the Spectrum for T >0 . . . 77

6 The Level Shift Operator 86

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.2 General Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.3 Application to the Eigenvalue Ei − Ej 6= 0 of the Atomic Liouvillian . . . . 95
6.4 Application to the Eigenvalue 0 of the Atomic Liouvillian . . . . . . . . . . 98
6.5 Magnons in a Ferromagnet . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

A Algebraic Techniques 114

B Operator Algebras 117

Bibliography 119
Acknowledgements

This diploma thesis was finished in the summer 2002 at ETH Zürich under the guidance of
Prof. Dr. Jürg Fröhlich. I would like to gratefully thank to Prof. Dr. Jürg Fröhlich for his
help, support and advice and for the time that he has found to guide me. Many grateful
thanks also go to Dr. Marco Merkli for his supporting help and advice. I would like to
thank also Dr. Walid Abou Salem for the revision of the thesis.

3
Abstract

We study a small system with a finite number of energy levels that interchanges energy with
a reservoir, consisting of a bosonic field at temperature T > 0. We will study the dynamics
of this system, which is given by a Liouvillian operator.
It has been proven, that if 0 is a simple eigenvalue of the Liouvillian then the system has the
property of return to equilibrium. That means that the system under small perturbations
approaches its unique equilibrium state, as time tends to ∞.
We will prove that 0 is a simple eigenvalue in leading order perturbation theory and we will
apply these results to spin waves in a ferromagnet.
We will see the principal methods and results used in studying the spectrum of the Liou-
villian.

4
Chapter 1

Introduction

In this diploma thesis we will deal with two classes of open systems, namely a finite di-
mensional open system coupled to a radiation field and an open system under local pertur-
bations. With ”Open system” we mean that the system energy is interchanged with the
exterior. In our case the exterior consists either of a reservoir of quantized energy, or of a
small perturbation of the dynamics of our system (e.g. a local impurity in a lattice or a local
perturbation of the temperature in a bosonic field). We are interested in the dynamical be-
havior of these open systems. In order to study it we will start with a quantum-mechanical
model for the open system and the exterior, write down a model Hamiltonian for the in-
teraction between the two, and then investigate the evolution of the system as determined
by the complete Hamiltonian. For systems at T > 0 we will see in chapter 2 that it will be
easier to use Liouvillians instead of Hamiltonians.
We will see that under appropriate conditions on the Hamiltonian/Liouvillian of the open
system and on the Hamiltonian/Liouvillian of the interaction, the system possesses the
property of return to equilibrium. That is to say the initial excited states of the open
system approach a ground state of equilibrium of the coupled system as the time tends to
infinite. These equilibrium states of the system are called KMS states. We will study also
the uniqueness of these stationary KMS states.

5
 6

1.1 Description of Physical Systems

Our principal model will be an atom or a molecule coupled to a radiation field. We will
study also a model of magnons in a ferromagnet (section 6.4 of chapter 6). As an open
system model under local perturbation we will study a model of a free scalar Bose field
perturbed by a local impurity (section 3.4 chapter 3).

1.1.1 Finite dimensional quantum system coupled to a radiation field

Under ”system” we will understand the system formed by the open system S and by the
reservoir R:

S R

Sytem = Small Open System + Infinitely Extended Reservoir

Fig.1 Diagram of the System

Small System: It consists of a confined atom or molecule with a finite number of energy
levels. The Hilbert space of this small system is:

Hat ' CN N <∞ (1.1)

the Hamiltonian of the small system Hat is given by:

Hat = −4 + V (x) (1.2)

We will suppose that V (x) ≥ 0 and V (x) → 0 as x → ∞. To this Hamiltonian we will

associate an atomic Liouvillian Lat . We will suppose that:

−1
σ(Hat ) = {Ei }N
i=0 E0 < E1 < · · · < EN −1 (1.3)

• • •• • • •
E0 E1 · · ·· EN −1

Fig.2 Spectrum of Hat

 7

We will suppose that the eigenvalues are simple and every eigenvalue Ej corresponds to an
eigenvector ϕj :
Hat ϕj = Ej ϕj (1.4)

These eigenvectors form a complete orthonormal system in Hat , with the scalar product
(·, ·) on Hat :
(ϕi , ϕj ) = δij i, j = 0, 1 . . . N − 1 (1.5)

Reservoir: It consists of the quantized electromagnetic field. The Hilbert space of a photon
is given by:

(1)
hf = L2 (R3 ⊗ C2 , d3 k) (1.6)

The Hilbert space of the reservoir coincides with the Fock space:
∞s
(n)
M
Hf ≡ F := C ⊕ hf h(n) := h⊗s n , n ≥ 1 (1.7)
n≥1

h⊗s n denotes a symmetric tensor product since we are dealing with bosons. The Hamiltonian
of the reservoir is: Z
Hf = dkω(k)a∗ (k)a(k) (1.8)

To this Hamiltonian we will associate a field Liouvillian Lf .

We will deal with physical photons, that means that the dispersion relation of ω(k) is
ω(k) := |k| and the dimension of the photon momentum space is three. Therefore the last
integral is understood as a three dimensional one, but we will write often dk in order to
simplify the notation.
•
0

Fig.3 Spectrum of Hf , σpp (Hf ) = 0 and σac (Hf ) = (0, ∞)

System: Its Hilbert space is:

H = Hat ⊗ Hf (1.9)

On one hand we will consider the uncoupled system (without interaction) with Hamiltonian:

H0 = Hat ⊗ + ⊗ Hf (1.10)
 8

To this Hamiltonian we will associate a Liouvillian L0 .

• • •• • • • •
E0 E1 0 · · ·· EN −1

Fig.4 Spectrum of H0 , σpp (H0 ) = {E0 . . . EN −1 }, σac (H0 ) = [E0 , ∞)

As it can be seen in Fig.4 the spectrum of H0 consists of the union of branches [Ej , ∞)
starting at the eigenvalues Ej which are thresholds of continuous spectrum.
On the other hand we have the coupled system (taking into account the interaction I) with
Hamiltonian:
Hg = H0 + gI g≥0 (1.11)

where g is the coupling constant. To this Hamiltonian Hg we will associate the Liouvillian
of the coupled system Lg .
Our interaction has the form:
Z
I := dk{G(k) ⊗ a∗ (k) + G∗ (k) ⊗ a(k)} G(k) ∈ MN ×N (1.12)

where G(k) is a complex N × N matrix. To the term I we will associate a term W (the
Liouvillian of the interaction).
We will suppose that the matrix G fulfil the following conditions, writing Gij (k) = hϕi |G(k)ϕj i:
H1: The functions Gij (e−θ k) for θ ∈ R extend analytically as functions of θ, on a domain
in C containing the strip:
Σϑ0 := {θ |Imθ| < ϑ0 } (1.13)

for some ϑ0 , independent of ~k ∈ R3 and i, j.

H2: We assume that there exists positive constants µ > 0 and M < ∞, such that for all
θ ∈ Σϑ and ~k ∈ R3 :
0
N
X
|Gi,j (e−θ k)| ≤ eM |Reθ| ω(k)µ (1.14)
i,j=1

where ω(k) = |~k|.

For theorem 2.2.11. we will need the following condition, which introduces a cutoff in the
interaction:
 9

H3: There exists a constant 0 < Λ < ∞ such that, for all θ ∈ Σϑ0 :
N Z
X
|Gi,j (e−θ k)|2 [ω(k) + ω(k)−3 ]dk ≤ e2M |Reθ| Λ2 (1.15)
i,j=1

Magnons in a ferromagnet

In this model we study a ferromagnetic material which suffers a perturbation that is shown
under the form of a spin wave. Our Hamiltonian is:
X X
H = HS + HW = −J Sbx · Sbx+y − j(x)~s0 · Sbx (1.16)
xy x

The first term represents the interaction of nearest neighbors between spin sites in the ferro-
magnetic lattice. The second term represents the interaction between the ferromagnet and
the exterior in form of a spin wave. We will see that under an appropriate transformation
the Hamiltonian Eq. (1.16) is transformed to the form of Eq. (1.11).

1.1.2 Open System under Local Perturbations

We will analyze the properties of infinite systems in or close to thermal equilibrium and their
behavior under small perturbations of their dynamics by coupling them to finite subsystems.

- Open System
Small
Perturbation

Fig.5 Diagram of the System

Free scalar Bose field perturbed by local impurity

In this case we have a lattice and a local perturbation under the form of phonons. We will
use this model to analyze the uniqueness of the equilibrium KMS state.
 10

1.2 Description of Problems to be Solved

Our principal purpose is to analyze the spectral properties of the Liouvillian of the perturbed
system (Lg ).
We know the spectrum of the atomic Liouvillian (Lat ) and the spectrum of the Liouvillian
of the reservoir (Lf ), but we do not know the spectrum of the coupled Liouvillian (Lg =
L0 + gW where L0 := Lat + Lf ). Therefore, there have been proposed several methods
to derive the properties of the Liouvillian of the coupled system from the properties of the
known Liouvillians.
We will assume that Lg is a self-adjoint operator which is proved in [7] or [56]. We will see
that the spectrum of Lg is absolutely continuous on R\0. At the point 0 the spectrum will
be more difficult to analyze. In order to analyze the simplicity of the eigenvalue 0 and the
uniqueness of the equilibrium KMS state there are two methods: the renormalization group
method and the positive commutator method. From the simplicity of the eigenvalue 0 it
follows the property of return to equilibrium. In chapter 6 we will show the simplicity of
the eigenvalue 0 in leading order perturbation theory by computing the level shift operator.
We will also calculate the resonance energies in low-order perturbation theory and we will
apply all these results to the spin relaxation in a magnon environment.
In chapter 3 we will show that the commutator method is also valid for matrix element
√ √
functions Gij (k) that behave as k and 1/ k as k tends to zero. We will also study the
uniqueness of the KMS state using the positive commutator method.
In chapter 5 we will present the renormalization group method and we will deduce the
expressions for the terms E, T, W contributing to our Liouvillian in low-order perturbation
theory. We will also show, by using the smooth Feshbach map, that the domain of the
spectrum of Lg (θ) in the proximity of 0 is cuspidal.

1.3 Summary of Methods to be Used

We will present the different methods used to analyze the spectrum of the Liouvillian of
the perturbed system (Lg ).
In chapter 2 we will present the mathematical formalism of the theory of open systems, the
results of which will be used in successive chapters (above all in chapter 6). In chapter 4
 11

we will do a review about the complex deformations and the Feshbach Map. The complex
deformation method is based on the equation (see Eq. (4.21)):

hϕ|(Lg − z)−1 ψi = hÛ (θ̄)ϕ|(Lg (θ) − z)−1 Û (θ)ψi (1.17)

where Lg (θ) is the complex deformed Liouvillian, this equation is the analyticity (in θ)
property of the complex deformation. If we show that the right hand side is analytic on
z near the real axis, then the left hand side is also analytic for these values of z. On the
other hand we know that since Lg is self-adjoint, it has real eigenvalues. Therefore we can
say that in the values of z on the real axis, where we can make an analytic continuation of
Lg (θ) from the upper half-plane (z + iε) to the lower one (z − iε) for z ∈ R, for these values
of z (Lg − z)−1 exists, with the meaning that the spectrum of Lg is absolutely continuous
for these values of z.
The Feshbach map will permit us to obtain a perturbative analysis for small g of the
spectrum of Lg (θ). We will get an expression of the form:

FP (Lg (θ) − z) = P (Lat − z)P + gP W (θ)P − g 2 Γ(2) (θ, z) + · · · (1.18)

where the spectrum of Lat is known, P is the projection onto the space of the desired
eigenvalue, W is the Liouvillian of the interaction, and Γ(2) (θ, z) is the level shift operator,
Γ(2) (θ, z) := −FP (Lg (θ, z))(2) defined on Eq. (6.26). Using the isospectrality property
(under a Feshbach map transformation the spectrum of our Liouvillian is conserved), we
see from Eq. (1.18) that the spectrum of Lg is approximated by P (Lat − z)P in order g 0
and by P (Lat − z)P + gP W P − g 2 Γ(2) (θ, z) in order g 2 .
 12

Renormalization
Group

Lg -Lg (θ) - F[Lg (θ)]

? - σ[Lg (θ)] -σ[Lg ]
6 6 6 6
6

Isospectrality
Complex F eshbach Analyticity
Def ormation M ap in θ ,Eq. (1.17)
(Level Shif t
Operator)

? - P ositive -
Commutator

Fig.6 Diagram of the methods and properties used for the analysis of the spectrum of Lg

Near 0 we must iterate this method, in other words we apply Feshbach map transformations
successively, which is the essence of the renormalization group method. In order that each
step of the iteration looks like the step before, we add two operations after each Feshbach
map transformation, namely a dilatation and a rescalation of the spectral parameter. The
set of these three operations constitute the renormalization transfomation.
The positive commutator method is based on the inequality:

E∆ (Lg )[Lg , A]E∆ (Lg ) ≥ γE∆ (Lg ) (1.19)

where γ > 0, [·, ·] is the commutator and E∆ (Lg ) is the spectral projector of Lg onto the
interval ∆, which contains only one eigenvalue εj 6= 0 of L0 . The idea is to construct a
antiself-adjoint operator that fulfills this inequality and together with the virial theorem
permits us to say that in ∆ the spectrum of Lg is absolutely continuous.
Chapter 2

Mathematical Theory of Quantum

Open Systems

The purpose of this chapter is to develop the formalism which is necessary in order to
work with the spectral properties of our system. We will use this formalism in coming
chapters when we discuss under which conditions of the spectrum, the states that are local
perturbations of equilibrium states, have the property of return to equilibrium.

2.1 Liouvillian as a Time Evolution Operator

We will work with infinite systems which we will construct as thermodynamic limit of finite
systems. Our C ∗ algebra of observables is given by:

A = ∪ AΛi (2.1)
i∈N

where AΛi is an algebra of observables of a system confined to Λi with AΛi ⊂ B(HΛ )

(B(HΛ ): bounded operators on HΛ ) and HΛ is a separable Hilbert space with Λ being a
bounded region of physical space. It is obvious that:

AΛi ⊆ AΛj ⊆ AΛ for Λi ⊆ Λj (2.2)

The dynamics of the system is determined by a Hamiltonian HΛ , which is a semibounded,

self-adjoint operator on HΛ . A mixed state of the system corresponds to a density matrix
ρ i.e. to a positive self-adjoint operator on HΛ of unit trace, ρ ∈ L1 (HΛ ) where:

13
 14

Definition 2.1.1. L1 (HΛ ) is the two sided-ideal of trace-class operators in B(HΛ )

Definition 2.1.2. K ≡ L2 (HΛ ) is the two sided-ideal of Hilbert-Schmitt operators in

B(HΛ ).

κ := ρ1/2 ∈ K since T r[κ2 ] = T r[κ∗ κ] = T r[ρ] = 1. K is a Hilbert algebra (i.e. as a

linear space it is a Hilbert space and as a algebra it is a ∗-algebra) with scalar product given
by:
hκ1 |κ2 i := T r[κ∗1 κ2 ] (2.3)

In chapter 3 we will see that this Hilbert space is isomorphic to HΛ ⊗ HΛ

Definition 2.1.3. We define a linear representation l of the algebra A on K by:

l[a]κ := aκ ∈ K a ∈ A,κ ∈ K (2.4)

Definition 2.1.4. We define an antilinear (r[za] = z̄r[a] , z ∈ C) representation r of the

algebra A on K by:
r[a]κ := κa∗ ∈ K a ∈ A,κ ∈ K (2.5)

If we had defined r[a]κ = κa (as a linear representation) then we would not have a
representation:
r[a]r[b]κ = r[a](κb) = κba = r[ba]κ 6= r[ab]κ (2.6)

To every element κ ∈ K we can associate a state of the system given by the density matrix:

ρ := hκ|κi−1 κκ∗ (2.7)

then the expectation value of an observable a ∈ AΛ in a state ρ is given by:

haiρ := T r[ρa] = hκ|κi−1 T r[κκ∗ a] = hκ|κi−1 T r[κ∗ aκ] = hκ|κi−1 hκ|l[a]κi (2.8)

We define the time evolution of an observable a ∈ A in the Heisenberg picture as usual by:

αt (a) := eitHΛ ae−itHΛ (2.9)

Let us see now how the states κ ∈ K evolve in the Schroedinger picture:

hκ1 |l[αt (a)]κ2 i = T r[κ∗1 αt (a)κ2 ] = T r[κ1 eitHΛ ae−itHΛ κ2 ] = (2.10)

= T r[(eitHΛ κ1 e−itHΛ )∗ a(eitHΛ κ2 e−itHΛ )] = hα−t (κ1 )|l[a]α−t (κ2 )i (2.11)
 15

Therefore, we can define the time evolution of an element κ ∈ K as:

κt = α−t (κ) := e−itHΛ κeitHΛ (2.12)

Now we define the Liouvillian operator LΛ on K, which for our physical systems is a self-
adjoint operator (for a proof see [7], [56]):

LΛ κ := [HΛ , κ] (2.13)

Proposition 2.1.5. The time evolution of an element κ ∈ K is also (besides of Eq. (2.12))
given by:
κt = e−itLΛ κ (2.14)

The representation on K of the dynamics of the observables is given by a unitary group

generated by L:
l[αt (a)] = eitLΛ l[a]e−itLΛ (2.15)

Proof. We define:
F (t) := e−itHΛ κeitHΛ (2.16)

and it follows:
dF (t) d2 F (t)
= e−itHΛ [κ, HΛ ]eitHΛ = e−itHΛ [ [κ, HΛ ], HΛ ]eitHΛ . . .
dt dt2
dn F (t)
= e−itHΛ [ [· · · [κ, HΛ ], HΛ ], HΛ ]eitHΛ
dtn
then from a Taylor series it follows:

t2
e−itHΛ κeitHΛ = κ − it[HΛ , κ] − [HΛ [HΛ , κ] + · · · = e−itLΛ κ (2.17)
2!
the second relation follows from:

eitLΛ l[a] e|−itL

{z κ} = e
Λ itLΛ
l[a]α−t (κ) = (2.18)
α−t (κ)

and from the definition of l[a]κ and Eq. (2.12):

= eitLΛ aα−t (κ) = αt (aα−t (κ)) = αt (a)κ = l[αt (a)]κ (2.19)

 16

2.2 KMS States and GNS Construction

We are interested in knowing what kind of equilibrium states an infinite system possesses.
Therefore we try to find a property of the equilibrium states of a bounded system which
can help us to characterize the unbounded ones.
From quantum statistics we know that the equilibrium states of a bounded system at inverse
temperature β are given by a density matrix (big canonical ensemble):
−1 −β(HΛ −Q)
ρβ,Q := Zβ,Q e (2.20)

where Q is a conserved charge of the system Q ∈ A0Λ , where A0Λ is the commutant of AΛ (i.e.
the von Neumann algebra of all bounded operators on HΛ that commute with all operators
in AΛ ) and Zβ,Q := T r[e−β(HΛ −Q) ] < ∞.
Taking the square root of ρ:
−1/2
κβ,Q = Zβ,Q e−β(HΛ −Q/2) U (2.21)

where U is an arbitrary unitary operator on HΛ .

The expectation value of an element a ∈ AΛ is from quantum statistics:
−1 −1
haiβ,Q = Zβ,Q T r[ρβ,Q a] = Zβ,Q T r[e−β(HΛ −Q) a] = T r[κβ,Q κ∗β,Q a] = (2.22)
= T r[κ∗β,Q aκβ,Q ] = hκβ,Q |l[a]κβ,Q i (2.23)

where we have used the cyclicity of the trace and in the last step we have used the definition
of scalar product in K Eq. (2.3)). The most important property of these equilibrium states
is:
−1
haαt (b)iβ,Q = Zβ,Q T r[e−β(HΛ −Q) aeitHΛ be−itHΛ ] = (2.24)

using the cyclicity of the trace and the fact that HΛ and b commute with Q
−1 −1
= Zβ,Q T r[eβQ eitHΛ be−(β+it)HΛ a] = Zβ,Q T r[e−β(HΛ −Q) e(β+it)HΛ be−(β+it)HΛ a] = hα−iβ+t (b)aiβ,Q
(2.25)
We make now the following definition:

Definition 2.2.1. A state ϕ over B(HΛ ) is called normal if it satisfies the following prop-
erty. For any bounded increasing net of positive operators Aν over B(HΛ ):

ϕ(sup Aν ) = sup ϕ(Aν ) (2.26)

 17

In the following proposition we see the equivalence between normal states and states
given by a density matrix:

Proposition 2.2.2. The functional ρ over B(HΛ ) (where ρ is a positive operator with trace
1 and the sum does not depend on the orthonormal bases {eν }):
X X
ρ(A) = T r(ρA) = heν |ρAeν i = hρeν |Aeν i (2.27)
ν ν

is a normal state over B(HΛ ). Conversely, any normal state φ over B(HΛ ) is of this form
with ρ uniquely determined by φ.

For a proof see [5].

Definition 2.2.3. We say that a normal state is a (αt , β)-KMS (Kubo-Martin-Schwinger)

state if it satisfies the KMS condition, namely:

haαt (b)iβ,Q = hα−iβ+t (b)aiβ,Q (2.28)

we will denote these KMS states by:

ωβ,Λ (a) := haiβ,Q (2.29)

where we use the parameter Λ, which is the size of the physical space (we need it in order
to take the thermodynamic limit). We recall that this expectation value depends on Λ, and
we omit the dependence on Q which is implicit in the definition of ρ. With this notation
the KMS condition becomes:

ωβ,Λ (aαt (b)) = ωβ,Λ (α−iβ+t (b)a) (2.30)

Proposition 2.2.4. The KMS-states have the following properties:

1. ωβ,Λ (aαt (b)) = ωβ,Λ (α−t (a)b)
2. ωβ,Λ (αt (a)) = ωβ,Λ (a) the KMS-states are time translation invariant.
3. ωβ,Λ (a∗ b) = ωβ,Λ (α−iβ/2 (b)α−iβ/2 (a)∗ )

Proof. the first follows immediately from the cyclicity of the trace:

−1 −1
ωβ,Λ (aαt (b)) = Zβ,Q T r[e−β(HΛ −Q) aeitHΛ be−itHΛ ] = Zβ,Q T r[e−itHΛ e−β(HΛ −Q) aeitHΛ b] =
−1
Zβ,Q T r[e−β(HΛ −Q) e−itHΛ aeitHΛ b] = ωβ,Λ (α−t (a)b)
 18

the second is a consequence of the first for a = the unit element of AΛ . The third property
follows from the time translation invariance and from the KMS condition:

ωβ,Λ (a∗ b) = ωβ,Λ (αiβ/2 (a∗ b)) = ωβ,Λ (αiβ/2 (a∗ )αiβ/2 (b)) =
ωβ,Λ (α−iβ/2 (b)αiβ/2 (a∗ )) = ωβ,Λ (α−iβ/2 (b)α−iβ/2 (a)∗ )

Remark 1. A KMS state has two parameters, namely the one parametric group of auto-
morphisms αt and the inverse temperature β = 1/kT (k: Boltzmann constant). For t = iβ
we can rewrite the KMS condition as:

ωβ,Λ (aαiβ (b)) = ωβ,Λ (ba) (2.31)

where we see that the value β = 0 is different from the other values of β. In this case a
KMS state is a trace state: ωβ,Λ (ab) = ωβ,Λ (ba).

As the size of the system increases, we will obtain a KMS state for an unbounded system
as a limit of a KMS state for a bounded system (Λi ⊆ Λj for i ≤ j):

ω β (a) = lim ωβ,Λi (a) (2.32)

i→∞

using also A = ∪ AΛi we have the following GNS theorem (Gelfand, Naimark and Segal):
i∈N

Theorem 2.2.5. For any time-translation invariant state ω β over a C ∗ algebra A, there
exists a Hilbert space Hβ , a representation l[·] of A on Hβ , a unit vector Ωβ and a continuous
one-parameter group of unitary operators {e−itL }t∈R where L is a self-adjoint operator on
Hβ , satisfying the following conditions:
1. For any a ∈ A
ω β (a) = hΩβ |l[a]Ωβ i (2.33)

2. l[αt (a)] = eitL l[a]e−itL

3. Ωβ is a cyclic vector of the representation l[·] i.e.

l[A]Ωβ = {l[a]Ωβ ; a ∈ A} (2.34)

 19

is dense in Hβ . If ω β is an αt -KMS state then there exists a antiunitary map J from Hβ

to Hβ satisfying:

J eitL J = eitL (2.35)

J l[·]J = r[·] (2.36)
J Ωβ = Ωβ (2.37)

where r is an antilinear representation.

A , αt , ω β - Hβ , e−itL , Ωβ , l[·]

Fig.7 GNS Construction

For a proof see [5], [16], [17].

2.2.1 Modular Operator and Modular Conjugation

We define the modular operator on Hβ (GNS Hilbert space) as:

S(l[a]|Ωβ i) := l[a]∗ |Ωβ i (2.38)

for a ∈ A, S is a bounded antilinear operator. We would be also interested in having an

antiunitary operator J on Hβ which relates the representation l and r. The question of
how to define it, is answered in the next proposition:

Proposition 2.2.6. Let be an operator J on Hβ with

hJ l[a]Ωβ |J l[b]Ωβ i = hl[a]Ωβ |l[b]Ωβ i (2.39)

then:
β
J l[a]|Ωβ i := e− 2 L l[a]∗ |Ωβ i (2.40)

The operator J is called modular conjugation.

 20

Proof.

hJ l[a]Ωβ |J l[b]Ωβ i = hl[a]Ωβ |l[b]Ωβ i = hl[b]Ωβ |l[a]Ωβ i (2.41)

recalling now the GNS construction in particular ω β (a) = hΩβ |l[a]Ωβ i:

hJ l[a]Ωβ |J l[b]Ωβ i = hΩβ |l[b]∗ l[a]Ωβ i = hΩβ |l[b∗ a]Ωβ i = ω β (b∗ a) (2.42)

now we use the KMS condition ω β (aαt (b)) = ω β (α−iβ+t (b)a) a, b ∈ A for t = 0:

ω β (b∗ a) = ω β (α−iβ (a)b∗ ) (2.43)

using now time-translation invariance ω β (αt (a)) = ω β (a) for t = iβ:

ω β (α−iβ (a)b∗ ) = ω β (aαiβ (b∗ )) = (2.44)

recalling now the expression of the time translation operator αt (k) = eitL k k ∈ K for
t = iβ and considering that L is a self-adjoint operator:

ω β (aαiβ (b∗ )) = hΩβ |l[a]l[αiβ (b)]∗ Ωβ i = hΩβ |l[a]e−βL l[b]∗ eβL Ωβ i

β β β β
= hΩβ |l[a]e−βL l[b]∗ Ωβ i = hl[a]∗ Ωβ |e− 2 L e− 2 L l[b]∗ Ωβ i = he− 2 L l[a]∗ Ωβ |e− 2 L l[b]∗ Ωβ i

where in the first line we have also considered that l[αt (b)] = eitL l[b]e−itL , in our case for
t = iβ l[αiβ (b)] = e−βL l[b]eβL . From the first to the second line we have noted that
L|Ωβ i = 0|Ωβ i = 0 and we have developed the exponential in a power series.
Our result is:
β β
hJ l[a]Ωβ |J l[b]Ωβ i = he− 2 L l[a]∗ Ωβ |e− 2 L l[b]∗ Ωβ i (2.45)

and from here it follows the assertion.

Proposition 2.2.7. The modular operator S and the modular conjugation operator J are
Involutions:
JJ = 1 SS = 1 (2.46)

Proof. From the definition of S Eq. (2.38)

SS(l[a]|Ωβ i) := Sl[a]∗ |Ωβ i = l[a]|Ωβ i (2.47)

 21

Because of the ciclycity of Ωβ each vector of Hβ can be written as l[b]Ωβ for b ∈ A. Let
ψn = l[an ]Ωβ be an orthonormal basis, then from the definition of J Eq. (2.40):
For Ωβ :
hΩβ |J J Ωβ i = hΩβ |Ωβ i = 1 (2.48)

For ψm , ψn , ψm = l[am ]Ωβ , ψn = l[an ]Ωβ :

hψm |J J ψn i = hl[a∗m ]Ωβ |J J l[an ]Ωβ i = (2.49)

= hJ −1 l[a∗m ]Ωβ |J l[an ]Ωβ i = he−β/2L l[am ]Ωβ |eβ/2L l[a∗n ]Ωβ i = (2.50)
= hl[am ]Ωβ |l[a∗n ]Ωβ i = hψm |ψn i = δm,n (2.51)

In particular for Hfβ (GNS Hilbert space of the electromagnetic field) we will use in the
next chapters the following simple results:

Definition 2.2.8. Let P be the polynomial algebra generated by {a(f ), a∗ (g)|f, g ∈ S0 (R3 )},
where a and a∗ are the creation and annihilation operators and S0 (R3 ) is the Schwartz space
of functions vanishing at the origin of R3 .

It is easy to see that P is a ∗ algebra for the operation defined by:

(a(f ))∗ := a∗ (τ f ) (2.52)

where τ is the complex conjugation operator.

Proposition 2.2.9. Let be Jf the modular conjugation operator and Lf the Liouvillian of
the field, then:
Jf Lf = −Lf Jf (2.53)

Proof. Let be a ∈ P, then:

β
h i∗
Jf Lf l[a]|Ωf i = Jf Hf , l[a] |Ωf i = e− 2 Lf l Hf , l[a] |Ωf i =
 
(2.54)

where we have considered the definition of the Liouvillian operator Lf l[a] = [Hf , l[a]] and
h i∗  ∗
Eq. (2.40). Now we see that l Hf , l[a] = Hf , l[a] :
β ∗ β β
= e− 2 Lf Hf , l[a] |Ωf i = e− 2 Lf l[a]∗ Hf |Ωf i − e− 2 Lf Hf l[a]∗ |Ωf i =

(2.55)
− β2 Lf
= −Hf e l[a]∗ |Ωf i = −Hf Jf l[a]|Ωf i = −[Hf , Jf l[a]]|Ωf i = Lf Jf l[a]|Ωf i (2.56)
 22

where we have considered Hf |Ωf i = 0 and again Eq. (2.40)

Proposition 2.2.10. Let be P the ∗-algebra before defined, then:

Jf |Ωf i = |Ωf i (2.57)

Proof. the proof is trivial, we use Eq. (2.40) with a = f then:

β β β
∗ β
Jf l[ f ]|Ωf i := e− 2 Lf l[ f ] |Ωf i == e− 2 Lf |Ωβf i = |Ωβf i (2.58)
β
where we have considered that l[ f ]|Ωf i = |Ωβf i and Lf |Ωβf i = 0

2.2.2 KMS States for the System Atom + Reservoir

As an example we will study the KMS state for the system described in section 1.1.1. First
we study the KMS states without perturbation and second with perturbation.
Our algebra would be:
A := Kat ⊗ P (2.59)

where Kat is the algebra of complex N × N matrices and P was defined in 2.2.8.
Without perturbation the KMS state is the tensor product of KMS states for the atom and
for the reservoir:
ω0β = ωat
β
⊗ ωfβ (2.60)

where:
N −1
β
X
−1
ωat = Zβ,0 e−βEj |ϕj ihϕj | (2.61)
j=0
−1
where {ϕj }N
j=1 are the eigenvalues of Hat corresponding to the eigenvalues E0 < E1 < · · · <
EN −1 . And where ωfβ is so defined that the expectation value of the number of photons is
given by (from quantum statistics):
δ(k − k 0 )
ωfβ (a∗ (k)a(k 0 )) = (2.62)
eβω(k) − 1
where a∗ (k), a(k) are the creation and annihilation operators for the photon field. From
this equation it is also easy to derive (using the commutation relations for the creation and
annihilation operators):
δ(k − k 0 ) 0
βω(k) δ(k − k )
ωfβ (a(k)a∗ (k 0 )) = δ(k − k 0 ) + ωfβ (a∗ (k 0 )a(k)) = δ(k − k 0 ) + = e
eβω(k) − 1 eβω(k) − 1
(2.63)
 23

From these considerations it is easy to see that a KMS vector in Hat ⊗ Hf is given by:
N −1
−1/2
Ωβ0 = Zβ,0
X
e−βEj /2 ϕj ⊗ Ωf (2.64)
j=0

where Ωf is the vacuum in Hf and:

ω0β (a) = hΩβ0 |l[a]Ωβ0 i (2.65)

with a ∈ A.
We define:

Lg,l := L0 + l[I] (2.66)

Lg,r := L0 − r[I] (2.67)

With perturbation we have the following theorem:

Theorem 2.2.11. Assume that G fulfils hypothesis H3, Eq. (1.15), then the vector Ωβ0 is
in the domain of the two unbounded operators e−βLg,l /2 and eβLg,r /2 and the vector:
−1/2 −1/2
Ωβg := Zβ,0 e−βLg,l /2 Ωβ0 = Zβ,0 eβLg,r /2 Ωβ0 (2.68)

defines a KMS state ωgβ on A := Kat ⊗ P for the time evolution given by αtg and we have:

Lg Ωβg = 0 (2.69)

that is to say Ωβg is an eigenvector of Lg with eigenvalue 0.

For a proof see [17]

2.3 Return to Equilibrium

The purpose of this section is to establish under which conditions of the spectrum, local
perturbations states (given by normal states) of the coupled system evolve to KMS states
in the course of time. The condition will be, that 0 is a simple eigenvalue of the spectrum
of Lg .
We have already seen that a normal state can be written as:
∞
X
ρ(a) = pn hψn |l[a]ψn i (2.70)
n=1
 24

where a ∈ A, ψn ∈ Hβ an orthonormal system and pn ≥ 0 ∀n. We know also that from the
GNS construction each of the vectors ψn can be approximated in norm by vectors of the
form l[b]Ωβg ( i.e. Ωβg is cyclic) for a specific b ∈ A.

Proposition 2.3.1. Assume that 0 is a simple eigenvalue of Lg corresponding to the eigen-

vector Ωβg and that the rest of the spectrum of Lg is absolute continuous. Let ρ be a normal
state (local perturbation) and a ∈ A. Then:
Z τ
lim ρ±t (a)dt = ωβ (a) return to equilibrium in ergodic mean sense (2.71)
τ →∞ 0

If σ(Lg )\{0} is absolutely continuous, then:

lim ρt (a) = ω β (a) return to equilibrium (2.72)

t→±∞

Proof. We begin with the equality (where w denotes weak-limit):

1 τ ±itLg
Z
w−lim e dt = |Ωβg ihΩβg | (2.73)
τ →∞ τ 0

In order to prove it, suppose first a vector ψ with PΩβ ψ = 0 then it is clear that:
Z τ
1
lim e±itLg dt|ψi = 0 (2.74)
τ →∞ τ 0

and the right side is:

|Ωβg ihΩβg |ψi = 0 (2.75)

Doing the same now with the cyclic vector Ωβg , then:

1 τ ±itLg
Z
lim e dt|Ωβg i = |Ωβg i
τ →∞ τ 0

and the right hand side

|Ωβg ihΩβg |Ωβg i = |Ωβg i (2.76)

where we have used that Lg |Ωβg i = 0.

Now we will prove Eq. (2.71):
Z τ Z ∞
τ X
1 1
lim ρ±t (a)dt = lim pn hψn |l[α±t (a)]ψn i (2.77)
τ →∞ τ 0 τ →∞ τ 0 n=1
 25

and now using that |ψn i = l[an ]|Ωβg i:

Z ∞ Z ∞
1 τX 1 τX
lim pn hψn |l[α±t (a)]ψn idt = lim pn hl[an ]Ωβg |l[α±t (a)]l[an ]Ωβg idt =
τ →∞ τ 0 τ →∞ τ 0
n=1 n=1
∞
Z τX
1
= lim pn hΩβg |l[an ]∗ l[α±t (a)]l[an ]Ωβg idt
τ →∞ τ 0
n=1

In the integrand we have hΩβg |l[an ]∗ l[α±t (a)]l[an ]Ωβg i = hΩβg |l[a∗n α±t (a)an ]Ωβg i = ω β (a∗n α±t (a)an ),
let us develop this expression using first the KMS condition and then that Lg |Ωβg i = 0:

ω β (a∗n α±t (a)an ) = ω β (α−iβ (an )a∗n α±t (a)) =

= hl[an ]l[αiβ (a∗n )]Ωβg |l[α±t (a)]Ωβg i = hl[an ]l[αiβ (a∗n )]Ωβg |e±itL l[(a)]Ωβg i

and finally we have:

Z ∞
τ X
1
lim pn ω β (a∗n α±t (a)an )dt = (2.78)
τ →∞ τ 0 n=1
Z ∞
τ X
1
lim pn hl[an ]l[αiβ (a∗n )]Ωβg |e±itL l[(a)]Ωβg idt (2.79)
τ →∞ τ 0 n=1

and using Eq. (2.73):

Z ∞
τ X
1
lim pn hl[an ]l[αiβ (a∗n )]Ωβg | e|±itL β
{z } l[(a)]Ωg idt = (2.80)
τ →∞ τ 0 n=1 


 (2.81)
∞ 
9
z }| {
X
pn hl[an ]l[αiβ (a∗n )]Ωβg | Ωβg ihΩβg | l[(a)]Ωβg i = ω β (α−iβ (an )a∗n )ω β (a) = (2.82)
n=1
ω β (a∗n an )ω β (a) = ω β (a) (2.83)
P
where we have used pn = 1 and:

ω β (a∗n an ) = hΩβg |l[a∗n an ]Ωβg i = hΩβg |l[a∗n ]l[an ]Ωβg i = (2.84)

hl[an ]Ωβg |l[an ]Ωβg i = hψn |ψn i = 1 (2.85)

For the case that σ(Lg )\{0} is absolutely continuous we have instead of Eq. (2.73):

w−lim e±itLg = |Ωβg ihΩβg | (2.86)

τ →∞

and the proof would be similar.

Chapter 3

Hilbert Spaces. Positive

Commutator Method

3.1 Introduction

In the last chapter we have worked with the Liouvillian of the perturbed system Lg on the
Hilbert space Kat ⊗ Hfβ . Now we would like to work on a Hilbert space related with our
initials Hat and F see Eq. (1.1) and Eq. (1.9). This will be possible with the use of two
isomorphism IC and IT as we will see later. We will see that the Hilbert spaces Kat ⊗ Hfβ
and Hat ⊗ Hat ⊗ F(L2 (R3 , d3 k)) ⊗ F(L2 (R3 , d3 k)) are isomorph.
We will study also the glued Hilbert space F(L2 (R × S 2 , du × dΣ)) which is isomorph to
F(L2 (R3 , d3 k)) ⊗ F(L2 (R3 , d3 k)). In this glued Hilbert space we will analyze the applica-
bility of the Commutator method to interesting physical cases.
In the last section of this chapter we will study the uniqueness of the KMS states with the
help of the positive commutator method.

3.2 The Araki-Woods Hilbert Space

The purpose of this section is to obtain the expression of the Liouvillian Lg on

Hat ⊗ Hat ⊗ F(L2 (R3 , d3 k)) ⊗ F(L2 (R3 , d3 k))

26
 27

where F(X) is the Fock space over X defined for bosons as:
M s⊗n
j=1
F(X) = C ⊕ Xj (3.1)
n≥1

where the s in the tensor product denotes a symmetric tensor product (Bose-Einstein statis-
tics).

Remark 2. The introduction of the polarization of the bosons would be possible with a Fock
space F = ∞ 2 3 3 2 ⊗s n but we will work with a simplified model.
L
n=0 (L (R , d k) ⊗ C )

On F ⊗ F we define the creation and annihilation operators:

a]l (f ) := a] (f ) ⊗ f (3.2)
a]r (f ) := f ⊗ T a] (f )T = f ⊗ a] (τ f ) (3.3)

where a] (f ) are the standard creation and annihilation operators which fulfil the following
commutation relations:

[a∗ (f ), a∗ (g)] = [a(f ), a(g)] = 0 [a(τ f ), a∗ (g)] = hf |gi f (3.4)

Z
]
where a (f ) := d3 kf (k)a] (k) (3.5)

where τ is the complex conjugation operator τ f = f for f ∈ L2 (R3 , d3 k). We define an

antiunitary operator T on F the second quantization of τ :

T Ω = Ω; T a] (f )T := a] (τ f ); T = T ∗ = T −1 (3.6)

where Ω is the vacuum vector characterized by the property:

a(f )|Ωi = 0 for all f ∈ L2 (R3 , d3 k) (3.7)

Theorem 3.2.1.
The Map: IT : Hβf 7−→ H
b := F ⊗ F (3.8)

√
IT lf [a(f )]IT−1 := al ( 1 + ρf ) + a∗r ( ρf )
p
defined by: (3.9)
√
IT rf [a(f )]IT−1 := a∗l ( ρτ f ) + ar ( 1 + ρτ f )
p
(3.10)
IT Ωfβ = Ω ⊗ Ω (3.11)

is an isometric isomorphism.
 28

Proof. From this definition it follows that IT l[a] (f )]IT−1 and IT r[a] (f )]IT−1 satisfy the same
canonical commutation relations like Eq. (3.4) which we will check with an example, taking
into account Eq. (3.4) and that different representations r and l commute :

√ √
[IT lf [a(f )]IT−1 , IT lf [a∗ (f )]IT−1 ] = (al ( 1 + ρf ) + a∗r ( ρf ))(a∗l ( 1 + ρf ) + ar ( ρf )) −
p p

√ √
−(a∗l ( 1 + ρf ) + ar ( ρf ))(al ( 1 + ρf ) + a∗r ( ρf )) = al ( 1 + ρf )a∗l ( 1 + ρf ) −
p p p p

√ √
−a∗l ( 1 + ρf )al ( 1 + ρf ) + al ((
p p p ((( p ((( (
(1(+(ρf
( )a
(r(( ρf ) − ar (((ρf
( )a
( l(( (1 + ρf ) +
√ ∗ hhhhh∗ √ ∗ √ √ √ ∗ √
a∗r ( ρf )a
hhh∗hp
h hp
l ( h1h+ hρf
h h) − al ( 1 + ρf )arh h) + ar ( ρf )ar ( ρf ) − ar ( ρf )ar ( ρf ) =
( hρf
h
(1 + ρ)hf |f i · f − ρhf |f i · f = hf |f i · f

It is easy to see the relations:

IT lf (p1 )IT−1 + IT lf (p2 )IT−1 = IT lf (p1 + p2 )IT−1 (3.12)

IT lf (p1 )IT−1 IT lf (p2 )IT−1 = IT lf (p1 p2 )IT−1 (3.13)

∀p1 , p2 ∈ P where P is the polynomial algebra generated by {a(f ), a∗ (g)|f, g ∈ L2 (R3 , d3 k)}.
With these relations we see that lf [P] is ∗-homomorphic to IT lf [P]IT−1 and in a similar way
rf [P] is ∗-homomorphic to IT rf [P]IT−1 . We will proof now that It is an isometry:

hΩ ⊗ Ω|IT lf [a∗ (k)]IT−1 IT lf [a(k 0 )]IT−1 Ω ⊗ Ωi =

√
√

hΩ ⊗ Ω| a∗l ( 1 + ρ) + ar ( ρ) al ( 1 + ρ) + a∗r ( ρ) Ω ⊗ Ωi =
p p

√((
hΩ ⊗ Ω|a∗l (( ∗ ∗(
p p (
(((
p (
(1(+(ρ)a
(( l ( 1 + ρ)Ω ⊗ Ωi + hΩ ⊗ Ω|al (( (1(+ (ρ)a(( r ( ρ)Ω ⊗ Ωi +
√ √ √
1 + ρ)Ω ⊗ Ωi + hΩ ⊗ Ω|ar ( ρ)a∗r ( ρ)Ω ⊗ Ωi =
p ((( (
hΩ ⊗ Ω|ar ((( ρ)a
( l
( ( ((

δ(k − k 0 )
ρ(k)ρ(k)δ(k − k 0 ) = = ωβf (a∗ (k)a(k 0 )) = hΩfβ |lf [a∗ (k)]lf [a(k 0 )]Ωfβ i (3.14)
p
=
eβω(k) − 1
likewise we would proof:

hΩ ⊗ Ω|IT rf [a∗ (k)]IT−1 IT rf [a(k 0 )]IT−1 Ω ⊗ Ωi = hΩfβ |rf [a∗ (k)]rf [a(k 0 )]Ωfβ i (3.15)

every vector ψ ∈ Hβf can be approximated by vectors of the form l[P]Ωfβ because of the
cyclicity of Ωfβ . Therefore the isomorphism between operators on Hβf and operators on
F ⊗ F means finally that IT is a isometry isomorphism between vectors in Hβf and vectors
in F ⊗ F.
 29

Now we will find the form of the Liouvillians Lf , Lat , Lg on F ⊗ F, Hat ⊗ Hat and
Hat ⊗ Hat ⊗ F ⊗ F respectively. Afterwards we will obtain the form of the interaction on
the gluing Hilbert space Hat ⊗ Hat ⊗ F(L2 (R × S 2 , du × dΣ)).

Proposition 3.2.2. The Liouvillian of the field Lf on the GNS Hilbert space of the field
Hβf becomes Lf on the Araki-Woods Hilbert space F ⊗ F with:
Z
Lf = dkω(k)[a∗l (k)al (k) − a∗r (k)ar (k)] (3.16)

Proof. As an operator Lf transforms like:

hΩ ⊗ Ω|Lf Ω ⊗ Ωi = hΩ ⊗ Ω|IT Lf IT−1 Ω ⊗ Ωi = hΩ ⊗ Ω| IT l[Hf ]IT−1 − IT r[Hf ]IT−1 Ω ⊗ Ωi


Z Z
dkω(k)IT l[a∗ (k)a(k)]IT−1 − dkω(k)IT r[a∗ (k)a(k)]IT−1 Ω ⊗ Ωi =

= hΩ ⊗ Ω|
Z
l[ab]=l[a]l[b]
= hΩ ⊗ Ω| dkω(k) IT l[a∗ (k)]IT−1 IT l[a(k)]IT−1 − IT r[a∗ (k)]IT−1 IT r[a(k)]IT−1 Ω ⊗ Ωi =


using now Eq. (3.9) and Eq. (3.10) and the commutation relations (using also that different
representations commute):

√
√

Z
= hΩ ⊗ Ω| ω(k)dk a∗l ( 1 + ρ) + ar ( ρ) al ( 1 + ρ) + a∗r ( ρ) −
 p p

√ √
− al ( ρ) + a∗r ( 1 + ρ) a∗l ( ρ) + ar ( 1 + ρ) Ω ⊗ Ωi =
p
p

(3.17)

√ √ √ √
Z
dkω(k) a∗l ( 1 + ρ)al ( 1 + ρ) − al ( ρ)a∗l ( ρ) + ar ( ρ)a∗r ( ρ)
 p p
hΩ ⊗ Ω|
√ √
−a∗r ( 1 + ρ)ar ( 1 + ρ) + a∗l ( 1 + ρ)a∗r ( ρ) + ar ( ρ)al ( 1 + ρ)
p p p p

√
√
−al ( ρ)ar ( 1 + ρ) − a∗r ( 1 + ρ) a∗l ( ρ) Ω ⊗ Ωi
p p
(3.18)

Z
dkω(k) (1 + ρ)a∗l (k)al (k) − ρal (k)a∗l (k) + ρar (k)a∗r (k)

= hΩ ⊗ Ω|

−(1 + ρ)a∗r (k)ar (k) + (1 + ρ)ρa

ph hhhh ph hhhh
(k)a
rh h− (1 + ρ)ρal (k)a
l (k) r (k)
hh hhh
h hh
∗ ∗ ∗ ∗
p (
( p (
(
− (1 (k)ar (k) Ω ⊗ Ωi =
( (
(+(ρ)ρa (k)al (k) + (1 (+(ρ)ρa
((r( ( ((l( (
( Z (
= hΩ ⊗ Ω| dkω(k)[a∗l (k)al (k) − a∗r (k)ar (k)]Ω ⊗ Ωi
 30

We can construct an isomorphism between Kat , the two sided-ideal of Hilbert-Schmitt

operators in B(Hat ) and Hat ⊗ Hat , where Hat is the Hilbert space of states of the small
open system. Namely:
IC : Kat 7−→ Hat ⊗ Hat (3.19)

If κ = |ψ1 ihψ2 | ∈ Kat then

IC κ := ψ1 ⊗ Cψ2 = |ψ1 ihCψ2 | ∈ Hat ⊗ Hat (3.20)

where C is an antiunitary involution on Hat , that is to say:

C2 = hCφ|Cϕi = hϕ|φi = hφ|ϕi (3.21)

Proposition 3.2.3. The Liouvillian of the atom Lat on the Kat Hilbert space becomes Lat
on the Hat ⊗ Hat Hilbert space with:

Lat = Hat ⊗ at − at ⊗ CHat C (3.22)

Proof. We recall the definitions of the representations of the observable algebra A on K

Eq. (2.4) and Eq. (2.5): l[a]κ := aκ and r[a]κ := κa∗ we have:

Lat = IC Lat IC−1 = IC (l[Hat ] − r[Hat ])IC−1 (3.23)

IC l[Hat ]κ = IC Hat |ψ1 ihψ2 | = |Hat ψ1 ihCψ2 | = (Hat ⊗ at )(|ψ1 ihCψ2 |) = (Hat ⊗ at )IC κ

(3.24)
IC r[Hat ]κ = IC |ψ1 ihψ2 |Hat = |ψ1 ihCHat ψ2 | = ( at ⊗CHat C)(|ψ1 ihCψ2 |) =( at ⊗CHat C)IC κ

(3.25)
Lat IC κ = IC Lat IC−1 IC κ = IC (l[Hat ] − r[Hat ])κ (3.26)

and substituting now Eq. (3.24) and Eq. (3.25)

Lat IC κ = (Hat ⊗ at )IC κ −( at ⊗ CHat C)IC κ (3.27)

In particular for C = T because of Hat T = T Hat and T 2 = at we have:

Lat = Hat ⊗ at − at ⊗ CHat C = Hat ⊗ at − at ⊗ Hat (3.28)

 31

The same procedure as in the proof shows us that:

IC (l[G] − r[G])IC−1 = G ⊗ at − at ⊗ CGC (3.29)

where l, r are defined in Eq. (2.4) and Eq. (2.5).

We will use the following notation:

Gl (k) := G ⊗ at (3.30)
Gr (k) := at ⊗ T G(k)T (3.31)

We will show now the last step in order to obtain the complete Liouvillian Lg . The Liou-
villian of the interacting system is defined on Hat ⊗ F by:

Lg := L0 + g(l[I] − r[I]) (3.32)

where: Z
I= dk[G(k) ⊗ a∗ (k) + G∗ (k) ⊗ a(k)] (3.33)

and:      
l lat lf
 = ⊗  (3.34)
r rat rf
I0 := IC ⊗ IT (3.35)

Proposition 3.2.4. The Liouvillian of the interacting system Lg on the Kat ⊗ Hβf Hilbert
space becomes Lg on the Hat ⊗ Hat ⊗ F ⊗ F Hilbert space with:

Lg = Lat + Lf + gI0 (l[I] − r[I])I0−1 (3.36)

where:
√ i √ io
Z n h h
I0 l[I]I0−1 dk G⊗ ⊗ a∗l (k) 1 + ρ+ar (k) ρ +G∗ (k)⊗ ⊗ al (k) 1 + ρ+a∗r (k) ρ
p p
=
(3.37)
√ √
Z n h i h io
I0 r[I]I0−1 ∗ ∗ ∗
p p
= dk ⊗T GT ⊗ al (k) ρ+ar (k) 1 + ρ + ⊗T G T ⊗ al (k) ρ+ar (k) 1 + ρ
(3.38)

Proof. Using Eq. (3.35):

−1
I0 Lat ⊗ f I0 = IC Lat IC−1 ⊗ f = Lat ⊗ f (3.39)
I0 at ⊗ Lf I0−1 = at ⊗ IT Lf IT−1 = at ⊗ Lf (3.40)
 32

Using now that:

Z
l[I] = dk[lat [G(k)] ⊗ lf [a∗ (k)] + lat [G∗ (k)] ⊗ lf [a(k)]] (3.41)

and a similar equation for r[I]. Transforming this operators with I0 :

Z
I0 l[I]I0−1 = IT dk IC lat [G(k)]IC−1 ⊗lf [a∗ (k)] + IC lat [G∗ (k)]IC−1 ⊗lf [a(k)] IT−1

(3.42)
| {z } | {z }
Gl (k) G∗l (k)
Z
I0 r[I]I0−1 = IT dk IC rat [G(k)]IC−1 ⊗rf [a∗ (k)] + IC rat [G∗ (k)]IC−1 ⊗rf [a(k)] IT−1 (3.43)

| {z } | {z }
Gr (k) G∗r (k)

dkGl (k) ⊗ lf [a∗ (k)]IT−1 = IT lf [a∗ (Gl (k))]IT−1 and similar equations. We
R
using now that IT
make the following definition (analog to Eq. (3.9), Eq. (3.10)):

√
IT lf [a(G)]IT−1 := al ( 1 + ρG) + a∗r ( ρG)
p
(3.44)

√
IT rf [a(G)]IT−1 := a∗l ( ρτ G) + ar ( 1 + ρτ G)
p
(3.45)

and substituting into Eq. (3.42)) and Eq. (3.43)):

I0 l[I]I0−1 = IT lf [a∗ (Gl (k))]IT−1 + IT lf [a(G∗l (k))]IT−1 (3.46)

I0 r[I]I0−1 = IT rf [a∗ (Gr (k))]IT−1 + IT rf [a(G∗r (k))]IT−1 (3.47)

Substituting this expressions and Eq. (3.30), Eq. (3.31) in the last equations give us the
desired results.

3.3 Glued Hilbert Space

The purpose of this section is to obtain the expression of the Liouvillian Lg on Hat ⊗ Hat ⊗
F(L2 (R × S 2 , du × dΣ)).

Theorem 3.3.1. There is a isometric isomorphism that identifies:

F(L2 (R3 , d3 k)) ⊗ F(L2 (R3 , d3 k)) ' F(L2 (R × S 2 , du × dΣ)) (3.48)

where S 2 is the unit sphere and dΣ is the element of solid angle.

 33

Proof. First we see that it exists a trivial isometric isomorphism between F(L2 (R3 , d3 k)) ⊗
F(L2 (R3 , d3 k)) and F(L2 (R3 , d3 k) ⊕ L2 (R3 , d3 k)) by doing the correspondence:

a∗ (f1 ) · · · a∗ (fm ) ⊗ a∗ (g1 ) · · · a∗ (gn ) 7−→ a∗ ((f1 , 0)) · · · a∗ ((fm , 0))a∗ ((0, g1 ) · · · a∗ ((0, gn ))
(3.49)
Now we will prove the following isometric isomorphism F(L2 (R3 , d3 k) ⊕ L2 (R3 , d3 k)) '
F(L2 (R × S 2 , du × dΣ)) we define the map jϕ,ψ : ∀f, g ∈ F(L2 (R3 , d3 k):

jϕ,ψ : (f, g) 7−→ h where: (3.50)


 eiϕ uf (u, α) : if u≥0
h(u, α) = (3.51)
 eiψ ug(−u, α) : if u<0

Isometry: Let be
k(f, g)k2L2 ⊕L2 := kf k2L2 + kgk2L2 (3.52)
Z
khk2L2 (R×S 2 ,du×dΣ) = |h(u, α|2 dudΣ =
R×S 2
Z Z
= |eiϕ |2 u2 |f (u, α)|2 dudΣ + |eiψ |2 u2 |g(−u, α)|2 dudΣ =
R+ ×S 2 R− ×S 2
Z Z
= |eiϕ |2 u2 |f (u, α)|2 dudΣ + |eiψ |2 u2 |g(u, α)|2 dudΣ = kf k2L2 + kgk2L2 (3.53)
R+ ×S 2 R+ ×S 2
−1
Isomorphism: Given a h we look for f, g so that jϕ,ψ (h) = (f, g) ∈ L2 (R+ ×S 2 )⊕L2 (R+ ×S 2 ),
in fact:
1
f (u, α) = e−iϕ h(u, α) for u > 0 (3.54)
u
1
g(u, α) = e−iψ h(−u, α) for u < 0 (3.55)
u

and finally we can write the expression of the Liouvillian Lg on Hat ⊗ Hat ⊗ F(L2 (R ×
S 2 , du × dΣ)):

Proposition 3.3.2. The expression of the Liouvillian Lg on Hat ⊗ Hat ⊗ F(L2 (R × S 2 , du ×

dΣ)) is given by:

Lg = Lat + Lf + a∗ (g1 ) + a(g1∗ ) + a(g2 ) + a∗ (g2∗ ) (3.56)

 34

where:
 p
 1 + ρ(u)uG(u, α) : if u ≥ 0
g1 (u, α) := (3.57)
 ρ(−u)uG∗ (−u, α) : if u < 0
p
 p
 ρ(u)uG(u, α) : if u ≥ 0
g2 (u, α) := (3.58)
 1 + ρ(−u)uG∗ (−u, α) : if u < 0
p

g1 , g2 ∈ F(L2 (R × S 2 , du × dΣ))

Proof. Using Eq. (3.51), and Eq. (3.2), Eq. (3.3) we can write our Liouvillian in the following
form:

√ √
Lg = Lat + Lf + a∗l ( 1 + ρGl ) + ar ( ρGl ) + al ( 1 + ρG∗l ) + a∗r ( ρG∗l ) −
p p
(3.59)
√ √ ∗ ∗

al ( ρGr ) + a∗r ( 1 + ρGr ) − a∗l ( ρGr + ar ( 1 + ρGr )
p
p
(3.60)

where we have used the definition:

Z
]
a (G) := d3 kG(k) ⊗ a] (k) (3.61)

the proof consists simply in transforming each of the eight terms of Eq. (3.60) using Eq. (3.2),
Eq. (3.3) and the gluing transformation:

 a∗ (√1 + ρuG ) : if

∗
p ∗
p l u≥0
al ( 1 + ρGl ) → (a ( 1 + ρGl ), 0) → (3.62)
 0 : if u<0


√ √ ∗  0 : if u≥0
a∗r ( ρG∗l ) → (0, a∗ ( ρGl ) → √ ∗ (3.63)
 a∗ ( ρuG ) : if u<0
l

From these two terms we obtain a∗ (g1 ).


√ √  0 : if u≥0
ar ( ρGl ) → (0, a( ρGl ) → √ (3.64)
 a( ρuGl ) : if u<0

 a(√1 + ρuG∗ ) : if

p ∗
p ∗ l u≥0
al ( 1 + ρGl ) → (a( 1 + ρGl ), 0) → (3.65)
 0 : if u<0
 35

From these two terms we obtain a(g1∗ ).

 a(√ρG ) : if

√ √ r u≥0
al ( ρGr ) → (a( ρGr ), 0) → (3.66)
 0 : if u<0

∗
 0 : if u≥0
ar ( 1 + ρGr ) → (0, a( 1 + ρG∗r )) →
p p
√ (3.67)
 a( 1 + ρG∗ ) : if u<0
r

From these two terms we obtain a(g2 ), taking into account that Gr = ⊗ T GT = ⊗G
∗
and G∗r = ⊗G .
 a∗ (√ρG∗ ) : if

√ ∗ √ ∗ r u≥0
a∗l ( ρGr ) → (a∗ ( ρGr ), 0) → (3.68)
 0 : if u<0

 0 : if u≥0
a∗r ( 1 + ρGr ) → (0, a∗ ( 1 + ρGr )) →
p p
√ (3.69)
 a∗ ( 1 + ρGr ) : if u<0

From these two terms we obtain a∗ (g2∗ ).

As a last remark we see from Eq. (3.51) that we can multiply by phase factors in the
expression of g1 and g2 :

 eiϕ p1 + ρ(u)uG(u, α) : if u≥0
g1 (u, α) := (3.70)
 eiψ ρ(−u)uG∗ (−u, α)
p
: if u<0

g2 (u, α) = −g1 (−u, α) (3.71)

3.4 Positive commutator Method

In this method we look for an antiself-adjoint operator A that fulfils the named
positive commutator estimate (PC estimate):

E4 (Lg )[Lg , A]E4 (Lg ) ≥ γE4 (Lg ) (3.72)

with γ > 0 and E4 is the spectral projector of Lg onto the interval 4, which contains an
eigenvalue εj 6= 0 of L0 but no other eigenvalue of L0 . In Eq. (3.72) [Lg , A] is understood
in the sense of quadratic forms on D(N 1/2 ) that is ∀ϕ, ψ ∈ D(Lg ) ∩ D(A)

hψ|[Lg , A]ϕi := hLg ψ|Aϕi − hAψ|Lg ϕi.

 36

If Eq. (3.72) is satisfied then Lg has not eigenvalues in 4, because let us suppose that
Lg ψ = λψ with λ ∈ 4 then E4 (Lg )ψ = ψ and the PC estimate gives hψ|[Lg , A]ψi ≥ γ, we
have taken kψk = 1. But on the other hand we have:

hψ|[Lg , A]ψi = hψ|[Lg − λ, A]ψi = h(Lg − λ)ψ|Aψi + hAψ|(Lg − λ)ψi =

= h(Lg − λ)ψ|Aψi + h(Lg − λ)ψ|Aψi = 2Reh(Lg − λ)ψ|Aψi = 0 (3.73)

which contradicts the PC estimate, that is to say in ∆ there is no eigenvalue of Lg , the

spectrum of Lg is absolutely continuous in ∆. However this proof is not completely valid
because we are dealing with unbounded operators (Lg and A). The problem is that ψ could
be in D(Lg ) but not in D(A) or viceversa and then [Lg , A] = Lg A − ALg can not be defined
for ψ. Therefore we study the ψ ∈ D(Lg ) ∩ D(A) and we hope that D(Lg ) ∩ D(A) is dense
in our Hilbert space. That is to say in order to study ψ[Lg , A]ψ with ψ 6∈ D(Lg ) ∩ D(A) we
take ψn ∈ D(Lg ) ∩ D(A) so that ψn → ψ.
In case that 4 contains the eigenvalue 0 we have to exclude this eigenvalue because we
know (see theorem 2.2.11.) that this eigenvalue remains eigenvalue of the perturbed system.
Therefore if Ωβλ is an eigenvector of the eigenvalue 0 and PΩβλ is the projector onto this
eigenvector, the PC estimate would be modified as:

E4 (Lg )PΩ⊥βλ [Lg , A]PΩ⊥βλ E4 (Lg ) ≥ γE4 (Lg )PΩ⊥βλ (3.74)

with again γ > 0.

For further reading see [56], [57]

3.5 Positive Commutator Method for physical cases

In [56], [57] it is proven that the commutator method is applicable for cases in which the
matrix elements of G(u) behave for u → 0 as up with p > 2. In this section we will see that
we can also apply the commutator method for cases in which our matrix element functions
√ √
G behave like u or 1/ u as u tends to zero. These two cases are usual in the physical
applications.
First we will try to simplify a little the expressions of g1 and g2 of Eq. (3.70) and Eq. (3.71)
 37

in the glued Hilbert space. We will take first ϕ = ψ = 0 and then we will vary these values:
 p
 1 + ρ(u)ug(u, α) : if u ≥ 0
g1 (u, α) = p (3.75)
 ρ(−u)uḡ(−u, α) : if u < 0
g2 (u, α) = −g1 (−u, α) (3.76)

1
where ρ(u) = , for u ≥ 0 we have:
eβu
−1
s
eβu u √
r
p βu/2
g1 (u, α) = 1 + ρ(u)ug(u, α) = βu
ug(u, α) = e βu
ug(u, α) (3.77)
e −1 e −1

and for u < 0:

r r
p 1 √ √ 1
g1 (u, α) = ρ(−u)uḡ(−u, α) = uḡ(−u, α) = (−1) −u −uḡ(−u, α) =
e−βu
−1 −1 e−βu
√ u √
r r
u βu/2
=− −uḡ(−u, α) = −e −uḡ(−u, α)
1 − e−βu eβu − 1
that is to say:
√

u≥0
r
βu/2 u  ug(u, α) : if
g1 (u, α) = e βu
(3.78)
e − 1  − (−u)ḡ(−u, α)
p
: if u<0
| {z }
h(u,α)
r
u 1
βu/2
by u → 0 we have only problems with h(u, α) because: e → √ . On the
−1 β eβu
other hand h(u, α) is an analytic function of u except maybe at 0 where we must study the
behavior. This behavior depends on how g(u, α) behaves at 0.
We will deal now with the two cases mentioned before:
u→0 √
1 Case: g(u, α) −→ u

√ √ √
 
 ug(u, α) : if u≥0 u→0
 u u : if u≥0 u→0
h(u, α) = p −→ p p −→
 − (−u)ḡ(−u, α) : if u<0  − (−u) (−u) : if u<0

u→0
 u : if u≥0 u→0
−→ −→ 0 (3.79)
 −u : if u<0

In this case for g1 (u, α) we do not have problems.

u→0 1
2 Case: g(u, α) −→ √
u
 38


√
 1 √
 ug(u, α) : if u≥0

 u√ : if u≥0
h(u) =
u→0
−→ u u→0
−→
p
 − (−u)ḡ(−u, α) p 1
: if u<0  − (−u) √
 : if u<0
−u

u→0
 1 : if u≥0
−→ (3.80)
 −1 : if u<0
and we have a discontinuity at point 0, which implies that the first derivative does not belong
to L2 . Therefore we could not apply the positive commutator method to this case. The
way to repair this problem, is to multiply by a phase factor our function h(u, α) without
changing its norm (this is possible because of the fact that the difference in behavior of
h(u, α) for u ≥ 0 and u < 0 is only a sign).
√

 eiϕ ug(u, α) : if u≥0
h(u, α) = p (3.81)
 −eiψ (−u)ḡ(−u, α) : if u<0

and we choose ϕ = 0 and ψ = π then we would have for u → 0:

√

 e0 ug(u, α) : if u ≥ 0 u→0
h(u, α) = p −→ 0 (3.82)
 −eiπ (−u)ḡ(−u, α) : if u < 0

 1 : if u≥0
(3.83)
 1 : if u<0
in the same way we see:
√

r
−u  −ug(−u, α) : if u≥0
g2 (u, α) = −g1 (−u, α) = −e−βu/2 (3.84)
e−βu − 1  − (u)ḡ(u, α) : if
p
u<0
| {z }
h0 (u)

u→0 √
1 Case: g(u, α) −→ u

 √−ug(−u, α) √ √
 
0 : if u ≥ 0 u→0  −u −u : if u≥0
h (u, α) = −→
 (−1)√uḡ(u, α) : if u<0  −√u√u : if u<0

u→0
 u : if u≥0 u→0
−→ −→ 0 (3.85)
 −u : if u<0
 39

In this case for g2 (u, α) we do not have problems.

u→0 1
2 Case: g(u, α) −→ √
u

 √−ug(−u, α) : if u ≥ 0
 
u→0
 1 : if u≥0
h0 (u) = −→ (3.86)
 −√uḡ(u, α) : if u < 0  −1 : if u<0

Which implies as before that the first derivative does not belong to L2 . We solve again this
problem with phase factors ϕ = 0 and ψ = π:

√
 
 −ug(−u, α) : if u ≥ 0 u→0  1 : if u≥0
h0 (u, α) = √ −→ (3.87)
 eiπ (−1) uḡ(u, α) : if u<0  1 : if u<0

3.6 Uniqueness of KMS states

With the positive commutator method we will study now the uniqueness of the KMS state
for a model without atomic liouvillian that is to say:

Lg = Lf + g(l[I 0 ] − r[I 0 ]) (3.88)

where I 0 is a local perturbation.

From theorem 2.2.11. we know that we can construct (by means of a Dyson-series) an
eigenvector of the operator Lg for the eigenvalue 0:

Lg Ωβg = 0 (3.89)

In the same way we construct a vector Ω̂βg = I0 Ωβg , with I0 := IC ⊗ IT as in Eq. (3.35), in
the Hilbert space Hat ⊗ Hat ⊗ Hf ⊗ Hf , without interaction we have (g = 0):

N −1
−1/2
Ω̂β0 = Zβ,0
X
e−βEj /2 ϕj ⊗ ϕj ⊗ Ωf ⊗ Ωf (3.90)
j=0

and with interaction:

−1/2 −1/2
Ω̂βg := Zβ,0 e−βLg,l /2 Ω̂β0 = Zβ,0 eβLg,r /2 Ω̂β0 (3.91)

We have:
0 = Lg Ωβg = I0−1 Lg I0 Ωβg = I0−1 Lg Ω̂βg (3.92)
 40

that means:
Lg Ω̂βg = 0 (3.93)

That is to say Ω̂βg is an eigenvector of Lg for the eigenvalue 0.

Let Pβ be the orthogonal projector onto Ω̂βg .
Suppose that we can construct an antiself-adjoint operator Aβ with:

Pβ⊥ [Lg , Aβ ]Pβ⊥ ≥ γPβ⊥ (3.94)

for a γ > 0.
From the relation:
∞ Z β/2 Z τ1 Z τn−1
hΩ̂β0 |e−βLg,l /2 Ω̂β0 i dτn hΩβ0 |I(τ1 )I(τ2 ) · · · I(τn )Ωβ0 i
X
n
= g dτ1 dτ2 · · ·
n=0 0 0 0
(3.95)
we see that:
 
 β β 
hΩ̂g |Ω̂0 i ≥ 1 − O(gβ) (3.96)

Let us suppose that we have more than one KMS vectors Ω̂β,j β,1 β
g for j = 1 . . . k with Ω̂g = Ω̂g

and kΩ̂β,j
g k = 1 i.e.:

Lg Ω̂β,j
g = 0, ∀j (3.97)

Proposition 3.6.1.
β 2
|hΩ̂β,j
g |Ω̂0 i| ≥ 1 − const(gβ)/γ, ∀j (3.98)

Proof. It is clear that:

hΩ̂β,j β,j
g |[Lg , Aβ ]Ω̂ i = 0 (3.99)

Using now that Pβ + Pβ⊥ = :

0 = hΩ̂β,j β,j
g |[Lg , Aβ ]Ω̂g i (3.100)
= hPβ Ω̂β,j β,j
g |[Lg , Aβ ]Pβ Ω̂g i (I) (3.101)
⊥ β,j
+2RehPβ Ω̂β,j
g |[Lg , Aβ ]Pβ Ω̂g i (II) (3.102)
+hPβ⊥ Ω̂β,j ⊥ β,j
g |[Lg , Aβ ]Pβ Ω̂g i (III) (3.103)

Let us analyze the different terms. We begin with I:

I = (1 − const(gβ)) hΩ̂β0 |[L0 , Aβ ]Ω̂β0 i +ghΩ̂β0 |[l[I] − r[I], Aβ ]Ω̂β0 i


(3.104)
| {z }
=0
 41

Now with II:

n
⊥ β,j
II = 2Re hPβ Ω̂β,j
g |Lg Aβ Pβ Ω̂g i (IIA) (3.105)
o
β,j ⊥ β,j
−hPβ Ω̂g |Aβ Lg Pβ Ω̂g i (IIB) (3.106)

β(j)
noting that L0 Pβ Ω̂g = 0, then:

⊥ β,j
IIA = 2Re hPβ Ω̂β,j
g |L0 Aβ Pβ Ω̂g i + (3.107)
| {z }
=0
⊥ β,j
+2gRehPβ Ω̂β,j
g |(l[I] − r[I])Aβ Pβ Ω̂g i = O(g) (3.108)

using Pβ⊥ = − Pβ :

IIB = −2Re hPβ Ω̂β,j β,j

g |Aβ Lg Ω̂g i + (3.109)
| {z }
=0
+2RehPβ Ω̂β,j
g |Aβ (L0 + g(l[I] − r[I])Pβ Ω̂β,j
g i = O(g) (3.110)

and finally III, applying the commutator estimate:

⊥ β,j
III ≥ γhΩ̂β,j
g |Pβ Ω̂g i (3.111)

Taking all the results together:

0 = I + II + III (3.112)
β 2
= O(g) + O(g) + γ[1 − |hΩ̂β,j
g |Ω̂0 i| ] (3.113)

and therefore:
β 2
γ[1 − |hΩ̂β,j
g |Ω̂0 i| ] ≤ const(gβ) (3.114)

and finally:
β 2
|hΩ̂β,j
g |Ω̂0 i| ≥ 1 − const(gβ)/γ (3.115)

From this proposition follows that taking an adequate g we can have:

β 3
|hΩ̂β,j
g |Ω̂0 i| ≥ ∀j (3.116)
4
 42

and from this inequality follows the uniqueness of the KMS state with the following reason-
ing:
We have Ω̂β,j β
g = αj Ω̂0 + χ
(j) with χ(j) ⊥ Ω̂β and |α | ≥
0 j
3
4 then, we have:

9 1
|hΩ̂β,i β,j (i) (j)
g |Ω̂g i| = |ᾱi αj + hχ |χ i| ≥ − |hχ(i) |χ(j) i| ≥ (3.117)
16 | {z } 8
7
≤ 16
 Chapter 4

Complex Deformations and

Feshbach Map

In this chapter we will see the method of complex deformations which consists in deforming
the spectrum of our Liouvillian in order to show that some parts of our spectrum are of
absolute continuous nature. The idea is that in case that the spectrum is of absolute nature
in a interval of the axis, we can make an analytical continuation of the matrix elements of
the resolvent of our Liouvillian near the axis in this interval from the upper half plane to
the lower half plane. We will see two of such deformations, namely the complex translations
and the complex dilatations.
Before we begin with the complex deformations, we will present the spectral information
that we know about the Liouvillians. We can deduce the following diagrams from Eq. (3.16)
and Eq. (3.28) and from Fig.2 and Fig.3:

σ(Lat ) = • • • • • • • ; Lat = Hat ⊗ − ⊗ Hat

0
; Lf = dkω(k)[a∗l (k)al (k) − a∗r (k)ar (k)]
R
σ(Lf ) = •
0
σ(L0 ) = • • • • • • • ; L0 = Lat + Lf
0
Fig.8

where the thick lines denote an absolute spectrum of the operator, for example Lat has
only pure point spectrum, Lf has pure point spectrum at 0 and the rest of the real axis

43
 44

is absolute continuous. For example if ϕi are eigenvectors of Hat for the eigenvalues Ei ,
i = 1 . . . N − 1 then:

Lat ϕi ⊗ϕj = {Hat ⊗ − ⊗Hat }ϕi ⊗ϕj = Hat ϕi ⊗ϕj −ϕi ⊗Hat ϕj = (Ei −Ej )ϕi ⊗ϕj (4.1)

which explains why the pure point spectrum of Lat and also for L0 are symmetric.

4.1 Complex Dilatation

For θ ∈ R we define the following transformation:

(u(θ)f )(x) := e−3θ/2 f (e−θ x) f ∈ L2 (R3 , d3 x) (4.2)

Proposition 4.1.1. For θ ∈ R is u(θ) an unitary transformation

Proof.
Z Z
ku(θ)f k2L2 = dxe−3θ |f (e−θ x)|2
dx|(u(θ)f )(x)| = 2
R3 R3
 Dx 
doing the substitution y = e−θ x and Jacobian   = e3θ
 
Dy
Z  Dx  Z
2  −3θ 2
ku(θ)f kL2 = dy  e |f (y)| = dy|f (y)|2 = kf k2L2

R3 Dy R3

We define now U (θ) as the operator obtained from u(θ) by the process of second quan-
tization. Let be:

ψ∈F i.e. ψ = {ψn }n≥0 with ψn (k1 , k2 , . . . kn ) ∈ L2 (R3n , d3n k) (4.3)

In the sector n the dilatation is defined as:

[U (θ)ψ]n (k1 , k2 , . . . kn ) := e−(3θ/2)n ψn (e−θ k1 . . . e−θ kn ) (4.4)

We now define the group of dilatations on the Araki-woods Hilbert space:

Û (θ) := at ⊗ at ⊗ U (θ) ⊗ U (θ)−1 (4.5)

Proposition 4.1.2. Under this unitary transformation the field-Liouvillian transforms like:

Lf (θ) := Û (θ)Lf Û (θ)−1 = e−θ (Hf ⊗ f) − eθ ( f ⊗ Hf ) (4.6)

 45

Proof. Let us see first how Hf transforms, we recall:

n
X
[Hf ψ]n (k1 , k2 , . . . kn ) = |kj |ψn (k1 , k2 , . . . kn ) (4.7)
j=1

[U (θ)Hf U (θ)−1 ψ]n (k1 , k2 , . . . kn ) = e−(3θ/2)n [Hf U (θ)−1 ψ]n (e−θ k1 , e−θ k2 , . . . e−θ kn ) =
Xn
−(3θ/2)n
=e |e−θ kj |[U (θ)−1 ψ]n (e−θ k1 , e−θ k2 , . . . e−θ kn ) =
j=1
−1
noting that U (θ) = U (−θ) we have:
n
X n
X
= e−(3θ/2)n |e−θ kj |e(3θ/2)n ψn (k1 , k2 , . . . kn ) = e−θ |kj |ψn (k1 , k2 , . . . kn ) =
j=1 j=1

= e−θ |Hf ψ|n (k1 , k2 , . . . kn )

and that means:

U (θ)Hf U (θ)−1 = e−θ Hf (4.8)

With this result we have:

 
Û (θ)Lf Û (θ)−1 = at ⊗ at ⊗ U (θ) ⊗ U (θ)−1 at ⊗ at ⊗ Hf ⊗ f − at ⊗ at ⊗ f ⊗ Hf ×
⊗ U (θ) ⊗ U (θ)−1 =
× at ⊗ at
 
( at ⊗ at ) ⊗ U (θ)Hf U (θ)−1 ⊗ U (θ)−1 U (θ) − U (θ)U (θ)−1 ⊗ U (θ)−1 Hf U (θ) =
=( at ⊗ at ) ⊗ (e−θ Hf ⊗ f − f ⊗ eθ Hf )

With this expression for Lf (θ) we can easily deduce the expression for L0 (θ):

L0 (θ) = Lat + Lf (θ) = Lat + e−θ (Hf ⊗ ) − eθ ( f ⊗ Hf ) (4.9)

We recall that the unitary transformation Û (θ) = at ⊗ at ⊗ U (θ) ⊗ U (θ)−1 will not affect
Lat because of the fact that the two first factors are unit operators. We can develop a little
more the expression of Lf using the equation eθ = cosh θ + sinh θ then:

Lf (θ) = e−θ (Hf ⊗ ) − eθ ( f ⊗ Hf ) = (4.10)

= (cosh θ − sinh θ)(Hf ⊗ ) − (cosh θ + sinh θ)( f ⊗ Hf ) = (4.11)
= cosh θ(Hf ⊗ − f ⊗ Hf ) − sinh θ(Hf ⊗ + f ⊗ Hf ) = (4.12)
= Lf cosh θ − Laux sinh θ (4.13)
 46

where we have defined:

Laux = Hf ⊗ f + f ⊗ Hf (4.14)

and finally we have:

L0 (θ) = Lat + Lf cosh θ − Laux sinh θ (4.15)

We will use this last equation in the successive chapters. Let us suppose now that θ ∈
/ R,
we take θ = iϑ with ϑ > 0. We will analyze the spectrum of the different terms of Eq. (4.6):

σ(Hf )
e−iϑ (Hf ⊗ ) : •
@ ϑ
@
@
@
@ −iϑ H ) = e−iϑ σ(H )
@ σ(e f f
@
@

σ(eiϑ Hf ) = eiϑ σ(Hf )

Kϑ
σ(Hf )
eiϑ ( f ⊗ Hf ) : •

Fig.9 Rotation of the spectrum of Hf . Top: rotated by an angle of −ϑ. Bottom: rotated by an angle of ϑ

Superposing both diagrams and taking into account the minus sign of Eq. (4.6) that
makes the second diagram go from the first quadrant to the third one we will have:
 47

0 0
σ(Lf (iϑ)) = • + • =
@
@
@
@
@
@
@
@ 0
= • ...
.......
...........
...............
...................
....................................
@
..............
...............................
...................................
.......................................
@
...........................................
. .......................................................................
..................................................... @
.........................................................
.............................................................
.................................................................
.................................................................... @
............................................................................................................
......................................
...............................................................................
................................................................................... @
.......................................................................................
...........................................................................................
................................................................................................................................................
@
..................................................
.......................................................................................................
...........................................................................................................
@
@

Fig.10 Top: Superposition of diagrams of Fig.1

Bottom: the shaded area corresponds to the spectrum of Lf (θ) for Reθ = 0 and Imθ = ϑ > 0

The shaded area of figure 10 is the spectrum of Lf (iϑ). In the same way we would obtain
the spectrum of L0 (iϑ) by doing the same procedure as before ( before it was done for the
eigenvalue 0) for each of the embedded eigenvalues of L0 .
−ε1 −ε0 0 ε0 ε1
[ ∆ ]

Fig.11 The shaded area corresponds to the spectrum of L0 (θ) for Reθ = 0 and Imθ = ϑ > 0

One can see that there is a dense domain D ⊂ H

b with the property that for every ψ ∈ D,
Û (θ)ψ is an analytic function of θ for θ ∈ Σ π2 , i.e. for |θ| < π/2. We can consider for
example:
D := span{ϕl ⊗ ϕr ⊗ ψl ⊗ ψr |ϕl/r ∈ Hat , ψl/r ∈ DGauss } (4.17)

where DGauss consists of translations and dilatations of Gaussians:

DGauss := {a∗ (f1 ) · · · a∗ (fn )Ω|fj (~k) = exp[−(~k − ~kj )2 /2σj2 , kj ∈ R3 , σj > 0} (4.18)

We consider now the function:

f (θ) = hÛ (θ̄)ϕ|(L0 (θ) − z)−1 Û (θ)ψi (4.19)

 48

for θ ∈ R, we see that:

hÛ (θ̄)ϕ|(L0 (θ) − z)−1 Û (θ)ψi = hϕ|Û (−θ)(L0 (θ) − z)−1 Û (θ)ψi =
 −1
= hϕ| Û (−θ)(L0 (θ) − z)Û (θ) ψi = hϕ|(Û (−θ)L0 (θ)Û (θ) − z)−1 ψi = hϕ|(L0 − z)−1 ψi

This equality is valid for all θ ∈ R. From the analytic continuation principle it follows that
it is also valid for θ ∈ C. On the other hand from the equation:

hϕ|(L0 − z)−1 ψi = hÛ (θ̄)ϕ|(L0 (θ) − z)−1 Û (θ)ψi (4.20)

we see that the r.s. is an analytic function of z for z ∈ ρ(L0 (θ) (resolvent of L0 (θ)). This
equation permits us to do an analytic continuation in z of the left hand side of Eq. (4.20)
for the values of z in the resolvent set of L0 (θ). With these considerations we have proved
the following proposition:

Proposition 4.1.3. Let ∆ be an interval, which does not contain any eigenvalue εi of L0 ,
that is to say:
for ∆ beginning on the positive half axis, there exists a j with ∆ ⊆ (εj + δ , εj+1 − δ)
for ∆ beginning on the negative half axis, there exists a j with ∆ ⊆ (−εj+1 + δ , −εj − δ)
with 0 < δ < mini,j |εi − εj | i, j = −1, 0 . . . and where ε−1 = −ε−1 = 0 (see Fig.11),
then L0 has an absolutely continuous spectrum on ∆.

We can do a similar analysis for the operator Lg in which case we will have instead of
Eq. (4.20):
hϕ|(Lg − z)−1 ψi = hÛ (θ̄)ϕ|(Lg (θ) − z)−1 Û (θ)ψi (4.21)

The problem now is that we do not know the spectrum of Lg (θ) = L0 (θ) + gW (θ). If we
knew the spectrum of Lg (θ) we could formulate a similar proposition as proposition 4.1.3.
for the intervals ∆ on the real axis, which do not contain any eigenvalue of Lg (θ). In order
to know the spectrum of Lg (θ) we will use the Feshbach map method. This method will
permit us to know this spectrum except near 0 (see Fig.14), where the Feshbach map only
gives us an incomplete information. In order to know the spectrum near 0 completely we
will use the renormalization group method.
Let us now analyze W (θ). The expression of W (θ) is given by the following proposition:
 49

Proposition 4.1.4. The expression of W (θ) on the Hilbert space Hat ⊗ Hat ⊗ F ⊗ F is:
Z q q
−3θ/2 ∗
W (θ) = e dk{ 1 + ρ(e−θ k)Gl (e−θ k) − ρ(e−θ k)Gr (e−θ k)}a∗l (k) +
Z q q
−3θ/2
+e dk{ 1 + ρ(e−θ k)G∗l (e−θ k) − ρ(e−θ k)Gr (e−θ k)}al (k) −
Z q q
−e 3θ/2
dk{ 1 + ρ(eθ k)Gr (eθ k) − ρ(eθ k)G∗l (eθ k)}a∗r (k) −
Z q q
3θ/2 ∗
−e dk{ 1 + ρ(eθ k)Gr (eθ k) − ρ(eθ k)Gl (eθ k)}ar (k) (4.22)

Proof. The proposition follows immediately from Eq. (3.37), Eq. (3.38) and from Û (θ) :=
at ⊗ at ⊗ U (θ) ⊗ U (θ)−1 .

we try now to write this expression in a easy way, therefore we define:

q
µθ (k) := ρ(e−θ k) (4.23)
q
νθ (k) := (1 + ρ(e−θ k) (4.24)

and defining:

∗
Gθ+,l := e−3θ/2 {νθ (k)Gl (e−θ k) − µθ (k)Gr (e−θ k)} (4.25)
Gθ−,l := e−3θ/2 {νθ (k)G∗l (e−θ k) − µθ (k)Gr (e−θ k)} (4.26)

Gθ+,r := e3θ/2 {µ−θ (k)G∗l (eθ k) − ν−θ (k)Gr (eθ k)} (4.27)
∗
Gθ+,r := e3θ/2 {µ−θ (k)Gl (eθ k) − ν−θ (k)Gr (eθ k)} (4.28)

and the expression of W (θ) becomes:

X X
W (θ) = a∗l (Gθ+,l ) + al (Gθ−,l ) + a∗r (Gθ+,r ) + ar (Gθ−,r ) = aστ (Gθσ,τ ) (4.29)
σ=± τ =l,r

We will use this notation in the next chapter with the renormalization group method.
We know the spectrum of L0 (θ) but we do not know the spectrum of W (θ). Being gW (θ)
a perturbation, we can hope that a perturbative analysis could help us and so it is. We will
see it in the section of the Feshbach map method.
 50

4.2 Complex Translation

−ε1 ε0 [ ∆ ] ε1
• • • • • R
6
δ∝T
? ? ?

Fig.12 Translated spectrum of L0

In this case the deformation of the spectrum consists in translating the continuous part of
the spectrum downwards and the pure point spectrum remains on the real axis. We define
the unitary translation group:

(u(a)f )(x) := f (x + a) ≡ f a (s) f ∈ L2 (R3 , d3 x) (4.30)

It is trivial to see that this defines an unitary transformation. We denote by U (a) = Γ(u(a))
the second quantization of u(a).

Definition 4.2.1. Let be Σδ := {z ∈ C : |Imz| < δ}, we define H 2 (δ, H) as the Hilbert
space of all functions f : Σδ 7−→ H which are analytic in Σδ and satisfy:
Z ∞
2
kf kH 2 (δ,H) := sup kf (x + ia)k2H dx < ∞ (4.31)
|a|<δ −∞

We need also the following definition:

Definition 4.2.2. Given a function on R3 we define a function fe on R×S 2 by the formula:


 u1/2 f (u, α) : if u ≥ 0
fe(u, α) = (4.32)
 −|u|1/2 f (−u, α) : if u < 0

The principal result with complex translations is:

Theorem 4.2.3. Suppose that the form factor G satisfies:

H1: (k + k −1 )G(k) ∈ L2 (R3 )
e ∈ H 2 (δ, H). Then:
H2: There exists δ > 0 such that G
There exists a dense subspace S ⊂ H and for each η ∈ (0, δ) a constant Λ(δ) > 0 such that
for g ∈ (−Λ(δ), Λ(δ)) and φ, ϕ ∈ S, the functions:

z 7−→ hφ|(Lg − z)−1 ϕi (4.33)

 51

have a meromorphic continuation from the upper half-plane onto the region:

O := {z : Im(z) > −η} (4.34)

Moreover, let εj be a simple eigenvalue of Lat , then the eigenvalues of Lg satisfy εj (g) =
εj + g 2 a2j + O(g 4 ), where aj is a number related to the level shift operator.

For a proof see [46].

From this theorem we can formulate a result similar to prop. 4.1.3. for an interval ∆, which
does not contain any eigenvalue (see Fig.12).
If we write the Liouvillian in the gluing Hilbert space as explained in chapter 3 in Eq. (3.56)
we obtain functions g 0 s of the form (see also Eq. (3.78)):
√

u≥0
r
βu/2 u  ug(u, α) : if
g1 (u, α) = e βu
(4.35)
e − 1  − (−u)ḡ(−u, α)
p
: if u<0

i.e.:
u 1/2

g1 (u, α) = eβu/2 ge1 (u, α) (4.36)
eβu − 1
where we have used the definition Eq. (4.32) of fe. Therefore we see that if ge1 ∈ H 2 (δ, H) for
some δ > 2π/β then g1 ∈ H 2 (δ − ε, H) for any 0 < ε < 2π/β but g1 ∈
/ H 2 (δ + ε, H) for any
ε. Therefore we need δ < 2π/β in order to satisfy the hypothesis of the last theorem. And
that means that our gap δ (see Fig.12) depends directly proportional on the temperature,
which forbids us the use of a limiting argument to analyze the zero temperature case. This
is one of the problems of this method (the non-uniformity in the temperature). There is
another one, namely the impossibility of applying the last theorem for form factors G(k)
that behave for k → 0 like:
|k|1/2+µ µ small (4.37)

in which case we have to use complex dilatation followed generally by one Feshbach map
or more Feshbach maps (RG). However for form factors which behave when k → 0 as |k|α
with α = ± 12 , 32 , 52 , · · · we do not need to change the method and the complex translations
with analytic perturbation theory give good results.
 52

4.3 Feshbach Map

The idea of the Feshbach map FP is that it permits us to restrict the study of the spectrum
of the operator Lg (θ) on H to the study of the operator FP (Lg (θ)) on a smaller space
P H without loss of information in the spectrum. In other words z ∈ σ(Lg (θ)) iff 0 ∈
σ(FP (Lg (θ) − z))). Besides, the form of the Feshbach map is quite appropriated in order
to do a perturbative analysis with the perturbation gW (θ).
Let Lg (θ) be a closed operator on a Hilbert space H and let P be a closed bounded projection
operator whose range is in the domain of Lg (θ). We define:

Lg (θ) := P Lg (θ)P P = −P (4.38)

Let z belong to the resolvent set of Lg (θ)|P H . We assume that the operators P Lg (θ)P |Lg (θ)−
z|−1/2 and |Lg (θ)−z|−1/2 P Lg (θ)P are bounded. Then we can define an operator FP (Lg (θ)−
z) at Lg (θ) − z associated to the projection P as an operator acting on the Hilbert space
P H:

FP (Lg (θ) − z) := P [(Lg (θ) − z) − g 2 P Lg (θ)P (Lg (θ) − z)−1 P Lg (θ)]P (4.39)

The following theorem characterizes the important property of the isospectrality of the
Feshbach map:

Theorem 4.3.1. Assume that L̄g (θ) = P̄ Lg (θ)P̄ is invertible on P̄ H and that
(P̄ Lg (θ)P̄ )−1 P̄ , P W P̄ (P̄ Lg (θ)P̄ )−1 , (P̄ Lg (θ)P̄ )−1 P̄ W P, P W P̄ (P̄ Lg (θ)P̄ )−1 P̄ W P, P W P
all extend to bounded operators on Hf ⊗ (Hat ⊗ Hat ), then:
a) z is an eigenvalue of Lg (θ) iff 0 is an eigenvalue of FP (Lg (θ) − z) and the multiplicity
of z ∈ σpp (Lg (θ)) is the same as the multiplicity of 0 ∈ σpp (FP (Lg (θ) − z)).
b) The operator FP (Lg (θ) − z) is invertible on P H iff Lg (θ) − z is invertible on H. In this
case [FP (Lg (θ) − z)]−1 = P (Lg − z)−1 P and for ϕ, ψ ∈ P H and z 6∈ σ(Lg (θ)):

hϕ|(Lg (θ) − z)−1 ψi = hϕ|[FP (Lg (θ)) − z]−1 ψi. (4.40)

For a proof see [14] or [8].

As projection, we will use:
P := Pεat ⊗ Pρaux (4.41)
 53

where ε is any eigenvalue of Lat (we recall ε = Ei − Ej where Ej are eigenvalues of Hat for
j = 0 . . . N − 1 ) and Pρaux is defined by:

Pρaux := χ[Laux < ρ] ρ>0 (4.42)

where χ is a characteristic function:


 1 : L
aux < ρ
χ[Laux < ρ] = (4.43)
 0 : otherwise

We define now a family of subsets which will help us studying the spectrum of Lg (θ). In
the next figure (Fig.13) all the domains are represented.
  sin ϑ
S := z ∈ CImz > − ρ0 (4.44)
4
we will use:
ρ0 := g 2− ρ1 := g 2+/2 (4.45)

with  small.
We divide this domain in different parts:
  ρ0
S> := z ∈ S  dist(Rez, σ[Lat ]) ≥ (4.46)
2
  ρ 0
Sij := z ∈ S  |Rez − Ei + Ej | ≤ (4.47)
2
  sin ϑ ρ0
S0> := z ∈ S  ρ1 ≤ |z| ≤ (4.48)
2 2
  sin ϑ
S0< := z ∈ S  |z| ≤ ρ1 (4.49)
2
It is evident that:
S ⊆ S> ∪ (∪i6=j Sij ∪ S0> ∪ S0< ) (4.50)

The easiest way to visualize these domains is to look at the next figure:

S0<
S> S> S>
S0>
S12 S21
6ρ sin ϑ/4
0
?

Fig.13 The subsets S> , S0> , S0< and Sij

 54

We will now expose the principal results of the analysis of the spectrum of Lg (θ) (σ[Lg (iϑ)]).
For a proof refer to [7].
We will take:
θ = iϑ 0 < ϑ00 ≤ ϑ ≤ ϑ0 < π/2 (4.51)

Theorem 4.3.2. For g > 0 sufficiently small and for any z ∈ S>

σ[Lg (iϑ)] ∩ S> = ∅ (4.52)

Theorem 4.3.3. For g > 0 sufficiently small and for any z ∈ Sij , it exists a positive
constant γ > 0, such that:

Sij ∩ {z ∈ C|Imz > −g 2 γ} ⊆ ρ[Lg (θ)] (4.53)

Theorem 4.3.4. For g > 0 sufficiently small and for any z ∈ S0> \C(ϑ0 ) where C is the
cone:
C(ϑ0 ) := {|Rez| ≤ − cot ϑ0 Imz} ϑ0 < ϑ (4.54)

then:
S0> ∩ {z ∈ S0> \C(ϑ0 )} ⊆ ρ[Lg (θ)] (4.55)

The mechanism of proving these last three theorems is always the same. We prove first,
by means of a Neumann series expansion, that in the desired domain the Feshbach map is
well defined. Then we analyze the invertibility of the Feshbach map (i.e. [FP (Lg (θ) − z)]−1 )
in this domain. The values of z where the FM is invertible belong to the resolvent set of
Lg (θ), ρ[Lg (θ)] (theorem 4.3.1.). The values of z where the FM is not invertible belong to
the spectrum of Lg (θ), σ[Lg (θ)] (theorem 4.3.1.).
With these results we infer that the spectrum of Lg (θ) in S is of the following form (where
the points indicate that the spectrum continues to the left and to the right in the same
form):
[ ∆ ]
6γg 2 6
?
ρ0 sin ϑ/4
D
?

Fig.14 The spectrum of Lg (θ) after the application of Theorems 4.3.2., 4.3.3., 4.3.4.
 55

From this diagram and similarly to the deduction of proposition 4.1.3. we deduce the
following proposition:

Proposition 4.3.5. The spectrum of Lg is absolutely continuous in each interval ∆ (see

Fig.14) on the real axis, if this interval does not contain the point zero and is sufficiently
far away from it (see Fig.14).

In order to analyze the domain D of Fig.14 we need new methods (the positive com-
mutator method and the renormalization group method). The question is to know whether
the N-fold eigenvalue 0 of L0 displaces also downwards or remains in the same position
under the complex deformation. In the next chapter we will show that the spectrum near
0 is cuspidal. We will also show in second order perturbation theory (in chapter 6) that 0
is a simple eigenvalue of Lg (θ). Therefore N-1 of the eigenvalues equal to 0 of L0 displace
downwards by the process of complex dilatation and one remains in the same position as
by L0 (θ) (see next figure).
[ ∆ ]
6γg 2 6
?
ρ0 sin ϑ/4

Fig.15 The spectrum of Lg (θ)

Chapter 5

Renormalization Group Method

5.1 Introduction

We have seen that the values z ∈ S0> \C(ϑ0 ) are in the resolvent set of Lg (θ). Now the
question is, what is the spectrum of Lg (θ) near 0 (forz ∈ S0< ) like? To answer this question
we use successive Feshbach maps and rescalations of z that permit us to analyze the spec-
trum of Lg (θ) in the proximity of 0. This structured set of operations repeated iteratively
is known as renormalization group method (RG), which we will present in this Chapter.
Mainly we will see it for T > 0 and we will see the differences with T = 0. Also we will define
the smooth Feshbach Method (SFM) and use it to show that the domain of the spectrum
of Lg (θ) is cuspidal near 0.

5.2 General Procedure for T > 0

The RG is an iterative process, that is to say, in each step we will repeat the same operations
as we did in the step before, over our in each moment effective Liouvillian. In this section
we will present this operations for the two first steps and for the case that T > 0. The
purpose is to enumerate both the operations and the generated terms. In the next section
we will make for the effective Liouvillian of step 0 a perturbative analysis in order g 2 and
we will see how these terms generate.

56
 57

L
e 0 [z] L0 [z] L
e 1 [z] L1 [z]
Lg (θ)
E
e0 Te0 Wf0 E0 T 0 W 0 E
e1 Te1 Wf1 E1 T 1 W 1

Fig.16 Renormalization Group Method

5.2.1 Step 0

Each step of the renormalization Group consists of the same three operations. In this sub-
section we will show them for the step 0:
1 Feshbach Map

FP0 (Lg (θ) − z) = P0 [(Lg (θ) − z) − g 2 P0 W (θ)P 0 (Lg (θ) − z)−1 P 0 W (θ)]P0 (5.1)

where:

P0 := P0at ⊗ χ[Laux < ρ0 ] (5.2)

N
X −1
P0at = |ϕi ⊗ ϕi ihϕi ⊗ ϕi | (5.3)
i=0

With this transformation we obtain a new effective Liouvillian L

e 0 [z] see Eq. (5.5). This is an
operator on a smaller vector space P0 H. This means that this transformation is a decimation
of degrees of freedom of particle and photons with energies ≥ ρ0 , where o < ρ0 < 1 was
defined in Eq. (4.45).
Because of the isospectrality condition z is eigenvalue of Lg (θ) if and only if 0 is eigenvalue
of FP0 (Lg (θ) − z).
This Feshbach Map is only defined where (L̄g (θ) − z)−1 exists. Now we develop Eq. (5.1)
using a Neumann series. The details will be presented in the next section:
∞
X P0
FP0 (Lg (θ) − z) = (L0 (iϑ) − z)P0 + (−1)ν−1 g ν P0 W ( W P0 )ν−1 (5.4)
L0 (θ) − z
ν=1
 58

defining:
e 0 [z] − z := FP (Lg (θ) − z)
L (5.5)
0

denoting L = (Lf , Laux ) and r = (rf , raux ); after doing a Wick ordering (see appendix), we
get an expression of the form:

L
e 0 [z] = P0 (E
e0 [z] + Te0 [z, L] + W
f0 [z])P0 (5.6)

with:

e0 [z] := 4E 0
E e0 [z] = ω
e0,0 [z, 0] (5.7)
0 0
Te0 [z, L] := Lf (θ) + 4Te0 [z, L] = Lf (θ) + {eω0,0 [z, r] − ω
e0,0 [z, 0]} (5.8)
X
f0 [z] := gW + 4W
W f0 [z] = gW + f (0)
W (5.9)
m,n
m+n≥1

e0 [z] ∈ C is a number, Te0 [z, Lf ] is a function of Lf and W

where E f0 [z] is a sum of Wick
monomials of the form:
Z
f (0) [z] :=
W m,n dk (m) de
k (n) a+ (k (m) )e 0
ωm,n [z; L; k (m) ; k (n) ]a− (e
k (n) ) (5.10)

where we use the notation:

a+ (k) := τ [a∗ (k)] (5.11)

a− (k) := τ [a(k)]; τ = l, r (5.12)

and ω 0
em,n is given by:

∞
X X M
X Z
0
ω
em,n [z; L; k (M ) ; k (N ) ] = − (−1)ν−1 (g)ν dx(p) Sm,n {Fb [x(p) , x(p) k (m) k (n) ]}dx̃(p)
p=0 b ∈Bm,n,p α1 ...αν−1
(5.13)
where Bm,n,p denotes the set of partitions b = (bk , bk̃ , bx , bx̃ ) of {1, 2, . . . , m + n + 2p},
such that the number of elements of bk is m , |bk | = m, |bk̃ | = n and |bx | = |bx̃ | = p.
bk , bk̃ , bx , bx̃ are pairwise disjoint subsets of {1, 2, . . . , m + n + 2p} the union of which give
{1, 2, . . . , m + n + 2p}. Denoting J := m + n + 2p the matrix valued function Fb is defined
as:

(0) (α ) (α ) (0)
Fb [x(p) , x(p) k (m) , k (n) ] := χat χat 1 · · · χat ν−1 χat hΩ|Gθb (1, X (p,p) , K (m,n) )×
 59

ab (1, X (p,p) , K (m,n) )fα1 [L + r] · · · fαν−1 [L + r]Gθb (J, X (p,p) , K (m,n) )ab (J, X (p,p) , K (m,n) )Ωi
(5.14)
where:



 Gθ+ (kl ) : if j is the lth member of bk

 Gθ (k̃ )

: if j is the lth member of bk̃
− l
Gθb (j, X (p,p) , K (m,n) ) := (5.15)
 θ
G+ (xl ) : if j is the lth member of bx



 Gθ (x̃ )

: if j is the lth member of bx̃
− l

The expressions for Gθ± were given in Eq. (4.29) and related equations, where we take:
X
Gθ± := G±,τ (5.16)
τ =l,r

and finally:



 1 : if j ∈ bk ∪ bk̃
ab (j, X (p,p) , K (m,n) ) := a+ (xl ) : if j is the lth member of bx (5.17)

a− (x̃l )

: if j is the lth member of bx̃


and where Sm,n denotes the symmetrization operator:

1 X
Sm,n := F [X (p,p) , kπ(1) . . . kπ(m) ; k̃π̃(1) . . . k̃π̃(m) ] (5.18)
m!n! π∈Sm
π̃∈Sn

In section 5.3. we will deduce these equations in order g 2 which can help to understand
them. 2 Dilatation and Rescalation of energies and momenta
Like in Eq. (4.2) we can define a general dilatation as:

(uρ f )(x) := ρ3/2 f (ρx) f ∈ L2 (R3 , d3 x) (5.19)

like in case of complex dilatation (ρ = e−θ ) it is easy to see that this is an unitary trans-
formation. We will take 0 < ρ < 1. We can define now Uρ by the process of second
quantization as:
[Uρ ψ]n (k1 , k2 , . . . kn ) := ρ(3/2)n ψn (ρk1 . . . ρkn ) (5.20)

and as in Eq. (4.8) we have:

Uρ Hf Uρ−1 = ρHf (5.21)
 60

It is easy to see that the same transformation is defined by:

3
Uρ a∗ (k)Uρ∗ = ρ− 2 a∗ (ρ−1 k) (5.22)
3
Uρ a(k)Uρ∗ = ρ− 2 a(ρ−1 k) (5.23)
Uρ Ωf = Ωf (5.24)

The interaction Eq. (1.12) transforms to:

Z
∗
dk G(k) ⊗ Uρ a∗ (k)Uρ∗ + G(k)∗ ⊗ Uρ a(k)Uρ∗ =

at ⊗ Uρ I at ⊗ Uρ =
Z
3 3
dk G(k) ⊗ ρ− 2 a∗ (ρ−1 k) + G(k)∗ ⊗ ρ− 2 a(ρ−1 k)

=

making the substitution k 0 = ρ−1 k and taking into account that the integrals are three-
dimensional:
Z
3 3
Uρ∗ ρ3 dk 0 G(ρk 0 ) ⊗ ρ− 2 a∗ (k 0 ) + G(ρk 0 )∗ ⊗ ρ− 2 a(k 0 ) =

at ⊗ Uρ I at ⊗ =
Z
3
ρ 2 dk G(ρk) ⊗ a∗ (k) + G(ρk)∗ ⊗ a(k)

= (5.25)

and the Hamiltonian Hg transforms to:

Z
3
Uρ Hg Uρ∗ = Hat ⊗ ρ 2 dk G(ρk) ⊗ a∗ (k) + G(ρk)∗ ⊗ a(k)

f +ρ at ⊗ Hf + (5.26)

In the same way we can also dilate the Liouvillian Lg and we would obtain results similar
to the ones obtained in chapter 4. With these examples we can see that the effect of this
dilatation is to rescale the photon momenta as k → ρk.
We apply now these results to Le 0 [z] Eq. (5.5), additionally we apply a rescalation by dividing
the result of the dilatation by ρ0 . The new effective Liouvillian after the dilatation and
rescalation is:
1
Pat ⊗ L0ef f [z] := Uρ0 (FP0 (Lg (θ) − z) + zχ(Lf < ρ0 ) Uρ∗0
 
(5.27)
ρ0
where Uρ0 is the dilatation operator.
3 Rescalation of the spectral parameter
Z : z → ρ1 z and the new effective Liouvillian is:

1
Uρ0 (FP0 (Lg (θ) − Z −1 (z)) + Z −1 (z)χ(Lf < ρ0 ) Uρ∗0
 
Pat ⊗ L0 [z] := (5.28)
ρ0
 61

The reason of this rescalation is that under a dilatation and a rescalation of 2, the process of
1
transformation of T would be Dilatρ0 (T ) = ρ0 T and Rescρ0 (T ) = ρ0 Dilatρ0 (T ) = T but the
1 1
way in which E transforms is not adequate since Rescρ (E · ) = ρ0 Dilatρ0 (E · )= ρ0 E ·
and with ρ0 < 1 we would have problems. Namely E/ρn0 would diverge after a big enough
number n of steps of this RG.
Applying these two operations 2 and 3 to Eq. (5.6) we obtain:

L0 [z] = P0 (E0 [z] + T0 [z, L] + W0 [z])P0 (5.29)

where:

E0 [z] = 4E0 [z] = ρ−1

0 ω
0
e0,0 [Z −1 (z), 0] (5.30)
T0 [z, L] = Lf + 4T0 [z, L] = Lf + ρ−1
0 {e
0
ω0,0 [Z −1 (z), ρ0 r] − ω 0
e0,0 [Z −1 (z), 0]} (5.31)
X
0
W0 [z] = gW + 4W0 [z] = gW + Wm,n (5.32)
m+n≥1

(0)
where in the same way as in Eq. (5.25) we can calculate the relation between ωm,n and
ω 0 :
em,n
3
(m+n)−1 (0)
(0)
ωm,n [z; r; k (m) ; e
k (m) ] := ρ02 em,n [Z −1 (z); ρ0 r; ρ0 k (m) ; ρ0 e
ω k (m) ] (5.33)
Z
Wm,n(0)
:= k (n) a∗ (k (m) )ωm,n
dk (m) de (0)
[z; r; k (m) ; e
k (n) ]a(e
k (n) ) (5.34)

The idea behind this RG is to repeat these three operations in a number of steps obtaining
after each step a new effective Liouvillian Ln [z], containing a term Wn [z] which is becoming
smaller. Because of the isospectrality condition of the renormalization group, the spectrum
of our initial Liouvillian Lg (θ) is the same as the spectrum of Ln [z] which is much easier
to analyze with the vanishing term Wn [z]. By decimating the degrees of freedom of the
system with this RG, the dynamics of the remaining degrees of freedom is approximated by
the free photons dynamics.

5.2.2 Step 1

In this step we apply the three operations formerly defined to the effective Liouvillian of
step 0 L0 [z]. The problem to be discussed now is the possibility of doing a new FM. If we
realize a new Feshbach Map (FM) onto our new Liouvillian L0 [z], that is to say:

FP1 (L0 [z] − z) = P1 [(L0 [z] − z) − g 2 P1 W (θ)P̄1 (L̄0 [z] − z)−1 P̄1 W (θ)]P1 (5.35)
 62

with:
P1 = Pkat ⊗ χ[Laux < ρ1 ] (5.36)
PN −1
where ρ1 := g 2+/2 (see Eq. (4.45)) and k = n=0 e−βEn /2 ϕn ⊗ ϕn and in theorem 6.3.3.
we prove that KerΓ0 = Ck, where Γ0 is the level-shift operator.
This FM would be defined only for the values of z where L̄0 [z] − z is invertible.
 −1
(P 1 L0 [z]P 1 − z)−1 = P 1 P0 (E0 [z] + T0 [z, L] + W0 [z])P0 P 1 − z (5.37)

using the polar decomposition, with R = |E0 [z] + T0 [z, L] − z|−1

E0 [z] + T0 [z, L] − z = U |E0 [z] + T0 [z, L] − z| = U R−1 (5.38)

(P 1 L0 [z]P 1 − z)−1 = R1/2 (U + R1/2 P 1 P0 W0 [z]P0 P 1 R1/2 )−1 P 1 P0 R1/2 =
∞
X
= R1/2 U ∗ (−R1/2 P 1 P0 W0 [z]P 1 P0 R1/2 U ∗ )n R1/2

(5.39)
n=0

∂T [z; ·]
and denoting by T 0 [z; ·] := we have:
∂raux
Lemma 5.2.1. Let be T0 [z; r] with kT00 [z; ·] − 1k∞ ≤ δ < 1/6 and 0 ≤ raux ≤ 1 then for
any z so that |z − E0 [z]| ≤ ρ/2 we have:
1
|T0 [z; r] + E0 [z] − z| ≥ (raux + ρ) (5.40)
6
Proof. from kT00 [z; ·] − 1k∞ ≤ δ we have −(T00 − 1) ≤ δ and T0 > raux (1 − δ) and therefore:
ρ 1−δ 1 1
|T0 [z; r]+E0 [z]−z| ≥ |T0 [z; r]|−|E0 [z]−z| ≥ (1−δ)raux − ≥( − )(raux +ρ) ≥ (raux +ρ)
2 2 4 6
(5.41)

QM QN
Lemma 5.2.2. Assume that |ωM,N [r; k (M ) ; k (N ) ]| ≤ εξ M +N j=1 ω(kj )
α e α
j=1 ω(kj ) then
for any 0 < ρ ≤ 1 we have:
2ε 2
k(raux + ρ)−1/2 P 1 P0 W0 P 1 P0 (raux + ρ)−1/2 k ≤ e (5.42)
ρ1/2
For a proof see [14]

Theorem 5.2.3. For any z so that |z − E0 [z]| ≤ ρ/2 the operator (P 1 L0 [z]P 1 − z) is
invertible on P 1 P0 .
 63

Proof. With these two lemmas and Eq. (5.39) we have:

kR1/2 P 1 P0 W0 [z]P 1 P0 R1/2 k ≤ k(raux + ρ)−1/2 P 1 P0 W0 P 1 P0 (raux + ρ)−1/2 k ×

raux + ρ 2e2 ε 1
×k k ≤ 1/2
T0 [z; r] + E0 [z] − z ρ 1/6

With this theorem we have the justification for doing a next FM FP1 (L0 [z] − z) and
then writing it in a Wick ordered form:
W icks theorem 1
FP1 (L0 [z] − z) := L
e 1 [z] + E0 [z] − z −→ ω
eM,N (5.43)

where
L
e 1 [z] = P1 (E
e1 [z] + Te1 [z, L] + W
f1 [z])P1

next we apply a dilatation and rescalation of the spectral parameter and we define the
Renormalization Map by:

Rρ1 (L0 )[z] − z := ρ−1 Uρ1 FP1 (L0 [Z −1 (z)] − Z −1 (z))Uρ∗1 (5.44)

where Uρ1 is a dilatation operator. We define:

L1 [z] − z := Rρ1 (L0 )[z] − z (5.45)

and substituting Eq. (5.43):

L1 [z] − z = ρ−1
1 Uρ1 (L1 [Z
e −1
(z)] + E0 [Z −1 (z)] − Z −1 (z))Uρ∗1 (5.46)

Applying now the Wick’s theorem:

L1 [z] = P1 (E1 [z] + T1 [z, L] + W1 [z])P1 (5.47)

where we obtain again a number E1 [z], a function of L, T1 [z, L] and a sum of Wick’s
monomials:

E1 [z] = 4E0 [z] = ρ−1

1 ω
1
e0,0 [Z −1 (z), 0] (5.48)
T1 [z, L] = ρ−1
1 T0 [Z
−1
(z); ρ1 r] + 4T1 [z, L] =
ρ−1
1 T0 [Z
−1
(z); ρ1 r] + ρ−1
1 {e
1
ω0,0 [Z −1 (z), ρ1 r] − ω 1
e0,0 [Z −1 (z), 0]} (5.49)
X
1
W1 [z] = gW + 4W1 [z] = gW + WM,N (5.50)
M +N ≥1
 64

5.3 Step 0 of the RG in Second Order in g for T > 0

The purpose of this section is to see the specific form in order g 2 of the expressions E,
e Te, W
f
seen in the last section (that is to say step 0 without dilatation and rescalation). In order
to do that we will develop the Feshbach Map in powers of g. We will restrict us to order g 2 ,
afterwards we will see how the terms E,
e Te, W
f mentioned in the last sections are generated
(until order g 2 ), using Wick’s theorem and lemma A.0.3 of Appendix A. We will use the
P −1
projector P0 := P0at ⊗ χ[Laux < ρ0 ] = P0at ⊗ χρ0 with P0at = N
i=0 |ϕi ⊗ ϕi ihϕi ⊗ ϕi | and
Hat ϕi = εi ϕi .
We proceed now to develop the FM:

FP0 (Lg (θ) − z) = P0 [(Lg (θ) − z) − g 2 P0 Lg (θ)P 0 (Lg (θ) − z)−1 P 0 Lg (θ)]P0 =
= P0 [(Lg (θ) − z) − g 2 P0 W P 0 (Lg (θ) − z)−1 P 0 W ]P0 (5.51)

where we have used P0 Lat P̄0 = 0 and now we use P 0 L0 (θ)P 0 = L0 (θ)P 0 to develop the
second term of Eq. (5.51) in a Neumann series expansion:
h
i−1
(Lg (θ) − z)−1 = (L0 (θ) − z + gW (θ))−1 = L0 (θ) − z 1 + (L0 (θ) − z)−1 gW (θ)

= (1 + (L0 (θ) − z)−1 gW (θ))−1 (L0 (θ) − z)−1

using now the Neumann series of (1 + (L0 (θ) − z)−1 gW (θ))−1 :

(Lg (θ) − z)−1 = (L0 (θ) − z)−1 − (L0 (θ) − z)−1 gW (θ)(L0 (θ) − z)−1 +
(L0 (θ) − z)−1 gW (θ)(L0 (θ) − z)−1 gW (θ)(L0 (θ) − z)−1 ∓ . . . =
X∞
(L0 (θ) − z)−1 (−gW (θ)(L0 (θ) − z)−1 )n (5.52)
n=0

1 2
noting that (L0 (θ) − z)−1 = P 0 and P 0 = P 0 we have:
L0 (θ) − z
∞
−1 1 X 1
n
(Lg (θ) − z) = P0 − gW P0 (5.53)
L0 (θ) − z L0 (θ) − z
n=0

inserting this result in Eq. (5.51) and using P0 Lf P0 = 0: KA

A
A
z }| {
FP0 (Lg (θ) − z) = P0 (Lg (θ) − z)P0 − g 2 P0 W (θ)P 0 (Lg (θ) − z)−1 P 0 W P0 =
∞
1 X 1
n
(L0 (θ) − z)P0 + gP0 W (θ)P0 − g 2 P0 W (θ)[ P0 − gW (θ) P 0 ]W (θ)P0 =
L0 (θ) − z L0 (θ) − z
n=0
 65

and reordering the terms:

FP0 (Lg (θ) − z) = (L0 (θ) − z)P0 + gP0 W (θ)P0 −

1 1 1
−g 2 P0 W (θ)[ P0 − P 0 gW (θ) P 0 + · · · ]W (θ)P0 =
L0 (θ) − z L0 (θ) − z L0 (θ) − z
∞
X 1
= (L0 (θ) − z)P0 + (−1)ν−1 g ν P0 W (θ)( P 0 W (θ))ν−1 P0 (5.54)
L0 (θ) − z
ν=1

P0
This Neumann series converges if k W (θ)P0 k < 1. We will work with the interac-
L0 (θ) − z
tion Liouvillian:

W = l[I]−r[I] = l[a∗ ]⊗(G⊗ )+l[a]⊗(G∗ ⊗ )−Jf l[a∗ ]Jf ⊗( ⊗T GT )−Jf l[a]Jf ⊗( ⊗T G∗ T )

(5.55)
and in the same way as by deducing Eq. (5.26), using Eq. (5.22) and Eq. (5.23) (for ρ = e−θ )
we can see:
Z
3
∗
W (θ) = U (θ)W U (θ) = dke− 2 θ {l[a∗ (k)] ⊗ (G(e−θ k) ⊗ ) + l[a] ⊗ (G∗ (e−θ k) ⊗ )}−
Z
3 ∗
− dke− 2 θ {r[a∗ (k)] ⊗ ( ⊗ G(e−θ k)) + r[a(k)] ⊗ ( ⊗ G (e−θ k))} (5.56)

and defining:
3
Gθ (k) := e− 2 θ G(e−θ k) (5.57)
− 32 θ
G∗θ (k) := e G∗ (e−θ k) (5.58)
− 32 θ
Gθ (k) := e G(e−θ k)) (5.59)
∗ 3 ∗
Gθ (k) := e− 2 θ G (e−θ k)) (5.60)

we can write it as:

Z
W (θ) = dk{l[a∗ (k)] ⊗ (Gθ (k) ⊗ ) + l[a] ⊗ (G∗θ (k) ⊗ )}−
Z
∗
− dk{r[a∗ (k)] ⊗ ( ⊗ Gθ (k)) + r[a(k)] ⊗ ( ⊗ Gθ (k))} (5.61)

For the convergence of the Neumann series we will need the following lemma:

Lemma 5.3.1. With the assumption on G:

Z 1/2
Λ1 := dk(1 + ω(k)−1 )|G(k)|2 <∞ (5.62)
 66

we have:
−1/2
k(Laux + ρ0 )−1/2 W (θ)(Laux + ρ0 )−1/2 k ≤ 4Λ1 ρ0 (5.63)

Proof. We have that:

Z
2
ka(G)ψk = ha(G)ψ|a(G)ψi = dkdk 0 G(k)G(k 0 )ha(k)ψ|a(k 0 )ψi ≤

|G(k)| |G(k 0 )| p
Z
dkdk 0 p |h |k|a(k)ψ| |k 0 |a(k 0 )ψi|
p
≤ p
|k| |k 0 |

applying now the inequality of Cauchy-Schwartz-Buniakovski:

|G(k)| p |G(k 0 )| p 0
Z Z
≤ dk p k |k|a(k)ψk dk 0 p k |k |a(k 0 )ψk ≤
|k| 0
|k |

Z |G(k)|2  p  Z  G 2  Z 
dkω(k)hψ|a∗ (k)a(k)ψi
p
dk k |k|a(k)ψ dkk |k|a(k)ψk2 = √

|k| ω 2
| {z }
1/2
kHf ψk2

and recalling that Laux = Hf ⊗ f + f ⊗ Hf , then we have:

G
ka(G)ψk ≤ √ kL1/2 ψk (5.64)

ω 2 aux
−1/2
and taking ψ := Laux χ where χ is a characteristic function we have:
G
ka(G)L−1/2 k ≤ √ (5.65)

aux
ω 2

and therefore:
1 G

k(Laux + ρ0 )−1/2 a(G)(Laux + ρ0 )−1/2 k ≤ √ √ (5.66)
ρ0 ω 2
for ka∗ (G)ψk we have:

ka∗ (G)ψk2 = hψ|a(G)a∗ (G)ψi = kGk22 kψk2 + hψ|a∗ (G)a(G)ψi = kGk22 kψk2 + ka(G)ψk2
(5.67)
then for both creation and annihilation operators we have:
Z
−1/2 ] −1/2 1  1 1/2
k(Laux + ρ0 ) a (G)(Laux + ρ0 ) k≤ √ dk(1 + )|G(k)|2 (5.68)
ρ0 ω(k)

and taking into account the four terms of Eq. (5.61) we obtain Eq. (5.63).
 67

Proposition 5.3.2. Let z be so that |Rez| < ρ0 /2 and |Imz| ≥ − 12 ρ0 sin ϑ where ρ0 := g 2−
for  small and 0 < ϑ < π/2. For these values of z and P0 = P0at ⊗ χρ0 , we have:
L
aux + ρ0 4

P 0 ≤ (5.69)
L0 (θ) − z sin θ

Proof. Using Eq. (4.15) with θ = iϑ:

L0 (θ) = Lat + Lf cos ϑ − iLaux sin ϑ (5.70)

and:

P 0 = P (Lat 6= 0) ⊗ χ(Laux ≤ ρ0 ) + P (Lat = 0) ⊗ χ(Laux > ρ0 ) + P (Lat 6= 0) ⊗ χ(Laux > ρ0 )

(5.71)
defining:

Q1 := P (Lat 6= 0) ⊗ χ(Laux ≤ ρ0 ) (5.72)

Q2 := at ⊗ χ(Laux > ρ0 ) (5.73)

then we have:
P 0 = Q1 + Q2 (5.74)

We will suppose that [L0 (θ), Q1,2 ] = 0. We have:

 
k(Laux +ρ0 )(L0 (θ)−z)−1 P 0 k = max k(Laux +ρ0 )(L0 (θ)−z)−1 Q1 k, k(Laux +ρ0 )(L0 (θ)−z)−1 Q2 k
(5.75)
We use now the inequalities: |z| ≥ |Imz| and |z| ≥ |Rez|. Let us analyze each of the two
expressions k(Laux + ρ0 )(L0 (θ) − z)−1 Q1 k and k(Laux + ρ0 )(L0 (θ) − z)−1 Q2 k:
Let A be the set of values (rat , rf , raux ) of (Lat , Lf , Laux ) so that:

 rat ∈ σ(Lat \{0})


rf ∈ R, |rf | < raux (5.76)


0 ≤ raux ≤ ρ0


 raux + ρ0 
k(Laux + ρ0 )(L0 (θ) − z)−1 Q1 k = sup ≤
 
rat + rf cos ϑ − iraux sin ϑ − z

(rat ,rf ,raux )∈A
 raux + ρ0 
sup (5.77)
 
rat + rf cos ϑ − Rez
 
(rat ,rf ,raux )∈A
 68

Looking at Fig.13 we see that in A:

3 1
|rat + rf cos ϑ − Rez| ≥ |rat − Rez| − |rf cos ϑ| ≥ ρ0 − ρ0 = ρ0 (5.78)
2 2
and:
|raux + ρ0 | ≤ 2ρ0 (5.79)

and then:
2
k(Laux + ρ0 )(L0 (θ) − z)−1 Q1 k ≤ 2ρ0 = 4 (5.80)
ρ0
Let B be the set of values (rat , rf , raux ) of (Lat , Lf , Laux ) so that:

 rat ∈ σ(Lat )


rf ∈ R, |rf | < raux (5.81)


raux > ρ0


 raux + ρ0 
k(Laux + ρ0 )(L0 (θ) − z)−1 Q1 k = sup ≤
 
r + r cos ϑ − ir sin ϑ − z

(rat ,rf ,raux )∈B at f aux
 raux + ρ0 
sup (5.82)
 
r sin ϑ + Imz
 
(rat ,rf ,raux )∈B aux

Looking at Fig.13 we see that Imz ≥ − 21 ρ0 sin ϑ and in B:

 raux + ρ0  raux + ρ0 raux + ρ0 raux + ρ0
≥ ≥ =
 
1
raux sin ϑ + Imz raux sin ϑ + Imz raux sin ϑ − 2 ρ0 sin ϑ sin ϑ(r aux − ρ0 /2)

(5.83)
and then:
 raux + ρ0  raux + ρ0
sup ≤ sup (5.84)
 
raux sin ϑ + Imz (rat ,rf ,raux )∈B sin ϑ(raux − ρ0 /2)

(rat ,rf ,raux )∈B

defining now y := raux − ρ0 /2 we have:

raux + ρ0 1 y + ρ0 /2 + ρ0 1 ρ0 /2 + ρ0
sup = sup = sup [1+ ]=
(rat ,rf ,raux )∈B sin ϑ(raux − ρ0 /2) sin ϑ y>ρ0 /2 y sin ϑ y>ρ0 /2 y
(5.85)
1 ρ0 /2 + ρ0 4
= (1 + )= (5.86)
sin ϑ ρ0 /2 sin ϑ
4
k(Laux + ρ0 )(L0 (θ) − z)−1 Q2 k ≤ (5.87)
sin ϑ
and finally:
4
4
k(L0 (θ) − z)−1 P 0 k ≤ max 4, = (5.88)
sin ϑ sin ϑ
 69

And now we can see the convergence of the Neumann series expansion:

Theorem 5.3.3. Let z be so that |Rez| < ρ0 /2 and |Imz| ≥ − 12 ρ0 sin ϑ where ρ0 := g 2− for
 small and 0 < ϑ < π/2. For these conditions and g sufficiently small P 0 (L − g(θ) − z)P 0
is invertible on RanP 0 . That implies that for these conditions and g sufficiently small the
Feshbach map FP0 (Lg (θ) − z) with P0 = P0at ⊗ χρ0 is well defined. The series expansion
Eq. (5.54)
∞
X 1
FP0 (Lg (θ) − z) = (L0 (θ) − z)P0 + (−1)ν−1 g ν P0 W (θ)( P 0 W (θ))ν−1 P0
L0 (θ) − z
ν=1

is for these conditions convergent.

Proof. The Feshbach map Eq. (5.51):

FP0 (Lg (θ) − z) = P0 [(Lg (θ) − z) − g 2 P0 Lg (θ)P 0 (Lg (θ) − z)−1 P 0 Lg (θ)]P0

is well defined for all the values of z where the inverse of (Lg (θ) − z) exists.
We will analyze the term Eq. (5.53) P 0 (Lg − z)−1 P 0 :
∞
−1 1 X 1
n
P 0 (Lg (θ) − z) P0 = P0 − gW (θ) P0 =
L0 (θ) − z L0 (θ) − z
n=0
(Laux + ρ0 )1/2
= P 0×
L0 (θ) − z
∞ n
X
 (Laux + ρ0 ) on  1 
(Laux + ρ0 )−1/2 (−gW )(Laux + ρ0 )−1/2 P0 (5.89)
n=0
L0 (θ) − z (Laux + ρ0 )1/2
this equation can be checked (for n=0 is trivial),
for n = 1:

(Laux + ρ0 )1/2
`

 (L`
aux +
```ρ0 )  1 
P 0 (Laux +ρ0 )−1/2 (−gW )(L` −1/2
` `
aux +ρ
`` 0 ) P 0
L0 (θ) − z L0 (θ) − z `ρ0 )1/2
```
(Laux +
`

for n = 2:

(Laux + ρ0 )1/2
 `
(L`
aux +
```ρ0 ) 
P 0 (Laux + ρ0 )−1/2 (−gW )(L` `ρ0 )−1/2
` `
aux +
` P0 ×
L0 (θ) − z L0 (θ) − z
−1/2 −1/2

 (Laux + ρ0 )  1 
× (L`
` `
aux +
`
` ρ 0 ) (−gW )(L aux + ρ 0 ) P 0
L0 (θ) − z (Laux + ρ0 )1/2
 70

and then we have

 (Laux + ρ0 )  4 16
k (Laux +ρ0 )−1/2 (−gW (θ))(Laux +ρ0 )−1/2 P 0 k ≤ (4gΛ1 )( )= gΛ1  1
L0 (θ) − z sin ϑ sin ϑ
(5.90)
which is  1 for g sufficiently small.

We will restrict us to the term in second order in g of Eq. (5.54):

P0
−g 2 P0 W (θ) W (θ)P0 (5.91)
L0 (θ) − z
we see that:
at at
P0 = − P0 = P 0 ⊗ χρ0 + P0at ⊗ χρ0 + P 0 ⊗ χρ0 =
at at
= P 0 ⊗ (χρ0 + χρ0 ) + P0at ⊗ χρ0 = P 0 ⊗ + P0at ⊗ χρ0 (5.92)

we define:
M
X
P 0 := χαat ⊗ χαf (0) (5.93)
0


f : for α = 1, 2 · · · M
where χαf [ω] := (5.94)
 χ[Laux + ω > ρ0 ] : for α = 0
(α)
χat := Pεatα = χεα [Lat ] (5.95)

and {ε0 · · · εM } = {Ei,j |0 ≤ i, j ≤ N − 1} with ε0 := 0 and M ≤ N (N − 1), that is we have

N zero eigenvalues (see Eq. (4.1)). These correspond to the differences Eij = Ei − Ej when
Ei = Ej and N 2 − N = N (N − 1) non-zero eigenvalues.
P0 P0
−g 2 P0 W (θ) W (θ)P0 = −g 2 χ0at χρ0 W (θ) W (θ)χ0at χρ0
L0 (θ) − z L0 (θ) − z
PM M
0 χαat ⊗ χαf (0) X χαf (0)
= −g 2 χ0at χρ0 W (θ) W (θ)χ0at χρ0 = −g 2
χ0at χρ0 χαat W (θ) W (θ)χ0at χρ0
L0 (θ) − z L0 (θ) − z
0

χαf (0)
Let us analyze the term W W with the interaction Eq. (5.55), considering that
L0 (θ) − z
T GT = G = (G∗ )T :
χαf (0)
W (θ)
W (θ) =
L0 (θ) − z
χαf (0) χαf (0) χαf (0) χαf (0)
= l(I) l(I) + r(I) r(I) − l(I) r(I) − r(I) l(I)
L0 (θ) − z L0 (θ) − z L0 (θ) − z L0 (θ) − z
| {z } | {z } | {z } | {z }
I II III IV
 71

We will develop now the different terms I, II, III, IV each of which will result in four terms
more and each of these will give also four terms. That is to say in total we will have
4 × 4 × 4 = 64 different terms but fortunately many of them are zero.
Taking into account Eq. (5.61), we have:

∗
χαf (0)
I = [l[a ] ⊗ (Gθ ⊗ ) + l[a] ⊗ (G∗θ ⊗ )] [l[a∗ ] ⊗ (Gθ ⊗ ) + l[a] ⊗ (G∗θ ⊗ )] =
L0 (θ) − z
χαf (0)
Z
= dkdl l[a∗ (k)] ⊗ (Gθ (k) ⊗ ) l[a∗ (l)] ⊗ (Gθ (l) ⊗ ) + (A1 ) (5.96)
L0 (θ) − z
χαf (0)
Z
+ dkdl l[a∗ (k)] ⊗ (Gθ (k) ⊗ ) l[a(l)] ⊗ (G∗θ (l) ⊗ ) + (B1 ) (5.97)
L0 (θ) − z
χαf (0)
Z
+ dkdl l[a(k)] ⊗ (G∗θ (k) ⊗ ) l[a∗ (l)] ⊗ (Gθ (l) ⊗ ) + (C1 ) (5.98)
L0 (θ) − z
χαf (0)
Z
+ dkdl l[a(k)] ⊗ (G∗θ (k) ⊗ ) l[a(l)] ⊗ (Gθ (l) ⊗ ) (D1 ) (5.99)
L0 (θ) − z

In order to become expressions like E

e0 , Te0 , W
f0 we will need the lemma A.0.3 of Appendix
A.
We will use the following notation:

r = (rf , raux ) (5.100)

L = (Lf , Laux ) (5.101)

L(θ) + r(θ) := (Lf + rf ) cosh θ − (Laux + raux ) sinh θ (5.102)

Applying lemma A.0.3 and the Pull-trough formula (lemma A.0.1) and with the Eq. (4.15)
L0 (θ) = Lat + Lf cosh θ − Laux sinh θ:
χαf (0)
Z
A1 = dkdl l[a∗ (k)] ⊗ (Gθ (k) ⊗ ) l[a∗ (l)] ⊗ (Gθ (l) ⊗ ) =
L0 (θ) − z
χαf (raux + ω(l))
Z D E
= dkdl l[a∗ (k)]l[a∗ (l)] Ωf |Gθ (k) ⊗ Gθ (l) ⊗ Ωf  + (5.103)

Lat + L(θ) + r(θ) + ω(l) − z r=L
χαf (raux )
Z D Z E
∗
+ dk l[a (k)] Ωf |Gθ (k) ⊗ dl l[a∗ (l)]Gθ (l) ⊗ Ωf  + (5.104)

Lat + L(θ) + r(θ) − z r=L

χαf (raux + ω(l))

Z D Z E
+ dl l[a (l)] Ωf | dk l[a∗ (k)]Gθ (k) ⊗
∗
Gθ (l) ⊗ Ωf  +

Lat + L(θ) + r(θ) + ω(l) − z r=L
(5.105)
Z
χα (r )
Z
aux
D E
f
+ Ωf | dk l[a∗ (k)]Gθ (k)⊗ dl l[a∗ (l)]Gθ (l)⊗ Ωf  (5.106)

Lat + L(θ) + r(θ) − z r=L
 72

From these terms Eq. (5.104), Eq. (5.105) and Eq. (5.106) are zero because of hΩ|l[a∗ (l)]Ωi =
hΩ|l[a∗ (k)]Ωi = hΩ|l[a∗ (k)]l[a∗ (l)]Ωi = 0. Doing the same procedure with B1 :
χαf (0)
Z
B1 = dkdl l[a∗ (k)] ⊗ (Gθ (k) ⊗ ) l[a(l)] ⊗ (G∗θ (l) ⊗ ) =
L0 (θ) − z

χαf (raux )
Z D E 
∗
dkdl l[a (k)] Ωf |Gθ (k) ⊗ G∗θ (l) ⊗ Ωf l[a(l)] + (5.107)

Lat + L(θ) + r(θ) − z r=L
χαf (raux )
Z D Z E 
+ dl Ωf | dk l[a∗ (k)]Gθ (k) ⊗ G∗θ (l) ⊗ Ωf l[a(l)]
+ (5.108)

Lat + L(θ) + r(θ) − z r=L
χαf (raux )
D Z Z E
+ Ωf | dk l[a∗ (k)]Gθ (k) ⊗ dl l[a(l)]G∗θ (l) ⊗ Ωf  + (5.109)

Lat + L(θ) + r(θ) − z r=L
χαf (raux )
Z D Z E
+ dk l[a∗ (k)] Ωf |Gθ (k) ⊗ dl l[a(l)]G∗θ (l) ⊗ Ωf  (5.110)

Lat + L(θ) + r(θ) − z r=L

because of the reasons mentioned before, there are vanishing terms and they are Eq. (5.108)
and Eq. (5.110), the term Eq. (5.109) do not vanish if k = l.
From now on we will write only the non-vanishing terms:
χαf (raux + ω(k) + ω(l))
Z D E 
∗
C1 = dkdl l[a (l)] Ωf |Gθ (k)⊗ Gθ (l)⊗ Ωf l[a(k)] +

Lat + L(θ) + r(θ) + ω(k) + ω(l) − z r=L
(5.111)
χαf (raux )
D Z Z E
+ Ωf | dk l[a(k)]Gθ (k) ⊗ dl l[a∗ (l)]Gθ (l) ⊗ Ωf 

Lat + L(θ) + r(θ) − z r=L
(5.112)
Z
χ α (r + ω(k))
aux
D E 
f
D1 = dkdl Ωf |G∗θ (k)⊗ Gθ (l)⊗ Ωf l[a(k)]l[a(l)]

Lat + L(θ) + r(θ) + ω(k) − z r=L
(5.113)
Using now the representation r(a) = Jf l[a]Jf and that G = T GT :

∗ χαf (0) ∗
II = [r[a∗ ] ⊗ ( ⊗ Gθ ) + r[a] ⊗ ( ⊗ Gθ ] [r[a∗ ] ⊗ ( ⊗ Gθ ) + r[a] ⊗ ( ⊗ Gθ )] =
L0 (θ) − z

χαf (0)
Z
dkdl r[a∗ (k)] ⊗ ( ⊗ Gθ (k)) r[a∗ (l)] ⊗ ( ⊗ Gθ (l)) + (A2 ) (5.114)
L0 (θ) − z
χαf (0)
Z
∗
+ dkdl r[a∗ (k)] ⊗ ( ⊗ Gθ (k)) r[a(l)] ⊗ ( ⊗ Gθ (l)) + (B2 ) (5.115)
L0 (θ) − z
χαf (0)
Z
∗
+ dkdl r[a(k)] ⊗ ( ⊗ Gθ (k)) r[a∗ (l)] ⊗ ( ⊗ Gθ (l)) + (C2 ) (5.116)
L0 (θ) − z
χαf (0)
Z
∗ ∗
+ dkdl r[a(k)] ⊗ ( ⊗ Gθ (k)) r[a(l)] ⊗ ( ⊗ Gθ (l)) (D2 ) (5.117)
L0 (θ) − z
 73

χαf (raux − ω(l))

Z D E
A2 = dkdl r[a∗ (k)]r[a∗ (l)] Ωf | ⊗Gθ (k) ⊗Gθ (l)Ωf 

Lat + L(θ) + r(θ) − ω(l) − z r=L
(5.118)
χαf (raux )
Z D E 
∗ ∗
B2 = dkdl r[a (k)] Ωf | ⊗ Gθ (k) ⊗ Gθ (l)Ωf r[a(l)] + (5.119)

Lat + L(θ) + r(θ) − z r=L

χαf (raux )
D Z Z E
∗
+ Ωf | dk r[a∗ (k)] ⊗ Gθ (k) dl r[a(l)] ⊗ Gθ (l)Ωf 

Lat + L(θ) + r(θ) − z r=L
(5.120)
χαf (raux − ω(k) − ω(l))
Z D E 
∗
C2 = dkdl r[a∗ (l)] Ωf | ⊗Gθ (k) ⊗Gθ (l)Ωf r[a(k)] +

Lat + L(θ) + r(θ) − ω(k) − ω(l) − z r=L
(5.121)
χαf (raux )
D Z Z E
∗
+ Ωf | dk r[a(k)] ⊗ Gθ (k) dl r[a∗ (l)] ⊗ Gθ (l)Ωf  (5.122)

Lat + L(θ) + r(θ) − z r=L

χαf (raux − ω(k))

Z D E 
∗ ∗
D2 = dkdl Ωf | ⊗Gθ (k) ⊗Gθ (l)Ωf r[a(k)]r[a(l)]

Lat + L(θ) + r(θ) − ω(k) − z r=L
(5.123)
Doing the same with III and maintaining the non-vanishing terms:
χαf (0) ∗
III = [l[a∗ ] ⊗ (Gθ ⊗ ) + l[a] ⊗ (G∗θ ⊗ )] [r[a∗ ] ⊗ ( ⊗ Gθ ) + r[a] ⊗ ( ⊗ Gθ ] =
L0 (θ) − z

χαf (0)
Z
dkdl l[a∗ (k)] ⊗ (Gθ (k) ⊗ ) r[a∗ (l)] ⊗ ( ⊗ Gθ (l)) + (A3 ) (5.124)
L0 (θ) − z
χαf (0)
Z
∗
+ dkdl l[a∗ (k)] ⊗ (Gθ (k) ⊗ ) r[a(l)] ⊗ ( ⊗ Gθ (l)) + (B3 ) (5.125)
L0 (θ) − z
χαf (0)
Z
+ dkdl l[a(k)] ⊗ (G∗θ (k) ⊗ ) r[a∗ (l)] ⊗ ( ⊗ Gθ (l)) + (C3 ) (5.126)
L0 (θ) − z
χαf (0)
Z
∗
+ dkdl l[a(k)] ⊗ (G∗θ (k) ⊗ ) r[a(l)] ⊗ ( ⊗ Gθ (l)) (D3 ) (5.127)
L0 (θ) − z
Now we are dealing with different representations l and r and therefore we must be careful
because of the relation:
β
r[a∗ (l)]Ωf i = l[a(l)]e 2
ω(l)
|Ωf i (5.128)

and other similar relations, see Eq. (6.53), Eq. (6.54), Eq. (6.55) and Eq. (6.56).
Taking into account Eq. (6.55):
χαf (raux − ω(l))
Z D E
A3 = dkdl l[a∗ (k)]r[a∗ (l)] Ωf |Gθ (k)⊗ ⊗Gθ (l)Ωf 

Lat + L(θ) + r(θ) − ω(l) − z r=L
(5.129)
 74

χαf (raux )
D Z Z E
+ Ωf | dk l[a∗ (k)]Gθ (k) ⊗ dl r(a∗ (l)) ⊗ Gθ (l)Ωf 

Lat + L(θ) + r(θ) − z r=L
(5.130)
Taking into account Eq. (6.53):
χαf (raux )
Z D E 
∗
B3 = dkdl l[a∗ (k)] Ωf |Gθ (k) ⊗ ⊗ Gθ (l)Ωf r[a(l)] +

Lat + L(θ) + r(θ) − z r=L
(5.131)
Taking into account Eq. (6.55):
χαf (raux − ω(k) + ω(l))
Z D E 
C3 = dkdl r[a∗ (l)] Ωf |G∗θ (k)⊗ ⊗Gθ (l)Ωf l[a(k)] +

Lat + L(θ) + r(θ) − ω(k) + ω(l) − z r=L
(5.132)
Taking into account Eq. (6.53):
χαf (raux + ω(k))
Z D E 
∗ ∗
D3 = dkdl Ωf |Gθ (k)⊗ ⊗Gθ (l)Ωf l[a(k)]r[a(l)]

Lat + L(θ) + r(θ) + ω(k) − z r=L
(5.133)
Z
χ α (r )
Z
f aux
D E
∗
+ Ωf | dk l[a(k)]G∗θ (k) ⊗ dl r[a(l)] ⊗ Gθ (l)Ωf 

Lat + L(θ) + r(θ) − z r=L
(5.134)
χ α (0)
∗ f
IV = [r[a∗ ] ⊗ ( ⊗ Gθ ) + r[a] ⊗ ( ⊗ Gθ )] [l[a∗ ] ⊗ (Gθ ⊗ ) + l[a] ⊗ (G∗θ ⊗ )] =
L0 (θ) − z

χαf (0)
Z
dkdl r[a∗ (k)] ⊗ ( ⊗ Gθ (k)) l[a∗ (l)] ⊗ (Gθ (l) ⊗ ) + (A4 ) (5.135)
L0 (θ) − z
χαf (0)
Z
∗
+ dkdl r[a (k)] ⊗ ( ⊗ Gθ (k)) l[a(l)] ⊗ (G∗θ (l) ⊗ ) + (B4 ) (5.136)
L0 (θ) − z
χαf (0)
Z
∗
+ dkdl r[a(k)] ⊗ ( ⊗ Gθ (k)) l[a∗ (l)] ⊗ (Gθ (l) ⊗ ) + (C4 ) (5.137)
L0 (θ) − z
χαf (0)
Z
∗
+ dkdl r[a(k)] ⊗ ( ⊗ Gθ (k)) l[a(l)] ⊗ (G∗θ (l) ⊗ ) (D4 ) (5.138)
L0 (θ) − z
Once more we deal with different representations r and l therefore we take into account
again Eq. (6.53), Eq. (6.54), Eq. (6.55) and Eq. (6.56).
Taking into account Eq. (6.54):
χαf (raux + ω(l))
Z D E
A4 = dkdl r[a∗ (k)]l[a∗ (l)] Ωf | ⊗Gθ (k) Gθ (l)⊗ Ωf 

Lat + L(θ) + r(θ) + ω(l) − z r=L
(5.139)
χαf (raux )
D Z Z E
+ Ωf | dk r[a∗ (k)] ⊗ Gθ (k) dl l[a∗ (l)]Gθ (l) ⊗ Ωf 

Lat + L(θ) + r(θ) − z r=L
(5.140)
 75

Taking into account Eq. (6.54):

χαf (raux )
Z D E 
B4 = dkdl r[a∗ (k)] Ωf | ⊗ Gθ (k) G∗θ (l) ⊗ Ωf l[a(l)] +

Lat + L(θ) + r(θ) − z r=L
(5.141)
Taking into account Eq. (6.56):

χαf (raux − ω(k) + ω(l))

Z D E 
∗ ∗
C4 = dkdl l[a (l)] Ωf | ⊗Gθ (k) Gθ (l)⊗ Ωf r[a(k)] +

Lat + L(θ) + r(θ) − ω(k) + ω(l) − z r=L
(5.142)
Taking into account Eq. (6.56):

χαf (raux − ω(k))

Z D E 
∗
D4 = dkdl Ωf | ⊗Gθ (k) G∗θ (l)⊗ Ωf r[a(k)]l[a(l)]

Lat + L(θ) + r(θ) − ω(k) − z r=L
(5.143)
χαf (raux )
D Z Z E
∗
+ Ωf | dk r[a(k)] ⊗ Gθ (k) dl l[a(l)]G∗θ (l) ⊗ Ωf 

Lat + L(θ) + r(θ) − z r=L
(5.144)
Recalling the expressions Eq. (5.7), Eq. (5.8), Eq. (5.9) and Eq. (5.10):

e0 [z] = 4E
E e0 [z] = ω
e0,0 [z, 0] (5.145)
Te0 [z, Lf ] = Lf + 4Te0 [z, Lf ] = Lf + {eω0,0 [z, r] − ω
e0,0 [z, 0]} (5.146)
X
f0 [z] = gW + 4W (0)
W f0 [z] = gW + W
fm,n (5.147)
m+n≥1
Z
f (0) [z] :=
W m,n dk (m) de
k (n) a+ (k (m) )e 0
ωm,n [z; Lf ; k (m) ; k (n) ]a− (e
k (n) ) (5.148)

In order to identify the different terms of these equations we will transform Eq. (5.109),
Eq. (5.112), Eq. (5.120), Eq. (5.122), Eq. (5.130), Eq. (5.134), Eq. (5.140) and Eq. (5.144).
The purpose of this transformation is to fix the point of reference for Te in r = 0, that is in
Te[z, r] we would have Te[z, 0] = 0 and doing this we obtain the terms of E.
e We will show
the procedure with the Eq. (5.112). For the other three the procedure would be the same.

χαf (raux ) χαf (0)

D Z n
Ωf | dk l[a(k)]Gθ (k) ⊗ − + (5.149)
L + L(θ) + r(θ) − z Lat + L(θ) − z
| at {z }
4Te0
χαf (0) oZ E
+ dl l[a∗ (l)]Gθ (l) ⊗ Ωf  (5.150)

L + L(θ) − z r=L
| at {z }
4E
e0
 76

In this way the terms Eq. (5.109), Eq. (5.112), Eq. (5.120), Eq. (5.122), Eq. (5.130),
Eq. (5.134), Eq. (5.140) and Eq. (5.144) make contributions to 4E
e0 and 4Te0 . The contri-
butions to 4W
f [z] are the following:

4W
f [z] = W11 + W20 + W02 (5.151)

with:

W11 = Eq. (5.107) + Eq. (5.111) + Eq. (5.119) + Eq. (5.121) + (5.152)
+Eq. (5.131) + Eq. (5.132) + Eq. (5.141) + Eq. (5.142) (5.153)
W20 = Eq. (5.103) + Eq. (5.118) + Eq. (5.129) + Eq. (5.139) (5.154)
W02 = Eq. (5.113) + Eq. (5.123) + Eq. (5.133) + Eq. (5.143) (5.155)

5.4 RG for T=0

The principal difference with respect to the case T > 0 is that we have to work with
Hamiltonians instead of Liouvillians and that the spectrum is a little different. That is:

L = (Lf , Laux ) −→ Hf (5.156)

Fig.17 Spectrum of H0 (θ)

In Fig.16 we see that instead of triangles like in Fig.10 and Fig.11 we have now lines because
of the fact that in Eq. (4.6) we had a difference of two terms but now with the Hamiltonian
we have only one, which means only one rotation. We have considered that our model
in this section consists of an atom with different levels: the ground state E0 , the excited
levels E1 , E2 ..... which could be degenerated and an ionization energy Σ. Until now we
have avoided degenerated eigenvalues and ionization energies in order not to complicate the
problem. We will consider them only in this section. We recall that θ = iϑ
 77

Fig.18 Spectrum of Hg (θ)

In Fig.17 we see the effect of including the perturbation in the Hamiltonian. As in the case
of T > 0 (but with other geometry) it produces a displacement of the eigenvalues to the
bottom half plane (movement associated with the Fermi-Golden’s rule) and to the sides left
or right (movement associated with the Lamb shift). The ground state remains on the axis
(but displaced to one side). In the Fig.17 we have supposed E2 is double and splits into
two different domains and that E1 is simple and all displaces to the left.

5.5 Smooth Feshbach Map. Cuspidal Domains of the Spec-

trum for T > 0

In this section we will present the Smooth Feshbach map (SFM) and then we will use it to
show that the spectrum of Lg (θ) is cuspidal near zero.
The SFM is a variant of the FM in which instead of using projections one uses smooth
partitions of unity on Hilbert space. We choose a smooth positive operator χ bounded
above by the identity on H and we define:
p
χ := 1 − χ2 (5.157)

Taking into account the analysis done before we can write our Liouvillian in each step of
the RG as:
L = T [L, z] − E[z] · + gW [z] (5.158)

where E[z] ∈ C is a number which is a purely conventional factor. T [L, z] is a closed

operator function of L, and W[z] (sum of Wick monomials) is a perturbation defined on the
entire domain of T. Our purpose is to understand the spectrum of Lg (θ) near 0. We will use
the isospectrality of the renormalization transformation and after a enough number of steps
 78

of the RG the term W[z] tends to zero, simplifying the analysis of the spectrum of Lg (θ).
The domain of the operator T is assumed to be invariant under χ and χ. T is assumed also
to commute with χ and χ. We define Lχ := T + χW χ and Lχ := T + χW χ, we assume
that both are bounded invertible on the range of χ and χ respectively. Let Lχ = U |Lχ | be
the polar decomposition of the operator Lχ . We assume also that |Lχ |−1/2 U −1 χW χ and
χW χ|Lχ |−1/2 U −1 extend to bounded operators on H. With these assumptions we define
the smooth Feshbach map as:

Fχ (L) := T [z; L] − E[z] + gχW χ − g 2 χW χL−1

χ χW χ (5.159)

For a more complete analysis of the SMF see [6]. The reason why we use SFM is the
smoothness of the derivative of these smooth characteristic functions. In contrast, by taking
derivatives of projections (case of the FM) we obtain δ-distributions and the calculations
are much more involved.

Curve4
 Curve3

Curve2

  
 Curve1
+ 
C3  +


 
+
C2 +


C1

Fig.19 Approximative spectrum of Lg (θ) near 0

Now we will use this SFM to show that the spectrum of Lg (θ) is cuspidal near 0. In
Fig.19 we show it in a graphical form after the application of three steps (step 0,1,2)
of the renormalization group method. We begin with the curve 1 which represents an
approximation of domain D of Fig.14. In the domain below of this curve we place the
center of the new circle C1. This circle correspond to U0in = {z ∈ D1/2 | |z| < ρ0 /2}. Now
we do the step 0 of the RG for values of z inside of the circle C1. For these values of z we
 79

have proven in Theorem 5.3.3. that the FM is well-defined. We analyze now the values of
z inside the circle C1 where R0 is invertible. Because of the isospectrality property of the
RG, these values of z are in the spectrum of Lg (θ). We deduce based on Theorem IV.2 of
[14] that above the curve 2 the FM is invertible. The invertibility of F above the curve 2
can also be seen in the following intuitive way, we know that:

Fχ (L) := T [z; L] − E[z] + gχW χ − g 2 χW χL−1

χ χW χ (5.160)

T [z; L] − E[z] corresponds in the step 0 to L0 (θ) − z and we know that this has a spectrum
like in Fig.11, i.e. L0 (θ) − z above the curve 2 is invertible, the other two terms gχW χ and
g 2 χW χL−1 µ 2 µ
χ χW χ are small in comparison with the first one (order O(gρ ) and O(g ρ ),
µ > 0 where ρ is the parameter of the energy scale). We take a point below of the curve 2
and this will be the center of the circle C2. Using Theorem 5.2.3. the FM is well-defined
for the points inside of U1in = {z ∈ D1/2 | |z − E0 | < ρ/2}. We do the step 1 of the RG
and we analyze where R1 is invertible. Based again on theorem IV.2 of [14] we obtain that
the values of z above the curve 3 are in the resolvent set of Lg (θ). We take a point below
of curve 3 and draw the circle C3, etc. In this way we obtain, after an enough number of
steps, a cuspidal domain as shown in Fig.19.
We will use the following smooth characteristic function:

χ := P0 ⊗ χρ [Laux ] (5.161)

where:
π
χρ [Laux ] = sin[ Θ(Laux /ρ)] (5.162)
2
where P0 is the orthogonal projection onto the space corresponding to the eigenvalue 0 and
Θ is a smooth, positive function on the interval [0, ∞), with 0 ≤ Θ ≤ 1, Θ ≡ 1 on [0, 43 ] and
Θ ≡ 0 on [1, ∞). We make the following definition:
q
χρ [Laux ] := 1 − χρ [Laux ]2 (5.163)

χρ [Laux ] is smooth as well. Doing the same analysis as with the FM we can make a Neumann
series expansion of the SFM, Eq. (5.159):

Fχ (L) := T [z; L] − E[z] + χW [z]χ − χW [z]χ(T [z; L] − E[z] + χW [z]χ)−1 χW [z]χ =

χW [z]χ
T [z; L] − E[z] + χW [z]χ − χW [z]χ(T [z; L] − E[z])−1 (1 + )−1 χW [z]χ =
T [z; L] − E[z]
 80

T [z; L] − E[z] + χW [z]χ − χW [z]χ(T [z; L] − E[z])−1 ×

 χW [z]χ χW [z]χ χW [z]χ 
1+ + + . . . χW [z]χ =
T [z; L] − E[z] T [z; L] − E[z] T [z; L] − E[z]

= T [z; L] − E[z] + χW [z]χ − χW [z]χ(T [z; L] − E[z])−1 ×

 χW [z]χ2 W χ χW [z]χ2 W χ χW [z]χ2 W χ2 W χ 
χW [z]χ + + + + ...
T [z; L] − E[z] T [z; L] − E[z] T [z; L] − E[z]
and the result is:
∞
X  χ2 L−1
= T [z; Laux ] − E[z] + (−g)L−1 χW [z] W [z] χ (5.164)
T [z; L] − E[z]
L=1

Let us analyze the expression of χ2 :

χ2 = P 0 ⊗ χ2ρ [Laux ] + P0 ⊗ χ2ρ [Laux ] + P 0 ⊗ χ2ρ [Laux ]

where we have considered χ2 = P02 ⊗ χ2ρ [Laux ] and P02 = P0 , and we obtain:

χ2 = P 0 ⊗ χ2ρ [Laux ] + (P 0 + P0 )χ2ρ [Laux ]

defining χ21 and χ22 as:

χ2 = P 0 ⊗ χ2ρ [Laux ] + at ⊗ χ2ρ [Laux ] = χ21 + χ22 (5.165)

We will also use the following expressions already seen ( see Eq. (5.93), Eq. (5.94) and
Eq. (5.95)) but now χρ is a smooth characteristic function:
M
X
χ2 := Pα ⊗ χαf (0)
0



f : for α = 1, 2 · · · M
where χαf [ω] :=
 χ2 [Laux + ω] : for α = 0
ρ

where Pηα is the projection onto the space generated by the eigenvector of the eigenvalue εα
and {ε0 · · · εM } = {Ei,j |0 ≤ i, j ≤ N − 1} with ε0 := 0 and M ≤ N (N − 1), that is we have
N zero eigenvalues. These correspond to the differences Eij = Ei − Ej when Ei = Ej and
N 2 − N = N (N − 1) non-zero eigenvalues. Now we do the same procedure (Wick ordering)
 81

as in Eq. (5.7), Eq. (5.8) and Eq. (5.9) and explained also in order g 2 in a past section (see
section 4.3). That is to say:

e 0 [z] − z := Fχ (Lg (θ) − z) = χ(E

L e0 [z] − z + Te0 [z; L] + W
f0 [z])χ (5.166)

The terms W are becoming smaller under successive steps of our RG analysis. That means
the process tends to En [z] + Tn [z; L]. We are interested in knowing if each of the terms
Ei + Ti generated in this process are bounded, otherwise we could not know the spectrum of
Li [z], which because of the isospectrality condition of the FM coincides with Lg (θ). Let us
analyze wether Te0 is bounded. From the past sections Eq. (5.8), Eq. (5.13) for m = n = 0
and with the notation of Eq. (5.15) and Eq. (5.17) we know that the expression of Te0 has
the following form:
 
Te0 [z, r] = P0 hΩ| Fχ (Lg (θ) + rf (θ) − z) − Fχ (Lg (θ) − z) ΩiP0 =
X X Z
n
= rf (θ) − (−g) dk (n) P0 Gθσ1 (k1 )Pα1 . . . Pαn −1 Gθσn (kn )P0 ×
σ1 ...σn =±
α1 ...αn−1 =0...M


×hΩ|{aσ1 (k1 )fα1 [L + r] · · · fαn−1 [L + r]aσn (kn ) − aσ1 (k1 )fα1 [L] · · · fαn−1 [L]aσn (kn )}Ωi

r=L
(5.167)
and taking into account the notation of Eq. (5.8):

0 0 0
Te0 [z, r] = rf (θ) + {e
ω0,0 [z, r] − ω
e0,0 [z, 0]} = rf (θ) + ∆e
ω0,0 [z, r] (5.168)

where:

L := (Lf , Laux ) r := (rf , raux ) (5.169)

rf (θ) := rf cos ϑ − iraux sin ϑ (5.170)
χαf (raux ) χαf (raux )
fα [rf , raux ] := = (5.171)
εα + rf (θ) − z εα + rf cos ϑ − iraux sin ϑ − z

∂fα [rf , raux ] ∂χαf (raux )/∂raux χαf (raux )(−1)(−i sin ϑ)
= + (5.172)
∂raux εα + rf cos ϑ − iraux sin ϑ − z (εα + rf cos ϑ − iraux sin ϑ − z)2
where εα are the eigenvalues of Lat .
∂ω 0 [z,r]
≤ O(g ε ) but before that, we need several
e0,0
In proposition 5.5.3 we will see that k ∂raux k
lemmas:
 82

Lemma 5.5.1. Let z be so that |Rez| < ρ0 /2 and |Imz| ≥ − 12 ρ0 sin ϑ (in particular for
z ∈ S0,< , see Fig.13) where ρ0 := g 2− for  small and 0 < ϑ < π/2. For these values
of z and χ := P0at ⊗ χρ0 [Laux ], where χρ0 [Laux ] is a smooth characteristic function as in
Eq. (5.162), then we have:
L
aux + ρ0 2 4

χ ≤ (5.173)
L0 (θ) − z sin ϑ

Proof. we use χ2 = χ21 + from Eq. (5.165):

χ21 := P 0 ⊗ χ2ρ [Laux ] (5.174)

χ22 := at ⊗ χ2ρ [Laux ] (5.175)

The proof is the same as the proof of Proposition 5.3.2. using χ21 and χ22 instead of Q1 and
Q2 .

From this lemma and Eq. (5.171) we see that:

4
(Laux + ρ0 )fα [L] ≤ (5.176)

sin ϑ

We will use this result in the next proposition.

In the same way we would prove:

Lemma 5.5.2. With the same conditions as before, then:

aux + ρ0 )∂χ2 /∂raux

(L 4
≤ (5.177)
L0 (θ) − z sin ϑ

(L 2 2
aux + ρ0 ) χ i sin ϑ 16

≤ (5.178)
(L0 (θ) − z)2 sin ϑ

From this lemma we see using Eq. (5.172) that:

∂fα [L] 20
(Laux + ρ0 )2 ≤ (5.179)

∂raux sin ϑ

which we use in the following proposition:

Proposition 5.5.3. Assume that z ∈ S0,< (see Fig.13) and g > 0 sufficiently small, then:
∂ω 0
e0,0 [z, r] ∂ Te0 [z, r] ∂rf (θ)

= − ≤ O(g ε ) (5.180)
∂raux ∂raux ∂raux

 83

Proof. If we take derivatives in Eq. (5.167)

∞ Z
∂ Te0 [z, r] ∂rf (θ) ∂ X X
= − (−g)n dk (n) P0 Gθσ1 (k1 )Pα1 . . . Pαn −1 Gθσn (kn )P0 ×
∂raux ∂raux ∂raux
n=2 σ1 ...σn =±
α1 ...αn−1 =0...M

×hΩ|{aσ1 (k1 )fα1 [L + r] · · · fαn−1 [L + r]aσn (kn ) − aσ1 (k1 )fα1 [L] · · · fαn−1 [L]aσn (kn )}Ωi

r=L

recalling that rf (θ) = rf cos ϑ − iraux sin ϑ and θ = iϑ, we have:

∞ Z
∂ Te0 [z, r] X X
= −i sin ϑ − (−g)n dk (n) P0 Gθσ1 (k1 )Pα1 . . . Pαn −1 Gθσn (kn )P0 ×
∂raux
n=2 σ ...σn =± 1
α1 ...αn−1 =0...M

n−1
X ∂fαj [L + r] 
× hΩ|{aσ1 (k1 )fα1 [L + r] · · · aσj (k1 ) · · · fαn−1 [L + r]aσn (kn )}Ωi

∂raux r=L
j=1
P∞ P
We analyze now one term (the nth ) of the sum n=2 σ1 ...σn =± , taking absolute value
α1 ...αn−1 =0...M
of it:
 n−1

θ ∂fαj [L + r]
 X  
n σ1 θ σj σn θ
(−g) hΩ|{a (Gσ1 )fα1 [L + r] · · · a (Gσj ) · · · fαn−1 [L + r]a (Gσn )}Ωi =
  
 ∂raux r=L 
j=1

 n−1
X
ρ0 (−g)n hΩ|(Laux + ρ0 )−1/2 {aσ1 (Gθσ1 )(Laux + ρ0 )−1/2 (Laux + ρ0 )fα1 [L + r](Laux + ρ0 )−1/2 · · ·

 | {z }
j=1
Eq. (5.176)
∂fαj [L + r]
(Laux + ρ0 )−1/2 aσj (Gθσj )(Laux + ρ0 )−1 (Laux + ρ0 )2 (Laux + ρ0 )−1 · · ·
∂raux
| {z }
Eq. (5.179)

 
−1/2 −1/2 σn −1/2
(Laux + ρ0 ) (Laux + ρ0 )fαn−1 [L + r](Laux + ρ0 ) a (Gθσn )(Laux + ρ0 ) }Ωi
 

r=L 
(5.181)
we can check this equation in the same way as with Eq. (5.89), where the ρ0 at the beginning
of the last equation comes from the following consideration:

ρ0 |hΩ|(Laux + ρ0 )−1/2 · · · · · · (Laux + ρ0 )−1/2 Ωi| (5.182)

and Laux |Ωi = 0. Using now Eq. (5.176) and Eq. (5.179):
4 n−2 20
(5.181) ≤ ρ0 (n − 1)g n ( ) k(Laux + ρ0 )−1/2 aσj (Gθσj )(Laux + ρ0 )−1 k ×
sin ϑ sin ϑ
Yn
k(Laux + ρ0 )−1 aσj+1 (Gθσj+1 )(Laux + ρ0 )−1/2 k k(Laux + ρ0 )−1/2 aσi (Gθσi )(Laux + ρ0 )−1/2 k
i=1
i6=j,j+1
 84

that is to say:

4 n−2 20
= ρ0 (n − 1)g n ( ) k(Laux + ρ0 )−1/2 aσj (Gθσj )(Laux + ρ0 )−1/2 kk(Laux + ρ0 )−1 k
sin ϑ sin ϑ
k(Laux + ρ0 )−1/2 aσj+1 (Gθσj+1 )(Laux + ρ0 )−1/2 k ×
n
Y
k(Laux + ρ0 )−1/2 aσi (Gθσi )(Laux + ρ0 )−1/2 k
i=1
i6=j,j+1

taking into account the estimate Eq. (5.63) of lemma 5.3.1.:

k(Laux + ρ0 )−1/2 aσi (Gθσi )(Laux + ρ0 )−1/2 k = O(ρ−1/2 ) (5.183)

we obtain finally:
∂ Te [z, r] ∞
0
X 1 g
+ i sin ϑ ≤ Kρ0 ( 1/2 )n (5.184)

∂raux ρ0 ρ

n=2 0
g g2
where K = (n − 1)( sin4 ϑ )n−2 sin
20
ϑ . Considering that ρ0 = g
2−ε , then (
1/2 )2 = g 2−ε
= g ε that
ρ0
is to say:
∂ Te [z, r]
0
+ i sin ϑ ≤ O(g ε ) (5.185)

∂raux

In the same way we would prove (see [7]):

Proposition 5.5.4. Assume that z ∈ S0,> . Then there exists a constant, C < ∞, such
that, for g > 0 sufficiently small:

kE0 [z] + ig 2 Γ0 k ≤ Cg 2+2 (5.186)

m n
(0)
 Cg m+n Y κ(kj ) Y κ(e kj )
kωm,n [z, r, K (m,n) ]k ≤ Cρ0 (5.187)
ρ0 ω(kj )1/2−µ
j=1 j=1 kj )1/2−µ
ω(e
m n
d3 kj d3 e
Z
(0)
Y Y kj
k∂raux ωm,n [z, r, K (m,n) ]k ≤ C(Cg)m+n (5.188)
Bρm+n
0 j=1
ω(kj )3/2−µ j=1 ω(kj )
e 3/2−µ

where Br := {k|ω(k) < r}

Similar bounds can be established for the step 1.

After each step of the RG we have expressions of the form:


Ln = Fχ (Ln−1 ) := Pn Tn [z; L] − En [z] + Wn [z] Pn (5.189)
 85

and then we handle Wn [z] as a perturbation. We develop Ln in a Neumann series and

we study the convergence of this Neumann series. The convergence will be secured for
sufficiently big values of Tn [z; L] − En [z]. Taking into account that:
(n−1) (n−1)
Tn [z; L] = Tn−1 [z; L] + {ω0,0 (z, L) − ω0,0 (z, 0)} (5.190)

and that we have seen in theorem 5.5.3.:

∂ (0) 0
{ω (z, L) − ω0,0 (z, 0)} ≤ O(g  ) = ∆0 (5.191)
∂raux 0,0
we could prove inductively that:
∂ (n−1) n−1
{ω (z, L) − ω0,0 (z, 0)} ≤ ∆n−1 (5.192)
∂raux 0,0
where the bound ∆n−1 depends only on:
(n−1)
sup kWM,N k ≤ Cg 2 ρµn (5.193)
N +M ≥1

From these considerations we can also prove inductively that:

1
|Tn [z; L]| ≥ sin ϑ(1 − δn )Laux |δn | ≤ ∀n (5.194)
2
From Eq. (5.185), we have:
1
T0 [z, r] ≥ sin ϑraux − Laux (5.195)
2
Suppose that it is true for n − 1:

|Tn−1 [z; L]| ≥ sin ϑ(1 − δn−1 )Laux (5.196)

then from Eq. (5.192) and Eq. (5.190):

4n−1
|Tn [z; L]| ≥ sin ϑ(1 − δn−1 − )Laux = sin ϑ(1 − δn )Laux (5.197)
sin ϑ
that is to say :
4n−1
δn = δn−1 + ; ∆0 = g  δ0 = 1/2 (5.198)
sin ϑ
this is the renormalization group flux equation.
Being the meaning of Wn [z] a perturbation and having seen that after the application of
an enough number of steps this Wn [z] becomes small (see Eq. (5.187) and Eq. (5.188)), we
see that our Ln tends to an unperturbed Liouvillian, the domain of which we have seen in
chapter 4 is cuspidal.
Chapter 6

The Level Shift Operator

6.1 Introduction

To uniformize the notation and make it applicable to the magnons in a ferromagnet we

will work on the Hilbert space (Hpp ⊗ Hpp ) ⊗ Hcβ , where Hcβ is the Hilbert space formed
by the GNS Hilbert space of the hamiltonian with an absolute continuous spectrum (that
would be the electromagnetic field hamiltonian Hfβ in the case of the atom coupled to a
radiation field) and by Hpp ⊗ Hpp , the latter is the Hilbert space obtained from K through
the isomorphism IC : K 7−→ Hpp ⊗ Hpp , where Hpp is the Hilbert space of the hamiltonian
whose spectrum consists of isolated eigenvalues (that would be the atomic Hilbert space Hat
in the case of the atom coupled to a radiation field). We take a general (minimal coupling
and dipole approximation) interaction of the form:
Z
I= dk[G(k) ⊗ a∗ (k) + G∗ (k) ⊗ a(k)] (6.1)

where G(k) is an operator on Hpp i.e. a CN ×N -matrix or in a shorter notation I = a∗ (G) +

a(G∗ ) our Liouvillian is:

Lg = L0 + gW (6.2)
L0 = Lc ⊗ pp + c ⊗ Lpp (6.3)
Lpp = Hpp ⊗ pp − pp ⊗ Hpp (6.4)
Z
Lc = dkω(k)[l(a∗ (k)l(a(k) − r(a∗ (k)r(a(k))] where r(a] ) = Jc l[a] ]Jc (6.5)

86
 87

where l and r are two representations of P on Hcβ . We know (see chapter 2) that these
two representations are related by r(a] ) = Jc l[a] ]Jc where Jc is the modular conjugation
operator on Hcβ . This antiunitary operator was defined in Eq. (2.40) as:
β
Jc l[a]|Ωβc i := e− 2 Lc l[a]∗ |Ωβc i

W = l[I] − r[I] (6.6)

l[I] = l[a∗ ] ⊗ (G ⊗ ) + l[a] ⊗ (G∗ ⊗ ) (6.7)
r[I] = Jc l[a∗ ]Jc ⊗ ( ⊗ T GT ) + Jc l[a]Jc ⊗ ( ⊗ T G∗ T ) (6.8)

We will write T GT = G = (G∗ )t

σ(Lpp ) = x x x x x x x
0
σ(Lf ) = x
0
σ(L0 ) = x x x x x x x
0
Fig.20

We will try to understand the stability of these eigenvalues.

- First we do a complex deformation, the equality Eq. (4.21):

hϕ|(Lg − z)−1 ψi = hÛ (θ̄)ϕ|(Lg (θ) − z)−1 Û (θ)ψi (6.9)

permits us to do an analytic continuation of the left side from the upper half-plane to the
lower half-plane for all the values of z where the right side exists. From now on we will take
θ = iϑ, with ϑ ∈ R+ .
- Next we use a Feshbach Map to explore the properties of σ(Lg (iϑ)) , 0 < ϑ < ϑ0 with the
help of a perturbative method, provided that we know σ(L0 (iϑ)).
As Projection we take:
P = P (Lpp = Eij ) ⊗ PΩc (6.10)

where Eij = Ei − Ej ; i, j = 1 . . . N − 1 are the eigenvalues of Lpp and where Ωf is the KMS
state on Hfβ .
We will use:
Lg (θ, z) := Lg (θ) − z (6.11)
 88

also:

Lg (θ, z) := Lpp (z) + Lc (θ) + gW (θ) (6.12)

L0 (θ, z) := Lpp (z) + Lc (θ) (6.13)
Lpp (z) := Lpp − z (6.14)

With these notations and from Eq. (4.39) the Feshbach Map results in:

FP (Lg (θ, z)) = P Lg (θ, z)P − g 2 P W (θ)P̄ (L̄g (θ, z))−1 P̄ W (θ)P (6.15)

where we have taken into account that:

P Lg (θ, z)P̄ = gP W (θ)P̄ (P commutes with L0 ) (6.16)

and P̄ = 1 − P, L̄g (θ, z) = P̄ Lg (θ, z)P̄ (6.17)

From the form of W, see Eq. (6.7), Eq. (6.8) we see that P W P = 0 also P Lc (θ)P = 0 and:

FP (Lg (θ, z)) = P Lpp (z)P − g 2 P W (θ)P̄ (L̄g (θ, z))−1 P̄ W (θ)P (6.18)

writing now Lg (θ, z) = L0 (θ, z) + gW (θ), we have:

h i−1
Lg (θ, z)−1 = (L0 (θ, z) + gW (θ))−1 = L0 (θ, z)(1 + L0 (θ, z)−1 gW (θ)) (6.19)
∞ 
X n
(1 + L0 (θ, z)gW (θ))−1 L0 (θ, z)−1 = L0 (θ, z)−1 − gW (θ)L0 (θ, z)−1 (6.20)
n=0

Substituting this in Eq. (6.18):

h ∞
−1
X
n i
2
FP (Lg (θ, z)) = P Lpp (z)P − g P W (θ)P L0 (θ, z) − gW (θ)L0 (θ, z)−1 P W (θ)P
n=0
(6.21)
One can show that the resolvent on the r.h.s. exists provided P is replaced by P0 = P (Lpp =
Eij ) ⊗ PLaux ≤ρ for some small ρ > 0 and Imz ≥ − 12 ρ and then we use a limit argument as
ρ → 0. From now on we will set z = Eij . The lowest order of Eq. (6.18) is from Eq. (6.21)
L0 (θ, z)−1 i.e. (recalling Eq. (6.13) and Eq. (4.15)):

0 0
L0 (θ, z) = P (Lpp + Lc cos ϑ − iLaux sin ϑ − z)P (6.22)
 89

with:
0
P = P (Lpp 6= Eij ) ⊗ PLaux ≤ρ + ⊗ P (Laux > ρ) := Q1 + Q2 (6.23)

with Q1 we have:
d
Q1 L0 (θ, z) = Q1 (Lpp + Lc cos θ − iLaux sin θ − z)Q1 = Q1 (Lpp − z) > (6.24)
2
where d is the minimal distance between eigenvalues of Lpp , therefore we have kQ1 L0 (θ, z)k <
2
d. For Q2 is easy to see that Q2 L0 (θ, z) is bounded.
We consider now only the term in second order in g of Eq. (6.21):

FP (Lg (θ, z))(2) = −P W (θ)P (L0 (θ, z))−1 P W (θ)P (6.25)

This term is defined as the Level-shift operator:

Γ(2) := −FP (Lg (θ, z))(2) = P W (θ)P (L0 (θ, z))−1 P W (θ)P (6.26)

We see that since P W (θ)P = 0 then P W (θ)P = P W (θ) and P W (θ)P = W (θ)P and we
have:
P W (θ)(L0 (θ, z))−1 W (θ)P = lim P W (θ)(L0 (θ) − z − iε)−1 W (θ)P (6.27)
ε→0+

This last function is analytic in θ, therefore:

P W (θ)(L0 (θ, z))−1 W (θ)P = lim P W (L0 − z − iε)−1 W P (6.28)

ε→0+

We are interested in the imaginary part of Γ(2) , which will yield information on life times
of resonances:
n1 o
Im(FP (Lg (θ, z))(2) ) = lim P W [(L0 − z − iε)−1 − (L0 − z + iε)−1 ]W P =
− g2
ε→0+ 2i
 1 2iε 
= P W lim − g 2 W P = −g 2 P W πδ(L0 − z)W P (6.29)
ε→0+ 2i (L0 − z)2 − ε2
where z = Eij . In the last equality we have used:
1 ε ε→0
2 2
−→ δ(x) (6.30)
πx +ε
See for example [35].
In the next sections we will calculate:

ImΓ(2) := −Im(FP (Lg (θ, z))(2) ) = πP W δ(L0 (z))W P (6.31)

 90

6.2 General Calculation

ImΓ(2) corresponds to an operator Σ(2) on Hpp ⊗ Hpp given by:


Σ(2) = hΩc |W δ(Lc + Lpp (z))W Ωc i (6.32)

Hpp

and with Eq. (6.6) we obtain:


Σ(2) = hΩc |l(I)δ(Lc + Lpp (z))l(I)Ωc i I (6.33)

H
 pp
+hΩc |r(I)δ(Lc + Lpp (z))r(I)Ωc i II (6.34)

Hpp

−hΩc |l(I)δ(Lc + Lpp (z))r(I)Ωc i − III (6.35)

H
 pp
−hΩc |r(I)δ(Lc + Lpp (z))l(I)Ωc i − IV (6.36)

Hpp

Now we develop Eq. (6.33), Eq. (6.34), Eq. (6.35) and Eq. (6.36):

I = hΩc |l(I)δ(Lc + Lpp (z))l(I)Ωc i =

Hpp
= hΩc |{(G ⊗ ) ⊗ l[a∗ ] + (G∗ ⊗ ) ⊗ l[a]}δ(Lc + Lpp (z)) ×

∗ ∗
×[(G ⊗ ) ⊗ l[a ] + (G ⊗ ) ⊗ l[a]]Ωc i (6.37)

Hpp

From these products we have only to consider the products that contain one a and one a∗ ,
all the others vanish because of:

Lemma 6.2.1.

hΩc |l[a(k)]l[a(k 0 )]Ωc i = hΩc |l[a∗ (k)]l[a∗ (k 0 )]Ωc i = 0 (6.38)

hΩc |r[a(k)]r[a(k 0 )]Ωc i = hΩc |r[a∗ (k)]r[a∗ (k 0 )]Ωc i = 0 (6.39)
hΩc |l[a(k)]r[a∗ (k 0 )]Ωc i = hΩc |l[a∗ (k)]r[a(k 0 )]Ωc i = 0 (6.40)
hΩc |r[a(k)]l[a∗ (k 0 )]Ωc i = hΩc |l[a∗ (k)]r[a(k 0 )]Ωc i = 0 (6.41)

and k,k 0 can be equal or different.

Proof. Eqs. (6.38) are trivial, for example hΩc |l[a∗ (k)]l[a∗ (k 0 )]Ωc i = hΩc |(k +k 0 )i = 0, where
we have considered that l[a∗ (k 0 )]|Ωc i = |k 0 i.
Eqs. (6.39) are similar deduced, for example:

hΩc |r[a(k)]r[a(k 0 )]|Ωc i = hΩc |Jc l[a(k)]Jc Jc l[a(k 0 )]Jc |Ωc i (6.42)
 91

where we have used Eq. (2.40) and now we use Eq. (2.57) Jc |Ωc i = |Ωc i and Eq. (2.46)
Jc Jc = 1 :

hΩc |l[a(k)]Jc Jc l[a(k 0 )]|Ωc i = hΩc |l[a(k)]l[a(k 0 )]|Ωc i = h(k + k 0 )|Ωc i = 0 (6.43)

Eqs. (6.40) and Eqs. (6.41) are similar deduced. We will deduce one of them, for the other
the procedure is similar:

hΩc |l[a(k)]r[a∗ (k 0 )]Ωc i = hΩc |l[a(k)]Jc l[a∗ (k 0 )]Jc Ωc i =

β β 0
= hΩc |l[a∗ (k)]e− 2 Lc l[a∗ (k 0 )]Ωc i = hΩc |l[a∗ (k)]l[a∗ (k 0 )]e− 2 (Lc +ω(k )) Ωc i = 0
β
where we have used hΩc |l[a(k)]Jc = hΩc |l[a∗ (k)]e− 2 Lc

Therefore from Eq. (6.37):

Z h 
I= dkdl (G(k) ⊗ )hΩc |l[a∗ (k)]δ(Lc + Lpp (z))l[a(l)]Ωc i(G∗ (l) ⊗ ) + (6.44)

Hpp
 i
+(G∗ (k) ⊗ )hΩc |l[a(k)]δ(Lc + Lpp (z))l[a∗ (l)]Ωc i(G(l) ⊗ ) (6.45)

Hpp

Now we apply Lemma Appendix A 3.1.2, f (Lc )|Ωc i = f (0)|Ωc i, Eq. (2.62) and Eq. (2.63):

hΩc |l[a∗ (k)]δ(Lc + Lpp (z))l[a(l)]Ωc i = hΩc |δ(Lc − ω(k) + Lpp (z))l[a∗ (k)]l[a(l)]Ωc i =
1
= δ(Lpp (z) − ω(k))hΩc |l[a∗ (k)]l[a(l)]Ωc i = δ(Lpp (z) − ω(k))δ(k − l) βω(k)
e −1
in the same way:

hΩc |l[a(k)]δ(Lc + Lpp (z))l[a∗ (l)]Ωc i = hΩc |δ(Lc + ω(k) + Lpp (z))l[a(k)]l[a∗ (l)]Ωc i =
1
= δ(Lpp (z) + ω(k))hΩc |l[a∗ (k)]l[a(l)]Ωc i = δ(Lpp (z) + ω(k))δ(k − l)e βω(k) βω(k)
e −1
and finally we get:
Z h 1 i
I= dkdl (G(k) ⊗ )δ(Lpp (z) − ω(k))δ(k − l) βω(k) (G∗ (l) ⊗ ) + (6.46)
e −1 Hpp

e βω(k)
Z h i
+ dkdl (G∗ (k) ⊗ )δ(Lpp (z) + ω(k))δ(k − l) βω(k) (G(l) ⊗ ) (6.47)
e −1 Hpp

and using polar coordinates:

Z
1 
I= dω 4πω 2 (G(k) ⊗ )δ(Lpp (z) − ω(k)) βω(k) G∗ (k) ⊗  + (6.48)

R + e − 1 Hpp

e βω(k)
Z 
+ dω 4πω 2 (G∗ (k) ⊗ )δ(Lpp (z) + ω(k)) βω(k) (G(k) ⊗ ) (6.49)

R + e − 1 Hpp
 92

Let us make the same procedure for II Eq. (6.35) now with the right representation on Hcβ :

II = hΩc |r(I)δ(Lc + Lpp (z))r(I)Ωc i
=

Hpp
h i
= hΩc | ( ⊗ T GT ) ⊗ Jc l[a∗ ]Jc + ( ⊗ T G∗ T ) ⊗ Jc l[a]Jc
h i
δ(Lc + Lpp (z)) ( ⊗ T GT ) ⊗ Jc l[a∗ ]Jc + ( ⊗ T G∗ T ) ⊗ Jc l[a]Jc Ωc i = (6.50)

Because of Eq. (6.39) only cross products are different from 0:

Z h i
∗
II = dkdl ( ⊗ G(k))hΩc |r[a∗ (k)]δ(Lc + Lpp (z))r[a(l)]Ωc i( ⊗ G (l)) + (6.51)
Hpp
Z h i
∗
+ dkdl ( ⊗ G (k))hΩc |r[a(k)]δ(Lc + Lpp (z))r[a∗ (l)]Ωc i( ⊗ G(l)) (6.52)
Hpp

We have used that r[a] = Jc l[a]Jc . Now we use the following calculation:
β
r[a(l)]Ωc i = Jc l[a(l)]Jc |Ωc i = Jc l[a(l)]|Ωc i = e − 2 Lc l[a(l)]∗ |Ωc i =
β β
= l[a(l)]∗ e − 2 (Lc +ω(l)) |Ωc i = l[a(l)]∗ e − 2 ω(l) |Ωc i (6.53)
β
where we have used Jc |Ωc i = |Ωc i, Jc l[a]|Ωc i = e − 2 Lc l[a]∗ |Ωc i, the pull-through formula
and Lc |Ωc i = 0.
In the same way we have:

hΩc |r[a∗ (k)] = hΩc |Jc l[a∗ (k)]Jc = hΩc |l[a∗ (k)]Jc =
β β β
= hΩc |l[a(k)]e − 2 Lc = hΩc |e − 2 (Lc +ω(k)) l[a(k)] = hΩc |e − 2 ω(k) l[a(k)] (6.54)

the other two similar results that we need are:

β
r[a∗ (l)]Ωc i = l[a(l)]e 2
ω(l)
|Ωc i (6.55)
β
hΩc |r[a(k)] = hΩc |e 2
ω(k)
l[a(k)]∗ (6.56)

with these results it follows from Eq. (6.51) and Eq. (6.52):
Z h i
β ∗
II = dkdl ( ⊗G(k))hΩc |δ(Lc +ω(k)+Lpp (z))l[a(k)]l[a∗ (l)]Ωc ie − 2 (ω(k)+ω(l)) ( ⊗G (l)) +
Hpp
Z (6.57)
h i
∗ β
+ dkdl ( ⊗ G (k))e 2
(ω(k)+ω(l))
hΩc |δ(Lc − ω(k) + Lpp (z))l[a∗ (k)]l[a(l)]Ωc i( ⊗ G(l))
Hpp
(6.58)
 93

and now with Eq. (2.62) and Eq. (2.63):

e βω(k)
Z h i
β ∗
II = dkdl ( ⊗ G(k))δ(Lpp (z) + ω(k))e − 2 (ω(k)+ω(l)) βω(k) δ(k − l)( ⊗ G (l)) (6.59)
e −1 Hpp
Z h
∗ β 1 i
+ dkdl ( ⊗ G (k))δ(Lpp (z) − ω(k))e 2 (ω(k)+ω(l)) βω(k) δ(k − l)( ⊗ G(l)) + (6.60)
e −1 Hpp

using polar coordinates:

Z
1 ∗

II = dω4πω 2 ( ⊗ G(k))δ(Lpp (z) + ω(k)) βω(k) ( ⊗ G (k)) (6.61)

R+ e −1 Hpp
Z
e βω(k) 
∗
+ dω4πω 2 ( ⊗ G (k))δ(Lpp (z) − ω(k)) βω(k) ( ⊗ G(k)) + (6.62)

R+ e −1 Hpp

For the term III, Eq. (6.35):


III = hΩc |l(I)δ(Lc + Lpp (z))r(I)Ωc i =

Hpp
h i
= hΩc | (G ⊗ ) ⊗ l[a∗ ] + (G∗ ⊗ ) ⊗ l[a]
h i 
δ(Lc + Lpp (z)) ( ⊗ T GT ) ⊗ Jc l[a∗ ]Jc + ( ⊗ T G∗ T ) ⊗ Jc l[a]Jc Ωc i (6.63)

Hpp

We use now Eq. (6.40) and Eq. (6.41), therefore the only surviving terms are the ones that
contain both creation operator or both annihilation operators:
Z h i
III = dkdl (G(k) ⊗ )hΩc |l[a∗ (k)]δ(Lc + Lpp (z))r(a∗ (l))Ωc i( ⊗ G(l)) + (6.64)
Hpp
Z h i
∗
+ dkdl (G∗ (k) ⊗ )hΩc |l[a(k)]δ(Lc + Lpp (z))r(a(l))Ωc i( ⊗ G (l)) (6.65)
Hpp

We take into account now the equations r[a(l)] = Jc l[a(l)]Jc , Jc |Ωc i = 0, Jc l[a(l)]|Ωc i =
β
L
e l[a∗ (l)]|Ωc i and similar equations for the creation operators:
2 c

Z h i
β
III = dkdl (G(k) ⊗ )hΩc |δ(Lc + Lpp (z) − ω(k))e 2 ω(l) l[a∗ (k)]l[a(l)]Ωc i( ⊗ G(l)) +
Hpp
Z (6.66)
h i
− β2 ω(l) ∗
+ dkdl (G∗ (k) ⊗ )hΩc |δ(Lc + Lpp (z) + ω(k))e l[a(k)]l[a∗ (l)]Ωc i( ⊗ G (l))
Hpp
(6.67)
considering now Eq. (2.62) and Eq. (2.63):
β
e 2 ω(l)
Z h i
III = dkdl (G(k) ⊗ )δ(Lpp (z) − ω(k)) βω(k) δ(k − l)( ⊗ G(l)) + (6.68)
e −1 Hpp

e βω(k)
Z h i
β ∗
+ dkdl (G∗ (k) ⊗ )δ(Lpp (z) + ω(k))e − 2 ω(l) βω(k) δ(k − l)( ⊗ G (l)) (6.69)
e −1 Hpp
 94

and finally using polar coordinates:

β
e 2 ω(k)
Z 
2
III = dω4πω (G(k) ⊗ )δ(Lpp (z) − ω(k)) βω(k) ( ⊗ G(k)) + (6.70)

R+ e −1 Hpp
β
e 2 ω(k)
Z 
2 ∗ ∗
+ dω4πω (G (k) ⊗ )δ(Lpp (z) + ω(k)) βω(k) ( ⊗ G (k)) (6.71)

R + e − 1 Hpp

The last term of the operator Σ(2) is Eq. (6.36):


IV = hΩc |r(I)δ(Lc + Lpp (z))l(I)Ωc i =

Hpp
h i
= hΩc | ( ⊗ T GT ) ⊗ Jc l[a∗ ]Jc + ( ⊗ T G∗ T ) ⊗ Jc l[a]Jc ×
h i 
×δ(Lc + Lpp (z)) (G ⊗ ) ⊗ l[a∗ ] + (G∗ ⊗ ) ⊗ l[a] Ωc i = (6.72)

Hpp

From Eq. (6.40) and Eq. (6.41) we see that the only surviving terms are the ones that contain
both creation operator or both annihilation operators. We also use hΩc |Jc l[a∗ (k)]Jc =
β β β
hΩc |l[a(k)]e− 2 Lc = hΩc |e− 2 (Lc +ω(k)) l[a(k)] = hΩc |e− 2 ω(k) l[a(k)] and similar equations for
the annihilation operator:
Z h i
β
IV = dkdl ( ⊗ G(k))hΩc |e − 2 ω(k) δ(Lc + Lpp (z) + ω(k))l[a(k)]l[a∗ (l)]Ωc i(G(l) ⊗ ) +
Hpp
Z (6.73)
h i
∗ β
ω(k) ∗ ∗
+ dkdl ( ⊗ G (k))hΩc |e 2 δ(Lc + Lpp (z) − ω(k))l[a (k)]l[a(l)]Ωc i(G (l) ⊗ )
Hpp
(6.74)
using now Eq. (2.62) and Eq. (2.63)

e βω(k)
Z h i
β
IV = dkdl ( ⊗G(k))δ(Lpp (z)+ω(k))e − 2 ω(k) δ(k −l)(G(l)⊗ ) + (6.75)
e βω(k) − 1 Hpp

e βω(k)
Z h i
∗ β
∗
+ dkdl ( ⊗ G (k))δ(Lpp (z) − ω(k))e 2 ω(k) δ(k − l)Ω c i(G (l) ⊗ ) (6.76)
e βω(k) − 1 Hpp

and finally expressing it in polar coordinates:

β
e 2 ω(k)
Z 
2
IV = dω4πω ( ⊗ G(k))δ(Lpp (z) + ω(k)) βω(k) (G(k) ⊗ ) + (6.77)

R+ e −1 Hpp
β
e 2 ω(k)
Z 
∗
+ 2
dω4πω ( ⊗ G (k))δ(Lpp (z) − ω(k)) βω(k) (G∗ (k) ⊗ ) (6.78)

R + e − 1 Hpp

So far we have considered projections onto a general Hpp atomic Hilbert space. Now we
will restrict our atomic projectors to smaller spaces in the two next sections. In our model
Hpp is finite dimensional as explained in chapter 1.
 95

6.3 Application to the Eigenvalue Ei − Ej 6= 0 of the Atomic

Liouvillian

We will analyze in this section the stability of an eigenvalue Ei − Ej of Lpp with i 6= j. We

will suppose that Ei − Ej is non-degenerated.
We will calculate the four terms I, II, III, IV and then we will see that our level-shift operator
is positive, which implies the instability of these eigenvalues (for i, j = 1 . . . N i 6= j).
We begin with I:
We have to project Eq. (6.46) and Eq. (6.47) onto {ϕi ⊗ ϕj } for i, j fixed (the space of
eigenvectors of Ei − Ej is one-dimensional because of the non-degeneracy of the eigenvalues
different from zero and we have:

hϕi ⊗ ϕj |Iϕi ⊗ ϕj i = (6.79)

XZ
= dkhϕi ⊗ ϕj |(G(k) ⊗ )δ(Lpp (z) − ω(k))ϕm ⊗ ϕp i ×
m,p
1
×hϕm ⊗ ϕp |(G∗ (l) ⊗ )ϕi ⊗ ϕj i + (6.80)
e βω(k) −1
XZ
+ dkhϕi ⊗ ϕj |(G∗ (k) ⊗ )δ(Lpp (z) + ω(k))ϕm ⊗ ϕp i ×
m,p

e βω(k)
×hϕm ⊗ ϕp |(G(l) ⊗ )ϕi ⊗ ϕj i (6.81)
e βω(k) − 1
now we use that (Hpp ⊗ − ⊗ Hpp )|ϕm ⊗ ϕp i = Em |ϕm i ⊗ |ϕp i − |ϕm i ⊗ Ep |ϕp i =
(Em − Ep )|ϕm ⊗ ϕp i, and therefore f (Lpp )|ϕm ⊗ ϕp i = f (Em − Ep )|ϕm ⊗ ϕp i writing the
matrix elements of G(k), Gij (k) = hϕi |G(k)ϕj i, we use also that Lpp (z) := Lpp − z with
z = Ei − Ej :
XZ 1
hϕi ⊗ϕj |Iϕi ⊗ϕj i = dkGim (k)⊗δjp δ(Em −Ep −Ei +Ej −ω(k)) G∗mi (k)⊗δpj +
m,p
e βω(k) −1
(6.82)
XZ e βω(k)
+ dkG∗im (k) ⊗ δjp δ(Em − Ep − Ei + Ej + ω(k)) βω(k) Gmi (k) ⊗ δpj (6.83)
m,p
e −1
taking into account that (G∗ )mi (k) = Gim (k):
XZ 1
hϕi ⊗ ϕj |Iϕi ⊗ ϕj i = dk|Gim (k)|2 δ(Em − Ei − ω(k))) + (6.84)
m
e βω(k) −1
XZ e βω(k)
+ dk|Gmi (k)|2 δ(Em − Ei + ω(k))) (6.85)
m
e βω(k) − 1
 96

and finally expressing the result in polar coordinates and using the dispersion relation for
phonons, which will be used in the next section ω(k) = k 2 , we have:
X 1
hϕi ⊗ ϕj |Iϕi ⊗ ϕj i = |Gim (Em − Ei )|2 2π(Em − Ei )1/2 + (6.86)
m>i
e β(Em −Ei ) −1
X e β(Ei −Em )
+ |Gmi (Ei − Em )|2 2π(Ei − Em )1/2 (6.87)
m<i
e β(Ei −Em ) − 1

from ω(k) ≥ 0 and if at least one of the elements Gim (Em − Ei )|m>i or at least one of the
elements Gmi (Ei − Em )|m<i is different from zero, then:

hϕi ⊗ ϕj |Iϕi ⊗ ϕj i > 0 (6.88)

P(2)
Now we calculate the second term II of , from Eq. (6.59) and Eq. (6.60):

hϕi ⊗ ϕj |IIϕi ⊗ ϕj i = (6.89)

XZ
= dkhϕi ⊗ ϕj |( ⊗ G(k))δ(Lpp (z) + ω(k))ϕp ⊗ ϕm i ×
m,p
1
×hϕp ⊗ ϕm |( ⊗ G∗ (k))ϕi ⊗ ϕj i + (6.90)
e βω(k) − 1

XZ
+ dkhϕi ⊗ ϕj |( ⊗ G∗ (k))δ(Lpp (z) − ω(k))ϕp ⊗ ϕm i ×
m,p

e βω(k)
×hϕp ⊗ ϕm |( ⊗ G(k))ϕi ⊗ ϕj i (6.91)
e βω(k) − 1
taking into account f (Lpp )|ϕm ⊗ ϕp i = f (Em − Ep )|ϕm ⊗ ϕp i as seen before, hϕi ⊗ ϕj |( ⊗
G(k))ϕp ⊗ ϕm i = δip (G(k))jm and (G∗ (k))mj = G(k)jm :
XZ 1
hϕi ⊗ ϕj |IIϕi ⊗ ϕj i = dkGjm (k)δip δ(Ep − Em − Ei + Ej + ω(k)) Gjm (k)δpi +
m,p
e βω(k) −1

XZ e βω(k)
+ dkGmj (k)δip δ(Ep − Em − Ei + Ej − ω(k)) Gmj (k)δpi (6.92)
m,p
e βω(k) − 1
finally using polar coordinates it becomes:
X 1
hϕi ⊗ ϕj |IIϕi ⊗ ϕj i = |Gjm (Em − Ej )|2 2π(Em − Ej )1/2 + (6.93)
m>j
e β(Em −Ej ) −1
X e β(Ej −Em )
+ |Gmj (Ej − Em )|2 2π(Ej − Em )1/2 (6.94)
m<j
e β(Ej −Em ) − 1
 97

from ω(k) ≥ 0 and if at least one of the elements Gjm (Em − Ej )|m>j or at least one of the
elements Gmj (Ej − Em )|m<j is different from zero, then:

hϕi ⊗ ϕj |IIϕi ⊗ ϕj i > 0 (6.95)

If we assume Gii = 0, ∀i then we see that:

hϕi ⊗ ϕj |IIIϕi ⊗ ϕj i = hϕi ⊗ ϕj |IV ϕi ⊗ ϕj i = 0 (6.96)

We will show it for example for III, from Eq. (6.68) and Eq. (6.69):

hϕi ⊗ ϕj |IIIϕi ⊗ ϕj i = (6.97)

XZ
= dkhϕi ⊗ ϕj |(G(k) ⊗ )δ(Lpp (z) − ω(k))ϕm ⊗ ϕp i ×
m,p
β
ω(k)
e 2
×hϕm ⊗ ϕp |( ⊗ G(k))ϕi ⊗ ϕj i + (6.98)
e βω(k) − 1
XZ
+ dkhϕi ⊗ ϕj |(G∗ (k) ⊗ ) × δ(Lpp (z) + ω(k))ϕm ⊗ ϕp i ×
m,p
β
ω(k)
∗ e 2
hϕm ⊗ ϕp |( ⊗ G (k))ϕj ⊗ ϕj i = (6.99)
e βω(k) − 1
taking into account the same considerations as for I and II:
β
XZ ω(k)
e 2
= dkGim (k)δjp δ(Ei − Ep − ω(k)) Gpj (k)δmi + (6.100)
m,p
e βω(k) − 1
β
XZ ω(k)
e 2
+ dkGmi (k)δjp δ(Ei − Ep + ω(k)) Gjp (k)δmj = 0 (6.101)
m,p
e βω(k) − 1

It vanishes because of the fact that all the matrix elements Gij in this expression are diagonal
(i = j).
Taking all the results together we have:

(Σ(2) )ij > 0 i 6= j (6.102)

and that means instability of the eigenvalue Ei − Ej . That is to say when we add the
interaction to our atomic Liouvillian, the non-degenerated real eigenvalues of Lpp different
from zero, move to the lower half plane. They become complex numbers with a negative
imaginary part which implies instability.
 98

6.4 Application to the Eigenvalue 0 of the Atomic Liouvillian

Let us consider now the eigenvalue 0 of Lpp and the N-dimensional Hilbert space Hpp =
−1
{ϕi ⊗ ϕi }N
i=0
In this case we have the matrix elements (see Eq. (6.46) and Eq. (6.47)):

Iij = hϕi ⊗ ϕi |Iϕj ⊗ ϕj i = (6.103)

XZ
= dkhϕi ⊗ ϕi |(G(k) ⊗ )δ(Lpp (z) − ω(k))ϕm ⊗ ϕp i ×
m,p
1
×hϕm ⊗ ϕp |(G∗ (l) ⊗ )ϕj ⊗ ϕj i + (6.104)
e βω(k) −1
XZ
+ dkhϕi ⊗ ϕi |(G∗ (k) ⊗ )δ(Lpp (z) + ω(k))ϕm ⊗ ϕp i ×
m,p

e βω(k)
×hϕm ⊗ ϕp |(G(l) ⊗ )ϕj ⊗ ϕj i (6.105)
e βω(k) − 1
taking into account f (Lpp )|ϕm ⊗ ϕp i = f (Em − Ep )|ϕm ⊗ ϕp i and hϕi ⊗ ϕi |(G∗ (k) ⊗ )ϕm ⊗
ϕp i = (G∗ (k))im δip :
XZ 1
= dkGim (k)δip δ(Em − Ej − ω(k)) Gjm (k)δpj + (6.106)
m,p
e βω(k) −1
XZ e βω(k)
+ dkGmi (k)δip δ(Em − Ej − ω(k)) Gmj (k)δpj (6.107)
m,p
e βω(k) − 1

now we use polar coordinates and integrate:

X 1
= δij Gim (Em − Ej ) β(E −E ) Gjm (Em − Ej )2π(Em − Ej )1/2 + (6.108)
e m j − 1
m
X e β(Ej −Em )
+δij Gmi (Ej − Em ) β(E −E ) Gmj (Ej − Em )2π(Ej − Em )1/2 = (6.109)
m
e j m −1
and that becomes:
X 1
= |Gim (Em − Ei )|2 2π(Em − Ei )1/2 + (6.110)
m>i
e β(Em −Ei ) −1
X e β(Ei −Em )
+ |Gmi (Ei − Em )|2 2π(Ei − Em )1/2 (6.111)
m<i
e β(Ei −Em ) − 1

and our result is:


 Eq. (6.110) + Eq. (6.111) : if i = j
Iij = (6.112)
 0 : otherwise
 99

In a similar way we calculate IIij (from Eq. (6.59) and Eq. (6.60)):

IIij = hϕi ⊗ ϕi |IIϕj ⊗ ϕj i = (6.113)

XZ
= dkhϕi ⊗ ϕi |( ⊗ G(k))δ(Lpp (z) + ω(k))ϕp ⊗ ϕm i ×
m,p
1
×hϕp ⊗ ϕm |( ⊗ G∗ (k))ϕj ⊗ ϕj i + (6.114)
e βω(k) − 1

XZ
+ dkhϕi ⊗ ϕi |( ⊗ G∗ (k))δ(Lpp (z) − ω(k))ϕp ⊗ ϕm i ×
m,p

e βω(k)
×hϕp ⊗ ϕm |( ⊗ G(k))ϕj ⊗ ϕj i (6.115)
e βω(k) − 1
taking into account f (Lpp )|ϕm ⊗ϕp i = f (Em −Ep )|ϕm ⊗ϕp i, hϕi ⊗ϕi |( ⊗G(k))ϕm ⊗ϕp i =
δim (G(k))ip and similar relations:
XZ 1
IIij = dkGim (k)δip δ(Ei − Em + ω(k)) Gjm (k)δpj + (6.116)
m,p
e βω(k) −1
XZ e βω(k)
+ dkGmi (k)δip δ(Ei − Em − ω(k)) Gmj (k)δpj (6.117)
m,p
e βω(k) − 1

taking polar coordinates and integrating:

X 1
IIij = δij Gim (Em − Ei ) Gjm (Em − Ei )2π(Em − Ei )1/2 + (6.118)
m
e β(Em −Ei ) −1
X e β(Ei −Em )
+δij Gmi (Ei − Em ) Gmj (Ei − Em )2π(Ei − Em )1/2 (6.119)
m
e β(Ei −Em ) − 1

our result is:

X 1
IIij = |Gim (Em − Ei )|2 2π(Em − Ei )1/2 + (6.120)
m>i
e β(Em −Ei ) −1
X e β(Ei −Em )
+ |Gmi (Ei − Em )|2 2π(Ei − Em )1/2 (6.121)
m<i
e β(Ei −Em ) − 1

i.e.:

 Eq. (6.120) + Eq. (6.121) : if i = j
IIij = (6.122)
 0 : otherwise
 100

We note that Eq. (6.122)=Eq. (6.112).

In the same way we calculate IIIij from Eq. (6.68) and Eq. (6.69):

IIIij = hϕi ⊗ ϕi |IIIϕj ⊗ ϕj i = (6.123)

XZ
= dkhϕi ⊗ ϕi |(G(k) ⊗ )δ(Lpp (z) − ω(k))ϕm ⊗ ϕp i ×
m,p
β
ω(k)
e 2
×hϕm ⊗ ϕp |( ⊗ G(k))ϕj ⊗ ϕj i + (6.124)
e βω(k) − 1
XZ
+ dkhϕi ⊗ ϕi |(G∗ (k) ⊗ ) × δ(Lpp (z) + ω(k))ϕm ⊗ ϕp i ×
m,p
β
ω(k)
∗ e 2
hϕm ⊗ ϕp |( ⊗ G (k))ϕj ⊗ ϕj i (6.125)
e βω(k) − 1
taking into account f (Lpp )|ϕm ⊗ϕp i = f (Em −Ep )|ϕm ⊗ϕp i, hϕi ⊗ϕi |(G(k)⊗ )ϕm ⊗ϕp i =
(G(k))im δip and similar relations:
β
XZ ω(k)
e 2
IIIij == dkGim (k)δip δ(Em − Ep − ω(k)) Gpj (k)δmj + (6.126)
m,p
e βω(k) − 1
β
XZ ω(k)
e 2 ∗
+ dkG∗im (k)δip δ(Em − Ep + ω(k)) Gpj (k)δmj (6.127)
m,p
e βω(k) − 1

After some simplifications with the δ of Kronecker:

β
Z ω(k)
e 2
IIIij = dk|Gij (k)|2 δ(Ej − Ei − ω(k)) + (6.128)
e βω(k) − 1
β
Z ω(k)
2 e 2
+ dk|Gji (k)| δ(Ej − Ei + ω(k)) (6.129)
e βω(k) − 1
taking polar coordinates and integrating:
 β
 e 2 (Ej −Ei )
 |Gij (Ej − Ei )|2 β(E −E ) 2π(Ej − Ei )1/2

 : if j>i
IIIij = e β j i −1 (6.130)
(Ei −Ej )
 e 2
 |Gji (Ei − Ej )|2 β(E −E ) 2π(Ei − Ej )1/2 : if j<i


e j i − 1
 101

and finally IV:

IVij = hϕi ⊗ ϕi |IV ϕj ⊗ ϕj i = (6.131)

XZ
= dkhϕi ⊗ ϕi |( ⊗ G(k))δ(Lpp (z) + ω(k))ϕm ⊗ ϕp i ×
m,p
β
ω(k)
e 2
×hϕm ⊗ ϕp |(G(k) ⊗ )ϕj ⊗ ϕj i + (6.132)
e βω(k) − 1
XZ ∗
+ dkhϕi ⊗ ϕi |( ⊗ G (k))δ(Lpp (z) − ω(k))ϕm ⊗ ϕp i ×
m,p
β
ω(k)
∗ e 2
×hϕm ⊗ ϕp |(G (k) ⊗ )ϕj ⊗ ϕj i (6.133)
e βω(k) − 1
taking into account f (Lpp )|ϕm ⊗ϕp i = f (Em −Ep )|ϕm ⊗ϕp i, hϕi ⊗ϕi |( ⊗G(k))ϕm ⊗ϕp i =
δim (G(k))ip and similar relations:
β
XZ ω(k)
e 2
IVij = dkGip (k)δim δ(Em − Ep + ω(k)) Gmj (k)δpj + (6.134)
m,p
e βω(k) − 1
β
XZ ω(k)
∗ e 2
+ dkGip (k)δim δ(Em − Ep − ω(k)) G∗mj (k)δpj (6.135)
m,p
e βω(k) − 1

simplifying:
β
Z ω(k)
e 2
IVij = dk|Gij (k)|2 δ(Ei − Ej − ω(k)) + (6.136)
e βω(k) − 1
β
Z ω(k)
2 e 2
+ dk|Gji (k)| δ(Ei − Ej + ω(k)) (6.137)
e βω(k) − 1
taking polar coordinates and integrating:
 β

2 e 2 (Ej −Ei )
 |Gij (Ej − Ei )| β(E −E ) 2π(Ej − Ei )1/2

 : if j>i
e j i − 1
IVij = β
(Ei −Ej )
(6.138)
 |Gji (Ei − Ej )|2 e 2

2π(Ei − Ej )1/2 : if j<i

e β(Ej −Ei ) − 1


We introduce a set of functions and another notation for the projection of the level shift
operator:

e β(Em −Ei )
φim (Em − Ei ) := 2|Gim (Em − Ei )|2 2π(Em − Ei )1/2 (6.139)
e β(Em −Ei ) − 1 
M := Σ(2)  (6.140)

N −1
(ϕi ⊗ϕi )i=0
 102

We can write the matrix elements of M from Eq. (6.112), Eq. (6.122), Eq. (6.130), Eq. (6.138):

Mii = Iii + IIii = 2Iii =

X 1
=2 |Gim (Em − Ei )|2 2π(Em − Ei )1/2 + (6.141)
m>i
e β(Em −Ei ) − 1
X e β(Ei −Em )
+2 |Gim (Ei − Em )|2β(Ei −Em ) − 1
2π(Ei − Em )1/2 = (6.142)
m<i
e
X X
= e β(Ei −Em ) φim (Em − Ei ) + φmi (Ei − Em ) > 0 (6.143)
m>i m<i

Mij = −IIIij − IVij = −2IIIij =

β
(Ej −Ei )
i<j e 2 β
= −2|Gij (Ej −Ei )| 2
2π(Ej −Ei )1/2 = −e − 2 (Ej −Ei ) φij (Ej −Ei ) ≤ 0 (6.144)
e β(Ej −Ei ) −1
β
(Ei −Ej )
i>j e 2 β
= −2|Gji (Ei −Ej )|2 2π(Ei −Ej )1/2 = −e − 2 (Ei −Ej ) φij (Ei −Ej ) ≤ 0 (6.145)
e β(Ei −Ej ) −1

 M >0 : if i=j
ii
i.e Mij = (6.146)
 Mij ≤ 0 : otherwise
Let us see now an easy example:

N = 2, E1 , E2
M11 = e −β(E2 −E1 ) φ12 (E2 − E1 ) + 0
M22 = 0 + φ12 (E2 − E1 )
β
(E2 −E1 )
e 2
M12 = −2|G12 (E2 − E1 )|2 2π(E2 − E1 )1/2
e β(E2 −E1 ) − 1

β
(E2 −E1 )
2 e 2
M21 = −2|G12 (E2 − E1 )| 2π(E2 − E1 )1/2
e β(E2 −E1 ) − 1
M is symmetric.

Definition 6.4.1. A bounded operator G on L2 is called positivity preserving if Gf is

positive whenever f ∈ L2 (x, dµ) is positive.

Definition 6.4.2. A bounded operator G on L2 is called ergodic if and only if it is positivity

preserving and for any u, v ∈ L2 that are both positive there is some n > 0 with hu|Gn vi =
6 0.
 103

P(2) 
Theorem 6.4.3. Let be M = −1
with Hpp ϕi = Ei ϕi i = 0 . . . N − 1, then:
(ϕi ⊗ϕi )N

i=0
a) M has an eigenvector k corresponding to the eigenvalue 0 and
P −1 − β Ei
k := Ni=0 e
2 ϕi ⊗ ϕi
b) (Perron-Frobenius) There is a γ ∈ R+ positive number that satisfies the condition M +
γ > 0 i.e. hϕi |M + γ ϕi i > 0 ∀i = 0 . . . N − 1. Assume G := (M + γ )−1 is ergodic then
0 is a simple eigenvalue of M and k is unique.

Proof. a)
X X
(M k)i = Mij kj = Mii ki + Mij kj =
j j6=i
X β X β
= e −β(Em −Ei ) φim (Em − Ei )e − 2 Ei + φmi (Ei − Em )e − 2 Ei +
m>i m<i
− β2 Ej β
X X
+ (−1)e −β(Ej −Ei ) φij (Ej − Ei )e + e −β(Ei −Ej ) φji (Ei − Ej )e − 2 Ej = 0
j>i j<i

b) We have seen in Eq. (6.4) that Mii > 0 and Mij ≤ 0 f or i 6= j. We can write M =
M d + M od with:

 M
ii : if i=j
(M d )ij = (6.147)
 6 j
0 : if i =

 0 : if i = j
(M od )ij = (6.148)
 Mij : if i 6= j

It is easy to see that it exists a γ so that M +γ > 0. It is enough to take γ = |mini6=j {Mij }|.
From the definition G := (M +γ )−1 , Mij is symmetric consequently G is also symmetric:

G = (M d + γ + M od )−1 = (M d + γ )−1 ( + (M + γ )−1 M od )−1 = (6.149)

∞
X n
= (M d + γ )−1 [(−M od )(M d + γ )−1 ] and : (6.150)
n=0

∞
X
Gij = (M d + γ )−1
ik1 (−M
od
) (M d + γ )−1 od d −1
k2 k3 . . . (−M )k2n k2n+1 (M + γ )k2n+1 j
| {z k1 k}2 |
n=0
| {z } {z } | {z } | {z }
>0 ≥0 >0 ≥0 >0
(6.151)
that is Gij ≥ 0 ∀i, j.
G is symmetric, therefore we can diagonalize it and then kGk ∈ σ(G) in fact kGk =
 104

maxi |gi | = maxi gi where gi ∈ σ(G).

Let be ψ (0) an eigenvector corresponding to the eigenvalue kGk then taking into account
that:
X X
|hψGψi| = | ψi Gij ψj | ≤ |ψi |Gij |ψj | = h|ψ|G|ψ|i we have
kGk = hψ (0) Gψ (0) i ≤ h|ψ (0) |G|ψ (0) |i ≤ kGk ⇒ G|ψ (0) | = kGk|ψ (0) | (6.152)

(0) (0) (0) (0)

Assume now that ∃ ψ1 , ψ2 eigenvectors to the eigenvalue kGk. |ψ1 |, |ψ2 | linear inde-
(0) (0)
pendent, Gn0 symmetric and h|ψ1 |, |ψ2 |i = 0

(0) (0) (0) (0)

0 = kGkn0 h|ψ1 |, |ψ2 |i = h|ψ1 |, Gn0 |ψ2 |i > 0 (6.153)

(0)
which is a contradiction, that is to say there is a unique ψ1 and

kGk = max |eigenvalues of G| = min |eigenvalues of M + γ | (6.154)

and we already know from part a) that 0 is eigenvalue of M and therefore it is simple.

6.5 Magnons in a Ferromagnet

We will begin now with an application of the last sections. Our Hamiltonian is :
X X
H = HS + HW = −J Sbx · Sbx+y − j(x)~s0 · Sbx (6.155)
xy x

where Sx and Sy are atomic-spins operators, with J > 0 and j(x) ≥ 0 which means that the
configuration of minimum energy corresponds to all the spins oriented in the same direction,
which we will take as z-axis. The first term of Eq. (6.155) represents the interaction of
the nearest neighbor between spins of the ferromagnet, y indicates precisely that the sum
runs over the nearest neighbors of each site x. The second term represents the interaction
of the external spin wave with the ferromagnet, in which j(x) is the amplitude of this
wave and s0 is an unit vector that shows the direction of the spin wave. At first we
make a Holstein-Primakoff transformation. We know that the spin-operators satisfy the
commutation relations:

[Sbα , Sbβ ] = iαβγ Sbγ , α, β = 1, 2, 3 [Sb+ , Sb− ] = 2Sb3 (6.156)

 105

where α, β, γ = 1, 2, 3 the later being the three directions of the space and αβγ represents
the antisymmetric tensor. We recall the relations:

Ŝx+ = Ŝx1 + iŜx2 , Ŝx− = Ŝx1 − iŜx2 (6.157)

On the other hand we know that a spin wave in a spin lattice is a kind of boson called
magnon, the creation and annihilation operators of which obey commutation relations
[a, a∗ ] = 1. The relation between these last operators and the spin operators is the Holstein-
Primakoff transformation (see [3]):
a∗x ax 1/2
Ŝx+ = Ŝx1 + iŜx2 = (2S)1/2 (1 − ) ax (6.158)
2S
a∗ ax
Ŝx− = Ŝx1 − iŜx2 = (2S)1/2 a∗x (1 − x )1/2 (6.159)
2S
Ŝx3 = S − a∗x ax (6.160)

where S is the spin of our system (typically 1/2) and where the square roots are to be
a∗x ax
understood purely formally as infinite series in powers of 2S . It is only in the subspace
a∗x ax
of the eigenvectors of the operator 2S which belongs to those eigenvalues of this operator
which are smaller or equal to unity, where Eq. (6.158), Eq. (6.159) and Eq. (6.160) can
be interpreted as the atomic-spin operators. Since it is there where the operators Ŝx+ , Ŝx−
are hermitian conjugates while the operator S 3 is self-adjoint. Our system is periodic and
we had better characterize the spin waves by a definite momentum, therefore we use the
Fourier transformed variables of ax , a∗x :
1 X ~
ax = e−ik·~x bk (6.161)
N 1/2 k
1 X ~
a∗x = eik·~x b∗k (6.162)
N 1/2 k

where N is the number of unit cells in the body and the wave vector k lies in the first
Brillouin zone. It is easy to verify that the operators bk , b∗k satisfy the same commutation
relations like ax and a∗x :

[bk , b∗k0 ] = δkk0 [bk , b0k ] = [b∗k , b∗k0 ] = 0 (6.163)

The operators bk and b∗k create and destroy a magnon or spin-wave excitation of the fer-
romagnet. These turn out to be excitations where the spins locally deviate only a small
 106

amount from their ground state values as the spin wave passes by. See figure 16:

6 6 6 6 6 6 6 6 6

  6 MBB KA
6 
 B A MB 6

 B A B

 B A B

Fig.21 Top: Ordered ground state. Bottom: Spin deviations in a magnon

Therefore it seems plausible taking into account that S 3 = S − a∗x ax considering a∗x ax as a
small perturbation of S in a series expansion of Eq. (6.158), Eq. (6.159):

a∗x ax 1/2 Sa∗x ax

Ŝx+ = (2S)1/2 (1 − ) ax ' (2S)1/2 ax (6.164)
2S
a∗ ax Sa∗x ax
Ŝx− = (2S)1/2 a∗x (1 − x )1/2 ' (2S)1/2 a∗x (6.165)
2S

From Ŝx+ = Ŝx1 + iŜx2 , Ŝx− = Ŝx1 − iŜx2 we see that:

−
Ŝx+ Ŝx+y = Ŝx1 Ŝx+y
1
+ Ŝx2 Ŝx+y
2
+ iŜx2 Ŝx+y
1 1 `
− iŜ` 2` `
x Ŝx+y
` (6.166)
+
Ŝx− = Ŝx+y
1
Ŝx1 + Ŝx+y
2
Ŝx2 − iŜx+y
1
Ŝx2 + Ŝx+y
2```
Ŝx1`
`
Ŝx+y (6.167)

That is to say:
1 −
Ŝx1 Ŝx+y
1
+ Ŝx2 Ŝx+y
2
= (Ŝx+ Ŝx+y +
+ Ŝx+y Ŝx− ) (6.168)
2
1 −
Ŝx · Ŝx+y = (Ŝx+ Ŝx+y +
+ Ŝx+y Ŝx− ) + Ŝx3 Ŝx+y
3
(6.169)
2
Taking into account this result, we will proceed to arrange our Hamiltonian Eq. (6.155) to
the typical form :
X X
H = −J Ŝx · Ŝx+y − j(x)~s0 · Ŝx =
xy x
X X
−J (Ŝx1 Ŝx+y
1
+ Ŝx2 Ŝx+y
2
+ Ŝx3 Ŝx+y
3
)− j(x)(s10 Ŝx1 + s20 Ŝx2 + s30 Ŝx3 )
xy x
 107

We note now the result of Eq. (6.169):

X 1 X 1X
− − +
H = −J (Ŝx+ Ŝx+y +
+ Ŝx+y Ŝx− ) + Ŝx3 Ŝx+y
3
− j(x) s30 M − j(x)(s+ −
0 Ŝx + s0 Ŝx )
xy
2 2
| x {z } x
j0

Taking into account now the Holstein-Primakoff transformation for S  a∗x ax , Eq. (6.164),
Eq. (6.165) and Eq. (6.160):
X n1 o
(2S)1/2 ax (2S)1/2 a∗x+y + (2S)1/2 ax+y (2S)1/2 a∗x + (S − a∗x ax )(S − a∗x+y ax+y ) −

H = −J
xy
2

1X
−j0 s30 M − j(x)(s+
0 (2S) ax + s−
1/2 ∗
0 (2S)
1/2
ax ) (6.170)
2 x
and simplifying:
Xn o
H = −J S 2 − Sa∗x ax − Sa∗x+y ax+y + Sax a∗x+y + Sax+y a∗x + a∗x ax a∗x+y ax+y −
xy
r
SX ∗ −
−j0 s30 M − j(x)(s+
0 ax + s0 ax ). (6.171)
2 x

We can neglect the last term of the first line in comparison with the others because of
S  a∗x ax , and taking into account that Sax a∗x+y = Sa∗x+y ax and Sax+y a∗x = Sa∗x ax+y :
Xn o
H = −J S 2 − Sa∗x ax − Sa∗x+y ax+y + Sa∗x+y ax + Sa∗x ax+y −
xy
r
SX ∗ −
−j0 s30 M − j(x)(s+
0 ax + s0 ax ). (6.172)
2 x

The last step in the way of obtaining our desired Hamiltonian expression, is to change the
creation and annihilation operators following Eq. (6.161) and Eq. (6.162):
Xn 1 X i~k·~x ∗ X −i~k0 ·~x 1 X i~k·(~x+~y) ∗ X −i~k0 ·(~x+~y)
H = −J S2 − S e bk e bk 0 − S e bk e bk 0 +
xy
N 0
N 0
k k k k
1 X i~k·(~x+~y) ∗ X −i~k0 ·~x 1 X i~k·~x ∗ X −i~k0 ·(~x+~y) o
+S e bk e bk 0 + S e bk e bk 0 −
N 0
N 0
k k k k

r
SX  1 X i~k·~x ∗ − 1
X ~
−ik·~

−j0 s30 M − j(x) s+
0 √ e b k + s0 √ e x
b k (6.173)
2 x N N
k k
 108

If n is the number of nearest neighbors and N the number of spin places, we have:

JN nS 2 1 X X X i(~k−~k0 )·~x ∗ 1 X X X i(~k−~k0 )·(~x+~y) ∗

H=− + JS e bk bk0 + JS e b k bk 0 −
2 N N
k k0 xy k k0 xy
| {z } | {z }
nN δkk0 nN δkk0
1 X X X i(~k−~k0 )·~x X i~k·~y ∗ 1 X X X i(~k−~k0 )·~x X −i~k0 ·~y ∗ o
−JS e e bk bk0 − JS e e bk b k 0 −
N N
k k0 x y k k0 x y
| {z } | {z }
N δkk0 N δkk0

r
S 1 XX ~ ∗ −i~k·~
x −

−j0 s30 M − √ j(x)eik·~x s+ b
0 k + j(x)e s b
0 k (6.174)
2 N x k

We define now the magnon dispersion function, which in this approximation depends only
on the positions of the nearest neighbor spins:

1 X i~k·~y
Υk := e (6.175)
n y

We assume that Υk = Υ−k , which is true for lattices with inversion symmetry. For example
for the simple cubic lattice (n = 6) in three dimensions with lattice constant a:

1 X i~k·~y 1 1
Υk := e = (eik1 a +e−ik1 a +eik2 a +e−iky a +eik3 a +e−ik3 a ) = (cos k1 a+cos k2 a+cos k3 a)
n y 6 3
(6.176)
which is evidently a even function of ~k.
We insert now this function in the expression of the Hamiltonian:

JN nS 2 X X X X o
H=− + nJS b∗k bk + nJS b∗k bk − nJS Υk b∗k bk − nJS Υ−k b∗k bk −
2
k k k k
r
S XX ~ ∗ −i~k·~
x −

−j0 s30 M − j(x)eik·~x s+ b
0 k + j(x)e s b
0 k (6.177)
2N x k

which can be simplified as:

JN nS 2 X
(1 − Υk )b∗k bk −

H=− + 2nJS
2
k
r
S XX ~ ∗ −i~k·~
x −

−j0 s30 M − j(x)eik·~x s+
0 bk + j(x)e s0 bk (6.178)
2N xk
 109

The most important magnons will be those with momenta close to the center of the Brioullin
zone, k ' 0 so we need to examine the small k-dispersion function. For a Bravais lattice,
like simple cubic, we obtain from Eq. (6.176):

1 ~k→0 1 (ka)2
Υk = (cos k1 a + cos k2 a + cos k3 a) ' 3 cos ka = 1 − (6.179)
3 3 2
and the expression of the Hamiltonian becomes:

JN nS 2 X  (ka)2
b∗k bk −

H=− + 2nJS
2 2
k
r
S XX ~ ∗ −i~k·~
x −

−j0 s30 M − j(x)eik·~x s+
0 bk + j(x)e s0 bk (6.180)
2N xk

That is to say:
r
JN nS 2 X S XX ~ −i~k·~
x −

H=− +na2 JS k 2 b∗k bk −j0 s30 M − j(x)eik·~x s+ ∗
0 bk +j(x)e s0 bk
2 2N x
k k
Z Z (6.181)
X V X 1
Doing the substitutions → 3 d3 k and → d3 x we have finally:
(2π) x
V
k
r
JN nS 2 na2 JSV
Z Z
2 ∗ 1 S 
∗ −

H=− −j0 s30 M + dkk b (k)b(k)− dk ĵ(k)s+
0 b (k)+ĵ(k)s0 b(k)
2 (2π)3 (2π)3 2N
(6.182)
where:
Z
~
ĵ(k) = dxj(x)eik·~x (6.183)
Z
~
ĵ(k) = dxj(x)e−ik·~x (6.184)

since j(x) ∈ R.
We can split our Hamiltonian into the following terms:

H = Hpp + Hc + I (6.185)
JN nS 2
Hpp = − − j0 M s30 (6.186)
2
na2 JSV
Z
Hc = dkk 2 b∗ (k)b(k) (6.187)
(2π)3
r Z
1 S 
+ ∗ −

I=− dk ĵ(k)s0 b (k) + ĵ(k)s 0 b(k) (6.188)
(2π)3 2N
 110

we recall:
     
0 1 0 −i 1 0
s10 =   s20 =   s30 =   (6.189)
1 0 i 0 0 −1
     
1 0 0 1 0 0
=  s+
0 =
  s−
0 =
  (6.190)
0 1 0 0 1 0

and then:  
2
− JN2nS − j0 M 0
Hpp =  2
 (6.191)
0 − JN2nS + j0 M
that is to say:

JN nS 2 JN nS 2
σ(Hpp ) = {−j0 M − , j0 M − } = {E0 , E1 } (6.192)
2 2

and our ”atomic” Liouvillian is Lpp = Hpp ⊗ − ⊗ Hpp , operator on C2 with spectrum:

σ(Lpp ) = {E0 − E1 , 0(twofold), E1 − E0 } = {−2j0 M, 0(twofold), 2j0 M } (6.193)

We recall our general form of the interaction Eq. (6.1):

Z
I= dk[G(k) ⊗ b∗ (k) + G∗ (k) ⊗ b(k)]

and comparing with Eq. (6.188):

r r
1 S 1 S
G(k) = − 3 ĵ(k)s+
0
∗
G (k) = − ĵ(k)s−
0 (6.194)
(2π) 2N (2π)3 2N

Eigenvalue 2j0 M
In this case i = 1, j = 0, considering Eq. (6.86) and Eq. (6.87):

e β(E1 −E0 )
hϕ1 ⊗ ϕ0 |Iϕ1 ⊗ ϕ0 i = 0 + |G01 (E1 − E0 )|2 2π(E1 − E0 )1/2 = (6.195)
e β(E1 −E0 ) − 1

1 1 β(2j0 M )
2 e
= 6 4N |ĵ(2j0 M )| β(2j0 M ) − 1
2π(2j0 M )1/2 > 0 (6.196)
(2π) e
considering now Eq. (6.93) and Eq. (6.94):

1
hϕ1 ⊗ ϕ0 |IIϕ1 ⊗ ϕ0 i = |G01 (E1 − E0 )|2 2π(E1 − E0 )1/2 + 0 = (6.197)
e β(E1 −E0 ) −1
 111

1 1 2 1
= 6 4N |ĵ(2j0 M )| β(2j
2π(2j0 M )1/2 > 0 (6.198)
e 0M ) − 1
(2π)
Both I and II are positive. This means that when we introduce the perturbation this
eigenvalue of Lpp becomes a complex eigenvalue of Lg with a negative imaginary part which
implies the instability of this new eigenvalue.
Eigenvalue −2j0 M
In this case i = 0, j = 1, considering Eq. (6.86) and Eq. (6.87):
1
hϕ0 ⊗ ϕ1 |Iϕ0 ⊗ ϕ1 i = |G01 (E1 − E0 )|2 2π(E1 − E0 )1/2 + 0 = (6.199)
e β(E1 −E0 ) −1
1 1 2 1
= 6 4N |ĵ(2j0 M )| β(2j0 M ) − 1
2π(2j0 M )1/2 > 0 (6.200)
(2π) e
considering now Eq. (6.93) and Eq. (6.94):

e β(E1 −E0 )
hϕ0 ⊗ ϕ1 |IIϕ0 ⊗ ϕ1 i = 0 + |G01 (E1 − E0 )|2 β(E −E )
2π(E1 − E0 )1/2 (6.201)
e 1 0 −1
1 1 β(2j0 M )
2 e
= |ĵ(2j0 M )| 2π(2j0 M )1/2 > 0 (6.202)
(2π)6 4N e β(2j0 M ) − 1
Both I and II are positive. As for the eigenvalue 2j0 M this means that when we introduce
the perturbation this eigenvalue of Lpp becomes a complex eigenvalue of Lg with a negative
imaginary part which implies the instability of this new eigenvalue.
Eigenvalue 0
Using Eq. (6.112):
 1
I00 = Eq. (6.110) + Eq. (6.111) = |G01 (E1 − E0 )|2 2π(E1 − E0 )1/2 + 0

i=j=0 eβ(E1 −E0 ) −1
(6.203)
 eβ(E1 −E0 )
I11 = Eq. (6.110) + Eq. (6.111) = 0 + |G01 (E1 − E0 )|2 2π(E1 − E0 )1/2

i=j=1 eβ(E1 −E0 ) − 1
(6.204)
I10 = I01 = 0 (6.205)

Using Eq. (6.122):

 1
II00 = Eq. (6.120) + Eq. (6.121) = |G01 (E1 − E0 )|2 2π(E1 − E0 )1/2 + 0

i=j=0 eβ(E1 −E0 ) −1
(6.206)
 eβ(E1 −E0 )
II11 = Eq. (6.120) + Eq. (6.121) = 0 + |G01 (E1 − E0 )|2 2π(E1 − E0 )1/2

i=j=1 eβ(E1 −E0 ) − 1
(6.207)
 112

II10 = II01 = 0 (6.208)

Using Eq. (6.130):

 eβ/2(E1 −E0 )
III01 = Eq. (6.130) i=0 = |G01 (E1 − E0 )|2 β(E −E ) 2π(E1 − E0 )1/2 (6.209)

e 1 0 − 1
j=1
 eβ/2(E1 −E0 )
III10 = Eq. (6.130) i=1 = |G01 (E1 − E0 )|2 β(E −E ) 2π(E1 − E0 )1/2 (6.210)

j=0 e 1 0 −1
III00 = III11 = 0 (6.211)

Using Eq. (6.138):

 eβ/2(E1 −E0 )
IV01 = Eq. (6.138) i=0 = |G01 (E1 − E0 )|2 β(E −E ) 2π(E1 − E0 )1/2 (6.212)

e 1 0 − 1
j=1
 eβ/2(E1 −E0 )
IV10 = Eq. (6.138) i=1 = |G01 (E1 − E0 )|2 β(E −E ) 2π(E1 − E0 )1/2 (6.213)

e 1 0 − 1
j=0

IV00 = IV11 = 0 (6.214)

and then we have the imaginary part of the level-shift operator Eq. (6.31),
ImΓ(2) := −Im(FP (Lg (θ, z))(2) ) = πP W δ(L0 (z))W P :
 
I00 + II00 −III01 − IV01
ImΓ(2) = π  = (6.215)
−III10 − IV10 I11 + II11

 
4π(E − E )1/2 eβ/2(E1 −E0 )
1 0
2
 2|G01 (E1 − E0 )| β(E1 −E0 ) −2|G01 (E1 − E0 )|2 β(E −E ) 4π(E1 − E0 )1/2 
 e −1 e 1 0 −1 

2 eβ/2(E1 −E0 ) 1/2 2 eβ(E1 −E0 ) 1/2

−2|G01 (E1 − E0 )| β(E −E ) 4π(E1 − E0 ) 2|G01 (E1 − E0 )| β(E −E ) 4π(E1 − E0 )
e 1 0 −1 e 1 0 −1

 
β/2(E1 −E0 )
2|G01 (E1 − E0 )|2 1 −e
= 4π(E1 − E0 )1/2   (6.216)
eβ(E1 −E0 ) − 1 −eβ/2(E1 −E0 ) eβ(E1 −E0 )

writing α = eβ/2(E1 −E0 ) and α2 = eβ(E1 −E0 ) we can calculate the eigenvalues of this last
matrix as:
 
 1 − λ −α
 
 = 0 ⇒ (1 − λ)(α2 − λ) − α2 = 0

 (6.217)
 −α α2 − λ
 

α2 − λ − λα2 + λ2 − α2 = 0 ⇒ λ(λ − (1 + α2 )) = 0 (6.218)
 113

with solutions λ1 = 0 and λ2 = 1 + α2 and the time of decay of these excitations is:
 2|G (E − E )|2
01 1 0 1/2
−1 1 eβ(E1 −E0 ) − 1 1
τ= −E
4π(E 1 − E 0 ) = =
β(E
e 1 0 −1) λ2 8π|G01 (E1 − E0 )| (E1 − E0 ) 1 + α2
2 1/2

eβ(E1 −E0 ) − 1 1
= (6.219)
8π|G01 (E1 − E0 )| (E1 − E0 ) 1 + e 1 −E0 )
2 1/2 β(E

substituting now the values of ε0 , ε1 and G01 (E1 − E0 ) we obtain finally:

e2βj0 M − 1 1
τ= 2 2βj0 M
(6.220)
1 + e
 q
1 S
8π − (2π)3 2N |ĵ(E1 − E0 )|2 (2j0 M )1/2
√
2( 2π)5 N e2βj0 M − 1
τ= (6.221)
S|ĵ(2j0 M )|2 (j0 M )1/2 1 + e2βj0 M
We could also write the KMS states for free dynamic and for the coupled dynamic on the
Hilbert space Hpp ⊗ Hpp ⊗ Hcβ ⊗ Hcβ . For the free dynamic, see Eq. (3.90):
1
−1/2
Ω̂β0 = Zβ,0
X
e−βEj /2 ϕj ⊗ ϕj ⊗ Ωc ⊗ Ωf (6.222)
j=0

where E0 and E1 are given in Eq. (6.192) and ϕ0 and ϕ1 are the corresponding eigenvectors.
With interaction we have Eq. (3.91):

−1/2 −1/2
Ω̂βg := Zβ,0 e−βLg,l /2 Ω̂β0 = Zβ,0 eβLg,r /2 Ω̂β0 (6.223)

where:

Lg,l = Lpp + Lc + I0 l[I]I0−1 (6.224)

Lg,r = Lpp + Lc − I0 r[I]I0−1 (6.225)

Lpp = Hpp ⊗ − ⊗ Hpp (6.226)

Z
Lc = dkk 2 [a∗l (k)al (k) − a∗r (k)ar (k)] (6.227)

where now the dispersion relation is:

ω(k) = k 2 (6.228)
Appendix A

Algebraic Techniques

This appendix is an adaptation of some results of the appendix of [14] for the case T > 0.
We will use the following notation: suppose that σj ∈ {+1, −1} and that τj ∈ {l, r}
a+ = τ [a∗ ], a−1 = τ [a], where a∗ , a are the creation and annihilation operators and
l, r are the left and right representation given by the GNS construction. We denote with
αj = (σj , τj ), so that aαj = a(σj ,τj ) = τj [aσj ]. We define:

 0 : for τ = l
π(τ ) = (A.1)
 1 : for τ = r

Lemma A.0.1. (Pull-through Formula): Let f : R+ → C be a measurable function,

obeying f [r] = O(r). Then f [Lf ] is defined on D(Lf ) and for all k ∈ Rd :

f [Lf ]τ [a∗ (k)] = τ [a∗ (k)]f [Lf + (−1)π(τ ) ω(k)] (A.2)

τ [a(k)]f [Lf ] = f [Lf + (−1)π(τ ) ω(k)]τ [a(k)] (A.3)

in the sense of operator-valued distributions. Moreover, extending f to the whole real line
by setting f ≡ 0 on R−
0 Eq. (A.2) and Eq. (A.3) extend to:

τ [a∗ (k)]f [Lf ] = f [Lf − (−1)π(τ ) ω(k)]τ [a∗ (k)] (A.4)

f [Lf ]τ [a(k)] = τ [a(k)]f [Lf − (−1)π(τ ) ω(k)] (A.5)

or in an even more compact notation we can write Eq. (A.2), Eq. (A.3), Eq. (A.4), Eq. (A.5)

114
 115

as:

f [Lf ]aαj (kj ) = aαj (kj )f [Lf + (−1)π(τj ) σj ω(kj )] (A.6)

aαj (kj )f [Lf ] = f [Lf − (−1)π(τj ) σj ω(kj )]aαj (kj ) (A.7)

Proof. We know that:

Z
Lf = l[Hf ] − r[Hf ] = dkω(k){l[a∗ (k)a(k)] − r[a∗ (k)a(k)]}

For the second equation Eq. (A.3) the proof is:

Z
l[a(k 0 )]Lf = dkω(k){l[a(k 0 )]l[a∗ (k)a(k)] − l[a(k 0 )]r[a∗ (k)a(k)]} (A.8)

using that the l and r representations commute and using the commutation relations, i.e.
l[a(k 0 )]l[a∗ (k)a(k)] = l[a(k 0 )a∗ (k)a(k)] = l[a∗ (k)a(k 0 )a(k) + δ(k 0 − k)a(k)] = {l[a∗ (k)a(k)] +
δ(k 0 − k)}l[a(k 0 )], then:
Z
0
l[a(k )]Lf = dkω(k){l[a∗ (k)a(k)] + δ(k 0 − k) − r[a∗ (k)a(k)]}l[a(k 0 )] = (Lf + ω(k))l[a(k 0 )]
(A.9)
We can develop the function f (Lf ) in a power series of Lf . We prove the equation for one
term of this series:

l[a(k 0 )]Lnf = (Lf + ω(k))l[a(k 0 )]Ln−1

f = . . . = (Lf + ω(k))n l[a(k 0 )] (A.10)

We Denote by a double point the normal ordering, that is to say the creation operators
are placed to the left of the annihilation operators. Taking also into account that different
representations commute, we have the following theorem:
Q QN
Lemma A.0.2. (Wick’s theorem) Denote N := {1, 2, . . . , N } and j∈A ≡ j=1 χ[j ∈
A] for any A ⊆ N . Then for any (σ1 , σ2 . . . σN ) ∈ {+1, −1}N and (τ1 , τ2 . . . τN ) ∈ {l, r}N ,
αj = (σj , τj ):
X D Y r r
Y E Y
aαj (kj ) = aαj (kj ) r aαj (kj ) r (A.11)
j∈N Q⊆N j∈N \Q j∈Q

Proof. For a proof we refer to [14], considering in this case that different representations
l, r commute.
 116

Lemma A.0.3. Denote N := {1, 2, . . . , N } and let be fj [r] = O(r+1) a measurable function
on R+ for any j = 1, 2, . . . , N . Let Ωf be the KMS state of the GNS field Hilbert space,
then:
N
α
Y X Y
{aj j (kj )fj [Lf ]} = a+
j (kj )×
j=1 Q⊆N j∈Q+

N j N
Y  αj χ(j ∈Q)
/
X
π(τi )
X 
(−1)π(τi ) ω(ki )] Ωf 

× Ωf | [aj (kj )] fj [Lf + r + (−1) ω(ki ) + ×

r=Lf
j=1 i=1 i=j+1
i∈Q− i∈Q+

Y
× a−
j (kj ) (A.12)
j∈Q−
α α α
where [aj j (kj )]χ(j ∈Q)
/ / Q and [aj j (kj )]χ(j ∈Q)
= aj j (kj ) for j ∈ / = 1 for j ∈ Q

Proof. Using the Pull-through formula (lemma 6.1.1.)Eq. (A.6):

N N N N
α α
Y Y Y X
{aj j (kj )fj [Lf ]} = aj j (kj ) fj [Lf + (−1)τi σi ωi ] = (A.13)
j=1 j=1 j=1 i=j+1

applying now the Wick’s theorem (double point denotes normal ordering):
N N
X D Y r rY
E Y
α α
X
= aj j (kj ) r aj j (kj ) r fj [Lf + (−1)τi σi ωi ] = (A.14)
Q⊆N j∈N \Q j∈Q j=1 i=j+1

D Y N
EY N
αj
X Y X X Y
= a+
j (kj ) aj (kj ) fj [Lf + (−1)τi
σ i ω i + (−1)τi
ω i ] a−
j (kj )
Q⊆N j∈Q+ j∈N \Q j=1 i=j+1 i∈Q− j∈Q−
(A.15)
now we move the fj0 s into the vacuum expectation value using that f [Lf ] = hf [Lf + r]i|r=Lf
and finally we note that:
N
X X N
X N
X X
(−1)τi σi ωi + (−1)τi ωi = (−1)τi ωi − (−1)τi ωi + (−1)τi ωi
i=j+1 i∈Q− i=j+1 i=j+1 i∈Q−
i∈Q i∈Q+ i∈Q−

N
X j
X
τi
= (−1) ωi + (−1)τi ωi (A.16)
i=j+1 i=1
i∈Q+ i∈Q−
Appendix B

Operator Algebras

Definition B.0.4. A set A is called a C ∗ − algebra if it has the following properties:

1. A is an algebra (Two intern operations, with the first it is a commutative group, and with
the second it is a semigroup. One extern operation so that with the first intern operation it
is a vector space.).
2. One can define on A a bijection a ∈ A 7−→ a∗ satisfying the following properties, where
a ∈ A, c ∈ C:

(c1 a1 + c2 a2 )∗ = c1 a∗1 + c2 a∗2 (B.1)

(a1 a2 )∗ = a∗2 a∗1 (a∗ )∗ = a (B.2)

(a algebra with such an map is called a ∗ − algebra)

3. We have a norm kak, and A is complete with respect to this norm (a complete normed
algebra is a Banach algebra).
4. The norm has the following property (the property of a C ∗ norm):

ka∗ ak = kak2 (B.3)

Definition B.0.5. A set of bounded linear operators N on a Hilbert space H satisfying the
following conditions is called a von Neumann algebra:
1. N is a ∗ − algebra.
2. N is closed in the weak operator topology. Namely, if a net of operators an ∈ N has a
weak limit a, satisfying:
lim(ψ, an φ) = (ψ, aφ) (B.4)

117
 118

for any vectors ψ, φ ∈ H, then a ∈ N .(We write a = w − lim an )

3. ∈N

B(H): Algebra of all bounded operators on H

Positive operator: A with hψ|Aψi ≥ 0 for each ψ ∈ H. AA∗ is a positive operator and also
|A| = (AA∗ )1/2
Trace class operator: A with kAk1 := T r|A| < ∞, we write A ∈ L1 (H)
Bibliography

[1] A.I.Akhiezer, V.G.Ba’ryakhtar, S.V.Peletminskii, Spin waves, North-holland publishing company

(1968).

[2] W.Amrein, A.Boutet de Monvel, V.Georgescu, C0 -groups, commutator methods and spectral theory
of N-body hamiltonians. (Basel-Boston-Berlin: Birkhaeuser, 1996).

[3] H.Araki, G.L.Sewel, KMS condition and local thermodynamical stability of quantum lattice systems.

[4] H.Araki, E.J.Woods, Representations of the canonical commutation relations describing a nonrelativis-
tic inifinnite free bose gas, J. Math. Phys. 4, 637-662 (1963).

[5] H.Araki, Mathematical theory of quantum fields, Oxford science publications, (1999).

[6] V.Bach, T.Chen, J.Fröhlich, I.M.Sigal, SmoothFeshbach map and operator-theoretic renormalization
groups methods Preprint (2002).

[7] V.Bach, J.Fröhlich, I.M.Sigal, Return to equilibrium, J.Math.Phys., Vol.41, No.6, (2000).

[8] V.Bach, J.Fröhlich, I.M.Sigal, Quantum electrodynamics of confined nonrelativistic particles, Adv.
Math. 137, 299-395 (1998).

[9] V.Bach, J.Fröhlich, I.M.Sigal, A.Soffer, Positive commutators and spectrum of nonrelativistic QED
(1997).

[10] V.Bach, J.Fröhlich, I.M.Sigal, Spectral analysis for system of atoms and molecules coupled to the
quantized radiation field, Commun. Math. Phys. 207, 249-290 (1999).

[11] V.Bach, J.Fröhlich, I.M.Sigal, Mathematical theory of non-relativistic matter and radiation, Lett.
Math. Phys. 34, 183-201 (1995).

[12] V.Bach, J.Fröhlich, I.M.Sigal, Mathematical theory of radiation, Found. Phys. 27, 227-237 (1997).

[13] V.Bach, J.Fröhlich, I.M.Sigal, A.Soffer Positive conmutators and the spectrum of Pauli-Fierz Hamil-
tonians of atoms and molecules, Phys.207, 557-587 (1999).

119
 120

[14] V.Bach, J.Fröhlich, I.M.Sigal, Renormalization group analysis of spectral problems in quantum field
theory, Adv. Math. 137,205-298 (1998).

[15] V.Bach, Spectral analysis of nonrelativistic quantum electrodynamics.

[16] O.Bratteli and D.Robinson, Operators algebras and quantum statistical mechanics 1, text and mono-
graphs in physics, 2nd ed. (Springer-Verlag, Berlin 1987).

[17] O.Bratteli and D.Robinson, Operators algebras and quantum statistical mechanics 2, text and mono-
graphs in physics, 2nd ed. (Springer-Verlag, Berlin 1996).

[18] D.Buchholz, P.Junglas, On the exstence of equilibrium states in local quantum field theory, CMP
121,255-270(1989).

[19] D.Buchholz, I.Ojima,H.Roos, Thermodynamic properties of non-equilibrium states in quantum field

theory, Preprint (2002).

[20] L.Bugliaro, J.Fröhlich, G.M.Graf, stability of quantum elctrodynamics with nonrelativistic matter,
Phys. Rev. Lett., Vol.77, number 17 3494-3497 (1996).

[21] L.Bugliaro Goggia, Stability of quantum electrodynamics with non-relativistic matter and magnetic
Lieb-Thirring estimates, Dissertation at ETH Zuerich 1998.

[22] C.Cohen-Tannoudji, J.Dupont-Roc, G.Grynberg, Photons and atoms, Introduction to quantum elctro-
dynamics (Wiley, New York 1991).

[23] C.Cohen-Tannoudji, Atoms in electromagnetic fields (World scientific 1994).

[24] H.Cycon, R.Froese, W.Kirsch, B.Simon, Schroedinger operators 1st ed. (Springer-Verlag, Berlin, 1987).

[25] E.B.Davies, Quantum theory of open system (Academic, New york 1976).

[26] J.Dereziński,V.Jaks̆ić, Spectral theory of Pauli-Fierz operators Preprint (2000).

[27] J.Dereziński,V.Jaks̆ić, Return to equilibrium for Pauli-Fierz systems Preprint (2002).

[28] G.G.Emch, Algebraic methods in statistical mechanics and quantum field theory (John Wiley and Sons
1972).

[29] C.Fefferman, J.Fröhlich, G.M.Graf, stability of ultraviolet-cutoff quantum elctrodynamics with non-
relativistic matter.

[30] J.Fröhlich, M.Griesemer, B.Schlein, Asymptotic electromagnetic fields in models of quantum-

mechanical matter interacting with the quantized radiation field (2000).

[31] J.Fröhlich, M.Griesemer, B.Schlein, Asymptotic completeness for Compton scattering (2001).
 121

[32] J.Fröhlich, M.Griesemer, B.Schlein, Asymptotic completeness for Rayleigh scattering (2001).

[33] J.Fröhlich, On the infrared problem in a model of scalar electrons and massless scalar bosons, Ann.
Inst. Henri Poincaré Vol.XIX, n1,1973, p.1-103.

[34] J.Fröhlich, Existence of dressed one elctron states in a class of persistent models, Fortschritte der
Physik 22, 159-198 (1974).

[35] I.M.Gelfand, C.E.Schilov, Generalized functions Volume I: properties and operations.

[36] V.Georgescu, C.Gérard, On the virial theorem in quantum mechanics, Commun. Math. Phys. 208,
275-281 (1999).

[37] G.M.Graf, D.Schenker, On the molecular limit of Coulomb gases CMP 174, 215-227 (1995).

[38] R.Haag, Local quantum physics: fields, particles, algebras. Text and monographs in physics (Springer-
Verlag Berlin, 1992).

[39] M.Huebner, H. Spohn, Radiative decay. nonperturbative approaches, Rev. Math. Phys. 7,363-387
(1995).

[40] W.Hunziker, Resonances, metastables states and exponential decay laws in perturbation theory, Com-
mun. Math. Phys.132, 177-188 (1990).

[41] V.Jaks̆ić, C.A.Pillet, A note on Eigenvalues of Liouvilleans, Preprint (2001).

[42] V.Jaks̆ić, C.A.Pillet, From resonances to master equations.

[43] V.Jaks̆ić, C.A.Pillet, Mathematical theory of non-equilibrium quantum statistical mechanics, Preprint
2001.

[44] V.Jaks̆ić, C.A.Pillet, On entropy production in quantum statistical mechanics, Preprint 2000.

[45] V.Jaks̆ić, C.A.Pillet, On a model for quantum friction I: Fermis golden rule and dynamics at zero
temperature, Ann. Inst. Henri Poincar: Phys. theor. 62, 47-68(1995).

[46] V.Jaks̆ić, C.A.Pillet, On a model for quantum friction II: Fermis golden rule and dynamics at zero
temperature, Commun. Math. Phys. 176, 619-643 (1996).

[47] V.Jaks̆ić, C.A.Pillet, On a model for quantum friction III: Ergodic properties of the spin-boson system,
Commun.Math. Phys. 178,627-651 (1996).

[48] V.Jaks̆ić, C.A.Pillet, Spectral theory of thermal relaxation.

[49] V.Jaks̆ić, C.A.Pillet, Non-equilibrium steady states of finite quantum systems coupled to thermal
reservoirs (2001).
 122

[50] V.Jaks̆ić, C.A.Pillet, Ergodic properties of classical dissipative systems I.

[51] J.M.Jauch and F.Rohrlich, The theory of photons and electrons (Springer-Verlag , 1976).

[52] T.Kato, Perturbation theory for linear operators (Springer-Verlag Berlin, 1966).

[53] E.H.Lieb, M.Loss, J.P.Solovej, Stability of matter in magnetic fields.

[54] E.H.Lieb, Self-energy of electrons in non-perturbative QED, Preprint (2000).

[55] Selecta of E.H.Lieb (Edited by W.Thirring) The stability of matter: from atoms to stars (Springer-
Verlag 2001).

[56] M.Merkli, Positive commutators in non-equilibrium quantum statistical mechanics, Com-

mun.Math.Phys.223, 327-362(2001).

[57] M.Merkli, Positive commutator method in nonequilibrium statistical mechanics Lett.Math-Phys. 57:
225-237, 2001.

[58] M.Reed and B.Simon, Methods of modern mathematical physics Vol.1: Funktional analysis (Academic
press 1972).

[59] M.Reed and B.Simon, Methods of modern mathematical physics Vol.2: Fourier analysis , self-
adjointness (Academic press 1975).

[60] M.Reed and B.Simon, Methods of modern mathematical physics Vol.3: Scattering theory (Academic
press 1979).

[61] M.Reed and B.Simon, Methods of modern mathematical physics Vol.4: Analysis of operators (Aca-
demic press 1978).

[62] D.W.Robinson, Return to equilibrium, Commun. Math. Phys. 31, 171-189 (1973).

[63] D.Ruelle, Statistical mechanics: Rigorous results (1988).

[64] D.Ruelle, Thermodynamic formalism (1978).

[65] D.Ruelle, Natural nonequilibrium states in quantum statistical mechanics.

[66] G.L.Sewel, Stability, equilibrium and metastability in statistical mechnics.

[67] G.L.Sewel, Metastable states of quantum lattice systems.

[68] G.L.Sewel, KMS conditions and local thermodynamical stability of quantum lattice systems II.

[69] I.M.Sigal, Renormalization group approach in spectral analysis and problem of radiation, Preprint
(2000).
 123

[70] H.Spohn, Dynamics of charged particles and their radiation field, Preprint (1999).

[71] J.Von Neumann, Mathematical foundations of quantum mechanics, Princenton university press 1995.

[72] K.Yosida, Theory of magnetism, Springer Verlag (1996).

View publica