Quantum Coulomb gases

Jan Philip Solovej

arXiv:1012.5178v1 [math-ph] 23 Dec 2010

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Classical point charges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
3 Charged quantum gases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1 Quantized particles and classical fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2 Statistics of identical particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3 Grand canonical picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.4 Second quantization and quantization of fields . . . . . . . . . . . . . . . . . . . . . . . 10
3.5 Quantization of the electromagnetic field . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.6 Non-relativistic QED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.7 Relativistic QED Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.1 Stability of the first kind for non-relativistic particles . . . . . . . . . . . . . . . . . . 16
4.2 Grand canonical stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.3 Existence of the thermodynamic limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5 Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.1 Examples of instability of the first kind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.2 Fermionic instability of the second kind . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.3 Instability of bosonic matter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1 Introduction

Ordinary matter is composed of electrons (negatively charged) and nuclei (positively

charged) interacting via electromagnetic forces. The electric potential between two
such particles of charges Q and Q located at r and r in R3 is QQ /|r r |. The
Coulomb potential poses two difficulties: (1) The local singularity and (2) its long
range. One has to understand why the local singularity does not cause instabilities
and why the long range does not have strong macroscopic effects. One of the first

Jan Philip Solovej

Department of Mathematics, University of Copenhagen
Universitetsparken 5, DK-2100 Copenhagen, Denmark, e-mail: solovej@math.ku.dk

1
2 Jan Philip Solovej

major triumphs of the theory of quantum mechanics is the explanation it gives of

the stability of the hydrogen atom (and the complete description of its spectrum)
and of other microscopic quantum Coulomb systems. It, surprisingly, took nearly
forty years before the question of stability of everyday macroscopic objects was
even raised. The rigorous answer to the question came shortly thereafter in what
came to be known as the Theorem on Stability of Matter proved first by Dyson
and Lenard and later, in a more simple and transparent way by Lieb and Thirring.
Since these seminal works, the proof of stability of matter has been extended to
several different settings, including relativistic systems and systems in the presence
of dynamic electromagnetic fields.
We will, in particular, discuss the importance of particle statistics, i.e., whether
the particles are bosons or fermions. Using Bogolubovs theory for Bose gases
Dyson concluded that charged bosonic particles would be macroscopically unsta-
ble. That Bogolubovs theory gives the correct description of charged Bose systems
was proved in a series of papers by the author in collaboration with E.H. Lieb.
Having proved stability of matter the next question is whether one can establish
the thermodynamics of charged systems, i.e., the existence of the thermodynamic
limit. This was originally achieved by Lieb and Lebowitz. We will describe a new
approach to the existence of the thermodynamic limit, which applies to many dif-
ferent quantum Coulomb systems, possibly in the presence of an underlying lattice
structure. This generalizes earlier work of Fefferman.
An important theme in these notes is the use of functional inequalities, among
which a prominent role is played by the Lieb-Thirring inequality.
Another important theme will be how to control the screening of the Coulomb
potential.
The notes are organized as follows. In Section 2 we derive the classical Hamil-
tonian for charged particles interacting with electromagnetic fields. In Section 3 we
discuss different ways of quantizing the system. We may quantize the particles and
leave the fields unquantized or quantize both the particles and the fields. Moreover,
we may consider the particles relativistically or classically. This will give a variety
of different models. In Section 4 we discuss stability both in the sense of stability
of individual atoms and in the sense of stability of macroscopic matter. Finally, in
Section 5 we discuss situations where stability fails, in particular the case of bosonic
matter.

2 Classical point charges

We consider N classical particles with charges Q1 , . . . , QN R situated at points

r1 , . . . , rN R3 . Dimension 3 is of course the physical space dimension, but a natu-
ral questions would be whether the discussion of charged system could be general-
ized to other dimensions. It is however not entirely clear what the correct physical
questions would be and we will there restrict the discussion to dimension 3.
Quantum Coulomb gases 3

We would ultimately like to consider the particles as point charges. Unfortu-

nately this will however lead to divergencies. In order to avoid this we will ini-
tially assume that each particle is given by a charge distribution Qi R (r ri ) where
R (r)R = R3 (r/R) and C0 (R3 ) (a continuous compactly supported function)
with R3 = 1.
We will start the discussion of this system of charged particles from the La-
grangian of the particles and the electromagnetic field. It is1 (using Gaussian units)
N  
LR (r j , A ,V ) = Ti (r j ) Q j r j A R(r j ) Q jV R (r j )
j=1
Z
1
+ (|t A + V |2 | A|2 ).
8
where
r j denotes the velocity of particle j.
T j (v) is the Legendre transform of the kinetic energy T j (p) as a function of mo-
mentum p. We assume that the kinetic energy functions T j are convex functions
on R3 . For a non-relativistic particle of mass m j we have T j (p) = 2m1 j p2 and thus
q
T j (v) = 12 m j v2 and for a relativistic particle it is T j (p) = p2 + m2j m j and

hence T j (v) = m j m j 1 v2. We are using units in which the speed of light is
1 (a convention we will use throughout these notes).
A is the vector potential and the magnetic field is B = A
V is the electric potential. The electric field is

E = t A V.

In order to go to a Hamiltonian description we will choose Coulomb gauge or more

precisely require that
t A = 0,
i.e., we assume that A is independent of time. We will also for simplicity assume
that A decays sufficiently fast that we are allowed to ignore boundary terms when
integrating by parts.
With the Coulomb gauge condition we see that

E = t A

is the divergence free part of the electric field. The total electric field is E = E V
and we have Z Z Z
E |2 = |E
|E E |2 + |V |2 .

1 Strictly speaking, even if we use a relativistic kinetic energy, this Lagrangian is not relativistically
invariant. The reason is that we consider the particles as rigid bodies, which do not Lorenz contract
as they move. We will here ignore this additional complication. The Lagrangian in the form given
here is that of the Abraham model of charged particles [32].
4 Jan Philip Solovej

The electric potential V is not a dynamic variable in the sense that that t V = V
does not occur in LR . The equation for V is

LR 1 N
0= = V Q j R (r r j ),
V 4 j=1

where we have used the Coulomb gauge condition. We recognize the equation for
V as Gauss law
N
4 Q j R (r r j ) = V = (t A + V ) = E .
j=1

In the Hamiltonian formalism this is a constraint equation. The solution is

N Z
V (r) = Qj R (r) |r r j |1 dr.
j=1

The canonical variable dual to r j is

p j = r j LR = T j (r j ) Q j A R(r j ).

The canonical variable dual to A is

LR 1
= t A = (4 )1E ,
t A 4
due to the Coulomb gauge condition. We then find the Hamiltonian function (where
we assume that T j is convex such that the double Legendre transform of T j = T j )

N Z
1
HR (r j , p j , A , E ) = p j r j
4
t A E LR (r j , A ,V )
j=1
N   N
= Tj p j + Q j A R(r j ) + Q jV R (r j )
j=1 j=1
Z Z
1 1
+ E |2 + | A|2 )
(|E |V |2
8 8
N   1 N
= T j p j + Q j A R(r j ) + Q jV R(r j )
j=1 2 j=1
Z
1
+ E |2 + | A|2 )
(|E
8
N  
= Tj p j + Q j A R(r j )
j=1
Quantum Coulomb gases 5
ZZ
1 N N
R (r ri )R (r r j )
+
2 Q j Qi |r r |
drdr
j=1 i=1
Z
1
+ E |2 + |B
(|E B|2 ).
8
If we subtract the (divergent) self-energy
N Q2j Z Z R (r)R (r ) N Q2 Z Z
(r) (r )

1 j
drdr = R drdr
j=1 2 |r r | j=1 2 |r r |

and assume that A is continuous at r j , we see that as R tends to zero HR converges

pointwise to the Hamiltonian
N
Qi Q j
H(r j , p j , A , E ) = Tj p j + Q j A (r j ) +
j=1 1i< jN |ri r j |
Z
1
+ E |2 + |B
(|E B|2 ). (1)
8
Here r j and p j are dual canonical variables as are the field variables A and
(4 )1E . Unfortunately, this Hamiltonian is only formal and suffers from sin-
gularities (the field A that solves Hamiltons equations will be singular at r j ) leading
to severe difficulties in describing the motion of classical point charges.
If external fields Vex and A ex are present, the energy is
N 
 Qi Q j
T j p j + Q j (A
A + A ex )(r j ) + Q jVex (r j ) +
j=1 1i< jN i r j |
|r
Z
1
+ E |2 + |B
(|E B|2 ).
8

3 Charged quantum gases

We now discuss the quantization of the Hamiltonian of charged point particles. We

emphasize that it has not been possible to define a fully relativistically invariant and
causal theory of quantum electrodynamics (QED) and all the models we describe
here are at best approximations to such a theory (if it exists). All models discussed
here are mathematically well defined (except when otherwise stated explicitly).
There are several levels of quantization that may be considered.
We can leave the fields A and E classical and quantize the particles, i.e., de-
scribe them by a square integrable wave function (r1 , . . . , rN ).
We may quantize the particles and the fields, i.e., describe the particles in terms
of a wave function and turn the fields into operator valued functions A and E .
This would require introducing some cut-off regularization in the fields.
6 Jan Philip Solovej

We may second quantize the particles, i.e., let also be an operator valued func-
tion. This procedure is necessary if we consider relativistic particles described by
the Dirac operator.

3.1 Quantized particles and classical fields

The variables of the system are the 3-dimensional vector fields A , E (assumed to
satisfy appropriate regularity and decay properties, at least implying that E and
B = A areN
square integrable) and the wave function , which is a normalized
function in Nj=1 [L2 ( )] j , where R3 (say an open set) and j is a positive
integer counting the number of internal degrees of freedom of particle j, (e.g. a
particle of spin s would correspond to j = 2s + 1). We shall write

= (r1 , s1 , . . . , rN , sN ), r j , s j = 1, . . . , j .

The energy of the system is given by

EN ( , A , E ) = h , HN (A
A, E ) i, (2)
NN j
where h , i refers to the inner product of , j=1 [L ( )]
2 and HN is the
(unbounded) operator (depending on A and E )
N 
 Qi Q j
HN (A
A, E ) = T j i j + Q j (A
A + A ex )(r j ) + Q j Vex (r j ) +
j=1 1i< jN |ri r j |
Z
1
+ E |2 + |B
(|E B|2 ). (3)
8
We will throughout be using units in which the reduced Planck constant h = 1.
The last integral above acts as a (A A and E dependent) scalar in the Hilbert space.
The Hamiltonian H(A A, E ) depends also on the exterior fields Vex and A ex , but we
suppress this in the notation as these fields usually remain fixed. In fact, we will
mostly, and unless otherwise explicitly stated, assume that the exterior fields vanish,
i.e., Vex = 0 and A ex = 0.
The expectation value EN ( , A , E ) is not defined for all in the Hilbert space
N
O
[L2 ( )] j .
j=1

We will here avoid the discussion of domains of self-adjointness of the operator

H(A A, E ). We will instead restrict attention to in the the subspace of smooth
N
functions with compact support, i.e, [C0 ( N )]1 N Nj=1 [L2 ( )] j .
One of the main issues we will discuss in these notes is the question of stability,
i.e., whether EN ( , A , E ) is bounded from below independently of (normalized),
Quantum Coulomb gases 7

A , and E . If such a lower bound holds the operator H(A A, E ) has a self-adjoint
Friedrichs extension and we are actually making claims about this extension. From
our point of view the only complication due to considering the restriction to C0 is
that a possible ground state (a state achieving the lowest possible energy) is most
likely not represented by an element in C0 , but only by an element in the Friedrichs
extended domain. We shall, however, not be concerned with the actual ground states,
but only the energy, so we ignore this issue.
Since the three coordinates of i j + Q j A (r j ) correspond to in general
non-
commuting operators, we must discuss the meaning of T j i j + Q j A (r j ) . We
will, in fact, only consider examples where the functions T j (p) can be written in
terms of (possibly matrix-valued) polynomial expressions of p in such a way that
the meaning of T j (at least on a suitable domain) will be clear. The examples we will
consider are
Non-relativistic kinetic energy operators, where T j (p) = (2m j )1 p2 , i.e., the op-
erator is

T j i j + Q j A (r j ) = (2m j )1 (i j + Q j A (r j ))2 . (4)
We will refer to particles with this kinetic energy as non-relativistic parti-
cles. This is the kinetic energy used when treating non-relativistic atoms and
molecules or ordinary matter.
Relativistic kinetic energy operators, where T j (p) = (p2 + m2j )1/2 m j , i.e., the
operator is


T j i j + Q j A (r j ) = ((i j + Q j A (r j ))2 + m2j )1/2 m j . (5)

The square root of an operator is here defined in the spectral theoretic sense2 .
We will refer to particles with this kinetic energy as relativistic (or sometimes
pseudo-relativistic) particles. Both relativistic and non-relativistic particles may
have internal degrees of freedom corresponding to j , being greater than one.
The non-relativistic and relativistic Pauli-operators. These are operators acting
on two-component vector valued functions given by inserting the operator

j (i j + Q j A (r j ))

into the kinetic energy functions above. Here = ( 1 , 2 , 3 ) is the vector of

2 2 Pauli matrices
     
01 0 i 1 0
1 = , 2 = , 3 = .
10 i 0 0 1

(The subscript j on above indicates that it acts on the internal degrees of free-
dom of particle j.) Thus in this case j = 2. The resulting kinetic energy operators
are

T j i j + Q j A (r j ) = (2m j )1 ( j (i j + Q j A (r j )))2 (6)
2 The operator inside the square root is defined as a self-adjoint operator by Friedrichs extending
it from the domain of smooth functions with compact support.
8 Jan Philip Solovej

for non-relativistic Pauli particles and


2 1/2
T j i j + Q j A (r j ) = j (i j + Q j A (r j )) + m2j mj (7)

for relativistic Pauli particles. For the Pauli operator we have the Lichnerowicz
formula

2
j (i j + Q j A (r j ) = (i j + Q j A (r j ))2 + Q j j B (r j ) (8)

and we see that the Pauli operator includes the coupling of the particle spin to the
magnetic field.
We could also consider the 4 4 Dirac operator

T j (p) = p + m j ,

.i.e.,
T j (i j + Q j A (r j )) = j (i j + Q j A (r j )) + m j j
where and are standard 4 4 Dirac matrices, e.g.,
   
0 I 0
= , =
0 0 I

using a 2 2-block notation. Thus in this case j = 4. In contrast to the other

types of operators the Dirac operator, however, is not positive, in fact, not
bounded below and we will therefore not be able to treat it unless we second
quantize the particle fields. We will discuss this briefly below. A different ap-
proach to deal with the unboundedness from below of the Dirac operator is to
restrict to the subspace of L2 (R3 )4 which corresponds to the positive spectral
subspace of the Dirac operator. This approach is called the no-pair theory, but we
will not discuss it further here.

3.2 Statistics of identical particles

Until now all particles have been considered as distinguishable, but if we have iden-
tical particles the issue of particle statistics plays an important role.
The N-particle space for N-identical particles moving in R3 and with
N
internal degrees of freedom is HN = N [L2 ( )] . On HN we define the orthogonal
projections PN

1
N!
(PN )(r1 , s1 , . . . , rN , sN ) = (1) (r 1 (1) , s 1 (1) , . . . , r 1 (N) , s 1 (N) ).
SN
Quantum Coulomb gases 9

They project onto the symmetric (+) or antisymmetric () subspaces. We denote

V
these subspaces HN = PN HN . We will also use the notation HN = N [L2 ( )] .
In case of N identical particles, i.e., if the operators T j and the charges Q j = Q
are the same for all j = 1, . . . , N, the Hamiltonian HN in (3) maps the subspaces HN
to themselves and it makes sense to restrict to these subspaces. We will write

EN ( , A , E ) (9)

to emphasize that we restrict to HN . In the symmetric case (+) we say that

we have a system of N bosons in the antisymmetric case () we say that we have a
system of N fermions.
It is of course also possible to have mixtures of several species of identical parti-
cles being fermions or bosons or even several species of identical particles together
with a number of distinguishable particles. We leave it to the reader to work out the
structure of the underlying Hilbert space and the Hamiltonian in the general case.
We will look at specific examples later.

3.3 Grand canonical picture

It is often useful to consider a situation where the particle number is not specified
at the outset, but where we would instead ask what the optimal particle number
is in a given situation, e.g., what number of particles minimizes the energy. This
picture is referred to as the grand canonical picture. The optimal particle number
may if necessary be adjusted by adding a term times the particle number to the
Hamiltonian. Such a parameter is called a chemical potential. We note that this
is not the same as adding a constant to the exterior electric potential Vext as such a
constant will multiply the total charge of the system.
In order to treat variable particle number we define the bosonic or fermionic Fock
spaces

M
F = F ((L2 ( )) ) = HN ,
N=0

with the convention that H0= C. The element 1 C = H0 is referred to as the

vacuum vector and will be denoted by |0i. For a normalized vector F we may
L
write =
N=0 N , where N HN with N=0 kN k = 1. We say that such a
2

vector represents a grand canonical state and we define the grand canonical energy
(with chemical potential included)

E ( , , A , E ) = EN(N , A , E ) + NkN k2. (10)
N=0

As before this energy is not definedLfor all , but we restrict to corresponding

to finitely many particles, i.e., = M
N=0 N for some finite integer M and where
each N is in C0 .
10 Jan Philip Solovej

Again it is possible to consider several species of identical particles in the grand

canonical picture. We leave it to the reader to write down the Hilbert space and the
energy (see also below).

3.4 Second quantization and quantization of fields

We shall here give a brief introduction to second quantization and discuss how to
quantize particle fields and the electromagnetic fields.
For f L2 ( ) we define the annihilation operator ae( f ) : HN HN1 for N =
1, . . . by
Z
a( f ) )(r1 , s1 , . . . , rN1 , sN1 ) =
(e N f (rN , sN ) (r1 , s1 , . . . , rN , sN )drN .
sN =1

The adjoint of this operator is ae ( f ) : HN1 HN given by

a ( f ) )(r1 , sn , . . . , rN , sN ) = N f (rN , sN ) (r1 , s1 , . . . , rN1 , sN1 ).
(e

We define the bosonic and fermionic annihilation operators a ( f ) : HN HN1

as the restriction of ae( f ) to the respective subspaces, i.e., a ( f ) = ae( f )|H . The
N

adjoints are a ( f ) : HN1 HN given by a ( f ) = PN ae ( f )|H .
N1
We may extend a ( f ) and a ( f ) to operators on the subspace of the Fock spaces
LM
F corresponding to finite particle numbers, i.e., span M=0

N=0 HN . They can-
not be extended as bounded operators on the full Fock spaces.
The extended operators satisfy the famous commutation (+) and anti-commuta-
tion () relations
[a ( f ), a (g)] = h f , giL2 ( ) I
where [A, B] = AB BA and I is the identity of Fock space (or rather its restriction
to the subspace corresponding to finite particle numbers.
If { f j } is an orthonormal basis in L2 ( ) we define the operator valued distri-
butions

(r, s) = f j (r, s)a ( f j ), (r, s) = f j (r, s)a ( f j ),
j=1 j=1

allowing us to write
Z Z
a ( f ) = f (r, s) (r, s)dr, a ( f ) = f (r, s) (r, s)dr.
s=1 s=1

for f L2 ( ) . Formally we have

[ (r, s), (r , s )] = ss (r r ).
Quantum Coulomb gases 11

We refer to and as field operators.

L
Using these field operators we may write the grand canonical Hamiltonian
N=0 (HN + N), corresponding to identical fermions or bosons, formally as

M Z h i
A, E ) + N) =
(HN (A (r, s) T ss (ir + QA
A(r)) + ss (r, s )dr
N=0 s,s =1
ZZ
1 Q2
2
+ (r, s) (r , s ) (r , s ) (r, s)drdr
ss
|r r |
Z
1
+ E |2 + |B
(|E B|2 ),
8

where T ss , s, s = 1, . . . , refer to the matrix components of the kinetic energy op-
erator. It is left as an exercise to the reader to check that the ordering of and
exactly gives the correct Hamiltonian with no-self interactions.
In this formalism it is easy also to write down the grand canonical operator cor-
responding to K different species of either fermions or bosons. For j = 1, . . . , K let
T j , Q j , and j represent the kinetic energy function, the charge, and the internal
degrees of freedom of species j which is either a fermion or a boson. The relevant
N
Hilbert space is H = Kj=1 F j ([L2 ( )] j ) where F j is the Fock space for species
j. Denoting the field operators for species j by j and j the corresponding grand
canonical Hamiltonian (with chemical potential included) is
K j Z h i
H( , A , E ) = j (r, s) T jss (ir + Q j A (r)) + ss j (r, s )dr
j=1 s,s =1
ZZ
1 K i j Qi Q j
2 i,
+ i (r, s) j (r , s ) j (r , s )i (r, s)drdr
j=1 s=1 s =1 |r r |
Z
1
+ E |2 + |B
(|E B|2 ).
8
The subspace on which H( , A , E ) is defined is the space corresponding to finitely
many particles and where the restriction to each particle sector is a smooth func-
tion with compact support. Although it is hopefully clear what this means it is
rather complicated to write it down explicitly. For the convenience of the reader
we will nevertheless do this now. N
The subspace
NN j 2
of H corresponding

to N j particles
of species j, , j = 1, . . . , K is Kj=1 Pj [L ( )] j , where Pj refers to the rel-
evant projection corresponding to the statistics of species j. We may consider this
N1 N
space a subspace of [L2 ( N1 +...+NK )]1 K k
. The subspace of smooth functions
N N
with compact support is [C0 ( N1 +...+NK 1 1 K k
)] . Thus the subspace on which we
define H( , A , E ) is
12 Jan Philip Solovej


[ M1
M MK
M K
O
N
Oj
\ N1 N
D = span Pj L2 ( ) j [C0 ( N1 +...+NK )]1 K k
.
M1 ,...,M j =0 N1 =0 NK =0 j=1
(11)
The energy of the system with particles being in a state represented by D is
denoted
E ( , , A , E ) = h , H( , A , E ) i. (12)
It is easy to check that this agrees with the definition (10) in the case of only one
species.

3.5 Quantization of the electromagnetic field

We will briefly discuss how to quantize the electromagnetic field. We will remain in
Coulomb gauge and quantize such that A = 0. This is most conveniently done in
momentum space.
For k R3 choose e1 (k), e2 (k) R3 such that e1 (k), e2 (k), k form an orthonormal
basis. e1 , e2 cannot be chosen continuously, but this will not cause problems for what
we want to say.
Let (r, ), = 1, 2 be a bosonic field operator with two internal degrees of free-
dom. They are field operators for the light quanta, i.e., photons. Define the Fourier
transformed operators (Of course they are also simply bosonic field operators)
Z Z
b(k, ) = (2 )3/2 eikr (r, )dr, b (k, ) = (2 )3/2 eikr (r, )dr.

We define the quantized magnetic vector potential as the operator valued distribution
s
Z
2
A (r) = (2 )3/2
R3 |k| e (k)(eikr b(k, ) + eikr b(k, ))dk (13)
=1,2

and the transversal electric field as

Z r
|k|
E (r) = i(2 )1/2 R3 2
e (k)(eikr b(k, ) eikr b(k, ))dk. (14)
=1,2

We then find the commutator between the conjugate variables

 
1
A i (r), E , j (r r) = Pi, j (r, r ),
4 +

where P(r, r ) is the 3 3-matrix valued integral kernel of the projection in L2 (R3 )3
projecting onto divergence free vector fields.
A straightforward (formal) calculation gives for the field energy
Quantum Coulomb gases 13
Z Z
1 1
2
E (r)|2 + | A(r)|2 dr =
|E |k|(b (k, )b(k, )
8 R3 R3

+b(k, )b (k, ))dk.

This expression however is infinite and we must normal order it to get a well-defined
operator: Z
|k|b (k, )b(k, )dk.
R3

This is the field energy operator of the electromagnetic field on the Fock space
F + (L2 (R3 )2 ).

3.6 Non-relativistic QED

We may now write down the Hamiltonian of non-relativistic QED, i.e., of the quan-
tized electromagnetic field coupled to quantized non-relativistic particles. The par-
ticles will be described by the non-relativistic kinetic energies (4) or (6), but since
A is now an operator valued distribution, these operators will not make sense un-
less we again introduce the extended charge distribution of the particles. The grand
canonical non-relativistic QED Hamiltonian for K species of identical particles is
then (ignoring for simplicity the chemical potential)
K j Z


H= j (r, s)T jss (ir + Q j A R(r)) j (r, s )dr
j=1 s,s =1
K i j Z Z
Qi Q j 1
+ i (r, s) j (r , s )
|r r |
j (r , s )i (r, s)drdr
i, j=1 2
s=1 s =1
Z
+ |k|b (k, )b(k, )dk.
R3

This operator is defined on the Hilbert space

O
K O
F j ([L2 ( )] j ) F + ([L2 (R3 )]2 ).
j=1

The operators j are field operators for the particles and is the field operator for
the photons. The energy may be calculated in a state represented by a in the
subspace of the Hilbert space consisting of C0 functions of finitely many particles
and photons (we will not write this explicitly this time). The energy is denoted

ENRQED ( ) = h , H i. (15)
14 Jan Philip Solovej

As written now the model depends on the regularization parameter R. The limit as
R tends to 0 is not well understood and will require at least to renormalize the bare
mass and charges of the particles.

3.7 Relativistic QED Hamiltonian

As already emphasized a Hamiltonian (or for that matter any non-perturbative) for-
mulation of QED is non-existent. Here we simply write down the formal expression
for the Hamiltonian for the electron-positron field (with charge e) interacting with
the electromagnetic field:
4 Z
HQED = e (r, a)( (i + eA
A(r)) + m )a,b e (r, b)dr
a,b=1
ZZ
e2 4 [e (a, r), e (a, r)]+ [e (b, r ), e (b, r )]+
+
8 |r r |
drdr
a,b=1
Z
+ |k|b (k, )b(k, )dk.
R3

Here e refers to the fermionic field operator for the electron-positron field and
is the bosonic field operator for the photon field. The operator A is given by (13).
Note that we have not distinguished between electrons and positrons, but that the
operator is written in a charge conjugation invariant way as the density is written as
the commutator 12 4a=1 [e (a, r), e (a, r)]+ .
The operator HQED is ill-defined unless regularizations are introduced and even
in this case it is very difficult to analyze. The no-photon situation was studied in the
mean-field approximation in [15].

4 Stability

In the previous section we discussed how to define the energy of states of charged
quantum gases in different models.
We have introduced the fixed particle number (or canonical) energy EN ( , A , E )
in (2) (or the bosonic or fermionic analogs in (9)) or the grand canonical en-
ergy E ( , , A , E ) in (12). We also defined the non-relativistic QED energy
ENRQED ( ) in (15).
We will say that a system is stable of the first kind or canonically stable if the
energy EN ( , A , E ) is bounded below independently of A , E , and normalized
. In this case we will call the infimum of EN ( , A , E ) the ground state energy
regardless of whether an actual minimizer (a ground state) exists or not. Thus the
canonical ground state energy of the system is
Quantum Coulomb gases 15

n  O
N 

EN ( ) = inf EN ( , A , E )  L2 ( ) j C0 ( N )1 ...N , k k = 1,
j=1
o
A , E C0 (R3 ; R3 ) .

Note that we are restricting the particles to be in the set whereas A and E are
unrestricted vector fields in R3 . It is immediate to see that we might take E = 0 in
the infimum, this will however not be the case for quantized fields below.
The ground state energy of course depends on the types of particles in the system.
We are suppressing this dependence in order not to overburden the notation.
The ground state is the state of the system at absolute zero temperature. It is of
course also of interest to study quantum gases at positive temperature corresponding
to minimizing the free energy we shall however not do this here.
We could also have chosen to consider the purely static Coulomb potential and
set A = 0, but as we shall see the inclusion of A does not really change the treatment
in the non-relativistic (and non-Pauli) case from the points of view discussed here.
We say that a system satisfies stability of the second kind or stability of matter
if N 1 EN ( ) is bounded below independently of N for all (open or in some cases
sufficiently regular) R3 . This is the version of stability mainly studied in [21].
We will here use a slightly stronger notion which we refer to as grand canonical
stability. We define the grand canonical ground state energy as
 o

E( , ) = inf{E ( , , A , E )  D, k k = 1, A , E C0 (R3 ; R3 ) ,

where E ( , , A , E ) was defined in (12). It of course depends on the species of

particles.
We say that a system is grand canonically stable (with chemical potential ) if

inf | |1 E( , ) > .
R3

The infimum here is over all open sets with bounded volume | | (or possibly
sufficiently regular sets if necessary, but we will not consider such cases here).
The original proof of stability of matter is due to Dyson and Lenard [7, 8] and
later by a simpler method by Lieb and Thirring [28]. We will present a proof of grand
canonical stability in a simple case relying on a combination of the two approaches.
For grand canonically stable systems it is of interest to consider whether the
thermodynamic limit
lim | |1 E( , ) (16)
R3

exists. The limit R3 can be given a precise meaning in different ways. Here we
shall simply take the simple situation of the family of scaled copies L of a fixed
set and let the real parameter L tend to infinity.
16 Jan Philip Solovej

4.1 Stability of the first kind for non-relativistic particles

We shall here prove the stability of the first kind for non-relativistic particles, i.e.,
particles with the kinetic energy (4).
Theorem 1 (Non-relativistic stability of the first kind).
For all C0 ( N )1 N and all vector fields A , E C0 (R3 ; R3 ) we have
D  N Z  E
1 Qi Q j 1
, 2m j (i j + Q j A (r j ))2 + |ri r j | + 8 B |2 )
E |2 + |B
(|E
j=1 1i< jN

Ck k2 ,

where the constant C > 0 depends only on the number of particles N and their
properties, i.e., on 1 , . . . , N N, m1 , . . . , mN > 0 and Q1 , . . . , QN R.
This theorem follows easily from the diamagnetic inequality and the Sobolev
inequality (see [20]).
Theorem 2 (Diamagnetic Sobolev inequality).
For all f C0 (R3 ) and all A C0 (R3 ; R3 ) there is a constant C > 0 such that
Z Z Z 1/3
|(i + A) f |2 || f ||2 C | f |6 .
R3 R3 R3

An immediate corollary of this result (using simply Holders inequality) is the fol-
lowing bound on one-body Schrodinger operators.
Corollary 1 (Lower bound on Schrodinger operator).
For all f C0 (R3 ), A C0 (R3 ; R3 ), 0 V1 L5/2 (R3 ), and 0 V2 L (R3 ) we
have

Z 
5/2
h f , (i + A)2 V1 V2 f iL2 C V1 + kV2k k f k2L2 .

We leave it to the reader to prove Theorem 1 from this corollary and the observation
that |r|1 L5/2 (R3 ) + L (R3 ).
The stability of the first kind holds even if the field energy
Z
1
E |2 + |B
|E B |2
8
is ignored. Moreover, it also holds if A is quantized, i.e., if we replace A (r) by the
operator (13). This last statement follows since A (r) is a commuting family (indexed
by r) and thus may be considered as a classical field.
Quantum Coulomb gases 17

4.2 Grand canonical stability

We turn to the question of grand canonical stability. We will study this in the simple
special case of two species of identical fermions with opposite charges. For grand
canonical stability it is not necessary that all particles are fermions. It is, in fact,
enough that all particles with one sign of the charge, i.e., say, all negatively charged
particles form a collection of finitely many species of fermions. Stability of matter
in this more general setting was proved in [7, 8, 28] (see also [18]) and the case of
grand canonical stability was treated in [17].
One of the main ingredients in the proof of grand canonical stability is the use
of the celebrated Lieb-Thirring inequality [28] (see also [21]) which replaces the
Sobolev inequality which we used in the proof of stability of the first kind.

Theorem 3 (Lieb-Thirring inequality).

Assume 0 V L5/2 (R3 ) and A C0 (R3 ; R3 ) then for all N we have on the anti-
V
symmetric subspace HN = N [L2 ( )] the operator inequality
N   Z
1
2m
(i j + A (r j )) V (r j ) Cm3/2 V 5/2 ,
2
j=1

for a universal constant C > 0. In particular, it is independent of the number N of

particles.

Note the apparent similarity between the Lieb-Thirring inequality and the Corol-
lary 1, to the Sobolev inequality. The important difference is that Corollary 1 would
only imply that
N   Z
1
2m
(i j + A (r j )) V (r j ) Cm3/2 N V 5/2 ,
2
j=1

which, in fact, holds on all of HN (left as an exercise for the reader). The lower
bound with a constant independent of N holds only on the fermionic subspace.
The Lieb-Thirring inequality relates the energy of a gas of independent parti-
cles to the corresponding classical energy. The classical energy (ignoring internal
degrees of freedom) would indeed be
ZZ Z
1 2 8
(p + A(r))2 V (r)drd p = m3/2 V 5/2 .
2m 15
1 2 A(r))2 V (r)0
2m (p +A

As a consequence of the Lieb-Thirring inequality we have the following lower

bound on the kinetic energy of N fermions confined to move in a bounded volume.
VN
Corollary 2. If the open set has finite volume | | then in [L2 ( )] we have a
universal constant C > 0 such that
18 Jan Philip Solovej

N
1
2m (i j + A(r j ))2 Cm1 2/3 N 5/3| |2/3. (17)
j=1

Proof. If we use the Lieb-Thirring inequality with a constant potential V we obtain

N
1
2m (i j + A(r j ))2 NV Cm3/2V 5/2| |
j=1

which gives the estimate above after optimization in V .

We now consider the situation with two species of identical non-relativistic fermions
with masses m > 0 and charges Q where Q > 0. For simplicity we assume that
there are no internal degrees of freedom, i.e., = 1. In this case the Hamiltonian
with particle numbers N for the two species is
N+ N+ +N
1 1
HN+ ,N = 2m+ (i j + Q+A (r j ))2 + 2m
(i j Q A (r j ))2 + VC
j=1 j=N+ +1
Z
1
+ (|E |2 + |B|2 )
8
where the Coulomb energy is
N+ N+ +N
Q+ Q Q2+ Q2
VC = + + .
j=1 i=N+ +1 |ri r j | 1i< jN+ |ri r j | N+ <i< jN+ +N |ri r j |

Note that we have numbered the positively charged particles 1, . . . , N+ and the nega-
tively charged particles N+ + 1, . . . , N+ + N . The Hamiltonian acts on the subspace
N N
!
+
^ +
^
D= L2 ( ) L2 ( ) C0 ( N+ +N ).

Theorem 4 (Simple case of grand canonical stability).

The grand canonical energy in the finite volume set R3
n
E( , ) = inf h , HN+ ,N i + (N+ N ) | D, k k = 1,
o
E , A C0 (R3 ; R3 )

satisfies the stability bound

E( , ) C( , m , Q )| |,

for a constant C( , m , Q ) > 0 depending only on , m , Q .

Proof. We define the distance from particle j to the nearest particle of the opposite
charge, i.e.,
Quantum Coulomb gases 19

j = j (r1 , . . . , rN+ +N )

mini=N+ +1,...,N+ +N |ri r j |, if j = 1, . . . , N+
= .
mini=1,...,N+ |ri r j |, if j = N+ + 1, , . . . , N+ + N

Let j = 6
1
j3 B(r j , j /2)
, where B(r j , j /2) denotes the ball centered at r j with radius
R
j /2 and 1B(r j , j /2) is its characteristic function. Note that j = 1.
We will use the following two observations:
Observation 1:
ZZ
N+ N+ +N
Q+ Q N+ N+ +N
j (r)i (r )
|ri r j | = Q+ Q
|r r |
drdr
j=1 i=N+ +1 j=1 i=N+ +1

Observation 2:
ZZ
Q2 j (r)i (r )
|ri +r j | Q2+ |r r |
drdr
1i< jN+ 1i< jN+

and likewise for the Q2 -terms .

The observations follow from Newtons Theorem:
Z  1
6 1 |r| , if |r| > /2 1
|r r | dr = 1 (3 4|r|2 2 ), if |r| < /2 |r| .
|r |< /2
3

From the two observations above we arrive at the following lower bound on the
Coulomb energy
ZZ
1 (r) (r ) 12 N+
12 N+ +N
VC
2 |r r |
drdr
5 Q2+ j1 5 Q2 j1 ,
j=1 j=N+ +1

where we introduced the smeared charge density

N+ N+ +N
(r) = Q+ j (r) Q j (r)
j=1 j=N+ +1

and used that ZZ

j (r) j (r ) 12

drdr = j1 .
|r r | 5
Using now the positive type (i.e., positivity of the Fourier transform) of the Coulomb
kernel we find
12 N+ 12 N+ +N 2 1
VC Q2+ j1
5 j=N
Q j .
5 j=1 + +1

A similar application of the positive type of the Coulomb kernel goes back to an
early paper of Onsager [31], who might have been the first to address the issue
20 Jan Philip Solovej

of grand canonical stability. Better lower bounds on the Coulomb energy can be
derived by more sophisticated use of the same ideas (see e.g. [1, 29, 21]).
We are led to the following lower bound on the Hamiltonian

HN+ ,N HN+ + HN + (N+ + N )

where
N+ N+
1 12
HN+ = 2m+ (i j + Q+A(r j ))2 5 Q2+ j1
j=1 j=1

and likewise for HN . Observe now that for j = 1, . . . , N+ the length j depends
on the position r j and the positions rN+ +1 , . . . , rN+ +N R of the negatively charged
particles but not on the positions of the other positively charged particles. In other
words we may write
N+ N+
12 12

5 Q2+ j1 = 5 Q2+ (r j )1,
j=1 j=1

where (r) = mini=N+ +1,...,N+ +N |ri r|. We thus have a potential parameterized
by the positions of the negatively charged particles. This observation allows us to
use the Lieb-Thirring inequality Theorem 3. If we choose a parameter R (to be
optimized over) and divide the space into the region where (r) < R (a union of
N possibly intersecting balls of radius R) and (r) > R we obtain from the Lieb-
Thirring inequality
N+
1 1
HN+
2 2m+ (i j + Q+A(r j ))2
j=1
 Z 
3/2
CQ5+ m+ N |r|5/2 dr + R5/2 | |
|r|<R
5/3 3/2
+ N+ | |
Cm1 CQ5+ m+ (N R1/2 + | |R5/2)
2/3

5/3 3/2 5/6
+ N+ | |
Cm1 CQ5+ m+ N | |1/6 ,
2/3
=

where we saved half of the kinetic energy in the first inequality and estimated it by
Corollary 2 in the second inequality. Finally, we optimized over the parameter R > 0.
Since the corresponding estimate holds for HN we finally get the lower bound
5/3 5/3
+ N+ | |
HN+ ,N Cm1 N | |
2/3
+ Cm1 2/3

3/2 5/6 3/2 5/6

CQ5+ m+ N | |1/6 CQ5m N+ | |1/6 + (N+ + N )
C( , m , Q )| |,

where we have minimized in N . We leave it to the reader to determine the exact

form of the constant C( , m , Q ).
Quantum Coulomb gases 21

The same proof would work also if periodic external electric and magnetic fields
were present, e.g., a situation describing a crystal structure.
As should also be clear from the proof the field energy
Z
1
E |2 + |B
|E B |2
8
plays no role for stability in the present case. Moreover, as in the case discussed for
stability of the first kind we could also have considered A quantized.

4.3 Existence of the thermodynamic limit

We will briefly discuss existence of the thermodynamic limit (16). This was first
proved by Lieb and Lebowitz [22] for the case of several species of particles where
all the species of, say, negatively charged particles are fermions. The method does
not allow for an exterior periodic potential or magnetic field. In particular, the
method does work in the case where the nuclei are confined to a periodic crystal
arrangement. This case was later treated by Fefferman in [9]. In [16, 17] an abstract
method was developed to conclude existence of thermodynamic limits for Coulomb
systems in great generality including periodic background potentials.
Indeed, the method relies on establishing general abstract properties of the energy
function that implies existence of the thermodynamic limit.
We will just give a brief overview of the method. For the details and more precise
definitions and assumptions we refer to [16, 17].
Let M = { R3 open and bounded} and consider a map E : M R with
the following properties. Given a function : [0, ) with lim () = 0, a subset
R M of sufficiently regular sets, constants , > 0, and a reference set R,
such that
(A1) (Normalization). E(0) / = 0.
(A2) (Stability). M , E( ) | |.
(A3) (Translation Invariance). R, z Z3 , E( + z) = E( ).
(A4) (Continuity). , R, with and d( , ) > ,
E( ) E( ) + | \ | + | | (| |).
(A5) (Subaverage Property). For all M , we have
Z

1
E( ) E g () d (g) | |r () (18)
|| R3 SO(3)

where d is the Haar-measure of R3 SO(3), (the group of isometries of R3 ) and

| |r := inf{| |, , R} is a regularized volume of .
If E : M R satisfies (A1A5) then it is not very difficult to show that the
thermodynamic limit
22 Jan Philip Solovej

lim ||1 E(g)

exists for all g R3 SO(3), i.e., it exists for all rotations or translations of the ref-
erence set . Under slightly more restrictive assumptions which we will not repeat
here the limit holds for a very large class of regular sets.
We see that (A2) is grand canonical stability. The difficult property to establish
for Coulomb systems is (A5). For being a simplex it is a consequence of the
following result of Graf and Schenker [14] generalizing a somewhat similar estimate
by Conlon, Lieb and Yau [5]:

Theorem 5 (Graf-Schenker inequality).

Let be a simplex in R3 . There exists a constant C such that for any N N,
Q1 , ..., QN R, r1 , . . . , rN R3 and any > 0,
Z
Qi Q j 1 Qi Q j 1g (ri )1g (r j )
d (g)
1i< jN |ri r j | || R3 SO(3) 1i< jN |ri r j |
N
C

Q2j .
j=1

This inequality follows by proving that the function

Z
F(r, r ) = 1g(ri )1g (r j )d (g),
R3 SO(3)

is of the form F(r, r ) = g(|r r |) where g is such that |r|1 (1 g(|r|) has positive
Fourier transform. Recall that for a function f of positive type
N
Qi Q j f (ri r j ) Q2j f (0).
1i< jN j=1

5 Instability

5.1 Examples of instability of the first kind

As an example of a system that can show instability of the first kind we consider two
relativistic particles with masses m1 = m2 = 1 and charges Q1 = 1 and Q2 = Q > 0.
The kinetic energy is given by (5) and we simply set A = 0. Thus the Hamiltonian
is
p p Q
H = 1 + 1 1 + 2 + 1 1
|r1 r2 |
acting on the smooth compactly supported functions in L2 (R3 ) L2 (R3 ). Let
C0 (R6 ) be normalized, i.e., its square integral is one. and define (r1 , r2 ) =
Quantum Coulomb gases 23

3 (r1 /, r2 /) for > 0. Note that is still normalized for all . Then
 p p  
Q
h , H i = 1
, 1 + + 2 +
2 2 .
|r1 r2 |

Thus if we let tend to zero

 p p  
Q
lim h , H i = , 1 + 2 .
0 |r1 r2 |

If Q is large enough we find that the right side is negative and hence for such a Q

lim h , H i =
0

and the system is not stable of the first kind.

On the other hand, if the negatively charged particles belong to a finite number
of fermionic species and if the number of fermionic species, the maximal negative
charge and the maximal positive charge satisfy appropriate bounds, then stability of
matter holds [4, 12, 21, 24, 29].
A similar situation happens for non-relativistic particles interacting with mag-
netic fields according to the Pauli operator (6). Consider as an example a Hamilto-
nian for two particles of mass m = 1 and charges Q1 = Q > 0 and Q2 = Q < 0:

1 1 Q2
A) =
H(A ( (i1 QA A(r1 )))2 + ( (i2 + QA
A(r2 )))2
2 2 |r1 r2 |
Z
1
+ | A|2 ,
8
where we have chosen E = 0 (which is the energetically best choice). The instability
in this case relies on the existence (see [13, 30]) of a non-zero
e L2 (R3 ) and a
R
e e
magnetic field A with | A| < such that
2

1
( (i1 A
e (r1 )))2
e = 0.
2
We may assume that
e is normalized. If for > 0 we set

(r1 , r2 ) = 3
e (r1 /)
e (r2 /)

(which is also normalized) and A (r) = (Q)1 A e (r/) we obtain for the energy
expectation
  Z
Q2 1
h , H(A
A ) ) = =1 , =1 + e |2 .
| A
|r1 r2 | 8 Q2

Again we see that if Q is large enough the right side is negative and hence for
such a Q we have as before lim h , H(A
A ) ) = . As for the relativistic case
24 Jan Philip Solovej

stability of matter also holds in this case under appropriate conditions [10, 25]. This
problem with a quantized field has been treated in [3, 11], the relativistic case with
classical fields is considered in [26], and the relativistic case with quantized field in
[23].

5.2 Fermionic instability of the second kind

As the final topic of these notes we will discuss instability of the second kind.
We will first make a very simple general remark about instability of many-body
systems with attractive interactions which has nothing to do with charged systems
and holds even for fermions.
Theorem 6 (Fermionic instability for attractive 2-body potentials).
Assume that the potential W : Rn R satisfies W (r) c < 0 for all r in a ball
around the origin. Consider the N-body operator
N
1
HN = 2j + W (ri r j )
j=1 1i< jN

V
acting in the fermionic Hilbert space N L2 (Rn ). If n 3 then HN cannot be
stable of the second kind, i.e., we can find a a sequence of normalized vectors
V
N N L2 (Rn ) such that

lim N 1 hN , HN N i = .
N

Proof. Assume that W (r) c < 0 on the ball of radius R centered at the origin.
Define N as the (normalized) Slater determinant

N (r1 , . . . , rN ) = (N!)1/2 det(ui (r j ))Ni,j=1

where u j , j = 1, . . . , N are orthonormalized eigenfunctions corresponding to the N

lowest eigenvalues of the negative Laplacian with Dirichlet boundary conditions for
the largest cube centered at the origin and contained in the ball of radius R. We
extend the functions to be 0 outside the cube. The functions u j are explicit and can
be written in terms of sines and cosines. It is a simple straightforward calculation to
show that in all dimensions n there is a constant Cn such that
N
1
hN , ( j )N i Cn N (n+2)/n R2 .
j=1 2

Comparing with Corollary 2 (written for the case n = 3) we see that there is always
a similar lower bound.
Thus
Quantum Coulomb gases 25

1
N 1 hN , HN N i Cn N 2/n R2 (N 1)c.
2
We see that instability occurs when n > 2.

5.3 Instability of bosonic matter

For matter consisting of charged particles we have discussed that the fermionic prop-
erty ensures grand canonical stability. In this final section we will show that the
fermionic property is indeed a necessity as stability fails for bosons.
We consider two species of bosons with masses m = 1, Q+ = Q = 1, A =
E = 0. We describe them by the standard Schrodinger kinetic energy (4). If we
have N+ positively charged particles and N negatively charged particles we may
write the Hamiltonian as
N+ +N
1 ei e j
HN+ ,N = j + ,
j=1 2 1i< jN+ +N |ri r j |

where e j = 1 if j = 1, . . . , N+ and e j = 1 if j = N+ + 1, . . ., N+ + N . The Hilbert

space is
N+
O N
O
HN+ ,N = PN++ L2 (R3 ) PN+ L2 (R3 ).
This system is not stable of the second kind, in fact, the energy behaves like the
number of particles to the 7/5-th power. The following precise asymptotics was
conjectured by Dyson in [6].
Theorem 7 (Dysons formula).
Let

E(N) = inf inf{ h , HN+ ,N i | HN+ ,N C0 (R3(N+ +N ) ), k k = 1}

N+ +N =N

then as N
 Z Z  Z 
E(N) 
lim = inf 1
2 | | I0
2
5/2  0 ,
 =1 ,
2
(19)
N N 7/5 R3 R3 R3

with I0 given by
Z
1/2 45/4 (3/4)
I0 = (2/ )3/4 1 + x4 x2 x4 + 2 dx = . (20)
0 5 1/4 (5/4)

From the Sobolev inequality (Theorem 2) it follows that the inf on the right of (19)
is finite. In [6] Dyson proved an upper bound on E(N) of the form cN 7/5 and thus
indeed proved the instability of the second kind. In [5] a lower bound of the form
CN 7/5 was established thus concluding that 7/5 is the correct power. The theorem
26 Jan Philip Solovej

was finally proved in [27, 33]. In [19] Lieb proved that if the positively charged
particles have infinite mass then the energy is much smaller, indeed, bounded above
by CN 5/3 a corresponding lower bound had already been proved in [7, 8].
The proof of Theorem 7 relies on an application of Bogolubovs theory of super-
fluidity [2]. The charged system, in fact, forms a superfluid state.
Dysons formula (19) is proved by establishing the corresponding two inequal-
ities. Establishing the lower bound is technically very involved and is beyond the
scope of these notes. It is the content of the paper [27]. We will here give a brief
sketch of the proof of the upper bound from [33]. The upper bound is proved by
finding an appropriate trial state. Here we are guided by Bogolubovs theory.
It turns out that it is significantly easier to write down a grand canonical trial
state than a canonical state. We are, however, interested in a canonical state. This
will not be a serious problem as we will eventually be able to show that the state we
construct is sharply peaked around the average particle number. We will ignore this
point here and simply work with the grand canonical state. We refer the reader to
[33] for details.
Another simplification is to consider the two species of bosons as one species
with two internal degrees of freedom corresponding to the two signs of the charge.
Constructing a trial state in this space will correspond to averaging over states with
different numbers of positively and negatively charged particles.
We are thus considering the Fock space F + = F + (L2 (R3 )2 ). We write a func-
tion f L2 (R3 )2 , as f = f (r, e), where e = 1 is the sign of the charge. Let |0i be
the vacuum vector in F + .
In constructing a bosonic trial state the first guess is to put all particles in the
same one-particle state, i.e., to have a condensate. Let this state be represented by
the (normalized) vector L2 (R3 )2 . Introduce first the normalized grand canonical
vector
 
N N n/2
| i = exp + Na+ ( ) |0i = eN/2 a+ ( )n |0i.
2 n=0 n!

The corresponding state is an average over states with varying occupation in the
condensate . The average particle number in is h |a+ ( )a+ ( )| i = N and the
variance is also

h |(a+ ( )a+ ( ))2 | i h |a+( )a+ ( )| i2 = N.

Thus this state is peaked around particle number N with a standard deviation N.
There is a unitary operator U on F + such that

U a+ ( f )U = a+ ( f ) + Nh , f i.

Using this unitary we may also write | i = U|0i.

A pure condensate like this will however not give the correct state. It is impor-
tant to build pair excitations too. This is achieved as follows. Let { f } =0 be an
orthonormal family in L2 (R3 )2 (they will represent the pair states). The normalized
Quantum Coulomb gases 27

vector F + representing our trial state may be abstractly written

!

 2

= (1 2 )1/4 exp a+ ( f ) Nh , f i | i (21)
=0 =0 2
!


= U (1 ) exp a+ ( f ) |0i.
2 1/4 2
=0 =0 2

We have introduced parameters 0 < < 1 with =0 < to control the occu-
2

pations in the pair states. For simplicity we will assume that and { f }
=0 are real
functions.
We encode the information about the pair states in the positive semi-definite trace
class operator on L2 (R3 )2

2
= | f ih f |.
2
(22)
=0 1

In terms of this operator a lengthy but straightforward calculation shows that

D E
, (a+ ( f ) Nh , f i)(a+ (g) Nhg, i) = hg, f i,

2 3 2

(the inner product

on the left is in F+ and the one on the right is in L (R ) ) and
, (a+ ( f ) Nh , f i) = 0. In particular,


, a+ ( f )a+ (g) = hg, (N| ih | + ) f i.

Or equivalently using the field operators from Section 3.4


, (r, e) (r , e ) = N (r, e) (r , e ) + (r, e; r , e ) (23)

where (r, e; r , e ) is the integral kernel of . Likewise,

 D  p  E
, a+ ( f )a+ (g) = g, N| ih | ( + 1) f ,

or

 p
, (r, e) (r , e ) = N (r, e) (r , e ) ( + 1)(r, e; r , e ). (24)

Moreover, the state represented by satisfies Wicks formula, which for the 4-point
function reads
* +
4
, (a+ (g j ) Nhg j , i )
# #
j=1
* +* +

= , (a#+ (g j ) Nhg j , i# ) , (a#+ (g j ) Nhg j , i# )
j=1,2 j=3,4
28 Jan Philip Solovej
* +* +

+ , (a#+ (g j ) Nhg j , i# ) , (a#+ (g j ) Nhg j , i# )
j=1,3 j=2,4
* +* +

+ , (a#+ (g j ) Nhg j , i# ) , (a#+ (g j ) Nhg j , i# ) .
j=1,4 j=2,3

Here # refers to either a (interpreted as complex conjugation on scalars) or no .

In particular, since is real this gives
D


, (r, e) N (r, e) (r , e ) N (r , e )


E
(r , e ) N (r , e ) (r, e) N (r, e)
p
= | ( + 1)(r, e; r , e )|2 + | (r, e; r , e )|2
+ (r, e; r, e) (r , e ; r , e ) (25)

Armed with these identities we can calculate the expectation of the energy in the
state represented by .
First we will explain, for the special case of the charged Bose system, how to
choose the condensate function and the trace class operator . More precisely, we
will specify their charge dependence. We set
r
1
(r, e) = 0 (r), (26)
2

where 0 is a real normalized function in L2 (R3 ). Thus the condensate function does
not depend on the charge. The operator on L2 (R3 )2 = L2 (R3 ) C2 will be chosen
to have the form  
1 1 1
= 0 ,
2 1 1
where 0 is a positive trace-class operator on L2 (R3 ). Put differently, the integral
kernel of is chosen to be
1
(r, e; r , e ) = ee 0 (r; r ). (27)
2
The charge part of this operator is a rank one operator and thus we also have
 
p p 1 1 1
( + 1) = 0 (0 + 1) .
2 1 1

It is now straightforward to calculate the expectation of the Coulomb potential in

the state represented by . From (2327) we obtain
* +
M
ei e j
, |r r j | =
M=0 1i< jM i
Quantum Coulomb gases 29
* ZZ
+
1
2 ee
, ee (r, e) (r , e ) |r r |1 (r, e) (r , e )
=1
 p 
= NTrL2 (R3 ) K (0 0 (0 + 1)) .

Here K is the operator with integral kernel

K (r, r ) = 0 (r)|r r |1 0 (r ).

The total energy expectation is

D
M E Z
N
, HN+ ,N = |0 |2
N+ ,N =0
2
1  p

+ Tr( 0 ) + NTr K 0 0 (0 + 1) .
2
The final step in the argument is to minimize the above expression over 0 . More
precisely this is done in a semiclassical approximation. We will only sketch this
argument. The rigorous argument can again be found in [33]. We assume that 0 is
the quantization of a classical symbol f (r, p) 0. The semiclassical approximation
to the energy is then
Z
N
|0 |2
2
ZZ p

p2
+(2 )3 f (r, p) + 4 N|p|20 (r)2 f (r, p) f (r, p)( f (r, p) + 1) drd p.
2
Minimizing this expression over f (r, p) and performing the p integration gives
Z Z
N
|0 |2 I0N 5/4 0 (r)5/2 dr,
2

where I0 is given in (20). If we introduce the rescaling 0 (r) = N 3/10 (N 1/5 r),
where is also normalized then the energy expression above becomes
Z Z 
N 7/5 | |2 I0 5/2 ,

which is exactly the expression conjectured by Dyson for the energy.

Note that the instability is also reflected in the shrinking of the linear dimen-
sion of the state with increasing N. According to the scaling of 0 above, the linear
dimension of the state behaves like N 1/5 .
30 Jan Philip Solovej

References

1. Baxter, John R. Inequalities for potentials of particle systems. Illinois J. Math., 24, (1980).
2. Bogolubov, N., On the theory of superfluidity, J. Phys. (U.S.S.R.) 11, 23, (1947).
3. Bugliaro, L. and Frohlich, J. and , Graf, G.M., Stability of quantum electrodynamics with
nonrelativistic matter, Phys. Rev. Lett., 77no. 17, 34943497, (1996).
4. Conlon, Joseph G., The ground state energy of a classical gas. Comm. Math. Phys. 94, no. 4,
439-458 (1984).
5. Conlon, Joseph G. and Lieb, Elliott H. and Yau, Horng-Tzer, The N 7/5 law for charged bosons.
Comm. Math. Phys., 116 , no. 3, 417-448 (1988).
6. Dyson, Freeman J., Ground state energy of a finite system of charged particles, Jour. Math.
Phys., 8, 15381545 (1967).
7. Dyson, Freeman J. and Lenard, Andrew, Stability of matter. I, Jour. Math. Phys., 8, 423434,
(1967).
8. Dyson, Freeman J. and Lenard, Andrew, Stability of matter. II, Jour. Math. Phys., 9, 698711,
(1968).
9. Fefferman, Charles, The Thermodynamic Limit for a Crystal. Comm. Math. Phys., 98, 289-
311 , (1985).
10. Fefferman, Charles, Stability of matter with magnetic fields, CRM Proc. Lecture Notes 12,
119133 (1997).
11. Fefferman, Charles, Frohlich, Jurg, and Graf, Gian Michele, Stability of nonrelativistic quan-
tum mechanical matter coupled to the (ultraviolet cutoff) radiation field, Proc. Natl. Acad. Sci.
USA 93, 1500915011 (1996); Stability of ultraviolet cutoff quantum electrodynamics with
non-relativistic matter, Comm. Math. Phys., 190, 309330, (1997).
12. Fefferman, Charles and de la Llave, Rafael, Relativistic stability of matter. I, Rev. Mat.
Iberoamericana, 2, 119213, (1986).
13. Erdos, Laszlo and Solovej, Jan Philip, The kernel of Dirac operators on S3 and R3 . Rev. Math.
Phys. 13, no. 10, 12471280, (2001).
14. Graf,Gian Michele and Schenker, Daniel, On the molecular limit of Coulomb gases. Comm.
Math. Phys., 174 , no. 1, 215227, (1995).
15. Hainzl, Christian and Lewin, Mathieu and Solovej, Jan Philip, The mean-field approximation
in quantum electrodynamics: the no-photon case, Comm. Pure Appl. Math. , 60, 546596,
(2007).
16. Hainzl, Christian and Lewin, Mathieu and Solovej, Jan Philip,The Thermodynamic Limit of
Quantum Coulomb Systems. Part I. General Theory. Advances in Mathematics. 221, 454487,
(2009).
17. Hainzl, Christian and Lewin, Mathieu and Solovej, Jan Philip, The Thermodynamic Limit of
Quantum Coulomb Systems. Part II. Applications. Advances in Mathematics. 221, 488546,
(2009).
18. Lieb, Elliott H., The Stability of Matter, Rev. Mod. Phys., 48, 553569, (1976).
19. Lieb, Elliott H., The N 5/3 Law for Bosons, Phys. Lett., 70A, 7173, (1979).
20. Lieb, Elliott H. and Loss, Michael, Analysis, Graduate Studies in Mathematics, 14, American
Mathematical Society (2001)
21. Lieb, Elliott H. and Seiringer, Robert, The stability of matter in quantum mechanics, Cam-
bridge University Press, Cambridge (2010).
22. Lieb, Elliott H. and Lebowitz, Joel L. , The constitution of matter: Existence of thermody-
namics for systems composed of electrons and nuclei. Advances in Mathematics, 9, 316398,
(1972).
23. Lieb, Elliott H. and Loss, Michael, Stability of a model of relativistic quantum electrodynam-
ics, Comm. Math. Phys. 228, 561588 (2002).
24. Lieb, Elliott H., Loss, Michael, and Siedentop, Heinz, Stability of Relativistic Matter via
Thomas-Fermi Theory, Helv. Phys. Acta, 69, 974984, (1996).
25. Lieb, Elliott H. and Loss, Michael and Solovej, Jan Philip, Stability of matter in magnetic
fields, Phys. Rev. Lett. 75, 985989 (1995).
Quantum Coulomb gases 31

26. Lieb, Elliott H. and Siedentop, Heinz and Solovej, Jan Philip, Stability and instability of rela-
tivistic electrons in classical electromagnetic fields, Jour. Stat. Phys. 89, 3759 (1997).
27. Lieb, Elliott H. and Solovej, Jan Philip, Ground state energy of the two-component charged
Bose gas, Comm. Math. Phys., 252, 485 534, (2004).
28. Lieb, Elliott H. and Thirring, Walter E. , Bound for the kinetic energy of fermions which
proves the stability of matter, Phys. Rev. Lett. 35, 687689, (1975).
29. Lieb, Elliott H. and Yau, Horng-Tzer, The stability and instability of relativistic matter, Comm.
Math. Phys., 118, 177213, (1988).
30. Loss, Michael and Yau, Horng-Tzer, Stability of Coulomb systems with magnetic fields. III.
Zero energy bound states of the Pauli operator, Comm. Math. Phys. 104, 283290 (1986).
31. Onsager, Lars, Electrostatic Interaction of Molecules, Jour. Phys. Chem. 43, 189196, (1939).
32. Spohn, Herbert, Dynamics of charged particles and their radiation field, Cambridge Univer-
sity Press, Cambridge (2004).
33. Solovej, Jan Philip, Upper Bounds to the Ground State Energies of the One- and Two-
Component Charged Bose Gases. Comm. Math. Phys., 266, No 3, 797818, (2006).