Bose-Einstein Condensation: Analysis of Problems and Rigorous Results

Download as pdf or txt
Download as pdf or txt
You are on page 1of 155

International School for Advanced Studies

Mathematical Physics Sector


October 2007

Ph.D. thesis

Bose-Einstein condensation:
analysis of problems and
rigorous results

by

Alessandro Michelangeli

supervisor external referee


prof. Gianfausto prof. Jakob
Dell’Antonio Yngvason
Contents

Introduction and outline 1

1 Physical preliminaries 7
1.1 Bose-Einstein condensation theoretically . . . . . . . . . . . . . . . . . . . 7
1.2 B.E.C. historically . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 B.E.C. experimentally . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2 Mathematical preliminaries 23
2.1 Density matrices: definition and general properties . . . . . . . . . . . . . 23
2.2 Trace of kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Partial trace and reduced density matrices . . . . . . . . . . . . . . . . . . 31
2.4 Distances of states and of reduced density matrices . . . . . . . . . . . . . 34

3 Formalisation of Bose-Einstein condensation 41


3.1 Many-body quantum systems of identical bosons . . . . . . . . . . . . . . 42
3.2 Limit of infinite systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3 Bose-Einstein condensation for finite size systems . . . . . . . . . . . . . . 50
3.4 B.E.C. in the limit of infinite systems . . . . . . . . . . . . . . . . . . . . 55
3.5 Proof of B.E.C. in a scaling limit: two examples.
Stationary Gross-Pitaevskiı̆ theory. . . . . . . . . . . . . . . . . . . . . . . 60

4 Role of scaling limits in the rigorous analysis of B.E.C. 65


4.1 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Ground state energy asymptotics for cold dilute Bose gases . . . . . . . . 66
4.3 The Gross-Pitaevskiı̆ scaling scenario . . . . . . . . . . . . . . . . . . . . . 68
4.4 From the GP to the Thomas-Fermi scaling . . . . . . . . . . . . . . . . . . 73
4.5 GP vs TF limit in investigating B.E.C. . . . . . . . . . . . . . . . . . . . . 77
4.6 Low-dimensional scalings . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.7 Appendix: scaling laws for the scattering length . . . . . . . . . . . . . . . 79

5 Strengthened convergence for reduced density matrices in the


derivation of the Gross-Pitaevskiı̆ equation 83
5.1 Time-stability of condensation . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2 Formal derivation of the time-dependent GP equation . . . . . . . . . . . 84
5.3 Rigorous derivation of the time-dependent CNSE . . . . . . . . . . . . . . 87
5.4 Scheme of derivation: method of the hierarchies . . . . . . . . . . . . . . . 91
5.5 Historic survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.6 Strengthened convergence in ESY theorem . . . . . . . . . . . . . . . . . . 100

i
ii Contents

5.7 Strengthened convergence in the AGT theorem . . . . . . . . . . . . . . . 101


5.8 Unified strengthened convergence from ESY/AGT theorems . . . . . . . . 104

6 Interparticle correlations in the condensate 111


6.1 Correlations in many-body condensed states . . . . . . . . . . . . . . . . . 112
6.2 Ground state correlations: Jastrow and Dyson factors . . . . . . . . . . . 114
6.3 Ground-state-like correlations lower the energy . . . . . . . . . . . . . . . 120
6.4 Correlations at the level of reduced density matrices . . . . . . . . . . . . 123
6.5 Trial function for the evolution of a condensate . . . . . . . . . . . . . . . 126
6.6 Correlations and energy discrepancies in the evolution of the condensate . 129

7 Equivalent characterizations of B.E.C. 131


7.1 B.E.C. of infinite systems at any k-particle level . . . . . . . . . . . . . . . 131
7.2 Asymptotic factorisation of the N -body wave function . . . . . . . . . . . 134

8 Concluding remarks 141

References 151
Introduction and outline

This work is an up-to-date and partially improved analysis of the basics of the mathe-
matical description of Bose-Einstein condensation in terms of first principles of Quantum
Mechanics.
The main aims are:

¬ to provide a compact, yet coherent overview

3 of the basic mathematical tools used to formalise Bose-Einstein condensation,


3 of the mathematical techniques for studying several features of this physical
phenomenon,
3 and of how such means emerge as the natural ones in connection with their
physical interpretation;

­ to discuss and to place in the above perspective some new contributions and im-
provements

3 on equivalent characterizations of Bose-Einstein condensation,


3 on the strength of the convergence for some currently available asymptotic
results,
3 and on the effects of interparticle correlations on the energy and the dynamics
of the many-body condensate.

Framework
As a general feature of the Physics of large (many-body) systems, mathematically precise
statements about the implications of the equations of Quantum Mechanics have to be the
necessary counterpart of formal derivations, perturbative approximations and numerical
treatments, which come all three into the game in the absence of exact solutions to the
equations, due to the overwhelming complexity of the problem. Indeed the validity of
a formal as well as a perturbative approach often remains to be studied and rigorously
proved and the feasibility of numerical solutions is heavily limited by the computational
power of even modern computers.
In this perspective, we will be concerned with the rigorous description of a large
number of bosons of certain chemical species, mutually interacting via a purely repul-
sive pair interaction of short range, confined in some container of macroscopic size, at a
suitably high dilution and low temperature. These are the systems that can be actually
considered in the experiments nowadays, after a long-lasting theoretical investigation,

1
2 Introduction and outline

and which exhibit bizarre properties that go under the name of Bose Einstein con-
densation (B.E.C.): not only they remain in a gaseous phase at all temperatures down
to absolute zero, but also they appear as if almost all particles behave as one, con-
densing in a condensed cloud (possibly surrounded by a thermal cloud) where, quantum
mechanically, particles are in the same one-body quantum state.
This is a purely quantum phenomenon that does not find any comprehension within
a classical picture and this makes B.E.C. one of the most striking evidences of Quan-
tum Mechanics. In fact, B.E.C. is a consequence of quantum statistics only, namely of
undistinguishability of identical bosons. In the simple case of an ideal quantum gas, all
thermodynamic properties of the system can be computed explicitly and the condensa-
tion mechanism can be clearly depicted and interpreted. In nature, however, particles
exhibit forces on each other and these interactions among the particles complicate the
theory immensely. In the presence of interaction, B.E.C. is far from being completely
understood, although one has a clear theoretical evidence of how it originates. Hence,
the comprehension in terms of first principles of all B.E.C. features showing up in the
most recent experiments is a major challenge from a Mathematical Physics viewpoint.
A new and reach mainstream in the mathematical analysis of B.E.C. and related
topics is boomed in the last decade, after the first experimental observations in 1995. It
is remarkable, instead, that the theoretical discover of this phenomenon dates back to the
1920s, even before the full establishment of Quantum Mechanics, while the mathematical
structure, before the current new flurry of interest, had not being improving significantly
since the semi-rigorous treatment of the 1950s and 1960s.
In this framework, a number of rigorous (and beautiful) results is already established
in the literature. They cover an amount of connected fields, such as

3 the stationary description of the ground state and the ground state energy of a
dilute Bose gas, both with short-range and with Coulomb interactions (‘charged’
gas),

3 the Bose-Einstein quantum phase transition in optical lattice models,

3 the emergence of vortices in rapidly rotating condensates and their characteristic


pattern

3 one- and two-dimensional limits of the three-dimensional theory,

3 the dynamics of a Bose gas freely expanding from an originally confined and con-
densed phase.

Altogether, they constitute a still now incomplete description – a general proof of Bose-
Einstein condensation for interacting gases still eludes us – and they all are asymptotic
results in some limit of infinite number of particles, for which a full control of the errors
is still lacking.
In the present work, our interest will be limited to the basics of such a rigorous
description:

¬ mathematical and physical well-posedness of the definition of B.E.C. and some


equivalent versions,

­ nature of the customarily adopted scaling limits,


Introduction and outline 3

® control of the convergence in the corresponding asymptotic results,

¯ emergence and persistence in time of correlations in the many-body condensate.

The physical picture will be that of Bose gases at zero temperature, i.e., in their ground
state. In real experiments the temperature is not exactly zero, of course, but low enough
to allow for a ground state description.
These topics, stemming from Condensed Matter Physics, Statistical Physics, and
Kinetic Theory, from the mathematical point of view are functional analytic problems
involving both standard mathematical techniques of advanced Functional Analysis, Cal-
culus of Variations, and Operator Theory, and ad hoc techniques specifically developed
in this context (generalized Poincaré inequalities, hard-to-soft potential transformation
by partially sacrificing the kinetic energy, localization of the energy, a priori estimates on
higher power of the energy with cut-off techniques, diagrammatic control of Duhamel-like
expansions, just to mention some of them).
Our investigation and contributions place themselves after the analysis and the re-
sults of the following contributors: Lieb, Seiringer, Solovej, and Yngvason, for the time-
independent picture, and Elgart, Erdős, Schlein, and Yau, and Adami, Mauser, Golse,
and Teta, for the time-dependent one. These, in turn, rely on a vast landscape of related
researches which will be thoroughly recalled.

Synopsis
The material is organized as follows.

Chapter 1 provides a short, self-consistent description of the essential physical features


of B.E.C. as they have been consolidated through a 80-years-long investigation. The
purpose here is to focus on the theoretical and experimental basis of the concept
of B.E.C. as a macroscopic occupation of a single-particle quantum state.

Chapter 2 is, in turn, a mathematical detour on the properties of the crucial and
ubiquitous tool throughout this work: the reduced density matrix. Physically, it
encodes the description of a subsystem of the original system one starts with. The
standard notions of kernels, trace, partial trace are here reviewed, together with
some other useful results for the following.

Chapter 3 presents the rigorous formalisation of B.E.C., from the mathematics to


describe large many-body boson systems, to the emergence of the reduced density
matrices as the natural tool, and the conceptual and technical need of a limit of
infinite number of particles. The outcoming notion of B.E.C. will be, from this mo-
ment on, an intrinsically asymptotic notion, accounting for the convergence of the
one-body reduced density matrix to the projection onto the one-body condensate
wave function. Some distinct natural topologies are here shown to be equivalent
for controlling such a convergence. To see how fertile this framework is, we report
the main achievements it has recently led to: the rigorous derivation of the sta-
tionary Gross-Pitaevskiı̆ theory for a dilute Bose gas and the proof of existence of
asymptotic B.E.C. for a model of bosons paired by a short-range and hard-core
interaction.
4 Introduction and outline

Chapter 4 puts the emphasis on the large-size limits used in studying many-body
systems. It is already well known that these are scaling limits, namely where the
interaction is scaled with the number of particles according to some prescriptions,
yet the subject deserves further consideration. By scaling, one handles a more
and more populated system which shares some physical features with the original
one. The collection of these features identifies the scaling of interest. In particular,
a scaling limit is not simply a thermodynamic limit. Any asymptotic result on
B.E.C., as well as its definition itself, depends actually on the scaling adopted
to derive it, namely under which physical regime it is derived, and should be
accompanied by a control of the error terms in the asymptotics, which is currently a
major open issue. In particular, the role of the Gross-Pitaevskiı̆ scaling is reviewed,
as the limit of ultra-high dilution which still enables one to study the dynamics of
the gas.
Chapter 5 is centred on the issue of the persistence in time of the condensed phase,
after the gas is released from the trap that was confining it. The underlying mathe-
matical problem is the derivation of the time-dependent non-linear equation for the
one-body condensate wave function, starting from the linear many-body dynam-
ics. A formal derivation, rather commonly accepted as satisfactory, together with
the fit of the experimental data, show that this equation is the cubic non-linear
Schrödinger equation known as the Gross-Pitaevskiı̆ equation. The path towards a
rigorous derivation, instead, snakes around a much longer sequence of intermediate
achievements, within a framework of classical and quantum kinetic equations which
is even more general than our B.E.C.-related problem. Such a derivation method
involves the study of an infinite hierarchy of evolutionary equations for reduced
density matrices, which is eventually solved by all tensor powers of the projections
onto the solution of the Gross-Pitaevskiı̆ equation. Physically, this corresponds to
describe the time evolution of all finite portions of the system of interest, while
letting it enlarge to infinity in some scaling limit. Both the mathematical scheme
of this method and the previous achievements are reviewed, until the most recent
and somehow conclusive results currently known. These are obtained within the
same conceptual scheme, but through different techniques: the core contribution of
this chapter is to provide some strengthened unified version for the convergence of
reduced density matrices, lifting it to a trace norm convergence at any fixed time.
Chapter 6 touches the problem of determining interparticle correlations establishing
in the true many-body state as a consequence of the interaction. If this state
undergoes B.E.C., correlations shows up in a peculiar form and one can monitor
their dramatic influence on the energy and the dynamics of the condensate. The
net effect is the presence of a typical short scale correlation structure built up (and
preserved in time) by the two-body process. This is actually a crucial issue in
understanding the properties of a condensate. In an asymptotic analysis where,
in the limit N → ∞, marginals factorise and correlations disappear, a strategy is
needed, both at the many-body and at the one-body level, to mime the true time
evolution of the system through some suitable trial state.
Chapter 7 deals with a number of similar characterizations of B.E.C. that one usually
ends up with because they are easier to handle, for technical reasons. The substan-
tial equivalence of all them is proved and discussed. First, it is pointed out that
Introduction and outline 5

the asymptotic convergence of the one-body reduced density matrix to a rank-one


projection is the same as the corresponding factorisation for any fixed number of
variables. After that, a couple of typical asymptotic factorisations of the many-
body state are proved to be equivalent to the standard definition of B.E.C. in terms
of reduced density matrices. This provides some homogeneity in the hypotheses of
several theorems of time-stability of condensation. All these mathematical charac-
terizations share the same physical interpretation of a condensate where essentially
all particles behave as one.

Chapter 8 collects some concluding remarks and outlines some open mathematical
problems related. Some of them are are just improvements with respect to the
currently known results, where certain non-trivial simplifications are believed to
be relaxable. Some others are indeed major open problems, even on basics of
rigorous B.E.C. Among them, a control of the errors in the asymptotic rigorous
results, a quantum statistical treatment which includes the temperature, a deeper
insight in the interparticle correlations, a full treatment of true delta interactions
at least in the one-dimensional case.
6 Introduction and outline
Chapter 1

Physical preliminaries

1.1 Bose-Einstein condensation theoretically


Bose-Einstein condensation is a familiar and rather well understood quantum phenomenon
that physicists have been dealing with for the last 80 years. One can refer to a number
of comprehensive references on the whole subject, which discuss the theoretical physical
background that will be understood throughout in this work. Our choice is unavoidably
towards the most recent and complete of them:

3 Castin [26] (2000),

3 Leggett [71] (2001),

3 Pethick and Smith [96] (2002),

3 Pitaevskiı̆ and Stringari [97] (2003),

3 Yukalov [111] (2004).

Not surprisingly, these are not Mathematical Physics references. Rigorous analysis on
B.E.C. has received a new flurry of interest in the last decade, parallel to the experimental
realizations that finally have been possible and to the comprehension of the most appro-
priate mathematical tools to investigate the subject. One could say that while from the
theoretical Condensed Matter Physics perspective B.E.C. realizations have represented
the apex of huge experimental efforts and the confirm of a long theoretical research, cul-
minating with the 2001 Nobel prize in Physics to Cornell, Ketterle, and Wieman [31, 63],
from the Mathematical Physics viewpoint they have stressed the need of a fully rigorous
comprehension of the phenomenology in terms of first principles.
It is difficult to forsee whether this recent and rich mathematical mainstream will
advance towards the understanding of the newest related phenomena (rotating conden-
sates, quantized vortices, B.E.C./B.C.S. transitions, optical lattices, coherence phenom-
ena, interference and Josephson effects, just to mention some of them). For instance,
temperature itself has still to enter in many of these rigorous treatments, which are set
at T = 0.
Instead, what can be emphasized now, and presumably for the future, is the pro-
lific interaction between this so-called “Mathematics of the Bose gas” and the Physics
behind. The former is developing new advanced tools combining their own abstract in-
terest with their physical applications. The latter provides true experimental data, as

7
8 Chapter 1. Physical preliminaries

Figure 1.1. Cover of Science magazine of December 22, 1995, declaring the Bose
condensate as the “molecule of the year”.

well as prescriptions and ansatz for the expected mathematical results, and benefits from
a comprehension and a predictive description on a rigorous basis.
The standard few-words description of B.E.C. is
the macroscopic occupation of the same single-
particle state in a many-body system of bosons.
This is illustrated by the cover of Science magazine of December 22, 1995, in which the
Bose condensate is declared “molecule of the year” and pictured as a platoon of soldiers
marching in lockstep: each particle in the condensate shares a quantum mechanical wave
function and so they all move as one; particles outside the condensate move faster and in
all directions (see Fig. 1.1). Actually this picture is misleading, rigorously speaking: the
condensate is described by an essentially factorised many-body wave function ΨN ∼ ϕ⊗N
which is interpreted as the occupation of the common one-body state ϕ. This, in turn,
extends to the whole region: one cannot distinguish among distinct ϕ’s marching close
together.
The ubiquitous and celebrated Fig. 1.2 gives the typical qualitative explanation of this
phenomenon. It provides at least a semi-quantitative prediction: occurrence of B.E.C.,
as a consequence of undistinguishability, namely, of quantum statistics, at a temperature
and a dilution such that the thermal (de Broglie) wavelength becomes comparable to the
mean interparticle distance. This is in contrast with other phase transitions (like melting
or crystallization), which depend on the interparticle interactions.
1.1. Bose-Einstein condensation theoretically 9

Figure 1.2. Semi-qualitative explanation for Bose-Einstein condensation

Ideal gas picture


A first analytical explanation of such a mechanism is customarily done for an ideal
Bose gas, that is, N undistinguishable non-interacting non-relativistic quantum particles,
which for concreteness are assumed to be in a box Ω with volume |Ω|. To treat the most
natural scenario, hereafter in this section, a three-dimensional setting will be assumed.
The emergence of this phenomenon, rigorously speaking, requires a thermodynamic
limit. In the grand-canonical description where the system is put in thermal equilibrium
at temperature T with a reservoir with chemical potential µ, as a consequence of the
statistics of bosons one finds that the total occupation number is
X 1
N = ≡ N 0 +N therm (1.1)
i>0
eβ(εi −µ) −1
1
N 0 := , (1.2)
eβ(ε0 −µ) − 1
where β = (kB T )−1 is the inverse temperature in terms of the Boltzmann’s constant kB
and εi is the energy of the i-th excited state of the single-particle Hamiltonian

p2
H (1) = , (1.3)
2m
while the Hamiltonian of the many-body system is
X (1)
X p2
k
H= Hk = . (1.4)
2m
k k
10 Chapter 1. Physical preliminaries

Here each pk is the quantum operator pk = −i ~∇rk acting on the one-body Hilbert space
L2 (Ω) with appropriate boundary conditions. As usual, ~ and m are the Plank’s constant
and the mass of each particle, respectively. For instance, the well known eigenvalues of
H (1) for a three-dimensional cubic box of side length L with periodic boundary conditions
are
(2π~)2 2
εn = n , n ∈ N3 . (1.5)
2mL2
Thus, one distinguishes among the occupancy N 0 of the ground state and the remaining
thermal component N therm , and in the thermodynamic limit ε0 → 0.
For fixed µ < 0, in the thermodynamic limit,
Z
N |Ω|→∞ ³ 2π ´3 1
−−−−→ p2
dp =: ρ (1.6)
|Ω| ~ R3 e β( 2m −µ)
−1
which is monotonously increasing with µ and bounded, as µ → 0, by
³ m ´3/2
ρcrit := g3/2 (1) . (1.7)
2π~2 β

Here Z √
+∞ X∞
2 x z`
g3/2 (z) := √ dx = , (1.8)
π 0 z −1 ex − 1 `3/2
`=1

whence g3/2 (1) ' 2.612. Thus, condensation arises as the phenomenon where particles
exceeding the above critical number all go into the lowest energy state. To fix the density
at some value larger that ρcrit one has to simultaneously let |Ω| → ∞ and µ → 0. Limit
(1.6) can be controlled separating the contribution from the lowest energy level and
approximating the contribution from the remaining terms by an integral, and the net
result is
ρ = ρcond + ρcrit (ρ > ρcrit ) (1.9)
where
N0
ρcond := lim (1.10)
|Ω|→∞ |Ω|

is the density of the condensate.


Transition between the non condensed regime and B.E.C. corresponds to the transi-
tion between a zero value and a positive value for ρcond , hence, it is determined by the
condition ρcrit = ρ in (1.7), which leads to the critical temperature

2π~2 ³ ρ ´2/3
Tc = . (1.11)
mkB g3/2 (1)

This corresponds to
1 1
lim =0 (1.12)
L→∞ L3 eβ(ε0 (L)−µ(L)) − 1
where L is the length of the side of the box and the dependence µ = µ(L) is determined
by (1.1), writing N = ρ L3 with fixed ρ. The critical value Tc is the smallest Tc for which
(1.12) is true. Condensation occurs for T 6 Tc . One usually restates this criterion in the
much more familiar form
ρ λ3dB = g3/2 (1) ' 2.612 (1.13)
1.1. Bose-Einstein condensation theoretically 11

where λdB is the thermal (or de Broglie) wavelength


s
2π~2
λdB := . (1.14)
mkB T
Condensation occurs when
ρ λ3dB & 2.612 , (1.15)
that is, below the critical value of the temperature such that the thermal wavelength is
of the same order of the mean interparticle distance ρ−1/3 . In the same picture one
proves that the ground state occupancy in the condensed regime is
N0 ³ T ´3/2
=1− (T 6 Tc ) . (1.16)
N Tc
This way, one recovers the interpretation of Figure 1.2.
Among other peculiar features, the condensate phase turns out to have infinite com-
pressibility and zero entropy and one has a first order gas/condensate phase transition.
In particular, at zero temperature there is 100% B.E.C.: all atoms are in the ground
state ϕ of the single-particle Hamiltonian H (1) . Therefore, ϕ is called the condensate
(or macroscopic) wave function.

Interacting gas picture


The explanation above is limited to ideal Bose gases. In the presence of interaction a
similar exact treatment is no longer possible. Since the many-body Hamiltonian is not
just the sum of single-particle Hamiltonians, the ground state many-body wave function
is not a product of N single-particle wave functions.
The first systematic and semi-rigorous treatment of interacting Bose fluids was due
to Bogolubov in 1947, based on a new perturbation technique (see the historical perspec-
tive in the next section, as well as that in Sec. 4.2 concerning the intense mainstream
after Bogolubov in the determination of the ground state energy of a cold dilute Bose
gas). Bogolubov’s theory relies on some unproved assumptions, yet provides a successful
description of the phenomenology. Today it has become a standard textbook subject
and the above references treat it extensively. For a comprehensive review in the current
case of cold and dilute Bose gases we also refer to the 2001 Physics Report exhaustive
presentation by Zagrebnov and Bru [113].
What one finds is that the ideal gas picture retains validity even when we include
weak interactions. Here below we summarize the main points.
¬ The dilute nature of the gas allows one to describe the effects of the interaction in
a rather fundamental way, via a single physical parameter, the s-wave scattering
length a of the two-body scattering process. In fact, an important simplification
for the alkalis and for H, is that the values of thermal energy kB T characterizing
B.E.C. conditions are small compared to the characteristic energies at which the
two-body scattering process takes place. As a consequence, in the description of
the scattering one legitimately neglects all partial waves, but the s-wave. The value
of a could be in principle determined if one knew the interatomic potential with
high precision. One can extract it from the experimental measurements; recently
it has become possible in some cases to tune the value of a by applying an external
magnetic potential, so to probe how theoretical results vary with a.
12 Chapter 1. Physical preliminaries

­ Under ultra-high dilution, that is, when the mean interparticle distance is far larger
than the scattering length,
ρ a3 ¿ 1 , (1.17)

one can neglects all but the two-body interactions among particles. The Hamilto-
nian is then
XN ³ 2 ´ X
pk
H = + U (rk ) + V (rk − rh ) , (1.18)
2m
k=1 16k<h6N

where the potential V has scattering length a and U is the trap. The ground state
energy of H, provided that the gas is sufficiently diluted (ρ a3 ¿ 1) and populated
(large N ), turns out to be1

4π~2 ρ a
E g.s. ∼ N. (1.19)
m

® A healing (or indeterminacy) length is defined as

1
` := √ . (1.20)
8πρ a

It is the length over which the perturbing effect of a confining wall is “healed”
(whence its name): in the presence of repulsive interactions, it is energetically
favourable to make the density of the gas nearly constant in the bulk of the box
and to fall off to zero exponentially with characteristic length ∼ `, as we approach
a wall. In the low density regime it is impossible to localize the particles relative
to each other (even though ρ is small), since

a ¿ ρ−1/3 ¿ `c (1.21)
−1/3 −1/3
(indeed ρ `c ∼ (ρa)
ρ 3 1/6 ¿ 1). Bosons in their ground state are therefore
−1/2 = (ρa )

smeared out over distances large compared to the mean particle distance and their
individuality is completely lost. They cannot be localized with respect to each
other without changing the kinetic energy enormously.

¯ As suggested by such a “delocalization picture”, at zero temperature the many-


body ground-state wave function turns out to be still, to a very good approximation,
a product of N copies of the same single-particle wave function ϕGP .

Ψ(r1 , . . . , rN ) = W (r1 , . . . , rN ) ϕGP (r1 ) · · · ϕGP (rN )


(1.22)
∼ ϕGP (r1 ) · · · ϕGP (rN )

This is the phenomenon of condensation. Such a ϕGP turns out to solve a nonlinear
Schrödinger equation, the so-called Gross-Pitaevskiı̆ equation

~2 4π~2 a
− ∆ϕ(r) + U (r)ϕ(r) + |ϕ(r)|2 ϕ(r) = µ ϕ(r) , (1.23)
2m m
1
It is worth noticing that, within Bogolubov’s theory, estimate (1.19) follows assuming B.E.C. Only
40 years after, the same result has been proved rigorously (see Sec. 4.2) for any dilute Bose gas with no
additional assumption of condensation.
1.1. Bose-Einstein condensation theoretically 13

R
with |ϕ(r)|2 dr = 1, and the many-body ground state energy turns out to be

4π~2 ρ a £ ¤
E g.s. ∼ N ∼ N E GP ϕGP (1.24)
m
where Z ³ 2
GP ~ 2π~2 a 4 ´
E [ϕ] := |∇ϕ|2 + U |ϕ|2 + |ϕ| dr (1.25)
2m m
and
R ϕGP actually minimizes such an energy functional among all ϕ’s such that
|ϕ(r)|2 dr = 1. Here µ is fixed by normalisation and has the meaning of a chemical
potential. The nonlinear term in (1.23) accounts for an on-site (i.e., local) self
interaction, as if each particle would be subject to an additional potential given
by the particle density itself, and describe the effect of two-body collisions. The
issue of the rigorous derivation of (1.23), beyond the typical formal treatment that
we will present Sec. 5.2, has been central in Mathematical Physics until the very
recent rigorous results reported in Sec. 3.5 and 5.3, and still many major related
problems remain open.

° In the event of condensation, (1.22) can still be interpreted quantitatively as the


presence of a “fraction N 0 /N of particles” in the condensate, that is, each in the
one-particle state ϕGP . The remaining “fraction of particles” 1 −N 0 /N is called
the quantum depletion. This fraction is positive even at zero temperature, due
to the presence of repulsive interaction, and one cannot have 100% B.E.C. like for
the ideal case. At T > 0 depletion is enhanced by thermal effects and one speaks
in addition of thermal depletion. In Bogolubov’s theory

8 p 3
quantum depletion = √ ρa , (1.26)
3 π

typically 1% or less for the alkali condensates. This means that even for the inter-
acting gases, we can, with 99% of accuracy, regard all the atoms to have the same
single-particle wave function.2

± This “imperfect Bose gas” turns out not to be infinitely compressible as the ideal
counterpart. The transition gas/condensate can be regarded to be of the second
order, provided that one neglects effects of higher order in a/λdB and ρaλ2dB .

Hence, summarizing, occurrence of B.E.C. at zero temperature for interacting Bose


gases is governed by the condition
(
negligible contribution of the l 6= 0 waves to the two-body cross section
(1.27)
ultra-high dilution ρ a3 ¿ 1

which replaces criterion (1.15) for ideal gases.


The density distribution of a condensate can be directly observed in a nondestructive
way. Such observations can be regarded as a direct visualization of the magnitude of the
macroscopic wave function. In B.E.C., one simultaneously realizes millions of identical
copies of the same one-body wave function and the entire many-body wave function can
2
This is in contrast to liquid helium in which the quantum depletion is about 90%.
14 Chapter 1. Physical preliminaries

thus be measured while affecting only a small fraction of the condensed atoms. The re-
sulting dramatic visualizations of wave functions are an appealing aspect of experimental
studies of B.E.C.
In turn, experimental data can be fit with the numerical results, with an excellent
agreement. Numerical analysis is essentially the only way to explicitly obtain thermody-
namic quantities of the many-body picture (when they are computationally accessible)
and the one-body condensate wave function. A number of techniques and tools has been
developed across a huge literature: we refer to [8, 86, 106].

1.2 B.E.C. historically


The basic idea of B.E.C. dates back to 1925 when Einstein, on the basis of a 1924 paper by
the Indian physicist Bose [18], devoted to the statistical description of the quanta of light,
predicted the occurrence of a phase transition in a gas of non-interacting atoms [38, 39].
This phase transition is the consequence of quantum statistical effects. Bose’s paper
dealt with the statistics of photons, for which the total number is not a fixed quantity.
He sent his paper to Einstein asking for comments.3 Recognising its importance, Einstein
translated the paper and submitted it for publication. Subsequently, he extended Bose’s
treatment to massive particles, whose total number is fixed.
For a long time these predictions had no practical impact and were viewed as a
mathematical curiosity with little experimental interest, however. In 1938 London [81],
immediately after the discovery of superfluidity in liquid helium, had the intuition that
superfluidity could be a manifestation of B.E.C. (an issue which was highly controversial
between him and Landau’s “quantum hydrodynamics”).
The first self-consistent theory of superfluids was developed by Landau [67] in 1941
in terms of the spectrum of elementary excitations of the fluid. In 1947 Bogolubov [15]
developed the first microscopic theory of interacting Bose gases, based on the concept of
B.E.C.: that theory, while intuitively appealing and undoubtedly correct in many aspects,
has major gaps and some flaws. In the 1950s, after Landau and Lifshitz [68], Penrose [94],
and Penrose and Onsanger [95] introduced the concept of the “off diagonal long range
order” and discussed its relationship with B.E.C., intense theoretical work was developed,
3
While presenting a lecture at the University of Dhaka on the photoelectric effect and the ultraviolet
catastrophe, Bose intended to show his students that the current theory was inadequate, because it
predicted results not in accordance with experimental results. During this lecture, Bose committed an
error in applying the theory, which unexpectedly gave a prediction that agreed with the experiments. The
error was a simple mistake – similar to arguing that flipping two fair coins will produce two heads one-
third of the time – that would appear obviously wrong to anyone with a basic understanding of statistics.
However, the results it predicted agreed with experiment and Bose realized it might not be a mistake at
all. He for the first time took the position that the Maxwell-Boltzmann distribution would not be true
for microscopic particles where fluctuations due to Heisenberg’s uncertainty principle will be significant.
Thus he stressed in the probability of finding particles in the phase space each having volumes h3 and
discarding the distinct position and momentum of the particles. Physics journals refused to publish
Bose’s paper. It was their contention that he had presented to them a simple mistake and Bose’s findings
were ignored. Discouraged, he wrote to Albert Einstein, who immediately agreed with him. His theory
finally achieved respect when Einstein sent his own paper in support of Bose’s to Zeitschrift für Physik,
asking that they be published together (see Ref. [30, 62]). The reason Bose’s “mistake” produced accurate
results was that since photons are indistinguishable from each other, one cannot treat any two photons
having equal energy as being two distinct identifiable photons. By analogy, if in an alternate universe
coins were to behave like photons and other bosons, the probability of producing two heads would indeed
be one-third (tail-head = head-tail).
1.2. B.E.C. historically 15

aimed to better understand the relationship between B.E.C. and superfluidity. While
theoretical intuition benefited greatly from this activity, the mathematical structure did
not significantly improve.
In the same years, experimental studies on superfluid helium had become more and
more refined, checking Landau’s predictions for the excitation spectrum and provid-
ing the first measurements of the condensate fraction through the determination of the
momentum distribution. An important development in the field took place with the
prediction of quantized vortices by Onsager [93] (1949) and Feynman [48] (1955), and
with their experimental discovery by Hall and Vinen [55] (1956).
Superfluid liquid 4 He is the prototype Bose-Einstein condensate and it has played
a unique role in the development of physical concepts. However, interaction between
helium atoms is strong and this reduces the number of atoms in the zero-momentum
state even at absolute zero. Consequently, it is difficult to measure directly the occupancy
(which is experimentally known today to be less than 1/10). The fact that interactions
in liquid helium reduce it dramatically led to the search for weakly interacting Bose
gases with a higher condensate fraction. The difficulty in most substances is that, at low
temperatures, they do not remain gaseous but form solids or, in the case of the helium
isotopes, liquids, and the effects of interaction thus become large. In other examples
atoms first combine to form molecules, which subsequently solidify.
As long ago as in 1959 Hecht [56] argued that spin-polarized hydrogen would be a
good candidate for a weakly interacting Bose gas. The attractive interaction between
two hydrogen atoms with their electronic spins aligned was then estimated to be so weak
that there would be no bound state. Thus, a gas of hydrogen atoms in a magnetic
field would be stable against formation of molecules and, moreover, would not form a
liquid, but remain a gas to arbitrarily low temperatures. Hecht’s paper was ahead of its
time and received little attention. The experimental studies on the dilute atomic gases
were developed much later, starting from the 1970s, profiting from the new techniques
developed in atomic physics based on magnetic and optical trapping, and advanced
cooling mechanisms. In a series of experiments hydrogen atoms were first cooled in a
dilution refrigerator, then trapped by a magnetic field and further cooled by evaporation,
coming very close to B.E.C., although this approach was still limited by recombination
of individual atoms to form molecules.
In the 1980s, laser-based techniques, such as laser cooling and magneto-optical trap-
ping, were developed to cool and trap neutral atoms. Laser cooling have been the major
experimental breakthrough, in contrast to so many other proposals which in practice work
less well than that predicted theoretically. Alkali atoms are well suited to laser-based
methods because their optical transition can be excited by available lasers and because
they have a favourable internal energy-level structure for cooling to very low tempera-
tures. Once they are trapped, their temperature can be lowered further by evaporative
cooling.
By combining the different cooling techniques, the experimental teams of Cornell and
Wieman at Boulder and of Ketterle at MIT eventually succeeded in 1995 in reaching the
temperature and the densities required to observe B.E.C. in vapours of 87 Rb [6] and 23 Na
[34], respectively. In the same year, first signatures of the occurrence of condensation in
vapours of 7 Li were also reported [19]. B.E.C. was later achieved in other atomic species,
including spin-polarized hydrogen, metastable 4 He [49], and 41 K [89].
In the end, the successful approach was to use laser cooling only as pre-cooling for
16 Chapter 1. Physical preliminaries

magnetic trapping and evaporative cooling. Laser cooling opened a new route to ultra-
low temperature physics. Also, the number of atomic species which can be studied
at ultra-low temperatures was greatly extended from helium and hydrogen to all of the
alkali atoms, metastable rare gases, several earth-alkali atoms, and others (the list of laser
cooled atomic species is still growing). In this sense, the 2001 Nobel prize in Physics to
Cornell, Ketterle, and Wieman “for the achievement of Bose-Einstein condensation in
dilute gases of alkali atoms, and for early fundamental studies of the properties of the
condensates” is along the same mainstream of the 1997 Nobel prize in Physics to Chu,
Cohen-Tannoudji and Phillips “for development of methods to cool and trap atoms with
laser light”.
Let us conclude this historical outlook by mentioning that in the course of time the
concept of B.E.C. have found applications in many system other than liquid He and
boson alkali atoms. For example, many aspects of the behaviour of superconductors
may be understood qualitatively on the basis of the idea that pairs of electrons form
a Bose-Einstein condensate (although their properties are quantitatively very different
from those of weakly interacting gas of pairs). B.E.C. of pairs of fermions is also observed
experimentally in atomic nuclei, where the effect of the neutron-neutron, proton-proton,
and neutron-proton pairing may be seen in the excitation spectrum as well as in reduced
moments of inertia. Theoretically, B.E.C. of nucleon pairs is expected to play an impor-
tant role in the interiors of neutron stars and observations of glitches in the spin-down
rate of pulsars have been interpreted in terms of neutron superfluidity. The possibility
of mesons, either pions or kaons, forming a Bose-Einstein condensate in the cores of
neutron stars has been widely discussed, since it would have far-reaching consequences
for theories of supernovae and the evolution of neutron stars. In the field of nuclear
and particle Physics the ideas of Bose-Einstein condensation also find application in the
understanding of the vacuum as a condensate of quark-antiquark (uū, dd̄, and ss̄) pairs,
the so-called chiral condensate. This condensate gives rise to particle masses in much
the same way as the condensate of electron pairs in a superconductor gives rise to the
gap in the electronic excitation spectrum.

1.3 B.E.C. experimentally


The current explosion of activity con ultra-cold atoms and Bose-Einstein condensation
is well documented by the huge number of groups working on it, partially listed at
http://www.uibk.ac.at/exphys/ultra-cold/atomtraps.html ,
and by the vast bibliography of several thousands papers related to B.E.C. written in
the lat decade, that one can find collected at
http://bec01.phy.georgiasouthern.edu/bec.html/bibliography.html .

So far B.E.C. has been realized in 7 Li, 23 Na, 41 K, 87 Rb, as well as in spin-polarized
H and metastable 4 He. The range of values for the relevant physical parameter in the
condensed regime are collected in Table 1.1.
In particular, in these systems B.E.C. shows up not only in momentum space but also
in coordinate space, making the direct experimental investigation of the condensation
feasible and providing new opportunities for interesting studies, like the temperature
dependence of the condensate, energy and density distributions, interference phenomena,
frequencies of collective excitations, and so on.
1.3. B.E.C. experimentally 17

Table 1.1. Typical values for the relevant physical parameters in the regime of
condensation (expressed in the convenient units for each quantity)

typical B.E.C. values


Temperature 50 nK ÷ 50 µK
Number of atoms 102 ÷ 1011
Trap size 10 ÷ 50 µm
Mean density 1011 ÷ 1015 cm−3
Scattering length 300 ÷ 300 a.u.
Dilution (ρa3 ) 10−4 ÷ 10−6

Trapping, cooling, detecting condensation


Making and probing Bose-Einstein condensates turns out to be a subtle art [64]. At the
ultra-low temperatures and densities of the experiments, the equilibrium configuration
would be the solid phase. Thus, in order to observe B.E.C., one has to preserve the
system in a metastable gas phase for a sufficiently long time.
In order to create a Bose condensate in a dilute gas, atoms must be cooled and
compressed in a trap until the thermal de Broglie wavelength is comparable to the spacing
between atoms. In addition, atoms must be thermally isolated from all material walls.
This is done by trapping atoms with magnetic fields or with laser light inside ultra-high
vacuum chambers. Such traps can store atoms for seconds or even minutes, which is
enough time to cool them. Efforts have to face the challenge of simultaneously achieving
effective laser cooling and trapping, which work best at low atomic densities, and efficient
evaporative cooling, which requires high densities.
Pre-cooling is a prerequisite for trapping because conservative atom traps can only
confine neutral atoms with a maximum energy of one kelvin at best (and in many cases
the trap depth is just a few millikelvins). The pre-cooling is done by laser cooling and the
final cooling by evaporation. These cooling techniques together reduce the temperature
of the atoms by a factor of a billion (from 102 ÷ 103 K in the oven, to 1 ÷ 10 µK after
laser cooling, to ∼ 102 nK after evaporative cooling). In almost all B.E.C. experiments,
pre-cooling is done in a magneto-optical trap loaded from a slowed beam.
To keep the atoms tightly compressed during cooling and hold the condensate for
study, the gas is confined by conservative atom traps that have a sufficiently small heating
rate. The requirements for the trap during cooling are more stringent than they are for
holding condensates. First, the time for cooling (typically 30 sec for evaporative cooling)
is usually much longer than the time for performing experiments on B.E.C., requiring low
heating and trap loss rates. Furthermore, for cooling, the trap needs to be sufficiently
deep to hold the initial (pre-cooled) cloud, and must accommodate a cooling scheme able
to reach B.E.C. temperatures. In particular, suitable magnetic traps (e.g., quadrupole-
type traps, Ioffe-Pritchard traps, etc.) accommodate the pre-cooled atoms and compress
them in order to achieve high collision rates and efficient evaporative cooling.
After that, evaporative cooling (like hot water that cools down in a bathtub or in a
cup of coffee) removes continuously the high-energy tail of the thermal distribution from
the trap. The evaporated atoms carry away more than the average energy, which means
that the temperature of the remaining atoms decreases. The high energy tail must be
18 Chapter 1. Physical preliminaries

constantly repopulated by collisions, thus maintaining thermal equilibrium and sustain-


ing the cooling process. Evaporation of atoms is controlled by radio frequency radiation
(“rf induced evaporation”) that flips the atomic spin. As a result, the attractive trap-
ping force turns into a repulsive force and expels the atoms from the trap. Evaporative
cooling proved to work much better for alkali atoms than for hydrogen, for which the
technique was originally developed. In fact, it turns out that for alkali atoms, in contrast
to hydrogen, the so-called good collisions (elastic collisions necessary for the evaporation
process) highly dominate over the so-called bad collisions (inelastic two- and three-body
collisions).
The resulting condensates can then be manipulated to study their properties in dif-
ferent shapes and symmetries and to explore their dynamic behaviour. Direct-contact
manipulation is excluded, due to the extremely low temperature of the condensate and
the high sticking probability of atoms on cold surfaces. Instead, special tools have been
developed to shape, slice, kick, shake, and stir condensates. Typically, by magnetic
fields one adiabatically expands the condensate and changes its aspect ratio. Focused
off-resonant laser beams are used for “microsurgery” on condensates. and to create
deformed (non-parabolic) trapping potentials.
We remark that everything we know about gaseous Bose condensates has been ob-
tained by optical diagnostics. “Contact probes” cannot be used because the samples
are much smaller (∼ 107 atoms) than even a 10 µm sized probe (∼ 1013 atoms), which
would cause the atoms to equilibrate with the probe rather than the opposite. With
the “cooling power” of rf evaporation one can cool at most 108 atoms per minute and
it would take several months just to cool the sensor tip. Fortunately, optical diagnostics
are very versatile and the ease with which light scattering methods are implemented for
dilute atomic samples is a major advantage over condensed matter systems.
The two most important techniques for observing Bose-Einstein condensates are in-
situ and time-of-flight imaging. In both cases, one obtains an image which reflects the
density distribution of the atoms either in a trapped state or in ballistic expansion.
Absorption imaging is done by illuminating the atoms with a laser beam and imaging
the shadow cast by the atoms onto a CCD camera. To image a transparent object,
information encoded in the phase shift of the light must be converted into intensity
information which can be detected by a photosensor. All dispersive methods rely on
the ability to distinguish scattered and unscattered components of the probe light and
manipulate them independently.
A discernible image with 30×30 pixels and 100 detected photons per pixel involves 105
photons. Even if each photon would “knock” an atom out of the condensate, this would
be “non-perturbative” if the sample had many millions of atoms. This explains why the
large Bose condensates can be observed non-destructively with absorption spectroscopy.
The current generation of low-noise CCD cameras meets all the needs of current
B.E.C. experiments. With about 5 electrons rms noise per pixel, the camera is already
shot-noise limited for more than about 50 detected photons per pixel. The low noise of
the camera usually requires a slow readout over several seconds. Rapid sequencing of
non-destructive images is achieved by storing the whole sequence on the chip and then
reading it out slowly. For this purpose, the sensor is divided into an area for exposure
and a much larger area for storage.
Next figures give a visible flavour of the very subtle experimental expedients to make
condensation possible. Fig. 1.3 refers to an observation of B.E.C. with metastable helium
1.3. B.E.C. experimentally 19

at Laboratoire Kastler Brossel, Paris.4 Fig. 1.4 describe a typical magnetic trap for
confining atoms in the cooling process. Fig. 1.5, 1.6, 1.7, and 1.8 survey the main steps
and features of an experiment of condensation with 87 Rb realized by the experimental
group of York University, Toronto.5

4
http://www.lkb.ens.fr/recherche/atfroids/anglais/activite an.html
5
http://www.yorku.ca/wlaser/projects/projects BEC.htm
20 Chapter 1. Physical preliminaries

(a) (b)

(c)

Figure 1.3. Observation of a condensate of metastable helium atoms. (a) 3D


image. (b) 1D profile of the same cloud after 4ms of expansion time. One can
clearly see the difference between the condensed and the thermal fraction which is
fitted by a bosonic distribution in the 1D profile. (c) Images of a pure condensate
containing 400 000 atoms for different expansion time after switching off the trap.
Initially the elliptical shape of the condensate corresponds to the geometry of the
trap. As the expansion time increases, the ellipticity of the condensate undergoes
an inversion due to mean-field interactions between atoms. This inversion is the
strongest evidence for the existence of a Bose-Einstein condensate.

Figure 1.4. QUadrupole-Ioffe Configuration (QUIC). It consists of two


quadrupole coils and Ioffe coil, with current directions shown by the black arrows
in the left side. The magnetic fields of the Ioffe coil and the quadrupole pair are
in the opposite direction on the z axis, so that they almost cancel. If the currents
in all the coils are equal then the design of the coil geometry means a harmonic
single well exists. If the Ioffe coil current is less than that of the quadrupole pair a
double-well potential is created. For scale, the quadrupoles are 80 mm in diameter.
The coils are driven at up to 27 Amps and are water cooled. The vacuum system,
which is at a pressure of less than 10−11 torr, extends from the left between the
two quadrupole coils.
1.3. B.E.C. experimentally 21

Figure 1.5. Apparatus: upper and lower vacuum chambers. Atoms are first
cooled in the upper vapour cell magneto-optical trap. A laser then pushes the
atoms into the lower chamber which is at a pressure of about 2 · 10−11 torr. The
atoms are then trapped in a second magneto-optical trap and subsequently loaded
into a so called QUIC trap (QUadrupole-Ioffe Configuration). Evaporative cooling
is finally used to achieve Bose Einstein Condensation.

Figure 1.6. Temperature determination of ultra-cold atoms. The atom tempera-


ture is determined by switching off the magnetic fields and observing the expansion
of the atom cloud. A hotter sample of atoms expands faster than a colder sample.
The data below show actual pictures (not modelled data) of such an expanding
cloud: from the plot of cloud radius vs time, temperature is found to be 52 mi-
crokelvins.
22 Chapter 1. Physical preliminaries

Figure 1.7. Evaporative cooling. The final stage of atom cooling is to switch off
all laser beams and apply a radio frequency signal using a small 1 loop antenna. The
purpose of this frequency is to flip the spins of the hot atoms which are then expelled
from the trap resulting in a cooler collection of trapped atoms. The radio frequency
is swept from 20 MHz to the lower frequencies shown in the figure. The cloud size
clearly gets smaller as the lower frequency decreases causing the temperature to
decrease accordingly.

Figure 1.8. Transition to B.E.C. (a) Thermal cloud with N = 1.9 · 106 atoms
at temperature T = 450 nK. (b) Mixed thermal atom cloud and B.E.C. where
N = 1.8 · 106 and T = 400 nK. (c) Pure condensate where N = 4.2 · 105 atoms and
T < 60 nK.
Chapter 2

Mathematical preliminaries

This chapter collects some general theoretic and preparatory material, which is suit-
ably re-organized with respect to standard references as [98, 99]. The scheme is that
of Ref. [83]. Density matrices will be introduced and discussed as special trace class,
Hilbert-Schmidt and compact operators on a Hilbert space, and the trace will be dis-
cussed in abstract, as well as concretely in terms of kernels (Sec. 2.1, 2.2). Then the
partial trace operation will be introduced and characterized, leading to the key notion
of reduced density matrix (Sec. 2.3). Such material is classic and statements will be
quoted without proof. Also we inserted the not-so-known Brislawn’s rigorous control of
the well-posedness of the trace of a kernel operator (a density matrix, for our purposes)
in terms of the integral of the diagonal of its kernel. The chapter is concluded by Sec. 2.4
with a discussion on the distance of states and of reduced density matrices.

2.1 Density matrices: definition and general properties


The aim of this section is to introduce the main notion of the present chapter, i.e., a
density matrix. Notoriously, it is the mathematical tool for describing states in quantum
systems. Density matrices are special trace class operators. Let us recall first some useful
basic facts about the trace class.
Let H be a separable Hilbert space with scalar product ( · , · ) and norm k · k. It
will be always assumed that the Hilbert structure is on C. Separability will be needed
throughout to guarantee the existence of a countable complete orthonormal system {ϕj }j
of H, usually referred to simply as an orthonormal basis. The finite or countable set the
index j runs in will be possibly omitted, so that we will write {ϕj }∞ ∞ M
j=0 , {ϕj }j=1 , {ϕj }j=0 ,
or simply {ϕj }j , depending on the context. In the absence of other specifications, we will
be working with an infinite-dimensional H . Denote by L(H) and Com(H) the spaces of
bounded and of compact operators on H respectively.
Definition–Theorem 2.1.1 (the trace class). Let T ∈ L(H). Then the sequence
{(ϕj , |T |ϕj )}j is summable or not independently of the orthonormal basis {ϕj }j of H.
Denote its sum by

X
Tr |T | := (ϕj , |T |ϕj ) . (2.1)
i=1

The trace class L1 (H) is the space

L1 (H) := {T ∈ L(H) : Tr |T | < +∞} (2.2)

23
24 Chapter 2. Mathematical preliminaries

Properties:

¬ L1 (H) is a ∗-ideal of L(H). It is closed under the usual Banach topology of L(H),
that is, the operator norm topology, iff dim H < ∞.

­ When endowed with the norm

kT kL1 := Tr |T | , (2.3)

then L1 (H) is a Banach space (in fact, a Banach ∗-algebra) and

kT k 6 kT kL1 . (2.4)

Also, the finite rank operators on H are k · kL1 -dense in L1 (H).

® L1 (H) ⊂ Com(H) densely in the operator norm topology.


P
If T ∈ L1 (H) then ∞ i=1 (ϕj , T ϕj ) converges absolutely and the limit is independent of
the orthonormal basis {ϕj }j . Thus, on the trace class one defines the operation of trace
as

L1 (H) −→ C

X (2.5)
T 7−→ Tr[T ] := (ϕj , T ϕj ) .
i=1

¯ Trace is linear; furthermore

Tr[T ∗ ] = Tr[T ]
Tr[T S] = Tr[ST ]
(2.6)
|Tr[T ]| 6 kT kL1
|Tr[T S]| 6 kSk kT kL1

∀ T ∈ L1 (H) and S ∈ L(H).

° The map T 7→ Tr[T · ] is an isometric isomorphism of L1 (H) onto Com(H)∗ , whence


L1 (H) = Com(H)∗ .

± The map S 7→ Tr[S · ] is an isometric isomorphism of L(H) onto L1 (H)∗ , whence


L(H) = L1 (H)∗ .

The following familiar notation will turn out to be useful in the sequel.

Definition–Theorem 2.1.2 (Dirac notation). ∀ ϕ, ψ ∈ H one defines the operator

|ϕihψ| : H −→ H
(2.7)
ξ 7−→ |ϕihψ|ξi := (ψ, ξ) ϕ .

Its range has dimension 0 or 1 and its norm is

k |ϕihψ| k = kϕk kψk . (2.8)


2.1. Density matrices: definition and general properties 25

Further,
|ϕihψ|∗ = |ψihϕ|
|λϕihψ| = λ |ϕihψ|
|ϕihλψ| = λ̄ |ϕihψ|
|ϕ + ϕ0 ihψ| = |ϕihψ| + |ϕ0 ihψ|
(2.9)
|ϕihψ + ψ 0 | = |ϕihψ| + |ϕihψ 0 |
|ϕ0 ihψ 0 | |ϕihψ| = (ψ 0 , ϕ) |ϕ0 ihψ|
S |ϕihψ| = |Sϕihψ|
|ϕihψ| S = |ϕihS ∗ ψ|
∀ ϕ, ϕ0 , ψ, ψ 0 ∈ H, ∀ λ ∈ C, ∀ S ∈ L(H).
Definition–Theorem 2.1.3 (bounded operators with canonical representation
P
j λj |ϕ j ihψj ||). Let {ϕj }j and {ψj }j be two orthonormal (not necessarily complete)
families of vectors in H and let {λj }j be a family of complex numbers, indexed by the
same set {j}. The sequence
{ λj |ϕj ihψj |ξi }j = { λj (ψj , ξ)ϕj }j
is summable for every ξ ∈ H iff {λj }j is bounded. Whenever this happens, the sequence
P
of operators {λj |ϕj ihψj |}j sums in L(H) to an operator denoted by j λj |ϕj ihψj |, with
norm °X °
° °
° λj |ϕj ihψj |° = sup |λj | . (2.10)
j j

In the usual notation X X


λj |ϕj ihψj | = λj (ψj , · ) ϕj . (2.11)
j j
Properties:
¬ One has X
λj |ϕj ihψj | = O ⇔ λj = 0 ∀ j (2.12)
j
whence X X
λj |ϕj ihψj | = µj |ϕj ihψj | ⇔ λj = µj ∀j . (2.13)
j j

­ If λj 6= 0 ∀ j, then
X
λj |ϕj ihψj | is injective ⇔ {ψj }j is complete
j
X (2.14)
λj |ϕj ihψj | has dense range ⇔ {ϕj }j is complete .
j
P
® Let S = j λj |ϕj ihψj | as above; then
X
S∗ = λ̄j |ψj ihϕj |
j
X

S S= |λj |2 |ψj ihψj | (2.15)
j
X
|S| = |λj | |ψj ihψj |
j
26 Chapter 2. Mathematical preliminaries

P
¯ S ∈ L(H) is an orthogonal projection iff S = j |ϕj ihϕj |, its range being the
subspace spanned by {ϕj }j .
P
° S ∈ L(H) is unitary iff S = j |ϕj ihψj | where both {ϕj }j and {ψj }j are orthonor-
mal basis.
P
± S ∈ L(H) is a partial isometry iff S = j |ϕj ihψj |, where {ϕj }j and {ψj }j are
not necessarily complete, and in this case the projections onto the initial and final
P P
space are S ∗ S = j |ψj ihψj | and SS ∗ = j |ϕj ihϕj |, respectively.
P
² A given S = j λj |ϕj ihψj | is invertible iff both {ϕj }j and {ψj }j are orthonormal
basis and 0 < c1 < |λj | < c2 < +∞ for some positive constant c1 , c2 and ∀ j; in
P
this case S −1 = j λ1j |ψj ihϕj |.

Theorem 2.1.4 (compacts). Let C ∈ L(H).

¬ C ∈ Com(H) iff it admits the following so called canonical form:


X
C= λj |ϕj ihψj |
j
(2.16)
{ϕj }j and {ψj }j orthonormal systems
λj > 0 ∀ j .

Series in (2.16) converges in operator norm. Both the orthonormal systems are not
necessarily complete in H, so that the j’s run on a finite or countably infinite set
of integers; in this last case, necessarily λj → 0. The λj ’s are called the singular
values of C. By (2.15), {λj }j is precisely the set of nonzero eigenvalues of |C|.

­ C ∈ {finite rank operators} iff for some positive integer m


m
X
C= λj |ϕj ihψj |
j=1
(2.17)
{ϕj }m m
j=1 and {ψj }j=1 orthonormal systems
λj > 0 ∀ j = 1, . . . , m .

® Let, in addition, C = C ∗ ∈ Com(H). Then there exists an orthonormal basis


{ϕj }∞j=1 of H of eigenvectors for C (i.e., Cϕj = λj ϕj ). Its nonzero eigenvalues have
finite multiplicity and λj → 0 as j → ∞. In Dirac notation

X
C= λj |ϕj ihϕj | . (2.18)
j=1

(This is the Hilbert-Schmidt theorem.)

Theorem 2.1.5 (trace class and compacts). ¬ ∀ ϕ, ψ ∈ H, |ϕihψ| ∈ L1 (H) and

Tr[ |ϕihψ| ] = (ψ, ϕ)


(2.19)
k |ϕihψ| kL1 = kϕk kψk .
2.2. Trace of kernels 27

P
­ Let T ∈ Com(H) and let T = j λj |ϕj ihψj | be its canonical form. Then T ∈ L1 (H)
P
iff j λj < +∞. If so,
X
Tr[T ] = λj (ψj , ϕj )
j
X (2.20)
kT kL1 = Tr|T | = λj .
j

With the background presented so far one introduces the main mathematical object
of this work.

Definition 2.1.6. A density matrix is a positive trace class operator on H with trace
one.

As a consequence of theorems 2.1.3, 2.1.4, and 2.1.5, one has the following.

Definition–Theorem 2.1.7 (pure and mixed states). ¬ Any density matrix γ


on H can be written in diagonal form as
X X
γ= λj |ψj ihψj | , λj > 0 , λj = 1 , (2.21)
j j

where {ψj }j is an orthonormal system of H and {λj }j is the corresponding sequence


of eigenvalues.

­ When in (2.21) at least two nonzero λj ’s enter, then γ is said to be a mixed state.
If so,
X
γ2 = λ2j |ψj ihψj |
j
X (2.22)
2
Tr[γ ] = λ2j < 1 .
j

® Conversely, if γ has only one nonzero eigenvalue, which is necessarily 1, then it


takes the form of the rank-one orthogonal projection γ = |ψihψ| = γ ∗ = γ 2 and
it is said a pure state. For pure states Tr[γ 2 ] = 1. Thus, there is a one-to-one
correspondence between pure states and rays of H, where by ray {ψ} one defines
the subspace generated by ψ ∈ H.

2.2 Trace of kernels


In the standard physical applications, the Hilbert space is H = L2 (Rn ) (or L2 (Ω) for
some measure space Ω), and the nature of density matrices as operators with an integral
kernel arises naturally. The perspective is then to regard density matrices as special
Hilbert-Schmidt operators. In this section Hilbert-Schmidt operators are reviewed as
kernel operators and the notion of trace is given in terms of kernels.

Definition–Theorem 2.2.1 (Hilbert-Schmidt operators). The Hilbert-Schmidt


class is the space
L2 (H) := {S ∈ L(H) : Tr[S ∗ S] < +∞} (2.23)
that is, S is Hilbert-Schmidt iff S ∗ S is trace class. Properties:
28 Chapter 2. Mathematical preliminaries

¬ L2 (H) is a ∗-ideal of L(H). It is closed under the usual Banach topology of L(H)
iff dim H < ∞.

­ T ∈ L1 (H) iff T = S1 S2 with S1 , S2 ∈ L2 (H). Thus, the product of two Hilbert-


Schmidt is traceable. When endowed with the scalar product

(S1 , S2 )L2 := Tr[S1∗ S2 ] , (2.24)

then L2 (H) is a Hilbert space. In fact, it is a Banach ∗-algebra with respect to the
1/2
inherited norm kSkL2 = (S, S)L2 . One has

(S2 , S1 )L2 = (S1 , S2 )L2


(RS1 , S2 )L2 = (S1 , R∗ S2 )L2
(2.25)
(S1 R, S2 )L2 = (S1 , S2 R∗ )L2
kRSkL2 6 kRk kSkL2

∀ R ∈ L(H) and ∀ S, S1 , S2 ∈ L2 (H). Further,

kT k 6 kT kL2 6 kT kL1 ∀ T ∈ L1 (H) (2.26)

and
{finite rank operators} ⊂ L1 (H) ⊂ L2 (H) ⊂ Com(H) (2.27)
where each possible inclusion of any space into any other space of the chain is dense
with respect to the topology of the larger space.

® Let S ∈ L(H). For any two orthonormal basis {ϕj }j , {ψj }j of H, the sequences

{kSϕj k2 }j , {kSψj k2 }j , {|(ϕi , Sψj )|2 }i,j

are simultaneously summable or not. S ∈ L2 (H) iff they are summable; if so,
X X X
kSk2L2 = kSϕj k2 = kSψj k2 = |(ϕi , Sψj )|2 . (2.28)
j j i,j

P
¯ Let S ∈ Com(H) and let S = j λj |ϕj ihψj | be its canonical form. Then S ∈ L2 (H)
P
iff j λ2j < +∞. If so,
X
kSk2L2 = Tr[S ∗ S] = λ2j . (2.29)
j

In particular, ∀ ϕ, ϕ0 , ψ, ψ 0 ∈ H,

( |ϕihψ| , |ϕ0 ihψ 0 | )L2 = (ϕ, ϕ0 ) · (ψ 0 , ψ)


(2.30)
k |ϕihψ| kL2 = kϕk kψk .

Definition–Theorem 2.2.2 (integral kernels). Let (M, dµ) be a measure space.


R
¬ Let K ∈ L2 (M × M, dµ ⊗ dµ). Then the integral M K(x, y)ϕ(y) dy is defined for
µ-almost all x and represents a function in L2 (M, dµ).
2.2. Trace of kernels 29

­ Let H = L2 (M, dµ) and S ∈ L(H). Then S ∈ L2 (H) iff ∃ KS ∈ L2 (M ×M, dµ⊗dµ)
such that
Z
(Sϕ)(x) = KS (x, y)ϕ(y) dµ(y) ∀ ϕ ∈ L2 (M, dµ) . (2.31)
M

If so, KS is said the integral kernel of S and

kSkL2 (H) = kKS kL2 (M ×M,dµ⊗dµ) . (2.32)

Also,
KcS+c0 S 0 = cKS + c0 KS 0 ∀ c, c0 ∈ C , ∀ S, S 0 ∈ L2 (H) (2.33)
and the map

L2 (M × M, dµ ⊗ dµ) −→ L2 (H)
(2.34)
KS 7−→ S

is an isomorphism of Banach spaces.


R
® If K1 , K2 ∈ L2 (M × M, dµ ⊗ dµ) and K(x, y) := M K1 (x, z)K2 (z, y) dµ(y), then
K ∈ L2 (M × M, dµ ⊗ dµ) as well. L2 (M × M, dµ ⊗ dµ), when endowed with the
•-product Z
(K1 • K2 )(x, y) := K1 (x, z)K2 (z, y) dµ(z) , (2.35)
M
is a Banach algebra. It is also a Banach-∗ algebra, when endowed with the involu-
tion such that K(x, y) 7→ K(y, x).

¯ If S, S1 , S2 ∈ L2 (H), then

KS1 S2 = KS1 • KS2 (µ ⊗ µ)-a.e.


(2.36)
KS ∗ (x, y) = KS (y, x) (µ ⊗ µ)-a.e.

Thus, (2.34) is also a ∗-isomorphism of Banach ∗-algebras. In particular,

S = S∗ ⇔ KS (y, x) = KS (x, y) (µ ⊗ µ)-a.e. (2.37)

° Let ϕ, ψ ∈ H: then the trace class operator |ϕihψ| has kernel ϕ(x)ψ(y). More
P
generally, let T ∈ L1 (H), with canonical form T = j λj |ϕj ihψj |, where λj > 0,
P
j λj = kT kL1 (H) , and {ϕj }j and {ψj }j orthonormal systems of H. Its kernel is
X
KT (x, y) = λj ϕj (x)ψj (y) , (2.38)
j

the expansion above converging in L2 (M × M, dµ ⊗ dµ).

When the Hilbert space is realized as a L2 -space, it is useful to get the trace of a trace
class operator directly from its kernel, instead of summing its singular values. This is
customarily done by integrating the kernel along the diagonal, modulo the technicalities
discussed here below.
One starts from the following classical result.

Theorem 2.2.3. Let H = L2 (M, dµ) as above and T ∈ L1 (H).


30 Chapter 2. Mathematical preliminaries

¬ Let S1 , S2 ∈ L2 (H) such that T = S1 S2 . In terms of the corresponding kernels,


KT = KS1 • KS2 (µ ⊗ µ)-a.e. Then ([50])
Z
Tr[T ] = (KS1 • KS2 )(x, x) dµ(x)
ZM (2.39)
= KS1 (x, z)KS2 (z, x) dµ(x) dµ(z) .
M ×M

­ In particular, if KT is continuous,
Z
Tr[T ] = KT (x, x) dµ(x) . (2.40)
M

(Sometimes (2.40) goes under the name of Mercer’s theorem, at least when M =
[0, 1] with the Lebesgue measure [107].)

One would be tempted to plug KT (x, x) = (KS1 • KS2 )(x, x) into (2.39), thus proving
(2.40) in general. Yet, by (2.36), KT (x, y) = (KS1 • KS2 )(x, y) holds (µ ⊗ µ)-a.e. in
M × M and, in the absence of continuity for KT , there is no guarantee that KT and
KS1 • KS2 agree on the diagonal {(x, x) ∈ M × M }, which has zero (µ ⊗ µ)-measure. A
P
similar argument is the following. Write T in the canonical form T = j λj |ϕj ihψj |. In
P
terms of kernels, KT (x, y) = j λj ϕj (x)ψj (y) (µ ⊗ µ)-a.e. One has
X X Z X
Tr[T ] = (ψj , T ψj ) = λj (ψj , ϕj ) = λj ϕj (x)ψj (x) dµ(x) (2.41)
j j M j

P
and again, in the absence of continuity for KT , one cannot plug KT (x, x) = j λj ϕj (x)ψj (x)
in, in order to prove (2.40) in general.
The way out goes through the following remark. Since T is traceable,
Z X Z X X
2
λj |ϕ(x)| dx = λj |ψ(x)|2 dx = λj < +∞ ,
M j M j j
P P
so the series j λj |ϕ(x)|2 and j λj |ψ(x)|2 are finite µ-a.e. Due to
X¯ ¯ ³X ´1/2 ³ X ´1/2
¯ λj ϕj (x)ψj (y) ¯ 6 λj |ϕ(x)|2 λj |ψ(x)|2 ,
j j j

P
series (2.38) converges absolutely (µ ⊗ µ)-a.e. and j λj ϕj (x)ψj (x) converges absolutely
µ-a.e. This pointwise limit, in general, is not KT (x, x), since {(x, x) ∈ M × M } has
zero (µ ⊗ µ)-measure: denote it by K e T (x, x). It turns out that Ke T (x, x) is obtained
from KT (x, y) by an averaging process over cubes centered on the diagonal in M × M
and letting the side of the cubes go to zero. This can be regarded as a way of selecting
a “smooth” R pointwise representative for the corresponding kernel. Thus, (2.41) gives
Tr[T ] = M K e T (x, x) dµ(x). This analysis has been carried out in [22, 23]. In the
concrete case (M, dµ) = (Rn , dx) the result is as follows.

Theorem 2.2.4. Let H = L2 (Rn ) for some positive integer n. Let T ∈ L1 (H), with
integral kernel KT ∈ L2 (R2n ). Let Cr = [−r, r]n be the n-dimensional cube of side 2r
centred at the origin of Rn and let |Cr | be its Lebesgue measure.
2.3. Partial trace and reduced density matrices 31

¬ The limit Z
e T (x, y) := lim 1
K KT (x + ξ, y + η) dξ dη (2.42)
r→0 |Cr |2 Cr ×Cr
exists (dx ⊗ dx)-a.e. and
e T (x, y) = KT (x, y)
K (dx ⊗ dx)-a.e. (2.43)
Ke T is uniquely determined by KT and it agrees with KT at each point (x, y) of
continuity of the latter.
e T (x, x) exists dx-a.e.
­ K

® T has trace Z
Tr[T ] = e T (x, x) dx .
K (2.44)
M

¯ Let S1 , S2 ∈ L2 (H) such that T = S1 S2 . Then


e T (x, x) = (KS • KS )(x, x)
K dx-a.e. (2.45)
1 2

P
° If, in addition, T > O and T = λj |ϕj ihϕj | is its canonical form, then
j
X
e T (x, x) =
K λj |ϕj (x)|2 dx-a.e. (2.46)
j

2.3 Partial trace and reduced density matrices


Let H and K be two separable Hilbert spaces. In the treatment of many-body systems
one systematically makes use of the operation that sends density matrices on H ⊗ K to
density matrices on H, whose definition and properties are collected in this section.
Definition–Theorem 2.3.1 (partial trace). Let H and K be two separable Hilbert
spaces. The partial trace with respect to K is the linear map
L1 (H ⊗ K) −→ L1 (H)
(2.47)
T 7−→ TrK [ T ]
such that, for an orthonormal basis {ξj }j of K,
X
(ϕ, TrK [ T ]ψ)H = (ϕ ⊗ ξj , T ψ ⊗ ξj )H⊗K ∀ ϕ, ψ ∈ H . (2.48)
j

One proves that if (2.48) holds, this happens independently of the basis {ξj }j . Equiva-
lently, TrK [ T ] is characterized by
£ ¤ £ ¤
Tr TrK [ T ] · A = Tr T · (A ⊗ 1K ) ] ∀ A ∈ L(H) (2.49)
where trace in the l.h.s. is on L1 (H), while trace in the r.h.s. is on L1 (H ⊗K). Properties:
Tr[ TrK [ T ] ] = Tr[ T ] ∀ T ∈ L1 (H) (2.50)
1
T > O ⇒ TrK [ T ] > O ∀ T ∈ L (H) (2.51)
1 1
TrK [ T1 ⊗ T2 ] = T1 · Tr[ T2 ] ∀ T1 ∈ L (H) , ∀ T2 ∈ L (K) (2.52)
In particular, by (2.50) and (2.51),
γ is a density matrix on H ⊗ K ⇒ TrK [ γ ] is a density matrix on H . (2.53)
32 Chapter 2. Mathematical preliminaries

In passing, we quote an alternative, algorithmic definition of partial trace, which


turns out to be equivalent to the previous one and is customarily used for finite rank
operators (matrices).

Theorem 2.3.2 (partial trace in terms of matrix elements). Let {ei }i and {fj }j be
any two orthonormal basis of the Hilbert spaces H and K respectively, so that {ei ⊗ fj }i,j
is an orthonormal basis of H ⊗ K. Let T ∈ L1 (H ⊗ K) and denote its matrix elements by

ti0,j 0 ;i,j := (ei0 ⊗ fj 0 , T ei ⊗ fj )H⊗K . (2.54)

Let Te := TrK [ T ] and denote its matrix elements by

t̃i0,i := (ei0 , TrK [ T ] ei )H . (2.55)

Consider the operators Ti0,i ∈ L(K) defined by

(fj 0 , Ti0,i fj )K := ti0,j 0 ;i,j . (2.56)

Then Ti0,i ∈ L1 (K) and


t̃i0,i = Tr[ Ti0,i ] (2.57)
(the last trace being computed on K). All this is independent of the choice of the basis.

When dim H = n < ∞ and dim K = m < ∞, last theorem reads as follows. Fixed
the orthonormal basis {ei }ni=1 and {fj }m
j=1 , T is the nm × nm matrix written in block
form as  
T1,1 T1,2 · · · T1,n
 T2,1 T2,2 · · · T2,n 
 
T = . .. . ..  (2.58)
 . . . . . . 
Tn,1 Tn,2 · · · Tn,n
in the basis {e1 ⊗ f1 , e1 ⊗ f2 , . . . , e1 ⊗ fm , e2 ⊗ f1 , . . . , en ⊗ fm−1 , en ⊗ fm } of H ⊗ K, where
the blocks Ti0,i ’s are in turn the m × m matrices
 
ti0,1;i,1 ti0,1;i,2 ··· ti0,1;i,m
 ti0,2;i,1 ti0,2;i,2 ··· ti0,2;i,m 
 
Ti0,i = . .. .. ..  . (2.59)
 .. . . . 
ti0,m;i,1 ti0,m;i,2 · · · ti0,m;i,m

Then in the basis {ei }ni=1 , TrK [ T ] is the n × n matrix


 
t̃1,1 t̃1,2 · · · t̃1,n
 t̃2,1 t̃2,2 · · · t̃2,n  m
X
 
TrK [ T ] =  . .. .. ..  with t̃i0,i = Tr[ Ti0,i ] = ti0,`;i,` . (2.60)
 .. . . .  `=1
t̃n,1 t̃n,2 · · · t̃n,n

When the Hilbert space under consideration is realized as L2 (Rn ), the operation of
partial trace takes a peculiar form in terms of kernels. Let us describe it in the case of
density matrices, being the objects of interest in the sequel. In fact, (2.53) makes the
following definition well-posed.
2.3. Partial trace and reduced density matrices 33

Definition 2.3.3. Let H and K be two separable Hilbert spaces and let γ be a density
matrix on H ⊗ K. The density matrix γ (H) := TrK [ γ ] is called reduced density
matrix (or marginal) of γ relative to H taken tracing out with respect to K.

Theorem 2.3.4. Let H = L2 (Rn ) and K = L2 (Rm ) for some positive integers n, m,
where the Lebesgue measure is understood. Let γ be a density matrix on H ⊗ K ∼ =
2
L (R n+m ). Denote its kernel by γ(x, x ; y, y ), where x and y run in R and x and y 0
0 0 n 0

run in R . Then γ (H) = TrK [ γ ] has kernel


m

Z
(H)
γ (x, y) = γ(x, z 0 ; y, z 0 ) dz 0 (2.61)
Rm

the integral being defined with the same technique as in theorem 2.2.4.

Equation (2.61) is actually how the reduced density matrix was introduced first by
Husimi [60].
To conclude this section, we point out that this construction is realized precisely in
view of the key operation

L(H) 3 O∗ = O 7−→ Tr[ γ O ] ∈ R , (2.62)

that in Quantum Mechanics is interpreted as the evaluation of the observable O (a


bounded selfadjoint operator on H) in the state γ (a density matrix on H) – see the
discussion at the beginning of Sec. 3.1.
When H = L2 (Rn ), then Tr[ γ O ] is given tracing the kernel of γ O ∈ L1 (H), accord-
ing to theorems 2.2.3, 2.2.4. In particular, if O ∈ L2 (H), with kernel O( · , · ) ∈ L2 (R2n ),
then a straightforward application of (2.35) and (2.40) gives
Z
Tr[ γ O ] = γ(x, y)O(y, x) dx dy . (2.63)
R2n

If γ is the pure state γ = |ψihψ|, this becomes


Z
Tr[ |ψihψ| O ] = (ψ, O ψ) = ψ(x)O(x, y)ψ(y) dx dy . (2.64)
R2n

Now, a typical physical situation is when the Hilbert space is isomorphic to H ⊗ K with
H = L2 (Rn ), K = L2 (Rm ) and the observables of interest act non-trivially only on the
subspace H ⊗ 1K , i.e., have the form O ⊗ 1K with O = O∗ ∈ L(H). Notice that such a
O⊗ 1K cannot be Hilbert-Schmidt and, hence, cannot have a L2 (R2(n+m) )-kernel, because

Tr[ (O ⊗ 1K )∗ (O ⊗ 1K ) ] = Tr[ O∗ O ] · Tr[1K ] = +∞ (2.65)

even if O is Hilbert-Schmidt. Yet, such a O ⊗ 1K is recognised to have a kernel-like action


whenever O ∈ L2 (H). Indeed, since on vectors ϕ ⊗ ϕ0 ∈ H ⊗ K it acts as

(O ⊗ 1K )(ϕ ⊗ ϕ0 ) = Oϕ ⊗ ϕ0 ,

then it is easily seen to act on a generic wave function ψ ∈ H ⊗ K ∼ = L2 (Rn+m ) as


Z
ψ(x; x0 ) 7−→ 0 0 0
n O(x, y) δ(x − y ) ψ(y; y ) dy dy
0
(2.66)
y∈R
y 0 ∈Rm
34 Chapter 2. Mathematical preliminaries

namely, with integral kernel

KO⊗1K (x, x0 ; y, y 0 ) = O(x, y) δ(x0 − y 0 ) . (2.67)

Thus, by general formula (2.63), the expectation of the observable O ⊗ 1K in the state γ
on H ⊗ K is given in terms of kernels by
Z
0 0 0 0 0 0
Tr[ γ (O ⊗ 1K ) ] = n γ(x, x ; y, y ) KO⊗1K (y, y ; x, x ) dx dx dy dy
x,y ∈R
x0 ,y 0 ∈Rm
Z
= γ(x, x0 ; y, y 0 ) O(y, x) δ(y 0 − x0 ) dx dx0 dy dy 0
x,y ∈Rn
x0 ,y 0 ∈Rm (2.68)
Z
= γ (H) (x, y) O(y, x) dx dy
x,y ∈Rn
(H)
= Tr[ γ O]
where Z
(H)
γ (x, y) := γ(x, z 0 ; y, z 0 ) dz 0 (2.69)
Rm
is the kernel of the reduced density matrix γ (H) – compare (2.68) with (2.49).
Of course, theorem 2.3.4 and the subsequent considerations can be easily restated
and hold true also when H = L2 (Ω) and K = L2 (Λ) for some domains Ω ⊂ Rn and
Λ ⊂ L2 (Rm ).

2.4 Distances of states and of reduced density matrices


This is a detour section where we prepare some results the application of which, in the
context of B.E.C., will be not discussed until Sec. 7.2. Due to their generality, it is
preferable to present them next to the previous sections which deals with the general
theory. The purpose is to compare distances among states with distances among the
corresponding marginals.
First, we focus on pure states. This is to point out that, although there is a one-to-
one correspondence between pure states and rays of the Hilbert space, yet some care is
needed when, to measure distances among two pure states, one chooses trace-norm for
density matrices or vector-norm for the representatives of the rays.
Theorem 2.4.1. Let H be a separable Hilbert space. Let ψ1 , ψ2 ∈ H with kψ1 k =
kψ2 k = 1 and set γi := |ψi ihψi | for i = 1, 2. Then
¡ ¢
k ψ1 − ψ2 k2H = 2 1 − Re(ψ1 , ψ2 ) (2.70)
2
¡ 2
¢
k γ1 − γ2 k 1 = 4 1 − |(ψ1 , ψ2 )| (2.71)
L (H)

k γ1 − γ2 kL1 (H) 6 2 k ψ1 − ψ2 kH (2.72)

and the constant in (2.72) is optimal.


Proof. (2.70) is obvious.
To prove (2.71) notice first that γ1 − γ2 acts non-trivially only on the subspace
generated¯ by ψ1 , ψ2 . So ⊥
¯ set H2 := SpanC {ψ1 , ψ2 }, whence H = H2 ⊕ H2 , γ1 − γ2 =
(γ1 − γ2 )¯H2 , (γ1 − γ2 )¯H⊥ = O. Moreover, notice that
2
¡ ¢
(γ1 − γ2 )2 = 1 − |(ψ1 , ψ2 )|2 1 , (2.73)
2.4. Distances of states and of reduced density matrices 35

p
thus, |γ1 − γ2 | = (γ1 − γ2 )2 in diagonal form reads
¯ µ ¶
¯ ¯ ¯¯ p 1 0
¯
¯(γ1 − γ2 ) H2 ¯ = 1 − |(ψ1 , ψ2 )|2 . (2.74)
0 1

This can be seen by direct computation. Indeed

(γ1 − γ2 )2 = ( |ψ1 ihψ1 | − |ψ2 ihψ2 | ) · ( |ψ1 ihψ1 | − |ψ2 ihψ2 | )


(2.75)
= |ψ1 ihψ1 | + |ψ2 ihψ2 | − (ψ1 , ψ2 )|ψ1 ihψ2 | − (ψ2 , ψ1 )|ψ2 ihψ1 | ,

then, in case |(ψ1 , ψ2 )| < 1 (while (2.74) is trivially true if |(ψ1 , ψ2 )| = 1), one expands
ψ ∈ H as X
ψ = a1 ψ1 + a2 ψ2 + ai ψi (2.76)
i>3

where {ψ3 , ψ4 , . . . } is an orthonormal basis of H2⊥ . Applying (2.75) to (2.76) one gets
¡ ¢
(γ1 − γ2 )2 ψ = 1 − |(ψ1 , ψ2 )|2 ψ . (2.77)

Alternatively, orthonormalisation of the basis {ψ1 , ψ2 } of H2 gives

ϕ1 = ψ1
ψ2 − (ψ1 , ψ2 ) ψ2 (2.78)
ϕ2 =
kψ2 − (ψ1 , ψ2 ) ψ2 k

and in the new basis {ϕ1 , ϕ2 } the two projections read


µ ¶
¯ 1 0
γ1 ¯
H2
=
0 0
 p  (2.79)
¯ |(ψ1 , ψ2 )|2 (ψ1 , ψ2 ) 1 − |(ψ1 , ψ2 )|2
γ2 ¯
H2
=  p
 .
(ψ1 , ψ2 ) 1 − |(ψ1 , ψ2 )|2 1 − |(ψ1 , ψ2 )|

Computing the eigenvalues λ1 , λ2 of (γ1 − γ2 )2 in this basis gives

λ1 = λ2 = 1 − |(ψ1 , ψ2 )|2 , (2.80)


p
thus, |γ1 − γ2 | = (γ1 − γ2 )2 in diagonal form is exactly (2.74). As a consequence,
p
k γ1 − γ2 kL1 (H) = Tr|γ1 − γ2 | = 2 1 − |(ψ1 , ψ2 )|2 , (2.81)

which proves (2.71).


To prove (2.72), look for an optimal α > 0 such that

k γ1 − γ2 kL1 (H) 6 α k ψ1 − ψ2 k . (2.82)

By plugging (2.70) and (2.71) into (2.82) squared, and setting x := Re(ψ1 , ψ2 ) and
y := Im(ψ1 , ψ2 ), one has

α2 α2
fα (x, y) := x2 − x+ − 1 + y2 > 0 (2.83)
2 2
36 Chapter 2. Mathematical preliminaries

to be minimized on the circle C = {x2 + y 2 6 1} of R2 . Since


( 2
( 2
0 = ∂x fα (x, y) = 2x − α2 x = α4
⇒ (2.84)
0 = ∂y fα (x, y) = 2y y= 0
2 2 2
then, if α < 2, fα reaches its minimum in ( α4 , 0) ∈ C; yet, fα ( α4 , 0) = −1 + α4 < 0, then
there exist norm-one vectors ψe1 , ψe2 ∈ H such that the corresponding projections violates
2
(2.82), that is, k γ e2 kL1 (H) > α k ψe1 − ψe2 k. On the other hand, if α > 2 then ( α4 , 0)
e1 − γ
does not belong to the interior of C, and fα must achieve its minimum in the boundary
∂C, actually in the point (1, 0):
¯ p α2
fα (x, y)¯∂C = fα (x, ± 1 − x2 ) = (1 − x) > fα (1, 0) = 0 . (2.85)
2
So the optimal α in order (2.83) to be satisfied ∀ (x, y) ∈ C is α = 2.

Notice, in passing, that a non optimal bound like (2.72) could be easily obtained in
the following direct way. First, since ∀ ϕ ∈ H with kϕk = 1

k (ψ1 , ϕ) ψ1 − (ψ2 , ϕ) ψ2 k 6 k (ψ1 , ϕ) ψ1 − (ψ2 , ϕ) ψ1 k + k (ψ2 , ϕ) ψ1 − (ψ2 , ϕ) ψ2 k


= | (ψ1 , ϕ) − (ψ2 , ϕ) | · kψ1 k + | (ψ2 , ϕ) | · k ψ1 − ψ2 k (2.86)
6 2 k ψ1 − ψ2 k ,

then
k γ1 − γ2 kL(H) = sup k (ψ1 , ϕ) ψ1 − (ψ2 , ϕ) ψ2 k 6 2 k ψ1 − ψ2 k . (2.87)
kϕk=1

Next observe that


p p p p p
k γ1 − γ2 kL1 (H) = λ1 + λ2 6 2 max{ λ1 , λ2 } = 2 max{λ1 , λ2 }
q q
= 2 k (γ1 − γ2 )2 kL(H) 6 2 k γ1 − γ2 k2L(H) (2.88)

= 2 k γ1 − γ2 kL(H)

whence
k γ1 − γ2 kL1 (H) 6 4 k ψ1 − ψ2 k . (2.89)
Yet (2.89) provides a non-sharp constant.
By theorem 2.4.1, if two unit vectors in H are close, so are the corresponding pro-
jections. The converse is false. For, if e.g. ψ2 = −ψ1 then k γ1 − γ2 kL1 (H) = 0 while

k ψ1 − ψ2 k = 2. Actually close rank-one projections (i.e., pure states) correspond to
close rays, instead of representatives, each unit vector in H determining a ray up to a
phase that cancels in the corresponding projection. This is the content of the following
definition 2.4.2 and theorem 2.4.3.

Definition–Theorem 2.4.2 (distance of rays). Let H be a separable Hilbert space


and denote by {ψ} a ray, where ψ ∈ H is any norm-one representative of it. In terms of
norm-one representatives, the two rays coincide and one writes {ψ1 } = {ψ2 }, iff ψ1 and
ψ2 are the same vector in H up to a phase. Define the distance between two rays to be
the quantity ¡ ¢ ° °
d {ψ1 }, {ψ2 } := min ° ψ1 − eiδ ψ2 ° . (2.90)
δ∈[0,2π] H
2.4. Distances of states and of reduced density matrices 37

The min is well-posed, since the function δ 7→ k ψ1 − eiδ ψ2 kH is clearly continuous on


the compact [0, 2π]. This is a true distance making the set of rays on H a metric space:
¡ ¢
d ¡{ψ1 }, {ψ2 }¢ = 0 ¡ ⇔ {ψ¢1 } = {ψ2 }
d ¡{ψ1 }, {ψ2 }¢ = d ¡{ψ2 }, {ψ1 }¢ (2.91)
¡ ¢
d {ψ1 }, {ψ2 } 6 d {ψ1 }, {ψ3 } + d {ψ3 }, {ψ2 } .

Theorem 2.4.3 (close pure state projections ⇒ close rays). Let H be a separable
Hilbert space, let ψ1 , ψ2 ∈ H with kψ1 k = kψ2 k = 1, representing the rays {ψ1 }, {ψ2 },
and set γi := |ψi ihψi | for i = 1, 2. Then
¡ ¢ 1
d {ψ1 }, {ψ2 } 6 √ k γ1 − γ2 kL1 (H) (2.92)
2
the constant being optimal.

Proof. One looks for an optimal β > 0 such that


¡ ¢
d {ψ1 }, {ψ2 } 6 β k γ1 − γ2 kL1 (H) . (2.93)

Notice preliminarily that with no loss of generality one can change ψ2 by a phase eiη so
that (ψ1 , ψ2 ) is real : indeed both kγ1 − γ2 kL1 (H) and d ({ψ1 }, {ψ2 }) are left unchanged
by (2.71) and (2.90), respectively. So, by (2.70),
£ ¡ ¢ ¤2 ° °2
d {ψ1 }, {ψ2 } = min ° ψ1 − eiδ ψ2 °
δ∈[0,2π] H
¡ iδ
¢
= 2 min 1 − Re [e (ψ1 , ψ2 )]
δ∈[0,2π] (2.94)
= 2 min (1 − x cos δ)
δ∈[0,2π]

= 2 (1 − x sign(x))

while, by (2.71),
k γ1 − γ2 k2 1 = 4 (1 − x2 ) (2.95)
L (H)

with x := Re(ψ1 , ψ2 ) = (ψ1 , ψ2 ) ∈ [−1, 1]. It is then clear that

2 (1 − x sign(x)) 6 4 β 2 (1 − x2 ) 1
⇒ β>√ (2.96)
∀ x ∈ [−1, 1] 2

and β = 1/ 2 is the sharp constant.

Next, we come to the main point: partial trace of density matrices are close if the
density matrices themselves are, the converse being false. In Sec. 7.2 this will be used
to claim that if two N -body states are close, then the same holds for the corresponding
one-particle reduced density matrices. Distance between the two states one starts with
is meant to be the trace-norm of the difference of density matrices; if, in particular, one
deals with pure states, then one indifferently means closeness of vectors, rays or density
matrices in the spirit of theorems 2.4.1 and 2.4.3.

Theorem 2.4.4. Let H and K be separable Hilbert spaces. Then, ∀ T ∈ L1 (H ⊗ K),


° °
° Tr [ T ] ° 1 6 k T kL1 (H⊗K) . (2.97)
K L (H)
38 Chapter 2. Mathematical preliminaries

Proof. From one side, recall from theorem 2.1.1, point °, that by the canonical isometric
isomorphism (L1 (H), k · kL1 ) = (Com(H), k · kL )∗ one has
¯ ¯
° ° ¯ Tr [ Tr [ T ] · C ] ¯
° Tr [ T ] ° 1 = sup K
. (2.98)
K L (H) kCkL(H)
C∈Com(H)
C6=O

From the other side, by characterization (2.49) of partial trace, and by (2.6),
¯ ¯ ¯ ¯
¯ Tr [ Tr [ T ] · C ] ¯ = ¯ Tr [ T · (C ⊗ 1K ) ] ¯ 6 kCkL(H) · k T kL1 (H⊗K) . (2.99)
K

So (2.97) is proved.

One can alternatively prove the key inequality (2.97) by controlling partial trace in
terms of matrix elements, according to theorem 2.3.2. This gives an explicit indication
of how the inequality can be strict – see (2.103).

Alternative proof of theorem 2.4.4. Diagonalize T so that in its block form (2.58) each
block (2.59) is
 (i) 
λ1 0 0 ···
 
 0 λ(i)
2 0 · · ·
Ti0,i = δi0,i 
 0
 (2.100)
λ3 · · ·
(i)
 0 
.. .. .. ..
. . . .
and X X ¯ (i) ¯
k T kL1 (H⊗K) = ¯λ ¯. (2.101)
j
i j

By (2.60)
 
µ1 0 · · ·
  X (i)
TrK [ T ] =  0 µ2 · · · , µi = Tr[ Ti,i ] = λj , (2.102)
.. .. . . j
. . .
and X X X ¯ (i) ¯
° °
° Tr [ T ] ° 1 = |µi | 6 ¯ λ ¯ = k T kL1 (H⊗K) ¤ (2.103)
K L (H) j
i i j

Our main application of theorem 2.4.4 (see theorem 7.2.3) will be in the form
° ° ° °
° Tr [ γ1 ] − Tr [ γ2 ] ° 1 6 ° γ1 − γ2 °L1 (H⊗K) (2.104)
K K L (H)

for some states γ1 , γ2 on H ⊗ K. (2.104) follows from (2.97) by linearity of TrK . In case
of pure states it will be useful to use (2.104) in the form
° °
° Tr [ γ1 ] − Tr [ γ2 ] ° 1 6 2 k ψ1 − ψ2 kH⊗K (2.105)
K K L (H)

To conclude this section, let us quote a useful inequality for the sequel, which allows
to control the distance between two pure states in terms of the distance between the first
one and an unnormalised version of the second.
2.4. Distances of states and of reduced density matrices 39

Theorem 2.4.5. Let ψ, ϕ ∈ H, with kψk = 1 and kψ − ϕk < 1. Then


° ϕ ° 2 kψ − ϕk
° °
°ψ − °6 . (2.106)
kϕk 1 − kψ − ϕk
¯ ¯
Proof. Since ¯ kψk− kϕk ¯ 6 kψ −ϕk and, consequently, 1−kψ −ϕk 6 kϕk 6 1+kψ −ϕk,
then
° ° ° ° °
° ϕ °° ° kϕk ψ − ϕ ° ° ( kϕk − 1 ) ψ + (ψ − ϕ) °
°ψ − °= =
kϕk kϕk kϕk
¯ ¯ (2.107)
¯ kϕk − kψk ¯ · kψk + kψ − ϕk 2 kψ − ϕk
6 6 . ¤
1 − kψ − ϕk 1 − kψ − ϕk
40 Chapter 2. Mathematical preliminaries
Chapter 3

Formalisation of Bose-Einstein
condensation

The mathematical investigation on B.E.C. this work is dealing with relies on a suit-
able rigorous definition of such a physical phenomenon, that is, a formalisation which is
mathematically well-posed, has a precise physical interpretation in terms of first princi-
ples, and is experimentally acceptable in the sense that it provides a clear signature of
condensation to be checked in the laboratory. Depending on the point one stresses (math-
ematics, theoretical interpretation, or experimental perspective), a complex of definitions
of B.E.C. has been established in the literature since Bose and Einstein’s “theoretical
discover”.
Following Ref. [83, 84], this chapter focuses on the appropriate mathematical tools
leading to a well-posed and physically meaningful formalisation of B.E.C., while the
experimental scenario, which in the last 12 years has become more and more rich, has
been already briefly sketched in the warm up of Chap. 1. Here the discussion is based
on a definition of B.E.C. which has been essentially well-established since half a century,
but which still deserves an appropriate interpretation.
Along this line, Sec. 3.1 and 3.2 provide the mathematical framework for the descrip-
tion of large many-body boson systems. The bosonic sector of a Hilbert space will be
introduced and described with suitable quantum labels. The need and the meaning of a
limit of infinite number of particle will be discussed, bridging the mathematical treatment
with its physical interpretation. In particular, stress will be put on the scaling nature of
any physically reasonable and mathematically well-posed limit of infinite systems.
The choice here is to place all the discussion within the standard quantum mechanical
picture, that is, without appealing to the formalism of second quantization. Both being
formalisms, the essence of the Physics behind is untouched. Moreover, both are standard
textbooks subjects. Admittedly, the second quantization would definitely avoid some
unnecessary complications in the notation, for instance, the emergence of combinatorial
factors or the use of labels for vectors in the many-body Hilbert space (the use of the
number operator N (ϕ) = a∗ (ϕ)a(ϕ) for one-particle state ϕ already does the job). On
the other side, most of the mathematical results and techniques presented in this chapter
and in the following ones do not involve second quantization or, at least, most of them
has been originally stated and derived in the Schrödinger formalism, and this motivates
the present choice.
In Sec. 3.3 and 3.4 we will revisit the classical Penrose and Osanger’s definition

41
42 Chapter 3. Formalisation of Bose-Einstein condensation

within our current formalism for an asymptotic analysis, i.e., in the limit of infinitely
many particles. Emphasis will be given to the crucial role that reduced density matrices
have in expressing condensation: in fact, the distribution of particles one observes in
the experiments is nothing but the diagonal part of the kernel of the one-body reduced
density matrix. To control (asymptotically) 100% B.E.C., such one-body marginal has
to approximate the projection onto the one-body condensate wave function: convergence
can be set in several natural but distinct topologies and we accounts for the equivalence
of a number of them.
Then in Sec. 3.5 we report two recent proofs of B.E.C. in the sense of the preceding
discussion, in the Gross-Pitaevskiı̆ scaling limit for a dilute system of weakly interacting
spinless bosons.

3.1 Many-body quantum systems of identical bosons


Notoriously, the mathematical apparatus outlined in Chap. 2 is the standard tool at the
basis of the description of a physical system in Quantum Mechanics with a finite number
of degrees of freedom. States of the systems are density matrices on some Hilbert space
H: pure states are rays of H, while mixed states are density matrices with rank > 2.
Observables are (possibly unbounded) selfadjoint operators on H and the expectation
value of an observable O in the state γ is the real number Tr[ γ O ]. Dynamics is given
by a strongly continuous one-parameter group {e−itH/~ }t∈R of unitary operators on H
generated by the selfadjoint H, the Hamiltonian of the system: a state γ evolves into
γt = e−itH/~ γ eitH/~ . In particular, a pure state ψ evolves into ψt = e−itH/~ ψ (the
infinitesimal version of which, is the Schrödinger equation), ψ being any norm 1 vector
in H representing the ray {ψ}.
Here the interest is towards special many-body quantum systems: systems of N
undistinguishable quantum particles. Thus, the building block is the Hilbert space H
of a single particle, while the N -body Hilbert space is HN := H⊗N . Actually the Spin-
Statistics theorem, which is a postulate in this non-relativistic setting, prescribes that
boson states in HN are symmetric with respect to permutation of the particles, while
fermion states are antisymmetric. In this section we will recall and develop the tools for
the study of the bosonic case, which is the one we are going to consider in the sequel
concerning Bose-Einstein condensation.

Notation. By H one will denote the single-particle Hilbert space. In view of our
next discussion on spinless particles, we will systematically choose H = L2 (Ω) for some
Ω ⊂ Rd : here Ω is the region of the d-dimensional Euclidean space where each particle
is allowed to stay (d = 1, 2, 3, for the cases of interest), be it the region of the laboratory
where an alkali gas is trapped and cooled or the region of space occupied by a star of
bosons. Ω will be chosen to be suitably regular, possibly the whole Rd . Thus, each
quantum particle is described by an infinite-dimensional Hilbert space. Yet the very
same results that we are going to discuss do hold, with a slightly simplified notation,
even in the finite dimensional case H ∼ = Cn . By HN = H⊗N ∼ = L2 (ΩN ) one will denote
N N d
the N -body Hilbert space, where Ω = Ω×· · ·×Ω ⊂ R is the N -th Cartesian product.
Starting from any orthonormal basis of H, say {ϕj }j , one builds the orthonormal basis

{ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN }j1 ,j2 ,...,jN (3.1)


3.1. Many-body quantum systems of identical bosons 43

of HN . For shortness, vectors of such a particular collection will be referred to as the


basis vectors of HN , with no further specification. In this notation, the basis vector

ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN (3.2)

of HN is the state of N particles, labeled by k = 1, . . . , N , where the k-th particle is in


the one-body state ϕjk . Although in general this is a physically forbidden state, since it
violates undistiguishability, yet it has a precise mathematical meaning. Alternative form
to (3.2) is the self-explanatory notation |j1 , j2 , . . . , jN i, which is familiar in the physical
literature. All this is to say that in (3.2) the particle ordering is kept into account. As
an element of L2 (ΩN ), (3.2) is the wave function

(ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN )(x1 , x2 , . . . , xN ) = ϕj1 (x1 )ϕj2 (x2 ) · · · ϕjN (xN ) (3.3)

where each variable xk ∈ Rd runs in Ω and ϕjk ( · ) ∈ L2 (Ω). The generic element of HN
will be denoted by ΨN . Also, denote by SN the symmetric group of order N , i.e., the
group of permutations of N symbols.
Our aim is to focus on the subspace of the total Hilbert space containing the only
physically admissible states for a bosonic system.

Definition–Theorem 3.1.1 (symmetrization – I). Let σ ∈ SN . The permutation


operator relative to σ is the operator Pσ : HN −→ HN such that

(Pσ ΨN )(x1 , x2 , . . . , xN ) := ΨN (xσ(1) , xσ(2) , . . . , xσ(N ) ) , (3.4)

so that σ 7→ Pσ is the representation of the permutation group SN on HN . On the basis


wave functions, Pσ acts as
¡ ¢
Pσ (ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ) (x1 , x2 , . . . , xN ) =
(3.5)
= ϕj1 (xσ(1) )ϕj2 (xσ(2) ) · · · ϕjN (xσ(N ) ) ,

i.e., in terms of abstract basis vectors,

Pσ (ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ) = ϕjσ−1 (1) ⊗ ϕjσ−1 (2) ⊗ · · · ⊗ ϕjσ−1 (N ) . (3.6)

Notice that both in (3.5) and in (3.6) the following is meant: after the action of Pσ on
the basis vector, the k-th particle, that was in the state ϕjk , is now in the state ϕjσ−1 (k) ,
while the state ϕjk is now occupied by the σ(k)-th particle. It is easily seen that Pσ is
unitary ∀ σ ∈ SN . Define also the symmetrizer on HN to be the operator

S : HN −→ HN
1 X (3.7)
S := Pσ .
N!
σ∈SN

One sees that


S = S∗ = S2 (3.8)
so S is an orthogonal projection. The Hilbert subspace it projects onto is said the sym-
metric or bosonic sector of HN , denoted by HN, sym or, in particular, by L2sym (ΩN ):

HN, sym := S HN . (3.9)


44 Chapter 3. Formalisation of Bose-Einstein condensation

It is a postulate in this context that physically admissible pure states for a system of
N undistinguishable bosons are invariant under S, i.e., are norm-one vectors in HN, sym ,
i.e., are eigenvectors of S with eigenvalue 1:
S ΨN = ΨN (admissible pure states for the bosonic system) . (3.10)
More generally, physically admissible (possibly mixed ) states are density matrices oper-
ators on HN which act non-trivially only on HN, sym :
HN = S HN ⊕ (1 − S) HN
≡ HN, sym ⊕ HN, asym
(3.11)
γN (HN, sym ) ⊂ HN, sym
¯
γN ¯ = O.
HN, asym

When HN = L2 (ΩN ), the bosonic sector is nothing but the subspace of permutation-
ally symmetric wave functions, i.e., wave functions such that
ΨN (x1 , x2 , . . . , xN ) = ΨN (xσ(1) , xσ(2) , . . . , xσ(N ) )
(3.12)
∀ σ ∈ SN , ∀ ΨN ∈ L2sym (ΩN ) ,
and bosonic states are density matrices such that their kernels are symmetric under
permutation of each set of variables,
γN (x1 , . . . , xN ; y1 , . . . , xN ) = γN (xσ(1) , . . . , xσ(N ) ; yσ(1) , . . . , yσ(N ) )
(3.13)
∀ σ ∈ SN , ∀ bosonic state γN .
As a consequence of the aforementioned postulate and of first principles of Quan-
tum Mechanics, physical observables do not connect the symmetric (bosonic) and anti-
symmetric (fermionic) sectors (for, otherwise one could provide a physical measure – a
projection – that would prepare a superposition of a bosonic + fermionic state, starting
from a bosonic state only). Thus, observables one shall be dealing with, i.e., bosonic
observables, are operators ON ∈ L(HN ) which act non-trivially only on HN, sym as γN in
(3.10). In particular, when ON ∈ L2 (L2 (ΩN )), namely, when ON has an integral kernel,
then it satisfies (3.13) as well. The Hamiltonian too of a system of N identical bosons
shares the same symmetry
[ S, e−itHN ] = O ∀ t ∈ R. (3.14)
Hence, any bosonic initial state evolves inside the bosonic sector at any later time.
The basis {ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN }j1 ,j2 ,...,jN of HN turns out not to be convenient for
describing the physical subspace HN, sym . Among them, only basis vectors that have the
form of a N -th tensor power are physical. We will exhibit soon a convenient basis of
HN, sym in equation (3.22). To this aim, first notice that HN, sym is certainly generated
by {S(ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN )}j1 ,j2 ,...,jN , so one needs to manage vectors of this kind.
First we adopt the following self-explanatory and rather useful notation.
Definition 3.1.2 (quantum labels for N -th tensor product vectors). A basis
vector ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN of HN is said to have quantum labels
|i1 , n1 ; i2 , n2 ; . . . ; ir , nr i (3.15)
(also written |ϕ1 , n1 ; ϕ2 , n2 ; . . . ; ϕr , nr i), for some r = 1, . . . , N and some positive integers
P
n1 , . . . , nr such that ri=1 ni = N and n1 > n2 > · · · > nr > 1, iff the following holds:
3.1. Many-body quantum systems of identical bosons 45

¬ there are r distinct indices i1 , . . . , ir in {j1 , j2 , . . . , jN } (namely, there are r distinct


vectors ϕi1 , . . . , ϕir among ϕj1 , ϕj2 , . . . , ϕjN ),

­ there are, in {j1 , j2 , . . . , jN }, n1 indices equal to i1 , n2 indices equal to i2 , . . . , and


nr indices equal to ir .
Equivalently, there is a suitable permutation operator Pσ such that

Pσ (ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ) = ϕi⊗n


1
1
⊗ · · · ⊗ ϕ⊗n
ir
r

for r distinct ϕi1 , . . . , ϕir out of ϕj1 , ϕj2 , . . . , ϕjN .


The well known spirit of this definition is simply that quantum labels (3.15) shall
be all that one is allowed to say on a physically admissible pure state: due to undis-
tinguishability, it is not possible to say which particles of the N -body state labeled by
|i1 , n1 ; . . . ; ir , nr i are, e.g., in the one-body state ϕi1 , but simply that there are n1 out of
N in that state ϕi1 . Thus, for instance, the three vectors in H3

ξ ⊗ξ ⊗η, ξ⊗η⊗ξ, η⊗ξ⊗ξ

(with kξk = kηk = 1 and (ξ, η) = 0 in H), are mathematically distinct objects, and
are all the possible basis vectors with quantum labels |ξ, 2; η, 1i. Also, not even one is
physically admissible, i.e., belongs to H3, sym , while the only physically meaningful pure
state one can build with them is (apart a normalisation)

S(ξ ⊗ ξ ⊗ η) = S(ξ ⊗ η ⊗ ξ) = S(η ⊗ ξ ⊗ ξ) =


1
= (ξ ⊗ ξ ⊗ η + ξ ⊗ η ⊗ ξ + η ⊗ ξ ⊗ ξ) .
3
More generally one remarks the following.
Theorem 3.1.3. ¬ Fix an integer r between 1 and N , fix r positive integers n1 , . . . , nr
Pr
with i=1 nr = N , and pick r distinct vectors ϕi1 , . . . , ϕir in the basis {ϕj }j of H.
Then the number of distinct vectors in the basis {ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN }j1 ,j2 ,...,jN
of HN which have the same quantum labels |ϕi1 , n1 ; . . . ; ϕir , nr i is
N!
n1 ! n2 ! · · · nr !
and, unless r = 1 (whence n1 = N ), none of them is in HN, sym .

­ All distinct basis vectors of HN with the same quantum labels are orthonormal in
HN and have the same image in HN, sym under the action of S.

® Any two basis vectors of HN with distinct quantum labels are orthonormal and
give orthogonal images in HN, sym under the action of S.

¯ Partition the basis {ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN }j1 ,j2 ,...,jN of HN into classes of vectors
with the same quantum labels. Pick a vector in each class, take their image under
S and renormalise them to 1. This way, one gets an orthonormal basis of HN, sym .

° Let ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN have quantum labels |i1 , n1 ; i2 , n2 ; . . . ; ir , nr i. Then


S(ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ) is a linear combination of all the N !/(n1 ! n2 ! · · · nr !) dis-
tinct basis vectors with the same quantum labels, each of these being weighted
with (n1 ! n2 ! · · · nr !).
46 Chapter 3. Formalisation of Bose-Einstein condensation

± Let ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN have quantum labels |i1 , n1 ; i2 , n2 ; . . . ; ir , nr i. Then


° °
° S(ϕj ⊗ ϕj ⊗ · · · ⊗ ϕj ) °2 = n1 ! n2 ! · · · nr ! . (3.16)
1 2 N HN N!
So it is natural to asopt the following definition (the well-posedness of which is guar-
anteed by theorem 3.1.3).

Definition 3.1.4 (quantum labels in the bosonic sector). Each orthonormal basis
of HN, sym built as in point ¯ of theorem 3.1.3 is said a canonical basis of HN, sym
associated to the basis {ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN }j1 ,j2 ,...,jN of HN . Each vector ΨN of such
a canonical basis is said to have quantum labels |ϕ1 , n1 ; . . . ; ϕr , nr i iff

ΨN = S (ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN )

where ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN is labeled by |ϕ1 , n1 ; . . . ; ϕr , nr i, or equivalently, iff ΨN is


the image, under S, of any basis vector of HN with those quantum labels, for instance

ΨN = S (ϕ⊗n
i1
1
⊗ · · · ϕ⊗n
ir ) .
r
(3.17)

The standard (and physically ubiquitous) example one has in mind concerns the
one-body Hilbert space H ∼
= C2 , where an orthonormal basis {e1 , e2 } is chosen. Then

{e1 ⊗ e1 , e1 ⊗ e2 , e2 ⊗ e1 , e2 ⊗ e2 }

is an orthonormal basis of H2 = H⊗2 and, via S (and after renormalisation), one builds
the canonical basis
1
{e1 ⊗ e1 , √ (e1 ⊗ e2 + e2 ⊗ e1 ), e2 ⊗ e2 }
2
of HN, sym , often referred to as triplet basis. The states of this triplet basis have quantum
labels |e1 , 2i, |e1 , 1; e2 , 1i and |e2 , 2i, respectively. The vector √12 (e1 ⊗ e2 − e2 ⊗ e1 ) of
(1 − S)H2 completes the triplet basis to an orthonormal basis of the whole H2 and is
usually referred to as singlet state.
Incidentally, renormalisation to 1 of each SΨN (provided that ΨN ∈ / ker S) is a stan-
dard operation that deserves a specific notation.

Definition–Theorem 3.1.5 (symmetrization – II). The symmetrized of any ΨN ∈


HN such that ΨN ∈
/ ker S is the norm-one vector
¡ ¢ SΨN
ΨN sym := ∈ HN, sym . (3.18)
kSΨN k

Any vector of the basis {ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN }j1 ,j2 ,...,jN of HN has a nonzero image
under S. Let ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN have quantum labels |i1 , n1 ; i2 , n2 ; . . . ; ir , nr i. Then,
according to (3.16),
r
¡ ¢ N!
ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN sym = S(ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ) . (3.19)
n1 ! n2 ! · · · nr !
Last equation can be simplified, by point ° in theorem 3.1.3, as
r
¡ ¢ n1 ! n2 ! · · · nr ! X
ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN sym = Pσ (ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ) . (3.20)
N! 0
σ∈SN
3.1. Many-body quantum systems of identical bosons 47

where S0N ⊂ SN consists of all the N !/(n1 ! n2 ! · · · nr !) distinct permutations of the N -ple
¡ ¢
i1 , . . . i1 , i2 , . . . i2 , . . . , ir , . . . ir .
| {z } | {z } | {z }
(n1 ) (n2 ) (nr )

The symmetrized of any basis vector ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN of HN with quantum labels
|i1 , n1 ; i2 , n2 ; . . . ; ir , nr i will be denoted also by
¡ ¢
|i1 , n1 ; i2 , n2 ; . . . ; ir , nr isym := ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN sym
¡ ¢ (3.21)
= ϕ⊗n
i1
1
⊗ ϕ⊗n
i2
2
⊗ · · · ⊗ ϕ⊗n
ir
r
sym
.

For example, when kξk = kηk = 1 and (ξ, η) = 0,


1
S(ξ ⊗ ξ ⊗ η) = (ξ ⊗ ξ ⊗ η + ξ ⊗ ξ ⊗ η + ξ ⊗ η ⊗ ξ
6
+ ξ ⊗ η ⊗ ξ + η ⊗ ξ ⊗ ξ + η ⊗ ξ ⊗ ξ)
1
= (ξ ⊗ ξ ⊗ η + ξ ⊗ η ⊗ ξ + η ⊗ ξ ⊗ ξ) ,
3
¡ ¢ 1
ξ ⊗ ξ ⊗ η sym = √ (ξ ⊗ ξ ⊗ η + ξ ⊗ η ⊗ ξ + η ⊗ ξ ⊗ ξ) .
3
Summarizing, one chooses an orthonormal basis {ϕj }j of the single-particle Hilbert
space H – and this has to be done in the most convenient way for the Physics of the
problem – which determines the orthonormal product basis (3.1) of the N -body space
HN . Then, partitioning (3.1) into classes of basis vectors with the same quantum labels,
and taking the ( · )sym in each class, one gets a canonical orthonormal basis of the bosonic
sector HN, sym . Thus, varying
3 r = 1, . . . , N ,

3 ϕi1 , . . . , ϕir distinct in {ϕj }j ,


Pr
3 (n1 , . . . , nr ) such that n1 > · · · > nr > 1 in N and i=1 ni = N,
one runs over all distinct quantum labels |ϕi1 , i1 ; . . . ; ϕir , ir i which label such a basis of
HN, sym . Written in terms of this basis, a generic N -body bosonic pure state reads
X ³¡ ¢ ´ ¡ ⊗n1 ¢
ΨN = ϕ⊗n
i1
1
⊗ · · · ⊗ ϕ ⊗nr
ir sym
, ΨN ϕi1 ⊗ · · · ⊗ ϕ⊗n
ir
r
sym
HN
|ϕi1 ,n1 ;...;ϕir ,nr i
X
= |ϕi1 , n1 ; . . . , ϕir , nr isym sym hϕi1 , n1 ; . . . , ϕir , nr |ΨN i
r (3.22)
i1 ,...ir
n1 ,...,nr
X
= C quantum |ϕi1 , n1 ; . . . ; ϕir , nr i
labels
{quantum
labels}

(the last line being just a suggestive condensed form).


Combining this knowledge of a basis of HN, sym with prescription (3.11) for bosonic
density matrices, it follows that the generic (possibly mixed) N -body bosonic state is a
density matrix
X (j) (j)
X
γN = λj |ΦN ihΦN | , λj > 0 , λj = 1 (3.23)
j j
48 Chapter 3. Formalisation of Bose-Einstein condensation

(j)
for some orthonormal basis {ΦN }j of HN, sym , which in general is not the canonical basis
entering (3.22). Expression (3.23) is nothing but the concrete realization of the diagonal
form (2.21) in this picture.

3.2 Limit of infinite systems


It is an unfortunate but true fact that essentially always many-body problems are too
hard to be solved explicitly, due to their overwhelming complexity. A standard perspec-
tive is to study the problem in some kind of infinite limit for the “large” system. To
give logical consistency to the present chapter, this section provides details of such an
approach in view of the following two sections dealing with gases of bosons in a possibly
condensed phase. This anticipates a more complete discussion on the main concept of
scaling limit which is the topic of the next chapter.
For concreteness, let S0 be a system of N0 undistinguishable bosons, possibly inter-
acting and confined in some region of the d-dimensional space. Associated to it, one
has a Hilbert space HN0 = H⊗N0 , H being the Hilbert space by which a single particle
is described, and a Hamiltonian HN0 which in general is not the sum of single-particle
Hamiltonians, due to the presence of some interaction.
The customary mathematical idealization is to model the “true” physical S0 by an
increasing sequence {SN }N of larger and larger N -body systems that share with S0 some
prescribed features, and to take the limit N → ∞ (in fact, in current experiments with
Bose-Einstein condensates, N ranges from some hundreds to 1011 – see Table 1.1). More
precisely, it has to be emphasized that each SN is not simply the same S0 , but with a
larger N instead of N0 , and that unless some prescriptions on the Hamiltonian HN of
SN are taken, the limit procedure is ambiguous.
What one assumes is that each SN is made by N particles of the same kind as in
SN0 , confined and coupled each other as particles are in S0 ,1 but with coupling strength
and size of the confinement depending on N in such a way that

¬ as N = N0 , the system SN is exactly S0 , i.e., HN |N =N0 = HN0 on HN = HN0 ,

­ for any N , or asymptotically as N → ∞, certain physical quantities of SN in the


state γN are the same as the corresponding ones for S0 in the state γN0 .

For instance, a typical case (discussed in Sec. 4.3) is a gas of spinless undistinguishable
bosons where, in the limit N → ∞, the ratio between interaction and kinetic energy
in the ground state is kept constant: such a ratio would not be constant with N if one
simply took SN to be exactly S0 with N instead of N0 , namely, without rescaling the
potential.
This way, each SN resembles by construction the original S0 at least for all those
physical quantities that are kept fixed with N .
Suppose now that one is interested in some physical quantity f (i.e., the expectation of
some observable) of the original system S0 in some state, e.g., its ground state. Normally
f is not accessible exactly, yet being in principle a well-defined function f (N, L, a, . . . )
of the labels of the quantum state under consideration and of the relevant physical
1
This, in principle, can change the shape of the trap: this is the case when the scaling rate is faster
along one spatial dimension than the others, with the aim to obtain a lower dimensional model. We will
be not discussing this possibility, so for us by scaling the trap mantains the same aspect ratio as for S0 .
3.2. Limit of infinite systems 49

parameters of the Hamiltonian, such as the number N of particles, the size L of the
system (e.g., the typical linear size of the trap where the gas is confined in), the s-wave
scattering length a of the interparticle interaction, etc. To obtain f , prescription ­ above
fixes in principle how L, a, . . . have to scale with N (say LN , aN , . . . ), as N → ∞, thus
making the limit for f unambiguous in the form limN →∞ f (N, LN , aN , . . . ).
The collection of all such prescriptions in HN identifies the so called scaling limit
under which, as N → ∞, quantities of interest of the form f (N, L, a, . . . ) are investigated,
to get explicitly their asymptotics. Thus, a given scaling is nothing but a curve going
to infinity in the direction of monotonically increasing N in the domain {(N, L, a, . . . )}
of the function f (see Fig. 4.1), such that at each point the corresponding system SN
exhibits a number of features in common with the original S0 . If in the scaling limit one
gets
lim f (N, LN , aN , . . . ) = F ,
N →∞

then such F is regarded as an approximation of the value f (N0 , L0 , a0 , . . . ) of the true


physical system under consideration, since the asymptotics has been obtained under the
same physical regime of S0 (see Sec. 4.3 for details).
So the N0 -body physical system is investigated via a sequence {SN }N of larger and
larger auxiliary systems each of which is described by a Hamiltonian HN acting on HN .
The original S0 in the state γN0 is regarded as a sequence of states {γN }N .
Our standard setting will be

HN = L2 (RN d ) ∼
= L2 (Rd )⊗N (3.24)

where d = 1, 2, 3 is the dimension of the system. If Ω ⊂ Rd is the region where the


original system is confined in, then each N -body system is confined in a region ΩN
possibly depending on N ,2 that is, the size of the domain of integration may scale with
N.
Each of the γN is well-defined, although their limit as N → ∞ is ill-posed. In fact,
the sequence of increasing systems SN cannot be thought to “converge” to the original
S0 . More precisely, HN acts on a larger and larger Hilbert space HN which eventually
becomes a far too large space to control the N = ∞ limit system (in the infinite tensor
product H⊗∞ separability is lost and distinct tensor power vectors ϕ⊗∞ are always
orthogonal).
Nevertheless, observables of interest, when the “large” system is modelled as infinite,
usually are observables only on finite parts of it and it is at this very point that the tool
of the reduced density matrices finds its application.
Observables on a fixed subsystem of k particles out of N0 will be lifted to the corre-
sponding observables on the same fixed subsystem of each SN . Boson symmetry makes
all choices of k particles equivalent, but from the mathematical viewpoint we will dis-
tinguish the variables ordering, so that the subsystem of k particles will be meant to be
the subsystem of the first k.
More precisely, for some fixed integer k with 1 6 k < N , one regards HN as

HN ∼
= Hk ⊗ HN −k ∼
= L2 (Rkd ) ⊗ L2 (R(N −k)d ) (3.25)
2
Notice the distinct notation: each ΩN ⊂ Rd is an element of the sequence {ΩN }N of enlarging
domains in Rd , while by ΩN one means the Cartesian product Ω × Ω × · · · × Ω ⊂ RN d .
50 Chapter 3. Formalisation of Bose-Einstein condensation

and one considers observables of the form O ⊗ 1N −k with O = O∗ ∈ L(Hk ). The


expectation of O ⊗ 1N0 −k in the state γN0 is thus investigated through the sequence of
expectations
(k)
Tr[ γN (O ⊗ 1N −k ) ] = Tr[ γN O ] (3.26)
(k)
where γN is the reduced density matrix of the k-particle subsystem. Then one takes
an appropriate scaling limit N → ∞ as discussed above. Notice that dealing with
observables on some fixed subsystem only allows one to overcome the ambiguity of the
(k)
limit in {γN }N : indeed the sequence {γN }N is all inside L1 (Hk ), hence taking N → ∞
makes sense, provided that a suitable topology for the limit is specified.
Natural questions are

(Q1 ) whether a limit exists for a reasonable, physically meaningful set of observables
(k)
on the k-particle subspace Hk ⊗ 1N −k and whether it is of the form Tr[ γ∞ O ] for
(k)
some density matrix γ∞ on Hk ,

(Q2 ) whether this depends or not on k.

Usually (Q1 ) is addressed to ∀ O ∈ Com(Hk ): if in this case it has an affirmative


(k) (k)
answer, this is equivalent to say that γN → γ∞ weakly-∗ in L1 (Hk ), by point ° in
theorem 2.1.1:
 (k)
 (k)
1  lim Tr[ (γN − γ∞ )C ] = 0
(k) weakly-∗ in L (Hk ) (k) N →∞
γN −−−−−−−−−−−−→ γ∞ ⇐⇒ (3.27)
N →∞ 

∀ C ∈ Com(Hk ) .

(k)
The quantum mechanical interpretation is that the state γ∞ is an approximation of the
(k)
“true” γN0 : the larger N0 , the better the approximation.
Question (Q2 ) deserves a special care when Bose-Einstein condensation is investigated
in the limit of infinite system and it will be discussed for our purposes in Sec. 7.1. In fact,
as we are going to remark, control of marginals at distinct k’s gives information on the
interparticle correlations, which, in turn, is central in the study of several B.E.C. features.
In conclusion, the twofold essence of this scaling approach is

3 to regard the original physical system under consideration, S0 , to be so large that


only observables on finite portions of it are of interest, so that one deals with
reduced density matrices only,

3 to perform a N → ∞ limit through an appropriate scaling, that is, keeping some


relevant physical quantities constant with N : S0 is mimed with an enlarging se-
(k) (k)
quence {SN }N , the possible limit γN → γ∞ is investigated at fixed k; within the
(k) (k)
scaling in use, γ∞ is an approximation of the true γN0 .

3.3 Bose-Einstein condensation for finite size systems


According to the physical scenario depicted in Chap. 1, a physically meaningful and
reasonably general mathematical formalisation of Bose Einstein condensation has to ac-
count for the macroscopic occupation of the same single-particle state, the characteristic
feature of such a phenomenon.
3.3. Bose-Einstein condensation for finite size systems 51

The experimental evidence of that relies, e.g., on the typical peak in the observed
particle space distribution (or momentum distribution) of a suitably cooled and trapped
alkali vapour (see, e.g., Fig. 1.8). From its shape one argues (we are coming back to this
point later, see the introductory considerations of Chap. 6) that with some approximation
the system is in a factorised state, all factors being the same one-body wave function.
For a non-interacting system it corresponds to a bound state of the trap for a single
particle. For interacting systems it is a modification of it – for instance instead of being
the ground state of the trap, it is the minimizer of a suitable functional. The density
profile is thus proportional to the density of each one-particle factor, called the condensate
wave function. Yet for interacting particles, only N -body states have physical meaning,
and one has to give meaning to a notion of occupation of a single-particle state in a
many-body state.
So, occupation number is the crucial point when formalising B.E.C. In this section
we are going to discuss it with respect to finite size systems, while in the next one we
will focus on the same problem in the context of the limit of infinite system. Here and
in the following the key remark will be that occupation number is a concept that can
be made physically and mathematically meaningful at the level of one-particle reduced
density matrices, instead of at the level of N -body states.
In fact,

3 in the limit of infinite system, the tool of fixed-k-particle reduced density matrices
emerges as the natural one, while the N -body state looses its meaning as N → ∞,
as we have already discussed in the previous section.

3 Even when dealing with B.E.C. for finite size systems, the many-body wave func-
tion is neither easily accessible, due to the complexity of the problem, nor easily
observable. Experimentally one observes, e.g., a spatial density distribution n(x),
that is, the average number of particles in some neighborhood of given point. By
undistinguishability
Z
¯ ¯
n(x) = N ¯ΨN (x, x2 , . . . , xN )¯2 dx2 · · · dxN = N γ (1) (x, x)
N
ΩN −1
R
(indeed Ω n(x) dx = N ), Ω ⊂ Rd being the region where the system is confined in.
(1)
Thus, one is actually dealing with the one-particle reduced density matrix γN of
the N -body state γN . This is another way to say that physical observables involved
are essentially one-particle observables.

So one needs to relate the standard description of the many-body states to properties
of the marginals, in order to manage a satisfactory formalisation of condensation and, in
particular, of the concept of macroscopic occupation of the same single-particle state.
For pure states (3.21), i.e., vectors of the canonical basis of the many-body bosonic
sector HN,sym , the notion of occupation numbers can be indifferently set at the level of
N -body states or 1-marginals. In the first case, the statement that in the state

|ϕi1 , n1 ; ϕi2 , n2 ; . . . ; ϕir , nr isym

there is a fraction ni1 /N of particles in the same one-particle state ϕi1 , etc., is definitely
unambiguous. The same information is present at the one-body level, due to the following
theorem.
52 Chapter 3. Formalisation of Bose-Einstein condensation

Theorem 3.3.1 (occupation numbers in a N -body pure state, [83]). Let γN =


|ΨN ihΨN | be the N -body pure state with definite quantum labels

ΨN = |ϕi1 , n1 ; ϕi2 , n2 ; . . . ; ϕir , nr isym . (3.28)

Then the corresponding one-particle reduced density matrix is the finite rank operator
r
X
(1) nk
γN = |ϕik ihϕik | . (3.29)
N
k=1

In particular, if |ϕi1 , n1 ; ϕi2 , n2 ; . . . ; ϕir , nr isym and |ϕi1 , m1 ; ϕi2 , m2 ; . . . ; ϕir , mr isym have
the same one-particle reduced density matrix then n1 = m1 , . . . , nr = mr .
Proof. Recall (see equation (3.20) and theorem 3.1.3) that |ϕi1 , n1 ; ϕi2 , n2 ; . . . ; ϕir , nr isym
is a linear combination of N !/(n1 ! n2 ! · · · nr !) orthonormal basis vectors of the form (3.2),
each weighted with the factor ((n1 ! n2 ! · · · nr !)/N !)1/2 : these are the vectors

ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN

of HN for all the possible distinct N -ples of indices (j1 , j2 , . . . , jN ) obtained by permu-
tation of the assigned N -ple
¡ ¢
i1 , . . . i1 , i2 , . . . i2 , . . . , ir , . . . ir .
| {z } | {z } | {z }
(n1 ) (n2 ) (nr )

When constructing γN = |ΨN ihΨN | one then gets diagonal terms of the form
n1 ! n2 ! · · · nr !
|ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ihϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN | (3.30)
N!
and off-diagonal terms of the form
n1 ! n2 ! · · · nr !
|ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ihϕj10 ⊗ ϕj20 ⊗ · · · ⊗ ϕjN0 | (3.31)
N!
with
(j1 , j2 , . . . , jN ) 6= (j10 , j20 , . . . , jN
0
) (labels of off-diagonal terms)
j1 , j2 , . . . , jN , © ª (3.32)
0 0 0 ∈ i1 , . . . i1 , i2 , . . . i2 , . . . , ir , . . . ir .
j1 , j2 , . . . , jN | {z } | {z } | {z }
(n1 ) (n2 ) (nr )

Denote by Tr[N −1] : H⊗N → H the partial trace over the last N − 1 particles (i.e.,
variables). One then sees that only diagonal terms contribute when partial trace is taken,
for necessarily (j2 , . . . , jN ) 6= (j20 , . . . , jN
0 ) for any off-diagonal term. Indeed if j = j 0
1 1
then (j2 , . . . , jN ) 6= (j2 , . . . , jN ) in order (3.32) to hold, while if j1 6= j10 then the equality
0 0

(j2 , . . . , jN ) = (j20 , . . . , jN
0 ) would contradict that the collections

{j1 , j2 , . . . , jN } and {j10 , j20 , . . . , jN


0
}

are made by the same N integers. Then, when evaluated, e.g., in terms of kernels, Tr[N −1]
R
in (3.31) gives products of integrals of the form Ω ϕjk (xk )ϕjk0 (xk ) dxk , k = 2, . . . , N and
there must be at least one k such that jk 6= jk0 whence, by orthonormality,
Z
ϕjk (xk )ϕjk0 (xk ) dxk = 0 . (3.33)

3.3. Bose-Einstein condensation for finite size systems 53

Thus, only diagonal terms (3.30) contribute to partial trace and each of them gives
· ¸
n1 ! n2 ! · · · nr !
Tr[N −1] |ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ihϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN |
N!
N Z
Y
n1 ! n2 ! · · · nr !
= |ϕj1 ihϕj1 | |ϕjk (xk )|2 dxk (3.34)
N! Ω k=2
n1 ! n2 ! · · · nr !
= |ϕj1 ihϕj1 | .
N!
Actually, |ϕj1 ihϕj1 | can be one of the r distinct projections |ϕi1 ihϕi1 |, . . . , |ϕir ihϕir | and it
remains to count, in the linear combination (3.20) giving |ϕi1 , n1 ; ϕi2 , n2 ; . . . ; ϕir , nr isym ,
the number νi1 of terms with ϕi1 in first position, the number νi2 of terms with ϕi2 in
first position, etc., for then
r
X
(1) n1 ! n2 ! · · · nr !
γN = νik |ϕik ihϕik | . (3.35)
N!
k=1

To do this counting, consider again the N !/(n1 ! n2 ! · · · nr !) distinct ways to place n1 >
1 copies of ϕi1 , . . . , and nr > 1 copies of ϕir in an ordered set of N = n1 + · · · + nr boxes.
Fixing, e.g., one copy of ϕ1 in first position, one ends up with all distinct dispositions of
n1 − 1 > 0 copies of ϕi1 , . . . , and nr > 1 copies of ϕir in an ordered set of N − 1 boxes:
their number is (N − 1)!/((n1 − 1)! n2 ! · · · nr !). Thus, in general,

(N − 1)!
νik = , k = 1, . . . , r . (3.36)
n1 ! · · · (nk − 1)! · · · nr !

In conclusion, (3.35) and (3.36) give (3.29).

On the other hand, for a generic N -body bosonic pure or mixed state – see (3.22) and
(3.23) respectively – one should deal with some kind of weighted means of occupation
numbers in terms of states with definite quantum labels. For instance, it is hard to say
at first glance what “macroscopic occupation of single-particle states” might be in the
(perfectly admissible) 8-body bosonic pure state
1 ¡ ¢
Ψ8 = √ |ϕ, 6; ξ, 2isym + |ϕ, 5; ξ, 2; η, 1isym + 2 |ϕ, 4; η, 4isym
6
(with ϕ, ξ, η orthonormal in the one-body Hilbert space). One might say, e.g., that the
one-body states ϕ, ξ and η have relative occupation in Ψ8 given by
6
8 + 58 + 2 · 48 19 2
+ 28
8 4 1
+ 2 · 48
8 9
= , = , =
1+1+2 32 1+1+2 32 1+1+2 32
respectively. Set γ8 := |Ψ8 ihΨ8 |: after a somewhat lengthy computation one finds

(1) 27 4 17 1 1
γ8 = |ϕihϕ| + |ξihξ| + |ηihη| + √ |ϕihη| + √ |ηihϕ|
48 48 48 8 6 8 6
that is, a different (although similar) set of candidate occupation numbers. In both
cases these numbers do not reconstruct Ψ8 uniquely, of course, although in both cases
one should agree, e.g., that the one-body state ϕ is “occupied” for a fraction which is
(1)
slightly more than one half. Actually, when diagonalizing γ 8 , one finds that the largest
54 Chapter 3. Formalisation of Bose-Einstein condensation

√ √
eigenvalue is λ0 = 22+48 31 ≈ 0.57 (the others being λ1 = 22−48 31 and λ2 = 48 4
) and it
1 5 1/2 1 5 1/2
corresponds to the (normalised) eigenvector ( 2 + 2√31 ) ϕ + ( 2 − 2√31 ) η, that is,
approximately, 0.97ϕ + 0.22η.
When addressing the same question of occupancy in the one-particle reduced density
(1)
matrix γN , instead of in the full γN , one can always have the decomposition (2.21)

(1)
X X
γN = λj |ϕj ihϕj | , λj > 0 , λj = 1 ,
j j

where {ϕj }j is an orthonormal system of the one-particle Hilbert space H. Each nonzero
eigenvalue λj deserves the role of occupation number in the one-body state {ϕj }j .
The above considerations serve as a motivation to the standard mathematical defini-
tion of B.E.C. for generic interacting N -body boson systems, which we are now discussing.
In it, emphasis is given to the one-particle reduced density matrix. Such a definition has
been given for the first time by Penrose and Onsager [95] half a century ago. In the
current notation it reads as follows.

Definition 3.3.2 (B.E.C. after Penrose-Osanger). Let γN be the state of a system


of N identical and possibly interacting bosons and let the corresponding one-particle
(1)
reduced density matrix γN have diagonal form

M
X M
X
(1)
γN = λj |ϕj ihϕj | , λj > λj+1 > 0 , λj = 1 , (3.37)
j=0 j=0

where M 6 ∞ and {ϕj }M j=0 is an orthonormal system of the one-particle Hilbert space H.
The N -body state γN is said to exhibit Bose-Einstein condensation iff the largest
eigenvalue λ0 is such that
λ0 ∼ 1 (3.38)
with an approximation that has to be specified for each specific system under considera-
tion. (The phrase “λ0 almost 1” is somewhat vague in a situation where there is no limit
in the number of particles, but in practice this does not usually lead to difficulty.)

Definition 3.3.3. One has simple condensation if one and only one eigenvalue of
(1)
γN is of order 1, all the others being negligible, and fragmented condensation if
more that one eigenvalue is of order 1, e.g., λ0 , λ1 ∼ 1.

Thus, B.E.C. is defined to be a property of the state under consideration: no reference


is made to the Hamiltonian of the system, that is, assigned a bosonic system with its
Hamiltonian, there may be or there may be not states of condensation.
A natural problem then arises on whether B.E.C. can be proved in some state, once
HN is given, for instance in the ground state. We are discussing this point (existence
of condensation) in Sec. 3.5 for a typical dilute gas of trapped and pairwise interacting
bosons.
Another issue is the persistence of condensation in an originally condensed state
which evolves in time: we are coming back to this point (time-stability of condensation)
in Chap. 5.
Since a rigorous treatment of these issues requires a limit of infinitely many particle,
in the next section we will restate the definition of B.E.C. for infinite systems.
3.4. B.E.C. in the limit of infinite systems 55

3.4 B.E.C. in the limit of infinite systems


So far we have considered finite systems of bosons and the operational definition 3.3.2
of B.E.C. Actually one would like to appeal to some kind of limit N → ∞ at least
for two main reasons: (i) in practice the number of particles is so large to justify the
assumption of infinite size; (ii) to prove condensation for interacting systems requires to
(1)
solve a formidably difficult many-body problem in order to access γN , while there might
be hopes to get asymptotic results when N goes to infinity.
Concerning the latter, recall that for interacting systems the ground state already
poses a challenging problem which is still largely unsolved. B.E.C. itself has, so far,
never been proved for many-body Hamiltonians with genuine interactions, except for the
special case of hard-core bosons on a lattice at half-filling (see, e.g., [4]).
Discussion in Sec. 3.2 shows that N → ∞ is not just a thermodynamic limit. Instead
a more careful treatment is needed: some scaling has to be specified. After this is done,
operational definition 3.3.2 takes the following asymptotic form.
Definition 3.4.1 (asymptotic B.E.C.). Let {SN }N be a sequence of systems, each of
N identical bosons, with Hamiltonian HN on the Hilbert space HN = H⊗N (where H is
the one-particle Hilbert space). Let each SN be in the state γN . The sequence {SN }N
is assumed to encode the prescriptions under which limit N → ∞ is taken in a N0 -body
system S0 in the state γN0 , according to the discussion of Sec. 3.2. Let
MN
X MN
X
(1) [N ] [N ] [N ] [N ] [N ] [N ]
γN = λj |ϕj ihϕj | , λj > λj+1 > 0, λj = 1, (3.39)
j=0 j=0

(1)
be the canonical diagonal form of the one-particle reduced density matrix γN relative
[N ]
to each γN , where MN 6 ∞ and {ϕj }M j=0 is an orthonormal system of H. Then the
N

system S0 in the state γN0 is said to exhibit


• ultra-weak asymptotic Bose-Einstein condensation iff
[N ]
∃ lim inf λ 0 =: λ0 (3.40)
N →∞

and 1 − λ0 is conveniently small, with an agreement depending on the specific


system S0 under consideration.

• weak asymptotic Bose-Einstein condensation iff


[N ]
∃ lim λ 0 =: λ0 (3.41)
N →∞

and 1 − λ0 is conveniently small, with an agreement depending on the specific


system S0 under consideration.

• strong asymptotic simple Bose-Einstein condensation iff


(1)
(1)
∃ lim γN =: γ∞ (3.42)
N →∞

(1)
in some topology for density matrices, where γ∞ is still a density matrix with
diagonal form X
(1)
γ∞ = λj |ϕj ihϕj | (3.43)
j
56 Chapter 3. Formalisation of Bose-Einstein condensation

and the largest eigenvalues λ0 is conveniently close to 1, with an agreement de-


pending on the specific system S0 under consideration. In other terms
(1)
γ∞ ∼ λ0 |ϕ0 ihϕ0 | . (3.44)

In this case λ0 is said the fraction of the condensate, while 1 − λ0 is said its
depletion, and the corresponding one-body eigenstate ϕ0 is said the condensate
wave function.

• strong asymptotic fragmented Bose-Einstein condensation iff (3.42)


holds and in (3.43) the two largest eigenvalues λ0 , λ1 (or more) are of order 1,
all the others being negligible, with an agreement depending on the specific system
S0 under consideration. In this case the condensate is said to be fragmented into
the condensate wave functions ϕ0 and ϕ1 .

• complete (or 100%) asymptotic Bose-Einstein condensation in the con-


densate wave function ϕ iff ∃ ϕ ∈ H, with kϕk = 1, such that
(1)
lim γ = |ϕihϕ| (3.45)
N →∞ N

in some topology for density matrices.

In this definition we have included also two “weak” notions of condensation that
sometimes one can find in the literature, but we will be dealing only with the customary
“strong” notion, i.e., controlling condensation in terms of reduced density matrices. Of
course, strong ⇒ weak ⇒ ultra-weak.
Unlike the finite-size case, asymptotic B.E.C. in definition 3.4.1 is not only a prop-
erty of the state, but also of the way how limit N → ∞ is taken. This is not often
mentioned and has to be recalled to prevent thinking that asymptotic results are exactly
features of the original S0 itself, instead of the auxiliary sequence {SN }N . This is a
quite usual misunderstanding when physicists of the Condensed Matter community read
results provided by the Mathematical Physics community. Instead, asymptotic results
have to be interpreted together with a control of the error terms depending on N and
these, in turn, reflect how good the choice of the scaling is to mime S0 with {SN }N . In
general, unfortunately, a control of the error terms is currently a major open problem
and, concerning rigorous results, one has often to content oneself with asymptotic results
only.
A typical example is the case of the so called Gross-Pitaevskiı̆ scaling limit, which is
systematically used to study highly diluted gases of very weakly interacting bosons. By
definition (see Sec. 4.3), this is a depletionless limit, in the sense that it cannot account
for depletion. Indeed in this scaling the quantity ρa3 is vanishing with N , where ρ is a
(suitably defined) mean density of the gas and a is the s-wave scattering length of the
interparticle interaction. Since from Bogolubov’s theory, according to (1.26), one knows
that depletion goes as (ρa3 )1/2 , one cannot find depletion in this limit. This does not
mean that such a scaling cannot describe B.E.C. with depletion, but only that asymptotic
results intrinsically do not account for it. Depletion is zero in the limit and (small and)
positive before the limit.
(1) (1)
We come now to the topology by which one controls the limit γN → γ∞ . We
have already argumented that a natural choice is the weak-∗ topology in the trace class
3.4. B.E.C. in the limit of infinite systems 57

L1 (H): this is exactly the convergence of the expectation values of all compact one-body
observables – see equation (3.27). This is actually the most frequently adopted topology
in the literature.
One may choose to control convergence in the much finer trace norm topology, but
one discovers that the two limits are equivalent in this special case, where a sequence of
density matrices converges to a density matrix. Indeed the following holds.

Theorem 3.4.2 (Weakly converging density matrices converge in trace norm).


Let {γN }N be a sequence of density matrices3 and let γ be a density matrix, all on some
Hilbert space H. Then γN → γ weakly-∗ in L1 (H) iff kγN − γkL1 → 0.

Theorem 3.4.2 is an old fact. It is a consequence of a general mechanism of con-


vergence in the ideals of compact operators, first proved by Simon in 1979 ([101], theo-
rem 2.20). As far as the trace ideal is concerned, Simon’s theorem implies trace norm
convergence, as N → ∞, whenever AN → A, A∗N → A∗ , |AN | → |A| weakly, and
kAN kL1 → kAkL1 . This is clearly the case for our sequence {γN }N , all density matrices
being positive and with trace equal to one. An improved version of this theorem has
been achieved two years after by Arazy [7] and Simon [102], after it had been conjectured
in [101]: they showed that it suffices that AN → A weakly and that kAN kL1 → kAkL1 to
conclude that kAN − AkL1 → 0. In this spirit, we choose to refer to the original weaker
Simon’s theorem and we give detail of such a convergence mechanism in our case with
density matrices.

Proof of theorem 3.4.2. Convergence in trace norm implies trace class weak-∗ conver-
gence. Indeed, by (2.6),

N →∞
Tr[(γN − γ)C] 6 kCk kγN − γkL1 −−−−→ 0 (3.46)

for every C ∈ Com(H), whence

N →∞
Tr|(γN − γ)C| −−−−→ 0 . (3.47)

Let us now prove the converse. First, trace class weak-∗ convergence implies weak
operator convergence. Indeed, by assumption, (3.47) holds ∀ C ∈ Com(H) and the special
choice C = |ψihψ|, ∀ ψ ∈ H, gives
¯ ¯ ¯ ¯ ¯ ¯ N →∞
¯ (ψ, (γN − γ)ψ) ¯ = ¯ Tr[ (γN − γ)|ψihψ| ] ¯ 6 Tr¯ (γN − γ)|ψihψ| ¯ −
−−−→ 0 , (3.48)

whence, by polarization,

N →∞
(ψ, (γN − γ)ϕ) −−−−→ 0 ∀ ψ, ϕ ∈ H . (3.49)

Next, choose any ε > 0. Since Tr[ γ ] = 1, one can choose a finite-dimensional
orthogonal projection P onto the span of a sufficiently large number of eigenvectors of γ
such that
k P γP kL1 > 1 − ε . (3.50)
3 (1)
Here we dropped the index k: all the γN ’s act on H. For our purposes the statement is: if γN →
(1)
|ϕihϕ| weakly-∗ in the trace class, then γN → |ϕihϕ| in the trace norm.
58 Chapter 3. Formalisation of Bose-Einstein condensation

Notice that P γP > O, whence k P γP kL1 = Tr[ P γP ]. Also, P is a positive compact (in
particular: finite rank) operator. By trace class weak-∗ convergence, ∃ Nε such that, as
N > Nε , | Tr[(γN − γ)P ] | 6 ε, so

k P γN P kL1 = Tr|P γN P | = Tr[γN P ]


= Tr[(γN − γ)P ] + Tr[γP ]
¯ ¯ (3.51)
> −¯ Tr[(γN − γ)P ] ¯ + k P γP kL1
> −ε + (1 − ε) = 1 − 2ε .

Define the orthogonal projection Q := 1 − P . Then QγN Q > O and, ∀ N > Nε ,

k QγN Q kL1 = Tr[γN Q]


= Tr[γN ] − Tr[γN P ]
(3.52)
= 1 − k P γN P kL1
6 1 − (1 − 2ε) = 2ε .

One analogously has


k Q γ Q kL1 6 ε . (3.53)
Thus,

k γN − γ kL1 =k (P + Q)(γN − γ)(P + Q) kL1


6 k P (γN − γ)P kL1 + k Q(γN − γ)Q kL1 (3.54)
+ k P γ Q kL1 + k Q γ P kL1 + k P γN Q kL1 + k Q γN P kL1 .

Due to weak operator convergence (3.49),

lim k P (γN − γ)P kL1 = lim k Q(γN − γ)Q kL1 = 0 , (3.55)


N →∞ N →∞

so it remains to control that last four terms in the r.h.s. of (3.54) are small too.
This is immediately done as follows. Take for instance k P γN Q kL1 :

k P γN Q kL1 6 kP k · k γN Q kL1 6 2ε (3.56)

due to (2.6) and to the following Schwartz inequality


√ √
k γN Q kL1 6 k γN kL2 · k γN Q kL2

= k γN kL2 · Tr[ QγN Q ]1/2 (3.57)

6 2ε

(where again it is crucial that density matrices are positive trace-one operators).
All other remaining terms in (3.54) are bounded analogously, so that

k γN − γ kL1 6 k P (γN − γ)P kL1 + k Q(γN − γ)Q kL1 + const · ε. (3.58)

By (3.55) and since ε is arbitrary, then

lim k γN − γ kL1 = 0 (3.59)


N →∞

that is, convergence in trace norm.


3.4. B.E.C. in the limit of infinite systems 59

Since Hilbert-Schmidt convergence implies weak operator convergence, one has also
the following.
Theorem 3.4.3 (Hilbert-Schmidt or trace class norm convergence for density
matrices is the same). Let {γN }N be a sequence of density matrices and let γ be a
density matrix, all on some Hilbert space H. Then kγN − γkL2 → 0 iff kγN − γkL1 → 0.
Here is a summary of equivalent ways of controlling factorisation of marginals (that
(1)
is, γN → |ϕihϕ|) for 100% asymptotic B.E.C.
Theorem 3.4.4 (factorisation of marginals for 100% asymptotic B.E.C.). Let
(1)
{γN }N be the sequence of one-particle reduced density matrices corresponding to the
sequence {γN }N of N -body bosonic states, each on the Hilbert space HN = H⊗N , where
H is some one-particle Hilbert space. Let ϕ ∈ H with kϕk = 1. Then the following are
equivalent:
(1) N →∞
i) γN −−−−→ |ϕihϕ| weakly-∗ in trace class
(1) N →∞
ii) γN −−−−→ |ϕihϕ| in the Hilbert-Schmidt norm
(1) N →∞
iii) γN −−−−→ |ϕihϕ| in the trace class norm
(1) N →∞
iv) (ϕ, γN ϕ) −−−−→ 1.
(k) N →∞
The same clearly can be restated also with respect to convergence γN −−−−→ |ϕihϕ|⊗k ,
where k is a generic positive integer.
(1)
Proof. (i) ⇒ (iv) because tracing the difference γN − |ϕihϕ| against the particular
compact |ϕihϕ| gives
(1) (1)
Tr[ (γN − |ϕihϕ|)|ϕihϕ| ] = (ϕ, γN ϕ) − 1 . (3.60)
(iv) ⇒ (ii) because
° (1) °
° γ − |ϕihϕ| °2 2 (1)
= Tr[ (γN − |ϕihϕ|)2 ]
N L (H)
(1) (1) (3.61)
= Tr[ (γN )2 ] − 2 Tr[ γN |ϕihϕ| ] + Tr[ (|ϕihϕ|)2 ]
¡ (1) ¢
6 2 1 − (ϕ, γN ϕ) .
(ii) ⇔ (iii) by theorem 3.4.3. (i) ⇔ (iii) by theorem 3.4.2. Actually, in the present case,
the limit point γ is the one-dimensional projection |ϕihϕ|, so in the proof of theorems
3.4.2 and 3.4.3 one can choose P = |ϕihϕ| = γ, thus improving the inequalities to
k P γP kL1 = 1, k P γN P kL1 > 1 − ε, k QγN Q kL1 6 ε, and k QγQ kL1 = 0.

Other topologies than the trace class weak-∗ one are present in the literature. This
is typical when one is analysing of the dynamics of the N -body system. In this case one
(1)
studies objects like γN,t , that is, time-dependent marginals corresponding to the state
γN,t of the system which is evolving in time and one wants to know whether in some
(1)
approximation γN,t ≈ |ϕt ihϕt | for N large enough: this way, one gets information about
the evolution of the condensate wave function ϕt .
So here the objects of interest are density matrices with kernels that are functions
(1)
both of time and of space, i.e., (t, x, y) 7→ γN,t (x, y). To manipulate them in the limit
N → ∞ it is technically convenient to regard them as elements in suitable Sobolev spaces
and this broadens the landscape of the possible topologies used to control the limit. This
scenario will be presented in Sec. 5.4.
60 Chapter 3. Formalisation of Bose-Einstein condensation

3.5 Proof of B.E.C. in a scaling limit: two examples.


Stationary Gross-Pitaevskiı̆ theory.
In this section we report two recent proofs of B.E.C. in the sense of definition 3.4.1, that
is, asymptotic B.E.C. in a scaling limit for an infinite system of interacting bosons. We
cite Ref. [75] as a comprehensive review. The first one is the case of a gas in a box with
periodic or Neumann boundary conditions.

Theorem 3.5.1 (B.E.C. for dilute homogeneous gases – Lieb and Seiringer
(2002), [74]). Let N0 undistinguishable spinless bosons of mass m in a three-dimensional
box Ω of side L with either periodic or Neumann boundary conditions be paired by a
two-body, non-negative, spherically symmetric potential V with finite s-wave scattering
length a, according to the Hamiltonian
N
~2 X0 X
HN0 = − ∆xi + V (xi − xj ) (3.62)
2m
i=1 16i<j6N0

(here each xi ∈ Ω ⊂ R3 ). Take the Gross-Pitaevskiı̆ scaling 4 in the Hamiltonian, that


is, N0 7→ N , L 7→ LN , and a 7→ aN , such that
N aN N0 a
= = const . (3.63)
LN L
This means that one considers systems with Hamiltonians
N
~2 X X
HN = − ∆xi + VN (xi − xj ) (3.64)
2m
i=1 16i<j6N

acting on HN = L2 ((ΩN )N ), with the same boundary conditions, where ΩN is the cube
of side LN , and
³ a ´2 ³ a ´
VN (x) := V x (3.65)
aN aN
so that the scaled interaction VN has scattering length aN . Let ΨN be the ground state
(1)
of HN , and let γN be the corresponding one-particle reduced density matrix. Then
Z
1 (1)
lim 3 γN (x, y) dx dy = 1 . (3.66)
N →∞ LN
ΩN ×ΩN

Let us add some explanatory comments to theorem 3.5.1.

3 First, the standard rigorous definition of scattering length can be found in appendix
A of [80] and it implies immediately that VN has scattering length aN .

3 Scaling (3.63) is a limit of infinite dilution and with a constant ratio between
kinetic and interaction energy. Among all infinite equivalent realizations of (3.63),
one can fix it as a “thermodynamic limit” ρN := N/L3N = N0 /L3 = const, whence
necessarily aN ∼ N −2/3 (due to the identity aN = (N aN /LN )(N/L3N )−1/3 N −2/3 ).
Actually it is more manageable to fix LN = L and this is what we will refer to.
4
a detailed presentation of which is in Sec. 4.3
3.5. Proof of B.E.C. in a scaling limit. Stationary GP theory. 61

So the scaling takes the form N aN = N0 a = const. Thus, we will have in mind
to fill the fixed box Ω with more and more particles, yet getting a more and more
−1/3
dilute system, because ρN a3N → 0, that is, the mean interparticle distance ρN is
infinitely larger than aN . (All this is discussed in detail in Sec. 4.3.)

3 Limit (3.66) is equivalent to

(1) 1
lim (ϕ0 , γN ϕ0 ) = 1 , ϕ0 (x) ≡ , (3.67)
N →∞ L3/2
which in turn is equivalent to
(1)
lim γN = |ϕ0 ihϕ0 | (3.68)
N →∞

in the trace norm (see theorem 3.4.4). So one has 100% asymptotic condensation
in the many-body ground state. Such ϕ0 , i.e., the condensate wave function, is
recognised to be the ground state wave function of the free particle in the box Ω.
(1)
3 By (3.68), γN (x, y) → L−3 at least in L2 (Ω × Ω)-sense. So the kernel of the one-
body reduced density matrix is approximately constant. In particular, it tends to
be a function of the difference x − y only, that is, to be translational invariant.
This feature justifies the name one customarily refers to (3.66), that is, the off-
(1)
diagonal long range order in the kernel of γN . Such a notion of B.E.C. too
has been introduced first by Penrose and Osanger [95].

An analogous result holds for inhomogeneous systems, e.g., a dilute gas of bosons
confined by some trapping potential. To state it properly, one has to define a suitable
notion of mean density of particles and of the size of the trap. Also, the condensate wave
function cannot be expected to be of the form (3.67), due to spatial non uniformity. We
enclose all these preliminaries – actually the rigorous basis of the Gross-Pitaevskiı̆ theory
– in the following statement.

Definition–Theorem 3.5.2 (GP theory – Lieb, Seiringer, and Yngvason (1999),


[77]). Let N0 undistinguishable spinless bosons of mass m be confined in a three-dimen-
sional trapping potential U ∈ L∞ 3
loc (R ) such that U (x) → +∞ as |x| → +∞. Define the
characteristic length L of the trap as
r

L := (3.69)

where
~2
µ := (3.70)
2m
and ~ω is the ground state energy of −µ∆ + U . Also, let the particles be paired by a
two-body, non-negative, spherically symmetric potential V with finite s-wave scattering
length a, according to the Hamiltonian
N0 ³
X ´ X
~2
HN0 = − ∆xi + U (xi ) + V (xi − xj ) (3.71)
2m
i=1 16i<j6N0

acting on L2 (R3N0 ) – each xi ∈ R3 .


62 Chapter 3. Formalisation of Bose-Einstein condensation

¬ Define the Gross-Pitaevskiı̆ energy functional


Z
GP
¡ ¢
E [ϕ] := µ|∇ϕ|2 + U |ϕ|2 + 4πµa|ϕ|4 dx (3.72)
R3

on the natural domain D of the normalised one-body wave functions on which it


makes sense,5 i.e.,
© ª
D := ϕ ∈ H 1 (R3 ) : U |ϕ|2 ∈ L1 (R3 ), kϕkL2 = 1 . (3.73)

It admits a unique (up to a phase) strictly positive minimizer ϕGP , which satisfies
the time-independent Gross-Pitaevskiı̆ equation

−µ∆ϕGP (x) + U (x)ϕGP (x) + 8πµaϕGP (x)3 = µGP ϕGP (x) , (3.74)

the GP chemical potential µGP (i.e., the Lagrange multiplier) being fixed by normal-
isation. Denote the minimum of the functional as the Gross-Pitaevskiı̆ energy

E GP := E GP [ϕGP ] . (3.75)

Such ϕGP turns out to be also in C 1 (R3 )∩L∞ (R3 ), with an exponential falloff of the
kind ϕGP (x) 6 Mα e−α|x| for any α > 0 and an appropriate Mα > 0; furthermore,
U ∈ C ∞ (R3 ) ⇒ ϕGP ∈ C ∞ (R3 ), and if U is spherically symmetric and monotone
increasing with |x|, then ϕGP is spherically symmetric and monotone decreasing.

­ Take the Gross-Pitaevskiı̆ scaling in the Hamiltonian with fixed L and a 7→ aN


such that
N aN = N0 a = const . (3.76)
This means that one considers systems with Hamiltonians
N ³
X ´ X
~2
HN = − ∆xi + U (xi ) + VN (xi − xj ) (3.77)
2m
i=1 16i<j6N

acting on HN = L2 (R3N ), with


³ N ´2 ³N ´
VN (x) = V x (3.78)
N0 N0
so that the scaled interaction VN has scattering length aN = aN0 /N .

® The gas is inhomogeneous. In view of describing its non uniform density as


N ϕGP (x)2 , define the Gross-Pitaevskiı̆ density as its average
Z
1 ¡ ¢2 ° °4
ρ̄N := N (ϕGP (x))2 dx = N °ϕGP °4 . (3.79)
N R3
Notice that ρ̄N ∼ N and ρ̄N a3N ∼ N −2 as in the corresponding homogeneous case.

¯ Let Ψg.s. g.s.


N be the ground state of HN and let EN be the ground state energy. Then
g.s.
EN
lim = E GP , (3.80)
N →∞ N

that is, the GP energy is the asymptotics of the many-body ground state energy
per particle.
5
in fact, H 1 (R3 ) ,→ L4 (R3 )
3.5. Proof of B.E.C. in a scaling limit. Stationary GP theory. 63

One can now formulate the second example of proof of asymptotic B.E.C. we an-
nounced: the case of complete condensation in the ground state of a trapped gas of
bosons.

Theorem 3.5.3 (B.E.C. for dilute trapped gases – Lieb and Seiringer (2002),
(1)
[74]). Assume the model and the scaling of theorem 3.5.2. Let γN be the one-particle
reduced density matrix corresponding to the ground state Ψg.s.
N of the N -body system.
Then
(1)
lim γN = |ϕGP ihϕGP | (3.81)
N →∞

in the trace norm.

Thus, also in this more realistic case, one has complete condensation in the ground
state, and the condensate wave function is the minimizer of the GP functional (3.72),
that is, a solution of the GP equation (3.74). Actually the GP energy functional in a box
Ω with periodic or Neumann boundary conditions is minimized by the wave function ϕ0
in (3.67), so that (3.81) reads as (3.68): in fact, we have split for convenience the unique
work [74] into the two theorems 3.5.1 and 3.5.3.
By (3.81),
(1)
γN (x, y) → ϕGP (x)ϕGP (y) (3.82)
(1)
at least in the L2 (R3 × R3 )-sense. Consequently γN (x, y) remains finite up to distances
|x − y| fixed by the extension of the function ϕGP . This is again that typical signature
of B.E.C. which goes under the name of off-diagonal long range order for the kernel of
the one-body marginal.
In standard textbooks treatment of B.E.C., off-diagonal long range order in space is
formally equivalent, in Fourier transform, to the presence of a singular term in momentum
distribution (see, e.g., Sec. 2.1 of [97]): the contribution of such a δ(p) term is thus
interpreted as a macroscopic occupation of the single-particle state with momentum
p = 0. The analysis in [74] provides a rigorous statement in this sense, in the following
form: under the assumptions of theorem 3.5.3,

b GP (p)|2
lim ρbN (p) = |ϕ (3.83)
N →∞

in L1 , where ϕ
b GP is the Fourier transform of ϕGP and
Z
(1)
ρbN (p) := γN (x, y) exp[ ip(x − y)/~] dx dy (3.84)
R3 ×R3

denotes the one-body momentum density in the ground state.


64 Chapter 3. Formalisation of Bose-Einstein condensation
Chapter 4

Role of scaling limits in the


rigorous analysis of B.E.C.

4.1 Motivations
Asymptotic analysis of condensation in many-body dilute Bose gases in the limit N → ∞
relies on the fact that (1) in practice the number of particles is so large (N ∼ 100 ÷ 1011 )
to justify the assumption of infinite size, and (2) condensation for interacting systems
(1)
requires to solve a formidably difficult many-body problem in order to access γN , while
there are techniques to get its asymptotics.
In Sec. 3.2 it has been pointed out that the nature of this limit is that of a scaling
limit: the true system S0 in the state γN0 and with Hamiltonian HN0 is regarded as an
enlarging sequence {SN }N of N -body systems, each in the state γN and with Hamiltonian
HN , made by N particles of the same kind as in SN0 , confined and coupled each other as
particles are in S0 , but with coupling strength and size of the confinement depending on
N in such a way that as N = N0 , SN is exactly S0 , i.e., HN |N =N0 = HN0 on HN = HN0 ,
and for any N , or asymptotically as N → ∞, certain physical quantities of SN in the
state γN are the same as the corresponding ones for S0 in the state γN0 .
Thus, one fixes some quantities Q(N, L, a, . . . ), i.e., the expectation values of some
observables on SN in the state γN (e.g. the ground state energy), to be constant
with N : they depend on parameters labelling the quantum state under consideration
and the Hamiltonian, such as the number N of particles, the size L of the system,
the s-wave scattering length a of the interparticle interaction, etc. The prescription
that these Q’s are kept constant in the limit fixes in principle how L, a, . . . have to
scale with N (say LN , aN , . . . ), thus making the limit for f unambiguous in the form
limN →∞ f (N, LN , aN , . . . ).
In this chapter, following Ref. [84], the role of two particular scaling limits is analysed
and commented, which turns out to be crucial in the rigorous study of B.E.C. First,
we consider the so called Gross-Pitaevskiı̆ (GP) limit. Its name itself has been
customarily adopted due to one of the main results achievable in this limit, namely,
the rigorous derivation of the Gross-Pitaevskiı̆ equation, that for half a century has been
known to be at the basis of the description of the condensate wave function in the diluted
regime.
To state it properly, a preliminary discussion is needed on the interaction and kinetic
average energies in a dilute Bose system, which is placed in Sec. 4.2. The definition of

65
66 Chapter 4. Role of scaling limits in the rigorous analysis of B.E.C.

the GP scaling and the mathematical and physical scenario behind it are then discussed
in Sec. 4.3. Remarkable rigorous results achieved through this scaling have been already
reported in Sec. 3.5. Such a discussion is meant to emphasize and clarify the centrality
of the GP scaling in the study of B.E.C.
An alternative (and a modification) to the GP scaling, is the so called Thomas-
Fermi (TF) limit: it is discussed in Sec. 4.4 in connection with its GP counterpart.
The whole treatment is in three dimensions, as the most natural choice when thinking
to the standard realizations of B.E.C. in the lab. The corresponding lower-dimensional
scenario is then sketched in Sec. 4.6.

4.2 Ground state energy asymptotics


for cold dilute Bose gases
The GP scaling has been introduced and eventually rigorously defined to be the “dy-
namical” limit of ultra-high dilution. That is, one wants that kinetic and interaction
energies are still comparable as N → ∞ – the limit is said dynamical because one can
investigate the dynamics also in the limit – and the dilution becomes infinite. To make
this precise, a control of the energy asymptotics is needed and it is not accidental that a
systematic use of such a scaling has been made only after the 1997 rigorous solution of
the ground state energy problem for cold dilute Bose gases – a task which covered the
whole second half of last century.
To fix the setting, let H be the Hamiltonian for N spinless undistinguishable bosons
of mass m enclosed in a three-dimensional cubic box Ω of side L, with suitable boundary
conditions to be specified, interacting by a spherically symmetric two-body repulsive
(i.e., non-negative) potential V , that is

N
~2 X X
H=− ∆xi + V (xi − xj ) (4.1)
2m
i=1 16i<j6N

(xi ∈ R3 ), acting on the bosonic sector (see (3.9)) of L2 (Ω⊗N , dx1 · · · dxN ), as is appro-
~2
priate for the statistics in use. Also, recall by (3.70) the position µ := 2m .
g.s.
Attempts to derive rigorously the ground state energy per particle E /N of (4.1) –
absolute and bosonic ground state energies coincide, H being permutationally symmetric
– in the physical regime of interest of high dilution and low temperature have been
performed intensively between the late 1940s and the early 1960s.
Bogolubov’s works [14, 15], perturbative derivations (based on the delta pseudopo-
tentials introduced by Fermi [47], Breit [21], and Blatt and Weisskopf [13]) both in the
simplified case of a Bose gas of hard spheres, due to Huang, Lee and Yang [58, 69, 70]
and Wu [110], and in a more general setting due to Brueckner and Sawada [25], Beliaev
[12], Hugenholtz and Pines [59], Girardeau and Arnowitt [53], as well as Lieb’s simplified
approach [72] (see Ref. [73] for a review of these early developments), all rely on special
a priori assumptions about the ground state (as the occurrence of condensation) or on
the selection of particular terms from likely divergent perturbation series.
g.s.
Dyson’s 1957 estimate [37] for an upper and a lower bound of E /N is the only
fully rigorous result of that period, though limited to the hard √ sphere case and with a
−1
lower bound that turns out to be off the mark by a factor (10 2) with respect to the
4.2. Ground state energy asymptotics for cold dilute Bose gases 67

expected asymptotics. Also, its techniques have turned out to have a crucial influence
even on the most recent mathematical treatments of BEC.
In the developments of these investigations, the non-trivial fact emerged that, as long
as the system is sufficiently diluted, the only other relevant length beyond L is the s-wave
scattering length a of the pair interaction (see appendix A of Ref. [80] for a standard
rigorous definition of a), through the dimensionless parameter ρa3 , ρ := N/L3 being the
density of the gas. Thus assumptions on V have to be made to guarantee that a exists
and is finite: it suffices (see
R [80]) and it will be assumed hereafter (with possibly more
stringent conditions) that |x|>R V (x) dx < ∞ for some R.
It was only in 1997 that the problem was closed in a wide generality, due to Lieb and
Yngvason [79] who completed the rigorous proof of an estimate of the form
¯ g.s. ¯
¯ E /N ¯
¯ ¯ 3
¯ 4πµρa − 1¯ 6 O(ρa ) . (4.2)

More precisely, after an innocent manipulation of the original results one gets the fol-
lowing.

Theorem 4.2.1 (Ground state energy upper and lower bounds for dilute Bose
g.s.
gases). Let HN be the Hamiltonian of a Bose gas as defined in (4.1) and let E be its
ground state energy. Then, as an upper bound ([77]),
à r !
g.s.
E /N ³ 1´ 4 ³ 1 ´
3
6 1− 1 + const πρa3 1 − (4.3)
4πµρa N 3 N

provided that periodic boundary conditions are imposed and that ρa3 is small enough in
the sense 43 πρa3 < (1 − 1/N )−1 , and, as a lower bound ([79]),
g.s. r
E /N 17 4
> 1 − const πρa3 (4.4)
4πµρa 3

provided that Neumann boundary conditions are imposed (plus the technical restriction
that V has finite range, otherwise the error term is possibly different) and that inequality
c1 N −51/3 < ρa3 < c2 is fulfilled for some suitable positive constants c1 , c2 .

Hence, due to the optimal choice of the boundary conditions (leading to the lowest
possible energy for the lower bound and the highest energy for the upper one), one has
g.s.
E /N
lim =1 (4.5)
N →∞ 4πµρa
ρa3 →0

for all boundary conditions and positive (compactly supported) potentials, with ρa3 → 0
not faster than N −51/3 .
Recall that ρa3 ¿ 1 is an expression of high dilution, because a ¿ ρ−1/3 , i.e., the
mean interparticle distance ρ−1/3 is much larger than the scattering length a of the
potential. Thus (4.5) amounts to E0 /N ≈ 4πµρa, provided that the gas is sufficiently
populated and diluted. Since limρ→0 E0 = 0, the system in its ground state does behave
as a gas (the ground state is not a many-body bound state).
Also, as already seen in Sec. 1.1, another length scale enters naturally the description
of the system, namely the healing length ` = (8πρa)−1/2 given by an uncertainty principle
68 Chapter 4. Role of scaling limits in the rigorous analysis of B.E.C.

µρa ∼ (~/`)2 /m. In the highly diluted regime ρa3 ¿ 1, one has ρ−1/3 /` ∼ (ρa3 )1/6 ¿ 1,
whence a ¿ ρ−1/3 ¿ `, that is, the impossibility to localize the particles relative to each
other. We remark, in passing, that ρ−1/3 ¿ ` is crucial in the proof of the lower bound
(4.4), where an intermediate length `˜ is fixed such that ρ−1/3 ¿ `˜ ¿ ` and particles
˜ which turn out to still be highly
in the box Ω are distributed into cubic cells of side `,
populated; furthermore, `˜ ¿ (ρa) −1/2 is necessary for a large enough spectral gap in
Temple’s inequality that is at the basis of inequality (4.4).
Next to this, there is the ground state kinetic energy per particle: the standard
textbooks computation for non-interacting particles gives

π2µ
E kin /N = (4.6)
L2
which amounts to the energy gap in the box, the natural energy unit of the problem.
In the more realistic case of an inhomogeneous gas confined into a trap, an analogous
picture can be proved [77] (see theorem 3.5.2), provided that the box side L and the
density ρ are substituted with the characteristic length of the trap and a suitably defined
mean density, respectively: so, as long as N is large enough and ρa3 is small enough, one
has total energy per particle ∼ ρa and kinetic energy per particle ∼ L−2 , and the ratio
between these two energies amounts to ρa/L−2 , up to a constant (4/π) of order 1.

4.3 The Gross-Pitaevskiı̆ scaling scenario


In this section the GP scaling is introduced and its main features are discussed.
The prescriptions (in the sense of Sec. 3.2 and 4.1) to this scaling are as follows.

(i ) It is a “dynamical” limit, where interaction and kinetic energies are still compara-
ble.

(ii ) It is a limit of “infinite dilution”.

If ρa3 → 0 within the rates admitted in the analysis reviewed in Sec. 4.2, then the
energies to be compared are asymptotically ρa and L−2 , so that the constraint is

ρa Na
2
= =: g = const (GP-lim) (4.7)
1/L L

or, slightly more generally, N a/L → g = const. The value of such a constant is fixed
by the parameters (N0 , L0 , a0 ) of the true physical system under consideration: g =
N0 a0 /L0 . Since
ρa3 = g 3 N −2 , (4.8)

this is actually a limit of infinite dilution as N → ∞. Also, ρa3 vanishes within the range
g.s.
of the validity of the asymptotics E /N ≈ 4πµρa reviewed in the previous section.
Conversely, notice that ρa3 ∼ N −2 is the only vanishing rate of the gas parameter
guaranteeing the ratio ρa/L−2 to be constant. To keep the quantity Q(N, L, a) := N a/L
(asymptotically) constant, both L and a have to be scaled accordingly: LN simply
accounts for the rate of the box enlargement with N , while aN scales by modifying the
interaction potential VN in a way that has to be precised later.
4.3. The Gross-Pitaevskiı̆ scaling scenario 69

Table 4.1. Typical experimental data of Bose-Einstein condensate of alkali


vapours. The gas being inhomogeneous, the quoted mean densities have been com-
puted by the authors by integrating over the whole condensate density profile. The
quoted scattering lengths have been determined in the same experiments, but the
87
Rb one which is taken from later measurements [109]. Scattering length are ex-
pressed in terms of a. u. (Bohr radius) aBohr . The ratio g between total and kinetic
ground state energy per particle is computed via (4.8).

N ρ̄ (cm−3 ) a (aBohr ) ρ̄a3 g


87 Rb Wieman, Cornell 4 · 106 2.5 · 1012 103 ± 5 4.0 · 10−7 186
et al (1995) [6]
23 Na Ketterle et al 5 · 105 4 · 1014 92 ± 25 4.7 · 10−5 226
(1995) [34]
41 K Inguscio et al 104 6 · 1011 78 ± 20 4.2 · 10−8 1.6
(2001) [89]

With no claim of an exhaustive review of the B.E.C. vast experimental landscape,


typical values of the ratio ρa/L−2 in experiments with alkali vapours are collected in
Table 4.1 , where three historically pivotal B.E.C. realizations are quoted.
By no means fixing the ratio (4.7) amounts to claim that the two energies have the
same order of magnitude. Actually, in a condensate the kinetic energy is usually highly
dominated by the interaction energy. One needs to keep these energies comparable,
although their ratio is far from 1, to end up in the limit with a system where the
dynamics can still be investigated. Instead, when one is not interested in the dynamical
properties of the Bose gas, another scaling may turn out to be useful, where the kinetic
term in the Hamiltonian is neglected. If this is the case when the ratio ρa/L−2 is sent to
infinity and the dilution ρa3 → 0, then the scaling goes under the name of Thomas-Fermi
limit: we discuss it in Sec. 4.4 in connection with the GP limit.
Let us denote the scaling of the scattering length by aN ∼ 1/ηN , with ηN > 0. In
principle ηN is allowed to diverge, be constant, or vanish with N . Then (4.7) fixes the
characteristic length, the mean density of the system, and the mean energy per particle
of the system, as in the first row of Table 4.2. The typical power-law choice ηN = N α is
made explicit in the second line of the same Table and then specified when α = 0, 32 , 1. In
particular, the α = 32 case is the thermodynamic limit (ρ = const) within the GP scaling;
notice that conversely the N/L3 = const prescription is not enough to fix alone the whole
scaling as N → ∞. In general, the choice of ηN reflects on the limiting behaviour of the
interaction energy per particle – see the right column of Table 4.2 – which is still clearly
visible in the limit, being by (4.7) comparable to the energy gap in the box. Anyway,
since a change ηN 7→ ηeN is just a rescaling of the length unit, any such realization turns
out to be mathematically equivalent before the limit.
Thus, the scaling is summarized as

a0 (N/N0 )L0
aN = , LN = , ηN0 = 1. (4.9)
ηN ηN
70 Chapter 4. Role of scaling limits in the rigorous analysis of B.E.C.

Table 4.2. Possible realizations of the GP scaling

aN LN ρN = N/L3N 1/L2N ∼ ρN aN
∼ 1/ηN ∼ N/ηN ∼ ηN3 /N 2 ∼ (ηN /N )2
∼ N −α ∼ N 1−α ∼ N 3α−2 ∼ N 2α−2
fixed ∼N ∼ N −2 ∼ N −2
∼ N −2/3 ∼ N 1/3 fixed ∼ N −2/3
N −1 fixed ∼N fixed

The scaling L0 7→ LN is implemented by modifying the domain of the Hamiltonian


N
X X
¡ ¢
HN = − µ∆xi + VN (xi − xj ) (4.10)
i=1 16i<j6N
¡ ¢
to be a suitable dense in L2 (ΩN )⊗N , dx1 · · · dxN , ΩN being the three-dimensional box
of side LN . Of course, for the analogous inhomogeneous case an appropriate confining
potential UN is taken into account, with characteristic length LN , and the Hamiltonian
is
XN X
¡ ¢
HN = − µ∆xi + UN (xi ) + VN (xi − xj ) (4.11)
i=1 16i<j6N

acting on L2 (R3N , dx1 · · · dxN ).


−1
The scaling a0 7→ aN ∼ ηN is implemented by some redefinition of the pair potential
VN with respect to the true positive compactly supported spherically symmetric pair
potential V with scattering length a0 . If aN → 0 one has a short-range limit and, in
order for the interaction not to disappear, a hard-core limit has to be realized as well. Due
to a straightforward rescaling in the definition of the scattering length, one immediately
sees that
2
VN (x) = ηN V (ηN x) (4.12)
has exactly scattering length aN = a0 /ηN , provided that V has scattering length a0 .
Assume, in addition, that V is bounded in the L1 norm: then ηN 3 V (η x) → b δ(x) in
N 0
−1
the sense of distributions, and VN ≈ ηN b0 δ(x), that is, VN is an
R approximate delta
−1
function on the same scale ηN of the scattering length; here b0 := V = kV k1 gives the
first Born approximation of a0 , and the Spruch-Rosenberg inequality [105] states

b0
a0 6 . (4.13)
8πµ

Let us comment on the most useful choice for ηN . The N -body wave functions which
HN acts on are supported (or essentially supported) in (ΩN )⊗N and a typical task in the
B.E.C. analysis is to express them as a product of suitable one-particle wave functions
supported (or essentially supported) in ΩN . As summarized in Table 4.2, depending on
whether N/ηN → 0 or N/ηN → ∞, the box (or the trap) where the gas is confined
in shrinks to a point or enlarges to become the whole R3 and, correspondingly, the
mean kinetic energy per particle diverges or vanishes. Thus, in these two mathematical
regimes, one deals with one-particle wave functions that are normalised in the L2 sense,
4.3. The Gross-Pitaevskiı̆ scaling scenario 71

but with H 1 norm that is necessarily either diverging or vanishing. Similarly, the fixed
L2 normalisation forces the L∞ norm of these one-particle wave functions to vanish or
to diverge respectively.
Then the choice
ηN = N/N0 (4.14)
emerges to be the most appropriate one for explicit computations, because the one-particle
wave function L2 , L∞ , H 1 norms do not scale with N . This corresponds to fix the box (or
the trap), filling it with more and more particles. This way, the scaled pair potential has
scattering length a0 N0 /N and is an approximate delta function on that scale, weighted
with a 1/N pre-factor sometimes interpreted as a mean field pre-factor – although its
origin is not a mean field treatment, where conversely the particle pairing would be
described with a long range potential with vanishing strength.
Let us also comment on possible alternatives to (4.12). There, the short-range +
hard-core limits are performed simultaneously with N , as N → ∞. A different choice of
reasonable generality would be to realize these limits with different rates with N , which
is the same to say that scattering length and effective range of VN scale differently. In
the literature, this is often done by scaling
1 3
VN (x) = ξ V (ξN x) (4.15)
ηN N
with ξN0 = ηN0 = 1 and ξN = O(ηN ) as ηN , ξN → ∞. In fact, under the further
assumption that V is bounded, one can prove that

 a0 if ξN /ηN → 1
lim ηN aN = b (4.16)
N →∞  0 if ξN = o(ηN ) .
8πµ
Proof of (4.16) is an adaptation of lemma A.1 in Ref. [42] and is postponed to the end
of this chapter (Sec. 4.7). Thus, ξN 3 V (ξ x) is an approximate delta function on the
N
−1 −1 3 −1
scale ξN and (4.15) defines a potential VN (x) = ηN ξN V (ξN x) ≈ b0 ηN δ(x) whose range
−1 −1
∼ ξN is possibly much larger than its scattering length ∼ ηN . When ηN = N/N0 , the
N -dependence that remains to be prescribed in ξN = O(N ) can be unrestrictively chosen
to be ξN = (N/N0 )β , with 0 < β 6 1: the pair potential has scattering length ∼ N −1
and range ∼ N −β and is an approximate delta function on the scale N −β , weighted with
the 1/N “mean field” pre-factor.
Despite the deep mathematical interest of taking the limit when the two-body inter-
action scales as (4.15), one has to be warned to consider it a proper GP limit. Indeed,
by (6.38) and (4.16), one has
N aN N →∞ N0 b0 N0 a0
−−−−→ ≥ (4.17)
LN L0 8πµ L0
that is, (4.7) is macroscopically violated by a finite quantity. What happens with the
interaction (4.15), and can be monitored concretely [42, 44] in the realization ηN ∼ N ,
is that the kinetic energy is decreased with respect to the total energy, when compared
with the case of the interaction (4.12): any particle feels only part of the potential when
scattering against another particle, having VN a much longer range than the scattering
length, and the effective two-body process is its semiclassical approximation. The mod-
ification of the N -body wave function when any two particles are close each other turns
72 Chapter 4. Role of scaling limits in the rigorous analysis of B.E.C.

Figure 4.1. Curves of scaling, as N → ∞, in the space of the relevant pa-


rameters (N, L, a, . . . ). The physical system under investigation lives at the point
(N0 , L0 , a0 , . . . ). Curves (i) and (ii) represent the GP scaling in the form aN ∼ N −1
and aN = const respectively (third and fifth rows of Table 4.2). A generic scaling
(N, L, a, . . . ) → ∞ may be curve (iii).

out to be less pronounced than in the much more singular (4.12) and, consequently, the
H 1 norm decreases.
To conclude this section, let us now discuss the topic of the validity of the scenario
described so far. Once the (GP) scaling is chosen and a limit result is possibly achieved,
the question arises on how well such a result approximates the corresponding physical
quantity of the true system under consideration. To this aim, a control is needed of
the error terms in the asymptotics of that specific quantity. Unfortunately, a rigorous
treatment of the error terms for any of the quantities investigated in the GP limit turns
out to be highly non-trivial and at present it is a major open problem. Rigorous results
currently known are essentially asymptotic results, usually obtained in quite weak or
indirect ways, though the task of addressing error terms is receiving more and more
interest.
The perspective one should have in checking the validity of the (GP) scaling, af-
ter a control of the errors has been established, is the following. In the notation of
Sec. 4.1, let again f be any of the physical quantities of interest and let f (N, L, a, . . . )
denote its dependence on the relevant parameters that in the original system amounts
to N0 , L0 , a0 , . . . . One seeks information on f (N0 , L0 , a0 , . . . ) through the limit

lim f (N, LN , aN , . . . )
N →∞

that is, through a special restriction of f to a curve parametrized by N according to the


specific scaling in use, GP in our case (4.9). Figure 4.1 gives a pictorial idea of this. Let
us assume that an asymptotic result is achieved of the form

lim f (N, LN , aN , . . . ) = F (4.18)


N →∞

with an error control of the form


|f (N, LN , aN , . . . ) − F|
6 E(N ) = o(1) (4.19)
|F|
4.4. From the GP to the Thomas-Fermi scaling 73

(of course, (4.19) is necessarily formal: |f − F| can be though as an absolute value


for scalars, like the energy, or a norm or a whatever weaker distance for reduced den-
sity matrices). Then the problem of to what extent the true f (N0 , L0 , a0 , . . . ) can be
approximated by F, relies on how satisfactorily small the error E(N0 ) is. Indeed, by
construction, 

f (N, LN , aN , . . . ) = f (N0 , L0 , a0 , . . . ) . (4.20)
N =N0

So the overall validity of the GP scaling is ultimately based on the possibility of achieving
small relative errors E(N0 ) for the largest number of physical quantities of interest.
Distinct scalings are reasonably expected to be more or less appropriate depending
on the different investigated quantities and the GP scaling itself should not be thought
to be the absolute best (although it is the scaling through which major results have
been achieved, as theorems 3.5.1, 3.5.2, 3.5.3, 5.3.1). Admittedly, the discussion above
should have pointed out that at least the GP scaling is central in the study of dilute
Bose systems, because it is the scaling which guarantees that the kinetic energy does not
disappear in the limit (with respect to the total energy) and allows an investigation of
the dynamics in the limit.
This leads to the last noticeable feature of this scenario in connection with B.E.C.,
namely, the GP scaling accounts only for depletionless condensation. Notoriously [33, 71],
when condensation occurs, due to the interaction a fraction of atoms do not occupy
the condensate even at zero temperature because of correlation effects: this quantum
depletion (to distinguish it from the thermal depletion that enhances this phenomenon at
positive temperatures), that is, this decrease in the condensate fraction, can be predicted
within the Bogolubov theory to be proportional to (ρa3 )1/2 . Hence, the GP limit cannot
take into account depletion.
As we have already commented, this does not mean that it cannot describe B.E.C. with
depletion, but only that asymptotic results by construction cannot be expected to ac-
count for it. Depletion is zero in the limit and (small and) positive before the limit.
If a physical quantity f (N0 , L0 , a0 , . . . ) has a sufficiently small error E(N0 ) in the GP
limit F, one has to deduce that depletion does not affect f (N0 , L0 , a0 , . . . ) strongly and
the GP scaling turns out to be an appropriate tool. However, one might expect that
other quantities of interest do depend heavily on the depletion and are not conveniently
accessible through the same scaling.

4.4 From the GP to the Thomas-Fermi scaling


One may adopt another kind of limit N → ∞ still guaranteeing infinite dilution, but
where the ratio between the interaction and the kinetic energy is diverging. In this case
the two-body interaction and the box (or trap) length have to scale with N according to
the following so-called Thomas-Fermi (TF) limit

ρa Na
2
= → ∞
1/L L (TF-lim) (4.21)
ρa3 → 0

the generic realization of which, for a homogeneous gas, is

a0 (N/N0 )L0
aN = , LN = , ηN0 = ξN0 = 1 (4.22)
ηN ξN
74 Chapter 4. Role of scaling limits in the rigorous analysis of B.E.C.

Table 4.3. TF scaling. First row: generic realization for a homogeneous Bose gas.
In particular: fixed box choice (second row); fixed box + power law choice (third
row). Fourth row: realization of the TF scaling for inhomogeneous Bose gases in a
fixed trap, with mean density (4.30).

aN LN ρN = N/L3N ρN a3N 1/L2N ρN aN


∼ 1/ηN ∼ N/ξN 3 /N 2
∼ ξN (ξN /ηN )3 /N 2 ∼ (ξN /N )2 ∼ ξN /(ηN N 2 )
3

∼ 1/ηN fixed ∼N ∼ N/ηN 3 →0 fixed ∼ N/ηN → ∞


∼ N −α fixed ∼N ∼ N 1−3α → 0 fixed ∼ N 1−α → ∞
∼ N −α fixed ∼ N s/(s+3) ∼ N s/(s+3)−3α → 0 fixed ∼ N s/(s+3)−α → ∞

with the conditions


ξN ξN
→ ∞, → 0. (4.23)
ηN ηN N 2/3
Indeed
N aN N0 a0 ξN
=
LN L0 ηN
is prescribed to diverge and
³ N a ´3 ³ ξ ´3
0 0 N
ρN a3N =
L0 ηN N 2/3
is prescribed to vanish.
The generic scalings L0 7→ LN and a0 7→ aN in the Hamiltonian HN are realized as
for the GP case. Again, it is natural to fix the realization ξN = N/N0 , in order to keep
the box side fixed and to deal with one-body wave function whose L2 , H 1 and L∞ norms
are not scaling with N (see the discussion at the end of Sec. 4.3). Then, with no loss
of generality, one can set ηN = (N/N0 )α : then prescriptions (4.23) imply 31 < α < 1.
A completely analogous scenario holds for inhomogeneous gases, where a suitable mean
density has to be adopted. The only difference with the GP treatment of inhomogeneous
gases, in this respect, is that such a mean density turns out to scale differently with N
(for instance differently than ρ̄ ∼ N in the fixed trap GP case), but anyway in such a
way that (4.21) is satisfied. Table 4.3 summarizes all that.
One can think the TF scaling as another kind of curve in the (N, L, a, . . . )-domain
starting from the physical point (N0 , L0 , a0 , . . . ) and going to infinity, just like curve
(iii) in Fig. 4.1. Thus, the same considerations hold concerning the validity of such a
scaling, as discussed in Sec. 4.3. In particular, when comparing the GP and the TF
scaling keeping the box (or the trap) fixed, the first corresponds to scale the scattering
length in such a way that N aN = const, whereas the second corresponds to scale it much
more slowly so that N aN → ∞.
In the GP limit, the ground state energy of dilute trapped gases is recovered by
minimizing the GP energy functional (3.72). In the TF limit, the same role is played
by the functional (4.24) obtained by neglecting the kinetic term in the previous one. Of
course, the two minima do not coincide for a fixed value of the coupling in front of the
quartic term.
Nevertheless, recall that such a coupling is constantly 4πµa0 = 4πµN aN /N0 and
that, when both the GP and the TF scaling are performed in the fixed box realization,
4.4. From the GP to the Thomas-Fermi scaling 75

the first amounts to N aN = const, the second to N aN → ∞. So the interest arises in


comparing the two minima (with and without the kinetic term in the functional) letting
N aN → ∞. This goes under the name of Thomas-Fermi limit of the GP energy and the
result is that the two energies are indeed asymptotically the same.

Definition–Theorem 4.4.1 (TF limit of the GP energy – Lieb, Seiringer and


Yngvason [78]). Let U ∈ L∞ 3
loc (R ) such that U (x) → +∞ as |x| → +∞. Also, let
U be locally Hölder continuous and asymptotically equal to some function U that is
homogeneous of some order s > 0, i.e., U(λx) = λs U(x) ∀λ > 0.1

¬ Let ν > 0; after suppressing the kinetic term in the GP functional (3.72), define
the Thomas-Fermi energy functional
Z
TF
¡ ¢
Eν [γ] := U γ + 4πµνγ 2 dx (4.24)
R3

on the domain D of the normalised one-body “densities”


 1 3 2 3

 γ ∈ L (R ) ∩ L (R ) : 
D := γ(x) > 0, U γ ∈ L1 (R3 ), . (4.25)
 
kγkL1 = 1

Notice that when ν is taken to be any νN = N aN /N0 , then the EνTFN


’s are a sequence
of functionals such that EνTF
N
has exactly the form of E GP without the gradient term.
0

­ For each ν, EνTF admits a unique minimizer γνTF given explicitly by


£ TF ¤
TF
µν − U (x) +
γν (x) = (4.26)
8πµν

where [ · ]+ is the positive part and the the TF chemical potential µTF
ν (i.e., the
Lagrange multiplier) is fixed by normalisation. Denote the minimum of each func-
tional as the Thomas-Fermi energy

EνTF := EνTF [γνTF ] . (4.27)

® The chemical potential has the explicit expression below and is given by a varia-
tional principle:
° TF °2
µTF TF
ν = Eν + 4πµν γN
° °
2
Z
¡ ¢
= inf U γ + 8πµνkγk∞ dx .
D R3

Furthermore, ν 7→ µTFν is a concave function that is monotonically increasing to


infinity and µTF
0 = 0.

¯ Compare each EνTF with the corresponding family of GP functionals EνGP where the
coupling is allowed to vary, namely,
Z
¡ ¢
EνGP [ϕ] := µ|∇ϕ|2 + U |ϕ|2 + 4πµν|ϕ|4 dx (4.28)
R3
1
This is the case, e.g., for a harmonic trap. Such an assumption is for technical convenience, see
Ref. [78] for a precise definition of locally Hölder continuity and homogeneity.
76 Chapter 4. Role of scaling limits in the rigorous analysis of B.E.C.

on the variational domain (3.73), and with minima EνGP . Notice that, as far as
ν = νN = N aN /N0 , (i) one recovers the original GP functional and GP energy
when N = N0 , (ii) along the GP scaling N aN = const = N0 a0 , these EνGP
N
’s and
EνGP
N
’s are identically the same.

° With this notation, one can probe the behaviour of the GP functional under other
types of scalings, in particular, under the TF scaling that in this setting is realized
as N aN → ∞ (“strong coupling” regime). The result of this TF limit of the GP
energy is that
E TF
lim νGP = 1 . (4.29)
ν→∞ Eν

As a consequence, when performing the TF scaling in the many-body Hamiltonian,


the ground state energy per particle is recovered in the limit just by minimizing the
energy functional where the kinetic term has been suppressed and the coupling diverges.
This can be made rigorous through the same techniques used in the treatment of dilute
trapped Bose gases in the GP limit [76] and the result is the following.
Definition–Theorem 4.4.2 (Ground state energy of dilute trapped gases in
the TF limit). Let N0 undistinguishable spinless bosons be confined in a trapping
potential U and be paired by a repulsive two-body interaction V . In particular, assume
U ∈ L∞ 3
loc (R ) such that U (x) → +∞ as |x| → +∞. Also, let U be locally Hölder
continuous and asymptotically
p equal to some function U that is homogeneous of some
order s > 0. Take L := 2µ/(~ω) to be the characteristic length L of the trap as in
(3.69). Also, assume V : R3 → R to be non-negative and spherically symmetric, with
finite s-wave scattering length a. The Hamiltonian of the system is then HN0 as in (3.71).
¬ Scale HN0 to HN , the N -body Hamiltonian (4.11), according to the TF prescription
(4.21) with the standard fixed-trap realization. To do that, a suitable mean density
of the inhomogeneous gas has to be defined: the appropriate choice (with the
notation of theorem 4.4.1) turns out to be the Thomas-Fermi density
Z ° °2
TF 1 ¡ TF ¢2 ° °
ρ̄N := N γ N a0 (x) dx = N °γ TF
N a0 ° . (4.30)
N R3 N0 N0 2

It can be seen explicitly how it scales with N , using the form (4.26) of the TF min-
imizer and the assumption that U is asymptotically homogeneous of some positive
order s: one finds
ρ̄TF
N ∼N
s/(s+3)
. (4.31)
Hence, the scattering length aN has to scale according to (4.21), in such a way that
N aN → ∞
(4.32)
ρ̄TF 3
N aN → 0
s
as N → ∞. An unrestrictive power-law choice aN ∼ N −α then imposes 3(s+3) <
1
α < 1. Notice that, since s > 0, it necessarily includes the range 3 < α < 1 that is
characteristic of the TF scaling for a homogeneous gas.
g.s. g.s.
­ Let EN be the ground state energy of HN . In the TF limit EN is recovered as
g.s.
EN /N
lim TF
= 1. (4.33)
N →∞ Eν
N
4.5. GP vs TF limit in investigating B.E.C. 77

NEAR-RESONANT SPATIAL IMAGES OF CONFINED . . . R55


0.4 C

Optical Density 0.2

0.4
B

0.2

0.0

0.4 A

0.2
Near-resonant images of column density profiles of so-
dium condensates. The solid curves result from mean-field calcula-
0.0 tions with N 0 580 000, 350 000, and 850 000 atoms in the conden-
0 200 400 600 800 1000 sates, respectively, after propagation through the optical system.
These number correspond to peak atomic column densities of 442,
Position (µm)
1072, and 1826 atoms/mm2. The asymmetric harmonic-oscillator
Density profiles of three samples of 41K potential has v x 5 v y 52050 rad/s and v z 5170 rad/s. Arrows
after 15 ms of expansion, showing the transition give Thomas-Fermi sizes in the z direction. Error bars represent
to BEC. (A) Thermal sample at T 5 250 nK. (B) statistical errors. By comparing the theoretical curves to the data we
Mixed sample at T 5 160 nK. (C) Almost pure obtain x 2 values of 0.71, 1.26, and 1.33. The upper and lower
condensate. The lines are the best fit with a dashed curves in each case correspond to N 0 values shifted by
Gaussian for the thermal component and with an 610%.
inverted parabola for the condensate component.

Figure 4.2. Fit of typical experimental data in the Thomas-Fermi regime. Left:
experiment in [89]. Right: experiment in [108].

4.5 GP vs TF limit in investigating B.E.C.

It has to be pointed out that, unlike the GP limit, the TF limit does not bring any
conclusion on B.E.C., defined in terms of reduced density matrices. Indeed, this tool
does not allow to deal with wave functions, but with density profiles only, namely, with
the diagonal of the kernel of marginals. It is remarkable, however, that experimentalists
often claim that their data on condensates fit the density profile of the TF density (4.26),
that has the form of the typical “inverted parabola” when the trap is given by a harmonic
potential – see, for example, Fig. 4 in Ref. [89], or Fig. 2 in Ref. [108], reproduced in
Fig. 4.2.
This happens to be the case because at least in certain experimental regimes the two
scalings turn out to be almost equivalent. In current terms, this means the following. If
one proceeds from the physical point (N0 , L0 , a0 ) along the GP curve (and if the usual
technical assumptions on the trapping and interaction potentials are fulfilled), then in
the limit N → ∞ the system Bose-condenses in the ground state, the condensate wave
function is the corresponding ϕGP , and the ground state energy per particle is given
by the GP functional evaluated in ϕGP . If one moves along the TF curve, then in the
limit N → ∞ the ground state energy is provided by the TF functional evaluated in
the corresponding γ TF . Some experimental regimes are such that both the two (a priori
distinct) limiting results are close enough to the true expectation value of the observable
of interest at the physical point (N0 , L0 , a0 ) and, hence, are close to each other, so that
|ϕGP (x)|2 ≈ γ TF (x). Thus, the condensate profile turns out to fit with the TF in good
agreement.
78 Chapter 4. Role of scaling limits in the rigorous analysis of B.E.C.

Actually, as discussed before in this section, what has to be stressed in this comparison
is the crucial role of the quantity N0 a0 /L0 for the true physical system. By (4.29), when
N aN /LN is large, then GP and TF scalings reproduce both the ground state energy. If
at the physical point (N0 , L0 , a0 ) where the two scalings start from (i.e., the two distinct
curves in the diagram of Fig. 4.1) the parameter N0 a0 /L0 is already “large enough”,
then one expects that taking N → ∞ while g = N aN /LN = const (GP lim) or while
N aN /LN → ∞ (TF lim) gives essentially the same description (though only the GP
scaling gives access to the full kernel of the reduced density matrices, by which one
defines B.E.C.).
Typical experimental values of N0 a0 /L0 are listed in the last column of Table 4.1.
Notice that in the aforementioned 41 K experiment [89], where the density profile of the
condensate fits with the TF inverted parabola, one has N0 a0 /L0 ∼ 1.6; hence, how large
this parameter has to be in order that GP and TF scaling give the same description is a
quite subtle matter.
Recall that in Table 4.1 such values are computed via (4.8), that is, N0 a0 /L0 =
(ρ0 a30 N02 )1/3 : this is because in the same experiment one could measure directly a0
and N0 , while for ρ0 one takes a mean value ρ̄ by averaging the experimental particle
distribution. To have a check, one can alternatively compute L0 itself via
r

L0 ∼ , (4.34)

that is, through the details of the trapping potential U . In [89] U is actually a magnetic
trap consisting of a Ioffe-Prichard potential in quadrupole Ioffe configuration (see [46]
for details), which in the current setting reads
m¡ ¢
U (x, y, z) = ωrad (x2 + y 2 ) + ωax z 2
2
with radial and axial frequencies ωrad /(2π) = 200 Hz and ωax /(2π) = 16 Hz, and with
m = atomic mass of 41 K. Then the ground state energy ~ω of −µ∆ + pU (r) is ~ω =
~ωrad + 12 ~ωax and by (4.34) the characteristic length of the trap is L0 = 2µ/(~ω) (one
neglects here the anisotropy of the trap). So altogether N0 a0 /L0 ∼ 1.54, comparable to
the value of 1.6 listed in Table 4.1.
An agreement between GP and TF density profiles is observed even in a much larger
regime of N0 a0 /L0 : this is the case, for instance, of the experiment with 23 Na in [108]:
there one has N0 = 8 · 104 , a0 = 27.5 Å, ωrad = 2050 s−1 , ωax = 170 s−1 , m = m(23 Na),
whence N0 a0 /L0 ∼ 193. (The scattering length in this case differs considerably from the
one quoted in Table 4.1 because of the distinct hyperfine levels of 23 Na involved in the
condensation phenomenon.)

4.6 Low-dimensional scalings


Despite that the present discussion has been centred in the natural three-dimensional
setting, the low-dimensional behaviour of dilute Bose gas in elongated or disc-shaped
traps deserves great interest.
The Thomas-Fermi scaling as discussed in Sec. 4.4 has a one- and two-dimensional
counterpart too – see Ref. [75, 78, 76]. It has also turned out to be a fundamental tool
in the study of trapped rotating condensates (see Ref. [32, 24] and references therein).
4.7. Appendix: scaling laws for the scattering length 79

In particular, although entering the peculiarities of the low-dimensional settings goes


beyond the scope of this chapter, where instead stress has been given to the role of
the scalings themselves, we mention that in the two-dimensional case the TF scaling is
especially relevant, for it is the natural and the more appropriate alternative to the GP
scaling. Indeed (see the discussion in [78]), in the latter the scattering length turns out
to decrease exponentially with N and the two-dimensional analog of the fixed quantity
g in (4.7) is typically very large, so that a limit g → ∞ as N → ∞ (i.e., the TF limit in
two dimensions) is meaningful.

4.7 Appendix: scaling laws for the scattering length


In this appendix, we give the proof of the technical property (4.16), essentially by adapt-
ing the original lemma A.1 in Ref. [42]. Restating it in the form (4.35), one can appreciate
the substantial content of the original result; that is, whatever the realization of a delta-
like potential ξN 3 V (ξ x) ≈ b δ(x) is, provided that ξ = o(η ), then the scattering length
N N N
−1 3
of ηN ξN V (ξN x) does not scale like a0 /ηN , as in the proper GP case ξN ∼ ηN , but instead
b
like ( 8πµ )/ηN , i.e., with the same behaviour but a higher constant. So this is a feature
which does not depend on the particular realization VN (x) = N 3β−1 V (N β x), 0 < β < 1.

Theorem 4.7.1. Let V : R3 → R be bounded, compactly supported, spherically sym-


−1 3
metric, positive, and smooth. Let b be its L1 -norm. Let VN (x) ≡ ηN ξN V (ξN x) with
ηN , ξN → ∞ as N → ∞ in such a way that ξN = o(ηN ). Then, if VN is regarded as a
~2
two-body quantum mechanical potential for particles of mass m and µ := 2m , the s-wave
scattering length aN has the asymptotic behaviour

b
lim ηN aN = . (4.35)
N →∞ 8πµ

Proof. Let supp(V ) = BR , the ball of radius R centred at the origin. Then supp(VN ) =
BR/ξN . One has
kV k1 b
bN ≡ kVN k1 = = ,
ηN ηN (4.36)
ξ 3
kVN k∞ = N kV k∞ ,
ηN
so that (6.38) reads
8πµηN aN 6 b . (4.37)

Also recall [80] that Z


8πµaN = VN (x)fN (x)dx (4.38)
R3

where fN is the spherically symmetric solution of


½ ¡ ¢
− 2µ∆ + VN fN = 0
. (4.39)
fN (x = ∞) = 1

Let r = |x|, vN (r) ≡ VN (x) and fN (x) ≡ gN (r)/r: then gN (r) > 0 ∀ r > 0,

00 1
−µgN (r) + vN (r)gN (r) = 0 ,
2
80 Chapter 4. Role of scaling limits in the rigorous analysis of B.E.C.

and for r > R/ξN one has gN (r) = r − aN (whence g 0 (r) = 1). For 0 6 r 6 R/ξN ,
R
− aN − gN (r) = gN (R/ξN ) − gN (r)
ξN
Z R/ξN
0
= gN (x)dx
r
Z R/ξN ½ Z R/ξN ¾
0 00
= gN (R/ξN ) − gN (y) dy dx
r x
³R ´ Z R/ξNZ R/ξN
vN (y)
= −r − gN (y) dy dx
ξN r x 2µ
R
6 −r,
ξN
then gN (r) > r − aN . So
(
1 − aN /|x| , |x| > aN
fN (x) > (4.40)
0 , |x| 6 aN

and (4.38) gives


Z ³ aN ´
8πµaN > VN (x) 1 − dx
|x|
|x|>aN
Z Z
aN
= VN (x)dx − VN (x) dx
|x|
|x|>aN |x|>aN
Z Z (4.41)
bN
> VN (x) dx − VN (x) dx
R3 |x|
N b
|x|> 8πµ
Z Z Z
dx
> VN (x) dx − VN (x)dx − bN VN (x) .
R3 R3 |x|
N b
|x|6 8πµ

Due to (4.36), the last two summands in the r.h.s. of the above inequality are estimated
as
Z
4 ³ bN ´3
VN (x) dx 6 π kVN k∞
3 8πµ
bN
|x|6 8πµ (4.42)
3 ³
kV k1 kV k∞ 1 ξN ´3 ³
C1 ξN ´3
= 3

384πµ ηN ηN ηN ηN
and as
Z Z
dx 1 3 dx
bN VN (x) = ξN V (ξN x)
R3 |x| R3 ηN |x|
Z
bN V (x) dx
=
ηN R3 |x|/ξN
Z
kV k1 ³ ξN ´ dx (4.43)
6 kV k∞
ηN ηN |x|
|x|6R
2πR2 kV k1 kV k∞ ³ ξN ´ C2 ³ ξN ´
= ≡ ,
ηN ηN ηN ηN
4.7. Appendix: scaling laws for the scattering length 81

respectively, so by plugging (4.42) and (4.43) into (4.41), one gets


³ ξ ´3 ³ξ ´
N N
8πµηN aN > b − C1 − C2 . (4.44)
ηN ηN

Since ξN /ηN → 0, (4.37) and (4.44) give (4.35).


82 Chapter 4. Role of scaling limits in the rigorous analysis of B.E.C.
Chapter 5

Strengthened convergence for


reduced density matrices in the
derivation of the Gross-Pitaevskiı̆
equation

5.1 Time-stability of condensation


An issue which is extensively discussed in the Mathematical Physics community working
on B.E.C.-related problems is that of time-stability of condensation. Its physical
formulation is: given a condensate Bose gas confined in a trap, investigate whether the
system continues evolving in time as a condensate once the trap is removed and the
gas is left to expand under the interparticle repulsive interaction only. In general this
turns out to depend on the chemical nature of the particles, on the conditions under
which the condensate is initially prepared (trapping potential, dilution, etc.), and on the
interaction among particles causing their expansion after removing the confinement.
Persistence of B.E.C. in a many-body system evolving in time, according to defini-
tions 3.3.2 and 3.4.1, means that the corresponding time-dependent one-particle reduced
density matrix is essentially a projection at any time, thus identifying the time-dependent
one-body condensate wave function, say ϕt .1 Experiments can reveal |ϕt |2 , thus its be-
haviour with t can be studied. Hence, one ends up with the mathematical counterpart
of this problem, that is, the problem of deriving the law by which the condensate wave
function evolves in time.
Heuristic arguments and theoretical investigations yet based on ill-posed derivations
or unproved assumptions, dating back the 50’s of last century, as well as the fit of most
recent experimental data, lead to a specific dependence: if suitably prepared and left
free to expand, the condensate preserves its condensed phase and the condensate wave
function is the solution of a cubic nonlinear Schrödinger equation which goes under the
name of time-dependent Gross-Pitaevskiı̆ equation. The mathematical task is then to
derive rigorously such a nonlinear one-body equation, starting from the linear many-body
Schrödinger equation governing the evolution of the whole system.
1
This is not a partial derivative, which instead would be denoted by ∂t ϕt . The established convention
in the related literature is to denote the time and the space dependence “asymmetrically”, by ϕt (x).

83
84 Chapter 5. Strengthened convergence for reduced density matrices

This chapter, following Ref. [85], reports the state of the art of this issue together
with some recent improvements.
In Sec. 5.2 we discuss the standard way to derive it formally. In comparison to that,
Sec. 5.3 reports the most recent rigorous results in the derivation of the cubic nonlinear
Scrödinger equation of interest in the present context. In the B.E.C. terminology, they
deal with a system where asymptotic 100% condensation is preserved in time, hence,
they provide the following diagram:
partial trace (k) N →∞
γN −−−−−−−→ γN −−−−→ |ϕihϕ|⊗k
  
many-body   GP−equation
linear dynamics y y y (5.1)
partial trace (k) N →∞
γN,t −−−−−−−→ γN,t −−−−→ |ϕt ihϕt |⊗k

The study of a diagram like (5.1) is part of an analysis which goes beyond this B.E.C.-
related topic and concerns the derivation of effective nonlinear (classical or quantum)
evolutionary one-body equations from the linear many-body dynamics. It involves a
general strategy which has been developed in a series of investigations in the last 30
years. We review the scheme of this analysis in Sec. 5.4, together with a historic survey
in Sec. 5.5.
The core of this chapter is then presented in Sec. 5.6, 5.7, and 5.8, where, following
[85], we show some improvements in the topologies for the convergence of time-dependent
marginals, providing a strengthened unified version of the recent results discussed in
Sec. 5.3.

5.2 Formal derivation of the time-dependent


Gross-Pitaevskiı̆ equation
Before proceeding along the main line of this chapter, towards the rigorous derivation
of the time-dependent Gross-Pitaevskiı̆ equation, this section presents the typical formal
derivation from the Condensed Matter Physics viewpoint (see, e.g., Chap. IV-V of [71],
Chap. 7 of [96], Chap. 5 of [97]). In the current formalism it goes at follows.
Chose for concreteness the three-dimensional setting. The main point is the formal
replacement of the true pair potential with a delta-interaction. In the N -body Hamilto-
nian
N
~2 X X
HN = − ∆xi + VN (xi − xj ) (5.2)
2m
i=1 16i<j6N

for a gas expanding under the repulsive interparticle interaction only, one substitutes
g
VN (x) 7→ δ(x) (5.3)
N
that is, a delta potential scaling in the mean field sense and with a magnitude fixed by
the coupling g. The choice of g that turns out to reproduce the correct GP equation is

4π~2 a
g = 8πµa = (5.4)
m
~2
(µ = 2m as usual), where a is the s-wave scattering length of the two-body process.
5.2. Formal derivation of the t-dependent GP equation 85

Ansatz (5.3) and (5.4) encodes the physical assumption that particles are coupled
by a hard-core and short-range two-body repulsive potential and it is even thought as
“possibly the most important result in the whole of the Physics of the dilute ultra-cold
alkali gases”.2
The outcoming Hamiltonian
N
(δ) ~2 X 1 X
HN =− ∆xi + g δ(xi − xj ) , (5.5)
2m N
i=1 16i<j6N

apart from any well-posedness question, can be treated formally in the Schrödinger equa-
tion
(δ)
i~ ∂t ΨN,t = HN ΨN,t . (5.6)

When regarding (5.6) as an Euler-Lagrange equation, its solutions ΨN,t are critical points
of the corresponding functional (namely, the many-body action) and, hence, are given
by the stationary condition
Z
¡ (δ) ¢
0=δ ΨN,t − i~ ∂t ΨN,t + HN ΨN,t . (5.7)
R3N

One then searches for critical points of the form

ΨN,t = ϕ⊗N
t , ϕt ∈ H 1 (R3 ) , kϕt kL2 (R3 ) = 1 ∀ t , (5.8)

and looks for the evolutionary equation that ϕ· must satisfy.


(δ)
By treating the delta operators in HN as distributions in the integral (5.7),
Z Z
¡ (δ) ¢
ΨN,t − i~ ∂t ΨN,t + HN ΨN,t = − ΨN,t i~ ∂t ΨN,t
R3N R3N
N Z
X X Z (5.9)
¡ ¢ g
+µ ΨN,t − ∆xi ΨN,t + ΨN,t δ(xi − xj )ΨN,t ,
R3N N R3N
i=1 16i<j6N

one gets
Z Z
ΨN,t i~ ∂t ΨN,t = N ϕt (x) i~ ∂t ϕt (x) dx (5.10)
R3N R3
N Z
X Z
¡ ¢
ΨN,t − ∆xi ΨN,t = N |∇ϕt (x)|2 dx (5.11)
i=1 R3N R3
X Z Z
N (N − 1)
ΨN,t δ(xi − xj )ΨN,t = δ(x − y)|ϕt (x)|2 |ϕt (y)|2 dx dy
R3N 2 R6
16i<j6N
Z
N (N − 1)
= |ϕt (x)|4 dx . (5.12)
2 R3

For N large enough (so that N − 1 ∼ N ), (5.7) becomes


· Z ³ ´ ¸
2 g 4
0=δ N − ϕt (x) i~ ∂t ϕt (x) + µ |∇ϕt (x)| + |ϕt (x)| dx , (5.13)
R3 2
2
Legget [71], Sec. IV.C
86 Chapter 5. Strengthened convergence for reduced density matrices

then, for the consistency of prescription (5.8), ϕ· has to be a critical point of the functional
(5.13), hence, a solution of the outcoming Euler-Lagrange equation

δE g [ ϕt ]
2 8πµ
i~ ∂t ϕt (x) = −µ ∆x ϕt (x) + g |ϕt (x)| ϕt (x) = (5.14)
δ ϕt
where Z
¡ ¢
Eν [ ϕ ] := µ |∇ϕ(x)|2 + 4πµ ν |ϕt (x)|4 dx . (5.15)
R3
With the choice g = 8πµa, (5.14) is the desired time-dependent GP equation (com-
pare it with (5.27)), and functional E g is the corresponding GP functional (compare it
8πµ
with (4.28)).
(δ)
What makes this derivation formal is the treatment of the Hamiltonian HN . First,
to fix it unambiguously, an explicit self-adjoint extension He (δ) of the symmetric operator
N
defined in (5.5) has to be chosen. This identifies the domain of the extension and the
action of the delta operators on it. This action is not the simple “multiplicative” action
that has been used in (5.9) by interpreting δ as a distribution. Instead, each extension is
related in a suitable sense to a choice of boundary conditions on the hyperplanes xi = xj
in R3N . 3
Moreover, what makes this derivation wrong is the occurrence of an Efimov effect:
indeed it can be proved that already for N = 3 any local 4 extension H e (δ) is unbounded
3
below, making the system unstable.5 Another instability shows up in two dimensions:
there one can prove ([35]) that He (δ) is bounded below, but one can argue that the infimum
N
of the spectrum goes to −∞ faster than N (presumably the behaviour is ∼ N 2 ).
It is noteworthy noticing that in the formal derivation the choice of g is made to let
(5.14) agree with the GP equation and one models the true interaction with an effective
1
VN (x) = 8πµa δ(x) ; (5.16)
N
on the other side, in the rigorous (although indirect) approach one scales the true inter-
action with the GP scaling and gets
b
VN (x) = N 2 V (N x) ∼ δ(x) (5.17)
N
b
where 8πµ is the Born approximation for a. Contradiction between (5.16) and (5.17)
is only apparent. In the rigorous argument one deals with the (unique) self-adjoint
P P
extension of i (−µ∆xi ) + i<j VN (xi − xj ), while in the formal argument one ignores
to fix one among the infinitely many self-adjoint extensions of (5.5) and no rigorous
comparison is known between the two operators in the limit. Moreover in the rigorous
treatment one does not plug VN ∼ Nb δ directly in the Schrödinger equation,6 but instead
keeps track of the action of VN in the corresponding hierarchy of density matrices (see
3
A precise meaning to the formal expression (5.5) was first given by Minlos and Faddeev [88] and
more recently by Dell’Antonio, Figari and Teta [35].
4
in the sense that locality is a property of the corresponding boundary conditions on the union of the
hyperplanes, see [5]
5
This serious and entirely new difficulty with respect to the N = 2 case, which falls instead within
the theory ([5]) of the one-particle Schrödinger equation with zero range potential, was first pointed out
by Minlos and Faddeev [87] and then studied by Dell’Antonio, Figari and Teta [35].
6
unless in the one-dimensional case where δ is easily controllable as an operator
5.3. Rigorous derivation of the time-dependent CNSE 87

Sec. 5.4) and then makes a highly non-trivial control of the limit N → ∞ in such a
hierarchy: this way the correct coupling 8πµa emerges in the limit.
Incidentally, (5.16) usually suggests to interpret the GP equation as a phenomenolog-
ical mean field type equation. Since (5.16) is only formal, this interpretation has not solid
grounds: Hamiltonian H e (δ) after the δ-replacement does not describe a stable system.
N
Conversely, one might emphasize that the same GP equation is rigorously recovered in
the GP limit, which being a short-range + hard-core limit is not a mean field scaling
(the latter is characterized, instead, by a long range interaction with vanishing strength).
Probably the most appropriate perspective is simply to declare the conditions of validity
of the GP equation, that is, the low density regime.

5.3 Rigorous derivation of the time-dependent


cubic nonlinear Schrödinger equation
(Gross-Pitaevskiı̆ equation)
Two independent construction of the commutative diagram (5.1) have been recently
achieved, by Erdős, Schlein and Yau [44, 43], and by Adami, Golse and Teta [3]. The
first is set in the d = 3 dimensional case, although it can be extended to d = 1, 2, whereas
the second is intrinsically set in d = 1. The two proofs are structured with the same
scheme, but they differ both in the techniques and in the topologies that control the
(k)
existence and the uniqueness of the limit γN,t → |ϕt ihϕt |⊗k . In this section, both the
results are reviewed, with some comments on their setting that will be needed in the
sequel. The year we indicate for each theorem stays for when the statement and the
proof was made public by the authors for the first time.
Let us start with the three-dimensional result by Erdős et al., which has been estab-
lished in [44] after the intermediate results [41], [40] (with Elgart) and [42] – see also
[100] for a comprehensive review.

Theorem 5.3.1 (Erdős, Schlein and Yau (2006), [44]). Let V : R3 → R be a


non-negative, smooth, compactly supported, spherically symmetric function. Regarding
it as a two-body quantum mechanical potential for spinless particles of mass m, let a be
its s-wave scattering length. Also, let V be small enough when measured in terms of the
2
dimensionless quantity α~2m , with
Z
2 V (x)
α := sup |x| V (x) + dx . (5.18)
x R3 |x|

Equivalently,
α = kr2 VkL∞ (R+ ) + 4πkrVkL1 (R+ ) (5.19)
where r = |x| and V(r) := V (x). Consider a three-dimensional system of N spinless
undistinguishable bosons of mass m interacting with a two-body potential VN defined by

VN (x) := N 2 V (N x) (5.20)

so that the N -body Hamiltonian is


N
~2 X X ¡ ¢
HN = − ∆xi + N 2 V N (xi − xj ) (5.21)
2m
i=1 16i<j6N
88 Chapter 5. Strengthened convergence for reduced density matrices

acting as a self-adjoint operator on L2 (R3N ). Assume that the system is prepared in the
initial pure state ΨN ∈ L2sym (R3N ), with kΨN kL2 = 1, satisfying either the two properties
(A1 ), (A2 ) or the two properties (B1 ), (B2 ) below.

A1 ) Uniformly bounded k-th moment of the energy per particle:


¡ ³ H ´k ¢
N
∃ C > 0 : ∀k > 1 ∀N > 1 ΨN , ΨN 6 C k (5.22)
N
(the bound on the k-th power of HN /N is uniform in N , but not in k).

A2 ) 100% asymptotic B.E.C.:

∃ ϕ ∈ L2 (R3 ) , kϕkL2 = 1 :
(k) (5.23)
∀k > 1 γN → |ϕihϕ|⊗k as N → ∞ .

B1 ) Uniformly bounded energy per particle:


HN
∃ C > 0 : ∀N > 1 (ΨN , ΨN ) 6 C . (5.24)
N

B2 ) Asymptotic factorisation:

∃ ϕ ∈ L2 (R3 ) , kϕkL2 = 1 :
∀ N > 1 , ∀k = 1, . . . , N
(5.25)
∃ Ξ(N −k) ∈ L2 (R3(N −k) ) , kΞ(N −k) kL2 = 1 :
kΨN − ϕ⊗k ⊗ Ξ(N −k) kL2 → 0 as N → ∞ .

(k) i
Then ϕ ∈ H 1 (R3 ). Moreover, if γN,t are the k-th marginals of ΨN,t := e− ~ HN t ΨN (the
time-evolution of ΨN under the Hamiltonian HN ), then
(k)
∀ k > 1 ∀ t ∈ R γN,t → |ϕt ihϕt |⊗k as N → ∞ (5.26)

where ϕt ∈ H 1 (R3 ) is the solution of the time-dependent Gross-Pitaevskiı̆ equa-


tion with initial datum ϕ :


 ~2 4π~2
 i~ ∂t ϕt = − ∆ϕt + a |ϕt |2 ϕt
2m m (5.27)

 ¯¯

ϕt t=0 ≡ ϕ

Limits in (5.23) and (5.26) are understood with respect to the weak-∗ topology of
L1 (L2 (R3k )), namely, in trace against any compact operator on L2 (R3k ).

Thus, ESY theorem essentially states that when the system is left to evolve under
HN , it still shows 100% asymptotic B.E.C. and the condensate wave function ϕt is the
evolution of ϕ under the GP equation.
Let us collect some explanatory comments.

3 To start with, the real crucial hypotheses on V , which is a major open problem to
relax, turn out to be positivity of V itself and smallness of the parameter (5.18),
while smoothness, compact support, and spherical symmetry are just technical
assumptions.
5.3. Rigorous derivation of the t-dependent CNSE 89

3 Recall that, according to the standard definition of the s-wave scattering length
(see, e.g., appendix A of [80]), VN has scattering length aN = a/N . Recall, also,
S0
that N VN (x) −→ b δ(x) as N → ∞, where b := kV kL1 : hence, the two-body VN is
an approximate delta function on the same scale N −1 of its scattering length. The
dependence N 7→ HN is actually the explicit realization of the GP scaling in use
(see Sec. 4.3).

3 The (B) hypotheses are interchangeable with the (A) ones, leading to the same
thesis: to relax the energy condition from (A1 ) to (B1 ) one introduces a technically
stronger factorisation property (B2 ) than (A2 ) (although this property is only ap-
parently stronger, as discussed in Sec. 7.2). Since it turns out that (B2 ) ⇒ (A2 ),
both have the meaning of B.E.C. according to definition 3.4.1. The (B) version of
the theorem follows from the (A) version by suitably smoothing an initial datum
satisfying (B) with a cut off of the high energy component. The modified initial
datum one ends up with, is admissible for the (A) hypotheses and the theorem
follows for it. Finally, the cut off is removed.

3 Two classes of physically interesting initial states satisfying the (B) assumptions
above are explicitly shown to exist: totally factorised wave functions ϕ⊗N , for an
arbitrary ϕ ∈ H 1 (R3 ), and almost product wave functions with ground state like
short-scale correlations (see [43] and appendices B and C of [44]). We will discuss
them in Chap. 6 in the context of the correlation structure in a condensate

In comparison with ESY theorem, Adami et al. have proved an analogous theorem
in [3], after the intermediate result [2] (with Bardos).

Theorem 5.3.2 (Adami, Golse and Teta (2005), [3]). Let V : R → R be a


non-negative function in the Schwartz space S(R) with L1 -norm b. Consider a one-
dimensional system of N spinless undistinguishable bosons of mass m interacting with a
two-body potential VN defined by

VN (x) := N β−1 V (N β x) (5.28)

with 0 < β < 1, so that the N -body Hamiltonian is

N
~2 X ∂ 2 X ¡ ¢
HN = − 2
+ N β−1 V N β (xi − xj ) (5.29)
2m ∂x
i=1 16i<j6N

acting as a self-adjoint operator on L2 (RN ). Assume that the system is prepared in


the totally factorised initial pure state ΨN = ϕ⊗N ∈ L2sym (RN ), with ϕ ∈ S(R) and
kϕkL2 = 1, such that any k-th moment of the energy per particle is uniformly bounded
in N : ³ ³ H ´k ´
N
∃ C > 0 : ∀k > 1 ∀N > 1 ΨN , ΨN 6 C k . (5.30)
N
(k) i
If γN,t are the k-th marginals of ΨN,t := e− ~ HN t ΨN (the time-evolution of ΨN under the
Hamiltonian HN ), then ∀ k > 1

(k)
γN, · → |ϕ · ihϕ · |⊗k as N → ∞ (5.31)
90 Chapter 5. Strengthened convergence for reduced density matrices

weakly-∗ in L∞ (R, L2reg ) – defined in (5.39) below – where ϕt ∈ H 1 (R3 ) is the solution
of the time-dependent cubic nonlinear Schrödinger equation (CNSE) with
initial datum ϕ : 

 ~2 ∂ 2
 i~ ∂t ϕt = − ϕt + b |ϕt |2 ϕt
2m ∂x2 (5.32)

 ¯
 ¯
ϕt t=0 ≡ ϕ .

So, AGT theorem provides an analogous persistence in time of the asymptotic fac-
(k)
torisation of marginals γN,t .
S0
Here, by scaling, N VN (x) −→ b δ(x) as N → ∞: VN is an approximate delta function
at the scale N −β . Adami et al. can control that factorised initial states ϕ⊗N satisfying
the energy bounds (5.30) exist, provided that 0 < β < 12 : for that it suffices that ϕ has
compactly supported Fourier transform. In the whole range 0 < β < 1 one can deal with
factorised initial states even without (5.30), through the approximation strategy set up
by Erdős et al. in the three dimensional problem [42] (details in the appendix of [3]).
Both equations (5.27) and (5.32) go under the name of cubic nonlinear Schrödin-
ger equation (CNSE). Its solution has the physical meaning of condensate wave func-
tion. In dimension d, each term of the CNSE has dimension
d
(mass) (length)2− 2 (time)−2 .

By solution of (5.27) or (5.32) here one means a strong H01 -solution, that is, a function

ϕ ∈ C(I, H01 (Rd )) ∩ C 1 (I, H01 (Rd )) (5.33)

such that ϕ satisfies the equation in H −1 (Rd ) for all t ∈ I, interval in R with I 3 0, and
prescribed initial value ϕt=0 ∈ H 1 (Rd ). Notoriously (see [27] for details),

3 the solution exists and is unique,

3 it is global, i.e., I = R,

3 its L2 -norm is conserved at any time, i.e., kϕt ( · )kRd = kϕt=0 ( · )kRd ,

3 its H 1 -norm is uniformly bounded in time.

Concerning the topologies for convergence in both theorems, let us recall from Sec. 2.1
that the Banach structure of the trace class L1 (L2 (R3k )) and the Hilbert structure of
the Hilbert-Schmidt class L2 (L2 (R3k )) are given, respectively, by

kAkL1 = TrHk |A| , (5.34)



(B, C)L2 = TrHk [B C] , (5.35)

∀A ∈ L1 (Hk ), ∀B, C ∈ L2 (Hk ). Their weak-∗ topology, namely the topology based on
the preduals, is determined by the dualities
¡ ¢ ¡ ¢∗
L1 (Hk ) , k kL1 = Com(Hk ) , k k
B∈L1
(5.36)
Com(Hk ) 3 C 7−→ TrHk [BC]
5.4. Scheme of derivation: method of the hierarchies 91

and
¡ ¢ ¡ ¢∗
L2 (Hk ) , k kL2 = L2 (Hk ) , k kL2
B∈L2
(5.37)
L2 (Hk ) 3 C 7−→ TrHk [B ∗ C]

respectively, Com(Hk ) being the space of compacts on Hk .


In particular, convergence (5.26) in ESY theorem reads
£ (k) ¤
lim TrHk C(γN,t − |ϕt ihϕt |⊗k ) = 0 (5.38)
N →∞

for any positive integer k, any time t and and any compact C. Convergence in AGT
theorem involves a Sobolev space of Hilbert Schmidt operators. To denote that an extra
regularity is asked to the density matrices under consideration, one introduces the space
© ª
L2reg (L2 (Rk )) := γ ∈ L2 (L2 (Rk )) : S12 · · · Sk2 γ ∈ L2 (L2 (Rk )) (5.39)

which is a Hilbert space when endowed with the scalar product

(ρ, γ)L2reg := ( S12 · · · Sk2 ρ , S12 · · · Sk2 γ )L2


£ ¤
= TrHk ( S12 · · · Sk2 ρ )∗ (S12 · · · Sk2 γ ) (5.40)
£
= TrHk ( S1 · · · Sk ρ S1 · · · Sk )∗ ( S1 · · · Sk γ S1 · · · Sk ) ]

where the Si ’s (i = 1, . . . k) are the positive operators

Si := (1 − ∂x2i )1/2 . (5.41)

Weak-∗ topology in L2reg is defined analogously to the L2 case (5.37). This, in turn,
(k)
defines the weak-∗ topology in L∞ (R, L2reg ) the theorem deals with: in fact, the γN,t ’s
and |ϕt ihϕt |⊗k are shown to have L2reg -norms that are uniformly bounded both in N and
in time. Thus, convergence (5.31) in AGT theorem reads
Z +∞ ¡ (k) ¢
lim dt ρt , γN,t − |ϕt ihϕt |⊗k L2 = 0
N →∞ −∞ reg (5.42)
1
∀ρ· ∈ L (R, L2reg )

∀k > 1.

5.4 Scheme of derivation: method of the hierarchies.


Effective nonlinear one-particle equations
from the linear many-body dynamics.
We review in this section the scheme of the rigorous derivation of the GP equation in
connection with the problem of the time-stability of condensation.
Actually, as we have already mentioned, from the mathematical point of view this
problem covers an even larger range of models than the aforementioned B.E.C.-related
origin. In general it arises, as a part of Kinetic Equations theory, whenever one wants to
extract, under suitable conditions and in some scaling limit N → ∞, the effective non-
linear one-particle equation emerging from a linear many-particle evolutionary equation.
92 Chapter 5. Strengthened convergence for reduced density matrices

The picture can be both classical and quantum. In the first case, the many-body dynam-
ics one starts with is given by the Liouville equation (or Boltzmann, or Vlasov equation,
depending on the context) and the one-body effective equation one derives is usually
referred to as the Poisson-Vlasov equation. In the latter, one starts from the many-body
Schrödinger equation to end up with a one-body nonlinear Schrödinger equation which
takes the name, depending on the kind of nonlinearity, of Hartree, Schrödinger-Poisson,
Gross-Pitaevskiı̆, . . . , equation. There is also the related issue of deriving the classical
results, both at the N -body and at the single-particle level, through an appropriate
semiclassical limit of the corresponding quantum equations.
Along this mainstream, classical and quantum perspective historically alternated in
the last three decades and specific tools and approaches have been developed, within
an essentially common strategy: rewrite the N -body equation in terms of a hierarchy of
evolutionary equations for marginals and then identify the limiting k-th marginals, at
each fixed k as N → ∞, as the solution of the corresponding infinite hierarchy.
This idea has been introduced in 1977 by Braun and Hepp [20] in the classical frame-
work and in 1980 by Spohn [103] for the quantum case (though this is not the only
difference among them, as we are discussing later). A historic survey of how it has been
developed and improved since then is postponed to the next section. To present it con-
cretely, according to the spirit of the present work, we choose the quantum perspective,
with the notation already introduced so far, of an initially condensed Bose gas for which
persistence of condensation is investigated. References to similar investigations will be
mentioned throughout.
The scheme of the analysis of the time-stability of B.E.C. goes as follows.
A N -body Bose system trapped in a confining potentials is given in a initial condensed
state γN , (like in the model of theorem 3.5.2), possibly with some additional conditions
(as in the hypotheses of theorems 5.3.1 and 5.3.2) expressing essentially a finite energy
per particle. At t = 0 the trap is instantaneously removed and the system evolves under
a two-body interaction Hamiltonian
N
~2 X X
HN =− ∆xi + VN (xi − xj ) . (5.43)
2m
i=1 16i<j6N

As usual, the pairing, apart from its regularity properties, is assumed to be spherically
symmetric and repulsive. The initial state is not invariant under such HN , then it evolves
according to the Heisenberg - von Neumann equation

i~ ∂t γN,t = [ HN , γN,t ] (5.44)


(k)
and at each time the corresponding marginals γN,t are defined.
The goal is to control whether condensation occurs at t > 0, that is, whether in some
(1)
approximation γN,t ≈ |ϕt ihϕt | for N large enough. In particular, some scaling has to be
specified as N → ∞, in the spirit of the discussion of Chap. 4 (and Sec. 3.2). One is
then interested, for any fixed k, in the limiting objects
(k) (k)
γ∞,t := lim γN,t , (5.45)
N →∞

in which sense they exist, and when they possibly factorise as a tensor product of projec-
tions onto some ϕt that plays the role of time-dependent condensate wave function. Fur-
thermore, ϕt is expected to be the solution of that cubic nonlinear Schrödinger equation
5.4. Scheme of derivation: method of the hierarchies. 93

(the Gross-Pitaevskiı̆ equation) confirmed by physical heuristic arguments and experi-


mental observations to be the evolutionary equation of the condensate wave function,
with initial condition given by the condensate wave function ϕ at time t = 0. So one
seeks to complete the following diagram for any positive integer k:

(k) N →∞
γN −−−−→ |ϕihϕ|⊗k
 
 GP−equation
BBGKY−hierarchyy y (5.46)
(k) N →∞
γN,t −−−−→ |ϕt ihϕt |⊗k

Following the seminal idea of Braun, Hepp and Spohn, the key tool to prove (5.46)
is represented by hierarchies of reduced density matrices. In fact, through the definition
of reduced density matrix (definition 2.3.3 and theorem 2.3.4), (5.44) is equivalent to a
(k)
finite hierarchy of first order linear PDE’s for kernels, where each ∂t γN,t is expressed in
(k) (k+1)
terms of γN,t and γN,t , the last equation of which is (5.44) itself. A rather established
although not unanimous use in the literature is to refer to it as the finite BBGKY
hierarchy after the works of Bogolubov [16], Born and Green [17], Kirkwood [65, 66],
and Yvon [112] – in fact, it resembles the BBGKY hierarchy of equations satisfied by
the k-particle distributions (i.e., probability densities) in the classical Kinetic Theory,
as a consequence of the Liouville equation (see [28, 29]). Sometimes it is also cited as
the finite Schrödinger hierarchy, to emphasize its quantum character, while its classical
analog is also called the Liouville hierarchy. Its explicit form in terms of kernels is

k
~2 X ¡ ¢ (k)
i~ ∂t (Xk , Yk ) = − ∆xi + ∆yi γN,t (Xk , Yk )
2m
i=1
X ¡ ¢ (k)
+ VN (xi − xj ) − VN (yi − yj ) γN,t (Xk , Yk )
16i<j6k
k Z
(5.47)
X ¡ ¢ (k+1)
+ (N − k) dz VN (xi − z) − VN (yi − z) γN,t (Xk , z, Yk , z)
i=1 Rd

k = 1, . . . , N

where d is the space dimension, each xi ∈ Rd and Xk := (x1 , . . . , xk ) ∈ Rkd , etc., and the
(k)
convention is that γN,t = O if k > N . Hence, the N -th and last equation of the hierarchy
is (5.44) itself, which in terms of kernels reads

N
~2 X ¡ ¢
i~ ∂t γN,t (XN , YN ) = − ∆xi + ∆yi γN,t (XN , YN )
2m
i=1
X ¡ ¢ (5.48)
+ VN (xi − xj ) − VN (yi − yj ) γN,t (XN , YN ) .
16i<j6N

Notice that the first term on the r.h.s. of the hierarchy (5.47) describes the kinetic energy
of the first k particles, the second term is associated with the interactions among the first
k particles, and the last term corresponds to interactions between the first k particles
and the other N − k particles.
94 Chapter 5. Strengthened convergence for reduced density matrices

Let us also quote other useful equivalent forms of (5.47). Its operator form is

~2 X h i h i
k X
(k) (k) (k)
i~ ∂t γN,t = −∆xi , γN,t + VN (xi − xj ) , γN,t
2m
i=1 16i<j6k
(5.49)
k
X h i
(k+1)
+ (N − k) Tr[k+1] VN (xi − xk+1 ) , γN,t
i=1

where Tr[k+1] : L1 (L2 (R(k+1)d )) → L1 (L2 (Rkd )) is the partial trace over the (k + 1)-th
variable. The corresponding integral forms are
k Z
X t h i X Z t h i
(k) (k) (k) (k)
γN,t = γN,s −i dr −∆xi , γN,r − i dr VN (xi − xj ) , γN,r
i=1 s 16i<j6k s

k Z
X t h i (5.50)
(k+1)
− i(N − k) dr Tr[k+1] VN (xi − xk+1 ) , γN,r
i=1 s

(s 6 t)

and
(k) (k) (k)
γN,t = UN (t) γN,0 +
k Z
X t h i (5.51)
(k) (k+1)
− i(N − k) dr UN (t − r) Tr[k+1] VN (xi − xk+1 ) , γN,r
i=1 0

where
(k) (k)
(k)
UN (t) γ := e−iHN t/~
γ eiHN t/~
(5.52)
k
(k) ~2 X X
HN := − ∆xi + VN (xi − xj ) . (5.53)
2m
i=1 16i<j6k

This way, the knowledge of a solution γN,t of (5.44) with the initial condition γN
(k)
is equivalent to the knowledge of the finite sequence {γN,t }N
k=1 solving hierarchy (5.47)
(k) (k)
with initial condition {γN }N k=1 . To control possible limiting marginals γ∞,t in a scaling
limit N → ∞, a twofold analysis is performed. The description we give here is modelled
on the concrete case of the recent investigation by Erdős, Schlein and Yau which has led
to theorem 5.3.1, that is, a class of initial states with asymptotic 100% B.E.C. for which
persistence of asymptotic 100% condensation at later time is proved in the GP scaling
limit. Many other related attempts share an analogous scheme.
(k)
¬ (Extraction of weak-∗ limits) – At each fixed k, the sequence {γN,t }N >k is
proved to be weakly-∗ compact in some suitable topology for trace class operators.
Then, via a Banach-Alaoglu argument, it admits weak-∗ limits for subsequences.
What ensures such a sequence to stay uniformly inside a ball of a suitable Banach
space with predual, is an “a priori estimate” following from assuming that the
initial state has a N -uniformly bounded energy per particle, together with the
same assumption for higher moments of the energy. Each (possibly not unique)
(k)
limit point γ∞,t is not a priori a density matrices, yet it is proved to be a positive
and permutationally symmetric trace class operator, with “good” trace properties
5.4. Scheme of derivation: method of the hierarchies. 95

for the subsequent steps (i.e., the weak-∗ limit belongs to the same ball as the
original sequence).
(k)
(Convergence to the infinite hierarchy) – Each infinite family {γ∞,t }k>1 of
weak-∗ limit points is proved to be the solution of an infinite hierarchy for trace class
operators with initial condition { |ϕihϕ|⊗k }k>1 . This can be obtained formally by
taking N → ∞ in the finite hierarchy (5.47). Such an infinite hierarchy maintains
the structure of coupled P.D.E.’s where the k-th element of the family is given in
terms of the (k + 1)-th one.

­ (Uniqueness of the solution) – One proves that the infinite hierarchy admits a
(k)
unique solution {γ t }k>1 with initial condition { |ϕihϕ|⊗k }k>1 and such that each
(k)
γ t is a positive and permutationally symmetric trace class operator inside the
(k)
same ball as γ∞,t (namely, with the same trace properties). By direct inspection,
one checks that { |ϕt ihϕt |⊗k }k>1 solves such infinite hierarchy with those properties,
iff ϕt is a solution of the time-dependent Gross-Pitaevskiı̆ equation with initial
(k)
condition ϕ. Hence, by uniqueness, the weak-∗ limit point is unique and γN,t →
|ϕt ihϕt |⊗k as N → ∞.

In the three-dimensional setting and under the (A) hypotheses of theorem 5.3.1, the
(k)
infinite hierarchy satisfied by the limiting sequence {γ∞,t }k>1 reads

~2 X ³ ´
k
(k) (k)
i~ ∂t γ t (Xk , Yk ) = − − ∆x` + ∆y` γ t (Xk , Yk )
2m
`=1
(5.54)
X k Z
¡ ¢ (k+1)
+ 8πµa dz δ(x` − z) − δ(y` − z) γ t (Xk , z, Yk , z)
`=1 R3

for k = 1, 2, 3, . . . , or equivalently, in integral operator form,

k Z
X t h i
(k) (k) (k)
γt =U (t) γ 0 − 8iπµa ds U (k) (t − s) Tr[k+1] δ(x` − xk+1 ) , γ (k+1)
s , (5.55)
`=1 0

where a1 is the s-wave scattering length of the unscaled (“true”) potential V and

k k
µ X µ X
i t ∆x` −i t ∆x`
~ `=1 ~ `=1
U (k) (t) γ (k) := e γ (k)
e . (5.56)

Here the action of the delta function on (kernels of) density matrices is well-defined (it
would not be for general density matrices) through an appropriate limiting procedure
(details in Sec. 8 of [42] and Sec. 7 of [44]), which is possible since the limiting objects
(k)
γ∞,t are proved to have sufficiently regular kernels. This gives

³ Z
£ ¤´ (k+1)
Tr[k+1] δ(x` − xk+1 ) γ (k+1)
s (X ,
k kY ) = dz δ(x` − z) γ t (Xk , z, Yk , z)
R3 (5.57)
= γ (k+1)
s (Xk , x` , Yk , x` )
96 Chapter 5. Strengthened convergence for reduced density matrices

and (5.54) reads

~2 X ³ ´
k
(k) (k)
i~ ∂t γ t (Xk , Yk ) = − − ∆x` + ∆y` γ t (Xk , Yk )
2m
`=1
(5.58)
Xk h i
(k+1) (k+1)
+ 8πµa γt (Xk , x` , Yk , x` ) − γt (Xk , y` , Yk , y` ) .
`=1

(k)
It is then straightforward to check that {γt = |ϕt ihϕt |⊗k }k>1 is a solution of (5.58)
with initial condition {|ϕihϕ|⊗k }k>1 iff ϕ· is a solution of the problem (5.27), that is, of
the GP equation with initial condition ϕ.
Hierarchy (5.54) is usually called the infinite BBGKY hierarchy, or sometimes
also infinite Schrödinger hierarchy to emphasize its quantum character, while its classical
counterpart goes usually under the name of infinite Vlasov hierarchy. Due to the physical
relevance of the coupling in front of the interaction term, (5.54) is also called infinite
GP hierarchy whenever this coupling is 8πµa, to distinguish it from the a generic
coupling α. Such α fixes the magnitude of the nonlinearity in the corresponding nonlinear
Schrödinger equation for ϕt , which correspondingly has the form

i~ ∂t ϕt = −µ∆x ϕt + α|ϕt |2 ϕt .

Only for α = 8πµa one gets the physically relevant GP equation.


The formal limit by which one obtains the infinite hierarchy (5.54) from the finite
one (5.47) is itself a major conceptual point. Recalling that in the three-dimensional GP
S0
scaling N VN −→ b δ (with b = kV kL1 ), whence VN ∼ N1 δ, it is clear heuristically that
the second term in the r.h.s. of (5.47) disappears in the limit N → ∞, since it contains
k(k − 1)/2 terms of the form ∼ N1 δ(xi − xj ), while the third term in the r.h.s. of (5.47)
consists of k terms with a pre-factor (1 − k/N ) and hence approaches asymptotically
the second term in the r.h.s. of (5.54). (The same can be argued for different kinds of
scalings, like mean field or modified GP.) Yet, this does not provide an even heuristic
justification of the the coupling 8πµa instead of the emerging b.
(2)
Since to close the infinite hierarchy one would need some relation between γ t and
(1) (2) (1)
γ t , any relation between γN,t and γN,t (before the limit) provides in principle a distinct
way to take the formal limit in the finite hierarchy. As we will be discussing in Chap. 6,
(2) (1) (1)
any other relation than the simple factorisation γN,t = γN,t ⊗ γN,t corresponds to certain
non-trivial correlations in the many-body state. Thus, the knowledge of correlations
present in the many-body state along its time evolution would allow one to perform
a distinct formal limit. Since having such a knowledge would, in turn, require to have
already solved the problem, one appeals to some ansatz on the basis of physical intuition.
For some years, attempts have been made without introducing short scale correlations
in the marginals: this ansatz leads to an infinite hierarchy with an unexpectedly larger
coupling constant than the GP one. Only when Erdős, Schlein, and Yau realized that
suitable ground-state-like short scale correlations persisting in time had to be plugged
(k)
into γN,t , one finally could perform a formal limit N → ∞ leading to the correct GP
hierarchy. We give details of this at the end of Sec. 6.4, in the framework of correlations
at the level of reduced density matrices (see also the most recent discussion in Sec. 3 of
[44], after the first observations in this direction expressed in the introductory sections
of [41] and [40]).
5.5. Historic survey 97

Physically, pairwise correlations are established in the two-body scattering process,


provided that a particle has enough time to feel the potential field of the other. The GP
scaling is indeed the scaling where the strength of the interaction (namely the scattering
length) is comparable with its spatial range (both scale as N −1 ) and, hence, it allows
to study emergence and persistence in time of short scale correlations. This is not the
case for other scalings like the mean field or the modified GP (4.15), where the range
of the interaction is respectively much larger than the scattering length or of the same
order than the spatial density of the gas. For those scalings one does not need further
correlation ansatz to derive formally the infinite hierarchy from the finite one.

5.5 Historic survey


We have already mentioned that the method of the hierarchies, both in classical and in
quantum Kinetic Theory, and the twofold task of “convergence + uniqueness” have a 30
years lasting history.
In 1977 Braun and Hepp [20] derived the Vlasov equation from a system of N classical
P
particles coupled by a mean field interaction N1 i<j V (xi − xj ). They assumed the
interacting potential bounded in C 2 and proved the convergence of the classical Liouville
hierarchy to the infinite Vlasov hierarchy. They avoided addressing the question of the
uniqueness of solutions of the infinite hierarchy, but nevertheless succeeded to prove that
the limits of N -particle marginals (in the large N limit) coincide with factorised products
of solution of the Vlasov equation. Their argument rely on a clever use of probabilistic
methods (such as the strong law of large numbers to establish the propagation of chaos
and the central limit theorem to analyze fluctuations about the mean field). Such an
argument does not seem to work for the limit ~ → 0, N → ∞ of a system of quantum
particles. This might explain why Narnhofer and Sewell [90] had to prove a uniqueness
theorem for the Vlasov equation. Since the corresponding operator is very singular,
stringent assumptions (equivalent to the analyticity of the potential) are required in that
paper. The regularity assumption was substantially relaxed by Spohn [104].
In 1980 Spohn [103] came up with the idea of using the trace norm to prove the
convergence of the solution of the N -body Schrödinger equation to the factorised N -body
state driven by the non-linear Schrödinger equation. His proof rests on the assumption of
a bounded interaction potential. The scaling in the interaction is just a mean field limit.
Being based on the Duhamel formula, his proof avoids the question of uniqueness for the
infinite Schrodinger hierarchy, although in a very different way than that in Braun and
Hepp. Along this quantum side of the problem, Spohn’s result is remarkable not only
because of the introduction of the method of the hierarchies for density matrices, but
also because it generalizes a previous (1974) result by Hepp [57], obtained in the context
of field operators, where the pair potential needed to be differentiable, instead of only
bounded as in Spohn.
Incidentally, Ginibre and Velo [51, 52] greatly extended in the meanwhile these results
to cover singular potentials (including Coulomb), but in a completely different approach
involving quasi-free states in the second quantized framework: this method requires
special initial states with no definite particle number.
The next key contribution was due to Bardos, Golse and Mauser [10] who in 2000
split the strategy into the twofold “convergence + uniqueness” task for the mean field
treatment of a gas of spinless undistinguishable bosons expanding under a repulsive pair
98 Chapter 5. Strengthened convergence for reduced density matrices

potential and prepared in a purely factorised initial state. The outcoming one-body
equation is the Hartree equation (also known as the Schrödinger-Poisson equation when
the interaction V is a repulsive Coulomb potential), that is, a nonlinear Schrödinger
equation with a non-local nonlinearity (V ∗ |ϕt |2 )ϕt , instead of the local nonlinearity
|ϕt |2 ϕt of the CNSE. Convergence of the N -body Schrödinger hierarchy to the infinite
Schrödinger hierarchy was shown under fairly general assumptions – a bounded-below
potential in L2 + L∞ , thus including repulsive Coulomb systems, which were not covered
by Spohn. Uniqueness, instead, needed stringent assumptions (boundedness and vanish-
ing at infinity) and was based on the abstract Cauchy-Kowalewski theorem by Nirenberg
[91] and Nisida [92] (a slight modification of such argument was subsequently obtained
by the same authors together with Erdős and Yau in [9]).
Technically, all uniqueness methods known until then relied on the boundedness of
the potential via the estimate Tr|V γ| 6 kV k∞ Tr|γ|. To include the Coulomb repulsion a
Hardy-like inequality Tr|V γ| 6 C(Tr[∇γ∇] + Tr|γ|) is needed (which is indeed a conse-
quence of the Hardy inequality 14 |x|−2 6 −∆ in three dimensions). It was Erdős and Yau
who first noticed it in 2001 the importance of such an estimate and realized that the so-
lution to the infinite BBGKY hierarchy should be unique under the right Sobolev norm:
this led to the full derivation (convergence + uniqueness) of the nonlinear Schrödinger
(Hartree) equation from a many-body Coulomb system [45]. The scaling was a mean field
and the initial datum was a completely factorised ϕ⊗N , but with an unnatural ϕ ∈ H 2 ,
instead of the natural H 1 space for the nonlinear Schrödinger equation. Convergence
was controlled in a weak-∗ sense for trace class valued bounded functions.
In fact, it is noteworthy noticing that in this scenario the choice of the topology
to control the limiting factorisation of the time-dependent marginals is forced by the
technique of such analysis. Depending on the topology by which one is able to control
uniqueness, one has to adopt the topology by which controlling the existence of the limit.
Explicitly, if to prove uniqueness one has to assume that the solution is bounded in some
(1)
norm, then the limiting γ∞,t has to be bounded in that norm, so one has to introduce
a topology for the convergence which allows to control that kind of boundedness in the
(1)
limit. It may happen that to prove uniqueness one needs to regard the kernel of γ∞,t as
a more regular object than simply a L2 -function and some suitable Sobolev spaces enter
the analysis. Thus, one ends up with some regularized weak-∗ topology.
All treatments until this point had been dealing with a mean field scaling, both in the
above quantum side and in the corresponding classical counterpart we are not concerned
in here. A complete review of the state of the art (until 2003) of the mean field limit for
the dynamics of large classical and quantum particle systems is Ref. [54] by Golse. An
analogous review for the quantum side only is Ref. [11] by Bardos and Mauser.
It is reasonable to say that the interest in developing the same conceptual scheme
in some GP-like scaling (instead of all previous mean field treatments), which started
only at this point, was driven by the increasing knowledge on B.E.C. of that time, both
at the experimental level and on the mathematical side. In fact, contributions in the
B.E.C.-driven directions, that is, towards the rigorous derivation of the CNSE, followed
soon after. The crucial intermediate steps to obtain the somewhat conclusive results
of theorems 5.3.1 and 5.3.2 have been (i) the identification of the appropriate infinite
hierarchy as N → ∞ in a suitable scaling limit (see the discussion concluding the previous
section) and (ii) a joint control of convergence + uniqueness.
In 2003 Adami, Bardos, Golse, and Teta [2] (see also [1]) addressed the problem
5.5. Historic survey 99

for a one-dimensional system of N spinless undistinguishable bosons interacting through


P
a mean field, repulsive, pairwise delta potential N1 i<j δ(xi − xj ), initially prepared
in a condensed state with N -uniformly bounded energy per particle, and proved the
convergence of (solutions of) the finite hierarchy to (solutions of) the infinite one in a
weakly-∗ regularized (Sobolev) trace class sense, where limit points solve the infinite
hierarchy in a distributional sense. A trivial solution of the infinite hierarchy is the
sequence of the factorised projections onto the solution of the CNSE, but uniqueness for
such a model is still missing today,
Concerning the three-dimensional setting, starting from 2004 a comprehension spread
over the obstructions and the technical solutions to obtain the true GP hierarchy, that
is, the infinite BBGKY hierarchy with the right coupling (8πµa) ensuring the emergence
of the GP equation (say the discussion at the end of the previous section).
In a first work [41] Erdős, Schlein, and Yau have proposed a model of spinless undis-
tinguishable bosons interacting through a sufficiently small potential in the k kL1 + k kL∞
sense, scaling with N in the GP sense. By considering a modified Hamiltonian H e N which
cuts off the pair interactions whenever at least three particles come into a region with
diameter much smaller than the typical interparticle distance, and a class of initial states
with a N -uniform bound both in the energy per particle H e N /N , and in (H
e N /N )2 , they
could prove convergence of (solutions of) the finite hierarchy to (solutions of) the infinite
GP hierarchy in a weighted weakly-∗ Hilbert-Schmidt sense, where limit points solve the
infinite hierarchy in a distributional sense. The above assumption on the energy selects
admissible initial states which are not necessarily condensates. Yet for the first time two
explicit classes of initial data have been recognised to satisfy the uniform bounds in the
energy, which can be both proved to have asymptotic 100% condensation (although this
is not explicitly noticed in the original work): purely factorised ϕ⊗N and wave func-
Q
tions i<j fN (xi − xj )ϕ(x1 ) · · · ϕ(xN ) (with enough regularity for ϕ), where correlations
are given by the zero energy scattering solution of the two-body process fN . From the
mathematical viewpoint, the modified Hamiltonian removes a technical obstacle: squar-
ing the true HN instead of H e N , the outcoming derivatives of the pair potential could not
be controlled by the kinetic energy operator in the rare situation when many particles
come close together, since such derivatives have no definite sign.
In a second work [40] Elgart, Erdős, Schlein, and Yau have taken into account the
full Hamiltonian (assuming a non-negative, smooth, compactly supported interaction),
but scaling it with a modification of the true GP scaling (see (4.15) in Sec. 4.3 and the
discussion thereafter) such that the interaction potential is a delta function at a much
larger scale than the scattering length. Consequently, the two-body process does not
establish correlations at the scale of the scattering length and formal limit leads to an
infinite BBGKY hierarchy with the first Born approximation coefficient in the interaction
term. For any initial state with a N -uniform bound on every k-th moment of the energy
per particle (i.e., Tr[(HN /N )k γN ] 6 C k ), convergence to such hierarchy is established in
the same sense as in [41].
In both these 2004 contributions uniqueness is missing. Only in 2005 the first joint
control of convergence and uniqueness in a GP-like scaling has been achieved. Since in
the GP-like scaling the interaction is asymptotically ∼ N1 δ, in two and three dimensions
this is such a singular potential that one cannot apply any longer the joint control
developed by Bardos et al. for the mean field scaling. In one dimension, instead, this is
still possible. Thus, by combining convergence technique of [40] and uniqueness technique
100 Chapter 5. Strengthened convergence for reduced density matrices

of [9], Adami, Golse and Teta [3] have completed the project of the rigorous derivation of
the CNSE in dimension one from the dynamics of pairwise coupled bosons (see theorem
5.3.2) where the interaction scales with a much larger range than the scattering length
(in analogy with the three-dimensional setting in [40]) and for any initial state with N -
uniform bounds on every energy moment – which includes, in particular, suitably regular
factorised state ϕ⊗N .
At the same time, Erdős, Schlein, and Yau [42] have completed the analogous three-
dimensional project (with a proof that can be extended to lower dimensions). Interaction
is scaled again as VN (x) = N 3β−1 V (N β x) for a non-negative, smooth, compactly sup-
ported and spherically symmetric V (but limited to the regime 0 < β < 12 , unlike the
regime 0 < β < 53 of the previous attempt [40]). The admitted initial states are of the
form ϕ⊗N . Convergence holds weakly-∗ in the trace class to an infinite hierarchy where
the coupling constant is the first Born approximation of the scattering length. Unique-
ness is achieved via a diagrammatic proof to control the Duhamel-type expansion of the
integral form (5.55) of the hierarchy. The whole proof deals with a regularization of the
initial data where the high energy component is cut off: this allows an appropriate a pri-
ori estimate for the regularized initial states guaranteeing the convergence + uniqueness
control, that holds true even after the removal of the cutoff.
Along this line, 2006 theorem 5.3.1 by Erdős, Schlein, and Yau [44] combines the
uniqueness technique of [42] with a new control of the convergence in the true three-
dimensional GP scaling VN (x) = N 2 V (N x). In the (A) hypotheses, admitted initial
states have asymptotic 100% B.E.C. and N -uniformly bounded energy moments. This
leads to the control of convergence in the weakly-∗ trace class sense and to the bounded-
(k)
ness of a suitable Sobolev trace norm for each limiting point γ∞,t , which is exactly that
(k)
norm under which the uniqueness theorem of [42] holds. For finite N , marginals γN,t
are recognised to contain a short scale structure emerging in the two-body process: the
many-body dynamics builds up and preserves short-scale correlations (see Sec. 6.5).

5.6 Strengthened convergence in ESY theorem


In this section it will be recalled how the trace class weak-∗ convergence in ESY theorem
actually implies convergence in the trace norm. This is due to the same mechanism used
in the proof of B.E.C. for dilute trapped gas by Lieb and Seiringer [74]. We have already
reported it in theorem 3.4.2 and the result is the following.

Theorem 5.6.1 (ESY strengthened convergence). Limit (5.38) in ESY theorem


implies, and hence can be lifted to,
° (k) °
lim ° γN,t − |ϕt ihϕt |⊗k °L1 = 0 (5.59)
N →∞

for any positive integer k and any time t ∈ R.

As a matter of fact, the most recent formulation of ESY theorem already mentions
trace norm (see theorem 1 in [43]), while its original version (see theorems 2.1 and 2.2 in
[44]), as well as the preliminary results along this mainstream (see theorem 2.1 in [41],
theorem 1 in [40], theorem 1 in [42]), all accounts for the weak-∗ topology only. One has
to recognise that the real power of ESY theorem is
5.7. Strengthened convergence in the AGT theorem 101

¬ to show that weak-∗ limits of sequence of marginals do exist via a Banach-Alaoglu


weak-∗ compactness argument; for this one needs that the k-th marginal is uni-
formly (in N ) bounded in a suitable Banach norm,

­ to prove that all these limits satisfy an infinite BBGKY hierarchy,

® to prove that the hierarchy admits a unique solution which is explicitly seen to be
tensor power of projections onto the solution of the GP equation.

In this perspective, stress in ESY theorem is not given to any stronger topology: the
primitive way to think to limit of marginals as N → ∞ (and to deal with them in
computations) is to take weak-∗ limits.

5.7 Strengthened convergence in the AGT theorem


In this section it will be shown how the L∞ (R, L2reg ) weak-∗ convergence in AGT theorem
actually implies convergence in a suitable time-weighted trace norm.

Theorem 5.7.1 (AGT strengthened convergence, [85]). For any positive integer
k, limit (5.42) in AGT theorem implies, and hence can be lifted to,
Z +∞ ° (k) °2
lim dt f (t) ° γN,t − |ϕt ihϕt |⊗k °L2 = 0
N →∞ −∞ (5.60)
1
∀ f ∈ L (R)

and also to
Z +∞ ° (k) °
lim dt f (t) ° γN,t − |ϕt ihϕt |⊗k °L1 = 0
N →∞ −∞ (5.61)
1
∀ f ∈ L (R) .

Proof. We give details for k = 1; generalization to k > 1 is elementary. Also, we consider


only the case f > 0 and kf kL1 (R) = 1: indeed, once (5.60) and (5.61) are proved under
this restriction, the general case follows writing any non zero f ∈ L1 (R) as the linear
combination
f+ f−
f = kf+ kL1 (R) − kf− kL1 (R) (5.62)
kf+ kL1 (R) kf− kL1 (R)

where the positive L1 (R)-functions f± are the positive and negative part of f respectively,
that is,
f± (x) = max{±f (x), 0} . (5.63)

Step 1. One chooses a suitable ρ · ∈ L1 (R, L2reg ) in (5.42):

ρt := f (t)S −2 |ϕt ihϕt |S −2 (5.64)

∀f ∈ L1 (R) such that f > 0 and kf kL1 (R) = 1, where S is the operator

³ d2 ´1/2
S := 1 − 2 . (5.65)
dx
102 Chapter 5. Strengthened convergence for reduced density matrices

Indeed S −2 |ϕt ihϕt |S −2 is a positive trace class operator with uniformly (in time)
bounded L2reg -norm:
° −2 ° ° °
° S |ϕt ihϕt |S −2 °2 2 = ° S 2 S −2 |ϕt ihϕt |S −2 °2 2
Lreg L
£ ¤
= Tr |ϕt ihϕt |S |ϕt ihϕt |S −2
−2

= (ϕt , S −2 ϕt )2L2 (R)


(5.66)
³ Z +∞ 1 ¯ ¯2 ´2
= ¯ϕbt (k)¯ dk
2 2
−∞ 1 + 4π k
bt k4L2 (R) = kϕt k4L2 (R) = 1 .
6 kϕ

(Here the unimportant constants entering the computations are due to the convention
Z +∞
ϕ(k)
b = e−2πikx ϕ(x)dx
−∞

to define the Fourier transform.)


The L2 (R)-norm of ϕt is preserved in time and it amounts to 1 as at t = 0. Then
ρ · ∈ L1 (R, L2reg ) because
Z Z
+∞ ° ° +∞ ° °
dt ° ρt ° L2reg
= dt f (t)° S −2 |ϕt ihϕt |S −2 °L2 6 1 . (5.67)
reg
−∞ −∞

Step 2. With this choice of ρ · , AGT convergence gives


Z +∞
° (1) °2
dt f (t) ° γN,t − |ϕt ihϕt | °L2 6
−∞
Z +∞
£ (1) ¤
6−2 dt f (t) TrH1 |ϕt ihϕt | (γN,t − |ϕt ihϕt |)
−∞
Z +∞ (5.68)
£ (1) ¤
=−2 dt TrH1 S 2 f (t)S −2 |ϕt ihϕt |S −2 S 2 (γN,t − |ϕt ihϕt |)
−∞
Z +∞ ¡ (1) ¢
=−2 dt ρt , γN,t − |ϕt ihϕt | L2
reg
−∞

the inequality following by


° (1) °
° γ − |ϕt ihϕt | °2 2 =
N,t L
£ (1) 2 (1) (1) ¤
= TrH1 (γN,t ) − γN,t |ϕt ihϕt | − |ϕt ihϕt |γN,t + (|ϕt ihϕt |)2
£ (1) ¤ (5.69)
6 2 − 2 TrH1 γN,t |ϕt ihϕt |
£ (1) ¤
= −2 TrH1 |ϕt ihϕt | (γN,t − |ϕt ihϕt |) .

Since, by assumption,
Z +∞ ¡ (1) ¢
lim dt ρt , γN,t − |ϕt ihϕt | L2 = 0
N →∞ −∞ reg (5.70)
1
∀ρ· ∈ L (R, L2reg ) ,

then Z +∞ ° (1) °2
lim dt f (t) ° γN,t − |ϕt ihϕt | °L2 = 0 . (5.71)
N →∞ −∞
5.7. Strengthened convergence in the AGT theorem 103

Step 3. Here one arrives to the trace class norm to prove (5.61). First one has
° (1) °
° γ − |ϕt ihϕt | ° 1 =
N,t L
° (1) °
°
= ( |ϕt ihϕt | + ( 1 − |ϕt ihϕt | ) ( γN,t − |ϕt ihϕt | ) °L1
° (1) ° ° (1) °
6 ° |ϕt ihϕt | ( γN,t − |ϕt ihϕt | ) °L1 + ° ( 1 − |ϕt ihϕt | ) γN,t °L1
° ° (5.72)
+ ° ( 1 − |ϕt ihϕt | ) |ϕt ihϕt | °L1
° (1) ° £ (1) ¤
= ° |ϕt ihϕt | ( γN,t − |ϕt ihϕt | ) °L1 + 1 − Tr |ϕt ihϕt | γN,t
° (1) ° £ (1) ¤
= ° |ϕt ihϕt | ( γN,t − |ϕt ihϕt | ) °L1 − Tr |ϕt ihϕt | ( γN,t − |ϕt ihϕt | ) ,

and, by (5.68),
Z +∞ £ (1) ¤
lim dt f (t) Tr |ϕt ihϕt | ( γN,t − |ϕt ihϕt | ) = 0 . (5.73)
N →∞ −∞

(1)
At any time t and for all N , all the operators |ϕt ihϕt | ( γN,t − |ϕt ihϕt | ) have rank one,
so their trace class and their Hilbert-Schmidt norms coincide and
Z +∞
° (1) °
dt f (t) ° |ϕt ihϕt | ( γN,t − |ϕt ihϕt | ) °L1 =
−∞
Z +∞ ° (1) °
= dt f (t) ° |ϕt ihϕt | ( γN,t − |ϕt ihϕt | ) °L2
−∞
Z (5.74)
+∞ ° (1) ° ° °
6 dt f (t) ° γN,t − |ϕt ihϕt | °L2 · ° |ϕt ihϕt | °L2
−∞
Z +∞ ° (1) °
= dt f (t) ° γN,t − |ϕt ihϕt | °L2 .
−∞

Write the r.h.s. as a scalar product in L2 (R)


Z p p
+∞ ° (1) °
dt f (t) · f (t) ° γN,t − |ϕt ihϕt | °L2
−∞
³ Z +∞ ´1/2 ³ Z +∞ ° (1) °2 ´1/2
6 dt f (t) dt f (t) ° γN,t − |ϕt ihϕt | °L2 (5.75)
−∞ −∞
³ Z +∞ ° (1) °2 ´1/2
= dt f (t) ° γ − |ϕt ihϕt | ° 2
N,t L
−∞

so that, by (5.71),
Z +∞ ° (1) °
lim dt f (t) ° |ϕt ihϕt | ( γN,t − |ϕt ihϕt | ) °L1 = 0 . (5.76)
N →∞ −∞

Then, plugging (5.73) and (5.76) into (5.72), one gets


Z +∞ ° (1) °
lim dt f (t) ° γN,t − |ϕt ihϕt | °L1 = 0 (5.77)
N →∞ −∞

which completes the proof.


104 Chapter 5. Strengthened convergence for reduced density matrices

5.8 Unified strengthened convergence


from ESY/AGT theorems
In this section a unified strong convergence will be shown of any k-th marginal to the
k-th tensor power of the projection onto the solution of the CNSE. In other words, it is
possible to obtain the same stronger convergence both if the starting point is the result
of ESY theorem and if one starts from AGT theorem. Thus, despite the difference in
the statement of such theorems (and in the techniques to derive them), they both admit
the corollary that convergence can be lifted to trace-norm convergence at any time.
Since half of this program has been completed in Sec. 5.6, all that remains to do is
to prove the following.

Theorem 5.8.1 (AGT strengthened convergence at any time, [85]). For any
positive integer k, limit (5.60) in the improved version of AGT theorem implies, and
hence can be lifted to, ° (k) °
lim ° γN,t − |ϕt ihϕt |⊗k °L1 = 0 (5.78)
N →∞

for any time t ∈ R.

Theorems 5.6.1 and 5.8.1 together provide a unification to the same strong con-
vergence at any time, starting from two distinct weak convergences which have been
obtained in ESY and AGT theorem, respectively, through a similar conceptual scheme
but different techniques.
(k)
Let us focus on the origin of such differences. In ESY theorem, marginals γN,t ’s
are proved to have a bounded trace class norm uniformly in N and in t. By weak-∗
(k)
compactness of the ball they all belong to, one extracts weak-∗ limit points γ∞,t : each
(k)
sequence {γ∞,t }k>1 of limit points is shown to solve the infinite BBGKY hierarchy with
(k)
initial data γ∞,0 = |ϕihϕ|⊗k . This is done at any time t in the space L1 (L2 (R3k )).
(k)
On the other hand, objects of interest in AGT theorem are not the operators γN,t ’s at
(k)
any fixed time, but instead the operator-valued functions γN,t ∈ L∞ (R, L2reg (L2 (R3k )))
(k) (k)
and one proves that γN,· → γ∞,· weakly-∗ in such a space. This choice, in turn, is
forced by the way one controls that each limit point is the unique solution of the infinite
hierarchy: this is not done at any t as in ESY theorem, but instead it is done in terms of
functions (t, Xk , Yk ) 7→ γ∞,t (Xk , Yk ): they are proved to solve the infinite hierarchy (for
kernels) in the sense of distributions D0 (R2k+1 ), that is, 2k space coordinates and one
time coordinate.
Let us prepare some preliminaries to the proof of theorem 5.8.1. The notation

fN (t) 6 const(N
/,\)
t (5.79)

will indicate that the sequence of functions {fN }N >N0 is uniformly in L∞ (R), i.e., fN (t)
is bounded by a constant independent of N and t, for all t ∈ R and for all N exceeding
a fixed N0 . Our strategy will be to estimate time derivatives by regularity in space:
for ϕt (x) this is done by shifting from the l.h.s. to the r.h.s. of the CNSE, while for
(1)
the kernel γN,t (x, y) this is done by shifting from the l.h.s. to the r.h.s. of the P.D.E. it
satisfies, namely the first equation of the BBGKY hierarchy for marginals. The regularity
properties we will need are summarized as follows.
5.8. Unified strengthened convergence from ESY/AGT theorems 105

Theorem 5.8.2. Under the hypotheses of AGT theorem, the following holds.
¬ The first equation of the finite BBGKY hierarchy reads, in terms of kernels,
(1) ~2 (1)
i~ ∂t γN,t (x, y) = (−∂x2 + ∂y2 ) γN,t (x, y) +
Z 2m (5.80)
£ (2)
+ (N − 1) VN (x − z) − VN (y − z)] γN,t (x, z; y, z) dz .
R

(k)
­ For each integer k, γN,t has bounded L2reg -norm uniformly in N and in t. When
k = 1, in terms of kernels this reads
Z
¯ ¯
¯(1 − ∂x2 )1/2 (1 − ∂y2 )1/2 γ (1) (x, y)¯2 dx dy < const(N
/,\)
t . (5.81)
N,t
R2
(1)
This, in turns, implies that the kernel γN,t (x, y) is bounded in H 1 (R2 ) and then
also in L2 (R2 ), uniformly in N and in t.

® Define
Z
£ (2)
σN,t (x, y) := (N − 1) VN (x − z) − VN (y − z)] γN,t (x, z; y, z) dz . (5.82)
R

Then σN,t (x, y) is the kernel of an operator σN,t that has bounded L2reg -norm, and
then also L2 -norm, uniformly in N and in t. Equivalently, the kernel σN,t (x, y) is
bounded in H 1 (R2 ) and then in L2 (R2 ), uniformly in N and in t.
Proof. (5.80) and (5.81) are obtained along the proof of AGT theorem (see [3], equa-
tions (1.20) and (1.14), respectively). Also, (5.81) gives immediately uniform H 1 (R2 )-
(1)
boundedness of γN,t (x, y):
Z
¯ ¯
const(N/,\)
t > ¯(1 − ∂x2 )1/2 (1 − ∂y2 )1/2 γ (1) (x, y)¯2 dx dy
N,t
2
ZR
(1) (1)
= γN,t (x, y) (1 − ∂x2 ) (1 − ∂y2 ) γN,t (x, y) dx dy
ZR
2
(5.83)
(1) 2 2 (1)
> γN,t (x, y) (1 − ∂x − ∂y ) γN,t (x, y) dx dy
R2
(1)
= kγN,t ( · , · )k2H 1 (R2 ) .

Finally, according to proposition 2.4. of [3], the operator with kernel


Z
(2)
σ
eN,t (x, y) := VN (x − z) γN,t (x, z; y, z) dz (5.84)
R

has uniformly bounded L2reg -norm estimated by


32π (2)
σN,t k2L2reg (L2 (R)) 6
ke 2
kV̂ k2L∞ (R) kγN,t k2L2reg (L2 (R2 )) , (5.85)
N
whence
(2)
kσN,t k2L2reg (L2 (R)) 6 64 π(1 − 1/N )2 kV̂ k2L∞ (R) kγN,t k2L2reg (L2 (R2 ))
(5.86)
6 const(N
/,\)
t

(recall that each k-th marginal turns out to be L2reg (L2 (Rk ))-bounded uniformly in N
and in t, the bound depending only on k).
106 Chapter 5. Strengthened convergence for reduced density matrices

We are now ready to prove theorem 5.8.1.

Proof (of theorem 5.8.1). We give details only for k = 1 case, the generic case being
analogous, after a natural modification of theorem 5.8.2 to higher values of k. Also, it
suffices to prove L2 -norm convergence at any fixed t: then one applies theorem 3.4.3 to
go from Hilbert-Schmidt to trace class norm.
The key point of the proof is the following. By assumption
Z +∞
° (1) °2
0 = lim dt f (t) ° γN,t − |ϕt ihϕt | °L2
N →∞ −∞
Z +∞ (5.87)
1
= lim dt f (t) GN (t) , ∀ f ∈ L (R)
N →∞ −∞

having set ° (1) °2


GN (t) := ° γN,t − |ϕt ihϕt | °L2 . (5.88)

To prove that GN → 0 pointwise, let t̃ be any fixed time and let I 3 t̃ be a finite measure
interval in R: we will show that

kGN kH 1 (I) < const , uniformly in N . (5.89)

If (5.89) holds, then {GN }N is a sequence of uniformly bounded functionals on H −1 (I).


Since L1 (I) is dense in H −1 (I), then (5.87) reads
Z
lim dt f (t) GN (t) = 0 , ∀ f ∈ H −1 (I) . (5.90)
N →∞ I

Choosing f (t) = δ( t − t̃ ) one has GN ( t̃ ) → 0 (notice that δ ∈ H −σ ∀σ > 21 ). Since t̃ is


arbitrary, then ° (1) °2
lim ° γN,t − |ϕt ihϕt | °L2 = 0 (5.91)
N →∞

∀ t ∈ R. Finally, by theorem 3.4.3,


° (1) °
lim ° γN,t − |ϕt ihϕt | °L1 = 0 (5.92)
N →∞

∀ t ∈ R.
So one needs to prove that
Z
¡ ¢
kGN k2H 1 (I) = |GN (t)|2 + |G0N (t)|2 dt < const (5.93)
I

uniformly in N . The first summand in the integrand of (5.93) is easily bounded:


° (1) °4
|GN (t)|2 = ° γN,t − |ϕt ihϕt | °L2
¡ ° (1) °4 ° °4 ¢ (5.94)
6 8 ° γN,t °L2 + ° |ϕt ihϕt | °L2 6 16

whence Z
|GN (t)|2 dt 6 16|I| (5.95)
I

|I| being the measure of the interval I. Thus, the non-trivial task is to estimate the
second summand in the integrand of (5.93).
5.8. Unified strengthened convergence from ESY/AGT theorems 107

We compute
Z
d ¯ (1) ¯
G0N (t) = ¯ γ (x, y) − ϕt (x)ϕt (y) ¯2 dx dy
dt R2 N,t
Z ³
d (1) (1) (1) (5.96)
= γN,t (x, y) γN,t (x, y) − γN,t (x, y) ϕt (x)ϕt (y)
dt R2
´
(1)
− γN,t (x, y)ϕt (x)ϕt (y) + |ϕt (x)|2 |ϕt (y)|2 dx dy

and we treat separately each of the four summands R in the r.h.s. of (5.96). The fourth,
actually, does not give any contribution, since R2 |ϕt (x)|2 |ϕt (y)|2 dx dy = 1, so its time
derivative is zero. Let us denote the others by
Z
d (1) (1)
I1 (N, t) := γ (x, y) γN,t (x, y) dx dy
dt R2 N,t
Z
d (1)
I2 (N, t) := γ (x, y) ϕt (x)ϕt (y) dx dy (5.97)
dt R2 N,t
Z
d (1)
I3 (N, t) := γN,t (x, y)ϕt (x)ϕt (y) dx dy = I2 (N, t) .
dt R2
The first term gives
|I1 (N, t)| =
¯Z ³ ´ ¯
¯ (1) (1) (1) (1) ¯
=¯ ∂t γN,t (x, y) · γN,t (x, y) + γN,t (x, y) · ∂t γN,t (x, y) dx dy ¯
R2
¯Z ¯
¯ (1) (1) ¯
6 2¯ γN,t (x, y) · ∂t γN,t (x, y) dx dy ¯
R2
Z
~ ¯¯ (1)
h
(1)
i ¯
¯ (5.98)
= ¯ γN,t (x, y) · (−∂x2 + ∂y2 ) γN,t (x, y) + σN,t (x, y) dx dy ¯
m R2
Z ³
~ ¯ ¯ ¯ ¯ ´
6 ¯∂x γ (1) (x, y)¯2 + ¯∂y γ (1) (x, y)¯2 dx dy
N,t N,t
m R2
¯ Z ¯
~ ¯ (1) ¯
+ ¯ γN,t (x, y) σN,t (x, y) dx dy ¯
m R2
where equation (5.80) has been used. By theorem 5.8.2,
Z ³
¯ ¯ ¯ ¯ ´
¯∂x γ (1) (x, y)¯2 + ¯∂y γ (1) (x, y)¯2 dx dy
N,t N,t
R2 (5.99)
(1)
< kγN,t ( · , · )k2H 1 (R2 ) < const(N
/,\)
t
and
¯Z ¯
¯ (1) ¯
¯ γN,t (x, y) σN,t (x, y) dx dy ¯
R2 (5.100)
(1)
< kγN,t ( · , · )kL2 (R2 ) kσN,t ( · , · )kL2 (R2 ) < const(N
/,\)
t
so that
|I1 (N, t)| < const(N
/,\)
t . (5.101)
The second and third terms give
|I2 (N, t)| = |I3 (N, t)| =
¯Z ³
¯ (1) (1)
=¯ ∂t ϕt (x) · ϕt (y) · γN,t (x, y) + ϕt (x) · ∂t ϕt (y) · γN,t (x, y) (5.102)
R2 ¯
´
(1) ¯
+ ϕt (x) · ϕt (y) · ∂t γN,t (x, y) dx dy ¯ .
108 Chapter 5. Strengthened convergence for reduced density matrices

Set
Z
(1)
J1 (N, t) := ∂t ϕt (x) · ϕt (y) · γN,t (x, y) dx dy
R2
Z
(1)
J2 (N, t) := ϕt (x) · ∂t ϕt (y) · γN,t (x, y) dx dy (5.103)
2
ZR
(1)
J3 (N, t) := ϕt (x) · ϕt (y) · ∂t γN,t (x, y) dx dy
R2

so that
|I2 (N, t)| = |I3 (N, t)| 6 |J1 (N, t)| + |J2 (N, t)| + |J3 (N, t)| . (5.104)

Let us now show the uniform boundedness of each Ji , i = 1, 2, 3. By the CNSE (5.32),
¯Z ¯
¯ (1) ¯
|J1 (N, t)| = ¯ ∂t ϕt (x) · ϕt (y) · γN,t (x, y) dx dy ¯
R2
¯Z ³ ~2 ´ ¯
¯ (1) ¯
=¯ − ∂x2 ϕt (x) + b1 |ϕt (x)|2 ϕt (x) ϕt (y) γN,t (x, y) dx dy ¯
R2 2m
Z (5.105)
~ ¯
2
¯ (1) ¯
¯
6 ¯ ∂x2 ϕt (x) ϕt (y) γN,t (x, y) dx dy ¯
2m R2
Z
~2 b1 ¯¯ (1)
¯
¯
+ ¯ |ϕt (x)|2 ϕt (x) ϕt (y) γN,t (x, y) dx dy ¯
2m R2

By Schwartz inequality and by theorem 5.8.2,


¯Z ¯
¯ (1) ¯
¯ ∂x2 ϕt (x) ϕt (y) γN,t (x, y) dx dy ¯
R2
¯Z ¯
¯ (1) ¯
=¯ ∂x ϕt (x) ϕt (y) · ∂x γN,t (x, y) dx dy ¯
R2
(1)
(5.106)
6 k∂x ϕt ( · )kL2 (R) kϕt ( · )kL2 (R) k∂x γN,t ( · , · )kL2 (R2 )
(1)
6 kϕt ( · )kH 1 (R) kγN,t ( · , · )kH 1 (R2 )
6 const(N
/,\)
t

and
¯Z ¯
¯ (1) ¯
¯ |ϕt (x)|2 ϕt (x) ϕt (y) γN,t (x, y) dx dy ¯
R2
Z
2 (1)
6 kϕt ( · )kL∞ (R) |ϕt (x)| |ϕt (y)| |γN,t (x, y)| dx dy
R2
(5.107)
(1)
6 kϕt ( · )k2L∞ (R) kϕt ( · )k2L2 (R) kγN,t ( · , · )kL2 (R2 )
6 const(N
/,\)
t .

The treatment of J2 is completely analogous. So

|J1 (N, t)| 6 const(N


/,\)
t
(5.108)
|J2 (N, t)| 6 const(N
/,\)
t .
5.8. Unified strengthened convergence from ESY/AGT theorems 109

We now estimate J3 :
Z
2m 2m ¯¯ (1)
¯
¯
|J3 (N, t)| = ¯ ϕt (x) · ϕt (y) · ∂t γN,t (x, y) dx dy ¯
~ ~ R2
¯Z h i ¯
¯ (1) ¯
=¯ ϕt (x) ϕt (y) (−∂x2 + ∂y2 ) γN,t (x, y) + σN,t (x, y) dx dy ¯
R 2
¯Z ¯
¯ (1) ¯
6¯ ∂x ϕt (x) ϕt (y) · ∂x γN,t (x, y) dx dy ¯ (5.109)
R 2
¯Z ¯
¯ (1) ¯
+¯ ϕt (x) ∂y ϕt (y) · ∂y γN,t (x, y) dx dy ¯
R 2
¯Z ¯
¯ ¯
+¯ ϕt (x) ϕt (y) · σN,t (x, y) dx dy ¯
R2

where we have used equation (5.80). By Schwartz inequality and theorem 5.8.2
¯Z ¯
¯ (1) ¯
¯ ∂x ϕt (x) ϕt (y) · ∂x γN,t (x, y) dx dy ¯
R2
(1)
6 k∂x ϕt ( · )kL2 (R) kϕt ( · )kL2 (R) , k∂x γN,t ( · , · )kL2 (R2 ) (5.110)
(1)
6 kϕt ( · )kH 1 (R) kγN,t ( · , · )kH 1 (R2 )
6 const(N
/,\)
t ,

and analogously
¯Z ¯
¯ (1) ¯
¯ ϕt (x) ∂y ϕt (y) · ∂y γN,t (x, y) dx dy ¯ 6 const(N
/,\)
t , (5.111)
R2

and also
¯Z ¯
¯ ¯
¯ ϕt (x) ϕt (y) · σN,t (x, y) dx dy ¯
R2
6 kϕt ( · )k2L2 (R) kσN,t ( · , · )kL2 (R2 ) (5.112)

6 const(N
/,\)
t .

So
|J3 (N, t)| 6 const(N
/,\)
t . (5.113)
Plugging (5.108) and (5.113) into (5.104), one gets

|I2 (N, t)| = |I3 (N, t)| 6 const(N


/,\)
t . (5.114)

Finally, by (5.101) and (5.114), one gets

|G0N (t)| 6 |I1 (N, t)| + |I2 (N, t)| + |I3 (N, t)| 6 const(N
/,\)
t . (5.115)

Call C this constant. This in turn implies that


Z
|G0N (t)|2 dt 6 C 2 |I| (5.116)
I

and (5.95) and (5.116) lead to equation (5.93). The desired H 1 (I)-boundedness of GN ,
uniformly in N , is proved.
110 Chapter 5. Strengthened convergence for reduced density matrices

It is noteworthy to remark that G0N (t) 6 const(N /,\)


t amounts to the uniform bound-
edness of Z
¯ (1) ¯2
∂t ¯ γN,t (x, y) − ϕt (x)ϕt (y) ¯ dx dy ,
R2
(1)
but neither ∂t γN,t ( · , · ) nor ∂t ϕt ( · ) are separately L2 -bounded, uniformly in N and in
t.
Indeed, since ϕ solves the CLSE (5.32), it is seen that L2 (R)-boundedness of ∂t ϕt ( · )
requires H 2 (R)-boundedness of ϕt ( · ). But it is known (see, e.g., theorem 4.8.1. in [27])
that for every initial datum in H 2 there exist T± > 0 and a unique maximal solution

ϕ ∈ C((−T− , T+ ), H 2 (R)) ∩ C 1 ((−T− , T+ ), L2 (R))

of (5.32), with H 2 -norm blow-up as t → T± , so that one cannot localise the interval I in
(5.93) around any time t and independently of the initial datum.
(1) (1)
Similarly, since γN,t solves (5.80), it is seen that L2 (R2 )-boundedness of ∂t γN,t ( · , · )
(1)
requires uniform H 2 (R2 )-boundedness of γN,t ( · , · ), while only uniform H 1 (R2 )-bounded-
ness is known.
(1)
Uniform L2 -boundedness of ∂t γN,t ( · , · ) and ∂t ϕt ( · ) would have helped the proof via
the estimate
¯d° °2 ¯¯
¯ ° (1)
0
|GN (t)| = ¯ γ − |ϕt ihϕt | °L2 ¯
dt N,t (5.117)
° (1) ° ° °
6 4 ° ∂t γN,t ( · , · ) °L2 (R2 ) + 4 ° ∂t ϕt ( · ) °L2 (R2 ) ,

the inequality following by


¯d° Z
¯ ° °2 ¯¯ ¯¯ d ¯ ¯ ¯
¯ °
ρN,t L2 ¯ 6 ¯ ¯ ρN,t (x, y) ¯2 dx dy ¯¯
dt dt R2
¯Z ³ ´ ¯
¯ ¯
=¯ ∂t ρN,t (x, y) · ρN,t (x, y) + ρN,t (x, y) · ∂t ρN,t (x, y) dx dy ¯
2
°R ° ° °
6 2 ° ∂t ρN,t ( · , · ) °L2 (R2 ) ° ρN,t ( · , · ) °L2 (R2 ) ,

(1) ° °
where ρN,t = γN,t − |ϕt ihϕt | (so that ° ρN,t ( · , · ) °L2 (R2 ) 6 2).
Instead of estimate (5.117), our strategy has been to perform the full computation
in (5.96), with the insertion of the CNSE or the first BBGKY equation in each of the
resulting summands (5.97). This enabled us to substitute each time derivative with a
double space derivative and, consequently, to deal with duality products of H −1 functions
tested on H 1 -functions, thus leading to the desired uniform boundedness.
Chapter 6

Interparticle correlations
in the condensate

Interaction among particles establishes correlations in the many-body state: their the-
oretical understanding is crucial both in view of their experimental observation and for
their influence on other features of the state under consideration. In the presence of
B.E.C., the question becomes more concrete, because one expects that the condensed
state is roughly speaking a totally factorised state.
In fact, in the trivial case of systems of non-interacting particles, B.E.C. shows up as
a totally factorised N -body state. In the more realistic case of even weakly interacting
systems, if condensation is present, e.g., in the pure state ΨN , and if the condensate
wave function is recognised to be some ϕ, one would still be tempted to make some
approximation ΨN ≈ ϕ⊗N . For instance (see Sec. 1.3), observation of B.E.C. is carried
on by illuminating the cloud of atoms of a cooled and trapped alkali gas by resonant
laser light and its shadow is imaged onto a CCD camera: the tacit assumption behind
this discrete counting is that “almost all” particles have their own individuality in the
one-body state ϕ, as it would be in the N -th tensor power ϕ⊗N .
So here the problem of correlations is the problem of to what extent a condensed state
differs from some ϕ⊗N , and in which sense this difference can be expressed and detected.
In this chapter, following the work in progress [82], we are discussing the reasonable way
to state and to approach to such a problem, together with some related results.
Discussing correlations at finite N , within an analysis where rigorous results are
asymptotic only and where the system appears as totally factorised as N → ∞ (at least
when observed at any fixed k-body level), turns out to be a rather subtle issue. In
Sec. 6.1 we will sketch it and its main difficulties, emphasizing that the discussion has to
be carried on simultaneously at the many-body state level and at the one-body marginal
level. This is done by introducing a trial wave function showing condensation at the level
of marginals and reproducing the energy of the true state.
In Sec. 6.2 we focus on the specific form of correlations emerging in the ground state
of a dilute Bose gas with B.E.C. Their effect is to lower by a macroscopic quantity the
expectation value of the energy in the corresponding uncorrelated state. This depends on
the presence of an appropriate correlation structure which modifies the purely factorised
ϕ⊗N , independently of the latter, as discussed in Sec. 6.3.
The counterpart at the level of marginals is a typical correlation structure which turns
out to be crucial in many aspect of the rigorous analysis on B.E.C., as the determination

111
112 Chapter 6. Interparticle correlations in the condensate

of the correct evolutionary hierarchy for reduced density matrices (Sec. 6.4).
In fact, emergence and persistence of short scale correlations is recognised in the evo-
lution of the condensate, although one still misses an understanding of the full correlation
structure accounting for the energy of a condensate (Sec. 6.5 and 6.5).

6.1 Correlations in many-body condensed states


Let us start with focusing on the main physical and technical obstructions emerging in
controlling the difference (ΨN − ϕ⊗N ) in some sense, where ΨN is the many-body state
with B.E.C. and ϕ is the one-body condensate wave function. There are difficulties both
at the many body level (“M”) and at the reduced one-body level (“B”).

(M1 ) ΨN is not accessible explicitly, the many-body problem is too hard.

(M2 ) One can instead introduce an explicit trial wave function Ψtrial
N , which has not the
form of a tensor product, but by construction is close in norm to ϕ⊗N . If one could
prove that ΨN and Ψtrial
N too are close in norm, then ΨN would be almost factorised
trial
as ΨN , which would provide a reasonable structure of the correlations in the true
N -body state. Yet, since the distance is taken with the k kHN -norm, the choice of
the trial function is extremely delicate, as N is very large. A “physically irrelevant”
change in Ψtrial
N can make it essentially orthogonal to the ΨN , or to the ϕ⊗N it was
close to. For example, if ϕ, ξ ∈ H are two orthonormal one-body states, then1
° ¡ ⊗(N −1) ¢ ° √
° ϕ ⊗ξ − ϕ⊗N °
= 2 (6.1)
sym H N

independently of N , while one would tend to say that the distinction among such
two N -body states is negligible as N → ∞, since they are two bosonic state differing
only by one particle out of N . So, evaluating distances in norm is “too detailed” –
it carries information on all particles – and does not provide a useful comparison.

(M3 ) Although example in (6.1) may be fictitious with respect to realistic condensed
states, which may indeed be close in norm to some ϕ⊗N , it is not trivial at all
that Ψtrial
N and ΨN are close also in the energy form. In fact, ϕ⊗N typically does
not approximates in this sense the “true” ΨN (the Hamiltonian being unbounded,
vectors close in norms are not necessarily close in the form of HN ) and this causes
difficulties in choosing a good trial function: correlations deeply influence the com-
putation of hHN i. We are discussing a concrete example of that in Sec. 6.2 and
6.3.

At the one-body level things seem more promising, at least because, unlike (M1 ), it
is easier to handle marginals and, unlike (M2 ), by computing distances of marginals one
1
To prove (6.1), recall that, by (3.20),
‚ ` ⊗(N −1) ´ ‚2
‚ ϕ ⊗ξ − ϕ⊗N ‚H =
sym N
‚ 1 1 ‚2
= ‚ √ ξ ⊗ ϕ ⊗ · · · ⊗ ϕ + √ ϕ ⊗ ξ ⊗ · · · ⊗ ϕ + · · · − ϕ ⊗ ϕ ⊗ · · · ⊗ ϕ ‚H ,
N N N

that is, the norm of a linear combination of N + 1 orthonormal terms: the first N terms contribute with
( √1N )2 · N , the last one contributes with 1, so that the r.h.s. amounts to 2.
6.1. Correlations in many-body condensed states 113

gets rid of physically irrelevant discrepancies between ΨN , Ψtrial


N , and ϕ
⊗N . This is the

case with respect to the example in (6.1): by theorem 3.3.1,

(1) N −1 1
γN = |ϕihϕ| + |ξihξ| (6.2)
N N

is the one-particle reduced density matrix corresponding to (ϕ⊗(N −1) ⊗ ξ)sym and

° (1) ° 1° ° 2 N →∞
° γ − |ϕihϕ| ° = ° |ξihξ| − |ϕihϕ| ° 6 −−−−→ 0 . (6.3)
N L1 (H) N L1 (H) N

Instead, one encounters another kind of difficulties in characterizing correlations


present in the many-body state.
(1)
(D1 ) For finite N , in general γN (or any k-th marginal) does not reconstruct γN
uniquely, therefore does not controls correlations in the N -body state. See, e.g.,
the elementary example discussed before definition 3.3.2.
(1)
(D2 ) As N → ∞, convergence of γN to a projection does not imply in general factori-
sation of ΨN in the limit. This is just the case of (6.3) vs (6.1).

(D3 ) It may happen that the effects of two-body correlations in the many-body wave
function essentially do not contribute to the first (or any k-th) marginal.

Thus, a rigorous analysis of correlations in the condensed many-body state relies on


a compromise (and a synergy) in adopting distinct tools.

General strategy. One tries to build explicit interparticle correlations in a trial wave
function Ψtrial
N in such a way that:

(1),trial
i) the corresponding γN is essentially |ϕihϕ| as it would be for the completely
factorised ϕ⊗N ,

ii) Ψtrial
N ≈ ΨN in norm,

iii) Ψtrial
N ≈ ΨN also in the energy form.

For i) to hold it suffices that Ψtrial


N ≈ ϕ⊗N in norm, since by (2.105)
° (1),trial ° ° °
°γ − |ϕihϕ| ° 6 2 ° Ψtrial
N −ϕ
⊗N °
.
N L1 (H) H
N

By the same estimates (2.72) and (2.105), i) and ii) prove that ΨN has complete con-
densation:
° (1) ° ° (1) (1),trial ° ° (1),trial °
° γ − |ϕihϕ| ° 6 ° γN − γN ° 1 + °γ − |ϕihϕ| ° 1
N L1 (H) L (H) N L (H)
° ° ° (1),trial °
6 2 ° ΨN − ΨN °H + ° γN
trial
− |ϕihϕ| °L1 (H) .
N

In all that, a limit N → ∞ enters to make asymptotic statements and only a final control
of error terms provides complete results for large finite N .
114 Chapter 6. Interparticle correlations in the condensate

6.2 Ground state correlations: Jastrow and Dyson factors


As an example, we turn back to the case of (asymptotic) 100% B.E.C. in a three-
dimensional dilute trapped Bose gas, after the analysis of Lieb, Seiringer and Yngvason
(Sec. 3.5). There, condensation is proved in the ground state Ψg.s.
N of the Hamiltonian
HN (3.77) and the condensate wave function turns out to be the minimizer ϕGP of the
GP energy functional (3.72).
N →∞
The ground state energy asymptotics (3.80), i.e., (Ψg.s. g.s.
−−−→ E GP , is
N , HN ΨN )/N −
obtained in [77] through a separate control from above and from below and the two
bounds are proved to converge to the same limit. For the upper bound, in particular,
the following trial function Ψlsy
N is used:

N
Y
WN (x1 , . . . , xN ) ϕGP (xi )
i=1
Ψlsy
N (x1 , . . . , xN ) := . (6.4)
° N
Y °
° GP °
° WN (x1 , . . . , xN ) ϕ (xi ) °
i=1

Here x1 , . . . , xN ∈ R3 , ϕGP is the GP minimizer discussed in theorem 3.5.2, and inter-


particle correlations are encoded in the term
N
Y ¡ ¢
WN (x1 , . . . , xN ) := FN ti (x1 , . . . , xi ) (6.5)
i=2

with
© ª
ti := min |xi − xj | , 1 6 j 6 i − 1 (6.6)

½
FN (r)/FN (bN ) r 6 bN
FN (r) := ; (6.7)
1 r > bN
bN is some N -indexed cut-off parameter of order
−1/3
bN ∼ ρN ∼ N −1/3 (6.8)

(that is, of the same order as the mean interparticle distance),

FN (r) := fN (x) , r := |x| , (6.9)

and fN is the spherically symmetric solution of the zero energy scattering problem
(
(−2µ∆ + VN ) fN = 0
lim fN (x) = 1 . (6.10)
|x|→∞

It can be proved that

0 6 fN (x) 6 1 (6.11)
0
FN (r) > 0 (6.12)
aN
fN (x) ∼ 1 − as |x| → ∞ (6.13)
|x|
Z
8πµaN = VN (x)fN (x) dx (6.14)
R3
6.2. Ground state correlations: Jastrow and Dyson factors 115

where aN is the s-wave scattering length of VN . Consequently, FN (r) is by construction


the constant 1 but at the scale r . N −1/3 , where it stays bounded below 1.
In view of the general strategy above, we point out that
(1),lsy
i) γN → |ϕGP ihϕGP |,
(1)
ii) γN → |ϕGP ihϕGP |, but nothing rigorous is known about Ψlsy
N ≈ ΨN in norm,

iii) Ψlsy
N ≈ ΨN in the energy form,
¡ GP ¢⊗N
iv) Ψlsy
N ≈ ϕ in norm,
¡ GP ¢⊗N
v) Ψlsy
N is far from ϕ in energy.

Point i) follows from iv), by (2.105):


° (1),lsy ° ° ¡ GP ¢⊗N °
°γ − |ϕGP ihϕGP | ° 6 ° Ψlsy
N − ϕ
° 2 3N . (6.15)
N L1 (L2 (R3 )) L (R )

Concerning point ii), the trial function (6.4) has not been proved to be close also in
norm to the true ground state (asymptotically in N ), so currently the general strategy
above is only partially addressed. Actually, in Lieb and Seiringer’s work [74] on B.E.C. for
(1)
dilute trapped gases, one controls γN by a detailed examination of ΨN , writing it (apart
from normalisation) as ΨN (x1 , . . . , xN ) ≡ ϕGP (x1 ) · · · ϕGP (xN )WN (x1 , . . . , xN ), where
WN is necessarily non-negative and permutationally symmetric and has the role of (but
it is distinct from) the analogous WN defined in (6.5). By combining the “localization of
the energy” technique and the generalised Poincaré inequality (see [74, 75] for details),
(1)
one finally gets B.E.C. as γN → |ϕGP ihϕGP |.
Point iii) is addressed in [77], as we have just mentioned above. There one proves
g.s. g.s.
that (Ψlsy lsy
N , HN ΨN )/N > (ΨN , HN ΨN )/N and that the difference between these two
quantities asymptotically vanishes, both converging to E GP : that is, Ψg.s. N and ΨN are
lsy

close in energy. (Although ΨN ∈ lsy 2 3N


/ HN,sym = Lsym (R ), it is nevertheless an acceptable
trial wave function, since bosonic and absolute ground state energies coincide.)
Point iv) is given by the following.

Theorem 6.2.1. Take Ψlsy


N as in (6.4). Then
° ¡ GP ¢⊗N °
lim ° Ψlsy
N − ϕ
° 2 3N = 0
L (R )
(6.16)
N →∞

Proof. We are going to provide an upper bound of k Ψlsy N − (ϕ


GP )⊗N k
L2 which vanishes
as N → ∞. We will be using the following: (a) ϕ GP is bounded and L2 -normalised,
(b) 0 6 FN (r) 6 1 and the support of 1 − FN is [0, bN ], which for our purposes can be
concretely taken to be [0, N −1/3 ], (c) a more refined estimate for 1 − FN available in the
literature, see (6.30).
Actually we will prove
° ¡ ¢ ° N →∞
° ΦN − ϕGP ⊗N ° 2 3N −−−−→ 0 (6.17)
L (R )

with
N
Y
ΦN (x1 , . . . , xN ) := WN (x1 , . . . , xN ) ϕGP (xi ) (6.18)
i=1
116 Chapter 6. Interparticle correlations in the condensate

(so that Ψlsy


N = ΦN /kΦN k), because, by (2.106),
° ¡ ¢⊗N °
° lsy ¡ GP ¢⊗N ° 2 ° ΦN − ϕGP ° 2 N →∞
° ΨN − ϕ ° 6 ° ¡
L
¢⊗N ° −−−−→ 0 . (6.19)
L2
1 − ° ΦN − ϕGP ° 2
L

First, let us re-write (6.5) as


N
Y
WN (x1 , . . . , xN ) = FN (|xi − xj(i) |) (6.20)
i=2

where by xj(i) one means the closest coordinate to xi among {x1 , x2 , . . . , xi−1 }. One has
° ¡ ¢ °
° ΦN − ϕGP ⊗N °2 2 3N
L (R )
Z ¯ YN ¯2 Y
N
¯ ¯ ¯ GP ¯
= ¯1 − FN (|xi − xj(i) |) ¯ ¯ϕ (xi )¯2 dx1 · · · dxN
R3N i=2 i=1
(6.21)
Z ³ N
Y ´Y N
¯ GP ¯
62 1− FN (|xi − xj(i) |) ¯ϕ (xi )¯2 dx1 · · · dxN
R3N i=2 i=1

because FN 6 1. Notice that one cannot use Lebesgue dominated convergence, because
the integration space enlarges with N .
Then set

ωN (t) := 1 − FN (t) (6.22)


(i)
ω := ωN (|xi − xj(i) |) (6.23)

Such ωN is supported in [0, N −1/3 ] with values in [0, 1]. The correlation term in the
r.h.s. integrand of (6.21) reads
N
Y N
X X
1− (1 − ω (i) ) = ω (i) − ω (i) ω (j) + · · · + (−)N ω (2) ω (3) · · · ω (N )
i=2 i=2 26i<j6N
(6.24)
N
X −1 X
k+1 (i1 ) (i2 ) (ik )
= (−) ω ω ···ω
k=1 26i1 <···<ik 6N

so that
N
X −1 X (N,k)
(6.21) 6 2 (−)k+1 Ii1 ,...,ik (6.25)
k=1 26i1 <···<ik 6N

with
Z k
Y N
Y ¯ GP ¯
(N,k)
Ii1 ,...,ik := ωN (|xis − xj(is ) |) ¯ϕ (xi )¯2 dx1 · · · dxN . (6.26)
R3N s=1 i=1

(N,k)
In each Ii1 ,...,ik the ωN -factors depend on the variables xi1 , xi2 , . . . , xik and other
variables xj(i1 ) , xj(i2 ) , . . . , xj(ik ) that, by construction, are among {x1 , x2 , . . . , xik }, since
ik is the largest index among {i1 , i2 , .R. . ik }. Variables xik +1 , . . . , xN are only in the
ϕGP -factors and are integrated out by R3 |ϕGP (x)|2 dx = 1.
6.2. Ground state correlations: Jastrow and Dyson factors 117

Next, partition the remaining domain of integration, that is, R3ik , into separated
regions where coordinates xj(i1 ) , xj(i2 ) , . . . , xj(ik ) are indexed by constant labels. In each
of these regions, take the coordinate change


 zi1 = xi1 − xj(i1 )

 zi = xi − xj(i )
2 2 2
. . (6.27)

 .
. .
.


zik = xik − xj(ik )

leaving all the other variables unchanged. In the new variables, the ωN -factors take the
Q
form ks=1 ωN (|zis |), then there are k factors each with the form |ϕGP (zis − xis )|2 and
all the others are expressed in the original variable. Bound each |ϕGP (zis − xis )|2 by
kϕGP k2∞ : this decouples the variables zi1 , . . . , zik from the others, so that the remaining
ϕGP -factors can be integrated out. This can be done for all the disjoint regions of the
integration domain.2 Thus,
Z k
Y
(N,k) ¡ ¢k
Ii1 ,...,ik 6 kϕGP k2∞ ωN (|zis |) dzi1 · · · dzik
R3k s=1
(6.28)
¡ ¢k ³Z ´k
= kϕGP k2∞ ωN (|z|) dz .
R3

A more refined estimate is now needed than


Z Z N −1/3 Z N −1/3
2 4π 1
ωN (|z|) dz = 4π ω(r) r dr 6 4π r2 dr = . (6.29)
R3 0 0 3 N

To this aim, let us make use of (7.24), which actually applies to a fN very similar to the
one we are dealing with, which is defined in (6.10). There fN solves the lowest energy
scattering problem in a ball, and is 1 outside; here it is assumed to solve the zero-energy
scattering problem on the whole space, with the condition to be 1 at infinity. It can be
easily checked that (7.24) in the current case leads to

FN (|z|) aN
ωN (|z|) = 1 − FN (|z|) = 1 − 6 1 − fN (z) 6 1B (z) (6.30)
FN (bN ) |z| + aN N
2
As an example, take
Z 6
Y ˛ GP ˛
(N,3)
I3,5,6 = ωN (|x3 − xj(3) |) ωN (|x5 − xj(5) |) ωN (|x6 − xj(6) |) ˛ϕ (xi )˛2 dx1 · · · dx6
(R3 )6 i=1

3 6
and in the region Ω of (R ) where, e.g., xj(3) = x1 , xj(5) = x1 , xj(6) = x5 , one has
Z
˛ ˛2 ˛ ˛2 ˛ ˛2
ωN (|z3 |) ωN (|z5 |) ωN (|z6 |) ˛ϕGP (x1 )˛ ˛ϕGP (x2 )˛ ˛ϕGP (z3 + x1 )˛ ·
(N,3)
I3,5,6 =

˛ ˛2 ˛ ˛2 ˛ ˛2
· ˛ϕGP (x4 )˛ ˛ϕGP (z5 + x1 )˛ ˛ϕGP (z6 + z5 − x1 )˛ dx1 dx2 dz3 dx4 dz5 dz6
Z
` ´3 ˛ ˛2 ˛ ˛2
6 kϕGP k2∞ ωN (|z3 |) ωN (|z5 |) ωN (|z6 |) ˛ϕGP (x1 )˛ ˛ϕGP (x2 )˛ ·

˛ ˛2
· ˛ϕGP (x4 )˛ dx1 dx2 dz3 dx4 dz5 dz6

so that Z
(N,3) ` ´3 “ ”3
I3,5,6 6 kϕGP k2∞ ωN (|z|) dz .
R3
118 Chapter 6. Interparticle correlations in the condensate

where 1BN is the characteristic function of the ball of radius bN = N −1/3 in R3 , aN =


N0 a/N by R(3.76) and, with no loss of generality, we take here aN = N −1 . Plugging
(6.30) into ωN , one gets
Z Z Z N −1/3 2
aN r
ωN (|z|) dz 6 = 4πaN dr
R3 |z|6N −1/3 |z| + aN 0 r + aN
Z 2/3 2
4π N s
= 3 ds
N 0 s+1 (6.31)
Z N 2/3

6 3 s ds
N 0

= 5/3
N
which is a refined version of (6.29). Plugging (6.31) into (6.28), one gets

(N,k) Ck
Ii1 ,...,ik 6 . (6.32)
N 5k/3
for some positive constant C.
P (N,k)
The whole sum I in (6.25), at fixed k and N , is in turn estimated noticing
¡N −1i¢1 ,...ik
that it is made of k summands, corresponding to all possible choices of arranging
(N,k)
indices in such a way that 2 6 i1 < · · · < ik 6 N , each leading to a Ii1 ,...,ik which is
estimated by (6.32). Then (6.25) and (6.32) give
N
X −1 X (N,k)
(6.21) 6 2 (−)k+1 Ii1 ,...,ik
k=1 26i1 <···<ik 6N
µ ¶ (6.33)
N
X −1 N
X −1
N − 1 Ck
6 2 (−)k+1 = (−)k+1 AN (k)
k=1
k N 5k/3 k=1

with µ ¶
N − 1 2 Ck
AN (k) := . (6.34)
k N 5k/3
In the above sum, terms alternate their sign, starting with AN (1) > 0. Moreover,
AN (k + 1) C
= 6 1, ∀ k = 1, 2, . . . , N − 1 (6.35)
AN (k) (k + 1)(N − k) N 5/3
(eventually as N → ∞), whence AN (k + 1) 6 AN (k). This means that
N
X −1
(6.21) 6 (−)k+1 AN (k)
k=1
= AN (1) − AN (2) + AN (3) + · · · + (−)N AN (N − 1)
(6.36)
N −1
6 AN (1) = 2 C
N 5/3
1 N →∞
6 const · 2/3 −−−−→ 0
N
and (6.17) is proved. (One can see that the refined estimate (6.31) is really necessary,
because the original estimate (6.29) would have led to a O(1) term in the r.h.s. of
(6.36).)
6.2. Ground state correlations: Jastrow and Dyson factors 119

Last, concerning point v) above, we shall now see in (6.40) that Ψg.s. lsy
N and ΨN have
GP ⊗N
lower expectation value of the energy than (ϕ ) : this is due to the lower probability
that two particles in the state Ψlsy N stay at a distance . N
−1/3 , with respect to the

purely factorised case. Also, correlations present in the ground state decrease its energy
with respect to any tensor power N -body wave function.
To state all that rigorously, let us tune the scattering length in the GP energy func-
tional (3.72), as previously done in (4.28), by introducing a parameter ν in the generalised
functional
Z ³
¡ ¢ ~2 ´
EνGP [ϕ] = µ|∇ϕ|2 + U |ϕ|2 + 4πµ ν|ϕ|4 dx µ= , (6.37)
R3 2m
on the same variational domain D (3.73). Also, let us recall that the scattering length a
of the pair interaction V is bounded above by its first Born approximation, according to
the Spruch-Rosenberg inequality [105]
b
a6 (6.38)
8πµ
where Z
b := V (x) dx = kV kL1 . (6.39)
R3

Theorem 6.2.2. Assume the hypotheses and take the notation of theorem 3.5.2, but
with the irrelevant simplification N0 = 1. Assume, also, that V has finite L1 -norm. Then
any ξ ⊗N (with ξ ∈ D ⊂ H, kξk = 1) has asymptotically higher expectation value of the
energy than the ground state ΨN :
³ HN ⊗N ´
lim ξ ⊗N , ξ GP
= Eb/(8πµ) [ ξ ] > E GP . (6.40)
N →∞ N
Assume, in addition, that b/(8πµ) > a: then inequality (6.40) becomes strict. Then
the best approximation in energy of ΨN given by a tensor power N -body state ξ ⊗N is
achieved by ξ = ψ GP , the latter being the minimizer of Eb/(8πµ)
GP defined in the same way
and with the same features as in theorem 3.5.2.

Proof. One has


Z N
Y ³X
N
¡ ⊗N ⊗N
¢ ¡ ¢
ξ , HN ξ = ξ(xr ) · − µ∆xi + U (xi ) +
R3N r=1 i=1
X ´ YN
+ VN (xi − xj ) · ξ(xs ) dx1 · · · dxN
16i<j6N s=1
N Z
X ¡ ¢
= ξ(xi ) − µ∆xi + U (xi ) ξ(xi ) dxi +
R3 (6.41)
i=1
X Z
+ ξ(xi ) ξ(xj ) VN (xi − xj ) ξ(xi ) ξ(xj ) dxi dxj
16i<j6N R3 ×R3
Z
¡ ¢
=N µ|∇ξ(x)|2 + U (x)|ξ(x)|2 dx +
R3
Z
(N − 1)N
+ VN (x − y) |ξ(x)|2 |ξ(y)|2 dx dy .
2 R3 ×R3
120 Chapter 6. Interparticle correlations in the condensate

N →∞
By (3.78), N VN −−−−→ b δ in the distributional sense, so
¡ ⊗N ¢ Z
ξ , HN ξ ⊗N N →∞ ¡ b ¢
−−−−→ µ|∇ξ(x)|2 + U (x)|ξ(x)|2 + |ξ(x)|4 dx
N R3 2 (6.42)
GP
= Eb/(8πµ) [ ξ ] .
GP
By (6.38), Eb/(8πµ) [ ξ ] > EaGP [ ξ ]. Since ϕGP is the minimizer of EaGP ≡ E GP , then
EaGP [ ξ ] > E GP [ ϕGP ] = E GP . Then (6.40) is proved. It is also clear that, if b/(8πµ) > a,
GP
then Eb/(8πµ) [ ξ ] > EaGP [ ξ ].

It deserves to be mentioned that trial functions of the form (6.4) and analogous, are
present throughout the recent literature on rigorous B.E.C. and are all inspired by 1957
Dyson’s pioneering work [37]. Before Dyson, attempts had been made by Dingle and
Q
Jastrow. A trial wave function of the form i<j f (xi − xj ) to calculate the N -body
ground state energy had already been suggested by Mott for the hard sphere Bose gas
and applied in the low density limit by Dingle [36]. Jastrow [61] then made a systematic
use of it examining by a variational method the ground state of bosons or of fermions
Q
coupled with a strong interaction of short-range. Actually, the term i<j f (xi − xj ) in
the trial wave function typically goes under the name of Jastrow factor and it is the
primary way for including electron correlations beyond the mean-field level, especially in
quantum mechanical simulations.
While Jastrow’s trial is symmetric under permutation of variables, Dyson’s one – as
well as (6.5) – is not. Also, the latter takes into account the interaction only between
Q
nearest neighbours, while in i<j f (xi −xj ) the product is taken over all pairs of particles.
However, Jastrow was not able with his wave function to obtain a rigorous upper bound
to the energy: his expression for the energy was a series of cluster integrals which he
was obliged to cut off without controlling the error. On the other hand, Dyson’s work
showed first that the “nearest neighbour” type of wave function led to a new approach
for the description of many-body systems with strong interactions, in which the typical
difficulties associated with cluster-integrals expansions do not arise.
In the presence of the Jastrow-like factor, LSY trial function turns out to be asymp-
totically equal to a product function in norm in the Gross-Pitaevskiı̆ limit. Such a
convergence is not expected for the ground state function (nor it is expected for ESY
trial function that we will be dealing with in Section 6.5, cointaining far more correlation
factors). Thus, it is interesting that Ψlsy
N can “afford” to have less correlations that one
might expect and nevertheless give the correct energy to the leading order.

6.3 Ground-state-like correlations lower the energy


The drop of hHN i in the trial function Ψlsy
N with respect to the factorised (ϕ
GP )⊗N is

a typical phenomenon due only to the Dyson-like correlation factor (6.5), independently
of the choice of the one-body condensate wave function. That is, the same happens also
for many-body states with higher energy expectation value, not necessarily eigenstates
of HN , provided that they have the same correlation structure as Ψlsy
N .
This is a consequence of the same analysis by Lieb, Seiringer, and Yngvason for the
upper bound of the ground state energy of a dilute Bose gas showing B.E.C. in the
condensate wave function ϕGP . By direct inspection of the proof of theorem III.1 in [77],
one can see that the same theorem asserts the following more general result.
6.3. Ground-state-like correlations lower the energy 121

Theorem 6.3.1. Let HN be the Hamiltonian (3.77), i.e., the Hamiltonian of the three-
dimensional model of N trapped and interacting bosons considered in theorem 3.5.2. Let
(ϕ)
ϕ ∈ H 1 (R3 ) ∩ L∞ (R3 ) such that kϕkL2 = 1 and let ΨN be the many-body wave function

(ϕ) WN (x1 , . . . , xN ) ϕ(x1 ) · · · ϕ(xN )


ΨN := °
°WN (x1 , . . . , xN ) ϕ(x1 ) · · · ϕ(xN )°
° (6.43)

where WN is the Dyson-like factor defined in (6.5). Let E GP be the Gross-Pitaevskiı̆


energy functional (3.72). Then
³ ´ ¡ ¢
(ϕ) HN (ϕ)
ΨN , ΨN 6 E GP [ϕ] 1 + O(N −2/3 ) (6.44)
N

hence, E GP [ϕ] is an upper bound to the expectation value of the energy per particle on
(ϕ)
ΨN .

Lieb et al. made the special choice ϕ = ϕGP , although it holds in general. As a
(ϕ)
consequence, combining theorems 6.2.2 and 6.3.1, the expectation value of HN in ΨN
is eventually less than that in ϕ⊗N ,
³ HN (ϕ) ´ ³ HN ⊗N ´
(ϕ)
ΨN , ΨN 6 ϕ⊗N , ϕ , (6.45)
N N
and this is due to the Dyson-like factor WN .

Proof of theorem 6.3.1. The scheme is the same as in the original proof of Lieb et al. To
(ϕ)
agree with the notation there, we let ~ = 2m = 1 and rewrite the above ΨN as
Q
(ϕ) WN (x1 , . . . , xN ) N g(xi )
ΨN = ° Qi=1 ° (6.46)
°WN (x1 , . . . , xN ) N
g(xi )°
i=1

with
ϕ
g := . (6.47)
k ϕ kL∞
¡ (ϕ) (ϕ) ¢
Then one estimates ΨN , HNN ΨN following exactly the same line of [77]. Direct com-
putation produces a number of terms (Eq. (3.11) of [77]) which have to be suitably
bounded. For the bound of the term in Eq. (3.19) of [77] a Cauchy-Schwarz inequality
is needed, regardless of g (but provided that g ∈ L4 (R3 ), as here). For the bound of the
term in Eq. (3.21) of [77], one needs only 0 6 |g(x)| 6 1, as it is in the present case.
The only difference is that here ϕ is no longer real-valued as for the minimizer of the GP
functional; after an innocent modification one gets the estimate
Z Z
2
³ ´ ẽN |g(x)| dx JN |g(x)|4 dx
(ϕ) H N (ϕ) R3 R 3
ΨN , Ψ 6Z +Nh Z i2
N N 2 2
|g(x)| dx − N IN |g(x)| dx − N IN
R3 R3 (6.48)
2 2 2
N KN
+hZ 3
i2
|g(x)|2 − N IN
R3
122 Chapter 6. Interparticle correlations in the condensate

which is the analog of Eq. (3.29) in [77], where constants ẽN , IN , JN , and KN are defined
as in the original work, namely
Z
¡ 2
¢
IN := 1 − fN (x) dx , (6.49)
RZ3
1 ¡ 0
¢
JN := 2 (fN (x))2 + VN (x)fN
2
(x) dx , (6.50)
2 R3
Z
0
KN := fN (x)fN (x) dx , (6.51)
R 3
Z
£ ¤
− g(x) ∆g(x) + U (x)|g(x)|2 dx
R3 Z
ẽN := . (6.52)
2
|g(x)| dx
R3

Now recall from (6.8) that the cut-off parameter used by Lieb et al. in the definition
of the Dyson-like factor WN is bN ∼ N −1/3 : following the same explicit choice as in
Eq. (3.39) of [77], let us take bN such that
Z Z
4
|g(x)| dx |ϕ(x)|4 dx
4π 3 R 3 R 3
ρ b =Z = =: c , (6.53)
3 N N 2 k ϕ k2L∞
|g(x)| dx
R3

where, in analogy with the original case,


Z
ρN := N |ϕ(x)|4 dx . (6.54)
R3

Indeed (6.53) and (6.54) give


³ 4π ´−1/3
bN = N k ϕ k2L∞ ∼ N −1/3 . (6.55)
3
In terms of this bN and of the scattering length aN
³ a3 ´
N
IN 6 4π + aN bN (bN − aN ) (6.56)
3
as in Eq. (3.34) of [77]. This, in turn, leads to
Z Z ³ ´
2 N IN
|g(x)| dx − N IN > |g(x)|2 1 − R 2
dx
R3 R3 R3 |g(x)| dx
Z ³ 4π
= |g(x)|2 1 − N a3N k ϕ k2L∞ +
R 3 3
´
− 4πN aN bN (bN − aN )k ϕ k2L∞ dx (6.57)
Z ³ a3 3 aN bN (bN − aN ) ´
= |g(x)|2 1 − 3N − dx
R3 bN bN
Z ³
2 aN ´3
> |g(x)| 1 − dx
R3 bN
as Eq. (3.40) of of [77]. In the r.h.s. of (6.57) we have used
³ 4π ´−1
k ϕ k2L∞ = N b3N (6.58)
3
6.4. Correlations at the level of reduced density matrices 123

which follows from (6.55).


This way, following [77], one has all the estimates to deduce from (6.48) the desired
bound, in analogy to Eq. (3.42) of [77]:
Z
£ ¤
³ ´ |∇ϕ(x)|2 + U (x)|ϕ(x)|2 dx
(ϕ) HN (ϕ) 3
ΨN , ΨN 6 R ³ +
N aN ´3
1−
bN
2 ³ aN ´ 2 ³ aN ´2 1 ³ aN ´3
Z 1+ − + (6.59)
c bN c bN 2c bN
+ 4πa |ϕ(x)|4 ³ ´8 dx
R 3 aN
1−
bN
Z
¡ ¢ £ ¤
= 1 + O(N −2/3 ) |∇ϕ(x)|2 + U (x)|ϕ(x)|2 + 4πa|ϕ(x)|4 dx
R3

where a is the unscaled scattering length.

6.4 Correlations at the level of reduced density matrices


We go back to the difficulties (D1 ), (D2 ), and (D3 ) one has in characterizing interparticle
correlations of the many-body condensed state through the analysis at the level of k-
particle marginals.
Correlations in the N -body wave function show up as modifications of the wave func-
Q
tion with respect to a purely factorised form ϕ(xi ) – see, e.g., (6.4). Hence, correlations
Q
at the level of marginals are modifications of purely factorised kernels ϕ(xi )ϕ(yi ).
When ΨN is thought as a point in an abstract HN,sym , correlations consist in writing
ΨN as some non-trivial expansion like (3.22) in a symmetric orthonormal basis – con-
structed as in theorem 3.1.3 – instead of simply as ϕ⊗N . At the level of marginals one
then detects them in the form of linear combinations of distinct tensor power projections
|ϕihϕ|⊗k , |ξihξ|⊗k , . . . , of mixed projection terms |ϕihϕ| ⊗ |ξihξ| ⊗ · · · , and of mixed
non-projection terms |ϕihξ| ⊗ · · · . Such a mechanism, in the particular case of N -body
pure states with definite quantum labels, goes as follows.

Theorem 6.4.1. Let γN = |ΨN ihΨN | be the N -body (N > 2) pure state with definite
quantum labels
ΨN = |ϕi1 , n1 ; ϕi2 , n2 ; . . . ; ϕir , nr isym . (6.60)

Then the corresponding two-particle reduced density matrix is

r
X
(2) nk nh − δk,h
γN = |ϕik ihϕik | ⊗ |ϕih ihϕih | (6.61)
N N −1
k,h=1

(δk,h being the Kronecker delta).

Proof. As in the proof of theorem 3.3.1, when constructing γN = |ΨN ihΨN | one gets
diagonal terms (3.30) and off-diagonal terms (3.31) and, by the same argument, the
latter do not contribute when Tr[N −2] is taken (the partial trace over the last N − 2
124 Chapter 6. Interparticle correlations in the condensate

positions). Instead, partial trace of each diagonal term gives


· ¸
n1 ! n2 ! · · · nr !
Tr[N −2] |ϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN ihϕj1 ⊗ ϕj2 ⊗ · · · ⊗ ϕjN |
N!
N Z
Y
n1 ! n2 ! · · · nr !
= |ϕj1 ihϕj1 | ⊗ |ϕj2 ihϕj2 | |ϕjk (xk )|2 dxk (6.62)
N! Ω
k=3
n1 ! n2 ! · · · nr !
= |ϕj1 ihϕj1 | ⊗ |ϕj2 ihϕj2 | .
N!
Among all terms of the linear combination (3.20), there is a fraction nk /N having ϕik
in first position; among these ones (the “tail” of which has N − 1 positions), there is a
fraction nh /(N −1) having ϕih in second position, if h 6= k, and a fraction (nk −1)/(N −1)
having ϕik also in second position, whence (6.61).
(2)
Thus, γN accounts not only for tensor power projections |ϕik ihϕik |⊗2 as for a totally
factorised state ΨN = ϕ⊗N , but also for mixed projection terms |ϕik ihϕik | ⊗ |ϕih ihϕih |,
k 6= h, which are a signature of correlations among particles. Notice that all the
ni1 , . . . , nir are positive integers, so terms like |ϕik ihϕik |⊗2 enter (3.29) only when nk > 2.
If ΨN has definite quantum labels, only distinct tensor power projections and mixed pro-
jection terms appear in its marginals, without mixed non-projection terms. If ΨN is in
turn a true linear combination, like (6.4), then also non-projection terms may appear in
the marginals.
For example, if

Ψ7 = |ϕ, 4; ξ, 2; η, 1isym
r r r
30 20 10
= (ϕ ⊗ ϕ ⊗ · · · ) + (ϕ ⊗ ξ ⊗ · · · ) + (ϕ ⊗ η ⊗ · · · )
105 105 105
r r r
20 5 5
+ (ξ ⊗ ϕ ⊗ · · · ) + (ξ ⊗ ξ ⊗ · · · ) + (ξ ⊗ η ⊗ · · · )
105 105 105
r r
10 5
+ (η ⊗ ϕ ⊗ · · · ) + (η ⊗ ξ ⊗ · · · )
105 105
then one can easily find
(2) 2 1
γ 7 = |ϕihϕ|⊗2 + |ξihξ|⊗2
7 21
4 2 4
+ |ϕihϕ| ⊗ |ξihξ| + |ϕihϕ| ⊗ |ηihη| + |ξihξ| ⊗ |ϕihϕ|
21 21 21
1 2 1
+ |ξihξ| ⊗ |ηihη| + |ηihη| ⊗ |ϕihϕ| + |ηihη| ⊗ |ξihξ|
21 21 21
and
(1) 4 2 1
γ 7 = |ϕihϕ| + |ξihξ| + |ηihη| .
7 7 7
On the other hand, as in the example before definition 3.3.2, if
1 ¡ ¢
Ψ8 = √ |ϕ, 6; ξ, 2isym + |ϕ, 5; ξ, 2; η, 1isym + 2 |ϕ, 4; η, 4isym
6
then
(1) 27 4 17 1 1
γ8 = |ϕihϕ| + |ξihξ| + |ηihη| + √ |ϕihη| + √ |ηihϕ| .
48 48 48 8 6 8 6
6.4. Correlations at the level of reduced density matrices 125

In the asymptotic analysis (N → ∞) and in presence of complete condensation,


(1) (1)
correlations disappear in the one-body density matrix: γN → γ∞ = |ϕihϕ|. (As we
(k)
shall see in Sec. 7.1, the same necessarily happens in any k-body density matrix: γN →
(k)
γ∞ = |ϕihϕ|⊗k ; by theorem 6.4.1, this is already clear in the special case of pure states
with definite quantum labels.) Further, their contribution to k-body marginals before
the limit may be essentially negligible
¡ for ¢many purposes: for example – see (6.3) – the
one-body marginal relative to ϕ⊗(N −1) ⊗ξ sym differs in trace norm from its asymptotics
|ϕihϕ| by an error O(N −1 ).
Yet, an accurate knowledge of correlations at the level of reduced density matrices
before the limit turns out to be crucial in many aspects of the rigorous analysis of B.E.C.
As an example of that, let us give here details of how an appropriate ansatz on the
correlation structure in the marginals determines that formal limit in the finite BBGKY
hierarchy which leads to the infinite hierarchy with the correct coupling (namely, the
two-body process scattering length, as already mentioned when sketching the scheme of
the “methods of the hierarchies” for the derivation of the GP equation, Sec. 5.4).
Let us recall that the first equation of the finite BBGKY hierarchy for the usual
model of N undistinguishable bosons expanding under the pairwise repulsion VN (x) =
N 2 V (N x):

(1) ~2 (1)
i~ ∂t γN,t (x1 , y1 ) = (−∆x1 + ∆y1 ) γN,t (x1 , y1 ) +
Z 2m (6.63)
£ (2)
+ (N − 1) VN (x1 − z) − VN (y1 − z)] γN,t (x1 , z, y1 , z) dz .
R3

(2) (1) (1)


The “no-correlation ansatz” γN,t = γN,t ⊗ γN,t , i.e.,

(2) (1) (1)


γN,t (x1 , x2 , y1 , y2 ) = γN,t (x1 , y1 ) γN,t (x2 , y2 ) , (6.64)

when plugged into (6.63), gives in the formal limit N → ∞ the following first equation
of the infinite BBGKY hierarchy:

(1) ~2 (1) ¡ ¢ (1)


i~ ∂t γ∞,t (x1 , y1 ) = (−∆x1 +∆y1 ) γ∞,t (x1 , y1 )+ Qt (x1 )−Qt (y1 ) γ∞,t (x1 , y1 ) . (6.65)
2m
In (6.65) one has set
Z
(1) (1)
Qt (x) := lim N VN (x − y) γN,t (y, y) = b γ∞,t (x, x) , (6.66)
N →∞ R3

(1)
last equality following if x 7→Rγ∞,t (x, x) is continuous. As usual, a is the s-wave scattering
1 1
length of V and 8πµ b = 8πµ R3 V (x) dx is its first Born approximation. This way, (6.65)
(1)
turns out to be solved by γ∞,t (x, y) = ϕt (x)ϕt (y) where ϕt (x) is the solution of

i~ ∂t ϕt (x) = −µ ∆x ϕt (x) + b |ϕt (x)|2 ϕt (x) . (6.67)

To recover the expected coupling in front of the cubic nonlinearity of (6.67), one
instead makes the ansatz
(2) (1) (1)
γN,t (x1 , x2 , y1 , y2 ) = fN (x1 − x2 )fN (y1 − y2 ) γN,t (x1 , y1 ) γN,t (x2 , y2 ) (6.68)
126 Chapter 6. Interparticle correlations in the condensate

modifying the pure product form (6.64) with correlations expressed in terms of the zero
energy scattering solution fN defined in (6.10). This, together with (6.14) and the
(1)
assumption that x 7→ γ∞,t (x, x) is smooth on scale N −1 , in the formal limit N → ∞
gives
Z
£ (2)
lim N VN (x1 − z) − VN (y1 − z)] γN,t (x1 , z, y1 , z) dz
N →∞ R3 (6.69)
¡ (1) (1) ¢ (1)
= 8πµa γ∞,t (x1 , x1 ) − γ∞,t (y1 , y1 ) γ∞,t (x1 , y1 )

for |x1 − y1 | À N −1 , thus recovering the GP equation with the correct dependence on
the scattering length.

6.5 Trial function for the evolution of a condensate


Here the setting changes to that of Sec. 5.3. From the time-dependent analysis of Erdős,
Schlein and Yau, a reasonable trial function emerges for the evolution of the originally
trapped condensate, expanding under some pairwise repulsion. Let us denote it by Ψesy N,t ,
lsy
to distinguish it from ΨN considered by Lieb et al. for the static case. In this section
we wish to present the motivation for introducing such a Ψesy N,t and in which sense it
can/cannot be considered close to the true evolution ΨN,t .
Actually, while Lieb et al. explicitly introduce their trial function, it has to be re-
marked that throughout the derivation of the ESY theorem 5.3.1 one does not deal with
any trial function at all, all the proof being based with a careful treatment of the true
marginals extracted from the unknown ΨN,t . Yet, an underlying Ψesy N,t is clearly present.
What motivates the choice of the trial function in this setting is the following key
property.

Theorem 6.5.1 (H 2 -energy estimate, [44]). Assume that H is the Hamiltonian of


HN N
the model considered in theorem 5.3.1. (In particular, α is the small enough parameter
defined in (5.18) to measure the strength of the interaction.) Let fN be the zero-energy
scattering solution defined in (6.10). Then, there exists a universal constant c > 0 such
that, for every fixed indices i, j ∈ {1, . . . , N } with i 6= j and ∀ ΨN ∈ L2sym (R3N ),
Z ¯ ¯2
¡ 2
¢ 2 ¯ ΨN ¯
ΨN , HN ΨN > (1 − c α)N (N − 1) fN (xi − xj )¯∇xi ∇xj ¯ . (6.70)
R3N fN (xi − xj )

Such an estimate involves a many-body function constructed by “extracting” from


ΨN the correlation fN (xi − xj ) at the short scale ∼ N −1 . It says that, if ΨN /fN (xi − xj )
is singular when |xi − xj | → 0, then ∇xi ∇xj (ΨN /fN (xi − xj )) cannot be L2 -integrable.
Now, recall that in ESY analysis, one assumes that at the initial time and, consequently,
at all later times the true many-body wave function satisfies the energy bound
¡ 2
¢
ΨN,t , HN ΨN,t 6 const · N 2 . (6.71)

Then the short distance behaviour of ΨN,t is given by fN (xi − xj ) when xi ∼ xj and this
holds at any time: there is a persistence of such a short scale structure during the time
evolution.
6.5. Trial function for the evolution of a condensate 127

There is actually more. Indeed ESY theorem accounts, among others, for purely
factorised initial data ϕ⊗N , where any correlation is absent. At later times, the many-
body state e−iHN t/~ ϕ⊗N must contain the above short scale structure. Thus, the many-
body dynamics builds up and preserves short-scale correlations emerging in the two-body
process (see [43, 44]).
On this basis one introduces
N
Y
WN (x1 , . . . , xN ) ϕt (xi )
i=1
Ψesy
N,t (x1 , . . . , xN ) := (6.72)
° YN °
° °
° WN (x1 , . . . , xN ) ϕt (xi ) °
i=1

where interparticle correlations are encoded in the Jastrow-like factor


Y
WN (x1 , . . . , xN ) := fN (xi − xj ) (6.73)
16i<j6N

and fN is the zero-energy scattering solution for the interaction VN . Here ϕt is the
evolution of a given ϕ under the time-dependent GP equation, where ϕ is assumed to be
the one-body condensate wave function at time t = 0. Instead, WN does not depend on
time: such a correlation structure is preserved during time evolution.
Currently, an explicit control that Ψesy
N,t ≈ ΨN,t in norm, or in energy form, is lacking.
Instead, Ψesy
N,t is believed to be close to the true ΨN,t in a much more indirect sense:
¬ Ψesy
N,t has the same short scale structure that ΨN,t must have, under the hypothesis
of boundedness of the square of the energy per particle;
(k),esy
­ the corresponding marginals, {γN,t }N k=1 , have the right correlations which enable
one to close the infinite hierarchy in the formal limit N → ∞ in such a way to
recover the expected dependence on the scattering length;
(k),esy
® estimates hold for the marginals such that each sequence {γN,t }N >k (at any
fixed k) is weakly-∗ compact in the trace class and any weak-∗ limit turns out to
be the unique solution of the infinite GP hierarchy.
Point ¬ is the consequence of the HN2 -energy estimate, as already noticed. Point ­

corresponds to the discussion concluding the previous section. Indeed one has
(2)
γN,t (x1 , x2 , y1 , y2 ) = fN (x1 − x2 ) fN (y1 − y2 ) ϕt (x1 ) ϕt (x2 ) ϕt (y1 ) ϕt (y2 )·
(2)
(6.74)
· AN,2 (x1 , x2 , y1 , y2 )
with
Z N
Y
(2)
AN,2 (x1 , x2 , y1 , y2 ) := fN (x1 − zj )fN (x2 − zj )fN (y1 − zj )fN (y2 − zj )·
j=3
R3(N −2) (6.75)
Y
2 2 2
· fN (zr − zs ) |ϕt (z3 )| · · · |ϕt (zN )| dz3 · · · dzN .
36r<s6N

(2)
Therefore, since AN,2 ≈ 1, the two-body trial marginal (6.74) has essentially the short
scale structure as in (6.68), which in turn leads to the correct GP hierarchy in the formal
limit N → ∞.
128 Chapter 6. Interparticle correlations in the condensate

Let us comment on point ®. In ESY theorem, one needs two fundamental bounds for
completing the “convergence+uniqueness” task of the method of the hierarchies. First,
the technique to prove the trace class weak-∗ compactness for each sequence of true k-
(k)
marginals {γN,t }N >k , is an Ascoli-Arzelà argument, for which the equicontinuity of each
(k)
t 7→ γN,t is needed, uniformly in N . This, in turn, is proved to be guaranteed by
£ (k) ¤
Tr (1 − ∆xi )(1 − ∆xj ) γN,t;(i,j) 6 const (6.76)

uniformly in t and eventually in N , and ∀ i, j ∈ {1, . . . , k} with i < j, where


(2)
(k) γN,t (x1 , . . . , xk , y1 , . . . , yk )
γN,t;(i,j) (x1 , . . . , xk , y1 , . . . , yk ) := . (6.77)
fN (xi − xj )fN (yi − yj )

The crucial bound (6.76) follows from the HN 2 -energy estimate above. This is a non-
(k)
trivial technical task: indeed, for finite N , marginals γN,t ’s do not satisfy estimates such
as (6.76), at least not uniformly in N , because they contain a short scale structure: only
after removal of this structure, as in (6.77), the bound holds. It is also remarkable that
(k) (k)
the same bound for γN,t , instead of γN,t;(i,j) , is recovered in the limit N → ∞, where the
correlation structure disappears: indeed it is another key achievement along the proof of
ESY theorem that £ (k) ¤
Tr (1 − ∆x1 ) · · · (1 − ∆xk ) γ∞,t 6 C k (6.78)
∀ k > 1, ∀ t ∈ [0, T ], where C and T are some positive constants. Furthermore, it has
to be emphasized here that uniqueness of the solution of the infinite GP hierarchy holds
(k)
provided that the solution {γ t }k>1 satisfies (6.78). Hence, summarizing, the “a priori
bounds” (6.76) and (6.78) are the essential ingredients for the existence of the limiting
marginals and for their uniqueness, respectively.
The content of point ® above is thus clear. For example, in the concrete case k = 2,
since essentially one has
(2),esy
γN,t ≈ fN (x1 − x2 ) fN (y1 − y2 ) ϕt (x1 ) ϕt (x2 ) ϕt (y1 ) ϕt (y2 ) , (6.79)

then
(2),esy
γN,t;(1,2) ≈ ϕt (x1 ) ϕt (x2 ) ϕt (y1 ) ϕt (y2 ) , (6.80)
whence
£ (2),esy ¤
Tr (1 − ∆x1 )(1 − ∆x2 ) γN,t;(1,2) ≈
¡ ¢
≈ ϕt (x1 )ϕt (x2 ) , (1 − ∆x1 )(1 − ∆x2 ) ϕt (x1 )ϕt (x2 )
³Z ´2 (6.81)
= ϕt (x) (1 − ∆x ) ϕt (x) dx
R3
= kϕt k4H 1

(1),esy
which is uniformly bounded in time. Then, bound (6.76) is satisfied and {γN,t }N >1
admits trace class weak-∗ limits, as for the case of the true marginals. The same holds
(k),esy
at higher k’s and the limit is recognised to be the sequence {γ∞,t = |ϕt ihϕt |⊗k }k>1 .
(k),esy
Notice that each γ∞,t satisfies the bound (6.78) with C = kϕt k2H 1 .
6.6 . Correlations and energy discrepancies in the time evolution 129

6.6 Correlations and energy discrepancies


in the evolution of the condensate
Despite ESY trial function Ψesy
N,t mimes the true many-body state ΨN,t in the sense that
it contains the correct correlation structure at the scale N −1 , it is a currently major
open problem to understand how good the approximation is in norm and in energy form.
Work in progress [82] is dealing with this issue and this section gives a brief account of
that.
Concerning the norm distance kΨN,t − Ψesy N,t k, estimates as in theorem 6.2.1 fail be-
cause ΨN,t contains a far larger number of correlation terms than Ψlsy
esy
N,t (respectively,
2
∼ N versus ∼ N ). A more refined estimate by means of the properties of fN seems
promising.
Concerning the difference (ΨN,t , HN ΨN,t ) − (Ψesy esy
N,t , HN ΨN,t ), it is clear that the first
term is conserved in time, because ΨN,t is the evolution of the initial datum ΨN,0 at
t = 0 under the dynamics generated by the Hamiltonian HN of the model considered in
theorem 5.3.1. On the other hand the second term fluctuates, since ESY trial is not an
eigenstate of HN . Although an exact asymptotic computation of (Ψesy esy
N,t , HN ΨN,t ) is still
missing (while one can bound this expectation value with some bound proportional to
N , through arguments similar to those of lemma B.1 in [44]), it is believed on the basis
of the corresponding one-body picture that
³ HN esy ´
Ψesy
N,t , Ψ ∼ Ea [ϕt ] = Ea [ϕ] (6.82)
N N,t
where, as usual, ϕt solves the time-dependent GP equation
δEa [u]
i~∂t ut = −µ∆x ut + 8πµa |ut |2 ut = (6.83)
δu
with initial datum ϕ and
Z
¡ ¢
Ea [u] := µ|∇u(x)|2 + 4πµ a|u(x)|4 dx (6.84)
R3

is the corresponding GP energy functional without trapping (compare with (5.14) and
(5.15)).
Thus, one understands that Ψesy
N,t cannot account correctly for the expectation value
of HN in the state ΨN,t , because of the possibility of choosing distinct initial data ΨN,0
with distinct energies. For instance, starting from an initial datum ΨN,0 = ϕ⊗N (which
is admissible in ESY theorem), the expectation value of the energy in the state ΨN,t =
e−iHN t/~ ϕ⊗N is
³ HN ´ ³ HN ⊗N ´ N →∞
ΨN,t , ΨN,t = ϕ⊗N , ϕ −−−−→ Eb/(8πµ) [ϕ] (6.85)
N N
due to the same computation as in the proof of theorem 6.2.2. Hence, for this choice of
the initial datum, ESY trial function underestimates hHN i.
Such an apparent discrepancy is reasonable because ground-state-like correlations
lower the expectation value of the energy, though it does not rely on those firm grounds
that the explicit computation of (Ψesy esy
N,t , HN ΨN,t ) would give.
In fact, the net effect of the presence of a Jastrow-like correlation factor WN in
ΨN,t , when compared to the initial datum ϕ⊗N , is to lower the expectation value of
esy
130 Chapter 6. Interparticle correlations in the condensate

the potential energy, because correlations reduce the probability that two particles stay
close together and the potential is repulsive, and to increase the expectation value of the
kinetic energy, because Laplacians act to fN -factors which are more and more singular in
the origin, as N → ∞. Yet, a detailed balance between these two opposite discrepancies
is not clear and it is of interest to have a full control of the energy loss
³ HN ´ ³ HN esy ´
∆EN,t := ΨN,t , ΨN,t − Ψesy
N,t , Ψ (6.86)
N N N,t
and to understand whether ∆EN,t > 0.
If this turned out to be the case, one should conclude that Ψesy N,t does not capture
the leading term in the energy of ΨN,t . Erdős, Schlein, and Yau’s tentative explanation
(see [43, 44]) is a picture where some energy exists on intermediate length scales of order
N −β , 0 < β < 1. This excess energy apparently does not participate in the evolution of
marginals on the macroscopic length scale.
The suspect is that one could recover the correct energy asymptotics by introducing
a modified trial function, say Ψ e esy , with higher energy than Ψesy : in Ψ
e esy one makes the
N,t N,t N,t
replacement
fN (xi − xj ) 7−→ fN (xi − xj ) + εN gN (xi − xj ) (6.87)
with a bounded gN (x) oscillating around zero and becoming more and more singular for
N →∞
large N , thus increasing the kinetic part of the Hamiltonian, while εN −−−−→ 0 in such
a way to guarantee that the modification of the potential part is negligible.
Summarizing, in this scenario for the evolution of a Bose gas initially prepared in
some state ϕ⊗N one may consider the following many-body wave functions:

3 the true ΨN,t = e−iHN t/~ ϕ⊗N where the expectation value of the energy per particle
is asymptotically Eb/(8πµ) [ϕ]; ΨN,t has the characteristic short scale correlations
imposed by the HN 2 -energy estimate and possibly certain incoherent excitations

living at an intermediate scale;

3 ϕ⊗N
t , in which hHN i is asymptotically the same as in ΨN,t , but does not solve the
N -body Schrödinger equation (in fact, correlations are absent);

3 Ψesy
N,t , a trial function which reproduces the correct correlations at the scale N
−1 and

matches with ΨN,t at the level of marginals, but has reasonably a lower expectation
value of the energy that the true ΨN,t ;

3 some modified Ψe esy to be constructed with criteria (see above) guaranteeing its
N,t
closeness to ΨN,t in energy form and in norm.
Chapter 7

Equivalent characterizations
of B.E.C.

7.1 B.E.C. of infinite systems at any k-particle level


Technical and physical reasons motivate to control condensation not only at the one-body
marginal level, but also through any k-particle reduced density matrix. In this section,
following Ref. [83], we give account of the substantial equivalence of these controls.
From the physical point of view, controlling one-particle marginals only does not give
full access to interparticle correlations, the knowledge of which turns out to be crucial in
determining, e.g., the expectation value of the energy in the state under consideration
(k)
(see Chap. 6). Moreover, the knowledge of the sequence {γN }N at level k allows one to
control the limit of the expectation values of k-particle observables.
From the mathematical point of view, when condensation is needed as an hypothesis
(e.g., assuming condensation at the initial time to investigate the dynamics at further
times), it is often necessary to assume it in the apparently stronger form of an asymptotic
(k)
factorisation of γN into some |ϕihϕ|⊗k , for any fixed positive integer k. The paradigmatic
case is the already mentioned ESY theorem 5.3.1.
The behaviour of the two-particle (or k-particle) density matrix has then to enter the
mathematical formalisation of B.E.C.
First of all, in this section we point out that to define and to control complete
condensation in the context of asymptotic B.E.C. the one-body level is enough.

Theorem 7.1.1. Let {ΨN }N be a sequence of bosonic pure states with the property of
asymptotic 100% B.E.C. and let ϕ be the corresponding condensate wave function (see
definition 3.4.1). That is,
(1)
lim γN = |ϕihϕ| (7.1)
N →∞
in the usual weakly-∗ or trace-norm sense (see theorem 3.4.2). Then
(k)
lim γN = |ϕihϕ|⊗k ∀k > 1 (7.2)
N →∞

in the same sense.

This fact was apparently first noticed and proved by Lieb and Seiringer – see the
remark after theorem 1 in [74] – within Bogolubov’s formalism of second quantization
and it applies both to pure and mixed states. This is actually a point where, to be

131
132 Chapter 7. Equivalent characterizations of B.E.C.

consistent with our choice of presenting the subject in an elementary way, a longer
proof is needed than that with creation and annihilation operators. (In that case, the
commutation relations give the upper bound and a convexity argument give a lower
bound leading to the limit (7.2).) For completeness, let us prove theorem 7.1.1 in the
present formalism. Generalization to mixed states is the content of the following theorem
7.1.2.

Proof of theorem 7.1.1. We give details of how (7.1) implies (7.2) for k = 2. The general
case follows along the same line.
Complete ϕ to an orthonormal basis {ϕh }h>1 of the one-body Hilbert space H, with
(N −1)
ϕ1 = ϕ. Let HN be the N -body Hilbert space H⊗N . For each N , let {Φj }j be an
(N −2)
orthonormal basis of HN −1 and let {Ξ` }` be an orthonormal basis of HN −2 .
By assumption (see theorem 3.4.4),
(1) N →∞
(ϕ, γN ϕ)H −−−−→ 1 , (7.3)
and, by characterization (2.48) of partial trace, and since γN = |ΨN ihΨN | (pure state),
X ¯¯ ¡ (N −1) ¢ ¯¯2 X ¡ (N −1) (N −1) ¢
¯ ϕ ⊗ Φj , ΨN H ¯ =
N
ϕ ⊗ Φj , γN ϕ ⊗ Φ j HN
j j (7.4)
(1) N →∞
= (ϕ, γN ϕ)H −−−−→ 1 ,
whence X ¯¯ ¡ (N −1) ¢ ¯¯2 N →∞
¯ ϕh ⊗ Φj , ΨN H ¯ −−−−→ 0 .
N
(7.5)
j
h6=1
What one needs to prove (see theorem 3.4.4) is
(2) N →∞
(ϕ⊗2 , γN ϕ⊗2 )H2 −−−−→ 1 , (7.6)
where, by (2.48),
(2)
X¡ (N −2) (N −2) ¢
(ϕ⊗2 , γN ϕ⊗2 )H2 = ϕ⊗2 ⊗ Ξ` , γN ϕ⊗2 ⊗ Ξ` H N
`
X ¯¯ ¡ (N −2) ¢ ¯¯2 (7.7)
⊗2
= ¯ ϕ ⊗ Ξ` , ΨN HN ¯ .
`
(N −2)
Since {ϕh ⊗ Ξ` }h,`
is an orthonormal basis of HN −1 , by (7.4)
X ¯¯ ¡ (N −2) ¢ ¯2
¯ (1)
X ¯¯ ¡ (N −2) ¢ ¯¯2
¯ ϕ⊗2 ⊗ Ξ` , ΨN H
N
¯ = (ϕ, γN ϕ)H − ¯ ϕ ⊗ ϕh ⊗ Ξ` , ΨN H ¯ (7.8)
N
` `
h6=1

and, since ΨN is symmetric under permutation of variables,


X ¯¯ ¡ (N −2) ¢ ¯¯2 X ¯¯ ¡ (N −2) ¢ ¯¯2
¯ ϕ ⊗ ϕh ⊗ Ξ` , Ψ N HN ¯ = ¯ hϕ ⊗ ϕ ⊗ Ξ ` , ΨN HN ¯
` `
h6=1 h6=1
X ¯¯ ¡ (N −2) ¢ ¯¯2
6 ¯ ϕh ⊗ ϕr ⊗ Ξ` , ΨN H ¯
`,r
N (7.9)
h6=1
X ¯¯ ¡ (N −1) ¢ ¯¯2 N →∞
= ¯ hϕ ⊗ Φj , ΨN HN ¯ −
−−−→ 0 .
j
h6=1
7.1. B.E.C. of infinite systems at any k-particle level 133

Then, plugging (7.3) and (7.9) into (7.8) as N → ∞, (7.6) follows.

Theorem 7.1.2. Let {γN }N be a sequence of bosonic (pure or mixed ) states with the
property of asymptotic 100% B.E.C. and let ϕ be the corresponding condensate wave
function,
(1)
lim γ = |ϕihϕ| . (7.10)
N →∞ N

Then
(k)
lim γN = |ϕihϕ|⊗k ∀k > 1. (7.11)
N →∞

(Limits in any of the usual senses.)

Proof. (Same notation of last theorem; details for k = 2 only.) Decompose each γN in
its diagonal form
X (i) (i)
X
γN = λi |ΨN ihΨN | , λi > 0 , λi = 1 , (7.12)
i i

(i)
where {ΨN }i is an orthonormal system of HN . By assumption (see theorem 3.4.4), and
by characterization (2.48) of partial trace,

N →∞ (1)
X¡ (N −1) (N −1) ¢
1 ←−−−− (ϕ, γN ϕ)H = ϕ ⊗ Φj , γN ϕ ⊗ Φ j H N
j
X X ¯¯ ¡ ¯2 (7.13)
(N −1) (i) ¢ ¯
= λi ¯ ϕ ⊗ Φj , ΨN H ¯ ,
N
i j

whence
X X ¯¯ ¡ (N −1) (i) ¢
¯2
¯
λi ¯ ϕh ⊗ Φj , ΨN H ¯ =
N
i j
h6=1 (7.14)
X ³ X ¯¯ ¡ (N −1) (i) ¢
¯2 ´
¯ N →∞
= λi 1 − ¯ ϕ ⊗ Φj , ΨN HN ¯ −−−−→ 0 .
i j

P
(since i λi = 1). On the other hand,

(2)
X¡ (N −2) (N −2) ¢
(ϕ⊗2 , γN ϕ⊗2 )H2 = ϕ⊗2 ⊗ Ξ` , γN ϕ⊗2 ⊗ Ξ` H N
`
X X ¯¯ ¡ (N −2) (i) ¢
¯2
¯
⊗2
= λi ¯ ϕ ⊗ Ξ ` , ΨN HN ¯
i `
X h¡ £ (i) (i) ¤ ¢
= λi ϕ, Tr[N −1] |ΨN ihΨN ϕ −
H
i (7.15)
X ¯¯ ¡ (N −2) (i) ¢
¯2 i
¯
− ¯ ϕ ⊗ ϕ h ⊗ Ξ` , ΨN HN ¯
`
h6=1
(1)
X X ¯¯ ¡ (N −2) (i) ¢
¯2
¯
= (ϕ, γN ϕ)H − λi ¯ ϕ ⊗ ϕh ⊗ Ξ` , ΨN H ¯ .
N
i `
h6=1
134 Chapter 7. Equivalent characterizations of B.E.C.

(i) (N −2)
Since each ΨN is symmetric under permutation of variables and {ϕr ⊗ Ξ` }r,` is an
orthonormal basis of HN −1 , by (7.14)
X X ¯¯ ¡ (N −2) (i) ¢
¯2
¯
λi ¯ ϕ ⊗ ϕh ⊗ Ξ` , ΨN H ¯ =
N
i `
h6=1
X X ¯¯ ¡ (N −2) (i) ¢
¯2
¯
= λi ¯ ϕh ⊗ ϕ ⊗ Ξ` , ΨN H ¯
N
i `
h6=1
X X ¯¯ ¡ (N −2) (i) ¢
¯2
¯
(7.16)
6 λi ¯ ϕh ⊗ ϕr ⊗ Ξ` , ΨN H ¯
N
i `,r
h6=1
X X ¯¯ ¡ (N −1) (i) ¢
¯2
¯ N →∞
= λi ¯ ϕ h ⊗ Φj , ΨN H ¯ −−−−→ 0 .
N
i j
h6=1

Thus, plugging (7.13) and (7.16) into the r.h.s. of (7.15), one has

(2)
lim (ϕ⊗2 , γN ϕ⊗2 )H2 = 1 (7.17)
N →∞

which, by theorem 3.4.4, is the same as the trace-norm limit (7.2).

We can thus conclude that the final content of theorems 3.4.4 and 7.1.2 is: 100%
B.E.C. into the one-body state ϕ, as an asymptotic property of a sequence {γN }N of
N -body bosonic states in the sense of definition 3.4.1, is the factorisation of a whatever
(k)
k-particle reduced density matrix γN into the k-th tensor power of the projection onto
(k)
ϕ, whatever the topology is for the convergence γN → |ϕihϕ|⊗k , be it weak-∗ trace class,
Hilbert-Schmidt in norm, or trace class in norm.

7.2 Asymptotic factorisation of the N -body wave function


There exist other types of complete condensation assumptions in the literature than
the asymptotic factorisation (7.2) of the k-th marginals. These are usually needed as
apparently stronger technical hypotheses on the initial state when one wants to analyse
its time evolution and are in the form of conditions on the N -body state, instead of on
the marginals. Following Ref. [83], in this section we deal with the two of them that
mostly entered the study of time-stability of condensates and the rigorous derivation of
the time-dependent Gross-Pitaevskiı̆ equation (Sec. 5.3) and we prove their equivalence
with the standard definition of asymptotic 100% B.E.C.
This provides a homogenization of the hypotheses appearing in the statements of a
number of recent theorems. Yet this does not cancel the merit of having introduced
such alternative assumptions of condensation which focus on the specific properties of
the initial state that are crucial to prove the desired statements.

Definition 7.2.1 (partial asymptotic factorisation). Let H be a separable Hilbert


space, set HN := H⊗N and let HN,sym be its bosonic sector. Let {ΨN }N be a sequence
of N -body bosonic pure states (kΨN k = 1 according to our definition), each ΨN ∈
HN,sym ⊂ HN . Let ϕ ∈ H with kϕk = 1. The sequence {ΨN }N is said to have the
7.2. Asymptotic factorisation of the N -body wave function 135

partial asymptotic factorisation (P.A.F.) property in the state ϕ iff

∀N > 1, ∀ k = 1, . . . , N
° °
∃ Ξ(N −k) ∈ HN −k , ° Ξ(N −k) °H =1 : (7.18)
° ° N −k
lim ° ΨN − ϕ ⊗ Ξ
⊗k (N −k) ° = 0.
HN
N →∞

Notice that the P.A.F. is a statement on the collection {ΨN }N compared with a
countable infinity of collections {ϕ⊗k ⊗ Ξ(N −k) }N with k = 1, 2, 3, . . . : each ΨN can
be approximated in norm with a ϕ⊗k ⊗ Ξ(N −k) for any k, the larger N , the better the
approximation. Unlike ΨN , the ϕ⊗k ⊗ Ξ(N −k) ’s are not necessarily in HN,sym : fixed any
positive integer k and provided that N is large enough, one can always find a vector close
to each ΨN which is completely factorised in the first k variables (actually it is a k-th
tensor power), tensor a tail in the remaining N − k. Choosing a different k, the same
holds, possibly in a different regime of large N ’s, that is to say that limit (7.18) is not
uniform in k.
From (7.18), it follows that ΨN is asymptotically close in norm also to the sym-
metrized of ϕ⊗k ⊗ Ξ(N −k) – see (3.18): indeed by (2.106)
° °
° ¡ ⊗k ¢ ° 2 ° ΨN − ϕ⊗k ⊗ Ξ(N −k) °
° ΨN − ϕ ⊗ Ξ(N −k) ° HN
6 ° ° (7.19)
sym HN ° ⊗k
1 − ΨN − ϕ ⊗ Ξ (N −k) °
H N

eventually with N . The converse is false, in general. Actually, in the applications


it is crucial that ΨN is controlled in norm with the non permutationally symmetric
ϕ⊗k ⊗ Ξ(N −k) .
Condition (7.18) has been introduced in 2006 by Erdős, Schlein, and Yau [44], first
by recognising it to be the convenient assumption which ensures that regularizing the
initial wave function ΨN by a cut-off on its high-energy component leads to a regularized
e N which exhibits condensation γ (k)
Ψ eN → |ϕihϕ|⊗k and has N -uniformly bounded energy
moments (see proposition 8.1 in [44]). Explicit {ΨN }N with P.A.F. are shown in appendix
B of [44]. Soon after it has been recognised by the same authors that P.A.F. holds for
the ground state of the Hamiltonian (3.64) of a trapped Bose gas, but with some more
restrictive conditions on the pair interaction, and factorisation holds with respect to the
one-body state ϕGP (see proposition C.2 in [44]).
Explicitly (see lemma B.1 in [44]), the following bosonic states can be proved to have
P.A.F. in the one-body state ϕ:

N
Y
WN (x1 , . . . , xN ) ϕ(xi )
i=1
ΨN (x1 , . . . , xN ) := (7.20)
° YN °
° °
° WN (x1 , . . . , xN ) ϕ(xi ) °
i=1
Y
WN (x) := fN (xi − xj ) (7.21)
16i<j6N

where now fN is defined to be identically 1 outside the ball BN in R3 of radius rN such


that aN ¿ rN ¿ 1, while inside BN it is defined to be the ground state solution of the
136 Chapter 7. Equivalent characterizations of B.E.C.

Neumann problem
 ¡ ¢
 − 2µ∆ + VN (x) fN (x) = EN fN (x) x ∈ BN
fN (x) = 1 x ∈ ∂BN (7.22)

∂radial fN (x) = 0 x ∈ ∂BN .

Here VN is the two-body interaction VN (x) := N 2 V (N x), where the unscaled V has
scattering length a and is assumed to be non-negative, smooth, spherically symmetric
and compactly supported. Then aN = a/N is the scattering length of VN (with respect
to notation of theorem 3.5.2 for trapped Bose gases, here we set unrestrictively N0 = 1).
P.A.F. for states (7.20) is verified with respect to the tail
Y N
Y
fN (xi − xj ) ϕ(xi )
k6i<j6N i=k+1
Ξ(N −k) (xk+1 , . . . , xN ) = ° ° . (7.23)
° Y N
Y °
° °
° fN (xi − xj ) ϕ(xi ) °
° °
k6i<j6N i=k+1

By lemma A.2 in [41] (see also appendix B in [44]), fN has the following behaviour:
it is spherically symmetric, and there exist positive constants C1 , C2 , independent of N
and depending only on V , such that
aN
1 − C1 1BN (x) 6 fN (x) 6 1
|x| + aN (7.24)
aN
|∇fN (x)| 6 C2 1B (x)
(|x| + aN )2 N

where 1BN is the characteristic function of the ball BN . Thus the WN -factor (7.21)
accounts for correlations at the scale rN in the wave function (7.20), analogously to
the Dyson-like correlations (6.5) used in Lieb and Seiringer’s trial function (6.4) for the
ground state of a trapped Bose gas.
By comparing (7.20) and (7.23), one sees that Ξ(N −k) = ΨN −k so that P.A.F. (7.18)
for {ΨN }N reads simply kΨN − ϕ⊗k ⊗ ΨN −k k → 0 as N → ∞. This motivates to
introduce the following.
Definition 7.2.2. Let H be a separable Hilbert space, set HN := H⊗N , and let HN,sym
be its bosonic sector. Let {ΨN }N be a sequence of N -body bosonic pure states (kΨN k = 1
according to our definition), each ΨN ∈ HN,sym ⊂ HN . Let ϕ ∈ H with kϕk = 1. The
sequence {ΨN }N is said to have the self-tail (S.T.) property with respect to the
one-body state ϕ iff ° °
lim ° ΨN +1 − ϕ ⊗ ΨN ° = 0 . (7.25)
N →∞

Notice that, as for the P.A.F. property, the S.T. property is a statement on the
collection {ΨN }N compared with {ϕ ⊗ ΨN }N : adding a particle ϕ to the N -th element
of the sequence {ΨN }N to get the (N + 1)-body state ϕ ⊗ ΨN , one essentially obtains the
subsequent ΨN +1 of the sequence, the larger N , the smaller the error in norm. Again,
the comparison is with a non permutationally symmetric N -body state and the variables
ordering in {ϕ ⊗ ΨN }N is crucial for the applications: {ΨN ⊗ ϕ}N would not play the
same role. Notice, also, that it is just a matter of a finite number of triangle inequalities
to prove that
° ° ° °
lim ° ΨN +1 − ϕ ⊗ ΨN ° = 0 ⇒ lim ° ΨN − ϕ⊗k ⊗ ΨN −k ° = 0 ∀ k (7.26)
N →∞ N →∞
7.2. Asymptotic factorisation of the N -body wave function 137

(the converse being obvious), so that one can refer to such a property evoking indifferently
(7.25) or the r.h.s. of (7.26).
What we now prove is that all these are equivalent notion of condensation.

Theorem 7.2.3 (equivalent definitions of B.E.C. for pure states).

asymptotic partial asymptotic self-tail


⇔ ⇔
100% B.E.C. factorisation property

In detail: let H be a separable Hilbert space, set HN := H⊗N , and let HN,sym be its
bosonic sector. Let {ΨN }N be a sequence of N -body bosonic pure states (kΨN k = 1
according to our definition), each ΨN ∈ HN,sym ⊂ HN . Let ϕ ∈ H with kϕk = 1. Then
any of the conditions of theorem 3.4.4, (7.18), and (7.25) are all equivalent.

Proof of: S.T. property ⇒ P.A.F. property. This is just a finite iteration of triangle in-
equality to get (7.26).

Proof of: P.A.F. property ⇒ asymptotic 100% B.E.C.. Assume (7.18) holds: it suffices
(1)
to use it for k = 1. Let γN be its one-particle reduced density matrix of ΨN and let
C ∈ Com(H), a generic compact on the one-body Hilbert space H. The one-particle
reduced density matrix of ϕ ⊗ Ξ(N −1) is clearly |ϕihϕ|. Then ,by characterization (2.49)
of partial trace,
¯ £ ¤¯
¯ Tr C(γ (1) − |ϕihϕ|) ¯ =
N
¯ £ (1) ¤¯
= ¯ Tr CγN ] − Tr[ C|ϕihϕ| ¯
¯ £ ¤ £ ¤¯
= ¯ Tr (C ⊗ 1N −1 ) |ΨN ihΨN | − Tr (C ⊗ 1N −1 ) |ϕ ⊗ Ξ(N −1) ihϕ ⊗ Ξ(N −1) | ¯
¯¡ ¢ ¡ ¢ ¯ (7.27)
= ¯ ΨN , (C ⊗ 1N −1 )ΨN H − ϕ ⊗ Ξ(N −1) , (C ⊗ 1N −1 )ϕ ⊗ Ξ(N −1) H ¯
° N ° ¡ ° N
° ¢
6 k C ⊗ 1N −1 kL(HN ) · ° ΨN − ϕ ⊗ Ξ(N −1) °H · 2 + ° ΨN − ϕ ⊗ Ξ(N −1) °H
N N
° ° N →∞
°
6 3 kCkL(H) · ΨN − ϕ ⊗ Ξ (N −1) °
−−−−→ 0
H N

(first inequality following because in general

| (x, Ay) − (y, Ay) | 6 kAk · kx − yk · (2 + kx − yk)

for any bounded operator A on some Hilbert space and any norm-one vectors x, y), that
(1)
is, γN → |ϕihϕ| weakly-∗ in L1 (H). This is exactly one of the equivalent ways to define
100% asymptotic B.E.C.

Alternative proof of: P.A.F. property ⇒ asymptotic 100% B.E.C.. By (2.72) – see the-
orem 2.4.1,
° ¯ ®­ ¯° ° °
° γN − |ϕihϕ| ⊗ ¯ Ξ(N −1) Ξ(N −1) ¯ ° 1 6 2 ° ΨN − ϕ ⊗ Ξ(N −1) °H , (7.28)
L (H ) N N

and by (2.104) – see theorem 2.4.4,


° (1) ° ° ¯ ®­ ¯°
° γ − |ϕihϕ| ° 6 ° γN − |ϕihϕ| ⊗ ¯ Ξ(N −1) Ξ(N −1) ¯ °L1 (H (7.29)
N L1 (H) N)

(1)
so that γN → |ϕihϕ| in trace norm.
138 Chapter 7. Equivalent characterizations of B.E.C.

Proof of: asymptotic 100% B.E.C. ⇒ P.A.F. property. Pick a positive integer k. By as-
sumption (see theorem 3.4.4),
¡ ⊗k (k) ⊗k ¢
lim ϕ , γN ϕ H
= 1. (7.30)
N →∞ k

Complete ϕ⊗k to an orthonormal basis {fi }i of Hk in such a way that f1 = ϕ⊗k . Choose
(N −k)
in each HN −k an orthonormal basis {gj }j : hence, each ΨN ∈ HN ' Hk ⊗ HN −k can
be written as X (N ) (N −k)
ΨN = αi,j fi ⊗ gj (7.31)
i,j

(N )
for suitable complex coefficients αi,j . Then
XX (N ) (N ) ¯ (N −k) ®­ (N −k) ¯
γN := |ΨN ihΨN | = α1,j α1,j 0 |ϕ⊗k ihϕ⊗k | ⊗ ¯ gj gj 0 ¯
j j0
XX (N ) (N ) ¯ (N −k) ®­ (N −k) ¯
+ α1,j αi0,j 0 |ϕ⊗k ih fi0 | ⊗ ¯ gj gj 0 ¯
j i0 6=1
j0
XX (N ) (N ) ¯ (N −k) ®­ (N −k) ¯ (7.32)
+ αi,j α1,j 0 | fi ihϕ⊗k | ⊗ ¯ gj gj 0 ¯
i6=1 j 0
j
XX (N ) (N ) ¯ (N −k) ®­ (N −k) ¯
+ αi,j αi0,j 0 | fi ih fi0 | ⊗ ¯ gj gj 0 ¯.
i6=1 i0 6=1
j j0

Taking the partial trace with respect to HN −k , according to (2.52), and using the first
of (2.19), gives
(k)
X ¯ (N ) ¯2
γN = |ϕ⊗k ihϕ⊗k | · ¯α ¯
1,j
j
X (N ) (N )
+ α1,j 0 α i0,j 0 |ϕ⊗k ih fi0 |
i0 6=1
j0
X
+
(N ) (N )
αi,j α1,j | fi ihϕ⊗k | (7.33)
i6=1
j
X (N ) (N )
+ αi,j αi0,j | fi ih fi0 | .
i6=1
i0 6=1
j

As a consequence, by (7.30),
X ¯ (N ) ¯2 ¡ ¢ N →∞
¯α ¯ = ϕ⊗k , γ (k) ϕ⊗k −−−−→ 1 , (7.34)
1,j N H k
j

whence
° X (N ) ° X ¯ (N ) ¯2 X ¯ (N ) ¯2 N →∞
° ⊗k (N −k) °2 ¯α ¯ = 1 − ¯α ¯ −−−−→ 0 . (7.35)
Ψ
° N − α 1,j ϕ ⊗ gj ° = i,j 1,j
HN
j i6=1 j
j
7.2. Asymptotic factorisation of the N -body wave function 139

Finally set X (N ) (N −k)


Ξ(N −k) := α1,j gj (7.36)
j

so that (7.35) reads kΨN − ϕ⊗k ⊗ Ξ(N −k) k → 0 as N → ∞: by the arbitrariness of k this
is just the P.A.F. property (7.18).

Remark, incidentally, that (7.36) accounts for the non uniqueness of Ξ(N −k) in the
P.A.F. property.

Proof of: P.A.F. property ⇒ S.T. property. One has


° ° ° °
° ΨN +1 − ϕ ⊗ ΨN ° 6 ° ΨN +1 − ϕ⊗(k+1) ⊗ Ξ(N −k) °+
° °
+ ° ϕ⊗(k+1) ⊗ Ξ(N −k) − ϕ ⊗ ΨN ° (7.37)
° ° ° °
= ° ΨN +1 − ϕ⊗(k+1) ⊗ Ξ(N −k) ° + ° ΨN − ϕ⊗k ⊗ Ξ(N −k) °

the r.h.s. vanishes as N → ∞ by assumption of P.A.F., hence (7.25) follows.

The mixed state version of theorem 7.2.3 holds too, although it is not of relevance in
the current literature. We state it for completeness. The proof is analogous to the pure
states case.

Theorem 7.2.4 (equivalent definitions of B.E.C. for mixed states). Let H be


a separable Hilbert space and set HN := H⊗N . Let {γN }N be a sequence of N -body
bosonic mixed states, each γN acting on HN . Let ϕ ∈ H with kϕk = 1. Then any of the
following conditions are equivalent:

• 100% asymptotic B.E.C.


(1)
lim γ = |ϕihϕ| (7.38)
N →∞ N

• partial asymptotic factorisation

∀N > 1, ∀ k = 1, . . . , N

∃ ρN −k bosonic density matrix on HN −k : (7.39)


° °
lim ° γN − |ϕihϕ|⊗k ⊗ ρN −k °L1 (H =0
N →∞ N)

• self-tail condition
° °
lim ° γN +1 − |ϕihϕ| ⊗ γN °L1 (H = 0. (7.40)
N →∞ N)
140 Chapter 7. Equivalent characterizations of B.E.C.
Chapter 8

Concluding remarks

Condensation is an “ancient” and “recent” phenomenon: theoretically predicted 80 years


ago, experimentally observed from 1995 on, mathematically studied since the 1950s, until
the renewed flurry of interest in the rigorous analysis of the last decade. B.E.C. is a
signature of Quantum Mechanics: no explanation is possible in a classical picture. It is
a consequence of the quantum statistics of Bosons only. Recent B.E.C. realizations have
represented the apex of huge experimental efforts and the confirm of a long theoretical
research. On the other side, they have re-fired the need of a fully rigorous comprehension
of the phenomenology in terms of first principles.
To conclude this work, let us first collect the key points developed throughout and
after outline some perspectives.

The mathematics of B.E.C.

1. Limit N → ∞ ∞. Formalisation and symptotic analysis of condensation in many-


body dilute Bose gases in the limit N → ∞ relies on the fact that, in practice, the number
of particles is so large (N ∼ 100 ÷ 1011 ) to justify the assumption of infinite size and
condensation for interacting systems requires to solve a formidably difficult many-body
(1)
problem in order to access γN , while there are techniques to get its asymptotics (for
interacting systems the ground state already poses a challenging problem which is still
largely unsolved: B.E.C. itself has, so far, never been proved for many-body Hamiltonians
with genuine interactions, except for the special case of hard core bosons on a lattice at
half-filling).

2. Scaling. Taking N → ∞ is not just a thermodynamic limit. Instead, a more


careful treatment is needed: some scaling has to be precised. The true system S0 in
the state γN0 and with Hamiltonian HN0 is regarded as an enlarging sequence {SN }N
of N -body systems, each in the state γN and with Hamiltonian HN . As N = N0 , SN
is exactly S0 , i.e., HN |N =N0 = HN0 on HN = HN0 . For any N , or asymptotically
as N → ∞, certain physical quantities of SN in the state γN are the same as the
corresponding ones for S0 in the state γN0 . This way, each SN resembles by construction
the original S0 at least for all those physical quantities that are kept fixed with N . By
scaling, one removes the ambiguity one would have in determining the asymptotic of the
expectation f (N, L, a, . . . ) of an observable, by computing limN →∞ f (N, LN , aN , . . . ).
The Gross-Pitaevskiı̆ scaling, in particular, is central in this analysis since it is the limit

141
142 Chapter 8. Concluding remarks

of ultra-high dilution which still enables one to study the dynamics of the gas, because
the ratio between interaction and kinetic energy is kept constant with N .
(1)
3. Role of one-body reduced density matrices. γN is the natural tool to for-
malise and investigate condensation. One-body marginals enter the description of the
many-body system because one usually deals with local observables only, i.e., observ-
ables of the form O ⊗ 1N −k acting on HN ∼ = Hk ⊗ HN −k . Their expectation values
(k) (k)
Tr[ γN (O ⊗ 1N −k ) ] = Tr[ γN O ] involve only the knowledge of γN . In particular, even
when dealing with B.E.C. for finite size systems, experimentally one observes a spatial
density distribution
Z
¯ ¯2 (1)
n(x) = N ¯ΨN (x, x2 , . . . , xN )¯ dx2 · · · dxN = N γN (x, x)

providing
R the average number of particles in some neighborhood of given point (indeed
n(x) dx = N ). As N → ∞, the many-body Hilbert space HN tends to an infinite tensor
power and objects like γN act on a far too large space to control. As a consequence,
when N is large, two “physically essentially equal” many-body states ΨN and ΦN are
essentially orthogonal, whereas evaluating their distance at the level of marginals turn
out to be indeed meaningful. (Compare, for instance, (6.1) with (6.3): the l.h.s. of
(1) (1)
kγN −ρN kL1 6 kΨN −ΦN k vanishes with N , the r.h.s. remaining constant.) Conversely,
(k)
the sequence {γN }N is all inside L1 (Hk ), hence taking N → ∞ makes sense, provided
that a suitable topology for the limit is specified.
4. Complete condensation. The ubiquitous form of B.E.C. in all currently known
rigorous results is that of asymptotic 100% B.E.C., namely, the asymptotic convergence
of the one-body marginal to some rank-one projection. The one-body state it projects
onto is the condensate wave function. This is because these results are obtained within
the Gross-Pitaevskiı̆ scaling, where condensation, if it happens to hold, is necessarily
depletionless. (In fact, in the definition of asymptotic B.E.C. one has also to specify how
the limit N → ∞ is taken.) Yet, in principle the case with depletion is not excluded:
depletion is zero in the limit and (small and) positive before the limit and a (highly
non-trivial, actually a major open problem) control of the errors would provide a control
(1)
of γN for finite large values of N and, hence, a description of the thermal component
around the condensate cloud. The rigorous proof of B.E.C. and of the whole GP theory
(1)
are today achieved within this scaling. Convergence γN → |ϕihϕ| can be equivalently
(1)
controlled in the trace class weak-∗ sense, or as (ϕ, γN ϕ) → 1, or in the Hilbert-Schmidt
(k)
norm, or in the trace class norm, and it implies the factorisation γN → |ϕihϕ|⊗k for
any positive integer k, both in case of pure and of mixed states. Moreover, complete
B.E.C. in the limit N → ∞ is equivalent to the partial asymptotic factorisation (7.18)
and to the self-tail property (7.25) of the many-body pure state (see (7.39) and (7.40)
for the mixed state counterpart).
5. Evolution of an originally trapped condensate. The goal is the rigorous
derivation of the time-dependent GP equation for the one-body condensate wave func-
tion. The strategy is to complete the scheme of derivation of the “method of hierarchies”.
ESY and AGT theorems are currently the most powerful tool to express rigorously the
evolution of the one-body condensate wave function according to the cubic nonlinear
Schrödinger equation. AGT asymptotic factorisation of marginals is now known to be
equivalent to the same trace norm convergence at any time, as for the ESY case.
143

6. Correlations in the condensed state. Interparticle interaction establishes


correlations in the many-body state. In the presence of condensation, a typical short
scale structure is recognised to be present as a modification of the purely factorised ϕ⊗N
and it is also seen to be preserved in time. To control it, one tries to build interparticle
correlations in a trial function Ψtrial
N , since the true many-body state is not accessible in
trial (1),trial
practice. Such a ΨN is chosen to reproduce the limit γN → |ϕihϕ| and, possibly to
be close to the true many-body state both in norm and in energy. This is a subtle task:
currently, a comprehension of the full correlation structure is missing because known
trial functions fail to capture the energy of the many-body state.

Major open problems


Even in the basics of the mathematics of B.E.C. some major problems remain open
(not to mention unsolved questions in other specific B.E.C.-related topics listed in the
Introduction, which did not enter this work).

3 Control of errors for finite N in the asymptotic results.


(1)
3 Condensate vs thermal cloud, through the control of γN before the limit.
(1) N →∞
3 Fragmented condensation (namely, γN −→ α1 |ϕ1 ihϕ1 | + α2 |ϕ2 ihϕ2 |).

3 Control of the correlation structure of a condensate evolving in time.

3 Relaxation of a number of simplifying hypotheses towards more singular interaction


potentials.

3 True δ-potentials in one dimension.

3 Overlap between the GP and the TF scaling regime.

3 Rigorous statistical treatment of B.E.C. at positive temperature.

As a matter of fact, great efforts are already ongoing in the Mathematical Physics
community to address some of these problems.
144 Chapter 8. Concluding remarks
References

[1] R. Adami, On the derivation of the Gross-Pitaevskii equation, Boll. Unione Mat.
Ital. Sez. B Artic. Ric. Mat. (8), 8 (2005), pp. 359–368.

[2] R. Adami, C. Bardos, F. Golse, and A. Teta, Towards a rigorous derivation


of the cubic NLSE in dimension one, Asymptot. Anal., 40 (2004), pp. 93–108.

[3] R. Adami, F. Golse, and A. Teta, Rigorous derivation of the cubic NLS in
dimension one, J. Statist. Phys., 127 (2007), pp. 1193–1220.

[4] M. Aizenman, E. H. Lieb, R. Seiringer, J. P. Solovej, and J. Yngvason,


Bose-Einstein quantum phase transition in an optical lattice model, Phys. Rev. A,
70 (2004), p. 023612.

[5] S. Albeverio, F. Gesztesy, R. Høegh-Krohn, and H. Holden, Solvable


Models in Quantum Mechanics, Texts and Monographs in Physics, Springer-Verlag,
New York, 1988.

[6] M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. Wieman, and


E. A. Cornell, Observation of Bose-Einstein Condensation in a Dilute Atomic
Vapor, Science, 269 (1995), pp. 198–201.

[7] J. Arazy, More on convergence in unitary matrix spaces, Proc. Amer. Math. Soc.,
83 (1981), pp. 44–48.

[8] W. Bao, D. Jaksch, and P. A. Markowich, Numerical solution of the Gross-


Pitaevskii equation for Bose-Einstein condensation, J. Comp. Phys., 187 (2003),
pp. 318–342.

[9] C. Bardos, L. Erdős, F. Golse, N. Mauser, and H.-T. Yau, Derivation of


the Schrödinger-Poisson equation from the quantum N -body problem, C. R. Math.
Acad. Sci. Paris, 334 (2002), pp. 515–520.

[10] C. Bardos, F. Golse, and N. J. Mauser, Weak coupling limit of the N -particle
Schrödinger equation, Methods Appl. Anal., 7 (2000), pp. 275–293.

[11] C. Bardos and N. J. Mauser, The weak coupling limit for systems of N → ∞
quantum particles: state of the art and applications, in Actes du CANUM, 2003.

[12] S. T. Beliaev, Energy spectrum of a non-ideal Bose gas, Sov. Phys. JETP, 7
(1958), pp. 299–307.

[13] J. M. Blatt and V. F. Weisskopf, Theoretical Nuclear Physics, John Wiley


and Sons, New York, 1952, ch. II, p. 74.

145
146 References

[14] N. N. Bogolubov, Energy levels of the imperfect Bose-Einstein gas, Bull. Moscow
State Univ., 7 (1947), pp. 43–56.

[15] , On the theory of superfluidity, J. Phys. (USSR), 11 (1947), pp. 23–32.

[16] , Problems of a Dynamical Theory in Statistical Physics, in Studies in Statis-


tical Mechanics, G. E. de Boer, J. and Uhlenbeck, ed., North-Holland, Amsterdam,
1962. English translation of J. Phys. (USSR), 10 (1946), pp. 256–265.

[17] M. Born and H. S. Green, A General Kinetic Theory of Liquids, Cambridge,


at the University Press, 1949.

[18] S. N. Bose, Plancks Gesetz und Lichtquantenhypothese, Z. Phys., 26 (1924),


pp. 178–181.

[19] C. C. Bradley, C. A. Sackett, J. J. Tollett, and R. G. Hulet, Evidence


of Bose-Einstein condensation in an atomic gas with attractive interactions, Phys.
Rev. Lett., 75 (1995), pp. 1687–1690.

[20] W. Braun and K. Hepp, The Vlasov dynamics and its fluctuations in the 1/N
limit of interacting classical particles, Comm. Math. Phys., 56 (1977), pp. 101–113.

[21] G. Breit, The scattering of slow neutrons by bound protons. I. Methods of calcu-
lation, Phys. Rev., 71 (1947), pp. 215–231.

[22] C. Brislawn, Kernels of trace class operators, Proc. Amer. Math. Soc., 104 (1988),
pp. 1181–1190.

[23] , Traceable integral kernels on countably generated measure spaces, Pacific J.


Math., 150 (1991), pp. 229–240.

[24] J. B. Bru, M. Correggi, P. Pickl, and J. Yngvason, The TF limit for


rapidly rotating Bose gases in anharmonic traps, 2007.

[25] K. A. Brueckner and K. Sawada, Bose-Einstein gas with repulsive interac-


tions: general theory, Phys. Rev., 106 (1957), pp. 1117–1127.

[26] Y. Castin, Bose-Einstein condensates in atomic gases: simple theoretical results,


in Lecture Notes of Les Houches Summer School, vol. VII C, EDP Sciences and
Springer-Verlag, 2001, pp. 1–136.

[27] T. Cazenave, Semilinear Schrödinger equations, vol. 10 of Courant Lecture Notes


in Mathematics, New York University Courant Institute of Mathematical Sciences,
New York, 2003.

[28] C. Cercignani, The Boltzmann equation and its applications, vol. 67 of Applied
Mathematical Sciences, Springer-Verlag, New York, 1988.

[29] C. Cercignani, R. Illner, and M. Pulvirenti, The mathematical theory of


dilute gases, vol. 106 of Applied Mathematical Sciences, Springer-Verlag, New York,
1994.

[30] S. D. Chatterjee, Biographical Memoirs, in Satyendra Nath Bose, Indian


Naional Science Academy, New Delhi, 1984, pp. 59–79.
References 147

[31] E. A. Cornell and C. E. Wieman, Nobel lecture: Bose-Einstein condensation


in a dilute gas, the first 70 years and some recent experiments, Rev. Mod. Phys.,
74 (2002), pp. 875–893.

[32] M. Correggi, T. Rindler-Daller, and J. Yngvason, Rapidly rotating Bose-


Einstein condensates in strongly anharmonic traps, J. Math. Phys., 48 (2007),
pp. 042104–042134.

[33] F. Dalfovo, S. Giorgini, L. P. Pitaevskii, and S. Stringari, Theory of


Bose-Einstein condensation in trapped gases, Rev. Mod. Phys., 71 (1999), pp. 463–
512.

[34] K. B. Davis, M. O. Mewes, M. R. Andrews, N. J. van Druten, D. S.


Durfee, D. M. Kurn, and W. Ketterle, Bose-Einstein condensation in a gas
of sodium atoms, Phys. Rev. Lett., 75 (1995), pp. 3969–3973.

[35] G. F. Dell’Antonio, R. Figari, and A. Teta, Hamiltonians for systems of


N particles interacting through point interactions, Ann. Inst. H. Poincaré Phys.
Théor., 60 (1994), pp. 253–290.

[36] R. B. Dingle, The zero-point of a system of particles, Phil. Mag., 40 (1949),


pp. 573–578.

[37] F. J. Dyson, Ground-state energy of a hard-sphere gas, Phys. Rev., 106 (1957),
pp. 20–26.

[38] A. Einstein, Quantentheorie des einatomigen idealen Gases, Sitzber. Kgl. Preuss.
Akad. Wiss., (1924), pp. 261–267.

[39] , Quantentheorie des einatomigen idealen Gases, II. Abhandlung, Sitzber. Kgl.
Preuss. Akad. Wiss., (1925), pp. 3–14.

[40] A. Elgart, L. Erdős, B. Schlein, and H.-T. Yau, Gross-Pitaevskii equation


as the mean field limit of weakly coupled bosons, Arch. Ration. Mech. Anal., 179
(2006), pp. 265–283.

[41] L. Erdős, B. Schlein, and H.-T. Yau, Derivation of the Gross-Pitaevskii hi-
erarchy for the dynamics of Bose-Einstein condensate, Comm. Pure Appl. Math.,
59 (2006), pp. 1659–1741.

[42] , Derivation of the cubic non-linear Schrödinger equation from quantum dy-
namics of many-body systems, Invent. Mat., 167 (2007), pp. 515–614.

[43] , Rigorous derivation of the Gross-Pitaevskii equation, Phys. Rev. Lett., 98


(2007), p. 040404.

[44] , Derivation of the Gross-Pitaevskii equation for the dynamics of Bose-


Einstein condensate, math-ph/0606017 (2006).

[45] L. Erdős and H.-T. Yau, Derivation of the nonlinear Schrödinger equation from
a many body Coulomb system, Adv. Theor. Math. Phys., 5 (2001), pp. 1169–1205.
148 References

[46] T. Esslinger, I. Bloch, and T. W. Hänsch, Bose-Einstein condensation in a


quadrupole-ioffe-configuration trap, Phys. Rev. A, 58 (1998), pp. R2664–R2667.

[47] E. Fermi, Ricerca sci., 7 (1936), pp. 13–52.

[48] R. P. Feynman, Application of Quantum Mechanics to liquid helium, in Progress


in Low Temperature Physics, vol. 1, Amsterdam: North-Holland, 1955, pp. 17–53.

[49] D. G. Fried, T. C. Killian, L. Willmann, D. Landhuis, S. C. Moss,


D. Kleppner, and T. J. Greytak, Bose-Einstein condensation of atomic hy-
drogen, Phys. Rev. Lett., 81 (1998), pp. 3811–3814.

[50] S. A. Gaal, Linear analysis and representation theory, Springer-Verlag, New York,
1973, ch. VI.7, pp. 376–387. Die Grundlehren der mathematischen Wissenschaften,
Band 198.

[51] J. Ginibre and G. Velo, The classical field limit of scattering theory for non-
relativistic many-boson systems. I, Comm. Math. Phys., 66 (1979), pp. 37–76.

[52] , The classical field limit of scattering theory for nonrelativistic many-boson
systems. II, Comm. Math. Phys., 68 (1979), pp. 45–68.

[53] M. Girardeau and R. Arnowitt, Theory of many-boson systems: Pair theory,


Phys. Rev., 113 (1959), pp. 755–761.

[54] F. Golse, The mean-field limit for the dynamics of large particle systems, in
Journées “Équations aux Dérivées Partielles”, Univ. Nantes, Nantes, 2003, pp. Exp.
No. IX, 47.

[55] Hall, H. E. and Vinen, W. F., The rotation of liquid helium II. I. experiments
on the propagation of second sound in uniformly rotating helium II, Proceedings of
the Royal Society of London. Series A, Mathematical and Physical Sciences, 238
(1956), pp. 204–214.

[56] C. E. Hecht, The possible superfluid behaviour of hydrogen atom gases and liquids,
Physica, 25 (1959), pp. 1159–1161.

[57] K. Hepp, The classical limit for quantum mechanical correlation functions, Comm.
Math. Phys., 35 (1974), pp. 265–277.

[58] K. Huang and C. N. Yang, Quantum-mechanical many-body problem with hard-


sphere interaction, Phys. Rev., 105 (1957), pp. 767–775.

[59] N. M. Hugenholtz and D. Pines, Ground-state energy and excitation spectrum


of a system of interacting bosons, Phys. Rev., 116 (1959), pp. 489–506.

[60] K. Husimi, Some formal properties of the density matrix, Proc. Phys. Math. Soc.
Jpn., 22 (1940), pp. 264–314.

[61] R. Jastrow, Many-body problem with strong forces, Phys. Rev., 98 (1955),
pp. 1479–1484.

[62] V. D. Kamble, Bosons - The Birds That Flock and Sing Together.
http://www.vigyanprasar.gov.in/dream/jan2002/article1.htm.
References 149

[63] W. Ketterle, Nobel lecture: when atoms behave as waves: Bose-Einstein con-
densation and the atom laser, Rev. Mod. Phys., 74 (2002), pp. 1131–1151.

[64] W. Ketterle, D. S. Durfee, and D. M. Stamper-Kurn, Making, probing


and understanding Bose-Einstein condensates, in Bose-Einstein Condensation in
Atomic Gases, S. M. Inguscio, ed., vol. 140 of proceedings of the Enrico Fermi
summer school in Varenna, 1999.

[65] J. G. Kirkwood, The statistical mechanical theory of transport processes I. Gen-


eral theory, The Journal of Chemical Physics, 14 (1946), pp. 180–201. Errata 14,
pp. 347.

[66] , The statistical mechanical theory of transport processes II. Transport in gases,
The Journal of Chemical Physics, 15 (1947), pp. 72–76.

[67] L. Landau, The Theory of Superfluidity of Helium II, J. Phys. USSR, 5 (1941),
p. 71.

[68] L. D. Landau and E. M. Lifshitz, Statisticheskaya Fizika, Fizmatgiz, Moscow,


1951.

[69] T. D. Lee, K. Huang, and C. N. Yang, Eigenvalues and eigenfunctions of a


Bose system of hard spheres and its low-temperature properties, Phys. Rev., 106
(1957), pp. 1135–1145.

[70] T. D. Lee and C. N. Yang, Many-body problem in Quantum Statistical Mechan-


ics. III. Zero-temperature limit for dilute hard spheres, Phys. Rev., 117 (1960),
pp. 12–21.

[71] A. J. Leggett, Bose-Einstein condensation in the alkali gases: some fundamental


concepts, Rev. Mod. Phys., 73 (2001), pp. 307–356.

[72] E. H. Lieb, Simplified approach to the ground-state energy of an imperfect Bose


gas, Phys. Rev., 130 (1963), pp. 2518–2528.

[73] , The Bose fluid, in Lecture notes in Theoretical Physics, vol. VII C, University
of Colorado Press Kluwer Acad. Publ., 1964, p. 175.

[74] E. H. Lieb and R. Seiringer, Proof of Bose-Einstein condensation for dilute


trapped gases, Phys. Rev. Lett., 88 (2002), p. 170409.

[75] E. H. Lieb, R. Seiringer, J. P. Solovej, and J. Yngvason, The mathematics


of the Bose gas and its condensation, vol. 34 of Oberwolfach Seminars, Birkhäuser
Verlag, Basel, 2005.

[76] , The quantum-mechanical many-body problem: the Bose gas, in Perspectives


in analysis, vol. 27 of Math. Phys. Stud., Springer, Berlin, 2005, pp. 97–183.

[77] E. H. Lieb, R. Seiringer, and J. Yngvason, Bosons in a trap: A rigor-


ous derivation of the Gross-Pitaevskii energy functional, Phys. Rev. A, 61 (2000),
p. 043602.
150 References

[78] , A rigorous derivation of the Gross-Pitaevskii energy functional for a two-


dimensional Bose gas, Comm. Math. Phys., 224 (2001), pp. 17–31.

[79] E. H. Lieb and J. Yngvason, Ground state energy of the low density Bose gas,
Phys. Rev. Lett., 80 (1998), pp. 2504–2507.

[80] , The ground state energy of a dilute two-dimensional Bose gas, J. Statist.
Phys., 103 (2001), pp. 509–526.

[81] F. London, On the Bose-Einstein condensation, Phys. Rev., 54 (1938), pp. 947–
954.

[82] A. Michelangeli, Analysis of interparticle correlations in dilute gases with


B.E.C., SISSA digital library, 66/2007/MP (2007).

[83] , Reduced density matrices and Bose-Einstein condensation, SISSA digital


library, 39/2007/MP (2007).

[84] , Role of scaling limits in the rigorous analysis of Bose-Einstein condensation,


J. Math. Phys., 48 (2007).

[85] , Strengthened convergence for reduced density matrices in the derivation of


the Gross-Pitaevskiui equation, SISSA digital library, 35/2007/MP (2007).

[86] A. Minguzzi, S. Succi, F. Toschi, M. P. Tosi, and P. Vignolo, Numerical


methods for atomic quantum gases with applications to Bose-Einstein condensates
and to ultracold fermions, Phys. Reports, 395 (2004), pp. 223–355.

[87] R. A. Minlos and L. D. Faddeev, Comment on the problem of three particles


with point interactions, Soviet Physics JETP, 14 (1962), pp. 1315–1316.

[88] , On the point interaction for a three-particle system in Quantum Mechanics,


Soviet Physics Dokl., 6 (1962), pp. 1072–1074.

[89] G. Modugno, G. Ferrari, G. Roati, R. J. Brecha, A. Simoni, and M. In-


guscio, Bose-Einstein Condensation of Potassium Atoms by Sympathetic Cooling,
Science, 294 (2001), pp. 1320–1322.

[90] H. Narnhofer and G. L. Sewell, Vlasov hydrodynamics of a quantum me-


chanical model, Comm. Math. Phys., 79 (1981), pp. 9–24.

[91] L. Nirenberg, An abstract form of the nonlinear Cauchy-Kowalewski theorem, J.


Differential Geometry, 6 (1972), pp. 561–576.

[92] T. Nisida, A note on a theorem of Nirenberg, J. Differential Geometry, 12 (1978),


pp. 629–633.

[93] L. Onsager, Statistical hydrodynamics, Nuovo Cimento (9), 6 (1949), pp. 279–287.

[94] O. Penrose, On the Quantum Mechanics of Helium II, Philos. Mag., (1951).

[95] O. Penrose and L. Onsager, Bose-Einstein Condensation and liquid Helium,


Phys. Rev., 104 (1956), pp. 576–584.
References 151

[96] C. Pethick and H. Smith, Bose-Einstein Condensation in dilute gases, Cam-


bridge University Press, 2002.

[97] L. Pitaevskii and S. Stringari, Bose-Einstein Condensation, Clarendon Press,


Oxford, 2003.

[98] M. Reed and B. Simon, Methods of Modern Mathematical Physics, vol. 1, New
York Academic Press, 1972.

[99] R. Schatten, Norm ideals of completely continuous operators, Ergebnisse der


Mathematik und ihrer Grenzgebiete. N. F., Heft 27, Springer-Verlag, Berlin, 1960.

[100] B. Schlein, Dynamics of Bose-Einstein condensates, arXiv.org:0704.0813 (2007).

[101] B. Simon, Trace ideals and their applications, vol. 35 of London Mathematical
Society Lecture Note Series, Cambridge University Press, Cambridge, 1979.

[102] , Convergence in trace ideals, Proc. Amer. Math. Soc., 83 (1981), pp. 39–43.

[103] H. Spohn, Kinetic equations from Hamiltonian dynamics: Markovian limits, Rev.
Mod. Phys., 52 (1980), pp. 569–615.

[104] , On the Vlasov hierarchy, Math. Methods Appl. Sci., 3 (1981), pp. 445–455.

[105] L. Spruch and L. Rosenberg, Upper bounds on scattering lengths for static
potentials, Phys. Rev., 116 (1959), pp. 1034–1040.

[106] S. Succi, F. Toschi, M. P. Tosi, and P. Vignolo, Bose-Einstein condensates


and the numerical solution of the Gross-Pitaevskii equation, Computing in Science
and Engineering, 07 (2005), pp. 48–57.

[107] F. G. Tricomi, Integral equations, Pure and Applied Mathematics. Vol. V, Inter-
science Publishers, Inc., New York, 1957, ch. 3.12, pp. 124–127.

[108] L. Vestergaard Hau, B. D. Busch, C. Liu, Z. Dutton, M. M. Burns,


and J. A. Golovchenko, Near-resonant spatial images of confined Bose-Einstein
condensates in a 4-dee magnetic bottle, Phys. Rev. A, 58 (1998), pp. R54–R57.

[109] J. Weiner, V. S. Bagnato, S. Zilio, and P. S. Julienne, Experiments and


theory in cold and ultracold collisions, Rev. Mod. Phys., 71 (1999), pp. 1–85.

[110] T. T. Wu, Ground state of a Bose system of hard spheres, Phys. Rev., 115 (1959),
pp. 1390–1404.

[111] V. I. Yukalov, Principal problems in Bose-Einstein condensation of dilute gases,


Laser Phys. Lett., 1 (2004), pp. 435–461.

[112] J. Yvon, La Théories Statistique des Fluides et l’Equation d’Etat, in Actualités


scientifiques et industrielles, no. 203, Hermann, Paris, 1935, p. 344.

[113] V. A. Zagrebnov and J.-B. Bru, The Bogoliubov model of weakly imperfect
Bose gas, Phys. Reports, (2001), pp. 291–434.

You might also like