,

,
%

%
,

P
hh
X
P
X
H
h
,




%
H
P
X
h
Q
((

(
(
l
eQ
@
Q
l
e
@
@l

Instituto de Fsica Te
orica
Universidade Estadual Paulista

IFT

DE MESTRADO
DISSERTAC
AO

IFTD.013/09

Evaluation of the partition function of fermions using Grassmann

coherent states without path integrals

Daniel Fernando Reyes Castillo

Orientador
Gastao Inacio Krein

Outubro de 2009

Agradecimentos

I want to take the opportunity to thank many people:

First, I want to thank my parents for all the support they gave me through all my
life, even with that little fear of not knowing what my life was going to be. Probably
they believed in me more than they should have.

To my family and friends as a whole for standing such a smartass as myself. And
for helping me understand (voluntarily but most of the time involuntarily) the importance of a more complete development as a human being.

To my advisor, for the patience he had (and still has) to understand my not even
portunhol, and for all the discussions we had that cleared up a lot of doubts I had.
And for solidifying in me the idea of the necessity (not only the importance) of
challenging established ideas.

To all the friends I made here that became my family, because these experiences
without them would be just an academic experience and not an humanistic one as
well.

And at last, but not the least important, to CAPES for the financial help, that
allowed me to study without any preoccupation.

Resumo

A presente dissertacao tem como objetivo principal fazer uma revisao sobre o
uso de estados coerentes para calcular a funcao de particao gran canonica de sistemas fermionicos, sem empregar integrais de trajetoria. Apos discutir um metodo
de calculo baseado numa expansao de altas temperaturas, formulamos uma teoria de perturbacao otimizada empregando campos auxiliares via transformacao de
Hubbard-Stratonovich. Aproximacoes nao perturbativas tradicionais de campo medio tipo Hartree-Fock e de BCS sao obtidas em ordem zero da teoria de pertubacao
otimizada. Correcoes nao perturbativas a` aproximacao de ordem zero sao implementadas usando uma expansao em potencias de uma interacao modificada, em que
os efeitos dos campos medios sao subtrados da interacao original do Hamiltoniano
da teoria.

Palavras Chaves: Estados coerentes, Integrais de trajetoria, Fermions, Algebra

de Grassmann, Transformacao de Hubbard-Stratonovich, Teoria de perturbacao
otimizada

Areas
do conhecimento: Fsica Nuclear, Teoria de Campos e Partculas Elementares, Fsica da Materia Condensada

ii

Abstract

The primary aim of the dissertation is to review the use of coherent states for
the calculation of the grand canonical partition function for fermion systems, without employing path integrals. After discussing a calculational method based on a
high temperature expansion, we formulate an optimized perturbation theory employing external fields via the Hubbard-Stratonovich transformation. Traditional
non-perturbative mean field approximations like Hartree-Fock and BCS are obtained
in zeroth order in the optimized perturbation theory. Non-perturbative corrections
to the zeroth order approximation are implemented through a power series expansion of a modified interaction,where the effects of the mean fields are subtracted
from the original interaction of the Hamiltonian of the theory.

iii

Indice

1 Introduction

2 Second Quantization Formalism

2.1

Many-particle bases . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

Many-body operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.3

Creation and annihilation operators . . . . . . . . . . . . . . . . . . . 12

2.4

Fock space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.5

Change of representation . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Coherent states and Path Integrals at Finite Temperature

19

3.1

Feynman path integral in quantum mechanics . . . . . . . . . . . . . 20

3.2

Coherent states for bosons . . . . . . . . . . . . . . . . . . . . . . . . 22

3.3

Grassmann algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.4

Coherent states for fermions . . . . . . . . . . . . . . . . . . . . . . . 31

3.5

Path integral for bosons and fermions . . . . . . . . . . . . . . . . . . 36

4 The Niter
oi Method

42

4.1

b s . . . . . . . . . . . 42
Coherent-state representation of the trace of ()

4.2

Explicit expression for  ( , ) . . . . . . . . . . . . . . . . . . . . . 44

4.3

Evaluation of the Grassmann integrals in Tr()s . . . . . . . . . . . . 45

5 Applications to simple problems and possible extensions

49

5.1

Non-interacting non-relativistic Fermi gas . . . . . . . . . . . . . . . . 49

5.2

Interacting nonrelativistic Fermions, canonical transformations . . . . 53

5.3

Perturbation on the mean field . . . . . . . . . . . . . . . . . . . . . . 56

5.4

Optimized perturbation theory

. . . . . . . . . . . . . . . . . . . . . 59

6 Conclusions and perspectives

64

A Two-body operators - Change of representation

67

iv

B Evaluation of U (rn+1 , ; rn , 0)

69

C Closure for bosonic coherent states

71

D Numerical values of the fundamental Grassmann integrals

73

E Symmetrical term for the exponential in the partition function

77

F The Feynman ordering label technique

79

Refer
encias

84

Captulo 1
Introduction

The present dissertation is primarily a review on the use of coherent states in the
evaluation of the quantum grand canonical partition function of a many-particle
system at finite temperature. The main focus of the dissertation are systems of
spin-1/2 fermions. The grand canonical partition function is the fundamental quantity in the mathematical treatment of many-body systems from which all physical
quantities can be derived [1, 2]. However, it can scarcely be calculated in closed
form, a fact that is not surprising in view of the intractability of the many-body
problem. On the other hand, there is great activity on the development of efficient
numerical methods for calculating the partition function non-perturbatively. Monte
Carlo methods have been central to such methods, in particular in the context of
quantum field theory problems. The basic strategy of the Monte Carlo (MC) [3]
method in field theory is to express the trace over field configurations in terms
of a path integral so that the problem is reduced to the evaluation of a multidimensional integral using importance sampling [4]. Path integral formulations of
fermion quantum fields involve the use of anti-commuting Grassmann variables [5],
providing a very useful tool for implementing covariant perturbation theory calculations in gauge theories [6] and super-symmetric field theories [7]. However, this
approach is problematic for nonperturbative approaches like the MC method. For
models (or theories) involving boson-fermion couplings, like Quantum Electrodynamics (QED) and Quantum Chromodynamics (QCD), invariably the application
of the MC method involves a formal, exact integration over the Grassmann variables in favor of determinants that depend only on the boson fields. For models
involving only fermion fields the application of the MC method involves the use of a
Hubbard-Stratonovich transformation [8]. This method introduces auxiliary boson
fields so that the self-interacting term becomes quadratic in order to the Grassmann
variables representing fermion fields can be integrated. Again this leads to determinants that involve only boson fields. In many cases the resulting determinants

can be rewritten as path integrals over additional boson fields and the problem is
then reduced to the evaluation of multidimensional integrals over boson degrees of
freedom. The problem with this approach is that the resulting determinants are
in general complex (when not complex, they might not be positive) and the use of
a MC approach becomes very inefficient or even inapplicable. This problem of a
non-positive determinant is known in the literature as the sign problem.
An alternative to the path integral formulation of the grand canonical partition function is the direct evaluation of the trace over Grassmann variables. A
particularly interesting novel approach in this direction was introduced a few years
ago by Thomaz and collaborators [9]. The method is based on the high temperature expansion of the Boltzmann factor in the partition function and makes use
of the coherent-state representation of the trace [1]. Each term of the expansion is
evaluated exactly exploiting the anti-commuting nature of the Grassmann numbers.
This novel method builds on previous experience in calculating the high-temperature
expansion of the partition function of an anharmonic fermionic oscillator on a lattice [10] and of the one-dimensional Hubbard model [11]. Crucial to the method
are two results obtained by Thomaz and collaborators in two separate publications.
First, Charret, de Souza and Thomaz [12] have shown that the moments of a Gaussian Grassmann multi-variable integral are related to the co-factors of the matrix
of the Gaussian exponential. This result is important because the expansion of the
Boltzmann factor requires the evaluation of a trace of multiple products of operators.
The trace of a product of operators can be expressed in terms of matrix elements in a
coherent-state representation and this leads to a multi-variable integral over Grassmann numbers. Second, I.C. Charret, Correa Silva, S.M. de Souza, O. Rojas Santos,
and M.T. Thomaz [13] have shown that the matrix related to the co-factors mentioned above can be diagonalized analytically through a similarity transformation.
This result is valid for any dimensionality of the matrix and is model independent,
in that it depends only on the kinematical aspects of the approach. This was a
tremendous achievement, since despite the closed form of the result of the multidimensional Grassmann integral in terms of co-factors, their explicit evaluation is still
a formidable task.
In the present dissertation we review this approach developed by Thomaz and
collaborators. We name this approach the Niteroi method. In addition to reviewing
the method, we indicate further developments beyond the high temperature expansion of the Boltzmann factor. In particular we make the case for using the method
in the context of improving mean field type of approximations through the combined use of the Hubbard-Stratonovich transformation and the ideas of optimized
perturbation theory (OPT) [14]. Specifically, the high temperature expansion of the

partition function can be re-summed in the case of a quadratic Hamiltonian, i.e.

for an Hamiltonian that involves the product of only two field operators. On the
other hand, a mean-field type of approximation is a non-perturbative method that
is able to bring the full Hamiltonian, which in general involves the product of four
field operators, into a quadratic form through a regrouping of the operators. Examples includes the well known Hartree-Fock and BCS approximation schemes [1, 2].
Initially we show explicitly that known mean field type of approximations can be
obtained trivially within the Niteoi method. In addition, we show that one can
reproduce standard formulas for perturbative corrections to the mean field approximations within the same method. As is well known, perturbative corrections to
mean field approximations, as with all kinds of perturbative calculations, become
very involved when higher order corrections are needed. We propose an approach
in that the high order corrections can be calculated in the context of OPT also
known in some contexts as the -expansion Ref. [15], or optimized -expansion [16].
A more complete list of references on this subject can be found in Ref. [17].
We envisage application of the proposed method in different fields. One immediate application is in the context of atomic fermionic gases [18]. The field of fermionic
gases is witnessing explosive interest, both in theoretical and experimental contexts,
and can be considered as a natural follow up of the first experimental realizations of
atomic Bose-Einstein condensates [19]. The first atomic experimental observation
of atomic Fermi gases occurred in 2003 [20] and others followed very soon afterwards [21]. A good review article is Ref. [22] and a more complete list of references
and discussions on recent experimental developments can be found at the sites mentioned in Refs. [24][25]. The excitement on the subject is due to the possibility of
exploring and manipulating experimentally matter composed of particles with no
classical analogue. Contrary to bosons, fermions cannot be described in terms of
the dynamical dynamical variables like position and momentum, they require new
dynamical variables that are not used in Classical Physics, Grassmann variables.
Another interesting aspect of Fermi systems is what became known as the unitarity limit. This is meant to be a limit in which much of the phenomena happening
in such systems are well described by assuming point-like fermions interacting very
strongly through very short-ranged interactions the unitarity limit is realized when
the scattering length characterizing the interaction strength is much larger than the
inter-particle spacing, so that the only scale relevant in the problem is the scattering
length. Such systems are encountered in different fields of physics [23], like in nuclear physics in the context of the low-energy properties of the atomic nucleus and
the structure of neutron stars, in astro-particle physics in studies related to quarkgluon plasma of the early Universe, in condensed matter physics in the context of

strongly correlated electron systems. Theoretical developments closely related to

these subjects and to the main theme of the present dissertation can be found in
Refs. [26]-[29]. These references deal with the use of coherent states in the combined
framework of path integrals and the Hubbard-Stratonovich transformation, mainly
in context of lattice formulations.
We believe that our proposed method has interest beyond pure academics. A major contemporary goal in the physics of atomic Fermi gases is to go beyond the
framework of mean field physics to access manifestations of strong interactions and
correlations. The experimental possibility of tuning the interaction using external
magnetic fields through Feshbach resonances [18] is a powerful experimental tool
to control physics beyond mean field and provides excellent opportunities to test
and understand applicability limits of traditional approximation schemes. Moreover, we also believe that our proposed method can be extended to more ambitious
situations of quantum field theory, like lattice QCD [30] [31]. Here we envisage the
applications in the strong coupling limit of the theory, a subject with renewed recent
interest [32] [33]. The strong coupling expansion of the QCD action resembles in
many respects the high temperature expansion and so the Niteroi method should
be of direct applicability.
A natural question that might arise is why one would give up the possibility of
obtaining an exact result and use, instead, an approximate scheme like OPT? The
exact result is actually a formal one, in that one still needs to perform Monte Carlo
simulations to integrate over the auxiliary scalar fields. An exact, numerical result
is of course preferred, but in many cases it does not bring understanding of the basic
processes responsible for observed features of the system. It is hoped that through
an expansion in a modified interaction one can capture most of the physics relevant
to the problem and that a milder or even no sign problem arises of course this we
will only know with explicit calculations. Also, it is important to understand how
correlations affect the zeroth-order mean field results, and a systematic expansion
that builds such correlations might be very useful for the insight one can get from
this. And finally, comparison with an exact solution will allow to measure the quality
of such an approximate scheme.
The dissertation is organized as follows. In the next Chapter we review the
second-quantization formalism as employed in the context of non-relativistic quantum many-body theory. The discussion is didactic and an effort is made to present
explicit derivation of important results. In Chapter 3, we review the use of coherent states for calculating traces over fermionic variables. We also discuss the path
integral representation of the partition function using coherent states. The Niteroi
method is discussed with detail in Chapter 4. As in the previous Chapters, our

discussion is deliberately didactic and detailed derivations are given whenever possible and adequate. In Chapter 5 we present applications of the Niteroi method to
simple problems. Initially, we consider the illustrative case of the free Fermi gas and
afterwards we consider mean field type of approximations to the interacting nonrelativistic Fermi gas. In Section 5.3 we discuss how to obtain the well known results
of perturbation theory on the top of the mean field approximation. In Section 5.4 we
propose to use the Niteroi method in connection with the Hubbard-Stratonovich [8]
transformation to implement high order optimized perturbation theory [14]-[17] to
improve on the mean field approximation. The aim here is to set up the approach
and no attempt is made to obtain explicit evaluations of high order corrections,
since this would require an specific model and some numerical work. This would
extrapolate the scope of the present dissertation and therefore we leave these issues
for future work. Our Conclusions and Perspectives are presented in Chapter 6. The
dissertation contains also five Appendices, where we collect some demonstrations
cited in the main text.

Captulo 2
Second Quantization Formalism

In the present Chapter we will present a very short review on the basics of the
second quantization formalism for a system of identical particles. At the cost of
being sometimes pedantic, our approach is deliberately didactic, in that we make
an effort to present explicit derivation of important results. Our discussion here
is strongly based on the book of Negele and Orland [1]. We will start discussing
the quantum mechanical description of many-particle systems making use of singleparticle basis states. Next the formalism of second quantization and the Fock space
is discussed. Finally, the important issue of changing representation is presented,
with emphasis on the change from the coordinate representation to the momentum
representation.

2.1

Many-particle bases

Let H be a Hilbert space for one particle and {|i i} a basis of dimension D. Let us
assume that the basis is orthonormal,
hi |j i = ij ,

(2.1)

and complete
D
P

|i ihi | = I.

(2.2)

i=1

We denote the space for N particles by

HN H N times H.

(2.3)

For |N i a vector of HN , it has to satisfy in the configuration space the condition

hN |N i =

d3 r1 d3 rN |N (r1 , , rN )|2 < .

(2.4)

A basis for this space can be taken as the external product of one-particle basis
|i1 iN ) |i1 i |iN i.

(2.5)

It is easily proved that this basis is orthonormal

(i1 iN |j1 jN ) = i1 j1 iN jN ,

(2.6)

and complete
D
P
{i}=1

|i1 iN )(i1 iN | = I,

(2.7)

where {i} denotes all the indices i.

For a system of identical particles, it is known that only completely symmetric
or antisymmetric states are observed in nature
(rP 1 , , rP N ) = P (r1 , , rN ),

(2.8)

where {rP 1 , , rP N } denotes a permutation of the indices r1 , , rN ; is equal to

1 for bosons and 1 for fermions; and the exponent P of indicates the parity
of the permutation. In this expression we have dropped the subindex N in the
wave function. If a state vector |1 N i is symmetric or antisymmetric under
a permutation of particles, it belongs to a Hilbert space of bosons (particles with
N
integer spin) H+
or to a Hilbert space of fermions (particles with half integer spin)
N
H , respectively. As we shall see in the following, the restriction to symmetric or
antisymmetric states implies restrictions on many-body observables.
It is useful to define a symmetrization operator S+ and a antisymmetrization
operator S as
S |1 i |N i

1 P P
|P 1 i |P N i,
N! P

(2.9)

where the factor 1/N ! is conveniently introduced so that S is also a projection

operator, i.e.

1 P
P

S2 |i =
+
2
0
0
N ! Peven
Podd

!
P

Peven

|i

Podd

1 P
P
P
P
|i
=
+
+
+
2
0
0 P
0
0 P
N ! Peven
Peven
Peven
Podd
Podd
Podd
even
odd
"

1 N! P
N! P
=
+
+
N !2 2 Peven
2 Peven
!
1
P
P
=
+
|i
N ! Peven
Podd
= S |i.

N! P N! P
+
2 Podd
2 Podd

!#

|i

(2.10)

The operators S are hermitian, as can be verified by comparing its matrix elements
with the ones of its hermitian conjugated. Explicitly, the matrix elements of S are

given by
(i1 iN |S j1 jN ) = hi1 | hiN |

1 P P
|jP 1 i |jP N i
N! P

1 P P
hi1 |jP 1 i hiN |jP N i
N! P
1 P P
i1 ,jP 1 iN ,jP N ,
=
N! P
=

(2.11)

while the matrix elements of S are given by

(i1 iN |S j1 jN ) = (i1 iN S |j1 jN )
1 P P0
hP 0 i1 |j1 i hP 0 iN |jN i
=
N! P0
1 P P0
P 0 i1 ,j1 P 0 iN ,jN .
=
N! P0

(2.12)

Since the sum over P 0 runs through all the permutations , we can make P 0 = P 1
(i1 iN |S j1 jN ) =

1 P P 1
P 1 i1 ,j1 P 1 iN ,jN ,
N! P

(2.13)

and this proves that both expressions are equal term by term, then
S = S .

(2.14)

A basis for the symmetric or antisymmetric HN spaces is

|i1 iN } S |i1 iN )
1 P P
=
|P i1 i |P iN i.
N! P

(2.15)
(2.16)

It should be noticed that this basis is over complete, since it has non-independent
elements
|i1 i2 i3 iN } = |i2 i1 i3 iN }.
(2.17)
The orthogonality of this basis follows from the two properties of S we have just
demonstrated, namely S2 = S and S = S ,
{i1 iN |j1 jN } = (i1 iN |S S j1 jN )
= (i1 iN |S2 j1 jN )
= (i1 iN |S j1 jN )
1 P P
(i1 iN |P j1 P jN )
=
N! P
1 P P
=
i1 ,P j1 iN ,P jN ,
N! P
8

(2.18)

this is zero if {i } 6= {j }. For the non-zero case and for fermions one cant have
repeated states, so we are going to have just one permutation that doesnt vanish
{i1 iN |j1 jN } =

(1)P
,
N!

instead, for bosons, if the k are repeated nk times so that

(2.19)
D
P
k=1

nk = N , one has

n1 ! nD !
.
N!

(2.20)

P n1 ! nD !{i},{j}
.
N!

(2.21)

{i1 iN |j1 jN } =
Summarizing both cases
{i1 iN |j1 jN } =
The closure of this basis is
D
P
{i}=1

|i1 iN }{i1 iN | = S .

(2.22)

To see why one has the symmetrizer operator appearing on the r.h.s., note that
D
P
{i}=1

|i1 iN }{i1 iN | = S

D
P
{i}=1

|i1 iN )(i1 iN |S

= S IS
= S2
= S ,

(2.23)

where we have used the completeness of the non symmetrized states and the properties S2 = S and S = S . If we think this thoroughly, S is actually the identity
in the symmetrized spaces, since when one applies this operator to any symmetrized
vector we obtain the same vector. In the future when we will mention the identity
I in a symmetric space context we would be referring to S . We can express S in
the original, unsymmetrized basis as
S = S I
D
P

= S

{i}=1

|i1 iN )(i1 iN |

D P
1 P
P |iP 1 iP N )(i1 iN |,
N ! {i}=1 P

(2.24)

or
S = IS
=

D
P
{i}=1

|i1 iN )(i1 iN |S

D P
1 P
P |i1 iN )(iP 1 iP N |.
N ! {i}=1 P

(2.25)

Finally to normalize the orthogonal basis we use the result in Eq. (2.21) and define
the final basis
s

N!
|i iN }
n1 ! nD ! 1
1
P P
= q
|P i1 P iN ).
N !n1 ! nD ! P

|j1 jN i

(2.26)

The orthonormality expressed in this basis is

hi1 iN |j1 jN i = P {i}{j} ;

(2.27)

n1 ! nD !
|i1 iN ihi1 iN | = S .
N!
{i}=1

(2.28)

and the completeness is

D
P

It is important to notice the different notation used to denote the several manyparticle basis we have discussed: the general many-particle state |j1 jN ), the
symmetrized orthogonal state |j1 jN }, and finally, the symmetrized and orthonormal state |j1 jN i.

2.2

Many-body operators

b
Let us consider a many-particle observable O.
We are going to use a physical
condition to know what property a symmetric operator should have. Using the fact
that a permutation operator (P ) is a unitary operator
b | i = h |O
b | i
h1 N |O

1
N
P1
PN

P1
PN
b P | i,
= h1 N |P O

1
N

(2.29)

that is,
b = P O
b P.
O

(2.30)

In other words, a symmetrized operator has to be invariant under any permutation.

If we write the operator using explicitly a basis
P

b =
O

{j,i}

|j1 jN )O{j,i} (i1 iN |,

we can write the r.h.s. of Eq. (2.30) as

b P =
P O

P
{j,i}

P
{j,i}

|P 1 j1 jN )O{j,i} (i1 iN P |
|P 1 j1 P 1 jN )O{j,i} (P 1 i1 P 1 iN |.

10

(2.31)

Since the indices are dummy, we can reorder them so that

b P =
P O

P
{j,i}

|j1 jN )O{P j,P i} (i1 iN |.

(2.32)

Finally the condition for a symmetrized operator would be

O{j,i} = O{P j,P i}. .

(2.33)

b (1) is said to be an one-body operator when

An operator O
b (1) =
O

N
P
b
O

i,

(2.34)

i=1

i.e., it is a sum of operators that depend on one single-particle label only. One
example of such an operator is the kinetic energy
Tb =

N
X
b 2i
p
i=1

2mi

(2.35)

The condition (2.33) for this type of operators, defined with Eq. (2.34), impose that
b =O
b
O
i
j

(2.36)

for every i, j = 1, ..., N . But still each one acting in its own space.
Another class of operators we are going to consider in the present dissertation is
the one formed by two-body operators, defined as
b (2) =
O

N
X

b ,
O
ij

(2.37)

i,j=1

i.e. it is a sum of operators that depend on two single-particle labels only. One
example of such an operator is the interaction potential energy between two particles
Vb =

N
N
X
1X
Vbij .
Vbij =
2 i6=j
i<j

(2.38)

Such a two-body operator is said to be local or velocity independent when it is

diagonal in configuration space, that is the matrix element of the operator in a
general two-particle state |ri rj ) is given by
b r ) = (r r ) (r r ) O(r , r ).
(r1 r2 |O|r
3 4
1
3
2
4
1 2

11

(2.39)

2.3

Creation and annihilation operators

For each single particle state |i i of the space H, we define a boson or fermion
creation operator ai (we are not going to use a hat on these operators) that acts on
a symmetrized vector state in the following way
aj |jN j1 } |j jN j1 }.

(2.40)

The action of ai on the N-particles state |jN j1 } which belongs to the Hilbert
space HN leads to a N +1-particles state |j jN j1 }, which belongs to the Hilbert
space HN +1
aj : HN HN +1 .

(2.41)

The action of ai over a normalized state can be deduced as following

aj |S jN j1 )
q

n1 ! nj ! nD !(nj

|S j jN j1 )
=q
,
n1 ! (nj + 1)! nD !
+ 1)

(2.42)

that is
aj |jN j1 i =

nj + 1 |j jN j1 i.

(2.43)

This leads to the definition of a vacuum state (a state with no particles) |0i such
that
ai |0i = |i i.

(2.44)

This state has to be distinguished from the zero-norm state of the H.

For the hermitian conjugated operator aj , or annihilation operator
aj : HN HN 1 ,

(2.45)

one can deduce its action applying it over an N -particles basis, i.e.
aj |i1 iN }.

(2.46)

Using the identity of HN 1 on the r.h.s. of Eq. (2.46), one has that
D
1
P
|k kN 1 }{k1 kN 1 |aj |i1 iN }.
(N 1)! {k}=1 1
(2.47)
Here we need {jN j1 |aj . This can be obtained considering the expression for
the dual of Eq. (2.40),

aj |i1 iN } =

{jN j1 |aj = {j jN j1 |.

12

(2.48)

Using this in Eq. (2.47), one obtains

aj |i1 iN } =

D
X
1
{j k1 kN 1 |i1 iN }|k1 kN 1 }, (2.49)
(N 1)! {k}=1

and using Eq. (2.21)

D P
1
P
P j,P i1 k1 ,P i2 kN 1 ,P iN |k1 kN 1 }
(N 1)! {k}=1 P
1
P P
j,P i1 |P i2 P iN }.
(2.50)
=
(N 1)! P

aj |i1 iN } =

Next, we just need to expand the sum and the permutations inside each ket to
obtain (N 1)! terms for each permutation of the deltas, and we can sum all of
them because of the property
|i3 i2 i4 iN } = |i2 i3 i4 iN }.

(2.51)

After some algebra, we get

N
X

aj |i1 iN } =

b ik iN },
k1 j,ik |i1

(2.52)

k=1

b i denotes a state removed from the ket at the indicated position. For an
where
orthonormal state, one has

aj |i1 iN i =

iN
1 P
b i iN i.
i1 j,i |i1
nj i=i1

(2.53)

The exchange symmetry of many-particle systems implies certain commutation

properties for the creation and annihilation operators. Namely,
aj ak |i1 iN } = |j k i1 iN }
= |k j i1 iN }
= ak aj |i1 iN },

(2.54)

or
aj ak ak aj [ak , aj ] = 0.

(2.55)

More explicitly, we have defined the commutator and the anticommutator as

[ak , aj ] = ak aj aj ak ,
[ak , aj ]+ = ak aj + aj ak ,

(2.56)

Taking the hermitean conjugate of Eq. (2.55), one has

[ak , aj ] = ak aj aj ak = 0.

13

(2.57)

To obtain the (anti)commutator of a and a , we apply them in sequence on the state

|i1 iN }
aj ak |i1 iN } = aj |k i1 iN }
= jk |i1 iN } +

N
P
l
l=1

j,il |k i1

b il iN }, (2.58)

and
ak aj |i1 iN } = ak
=

N
P
l1
l=1

N
P
l1

j,il |i1

j,il |k i1

l=1

b il iN }

b il iN }.

(2.59)

Using this last result into the first one

aj ak |i1 iN } = jk |i1 iN } + ak aj |i1 iN }.

(2.60)

Therefore, we arrived at the result

[aj , ak ] = aj ak ak aj = j k .

2.4

(2.61)

Fock space

Let us define the Fock space as the space in which the creation and annihilation
operator act
N
H
N =0 H

= H0 H1 H2 ,

(2.62)

with
H0 = |0i.

(2.63)

A basis for this space can be the union of all the symmetrized basis, normalized
{|0i} {|i i} {|i1 i2 i} ,

(2.64)

{|0)} {|i }} {|i1 i2 }} .

(2.65)

or not normalized
These are in fact orthogonal basis, because every state in HN is orthogonal with
0
every state in HN with N 6= N 0 . We are not going to give a general proof of this,
but the following example will suffice


{i |j k } = h0|ai aj ak |0i = h0| i j + aj ai ak |0i



= i j h0|ak |0i + h0|aj i k + ak ai |0i



= i j h0|ak |0i + i k h0|aj |0i + h0|aj ak ai |0i

= 0.

(2.66)

14

The closure condition is going to be just the sum of the completeness relations of
every HN
D
1 P
|i1 iN }{i1 iN S |
N =1 N ! {i}=1
D

1 P
P
n1 ! nD !|i1 iN ihi1 iN |.
= |0ih0| +
N =1 N ! {i}=1

I = |0ih0| +

2.5

(2.67)

Change of representation

Let us consider a change of basis, from {|i i} to a new basis {|i i},
|i i =

D
X

hj |i i|j i.

(2.68)

j=1

By definition, one has that

aj |jN j1 } |j jN j1 }
=
=

D
X
i=1
D
X

(2.69)

hi |j i|i jN j1 }

(2.70)

hi |j iai |jN j1 }.

(2.71)

i=1

Therefore, the creation and annihilation operators behave under this change of transformation as
aj =

D
X

hi |j iai ,

(2.72)

i=1

and
aj =

D
X

hj |i iai .

(2.73)

i=1

The commutation relation between a creation and annihilation operator in the new
basis follows straightforwardly
[aj , ak ] =
=
=

D
X

D
X

i=1
D
X

D
X

hj |i i

l=1

hj |i i

i=1
D
X

hl |k i[ai , al ]
hl |k ii l

l=1

hj |i ihi |k i = hj |k i = j k .

(2.74)

i=1

The commutation relations between two annihilation operators and two creation
operators are easily found to be zero, following exactly the same procedure as above
[aj , ak ] = 0[aj , ak ] = 0.

15

(2.75)

As an example of change of representation, let us assume that we start with

creation and annihilation operators in the momentum representation, i.e. these
create or annihilate particles with defined momentum p (other quantum numbers
might be added when needed), and we want to change to creation and annihilation
operators in the coordinate representation, where they create and annihilate particles
at a definite position r. This can be accomplished using Eq.s (2.72) and (2.73)
b (r) =

hp|ri ap =

p (r) ap ,

(2.76)

p (r) ap ,

(2.77)

and
b
(r)
=

hr|pi ap =

b
where, we have introduced the field operators b (r) and (r);
and

hr|pi = p (r) =

eipr/h
.
(2h)3/2

(2.78)

As it can be seen these equations matches the well known Fourier Transform of
functions.
The commutation relations of the field operators are given by
b
b 0 )] = 0,
[(r),
(r

(2.79)

[b (r), b (r0 )] = 0,

(2.80)

b
[(r),
b (r0 )] = (r r0 ).

(2.81)

All operators of the theory can be written in terms of creation and annihilation
operators. An easy way to express a general operator in terms of creation and annihilation operator is to use a basis in which the operator is diagonal. The expression
of the operator in another basis, in which the operator is not diagonal, can be obtained by a change of representation. To help us do that, let us define the number
operator
b i a ai .
n
(2.82)
i
This operator, when acting on a state |i1 iN }, gives the number of particles in
the state with quantum number i . This can be shown making use of Eqs. (2.52)
and (2.40),
b j |i1 iN } = a
n
j

=
=

N
P
k1
k=1

N
P
k1
k=1
N
P
k=1

j,ik |i1

j,ik |j i1

b i k iN }

b ik iN }

j,ik |i1 ik1 j ik+1 iN }

= nj |i1 iN }.

16

(2.83)

Naturally, the operator that counts all the particles is

D
P

c=
N

b i =
n

i=1

D
P
a
i=1

i a i .

(2.84)

b such that all the O

b , see
For simplicity, let us consider first a one-body operator O,
i
Eq. (2.34), are equal and diagonal in the single particle basis {|j i}
b | i = O | i.
O
i j
j j

(2.85)

In an arbitrary element of this basis

b
O|

j1

N
X

b 1 P P | i |
O
P j1
P jN i
i
N! P
i=1
N
1 P P P
b | i |
=

O
i P j1
P jN i
i=1
N! P
N
1 P P P
=

OP ji |P j1 i |P jN i
i=1
N! P
D
1 P P P
=

nk Ok |P j1 i |P jN i
k=1
N! P
!

jN } =

D
P

k=1
D
P

k=1

nk Ok |j1 . jN }

b k |j1 jN },
Ok n

then

D
P

b =
O

k=1

(2.86)

Ok ak ak .

(2.87)

Next, the transformation to another basis |{i i} (in general of different dimension
D0 )
b =
O

D
P

Ok

D
X

k=1

l=1

D
X

hl |

D
X

Ok

k=1

D0

D
X

D0

D
P

hl |k ial ak =

l=1

D
X

b | ia a
hl |O
i k l k

k,l=1

D0
P

b
|p ihp | O
i

|q ihq | |k ial ak

q=1

p=1

k,l=1

kl al ak =

b | ih | ia a
hl |p ihp |O
i q
q k l k

k,l=1 p,q=1
0

D
X

b | i
hp |O
i q

p,q=1

D
P

hl |p ial

l=1

D
P

hq |k iak .

(2.88)

k=1

Using Eqs. (2.72) and (2.73), one obtains finally

b
O

D0
P

b | ia a .
hp |O
i q p q

p,q=1

17

(2.89)

For example in the configuration representation, one has

b (r )(r
b ).
b = d3 r d3 r hr |O
b |r i
O
1
2
1
i 2
1
2
R

(2.90)

In the case of the kinetic energy operator,

Tb =

b2
p
,
2m

(2.91)

Eq. (2.90) becomes

Tb

h
2 2r2 b
b )
d r1 d r2 hr1 |r2 i
(r1 )(r
2
2m

h
2 Z 3 3
b )
d r1 d r2 (r1 r2 ) b (r1 ) 2r2 (r
2
2m
h
2 R 3 b
b
=
d r (r) 2 (r).
2m
=

(2.92)

For a two body operator, we can do an analogous procedure (see Appendix A),
obtaining the result
0

D
X

b =
O

b | )a a a a .
(r s |O
ij t u r s u t

(2.93)

r,s,t,u=1

In the configuration representation, one will have

b
O

4
Y

!
b )(r
b ).
b (r )
b (r )(r
b |r r )
d rk (r1 r2 |O
2
4
3
ij 3 4
1
3

(2.94)

k=1

For a local or velocity independent operator, see Eq. (2.39), one has
b =
O

4
Y

!
3

d rk

b )(r
b )
(r1 r3 ) (r2 r4 ) O(r1 , r2 ) b (r1 )b (r2 )(r
4
3

k=1

b )(r
b ).
d r1 d3 r2 O(r1 , r2 ) b (r1 )b (r2 )(r
2
1
3

18

(2.95)

Captulo 3
Coherent states and Path Integrals at Finite
Temperature

In the present Chapter we will present a review on the use of coherent states in
the evaluation of the grand canonical partition function. We will show how these
states can be used to obtain a path integral representation of the partition function.
We will also show how they can be used to calculate directly the trace defining the
partition function, without the use of path integrals.
In quantum statistical mechanics description of many-particle systems, the use
of field theoretic methods in Fock space is common practice. In such a formulation,
the use of the grand canonical ensemble is a natural choice, since in Fock space
one deals with states with an indefinite number of particles. The sum over all the
microstates can be written as the trace of the operator in the Fock space as
Z =

h|e (HN ) |i
b

= Tr e (HN ) ,
b

(3.1)

where |i is representing an element of a symmetrized many particle basis. Z is

the grand canonical partition function. All possible information on the macroscopic
states of a many-body system can be derived in principle from Z.
It is striking the resemblance with the trace of the well known evolution operator
of Quantum Mechanics
b
Tr U = Tr eitH/h .
(3.2)
In the next Section we will briefly review the Feynman path integral in quantum
mechanics. Although out of the main scope of the present dissertation, the subject is
included here for two main reasons. First, to motivate the similarities between path
integrals in quantum mechanics and in statistical mechanics. Second, to motivate
future developments of the Niteroi method to problems in quantum field theory, as
will be discussed in a later Chapter in this dissertation.

19

3.1

Feynman path integral in quantum mechanics

The starting point of the Feynman path integral in quantum mechanics is the probability amplitude of finding a particle at position rf at time tf , knowing that it was
at position ri at tf . Specifically, this probability amplitude is given by
U (rf , tf ; ri , ti ) Hhr, tf |r, ti iH = hrf |ei(tf ti )H/h |ri i,

(3.3)

c is
where the sub-index H in Hhr, tf |r, ti iH means Heisenberg representation and H
the hamiltonian of the particle
c = H(p
b, b
H
r)
2
b
p
+ V (br).
=
2m
The next step is to divide the time interval tf ti into M equal parts

tf ti
,
M

(3.4)

(3.5)

so that
tn ti +  n,

n = 0, 1, , M 1.

(3.6)

and


b h
iH/

U (rf , tf ; ri , ti ) = hrf | e

M

|ri i = hrf |eiH/h eiH/h eiH/h |ri i.

b

(3.7)

Inserting M 1 closures between the exponentials



b h
iH/

U (rf , tf ; ri , ti ) = hrf | e

M

|ri i =

Z MY
1

d3 rj hrf |eiH/h |rM 1 i

b

j=1

hrM 1 |eiH/h |rM 2 ihrM 2 | |r1 ihr1 |eiH/h |ri i.

b

(3.8)

Denoting r0 ri and rM rf , the probability amplitude can be written in the more

compact form
U (rf , tf ; ri , ti ) =

Z M 1
1
Q 3 MQ
dr
hr

k+1 |e

j=1

b h
iH/

|rk i,

(3.9)

k=0

and one needs therefore to evaluate the matrix elements of the form
U (rk+1 , ; rk , 0) = hrk+1 |eiH/h |rk i.
b

(3.10)

This is calculated in the appendix ??

b2
p
+ V (br) |rk i
2m
"
!#
Z
b2
i p
3
=
d pk hrk+1 |pk ihpk | exp
+ V (br) |rk i. (3.11)
h
2m
"

i
U (rk+1 , ; rk , 0) = hrk+1 | exp
h

20

!#

For small enough , one has

e

b h
iH/

b2
p
+ V (br) ,
2m
!

i
'I
h

(3.12)

and replacing
U (rk+1 , ; rk , 0) =
=

"

p2k
+ V (rk )
2m

i
d pk hrk+1 |pk ihpk | exp
h

d3 pk hrk+1 |pk ieiH(pk ,rk )/h hpk |rk i

!#

|rk i

eirk+1 pk /h iH(pk ,rk )/h eirk pk /h

e
2h "
2h
#!
Z 3
i (rk+1 rk )
d pk
exp
pk + H(pk , rk ) . (3.13)
=
2h
h


Z

d3 pk

Finally, putting all factors together, one obtains for the probability amplitude the
expression
U (rf , tf ; ri , ti ) =

!
1
d3 p0 Z MY
d3 rj d3 pj
2
h
2h
j=1

1
i MX
(rk+1 rk )
exp
pk + H(pk , rk )
h
k=0


"

#!

(3.14)

In the limit of M ,  0, one recognizes that the exponent is just the i/h
times the integral of the classical Lagrangian of the particle
1
i MX
i Z tf
(rk+1 rk )
dt L(r, r ).

pk + H(pk , rk )
h
k=0

h
ti

"

(3.15)

Denoting
1
d3 p0 MY
d3 rj d3 pj
2
h j=1
2h

[dr][dp],

(3.16)

one can write for the probability amplitude

U (rf , tf ; ri , ti ) =

i Z tf
[dr][dp] exp
dt L(r, r ) .
h
ti


(3.17)

We note that one could integrate over the momenta and obtain the traditional
Feynman path integral that involves only integrals over de coordinates. We decided
to leave the integrals over the momenta variables because in the next sections, when
discussing the path integral in terms of coherent states, we will arrive at expressions
involving two coordinates that can formally be related to generalized coordinates
and momenta.

21

The above derivation of the path integral representation of the partition function
is not adequate when the Hamiltonian and number operators are given in the second
quantization representation. Specifically, in the second quantization representation
the grand canonical potential operator
b H
c N
c,

(3.18)

that appears in the Boltzmann factor in Eq. (3.1) is given in terms of creation and
annihilation operators a and a (or field operators and ). In this case, coherent
states provide an adequate framework to express the partition function in terms of
c-number functions. Coherent states are eigenstates of the annihilation operator and
a qualitative understanding of why they are useful is as follows. In the derivation
of the path integral above, we have made repeated use of the completeness of the
momentum eigenstates because in the exponent of the evolution operator one has
the momentum operator. In order to use the same trick with the grand canonical
potential operator in second quantization, which involves in general the operators
in normal order (i.e. all annihilation operators appear to the right of all creation
operators), one would need eigenstates of the second quantized operators. This is
the subject of our next Sections.

3.2

Coherent states for bosons

For convenience we are going to use the occupation number representation. A generic
many-particle state can be represented as
|i =

n1 nD |n1 nD i,

(3.19)

{n }=0

where the n1 nD are complex numbers and

|n1 nD i =

D
Y

ai
q

i=1

n

|0i.

(3.20)

ni !

Since we are not going to perform any change of basis, there should be no source of
confusion if one simplifies the notation as
ni ni ,
so that, for example,
|i =

n1 nD |n1 nD i,

{n}=0

22

(3.21)

(3.22)

and
|n1 nD i =

D
Y
i=1

ni

ai

|0i.
ni !

(3.23)

We define a coherent state |i as the eigenstate of the annihilation operators ai

ai |i = i |i,

(3.24)

where i is the respective eigenvalue, in general a complex number. Before obtaining

an explicit expression for the eigenstates we should notice the following. Using the
generic notation for the commutation relation for boson and fermion operators, one
sees that
0 = [ak , aj ] |i = [k , j ] |i,

(3.25)

which implies that

[k , j ] = 0.

(3.26)

We see that if we were working with fermions, the numbers i would anticommute,
i.e. they would not be ordinary complex numbers and the concept of anticommuting
c-numbers, known as Grassmann numbers, is required. Here we will concentrate on
bosons, for which the eigenvalues i are ordinary complex numbers.
Let us come back to Eq. (3.24). From the l.h.s. of this equation, using Eq. (3.22)
one has
ai |i =
=
=

X
{n}=0

X
{n}=0

ai n1 nD |n1 ni nD i

n1 nD ni |n1 (ni 1) nD i

n1 (ni +1)nD ni + 1 |n1 ni nD i.

(3.27)

{n}=0

On the other hand, from the r.h.s. of Eq. (3.24), one has
i |i =

i n1 nD |n1 nD i.

(3.28)

{n}=0

From this and Eq. (3.27), one obtains the following recursive relation for the coefficients n1 nD , for every ni

n1 (ni +1)nD ni + 1 = i n1 ni nD
n n n
n1 (ni +1)nD = i 1 i D .
ni + 1
23

(3.29)

This can be solved fixing arbitrarily one of such coefficients. The simplest choice is
n1 nD |{n}=0 = 1,

(3.30)

so that
D
Y
(D )nD
(i )ni
(1 )n1
.

=
n1 nD =
n1 !
nD !
ni !
i=1

(3.31)

In view of this result, the many-particle state in Eq. (3.22) can be written as
|i =

(i )ni
|n1 nD i,
ni !
{n}=0 i=1
Y
D
X

(3.32)

and, because of Eq. (3.23),

|i =

Y
D
X

i ai

|0i

ni

ni !

i=1 ni =0
D
Y

ni

ni !

{n}=0 i=1
D X

i ai

|0i

exp i ai |0i.

(3.33)

i=1

Since we are working with bosons,

[i ai , j aj ] = i j [ai , aj ] = 0,

(3.34)

the product of exponentials can be written as a single exponential as

|i = exp

D
X

i ai

|0i.

(3.35)

i=1

This is the final general expression for the eigenstates of the annihilation operators.
It is important to note that this result is valid for any complex numbers i .
In order to obtain a path integral representation, we need a closure relation for
the coherent states. Here we shall simply give the final expression and in Appendix C
corroborate its correctness. Explicitly, the resolution of the identity for bosonic
coherent states is

D
D
X
1 Z Y
I=
di di exp
j j |ih|,
N i=1
j=1

(3.36)

N = (2i)D .

(3.37)

where

24

We shall need also expressions for the internal product and for operators in the
coherent representation. A general many-particle state can be written as

|gi =

gn1 nD |n1 nD i

{n}=0

gn1 nD

D
Y
i=1

{n}=0

ni

ai

|0i.
ni !

(3.38)

If one defines for every state a function

g(x)

gn1 nD

{n}=0

(x)ni
,
ni !
i=1
D
Y

(3.39)

we have in the coherent representation the state |gi is given by

h|gi = h|g(a )|0i = g( )h|0i,

(3.40)

where we have used the eigenvalue equation Eq. (3.24). Now, because of the normalization choice in Eq. (3.30), we have that

h|gi = g( ) =

gn1 nD

{n}=0

(i )ni
,
ni !
i=1
D
Y

(3.41)

where the last equality follows from the definition in Eq. (3.39). This allows us to
obtain immediately that the inner product of two general many-particle states |f i
and |gi as

D
D
X
1 Z Y

j j hf |i h|gi
hf |gi =
d
d
exp
i
i
(2i)D i=1
j=1
D
D
X
1 Z Y

=
j j [f ( )] g( )
d
d
exp
i
i
D
(2i)
j=1
i=1
D
D
X
1 Z Y

=
d
d
exp
j j f ()g( ),
i
i
(2i)D i=1
j=1

(3.42)

where, of course,

f (x)

fn1 nD

{n}=0

(x)ni
.
ni !
i=1
D
Y

(3.43)

Let us now discuss the coherent-state representation of a general operator given

in terms of creation and annihilation opertors a and a ,
b = O(a , a).
O

25

(3.44)

This representation is most easily obtained when all the creation operators are on
the left of the annihilation operators. Such a repositioning of the operators is always
possible for any operator using the commutation rules. We will call such an ordering
as simple order. As an example, for a one-dimensional Hilbert space we will define
the specific function of operators already ordered by




O (a , a) O00
+ O10
a + O01
a + O11
a a.

(3.45)

Note that in essence the operator is the same

b = O  (a , a),
O

(3.46)

O is just a specific function of a and a. In other words, although

b , a) = O
b  (a , a),
O(a

(3.47)

O(x, y) 6= O (x, y),

(3.48)

in general

with x and y beign any type of variable.

For example, suppose one has an operator of the form
c=
K

K(i, j) ai aj .

(3.49)

i,j

Assuming the following commutation relations, [ai , aj ] = ij , and K(i, j) = K(j, i),
one would have
c = K
c =
K

K(i, j) ai aj

i,j

K(i, j) i,j + aj ai =

i,j

K(i, i) +

K(i, j)ai aj ,

(3.50)

i,j

and so

K00
=



Ki0
= 0 = K0i
,

K(i, i),

Kij = K(i, j).

(3.51)

Having introduced the notion of operators in simple order, one can write their
coherent-state representation as
b 0 i = h|O  (a , a)|0 i = O  ( , 0 )h|0 i.
h|O|

Here we need the scalar product h|0 i



n 0

D
0j
(i )ni Y
0
q

h| i =
hn1 nD |n01 n0D i
0
n
!
nj !
i j=1
{n,n0 }=0 i=1

D
Y

26

(3.52)

nj

D
Y
D
0j
X
(i )ni Y
(i 0i )ni
q

=
=
ni !
ni ! j=1 nj !
{n}=0 i=1
{n}=0 i=1
Y
D
X

D
X
(i 0i )ni
=
= exp
i 0i .
n
!
i
i=1 ni =0
i=1
D X

This leads to
b 0i
h|O|

= O ( , ) exp

D
X

(3.53)

i 0i

(3.54)

i=1

Following similar steps, one can obtain a coherent-state representation of the

trace of an operator
b =
Tr O

b
hn1 nD |O|n
1 nD i

{n}=0

Z Y

D
D
X
X
1

hn

n
|
=
d
d
exp

j j |ih|O|n
1
D
i
1 nD i
i
D
(2i) {n}=0
i=1
j=1

D
D
X
X
1 Z Y

h|O|n

d
d
exp
=
1 nD ihn1 nD |i
j
i
j
i
(2i)D i=1
j=1
{n}=0
D
D
X
1 Z Y

=
j j h|O|i.
d
d
exp
i
i
(2i)D i=1
j=1

(3.55)

Using the operator in simple order, one obtains the final expression
b
Tr O

D
1 Z Y
=
di di O ( , ).
D
(2i)
i=1

(3.56)

Another case in which we can express an operator in the coherent representation

is when the operator is normal ordered, i.e. all creation operators are put on the
left of all annihilation operators, without using the commutation relations (in case
of fermions, one must keep track of minus signs). The operation of normal ordering
b is denoted by : O
b :, and the coherent-state matrix element of : O
b : is
an operator O
given as (for bosons)
0

b : | i = O( , ) exp
h| : O

D
X

i 0i

(3.57)

D
1 Z Y
di di O( , ).
(2i)D i=1

(3.58)

i=1

and its coherent-state trace is

b :) =
Tr(: O

27

3.3

Grassmann algebras

When we discussed Eq. (3.26) we faced the need for anticommuting numbers when
dealing with fermions. Grassmann Algebra is the mathematical framework universally used to address the anticommuting properties of the eigenvalues of the fermionic
annnihilation operators as observed in Eq. (3.26). Here we will review the minimal
material of Grassmann algebras necessary to build fermionic coherent states.
The generators of the Grassmann algebra, or Grassmann numbers(GN) are a set
of objects {i } with i = 1, ..., n such that
[i , j ]+ = 0.

(3.59)

i2 = 0 .

(3.60)

Note that, in particular

We call a basis of a Grassmann algebra all the linearly independent products of the
generators, i.e.
{1, 1 , , n , 1 2 , , 1 n , 2 3 , , 2 n , , 1 2 3 , , 1 2 n , , 1 n }.
(3.61)
The number of elements of the basis, or the dimension, is 2n , since each generator
has just two possibilities, or it appears once or it does not appear at all in the
set above, because of the property in Eq. (3.60). An element of the algebra is
any linear combination of the elements of the basis with complex coefficients, this
elements can be labeled as functions of this generators f (). They are going to be
used as the coherent state representation of a state |f i. Note that a function of just
one generator can only be linear
f () = f0 + f1 .

(3.62)

We have to define how these new variables are going to behave under the adjoint
(i ) . As we are using them as numbers, we will need an analog to the complex
conjugate and we will call it conjugation operation ( )
(i ) = (i ) .

(3.63)

Using the same symbol as the complex conjugate gives us a more friendly notation,
but the two operations will be different by definition. No confusion should arise when
is used because its apearance in both cases are excludent. Both came from taking
hermitic conjugate , but in one case is over complex numbers and in the other one
is over Grassmann numbers. To actually define the operation properly, let a GA

28

with an even number of generators n = 2p. We select p generators and through this
operation we associate to each element only one element of the remaining p elements
as
i (i ) = i+p .

(3.64)

In order to avoid conflict when we use the symbol for both operations, we need
the following auxiliary properties
(i ) = i ,

(i )

(3.65)
i

(3.66)

where is an ordinary complex number. Now, since the adjoint of the product of
noncommuting mathematical objects in general is given as
(i j ) = (j ) (i ) ,

(3.67)

(i j ) = j i .

(3.68)

we need

As an example for all these initial definitions, lets consider the simplest GA with
conjugation operation, n = 2. The dimension of this GA is 22 , with generators
and that satisfy anticommutation relations according to Eq. (3.59),
[, ]+ = 0,

[, ]+ = 0,

[ , ]+ = 0.

(3.69)

A basis of the algebra is

{1, , , }.

(3.70)

The conjugated of f ( ) would be

[f ( )] = (f0 + f1 )
= f0 + f1
= f (),

(3.71)

as in the Boson case.

We are not going to need a derivative with respect to GNs too much, but we
are going to see a little of it just to get familiar with such an operation. Since there
is no analog to an infinitesimal differential ( 0), one defines the Grassmann
derivative (GD) as

0
and
1,
(3.72)

29

where is an ordinary complex number, and is a GN. As GN are anticommuting

variables, we need a rule on how to operate with a derivative on a product of two
GNs, and the convention is
0

( ) =
( 0 ) = ( 0 ) =

0 = 0 .

(3.73)

In contrast to derivatives, we are going to need integrals over GNs extensively.

Even though there is no analog to the familiar sum motivating the Riemann integral,
and neither can we work with integration limits, we do have three guiding principles
that we can use to define integrals of GN in analogy to ordinary indefinite integrals.
Let us suppose we are able to define coherent states |i such that
ai |i = i |i,

(3.74)

with a a closure relation in the form

I=

Z Y
D

di di |i k(, ) h|,

(3.75)

i=1

where k is a general function

k( , ) k0 + k1 + k2 + k3 .

(3.76)

We get the inner product in coherent representation using the closure relation as
Z Y
D

hf |gi =

di di hf |ik( , )h|gi

i=1

Z Y
D

di di f ()k( , )g( ).

(3.77)

i=1

It is clear that our choice in the definition of the integrals will affect the inner
product, but we want to keep the scalar product representation independent. In
Appendix D, we will make the case for explicitly choosing the following definition
of an integral
Z
Z
d = 0
and
d = 1,
(3.78)
where is a complex number. This definition is not only for simplicity, it is also for
convenience, as we shall see. Note that the definition above for the integral leads to
results numerically identical to the corresponding derivatives. That means that any
formula obtained for derivatives is valid for integrals as well. It is also customary to
introduce a rule for the integration of products of GNs similar to the rule for the
derivatives as
Z

d =

d ( ) =

d =

30

Z

d 0 = 0 .

(3.79)

3.4

Coherent states for fermions

In this Section we will obtain an explicit expression for |i. In order to do so, notice
that we need twice as many generators for any possible state in the one-particle
Hilbert space, in other words
p = D.
(3.80)
In the following, for simplicity of presentation will consider just for one generator
(D = 1). The general form for a fermionic coherent state in this case is
|i = f ()|0i + g()|1i,

(3.81)

a|i = |i.

(3.82)

with the defining condition

As we dont have any criterion to determine whether the annihilation operators and
its eigenvalues commute or anticommute,
[a, ] = 0,

(3.83)

we are going to proceed considering both possibilities. Recalling that in general one
can have
f () = f0 + f1 ,
g() = g0 + g1 ,
(3.84)
the l.h.s. of Eq. (3.82) can be written as
a|i = a [f ()|0i + g()|1i] = f ()a|0i + g()a|1i
= g()|0i = (g0 g1 ) |0i,

(3.85)

while its r.h.s. as

|i = [f ()|0i + g()|1i] = f0 |0i + g0 |1i.

(3.86)

Therefore,
g0 = 0,

g1 = f0 .

(3.87)

On the other hand, we have

h0|i = f ()h0|0i + g()h0|1i = f ().

(3.88)

Therefore, fixing the iterative constant h0|i

h0|i = 1

31

f () = 1,

(3.89)

one then has

f0 = 1,

f1 = 0,

g1 = 1.

g0 = 0,

(3.90)

Replacing these results in Eq. (3.81), one finally has

|i = |0i |1i = (1 a )|0i = ea |0i,

(3.91)

so that

|i = ea |0i.

[a, ] = 0

(3.92)

As said in the previous Section, the choice of definition of the Grassmann integrals determine the final expression for closure. Using the definitions given in
Eq.(3.78) see Appendix D, the closure is
I=

d de |ih|.

(3.93)

Its generalization for many generators is

I=

Z Y
D

di di exp

D
X

j j |ih|,

(3.94)

j=1

i=1

where, for this case, the coherent states are given by

|i =

D
Y

exp i ai |0i,

(3.95)

i=1

where the in the exponent depend on the choice for

[ai , j ] = 0,

(3.96)

as discussed above. Here we use the common choice

[ai , j ]+ = 0.

(3.97)

In a vague common sense, the rational for such a choice is that, since we have two
sets of different mathematical objects that anticommute separately (here, the as
and the s), the most natural behavior seems to be that all of them anticommute
among themselves. There is no profound physical or mathematical reason for such a
choice and one could equally well pick the other option without any inconvenience.
With this choice, the coherent state is given by
|i = exp

D
X
i=1

32

i ai

|0i,

(3.98)

and the corresponding bra by

h| = h0| exp

D  
X

ai

D
X

= h0| exp

(i )

i=1

ai i .

(3.99)

i=1

Now, to obtain a coherent-state representation of the inner product of |f i and

|gi we use Eq. (3.94) so that
hf |gi =

Z Y
D

di di exp

i=1

Z Y
D

D
X

j j hf |ih|gi

j=1

di di exp

i=1

D
X

j j f ()g( ),

(3.100)

j=1

where f and g are now functions of many generators

g( ) = h|gi = h|

gm1 ,,mD |m1 mD i

(3.101)

m1 ,...,mD =0,1

gm1 ,,mD h| a1

m1

aD

mD

|0i

{m}=0,1
mD
gm1 ,...,mD (1 )m1 (D
=
)

{m}=0,1

g{m}

D
Y

(i )mi , (3.102)

i=1

{m}=0,1

and similarly for f (),

f () = [f ( )] =

mD
fm1 ,...,mD (1 )m1 (D
)

{m}=0,1

fm
mD 1m1 =
1 ,,mD D

{m}=0,1

1
Y
mi

f{m}

i .

i=D

{m}=0,1

Using these in the expression for the inner product above, one obtains
hf |gi =

Z Y
D

di di

exp

i=1

1
X

g{n}

X
{m,n}=0,1

j j

j=1

{n}=0

D
X

D
Y

f{m}

{m}=0,1

1
Y

kmk

k=D

(l )nl

l=1

f{m}
g{n}

Z Y
D

di di exp

i=2

d1 d1 exp (1 1 ) 1m1 (1 )n1

D
X
j=2

D
Y
l=2

33

(l )nl

j j

2
Y
k=D

kmk

(3.103)

Z Y
D

f{m}
g{n} m1 n1

Z

di di

exp

i=3

{m,n}=0,1

D
X

j j

j=3

d2 d2 exp (2 2 ) 2m2 (2 )n2

Y
D

3
Y

kmk

(3.104)

k=D

(l )nl

l=3

f{m}
g{n}

D
Y
i=1

{m,n}=0,1

fm
g
.
1 ,,mD m1 ,,mD

mi ni =

(3.105)

{m}=0,1

To make it clear, the coherent-state representation of the inner product of two

many-fermion states gives the appropriated value

fm
g
.
1 ,,mD m1 ,,mD

hf |gi =

(3.106)

{m}=0,1

For the fermion coherent-state representation of a simple-ordered operator

b 0 i = h|O
b  (a , a)| 0 i = O  ( , 0 )h| 0 i,
h|O|

(3.107)

one needs to evaluate h| 0 i. This is given as follows,

h| 0 i = h0|

D
Y

exp (ai i )

i=1

= h0|

D
Y

D
Y

exp j0 aj |0i = h0|

j=1

D
Y

(1 + i ai ) 1 i0 ai |0i = h0|

= h0|

D
Y

(1 i0 ai + i ai i ai i0 ai )|0i

i=1

(1 + i i0 ai ai )|0i = h0|

i=1
D
Y

exp (i ai ) exp i0 ai |0i

i=1

i=1

= h0|

D
Y

D
Y

[1 + i i0 (1 ai ai )]|0i

i=1

(1 +

i i0 )|0i

D
X

= exp

i=1

i i0

(3.108)

i=1

Therefore, the result for the fermion coherent-state representation of a simpleordered operator is given by
0

b i = O ( , ) exp
h|O|

D
X

i i0

(3.109)

i=1

Finally, we consider the coherent-state representation of trace of operators. First,

let us consider the trace of an unordered operator
b =
Tr O

b
hn1 nD |O|n
1 nD i

{n}=0

34

hn1 nD |

Z Y
D

di di exp

i=1

Z Y
D

Z Y
D

D
X

di di exp

D
X

di di exp

i=1

D
X

j j

b
j j |ih|O|n
1 nD i

b
hn1 nD |ih|O|n
1 nD i

{n}=0

j j

j=1

D
X
j=1

j=1

i=1

di di exp

i=1

{n}=0

Z Y
D

b
h|O|n
1 nD ihn1 nD |i

{n}=0

b
j j h|O|i,

(3.110)

j=1

where the sign change in the bra h| above comes from the exchange of the relative
b
positions of hn1 nD |i and h|O|n
1 nD i under the integral this can be shown
by writing each of such factors as in Eq. (3.84) and then regrouping them in reverse
order. One can go a bit further in the evaluation of the above trace taking the
operator in simple order, since then
b =
Tr O

Z Y
D

di di exp

di di

exp

Z Y
D

b  (a , a)|i
j j h|O

D
X

j j O ( , ) exp

di di exp 2

D
X

D
X

i i

i=1

j=1

i=1

j=1

i=1

Z Y
D

D
X

j j O ( , ).

(3.111)

j=1

i=1

Now, making , one obtains finally

b =
Tr O

Z Y
D

di di exp 2

D
X

j j O ( , ).

(3.112)

j=1

i=1

For a normal ordered operator

h|

b:
:O

| i = : O( , ) : exp

D
X

i i0

(3.113)

i=1

one has the result



b: =
Tr : O

Z Y
D

di di

exp 2

i=1

D
X

j j

: O( , ) : .

(3.114)

j=1

Note that now it is important to indicate the normal ordering in : O( , ) : since

and anticommute and as such their relative positions in the function O( , )
matters note that this was not necessary in the case of operators given in terms
of boson operators, since and are complex numbers and therefore commute.

35

3.5

Path integral for bosons and fermions

In the present Section we make use of the formalism of coherent states developed
above to obtain a path integral representation for the grand-canonical partition function of many-particle systems. The grand-canonical partition function was defined
b Eq. (3.18), as
in Eq. (3.1) and can be written in terms of the grand-potential ,
Z = Tr e .

(3.115)

Whenever possible we shall use a common notation for bosons and fermions. To this
extent we denote by |i a generic coherent state of bosons or fermions, i.e.
ai |i = i |i,

|i = exp

D
X

i ai

|0i,

(3.116)

i=1

with = 1 for bosons and = 1 for fermions, ai and ai denote creation and
annihilation operators that satisfy the generic commutation relations
[ai , aj ] = ij ,

[ai , aj ] = 0 = [ai , aj ] ,

(3.117)

and the parameters i satisfy

[i , j ] = 0.

(3.118)

The content of this last equation is that for bosons (and = 1), the i s are ordinary
complex numbers (previously denoted as i ) so that they commute trivially. In
addition, the closure relation is denoted as
I=

i=1

with
N

D
X
di di
exp
j j |ih|,
N
j=1

Z Y
D

2i
1

for bosons,
.
for fermions,

(3.119)

(3.120)

b is
The coherent-state matrix elements of an operator O
b 0i
h|O|

= O ( , ) exp

D
X

i i0

(3.121)

b
j j h|O|i,

(3.122)

i=1

b as
and the trace of O
b =
Tr O

Z Y
D
i=1

di di
N

exp

D
X
j=1

36

or
Z Y
D

b =
Tr O

i=1

D
X
di di
exp ( 1)
j j O ( , ),
N
j=1

Z Y
D

D
D
X
X
di di
exp
j j O ( , ) exp
j j
N
j=1
j=1

i=1

(3.123)

b is put in simple order.

when O
Having set the notation, one has that the coherent-state representation of the
trace in Eq. (3.115) for both bosons and fermions can be written as

Z=

Z Y
D
i=1

D
X
di di
b
exp
j j h|e |i.
N
j=1

(3.124)

Now, following the same strategy used in the derivation of the quantum-mechanical
path integral, the interval [0, ] is partitioned into M equal pieces as

,
M

(3.125)

so that the matrix element h|e b |i appearing in the integral in Eq. (3.124) can
be written as
h|e |i = h|[e ]M |i = h|e e e e |i.
b

(3.126)

Next, we introduce the decomposition of the identity, Eq. (3.119), between any two
consecutive exponentials so that
Z =

Z Y
D
i=1

b


h|e

|2 ih2 |

Z Y
D
i=1

i=1

(3.127)

D
X
di,M
1 di,M 1

j,M
exp
1 j,M 1 |M 1 ihM 1 |
N
j=1

Z Y
D
i=1

Z Y
D

D
X
di di
exp
j j
N
j=1

D
X
di,1
di,1
b

exp
j,1
j,1 |1 ih1 |e |i
N
j=1

!
!

D
D M
1
X
Y
di,k
di,k
di di Z Y
exp
j j
N
N
i=1 k=1
j=1

exp

D M
1
X
X

j, l j, l h|e |M 1 i h2 |e |1 ih1 |e |i. (3.128)

b

j=1 l=1

Here we have introduced the notation i, l to indicate that for every insertion of
the identity decomposition one needs a different dummy variable i , so that l =
1, , M 1 because we have M 1 insertions. Defining
i,0 i ,

i,M i ,

37

(3.129)

one can rewrite Eq. (3.128) more succinctly as

Z=

Z Y
D Y
M
i=1 k=1

D X
M
1
X
Y
di,k
di,k
b
exp
j, l j, l
hm |e (a ,a) |m1 i. (3.130)
N
j=1 l=1
m=M

,a)

Next, we have to evaluate the matrix elements hm |e b(a

exponent to first order in ,
,a)

e (a
b

|m1 i. We expand the

b , a) = 1 
b  (a , a),
' 1  (a

(3.131)

so that
,a)

hm |e (a
b

b  (a , a) |
|m1 i ' hm | 1 
m1 i
h

= hm |m1 i 1   (m
, m1 )
 ( ,
m m1 )

' hm |m1 i e

= exp

D
X

 ( ,
m m1 )

e
p,
m p, m1

(3.132)

p=1

Using this result in Eq. (3.130), one obtains

Z =

i=1 k=1

1
Y

M
D X
X
di,k
di,k
j, l j, l
exp
N
j=1 l=1

Z Y
D Y
M

exp

D
X

p,
m p, m1 e

m=M

p=1

Z MY
1 Y
D

di,k
di,k
N

k=0 i=1

,
  (m
m1 )

eS(

,)

(3.133)

where S( , ) is the result of combining the exponentials as

S( , ) =

M X
D
X

j,k

(j,k j,k1 ) + 

k=1 j=1
M
D
X
X

j,k

k=1

j=1

 (k , k1 )

k=1

M
X

(j, k j, k1 )
+  (k , k1 ) .


(3.134)

Now, if one imagines M , then  0, so that the index k in the exponential

becomes a continuous variable ,
i,k i ( ),

38

(3.135)

and one recognizes in the same limit that

(j,l j,l1 )
j ( )

and

M
X

f (i, k )

Z
0

k=1

d f (i ( )).

(3.136)

In addition, from (3.129) one has that the continuous variables i ( ) satisfy the
boundary condition
i () = i (0).
(3.137)
Therefore, the partition function can be written as
Z=

Z ()

D[ ( )( )] eS(

,)

(3.138)

(0)

where
S( , )

Z
0

D
X

j ( )
d j ( )
+  ( ( ), ( )) ,

j=1

(3.139)

and we used the notation

di,k
di,k
N

Z Y
D Y
M
i=1 k=1

Z ()

D[ ( )( )],

(3.140)

(0)

This is the final result for the path integral representation for the grand-canonical
partition function.
Sometimes it is useful to have an expression for S( , ) that is symmetrical in
the derivatives. This can be achieved averaging both expressions (see Appendix E)
M
1
X

S( , ) 

k=0

j,k+1
j,k
D
1 P
(j,k0 +1 j,k0 )
0

j,k
2 j=1 j,k +1



(k+1
, k )

(3.141)

In the trajectory notation

R( , )

D
j ( )
1 P
j ( )
d
j ( )

j ( )
2 j=1

"

+ ( ( ), ( )) .

Z
0

(3.142)

d
j ( )
j +  ( ( ), ( )) ,

j=1

D
X

(3.143)

Finally, we note the formal analogy with the quantum mechanical path integral
derived in Section 3.1. First, we note that if one defines in the quantum mechanical

39

path integral variables i pi + iri , and i pi iri , the integration measure would
simply be replaced by
(3.144)
D[ri pi ] D[i i ].
Second, making t i , the exponent in Eq. (3.17) would be precisely of the
form given above with the exception that there is no finite limit on the time
integral. This formal analogy means that the path integral for the partition function
in statistical mechanics is equivalent to the quantum mechanical path integral for
the probability amplitude in imaginary time with the addition of the fields being
periodic or antiperiodic in .
One path integral one will need in Chapter 5 is the following

I(A, a , a) =

[d d] e

Z "Y
N

A+a + a

di di exp

i=1

N X
N
X

i Aij j +

i=1 j=1

N
X

(ai i + i ai ) ,(3.145)

i=1

where A is an N N matrix and a and a are vectors of N components. Let us

start with the following integral (b is an ordinary number)
I(b) =

= b

d d e
Z

d d (1 b ) =

d d (1 + b )

d d = b.

(3.146)

Next, we consider the integral (A is an N N matrix)

Z "Y
N

I(A) =

di di exp

N
N X
X

i Aij j .

(3.147)

i=1 j=1

i=1

Initially we prove that an integral over the set of Grassmann variables {i , i , i =

1, , N } is invariant under unitary transformations. Let U be an N N unitary
matrix and define new Grassmann variables i0 = Uij j (here and in the following
we use the convention that there is a sum over repeated indices), then using the
completely antisymmetric symbol in N -dimensions ijl we can write
N
Y
i=1

1 ijl 0 0

i j l0
N!
1 ijl
=

Uii0 i0 Ujj 0 j 0 Ull0 l0
N!
N
1 ijl
0 0
0 Y
=

Uii0 Ujj 0 Ull0 i j l
i
N!
i=1

i0 =

= (det U )

N
Y

i =

i=1

N
Y
i=1

40

i ,

(3.148)

since U is unitary. Therefore, if we take U as being the unitary matrix that diagonalizes the matrix A, and i are the eigenvalues of A, we obtain the result
I(A) =

Z "Y
N

di di exp

i=1

Z "Y
N

i Aij j

i=1 j=1

di di exp

i=1
N
Y

N X
N
X

N
X

i i i

i=1

i = det A.

(3.149)

i=1

With this result we can now consider the integral I(A, a , a). If one make the change
of variables
= 0 + a A1 ,
= 0 + A1 a,

(3.150)

one has
A + a + a = ( 0 + a A1 )A( 0 + A1 a)
+ a ( 0 + A1 a) + ( 0 + a A1 )a
= 0 A 0 0 AA1 a a A1 A 0 a A1 AA1 a
+ a 0 + a A1 a + 0 a + a A1 a
= 0 A 0 + a A1 a,

(3.151)

and therefore
I(A, a , a) =

= e

[d d] e

a A1 a

A+a + a

[d d] e
A1 a

= (det A) ea

(3.152)

Another result we will need in Chapter 5 is related to the functional derivative

of Grassmann functions. In analogy to the bosonic functional derivatives, one has
that
Z Y 0
(3.153)
[ dj ]f (j0 ) = f [i ].
i
j

41

Captulo 4
The Niter
oi Method

In the present Chapter we will review the Niteroi method for obtaining the partition
function of a many-fermion system without resorting to the path integral representation.
The aim is to evaluate the partition function
Z = Tr (e )
b

(4.1)

b has been defined in Eq. (3.18), given by

b H
c N
c. As said above, the
where
starting point is a high-temperature expansion of the exponential in Eq. (4.1) in the
form

X
()s
b s.
Tr ()
(4.2)
Z=
s!
s=0

The first term s = 0 of the expansion is trivial. In essence, the method consists
in calculating the remaining traces exclusively in the coherent-state representation.
For a Hamiltonian that is quadratic in the creation and annihilation operators, as we
shall see, the series can be summed and therefore the validity of the result obtained
is not restricted to the high temperatures only.

4.1

c s
Coherent-state representation of the trace of ()

b
In order to grasp the essential steps to evaluate the traces of multiple products of
in Eq. (4.2), we start the discussion for D = 1. In this case, one has that the trace
b is given by Eq. (3.112), which translates for D = 1 to
of an operator O
b
Tr O

dd e2 O ( , ).

(4.3)

We need the trace of the product of s operators in Eq. (4.2). But let us first consider
the trace of the product of two operators using the expression for the trace of an
unordered operator given in Eq. (3.110),
b O
b
Tr (O
1 2) =

b O
b
d d e h|O
1 2 |i.

42

(4.4)

b operators in this equation the decomposition of

Next we insert between the two O
the identity given in Eq. (3.93)

I=

d d e |ih|,

(4.5)

so that one obtains,

b O
b
Tr (O

2)

b | i h |O
b | i.
d1 d1 d2 d2 e1 1 e2 2 h1 |O
1 2
2
2 1

(4.6)

Now, for each coherent-state matrix element above we use the D = 1 equivalent of
Eq (3.109),
b 0 i = e 0 O  ( , 0 ),
h|O|
(4.7)
which results in
b O
b
Tr (O
1 2) =

d1 d1 d2 d2 e1 1 e2 2 e1 2 e2 1 O1 (1 , 2 ) O2 (2 , 1 )

d1 d1 d2 d2 e1 1 e2 2 e1 2 e2 1 O1 (1 , 2 ) O2 (2 , 1 ).

(4.8)

Making the change of variables 1 2 and rearranging the integration measure,

one obtains the final result
Z

b O
b
Tr (O
1 2) =

d1 d1 d2 d2 e1 (1 +2 ) e2 (2 1 ) O1 (1 , 1 ) O2 (2 , 2 ).

(4.9)

It is not difficult to show that this result generalizes for the product of s operators
to
b
Tr (O

b
O

s)

d1 d1 ds ds e1 (1 +s ) e2 (2 1 ) es (s s1 )

O1 (1 , 1 ) O2 (2 , 2 ) Os (s , s ) .

(4.10)

Note that in the first exponential one has a positive sign on s . This is due to the
antiperiodic boundary condition of the endpoint Grassmann number in the trace.
The result coincides with the one given by Creutz in Ref. [34] Eq. (13) of this
reference.
Now, the result for a single-particle Hilbert space of arbitrary dimension D can
be easily generalized from the result for D = 1 in Eq. (4.10). For that matter,
one should notice that instead of only one we will have D 0 s. Therefore, one
can consider that each in the integral in Eq. (4.10) is actually a vector with D
components, so that i = (i,1 , i,2 , , i,D ), for i = 1, 2, , s, and the products
1 (1 + s ) and 2 (2 1 ), become
1 (1 + s ) =

D
X

1,l
(1,l + 2,l ),

l=1
D
X

i (i i1 ) =

l=1

43

i,l
(i,l i1,l ),

(4.11)
(4.12)

With i = 2, ..., s. The product of the exponentials can then be rewritten as a single
exponential with an exponent given by the sum
D X
s
X

(i,l i1,l ),
i,l

(4.13)

l=1 i=1

with the boundary condition

0,l = s,l .

(4.14)

b s is simply
With this, the final result for the trace of ()
b s
Tr ()

Z Y
D Y
s

di,k di,k

exp

k=1 i=1

"D s
XX

(i,l
i,l

i1,l )

l=1 i=1

b  ( , , )
b  ( , , ).

1,D
s,D
1,1
s,1

(4.15)

This is the final expression for the trace of ()s . In the next section we will obtain
the expression for the matrix elements  ( , , ) in terms of the Grassmann
numbers and .

4.2

Explicit expression for  ( , )

Evaluation of the multiple integral in Eq. (4.15) requires an explicit expression for
 ( , ) in terms of the the Grassmann numbers and . This can be obtained
b , a) is not already in
as follows. Let us start again with the D = 1 case. When (a
normal order, it can be put in such order through
b , a) =
(a

m
n

m,n (a ) (a) ,

(4.16)

m=0,1 n=0,1

b
where 
m,n can be obtained from the explicit representation of (a , a) for an
example, see Eqs. (3.49)-(3.51). Now, from Eq. (4.7)

b
 ( , ) = e h|O|i,

(4.17)

and inserting Eq. (4.16) there

 ( , ) = e

m
n

m,n h|(a ) (a) |i

m=0,1 n=0,1

= e

m
n

m,n ( ) () h|i.

(4.18)

m=0,1 n=0,1

Using the result given in Eq. (3.108), one has the final result
 ( , ) =

m=0,1 n=0,1

44

m
n

m,n ( ) () .

(4.19)

This can now be extended to a D-dimensional space. First, Eq. (4.16) becomes
b , a) =
(a

mD

(a1 )m1 (a1 )n1 (aD )nD .
m1 ,,mD ;n1 ,,nD (aD )

{m}=0,1 {n}=0,1

(4.20)
Therefore,

b  ( , , ,

1 , , D )

is given by

b  ( , , , , , ) =

D
1
D 1


m1 ,,mD ;n1 ,,nD

{m}=0,1 {n}=0,1
mD
(D
)
(1 )m1 (1 )n1 (D )nD .

4.3

(4.21)

Evaluation of the Grassmann integrals in Tr()s

The multidimensional Grassmann integral in Eq. (4.15), with the 

m1 ,,mD ;n1 ,,nD
given by Eq. (4.21), is not trivial when D > 1. However, it can be done more easily
by condensing the notation. Specifically, instead of using the two sets of indices
{k = 1, D} and {i = 1, , s}, Eq. (4.15) can be simplified by using a single set
of indices, {I = 1, , sD}, for each Grassmann number.
Let us consider first the example of D = 2. This corresponds to a system with
two degrees of freedom, e.g. k = 1 = and k = 2 =. For this case, we have that
the exponent in the integral is given by
D X
s
X

i,l
(i,l

i1,l ) =

l=1 i=1

s
X

i,
(i,

i1, ) +

s
X

i,
(i, i1, ). (4.22)

i=1

i=1

The sum over terms can be organized as follows:

s
X

i,
(i, i1, ) = + 1,
1, + 0 + + 0 + 1,
s,

i=1

1, + 2,
2, + 0 + + 0
2,

+ 0 + 0 + s,
s1, + s,
s,

s
X

i,
(A )i,j j,

i,j=1

s
X

i (A )i,j j ,

(4.23)

i,j=1

where we have changed the notation i, i , and the matrix A is, by inspection

45

of the sum above, given by

A =

..
.

1
0 0
1 1 0
0 1 1
..
.. ..
.
. .
0
0 0
0
0 0

0
0
0
..
.

1
0
0
..
.

1 0
1 1

elements

(4.24)

Note that we are using the transpose of the matrices used by Charret et al. in
Ref. [9]. It is clear that the matrix corresponding to the sum over terms, A , is
precisely the same as A , and the sum over terms is organized as follows:
s
X

i,
(i,

i1, ) =

i=1

s
X

i,

2s
X

(A )i,j j,

i=1

i (A )i,j j .

(4.25)

i,j=s+1

Moreover, the matrices A and A are null.

Therefore, if one considers a matrix of the form

A=

A
0

0
A

(4.26)

one can write Eq. (4.15) as

s

b =
Tr ()

Z Y
N

dI dI

exp

I=1

N X
N
X

AJK K  (1 , 1 )  (N
, N ) (4.27)

J=1 K=1

where N = sD. This integral requires further elaboration. Because the matrix A is
block diagonal, the integral factorizes in a product of 2 multidimensional integrals of
dimension s. The integrand contains the exponential factor and products of Grassmann numbers coming from the product of s operators  (1 , 2 , 1 , 2 ). Therefore,
in general, each of the s-dimensional integrals is a moment of a Grassmann Gaussian
integral of the form
M (L, K) =

Z Y
s
i=1

di di

exp

s X
s
X

A
JK

l1 k1 lm km ,

(4.28)

J=1 K=1

where L = l1 , , lm and K = k1 , , km , with m s. We also assume that the

product of and are ordered as l1 < l2 < < lm and k1 < k2 < < km .
Obtention of an explicit solution of an integral like Eq (4.28) is quite involved but
it is possible. In a very detailed and clear derivation, Charret, Correa Silva, Souza,
O. Rojas Santos, Thomaz, and Carneiro in Ref. [9] obtained the explicit expression
M (L, K) = (1)l1 ++lm (1)k1 ++km A(L, K),

46

(4.29)

where A(L, K) is a co-factor of the matrix A with respect to the rows l1 , , lm

and the columns k1 , , km the determinant of A with the rows l1 , , lm and
the columns k1 , , km removed.
For a system with D degrees of freedom, the dimension of the matrix A is N = sD
and its the structure remains block-diagonal, where each matrix on the diagonal is
equal to the matrices A and A instead of the numbers 1 and 0, one actually
have the N N identity and null matrices. In addition, the integrals necessary to
b s , Eq. (4.27), have as solutions Eqs. (4.28) and (4.29), the
evaluate the trace of ()
only difference as compared to the D = 2 case, is that being that now there are D
of such integrals.
Despite of the closed form of the result of the multidimensional Grassmann
integral, explicit evaluation of the remaining determinants is still a formidable task.
Fortunately, Charret, Correa Silva, Souza, Rojas Santos and Thomaz [13] were able
to diagonalize the A through a similarity transformation, exploiting the blockstructure of these matrices. This was a tremendous achievement, in that it opens
the possibility for explicit evaluation of traces of products of an arbitrary number
of normal-ordered fermionic operators without resorting to large scale computer
evaluations.
This completes the review on the Niteroi method. As already commented, the
highlight of the method is its ability of calculating the partition function of a manyfermion system at finite temperature and chemical potential in closed form in the
high temperature limit, i.e. in principle each term of the series expansion in ,
Eq. (4.2), is known explicitly. Of course, the evaluation of each term becomes
complicated for high powers s and therefore it will be useful for temperatures
sufficiently high such that the series can be truncated at some low value of s. On
b in the Boltzmann factor, Eq. (4.1), is quadratic
the other hand, when the operator
in the field operators, one can re-sum the series expansion in and so the validity
of the result is not restricted to high temperatures.
b represent essentially noninteracting partiOne might think that quadratic s
cles, since interaction terms involve at least four field operators. However, this
is not the case, since many approximate, nonperturbative methods in many-body
b that is quadratic in the field operators
physics have as starting point an ansatz for
and corrections to the ansatz are then calculated perturbatively. Examples include
Hartree-Fock and BCS-type of methods, optimized perturbation theory, etc. Another important example is the use of the Hubbard-Stratonovich [8] transformation
which, with the introduction of auxiliary bosonic fields, allows us to express a fourfermion interaction into a quadratic Hamiltonian. Therefore, once one calculates
the trace over the fermionic fields, there remains a trace over bosonic fields that

47

can be calculated perturbatively or numerically. Moreover, one ab initio method

of approaching many-body problems is through Monte Carlo simulations on a Euclidean space-time lattice. However, such methods lose accuracy for large values of
the chemical potential because of the non-positivity of the Euclidean action, commented in the Introduction. The Niteroi method, on the other hand, can in principle
evade this problem when used in connection with some approximation scheme.
In the next Chapter we will discuss some simple applications of the Niteroi
method. In addition, we will indicate further, novel applications of the method to
many systems of modern interest.

48

Captulo 5
Applications to simple problems and possible
extensions

In the present Chapter we present applications of the Niteroi method to simple

problems. Initially, in the next Section, we consider the trivial, but illustrative case
of the free Fermi gas. Next, in Section 5.2 we consider mean field type of approximations to the interacting non-relativistic Fermi gas. Here the aim is to reobtain the
well known results of the traditional Hartree-Fock and BCS approximations [1] [2].
Here we make use of the Bogoliubov-Valatin [36] canonical transformation to bring
the Hamiltonian to a diagonal form such that the Niteroi method can be applied
directly and trivially. In Section 5.3 we discuss how to obtain the well known results of perturbation theory on the top of the mean field approximation. All these
results are not new and actually can be obtained with other methods. It should
be understood that the aim here is to show that the method is reliable and should
be useful in other contexts. With this in mind, in Section 5.4 we propose to use
the Niteroi method in connection with the Hubbard-Stratonovich [8] transformation
to implement high order optimized perturbation theory [14]-[17] to improve on the
mean field approximation. And finally, we discuss the application of the method to
the lattice formulation of field theories.

5.1

Non-interacting non-relativistic Fermi gas

Here we consider the problem of calculating the grand-canonical partition function of

the nonrelativistic Fermi gas. We shall initially consider the traditional calculation
of direct evaluation by means of the second quantization formalism. Then we will
reobtain the result using the Niteroi method just discussed above.
The Hamiltonian for a system of noninteracting nonrelativistic fermions is given

49

in terms of field operators (x) and (x) as

c=
H

XZ

d3 x (x)

1 2
(x),
2m x


(5.1)

where =, . Here and elsewhere we use h

= 1. The field operators satisfy the
usual anticommutation relations
{ (x), 0 (x0 )} = 0 (x0 x),
{ (x), 0 (x0 )} = 0,

{ (x), 0 (x0 )} = 0.

(5.2)

It is convenient to work in momentum space, so we perform a Fourier decomposition

of the fields in a large volume V of the space
1 X

a (k) eikx ,
V 1/2 k
1 X
a (k) eikx .
(x) =
V 1/2 k
(x) =

(5.3)
(5.4)

Using Eq. (5.2), one can easily shown that a (k) and a (k) satisfy the anticommutation relations
{a (k), a0 (k0 )} = 0 (k0 k),
{a (k), a0 (k0 )} = 0,

{a (x), a0 (k0 )} = 0.

(5.5)

Replacing Eqs. (5.4) in the Hamiltonian, one obtains

c=
H

XX

k2
n
(k),
2m

(5.6)

where
n
(k) = a (k) a (k).

(5.7)

c,
The number operator N
c=
N

XZ

d3 x (x) (x),

(5.8)

in momentum space is given by

c=
N

XX

n
(k).

(5.9)

c and N
c, the operator
b in momentum space can be written as
Given H
b =

b =

b (k) =

XX

k,

50

(k) n
(k),

(5.10)

where, for later convenience, we have defined the quantities

b =

b (k),

b (k) = (k) n

(k),

(k) =

k2
.
2m

(5.11)

The dependence on in allows the possibility for different numbers of particles

with spin and in the system. The grand canonical partition function is given by
Z = Tr e = Tr e
b

= Tr e( + ) .
b

(5.12)

b ,
b ] = 0, the partition function is given by the product
Since [

Z = Tr (e ) Tr (e ).

(5.13)

This can be easily shown using as basis the eigenstates {|n (k)i} of the operator
n
(k)
n
(k)|n (k)i = n (k)|n (k)i,

with

n (k) = 0, 1.

(5.14)

Moreover, using this same basis, one can also show that
Tr e = Tr e
b

(k)b
n (k)

Tr e (k)bn (k) ,

(5.15)

and a similar expression for Tr e b .

One can calculate the traces above directly using the basis of Eq. (5.14) as
Tr e =
b

Tr e (k)bn (k)

(5.16)

e (k)n (k) =

k n (k)=0,1

Y

1 + e (k) ,

(5.17)

replacing in the partition

Z =

Y

1 + e (k1 )

k1

 Y

1 + e (k2 ) .

(5.18)

k2

This is the final result for the grand-canonical partition function for a polarized (i.e.
different number of particles with spin and ) noninteracting, nonrelativistic Fermi
gas. When = = , we have have (k) = (k) = k2 /2m and obtain the
well-known result for the partition function of a free Fermi gas
Z=

(
Yh

1+e

(k2 /2m)

51

)
i 2

(5.19)

Next, we reobtain the above results using the Niteroi method. We start from
the general result of Eq. (5.16) and expand the exponential in powers of
Tr e =

Tr e (k)bn (k) =

Tr

[ (k)]s
b (k)]s
[n
s!
s=0

YX

[ (k)]s
b (k)]s
Tr [n
s!
s=0

[ (k)]s
b (k)]s .
Tr [n
Tr (1) +
s!
s=1

(
Y

(5.20)

The trace of 1 is simply

Tr (1) =

dd exp (2 ) = 2.

(5.21)

b (k)]s is to be calculated using Eq. (5.22). We note that since the trace
The trace of [n
is calculated for k held fixed, we can simply ignore k in the Grassmann numbers
when applying Eq. (5.22). Specifically,
b ]
Tr [n

Z Y
s

dI dI

s X
s
X

exp

I=1

Z Y
s

A
JK

A
JK

K n
(1 , 1 ) n (s , s )

J=1 K=1

dI dI

s X
s
X

exp

I=1

K 1 1 s s ,

(5.22)

J=1 K=1

where we have used the fact that n

( , ) = . The value of the integral is given
by Eq. (5.23), that for the present case is given by

M (, K) = (1)1++s (1)1++1 A(L, K).

(5.23)

Removing the first s lines and s first columns from the matrix A , one obtains that
the determinant of the remaining matrix is simply 1, and so
T r e (k) = 2 +
b

[ (k)]s
s!
s=1

= 1 + e (k) .

(5.24)

Therefore, one obtains precisely the same value for the partition function as calculated directly.
b one can obtain the trace of e directly from
Note that for a quadratic
b =
Tr O

dd e2 O ( , ).

This will be clear in the discussions of the next Section.

52

(5.25)

5.2

Interacting nonrelativistic Fermions, canonical transformations

In this Section we will discuss the application of the Niteroi method to an interacting
many-fermion system. However, our aim here is not to use the high temperature expansion for the entire interacting Hamiltonian. Rather, our aim is to use the method
in conjunction with some nonperturbative approximation schemes. In particular,
here we will use canonical transformations to obtain the traditional mean-field approximations and to improve on such mean-field approximations within the Niteroi
method. We remark that we are not going to obtain new results, rather we will show
how to obtain known results within the Niteroi method and indicate future directions
for obtaining new results. The point of showing this is that in some circumstances
when dealing with many-fermion systems, e.g. in a lattice formulation of relativistic
field theory, as commented earlier, the traditional Monte Carlo method cannot be
applied because the Euclidean action is not real and positive. Therefore, having an
approximation scheme that does not suffer from such a problem is welcomed.
Initially we will start reobtaining the traditional finite-temperature BardeenCooper-Schrieffer (BCS) gap equation [35] for a many-fermion system. This will
be obtained employing a Bogoliubov-Valatin canonical transformation [36]. The
quadratic part of the resulting Hamiltonian can then be diagonalized and used to
obtain the grand canonical partition function. Certainly there is nothing new here,
however, this provides a stating point for us to discuss possible ways on how to
improve on this in the context of the Niteroi method. Notation and the BogoliubovValatin transformation we are going to use here is heavily based on Section 37 of
the book of Fetter and Walecka [2].
When considering interactions, the most general, nonrelativistic Hamiltonian for
a system of fermions interacting through a local two-body potential Vb can be written
in terms of the field operators (x) and (x) as
c = H
c +V
H
0

XZ

d3 x (x)

1 2
(x)
2m x


1 Z 3 3
+
d xd y 1 (x) 2 (y) V1 2 ,3 4 (x y) 4 (y)3 (x),
2

(5.26)

where V1 2 ,3 ,4 (xy) = V3 4 ,1 2 (yx) for identical particles. For the purposes of

simplifying the notation and explaining the main ideas, we will take an interaction
of the form [38]
V1 2 ,3 4 (x y) = 1 3 2 4 1 2 V (x y).

53

(5.27)

As in the noninteracting case, we perform a Fourier transformation and obtain for

b the following expression
the grand canonical operator
b

XX

k2

2m

n
(k)

X
1 X
hk1 k2 |V |k3 k4 i a (k1 ) a (k2 ) a (k2 ) a (k4 ),
2 1 2 k1 +k2 =k3 +k4

(5.28)

where
1 Z 3 3 i(k1 x+k2 y)
d xd y e
V (x y) ei(k3 x+k4 y) .
(5.29)
V2
Next, we make a canonical transformation on this Hamiltonian by introducing
new creation and annihilation operators as [2] [37]
hk1 k2 |V |k3 k4 i =

b1 (k) = u(k) a (k) v(k) a (k),

b2 (k) = u(k) a (k) + v(k) a (k),

(5.30)

where u(k) and v(k) are c-number functions to be determined, such that
{b1 (k), b1 (k0 )} = (k0 k),
{b2 (k), b2 (k0 )} = (k0 k),

(5.31)

and all other anticommutators are zero. This leads to the condition that
u2 (k) + v 2 (k) = 1.

(5.32)

This condition is not sufficient to determine u and v and more input is needed. This
will be provided by the imposition that part of the resulting Hamiltonian should be
diagonal, which results in addition into a gap equation.
The canonical transformation proceeds in replacing the original operators a (k)
and a (k) in favor of the b1 (k), b1 (k), b2 and b2 in the Hamiltonian, and then
arranging them in normal ordering (i.e. putting the resulting Hamiltonian in simple
order). The idea behind this procedure is that such a transformed Hamiltonian can
be brought to a diagonal form, plus an interaction term that should be small. The
meaning of small is in the sense that the main features of the original Hamiltonian
are captured by the diagonal part of the new Hamiltonian. There are different ways
to implement this and as mentioned by Fetter and Walecka [2] the easiest way is to
use Wicks theorem. We will not review this here, simply quote the final result (we
use = )
b = U+

E(k) [b1 (k)b1 (k) + b2 (k)b2 (k)] + : Vb (b) :

F (k) [b2 (k)b2 (k) + b1 (k)b2 (k)],

54

(5.33)

where U is a constant (not relevant for our purposes here), and

E(k) = [u2 (k) v 2 (k)](k) + 2u(k)v(k)(k),

(5.34)

F (k) = 2u(k)v(k)(k) [u2 (k) v 2 (k)](k),

(5.35)

with
X
k2
+ hkk0 |V |kk0 iv 2 (k 0 ) [1 2f (k0 )] ,
(k) =
2m
kk0
X
1
hk k|V |k0 k0 i u(k 0 )v(k 0 ) [1 2f (k0 )] ,
(k) =
2 k0

(5.36)

where hkk0 |V |kk0 i = hkk0 |V |kk0 ihkk0 |V |k0 ki+hkk0 |V |kk0 i gives the HartreeFock mean field contribution to the energy, and f (k) is the thermal average of b b,


f (k) = hbi (k)bi (k)i = Tr bi (k)bi (k) e ,

b

i = 1, 2.

(5.37)

In addition, : Vb (b) : is obtained from the original Vb in Eq. (5.26) by replacing the
a and a with
a (k) = u(k) b1 (k) + v(k) b2 (k),
a (k) = u(k) b2 (k) v(k) b1 (k),

(5.38)

and then normal ordering the resulting expression.

b neglecting : Vb (b) : on the assumption that the
The idea now is to diagonalize ,
main effect of the interaction has been captured by the diagonal part, as discussed
above. This is achieved by demanding that the anomalous Bogoliubov term be equal
to zero, i.e.
F (k) = 0 [u2 (k) v 2 (k)] (k) = 2u(k)v(k)(k).
(5.39)
Note that this equation has to be solved under the constraint of Eq. (5.32). As
shown explicitly in Fetter and Walecka [2], this leads to
i1/2

E(k) = 2 (k) + 2 (k)

(5.40)
"

(k)
1
(k)
, u2 (k) =
1+
,
2E(k)
2
E(k)
"
#
1
(k)
v 2 (k) =
1
,
2
E(k)
u(k)v(k) =

(5.41)

and (k) satisfies the gap equation

(k) =

1X
(k 0 )
hk k0 |V |k0 k0 i
[1 2f (k 0 )] .
2 k0
E(k 0 )
55

(5.42)

All thermodynamic properties of the system can then be obtained from the
b which
partition function calculated with the quadratic grand potential operator ,
can be calculated analytically, given by
Z=

"
Y

E(k)

1+e

#
 2

(5.43)

Eq. (5.42) always admits a trivial solution, i.e. (k) = 0. In this case one would have
then the familiar Hartree-Fock approximation, which still requires a self-consistent
solution, since the thermal average f (k) appears in in (k), see Eq. (5.36).
The central question now is how to include the effects of the neglected term,
b
: V (b) :, the normal ordered potential. Of course, in case these effects are small,
but not entirely negligible, perturbation theory would be the natural approach ro
pursue. In the next section we discuss this issue.

5.3

Perturbation on the mean field

Let us start reobtaining the traditional perturbative improvement on the mean field
b of Eq. (5.33) (without the irrelevant constant U and with
result. We rewrite the
the anomalous Bogoliubov term already made zero) as
b =
b + : Vb :,

(5.44)

where
b =

E(k) [b1 (k)b1 (k) + b2 (k)b2 (k)],

(5.45)

and where : Vb : contains all possible normal ordered products of two creation operators b1 and b2 , and two annihilation operators b1 and b2 . The parameter is just
for bookkeeping purposes, it will be taken = 1 at the end. The partition function
is then
Z = Tr e(0 + :V :) .
b

(5.46)

b and : Vb : do not
The idea is to expand the exponent in powers of . Since
0
commute, the expansion is not straightforward. However, it is well known how such
an expansion can be made, by going to the interaction representation, as explained
in Refs. [1][2]. Specifically, the expansion is (see Section 24 of Fetter and Walecka,
Ref. [2] )

Z = Tr e(0 + :V :)
b

56

"

= Tr e

b0

Z 1
Z n1
()n Z
1+
d1
d2
dn
n!
0
0
0
n=1

Vb (1 ) Vb (2 ) Vb (n )
Z 1
Z n1
()n Z
d1
d2
dn
= Tr 1 +
n!
0
0
0
n=1

"

Vb (

) Vb (

2)

Vb (

e 0

n)

Tr P eV () e 0 = Tr Vb e 0 ,
b

(5.47)

where P denotes path ordering in , the ordered chain of integrals in 1 , , n ,

and
Vb () = e0 : Vb : e0 .
b

(5.48)

We want to use the coherent-state representation of the trace. First, let us simplify
notation writing
b =

E(k) [b1 (k)b1 (k) + b2 (k)b2 (k)]

Ei bi bi ,

(5.49)

where we have condensed the indices k and 1 and 2 into i. Next, we use Eq. (4.9)
to write the trace of the product of the two operators Vb and Ub e b0 as
Tr (Vb e 0 ) =

dd d 0 d 0 e

dd e V  ( , )

d 0 d 0 e

(+ 0 )

0 ( 0 )

V  ( , ) U  ( 0 , 0 )

0 ( 0 )+ 0

U  ( 0 , 0 ),

(5.50)

where it should be clear here that , 0 , refer to sets {i , i0 , }, where the index
P
i stands for k and 1 and 2, and products like 0 ( 0 ) mean i 0 i (i0 i ). Let us
use source terms for the first integral above

V  ( , ) = V  ( , ) e(Ji i +i Ji ) 

(5.51)

J,J=0

where Ji and Ji are Grassmann numbers. Then

V  ( , ) = V 

,
Ji Ji

57

(Ji i +i Ji ) 

J,J=0

(5.52)

where /Ji and / Ji are Grassmann functional derivatives see Eq. (3.153). Then,
the expression for the trace can be written as
b
Tr (Vb e 0 )

= V
Z

,
Ji Ji

d 0 d 0 e

!Z

dd e

Ji i +i Ji 

J,J=0

0 ( 0 )+ 0

U  ( 0 , 0 ).

(5.53)

The remaining integrals can be integrated analytically. First, let us perform the
integral over 0 and 0 . The ordered form of the operator Ub = e b0 can be
obtained as [34]
e 0 = e
b

P
i

Ei bi bi

e Ei bi bi

Y
i

Yh

1 + e Ei 1 bi bi =

: e(e

Ei 1)b b
i i

"

= : exp

X

bi bi

"
X

Ei

Ei

:,

(5.54)

and U  ( , ) is given by


U ( , ) = exp

i i

(5.55)

Therefore, and using Eq. (3.152) in the second line, one has
Z

d 0 d 0 e

0 ( 0 )+ 0

U  ( 0 , 0 ) =

=
=

d 0 d 0 e
Z

0 D 1 0 + 0 ()+ 0

d 0 d 0 e

0 D 1 0 + 0 + 0

e D ,
Det D

(5.56)

where D is the diagonal matrix

Dij = eEi ij .

(5.57)

Inserting this result into the integral over and and using Eq. (3.152) again, one
obtains
1 Z
1 Z

Ji i +i Ji D
dd e e
dd e (1+D)+Ji i +i Ji
e
=

Det D
Det D
Det [(1 + D)] J (1+D)1 J
e
. (5.58)
Det D
With this, we arrived at the well known result for the perturbative expansion of the
partition function
=

= V
,
Z0
Ji Ji

58

P
i,j

Ji (1+D)1
i,j Ji

(5.59)

Plain perturbation theory would be simply to calculate physical quantities using

the partition function in Eq. (5.59) with D = eEi where Ei the energy calculated as
in the previous section. A better perturbation theory corresponds to recalculate the
energy in a self-consistent way from the perturbed partition function, Eq. (5.59) i.e.
D is to be determined from the partition function defined in terms of the unknown
D. Of course, in any practical sense, one would be able to do so by restricting the
expansion in Eq. (5.47) to lowest orders in .

5.4

Optimized perturbation theory

At this point it should be clear that one does not need the Niteroi method to do
perturbation theory on a mean field approximation, since Eq. (5.59) is known from
other techniques. But, having shown how to derive the traditional perturbative series, we will make use of part of the material above to motivate a method that might
be more useful for practical results. The method we propose makes interchangeable
use of the path integral method and the direct calculation of the trace by the Niteroi
method.
The basic idea of optimized perturbation theory in one of its simplest forms
known as the optimized linear expansion is the following [14]-[17]. Suppose the
Hamiltonian of the system is of the generic form
c=K
c + Vb
H

(5.60)

c is quadratic in the field operators and Vb contains more than two field
where K
operators. Then, one adds and subtracts a term for the Hamiltonian in the form


c= K
c + O(p
b
b
b
H
1 , p2 , ) + V O(p1 , ps , ) ,

(5.61)

b
where O(p
1 , p2 , ) is given in terms of field operators, p1 , p2 , are functions of
the coordinates in general, and is a dimensionless parameter that is introduced for
bookkeeping purposes and is to be taken equal to 1 at the end of calculation. The
b judiciously so that the theory with K
c+O
b is exactly soluble,
idea is to choose O
and corrections to this solution are calculated in perturbation theory in . If one
is able to calculate in all orders of , we of course would have obtained the exact
solution and the result would be independent of p1 , p2 , . Since we are going to
truncate the expansion in some power of , the results do depend on p1 , p2 , . In
order to minimize the sensitivity of the results on V  , we use the principle of minimal
sensitivity (PMS) [14], in which one requires that a physical quantity Q(p1 , p2 , )
satisfy
Q(p1 , ps , )
= 0.
(5.62)
pi

59

This physical quantity can be the grand canonical potential,

e = Z = 1 ln Z.

(5.63)

In order to explain the approach we propose in the perspective of the Fermi

b the following
gases in the unitarity limit discussed in the Introduction, we use for
expression [38]
b

XZ

x (x) K(x) (x)

d3 x (x) (x) (x) (x),

(5.64)

where

1 2
,
(5.65)
2m x
and g is the coupling strength the minus sign is used for later convenience. The idea
behind of the Hubbard-Stratonovich is to linearize the interaction term through
the introduction of auxiliary fields. This is done making use of the identity
K (x) =

1 Z
2
2

d e /2a = ea /2 .
2

(5.66)

In the present case, a would stand for the pairs of field operators (x) (x),
(x) (x), (x) (x) and (x) (x). For each of these pairings, one would introduce one auxiliary field (x). There is, however, one slight complication because we
b given in Eq. (5.64), because
cannot use directly the formula of Eq. (5.66) with the
the kinetic and the interaction terms of this operator do not commute. One way to
proceed is to change to the interaction representation as done in Section 5.3 that
effectively factorizes the exponential into two exponentials in which one of them is
path ordered. For our purposes here however, and for clarity of presentation, the
easiest way to proceed is like Hubbard proceeded in Ref. [8], namely he used the
ordering label technique of Feynman [39]. This is simply a simplification on the notation, the final result is not different from the traditional path-ordering used above
in Appendix F we have reviewed this technique. To further simplify the discussion
and for better exposition, we will perform the Hubbard-Stratonovich transformation
using only the diagonal pairings, i.e. we will linearize the interaction using only
the (x) (x) and (x) (x) pairs. In addition, we will take = = . This
will simplify our discussion because we will need only one auxiliary field. In this
case, the partition function can be written as
Z = Tr e
b

= Tr exp

Z
0

ds

"
XZ

dx (x, s) K(x, s) (x, s)

dx (x, s) (x, s) (x, s) (x, s)

60

(5.67)

where now the kinetic and potential terms in the exponential can be taken as commuting the integral over s is due to Feynman ordering technique, see Appendix F.
Now we introduce an auxiliary field through the Hubbard-Stratonovich transformation as

Z =

"

1Z Z
dx 2 (x, s)
[D(x, s)] exp
2 0

Tr exp

ds

(
X

dx

+ g

1/2

(x, s) K(x, s) (x, s)

(x, s)

(x, s) (x, s)

(x, s) (x, s)

)
i

(5.68)

At this point, since the exponential in the trace is diagonal in the Fermi fields,
the trace can be evaluated exactly. The result of the trace would be the square of
the determinant of the matrix K + g 1/2 and what remains is an integral over the
auxiliary field . Had we used 6= , one would have obtained the product of two
determinants. It is here where the sign problem enters, since it is not guaranteed
that the product of the determinants is positive. Our aim is not to proceed this
route, we want to implement a mean field approximation and to use optimized
perturbation theory on the top of the mean field, as explained in the Introduction.
Why one would give up the possibility of obtaining an exact result in favor of an
approximate scheme? First, as said previously, it is important to understand how
corrections affect the zeroth-order mean field results. Second, the exact solution can
be very involved due to the sign problem and also requires intensive use of Monte
Carlo methods. It is hoped that through an expansion in a modified interaction one
can capture most of the physics relevant to the problem and that a milder or even
no sign problem arises of course this we will only know with explicit calculations.
We implement the step indicated in Eq. (5.61) using the following identity
(x) (x) (x) (x) = 20 (x) + 0 (x) (x) (x) + 0 (x) (x) (x)
+

ih

(x) (x) 0 (x) (x) (x) 0 (x) ,

(5.69)

where the c-number 0 (x) is the mean field. Now we replace this into the expression
for Z in the following way, already using the Hubbard-Stratonovich transformation
of the four-field term
g

Z(0 ) = e

Tr e

dx 20 (x)

R
0

ds

[D(x, s)] e

dx K0 (x,s,0 )

61

R
0

21

ds

R
0

ds

dx 2 (x,s)

dx V(x,s,0 )

(5.70)

where
K0 (x, s, 0 ) =

(x, s) [K(x, s) g 0 (x)] (x, s),

(5.71)

and
h

ih

V(x, s, 0 ) = g 1/2 (x, s) g 1/2 0 (x) (x, s) (x, s) + (x, s) (x, s)

+ g 20 (x).

(5.72)

Up to here, there have been no approximations and all expressions are exact. The
implementation of the optimized perturbation theory proceeds in the following way.
One expands the second exponential in the trace in Eq. (5.70) up some order of ,
perform the trace using the Niteroi method, then perform the path integral over
, and finally use the principle of minimal sensitivity to determine 0 . Since both
exponentials commute due to the Feynman ordering notation, and since both exponentials are quadratic, the trace should be easily evaluated.
One could think that it would be easier to evaluate the complete trace and then
expand the resulting determinant in powers of . Obviously this can be done in
some cases, but the Niteroi method is of general applicability and would not require
the expansion of a complicated determinant.
Let us examine the O( 0 ) approximation to the partition function. In this case,
one would have simply
Z0 (0 ) e g

Tr e

dx 20 (x)

R
0

ds

[D(x, s)] e 2

R
0

ds

dx 2 (x,s)

dx K0 (x,s,0 )

(5.73)

Since the trace does not depend upon , the functional integral gives simply 1. Also,
the Feynman ordering becomes is trivial, and one would get for Z0 (0 )
Z0 (0 ) e g

dx 20 (x)

Tr e

dx K(x,0 )

(5.74)

The mean field 0 is determined from the grand canonical potential

(0 ) = 1 ln Z0 (0 )
= g

dx 20 (x) 1 ln Tr e

dx K(x,0 )

(5.75)

from the PMS condition of Eq. (5.62)

(0 )
= 0.
0 (x)

(5.76)

Applying this to the result in Eq. (5.75), one obtains

2 g 0 (x)

R
X
1
g Tr e dx K(x,0 )
(x) (x) = 0.
Z0 (0 )

62

(5.77)

That is, the mean field is determined by the gap equation

0 (x) = h (x) (x)i0 = h (x) (x)i0 .

(5.78)

This is a self-consistent equation for 0 .

The result is reassuring, in that it is the traditional mean field (Hartree) result
and it has been obtained in an elegant and relatively easy way. As explained,
corrections to this mean field solution are obtained by expanding in powers of
the complete partition function. We will not proceed to evaluate such corrections,
since the aim here was to set up the approach and an explicit evaluation of high
order corrections would require an specific model. In addition, for higher order
corrections, the functional integral over and Feynman ordering are not trivial as
for the O( 0 ). The evaluation of the functional integral most likely will require a
Monte Carlo method when working at high orders in an interesting question is to
see whether the sign problem is as severe as in the exact evaluation. Moreover, the
point-like interaction used here requires renormalization and further elaboration is
need to obtain numerical results. All this would extrapolate the scope of the present
dissertation and therefore we leave these issues for future work.
We finalize this Chapter mentioning that pairings of the type (x) (x) and
(x) (x) can be incorporated by introducing extra mean fields and extra HubbardStratonivich auxiliary fields. Although straightforward, the algebra is a little heavier
and more gap equations through the PMS condition would arise. Proceeding as
above, the traditional BCS solution would follow form the O( 0 ) result. Corrections
would then be calculated in the same way as explained above.

63

Captulo 6
Conclusions and perspectives

The primary aim of the present dissertation was to present a review on the use of
coherent states in the evaluation of the quantum grand canonical partition function
of spin-1/2 fermions at finite temperature. The main motivation for studying such
systems is due to the recent experimental developments in the area of fermionic
atoms. These developments have created great excitement in the Physics community
in view of the possibility of exploring and manipulating matter composed of particles
with no classical analogue. Contrary to bosons, fermions cannot be described in
terms of the dynamical variables like position and momentum, they require new
dynamical variables that are not familiar to the common sense, like Grassmann
variables. In addition, fermionic systems at the unitarity limit when the scattering
length characterizing the interaction strength is much larger than the inter-particle
spacing are encountered in different fields of physics, like in nuclear physics, astroparticle physics and condensed matter physics. In view of this, this is a fascinating
subject and it is important to develop mathematical methods to study such systems
in different contexts.
The fundamental quantity in the mathematical treatment of many-body systems
at finite temperature is the grand canonical partition function. It can very seldom
be calculated in closed form and approximation schemes and numerical methods
have been developed to study this quantity. In particular, there is great activity
in the field of Monte Carlo simulations of the path integral representation of the
partition function. Path integral formulations of fermion fields involve the use of
anti-commuting Grassmann variables but tricks need to be employed for their numerical evaluation, since Grassmann numbers cannot be generated directly in a
computer. Problems arise in implementing many of such tricks, like the sign problem, which arises when one formally integrates over such variables and re-express
the result in terms of path integrals over auxiliary fields.
In the present dissertation we have concentrated on an alternative to the path

64

integral formulation of the grand canonical partition function, namely the Niteroi
Method. This method is based on the direct evaluation of the trace over Grassmann variables using a high temperature expansion of the Boltzmann factor in
the partition function. The method makes use of the coherent-state representation
of the trace, and each term of the expansion is evaluated exactly exploiting the
anti-commuting nature of the Grassmann numbers. A good part of the present
dissertation has concentrated on the review of this novel method, the main results
are collected in Chapter 4. This was done after a brief discussion on the second
quantization formalism for many-particle systems and the Feynman path integral
in Chapter 2, and a short review in Chapter 3 on the use of coherent states for
calculating traces over fermionic variables and have also discussed the path integral
representation of the partition function using coherent states.
In addition to reviewing the Niteroi method, we indicated further developments
beyond the high temperature expansion of the Boltzmann factor. In particular we
make the case for using the method in the context of improving mean field type of
approximations through the combined use of the Hubbard-Stratonovich transformation and the ideas of optimized perturbation theory. Our approach starts from a
mean field type of approximation and then corrections are implemented making use
of an expansion in powers of a modified interaction, where the effects of the mean
field have been subtracted from the original interaction. In Chapter 5, initially we
have shown that known mean field type of approximations can be obtained trivially
within the Niteoi method, and have also shown that one can reproduce standard formulas for perturbative corrections to the mean field approximations within the same
method. Since perturbative corrections to mean field approximations become very
involved when higher order corrections are needed, we have proposed an alternative
so that high order corrections can be calculated in the context of an optimized perturbation theory, in that terms are added and subtracted from the original Hamiltonian
and rearranged in way that a remaining interaction can be expanded perturbatively.
The results are optimized in the sense that parameters introduced via the new terms
in the Hamiltonian are fixed through a principle of minimal sensitivity, i.e. they are
fixed by requiring that the grand canonical potential is stationary with respect to
variations in these parameters.
As perspectives, we envisage application of the proposed method in different
fields. One immediate application is in the field of atomic fermionic gases, in which
a contemporary goal is to go beyond the framework of mean field physics to access
manifestations of strong interactions and correlations. Work in this direction has
been started. We also believe that our proposed method can be extended to strong
coupling lattice QCD. The strong coupling expansion of the QCD action resembles

65

in many respects the high temperature expansion and as such the Niteroi method
should be of direct applicability. Work in this direction is underway.

66

Ap
endice A
Two-body operators - Change of representation

b is diagonal in the basis

Let a two body operator, see (2.37), such that every O
ij
{|k1 kN )}
b | ) = O
O
ij k1
kN
ki kj |k1 kN ),

(A.1)

b are equal, but

and as we are going to consider identical particles again, all the O
ij
acting specifically in its respective i, j-subspace.
On a general element of the symmetric basis
b
O|
k1 kN } =

=
=
=
=
=

b 1 P P |
O
ij
P k1 i |P kN i
N! P
i,j=1(i6=j)
N
1 P P P
b |

O
ij P k1 i |P kN i
P
N!
(i6=j)
N
1 P P P

OP ki P kj |P k1 i |P kN i
N! P
(i6=j)
D
1 P P P

(nl nm lm nl ) Olm |P k1 i |P kN i
l,m=1
N! P
D
1 P P P
b l n
b m lm n
b l ) Olm |P k1 i |P kN i

(n
l,m=1
N! P
N
P

D
P
l,m=1

b l n
b m lm n
b l ) |k1 kN }.
Olm (n

(A.2)

By linear independence
b =
O

D
P
l,m=1

D
P
l,m=1

D
P
l,m=1

D
P
l,m=1

Olm al al am am lm al al


Olm al al am lm am


Olm al am al am
Olm al am am al .

67

(A.3)

Now to the M -dimensional basis {|i i}

b =
O

D
P

b | )a a a a
(p q |O
ij l m p q m l

l,m,p,q=1

D
P

(p q |

l,m,p,q=1

D0
P
r,s=1
D0
P
t,u=1

D0
P
D0
P

b
|r s )(r s | O
ij
!

|t u )(t u | |l m )ap aq am al

b | )
(r s |O
ij t u

r,s,t,u=1

D
P

(p q |r s )ap aq

p,q=1

b | )a a a a .
(r s |O
ij t u r s u t

r,s,t,u=1

68

D
P

(t u |l m )am al

l,m=1

(A.4)

Ap
endice B
Evaluation of U (rn+1, ; rn, 0)

In the present Appendix we evaluate explicitly the matrix element U (rn+1 , ; rn , 0)

that we used in Sectiion 3.1.
Let us start from
b2
i p
+ V (br) |rn i
(B.1)
U (rn+1 , ; rn , 0) = hrn+1 | exp
h
2m
!#
"
Z
b2
d3 pn
i p
=
+ V (br) |rn i.(B.2)
hrn+1 |pn ihpn | exp
h
2m
(2h)3/2
!#

"

We expand the exponential as

e

b h
iH/

1
i
=

j=0 j!


j

!j

b2
p
+ V (br)
2m

(B.3)

b2
i p
+ V (br)
= I
h
2m
!
b4
b2
b2
2
p
p
p
2

+
V (br) + V (br)
+ V (br) + .
2m
2
h2 4m2 2m
!

(B.4)

Next, we order the operators in a symmetric way



b h
iH/

(Ord)

1
i
=

j=0 j!


j X
j

j!
k=0 k!(j k)!

b2
p
2m

!jk

[V (br)]k

b2
i p
= I
+ V (br)
h
2m
!
b4
b2
p
p
2
2
+2

V (br) + V (br) + .
2m
2h2 4m2

(B.5)

(B.6)

Therefore, one has that

e

b h
iH/

= e

b h
iH/

and neglecting terms of O(2 ), one obtains

69

(Ord)

+ O(2 ).

(B.7)

U (rn+1 , ; rn , 0) '
'

d3 pn
3/2

(2
h)
Z
3
d pn

b (Ord) /h

hrn+1 |pn ihpn |eiH

|rn i

(B.8)

eirn+1 pn /h h iH (Ord) (pn ,rn )/h i

e
,
(2
h)3/2 (2h)3/2

(B.9)

where


eiH(pn ,rn )/h

(Ord)

= eiH(pn ,rn )/h .

(B.10)

Then, finally
U (rn+1 , ; rn , 0) '

d3 pn
irn+1 pn i

3 exp
h

(2
h)
"

70

p2n
+ V (rn )
2m

!#

(B.11)

Ap
endice C
Closure for bosonic coherent states

In the present Appendix we show the correctness of Eq. (3.36). We start from the
expression given in the main text,
!
D
D
1 Z Y
P

d di exp
j j |ih|
1 =
N i=1 i
j=1
!
D
D
1 Z Y
P

j j
=
d di exp
N i=1 i
j=1
ni

D
Y
X

n 0

(i )
q
|n1 nD i

hn01 n0D |.
0
n
!
j=1
n
!
0
i
{n}=0 i=1
{n }=0
j
Making the change of variables
=

D
Q

ei ,

(C.1)

(C.2)

with and real, one has

!

= ,

e2i

.
= = ln

(C.3)

To obtain the the Jacobian (J) of the transformation, we note that

= ,

= ,

i ( )2
=

i
i1
= = ,

2
2
and then

J =

i 1
2

1 =

i
N

Z Y
D
i=1

di di exp

D
X

i 1
,
2

(C.5)

i1

= i.
2

In terms of the new variables, one obtains

D

(C.4)


j=1

71

(C.6)

Y
D
X
{n}=0 k=1

k eik

nk !

n k

|n1 nD i

D
X
Y

j eij
q

{n0 }=0 j=1

n0

n0j !

hn01 n0D |

0
0
D
D

D nj +nj eij (nj nj )

X
X
Y
iD Z Y
j
q
=
di di exp
j
N i=1
nj !n0j !
0
j=1
j=1
{n,n }=0

|n1 nD ihn01 n0D |.

(C.7)

Using the results

Z2

i(nn0 )

d e

= 2

n,n0

d e n = n!,

(C.8)

we get

n
D
D
Y
D
X
X
j j
(2i)D Z Y

1 =
|n1 nD ihn1 nD | (C.9)
di exp
j
N
n!
i=1
j=1
{n}=0 j=1 j

(2i)D
(2i)D X
|n1 nD ihn1 nD | =
.
N {n}=0
N

(C.10)

Therefore, with N = (2i)D given in Eq. (3.37), one has proven the completeness
relation.

72

Ap
endice D
Numerical values of the fundamental Grassmann
integrals

For simplicity the following argumentation will be for one generator only. We know
that the l.h.s. of (3.77) is
hf |gi = f0 g0 + f1 g1 ,

(D.1)

but the r.h.s. needs some work before, lets define it as

(f, g)
=
=

d d f ()k( , )g( )

d d(f0 + f1 )(k0 + k1 + k2 + k3 )g( )

d d(f0 k0 + f0 k1 + f0 k2
+f0 k3 + f1 k0 + f1 k1 )g( ).

(D.2)

Because of we are working with anticommuting variables we are going to impose,

analogously to the derivative, that any variable has to be next to the integral before
to operate
(f, g) =
=

d d(f0 k0 + f0 k1 + f0 k2 f0 k3 + f1 k0 + f1 k1 )g( )

d (f0 k0 I0 + f0 k1 I0 + f0 k2 I1
f0 k3 I1 + f1 k0 I1 + f1 k1 I1 )g( ),

(D.3)

where we have defined

I0 ()

I1 ()

and

d ,

(D.4)

and beign functions of one generator the Is are of the form

I() = a + b.

73

(D.5)

Continuing
(f, g) =

d [(f0 k0 I0 + f0 k2 I1 + f1 k0 I1 )
+(f0 k1 I0 f0 k3 I1 + f1 k1 I1 ) ](g0 + g1 )

d [(f0 k0 I0 + f0 k2 I1 + f1 k0 I1 )g0 + (f0 k0 I0 + f0 k2 I1 + f1 k0 I1 )g1

+(f0 k1 I0 f0 k3 I1 + f1 k1 I1 )g0 ]

d {[f0 k0 I0 () + f0 k2 I1 () + f1 k0 I1 ()]g0
+ [f0 k0 I0 () + f0 k2 I1 () + f1 k0 I1 ()]g1
+ [f0 k1 I0 () f0 k3 I1 () + f1 k1 I1 ()]g0 }

= I0 ( )[f0 k0 I0 () + f0 k2 I1 () + f1 k0 I1 ()]g0
+I1 ( )[f0 k0 I0 () + f0 k2 I1 () + f1 k0 I1 ()]g1
+I1 ( )[f0 k1 I0 () f0 k3 I1 () + f1 k1 I1 ()]g0 ,

(D.6)

for simplicity
I00 I0 ( )

I10 I1 ( ),

and

(D.7)

replacing
(f, g) = f0 g0 [k0 I00 I0 () + k2 I00 I1 () + k1 I10 I0 () k3 I10 I1 ()]
+f1 g0 [k0 I00 I1 () + k1 I10 I1 ()] + f0 g1 [k0 I10 I0 () + k2 I10 I1 ()]
+f1 g1 [k0 I10 I1 ()] .

(D.8)

If we equate this to Eq. (D.1) we get four equations

k0 I00 I0 () + k2 I00 I1 () + k1 I10 I0 () k3 I10 I1 () = 1,
k0 I00 I1 ()

k1 I10 I1 ()

(D.9)

= 0,

(D.10)

k0 I10 I0 () + k2 I10 I1 () = 0,

(D.11)

k0 I10 I1 () = 1.

(D.12)

From the last one

k0 6= 0

(D.13)

and
I10 , I1 C {0}

I10 = I1 .

(D.14)

Rewriting the four equations considering those two last conditions

k0 I00 I0 () + k2 I00 I1 + k1 I1 I0 () k3 I12 = 1,

(D.15)

k0 I00 I1 + k1 I12 = 0,

(D.16)

k0 I1 I0 () + k2 I12 = 0,

(D.17)

k0 I10 I1 = 1.

(D.18)

74

From the third one and because of k0 6= 0

I0 C

I00 = I0 ,

(D.19)

so
I0 (k0 I0 + k2 I1 ) + k1 I1 I0 k3 I12 = 1,

(D.20)

(k0 I0 + k1 I1 )I1 = 0,

(D.21)

I1 (k0 I0 + k2 I1 ) = 0,

(D.22)

k0 I12 = 1.

(D.23)

As I1 6= 0 and using the third equation in the first one

I1 (k2 I0 k3 I1 ) = 1,

(D.24)

(k0 I0 + k1 I1 ) = 0,

(D.25)

(k0 I0 + k2 I1 ) = 0,

(D.26)

k0 I12 = 1

(D.27)

we find
k0 =

1
,
I12

k1 = k2 =

(D.28)
I0
,
I13

(D.29)

I02
1
= 2 1+ 2 .
I1
I1
!

k3

(D.30)

So far we have found strong similarities beetwen all the Boson and Fermion equations, our third reason is related to this. We are going to fix these new parameters
in order to keep this symmetry. From the previous section we know the form of the
inner product of two functions in the Boson coherent representation (3.42), that tell
us to do
k0 1,

(D.31)

k1 = k2 0,

(D.32)

k3 1.

(D.33)

With those we find the original function k

k( , ) = 1

= e ;

75

(D.34)

the inner product

hf |gi =

d df ()e g( );

(D.35)

and then, the closure

I =
=

d d |ie h|

d de |ih|,

(D.36)

it is worth notice that it doesnt have any normalization factor. Now the integrals
are almost fixed
I0 = 0,

(D.37)

I1 = 1.

(D.38)

We are going to choose

I0 =

d1 = 0

and

I1

d = 1,

(D.39)

not only for simplicity, but in order to have the same numerical values than of the
derivatives too.

76

Ap
endice E
Symmetrical term for the exponential in the
partition function

In Section 3.5, Path Integral for bosons and fermions, we mentioned that one can
arrive at two choices for the calculation of the partition function in the ordering of
the Grassmann variables, the second choice is
,a)

hm+1 |e(a

|m i

with m = 0, , M 1.

(E.1)

We are going to follow the exact same steps. After we introduce the M 1 identities
(3.119)
Z=

Z Y
D M
1
Y
i=1 k=0

M
1 X
D
0
X
Y
di,k
di,k

hm+1 |e(a ,a) |m i, (E.2)

j,l j,l
exp
N
m=M 1
l=0 j=1

we simple order it, and use (3.121)

Z =

M
1 X
D
X
di,k
di,k

j,l
j,l
exp
N
l=0 j=1

Z MY
1 Y
D

k=0 i=1

0
Y

exp

m=M 1

D
X

 (m+1
,m )

p,m+1
p,m e

(E.3)

p=1

Putting together the exponentials, the argument would be

M
1
X

D 
X

l=0

= 

j,l
j,l  (l+1
, l )
j,l+1

j=1

M
1
X
l=0

D
X

j,l+1
j,l

j=1

j,l +  (l+1
, l ) .

(E.4)

Replacing it in the partition

Z=

Z MY
1 Y
D
k=0 i=1

di,k
di,k R( ,)
e
,
N

77

(E.5)

with

R( , ) 

M
1
X

D
X

j,k
j,k+1

j=1

k=0

j,k +

(k+1
, k ) .

(E.6)

Taking the limit M

lim Z =

Z ()

D[ ( )( )] eR(

,)

(E.7)

(0)

where
Z ()

Z M
D
Y Y

D[ ( )( )] lim

(0)

and
R( , )

Z
0

k=0

i=

di,k
di,k
,
N

(E.8)

| ( )
| ( ) +  ( ( ), ( )) .

D
X

|=

(E.9)

In which
M
1
X

lim

d,

(E.10)

j ( )
.

(E.11)

l=0

and


lim
0

j,l1
j,l

78

Ap
endice F
The Feynman ordering label technique

In the present Appendix we will explain Feynmans ordering label technique that
we used in Section 5.4. As remarked there, this technique provides a simplification
on the notation, the final result is not different from the traditional path-ordering
we used in section 5.3. The issue is in some sense trivial, but nevertheless we believe
it is instructive to have an Appendix to to clarify the notation. We shall follow the
original exposition of Feynman in Ref. [39].
appears because their algebra
The problem of ordering two operators A and B
6= AB.
The rules of ordinary algebra and
is noncommutative in general, that is AB
analysis for ordinary numbers are not directly applicable. For a single operator
,
there is no problem in defining a function of
, like through its power series
1
1

+
A = e 1 +
+
2
3!
1
1
= 1+
+ (
)2 + (
)3 + .
2
3!

(F.1)

In these cases the rules of ordinary algebra and analysis apply as for ordinary
with
numbers. The situation complicates when functions of another operator ,
then in general B
= exp ,
A 6= AB,
i.e.

6=
, are considered. Suppose B

A = e e 6= e+
B

for

6=
.

(F.2)

The
Let us suppose that small and expand the exponential to first order in .
zeroth order in is of course exp
, but the first order is neither (exp
) nor

(exp

).
Feynman devised a method to indicate the order in which operators are to operate
so as to free them of their noncommutative aspects. Feynman proposed to attach
an index to the operator with the convention that the operator with higher index
1 A0 and A0 B
1 mean B
A,
i.e. the order of the
operates later. For example, both B
1 were ordinary numbers.
indexed operators can be commuted freely as A0 and B

79

 one can safely write

= exp ,
Thus, if A = exp
and B
A = e 0 +1 ,
B

(F.3)

since there is only one way to interpret the exponential, namely

2
1 

0 + 1 +

0 + 1 +
e 0 +1 = 1 +
2

(F.4)

1  2
= 1+
0 + 1 +

0 + 2
0 1 + 12
2

(F.5)

1  2
= 1+
+ +

+ 2
+ 2 + .
2

(F.6)

This is the correct answer, since

e e =

1+
+

1 2

+
2



1 + +

1 2
+
2

(F.7)
1
1 2
= 1 + + 2 +
+
+
+
2
2
(F.8)

1  2
= 1+
+ +

+ 2
+ 2 + .
2

(F.9)

Let us write the exponential of the sum

+ as
=
exp(
+ )

n

+ )
lim exp (
n
n


n

= n
lim 1 + (
+ )
n


= lim 1 + (
+ )
n
n
|


1
1 + 1 (
. (F.10)
1 + (
+ )
+ )
n {z
n
}



n times

we replace
In each factor 1 + 1/n (
+ )
+ by
i + i , with 1 i n, and write
=
exp(
+ )

1
lim 1 + (
n + n )
n
n


1
1
1 + (
n1 + n1 ) 1 + (
1 + 1 )
n
n



1
1
1
(
n + n ) + (
n1 + n1 ) + + (
1 + 1 )
n
n
n

lim 1 +
n

n
1X
(
i + i ) .
= lim exp
n
n i=1

"

80

(F.11)

In the limit of n , the discrete index becomes continuous i/n s, 0 s 1,

with 1/n ds,

i
(s), i (s),
and
= exp
exp(
+ )

Z 1

ds [
(s) + (s)].

(F.12)

As Feynman remarked, it is evident that this expression is correct: calling

(s) +

(s) = (s), with a definite operator operating at order s, one has that
Z 1

Z 1

ds ,

ds (s) = exp

exp

(F.13)

since does not need to have the s dependence, it commutes with itself the integral,
and therefore
Z 1
Z 1
ds (s) =
ds = .
(F.14)
0

The expression in Eq. (F.12) at itself is trivial, but its nontrivial feature is that

the r.h.s. can be manipulated as

(s) and (s)
were ordinary functions of s, since
the order of operations will always be automatically specified by the index. For
example, in particular
= exp
exp(
+ )

Z 1

ds [
(s) + (s)]

Z 1

= exp

ds1
(s1 ) +

exp

Z 1
0

Z 1
0

 Z 1

ds1
(s1 )

2)
ds2 (s

2) .
ds2 (s

(F.15)

Let us check this for the lowest nontrivial order. The l.h.s. of this expression
gives

1  2
=1+
exp(
+ )
+ +

+
+
+ 2 + .
(F.16)
2
The ordering issue here is with respect to the term (
+
)/2 . Let us see how
this ordering appears with the integrals on the r.h.s. of Eq. (F.15). The ordering
R
R
2 ), which can be written
comes from the product of integrals 01 ds1
(s1 ) 01 ds2 (s
as
Z 1
0

ds1
(s1 )

Z 1
0

2) =
ds2 (s
+

Z 1
0

Z 1
0

ds1
(s1 )
ds1
(s1 )

Z s1
0

Z s1
s1

2)
ds2 (s
2 ).
ds2 (s

(F.17)

2) =
In the first integral on the r.h.s. one has s1 > s2 , so that
(s1 )(s
and the
value of the integral is
Z 1
0

ds1
(s1 )

Z s1
0

2) =
ds2 (s

81

Z 1
0

ds1

Z s1
0

ds2

Z 1
0

ds1 s1

.
(F.18)
2
2 ) =
In the second integral, one has s2 > s1 , then
(s1 )(s
, and the value of the
integral is
=

Z 1
0

ds1
(s1 )

Z 0
s1

2 ) =
ds2 (s

Z 1
0

Z 1
0

ds1

Z 1

ds2

s1

ds1 (1 s1 )

=
1

1
2

.
2

(F.19)

A case of interest to the present dissertation is when we want to expand exp(

+)

for small. Let us consider the expansion to first order in ,

=
exp(
+ )

exp

Z 1
0

exp

Z 1
0

= exp

Z 1
0

 Z 1

ds1
(s1 )

2)
ds2 (s



ds1
(s1 )

1+

Z 1
0

ds1
(s1 ) + exp

Z 1
0

2)
ds2 (s

Z 1

ds1
(s1 )

2 ). (F.20)
ds2 (s

The first term is trivial and gives exp

. In the second term, we break the integral
over s1 as


exp

Z 1
0

ds1
(s1 ) = exp

Z 1
s2



ds1
(s1 )

exp

Z s2
0

ds1
(s1 ) .

(F.21)

So, we have to consider the ordering of



exp

Z 1
s2



ds1
(s1 )

exp

Z s2
0

2 ).
ds1
(s1 ) (s

(F.22)

2 ) and is for all effects independent of s1 ,

The
(s1 ) in the first factor acts after (s
2 ). So, if we write (s
2 ) between
but
(s1 ) in the second term is to act before (s
both terms and imply the usual convention, the
(s1 )s become independent of s1
in the range from 0 to s2 and one may perform the integrals


exp

Z 1
0

ds1
(s1 )

Z 1
0

2 ) = [exp
ds2 (s
]
= e

Z 1
0

82

Z 1

ds [exp(s)
] [exp s
]

ds es es .

(F.23)

That is, we have obtained the result

e

=e

1+

Z 1

ds e

es + ,

(F.24)

which coincides with Eq. (5.47) to first order.

R
It is important to note that in a practical calculation, an expression like 01 ds
(s)
standing alone is obviously equal to , but it is far from trivial when multiplied by
R

factors involving 01 ds (s).

This means one must consider the complete expression

as a functional of
(s), (s),
etc.
One last remark: if one has

exp[(
+ )],

(F.25)

where is an ordinary number, Feynmans formula is

= exp
exp[(
+ )]

ds [
(s) + (s)]

"

exp

Z
0

as can be easily checked.

83

ds1
(s1 )

# "Z
0

2) ,
ds2 (s

(F.26)

Refer
encias

[1] J.W. Negele and H. Orland, Quantum Many-Particle Systems, Westview

Press, 1988.
[2] A.L. Fetter and J.D. Walecka, Quantum Theory of Many-Particle-Particle
Systems, McGraw Hill, 1971.
[3] C. Morningstar, The Monte Carlo method in quantum field theory, Lectures
given at the 21st annual Hampton University Graduate Studies (HUGS) Program at Jefferson Lab, Newport News, VA, June 5-23, 2006; arXiv:heplat/0702020.
[4] N. Metropolis, A.W. Rosenbluth, M.N. Rosenbluth, A.H. Teller and E. Teller,
J. Chem. Phys. 21, 1087 (1953).
[5] F.A. Berezin, The method of second quantization, Academic Press, 1966.
[6] C. Itzykson and J. B. Zuber, Quantum Field Theory, McGraw-Hill, 1980.
[7] J. Wess and J. Bagger, Supersymmetry and Supergrauity, Princeton University
Press, 1983.
[8] R.L. Stratonovich, Soviet Physics Doklady 2, 416 (1958); J. Hubbard, Phys.
Rev. Lett. 3, 77 (1959).
[9] I.C. Charret, E.V. Correa Silva, S.M. de Souza, O. Rojas Santos, M.T.
Thomaz, and C.E.I. Carneiro, Physica A 264, 204 (1999).
[10] S.M. de Souza and M.T. Thomaz, J. Math. Phys. 32, 3455 (1991).
[11] I.C. Charret, E.V. Correa Silva, S.M. de Souza and M.T. Thomaz, J. Math.
Phys. 36, 4100 (1995).
[12] I.C. Charret, S. M. de Souza and M.T. Thomaz, Braz. J. Phys. 26, 720 (1996).
[13] I.C. Charret, E.V. Correa Silva, S.M. de Souza, O. Rojas Santos, e M.T.
Thomaz, J. Math. Phys. 40, 4944 (1999).

84

[14] P. M. Stevenson, Phys. Rev. D 23, 2916 (1981).

[15] A. Okopinska, Phys. Rev. D 35, 1835 (1987).
[16] G. Krein, R. S. Marques de Carvalho, D. P. Menezes, M. Nielsen and
M. B. Pinto, Eur. Phys. J. A 1, 45 (1998) [arXiv:nucl-th/9709049].
[17] R. L. S. Farias, G. Krein and R. O. Ramos, Phys. Rev. D 78, 065046 (2008)
[arXiv:0809.1449 [hep-ph]].
[18] W. Ketterle and M.W. Zwierlein, Making, probing and understanding ultracold
Fermi gases, International School of Physics Enrico Fermi, Course CLXIV,
Varenna, 20 - 30 June 2006, edited by M. Inguscio, W. Ketterle, and C.
Salomon, IOS Press, 2008; arXiv:0801.2500v1
[19] M.H. Anderson, J.R. Ensher, M.R. Matthews,C.E. Wieman, and E.A. Cornell,
Science 269, 198 (1995); K.B. Davis, M.-O. Mewes, M.R. Andrews,N.J. van
Druten, D.S. Durfee, D.M. Kurn, and W. Ketterle, Phys. Rev. Lett. 75, 3965
(1995).
[20] J. Cubizolles, T. Bourdel, S.J.J.M.F. Kokkelmans, G.V. Shlyapnikov, and C.
Salomon, Phys. Rev. Lett. 91, 240401 (2003).
[21] S. Jochim, M. Bartenstein,A. Altmeyer, G. Hendl, C. Chin, J. HeckerDenschlag, and R. Grimm, Phys. Rev. Lett. 91, 240402 (2003) ; C.A. Regal,
M. Greiner, and D.S. Jin, Phys. Rev. Lett. 92, 083201 (2004).
[22] S. Giorgini, L. P. Pitaevskii, and S. Stringari, Rev. Mod. Phys. 80, 1215 (2008).
[23] T. Schaefer and D. Teaney, Nearly Perfect Fluidity: From Cold Atomic Gases
to Hot Quark Gluon Plasmas, [arXiv:0904.3107v2 [hep-ph] 17 Jul 2009]
[24] Atomcool: http://atomcool.rice.edu/ .
[25] CUA: http://cuaweb.mit.edu/Pages/Home.aspx .
[26] D. Lee, Phys. Rev. C 78, 024001 (2008) [arXiv:0803.1280 [nucl-th]].
[27] D. Lee and T. Schafer, Phys. Rev. C 73, 015202 (2006) [arXiv:nuclth/0509018].
[28] D. Lee and T. Schafer, Phys. Rev. C 73, 015201 (2006) [arXiv:nuclth/0509017].

85

[29] D. Lee, Phys. Rev. B 73, 115112 (2006) [arXiv:cond-mat/0511332].

[30] I. Montvay and G. M
unster, Quantum fields on a lattice, Cambridge University
Press, 1997.
[31] H.J. Rothe, Lattice gauge theories: an introduction, World Scientific, 2005 .
[32] B. Bringoltz, Dense hadronic matter in strong coupling lattice gauge theory,
PhD Thesis, Tel Aviv University, 2004; B. Bringoltz and B. Svetitsky, Phys.
Rev. D 68, 034501 (2003).
[33] Ph. de Forcrand and M. Fromm, Nuclear Physics from lattice QCD at strong
coupling, arXiv:0907.1915 [hep-lat].
[34] M. Creutz, Found. Phys. 30, 487 (2000) [arXiv:hep-lat/9905024].
[35] J. Bardeen, L.N. Cooper, and J.R. Schrieffer, Phys. Rev. 108, 1175 (1957).
[36] N.N. Bogoliubov, Sov. Phys.-JETP 7, 41 (1958), J.G. Valatin, Nuovo Cimento
7, 843 (1958).
[37] H. Lipkin, Quantum Mechanics: New Approaches to Selected Topics, Dover,
2001.
[38] X.-J. Liu, H. Hu and P. Drummond, Phys. Rev. A 75, 023614 (2007).
[39] R.P. Feynman, Phys. Rev. 84, 108 (1951).

86