Static and dynamic properties of the spinless Falicov-Kimball model
Extracted source text ยท arXiv cond-mat/0701433
\documentclass[preprint,aps,prb,showpacs,amsmath,amssymb]{revtex4}
\usepackage{graphics}
\usepackage{epsf,graphicx,pstcol,epsfig,psfrag}
\usepackage{color}
\begin{document}
\title{
Static and dynamic properties of the spinless Falicov-Kimball model}
\author{K.W.~Becker$^{a}$, S.~Sykora$^{a}$, and V.~Zlatic$^{a,b}$}
\affiliation{NN
$^{a}$Institut f\"{u}r Theoretische Physik,
Technische Universit\"{a}t Dresden, D-01062 Dresden, Germany \\
$^{b}$Institute of Physics, Bijenicka c.~46, P.O.B 304, 10000
Zagreb, Croatia
}
\date{\today}
\begin{abstract}
The spinless Falicov-Kimball model is studied by the use of a recently developed
projector-based renormalization method (PRM) for many-particle Hamiltonians.
The method is used to evaluate static and dynamic quantities of the one-dimensional
model at half-filling. To these belong the quasiparticle excitation energy
$\tilde{\varepsilon}_k$ and the momentum distribution $n_k$ of the
conduction electrons and spatial correlation functions of the localized electrons.
One of the most remarkable results is the appearance of a gap in
$\tilde{\varepsilon}_k$ at the Fermi level of the order of the
Coulomb repulsion $U$, which is accompanied by a smooth behavior for $n_k$.
The density of states for the conduction electrons and the one-particle
spectral functions for the localized electrons are also
discussed. In both quantities a gap opens with increasing $U$.
\end{abstract}
\pacs{71.10.Fd, 71.27.+a, 75.30.Mb}
\maketitle
%%%%%%%%%%%%%%%%%%%
\section{Introduction}
The Falicov-Kimball model is a widely used model to study the properties of
interacting fermions\cite{falicov_kimball}
and has also been used to describe the valence change transition
in YbInCu$_4$ and similar compounds\cite{freericks.98}.
(For a recent review see Ref.~ \onlinecite{freericks.03}).
The model considers a lattice of localized $f$-sites, which are either
empty or singly-occupied site, and of conduction electron sites,
which are delocalized by the nearest-neighbor hopping.
There are $n_f$ $f$-electrons and $n_c$ conduction electrons
interacting by an on-site Coulomb interaction $U$.
The two types of electrons share the common chemical potential $\mu$,
which is always adjusted so as to keep $n_f+n_c=1$.
The local $f$-charge is a constant of motion but thermal fluctuations
can change the average $f$-occupation by transferring
electrons or holes from the conduction band to $f$-states.
The relative occupation of $f$-states and conduction states
is determined by the competition between the entropy of the $f$- and
the conduction states, the excitation energy of the f-states, the kinetic
energy of band states, and the interaction energy.
The general solution of the Falicov-Kimball model
is not known except in infinite dimension
\cite{freericks.03,brandt.89,dongen.90}.
In low dimensions, the zero-temperature phase diagram
is highly non-trivial and has attracted a lot of theoretical
attention (see (\onlinecite{freericks.03}) for the list of references).
In one dimension the thermodynamic properties of the model
are more or less understood, whereas dynamical properties have
been much less studied.
Kennedy and Lieb \cite{kennedy.86,lieb.86} proved that at low enough
temperatures the half-filled Falicov-Kimball model for dimensions $d\geq 2$
possesses long-range order, i.e., the ions form a checkerboard pattern,
the same as in the ground state. In the ordered phase, the lattice can be divided
into two inter-penetrating sublattices A and B in such a way that all nearest
neighbors of a site from sublattice A belong to sublattice B and {\it vice versa}.
This result holds for any value of the
Coulomb interaction $U$. For one dimension longe range order exists
only at temperature $T=0$.
In what follows, we first describe the Hamiltonian and the projector-based
renormalization method (PRM) which was recently introduced (Sect.~I).
In Sect.~II the method is then applied to the one-dimensional Falicov-Kimball model
and renormalization equations are derived for the parameters of the Hamiltonian.
The results for static and dynamic quantities are discussed in Sect. III and IV.
To simplify the calculation, we consider only the case with translation symmetry.
This excludes the discussion of physical properties which
depend on lattice sites $i$, i.e. we do not discuss the ordered phase
which emerges at low temperatures. However, as discussed in Sect.~V,
we can still evaluate physical quantities which are given by averages over
all lattice sites. Note that the PRM approach can also be extended to
situations with long range order.
This was shown for instance in Ref.~(\onlinecite{sykora.05}),
where the quantum phase transition for the spinless Holstein model
was discussed as function of the electron-phonon coupling.
%
\subsection{Model\label{FK}}
The spinless Falicov-Kimball model is described by the Hamiltonian
\begin{eqnarray}
\label{G1}
{\cal H} &=&
\sum_{<i,j>} t_{ i,j} \;
c_i^\dagger c_j
+
\varepsilon_f\sum_i n_i +
U \sum_i n_i n_i^c ,
\end{eqnarray}
where, $c_i^\dagger$ and $ c_i$ are the creation and
annihilation operators for
conduction electrons at site $i$, $t_{ i,j}$ are the hopping matrix
elements, $\varepsilon_f$ is the energy level of the localized electrons,
and $n_i= f_i^\dagger f_i$ and $n_i^c= c_i^\dagger c_i $ are the occupation
number operators of the localized and
conduction electrons at site $i$.
The summation runs over the $N$ sites of a periodic lattice,
and $U$ is the interaction strength of the local Coulomb repulsion at sites
$i$ between conduction and localized electrons.
Note that $n_i$ at each site $i$
commutes with the Hamiltonian. Thus, the $f$-electron occupation number
is a good quantum number, taking only two values 1 or 0, according to whether
or not site $i$ is occupied by the localized $f$-electron.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Projector-based renormalization method (PRM)}
\label{PRM}
Let us first discuss a recently introduced many-particle
method, the projector-based renormalization
method (PRM) \cite{becker},
which provides an approximate solution to many-particle
problems defined by the Hamiltonian $\mathcal{H}$,
\begin{eqnarray}
\label{G2}
{\cal H} &=& {\cal H}_{0} + {\cal H}_{1}.
\end{eqnarray}
Here, ${\cal H}_{0}$ is the unperturbed part with a known eigenvalue
spectrum,
$ {\cal H}_{0} |n\rangle = E^{(0)}_n |n\rangle$,
and $ {\cal H}_{1}$ is the perturbation which does not commute
with ${\cal H}_{0}$ and has off-diagonal matrix elements,
$ \langle n|{\cal H}_{1}|m\rangle\neq 0$, but no diagonal elements.
%
In a usual perturbative approach one evaluates the corrections
to the eigenstates and eigenvalues of ${\cal H}_{0}$ due
to ${\cal H}_{1}$. The n-th order correction requires
the evaluation of all matrix elements
up to $ \langle n| {({\cal H}_{1})}^n |m\rangle$, which are
difficult to calculate.
An alternative insight is obtained by making a unitary transformation
to a new basis which has no transition matrix elements with energy differences
larger than some chosen cutoff $\lambda<\Lambda$. Here $\Lambda$ is the
largest energy difference between any two eigenstates
of ${\cal H}_0$ which are connected by ${\cal H}_1$.
This generates a new Hamiltonian
\begin{eqnarray}
\label{G4}
{\cal H}_{\lambda} &=&
e^{X_{\lambda}} \, {\cal H} \, e^{-X_{\lambda}} ,
\end{eqnarray}
which can be written as a sum of two terms,
\begin{eqnarray}
\label{G5}
\mathcal{H}_{\lambda} &=& \mathcal{H}_{0,\lambda} + \mathcal{H}_{1,\lambda},
\end{eqnarray}
such that
$\langle m| \mathcal{H}_{1,\lambda} |n\rangle=0$
for $|E^{\lambda}_n-E^{\lambda}_m| > \lambda$, where
$E^{\lambda}_n$ and $|n\rangle$ are the new eigenvalues and eigenstates
of $\mathcal{H}_{0,\lambda}$. $\mathcal{H}_{1,\lambda}$
is chosen in such a way that it has no diagonal
elements with respect to $\mathcal{H}_{0,\lambda}$.
%
The generator of the unitary transformation is denoted by $X_\lambda$
which has to be anti-hermitian
$X^{\dagger}_{\lambda}=-X_{\lambda}$. Note that
the elimination of high energy transitions may also generate new interaction
terms which have operator structures different
from that of ${\cal H}_1$. However,
they do not connect states with an energy separation larger than
$\lambda$.
In the PRM method
the elimination of transition matrix elements is carried out
by defining a generalized projection operator ${\bf P}_{\lambda}$
which removes from any operator $\mathcal{A}$ those parts which give
rise to 'forbidden' transitions, i.e.,
\begin{eqnarray}
\label{G6}
{\bf P}_{\lambda}\mathcal{A} &=&
\sum_{
\genfrac{}{}{0pt}{1}{
\genfrac{}{}{0pt}{1}{m,n}{
\left| E_n^{\lambda} - E_m^{\lambda} \right| \leq \lambda
}
}{}
}
|n\rangle \langle m| \, \langle n| \mathcal{A} |m\rangle ,
\end{eqnarray}
i.e.~we retain in \eqref{G6} only the states $|m\rangle$ and $|n\rangle$
with $|E_n^{\lambda} - E_m^{\lambda} |\leq \lambda$.
The orthogonal complement of ${\bf P}_{\lambda}$ is
$ {\bf Q}_{\lambda} = {\bf 1} - {\bf P}_{\lambda}$. Here only dyads
$|n\rangle \langle m|$ contribute with $|E_n^{} - E_m|>\lambda$.
The generator ${X_{\lambda}}$
in \eqref{G4} is determined by the
condition
\begin{eqnarray}
\label{G8}
{\bf Q}_{\lambda}{\cal H}_{\lambda}=0,
\end{eqnarray}
as can be seen by expanding the exponentials in \eqref{G4}.
This gives
\begin{eqnarray}
\label{G9}
{\bf Q}_{\lambda}
\left\{
{\cal H}_{1}
+
\sum_{n=1}^\infty \frac{1}{n!}\;
{{\bf X}_{\lambda}}^n {\cal H}
\right\}
&=&
0 ,
\end{eqnarray}
where ${\bf X}_{\lambda}$ is a superoperator defined by
${\bf X}_{\lambda} {\cal H}=[X_{\lambda}, {\cal H}]$,
${{\bf X}_{\lambda}}^2 {\cal H}=[X_{\lambda}, [X_{\lambda}, {\cal H}]]$,
etc.
In lowest order, ${\bf Q}_\lambda {\cal H}_\lambda=0 $ reduces to
${\bf Q}_\lambda \left\{ {\cal H}_{1} -
{\bf L}_0 {X}_{\lambda}^{(1)} \right\} = 0$,
where ${\bf L}_0 {\cal A}=[{\cal H}_{0}, {\cal A}]$
defines the unperturbed Liouville operator ${\bf L}_0$,
which commutes with ${\bf Q}_{\lambda}$.
Obviously, the condition ${\bf Q}_{\lambda}{\cal H}_{\lambda}=0$ is satisfied
by the expression
${X}_{\lambda}^{(1)}={\bf L}_0^{-1}{\bf Q}_{\lambda}{\cal H}_{1}$,
i.e.,~to lowest order, ${X}_{\lambda}^{(1)}$ can be obtained from the
decomposition of ${\cal H}_1$ into the eigenmodes of ${\bf L}_0$.
The higher order correction terms to ${X}_{\lambda}$ follow systematically
from the higher order commutators of \eqref{G9}
(for details see, e.g., Ref.~\onlinecite{becker}).
We see from Eq.~\eqref{G9} that all transitions in ${\cal H}_1$ with
energy transfers between the original cutoff $\Lambda$ and the new cutoff
$\lambda$ are eliminated in one step. However it is more convenient
to perform the elimination procedure step-wise,
such that each step reduces the cutoff energy $\lambda$
by a small amount $\Delta \lambda$.
Thus, the first step removes all the transitions
which involve energy transfers
between (the original cutoff) $\Lambda$
and $\Lambda- \Delta \lambda$, the subsequent steps remove all transitions
larger than $\Lambda- 2\Delta \lambda$, $\Lambda- 3\Delta \lambda$,
and so on. The unitary transformation for the
step from an intermediate cutoff $\lambda$ to the
new cutoff $\lambda - \Delta \lambda$
reads (in analogy to \eqref{G4})
\begin{eqnarray}
\label{G11}
\mathcal{H}_{\lambda-\Delta\lambda} &=&
e^{X_{\lambda,\Delta\lambda}}
\, \mathcal{H}_{\lambda} \,
e^{-X_{\lambda,\Delta\lambda}}
\end{eqnarray}
where $\mathcal{H}_{\lambda} = {\cal H}_{0,\lambda}+{\cal H}_{1,\lambda}$
and
$\mathcal{H}_{\lambda-\Delta\lambda}=
{\cal H}_{0,\lambda-\Delta \lambda}+
{\cal H}_{1,\lambda- \Delta \lambda}$.% In analogy to \eqref{G8}
The generator $X_{\lambda,\Delta\lambda}$ is now fixed by the condition
\begin{eqnarray}
\label{G12}
{\bf Q}_{\lambda-\Delta\lambda} {\cal H}_{\lambda-\Delta\lambda} &=& 0.
\end{eqnarray}
which specifies that ${\cal H}_{\lambda - \Delta \lambda}$ contains
no matrix elements which connect eigenstates of ${\cal H}_{0, \lambda -\Delta
\lambda}$ with energy differences larger than $\lambda - \Delta \lambda$.
Let us assume that the operator structure of
$\mathcal{H}_{\lambda}$ is invariant with respect to
further unitary transformations. Then, we can use
Eq.~\eqref{G9} to derive difference equations for
the $\lambda$-dependence of the coupling constants of the
Hamiltonian. These equations connect
the parameters of the Hamiltonian with cutoff $\lambda$ to those with cutoff
$\lambda - \Delta \lambda$. By using a finite number of steps we can
proceed to $\lambda\rightarrow 0$
and obtain a set of nonlinear equations for the renormalized parameters.
The solution determines the final, fully renormalized Hamiltonian
${\cal H}_{\lambda \rightarrow 0}={\cal H}_{0,\lambda
\rightarrow 0}$, which depends on the initial parameter values of the original
model at cutoff $\Lambda$.
In this limit all transitions due to the interaction with nonzero
transition energies have been
eliminated so that ${\cal H}_{1,\lambda \rightarrow 0}$ identically
vanishes, i.e.~${\cal H}_{1,\lambda \rightarrow 0}=0$.
The underlying idea of the PRM, namely the elimination of the interaction terms
by the unitary transformations, has been used before in the literature.
For instance, in Ref.\onlinecite{kohn.64}
a unitary transformation has been employed to eliminate
off-diagonal matrix elements of the Coulomb interaction
in order to study ground state properties of the Hubbard model.
Similarly, in the
Schrieffer-Wolff transformation \cite{schrieffer-wolff}, which maps the
Anderson to the Kondo model,
matrix elements connecting different charge configurations
of magnetic ions are eliminated also by use of a unitary transformation.
Note however that in the
PRM approach unitary transformations are performed in small steps
in contrast to earlier applications. Note also that the approach removes
high energy transitions but does not decimate the Hilbert space.
This is different to the poor man's
scaling\cite{anderson} which removes high energy states.
It should also be mentioned that the
PRM resembles the previous flow equation
method \cite{wegner} by Wegner and the similarity
renormalization approach\cite{wilson}
by Glazek and Wilson which can be considered as continuous versions of the
PRM method.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Renormalization of the Falicov-Kimball model}
In order to apply the PRM we express the Falicov-Kimball
interaction in terms of the fluctuations with respect to the
thermal averages of $n_i^c$ and $n_i$,
and write the diagonal and off-diagonal part of ${\cal H}$ as
%${\cal H}_0+{\cal H}_1$ as
\begin{eqnarray}
\label{G13}
{\cal H}_0 &=& \sum_i \big( \varepsilon_f + U \big< n_i^c \big> \big)\;
n_i +
\sum_k \big( \varepsilon_k + U \big< n_i \big> \big) \;
c_{k}^\dagger c_k - UN \big<n_i\big>\; \big< n_i^c \big> ,
\end{eqnarray}
and
\begin{eqnarray}
\label{G14}
{\cal H}_1 &=& \frac{U}{N} \sum_{i k {k'}}\;
e^{-i(k - {k'}){R}_i}\;
\delta n_i\; \delta (c_k^\dagger c_{k'}),
\end{eqnarray}
where $\delta {n_i^c}= {n_i^c}- \big< {n_i^c}\big>$,
$\delta {n_i}= {n_i}- \big< {n_i}\big>$,
$\big< {\cdots}\big>$ denotes the thermal averaging with the full
Hamiltonian ${\cal H}$, and
$
c_{k}^\dagger =
({1}/{\sqrt N}) \sum_i e^{ik{R}_i}c_{i}^\dagger
$
is the Fourier transform of $c_i^\dagger$.
Of course, due to the assumed translation invariance
$\langle n_i \rangle$ is equivalent to
$(1/N)\sum_i \langle n_i \rangle$, where $N$ is the number of lattice sites.
The sum in \eqref{G14} is restricted to $k\neq k'$,
since the diagonal part of the Coulomb repulsion is included in ${\cal H}_0$.
From translation invariance it follows
$\delta(c_k^\dagger c_{k'}) = c_k^\dagger c_{k'}
- \big< c_k^\dagger c_k\big> \delta_{k, k'}$.
In principle, the average values in \eqref{G13} can be defined
with an arbitrary ensemble since Eq.~\eqref{G1} and
Eqs.~\eqref{G13}, \eqref{G14} are equivalent.
It has turned out that the best
choice is to evaluate the averages with respect to the
original full
Hamiltonian ${\cal H}= {\cal H}_{\Lambda}$, which can be done by
the procedure explained below.
The perturbation ${\cal H}_1$ causes transitions between the
eigenstates of ${\cal H}_0$ by creating electron-hole pairs
in the conduction band, whereas the number of $f$-electrons is conserved
at each site. Using
${\bf L}_0 \; \delta n_i \; \delta(c_k^\dagger c_{k'})
= (\varepsilon_k - \varepsilon_{k'}) \;
\delta n_i \; \delta(c_k^\dagger c_{k'})$,
${\cal H}_1$ can be decomposed into a sum of
eigenmodes of the unperturbed Liouville operator.
The lowest order solution of the generator of the unitary
transformation is
$X_\lambda^{(1)}={\bf L}_0^{-1} {\bf Q}_{\lambda} {\cal H}_1 $,
which is given by
\begin{eqnarray}
\label{G16}
X_\lambda^{(1)}
&=&
\frac{U}{N} \sum_{i k k'}\;
\frac{e^{i(k - k'){R}_i}}
{\varepsilon_k - \varepsilon_{k'}}\;
\Theta({|\varepsilon_k - \varepsilon_{k'}|}- \lambda)\;
\delta n_i\; c_k^\dagger c_{k'}.
\end{eqnarray}
Here
$\Theta({|\varepsilon_k - \varepsilon_{k'}|}- \lambda)$
is the projection operator which removes from ${\cal H}_1$
all transitions with energy
transfers larger than $\lambda$. Note that the
same operator form also appears in the non-perturbative calculation
below.
%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Renormalized Hamiltonian\label{renormalizedH}}
%%%%%%%%%%%%%%%%%%%%%%%%%
The unitary transformation leads to the renormalized Hamiltonian
${\cal H}_\lambda = e^{X_\lambda} {\cal H} e^{-X_\lambda}$,
which can be written as
${\cal H}_\lambda ={\cal H}_{0,\lambda} + {\cal H}_{1,\lambda}$,
where ${\cal H}_{1,\lambda}$ gives rise to the transitions between the
eigenstates of ${\cal H}_{0,\lambda}$ but has no diagonal elements.
The choice of ${\cal H}_{0,\lambda}$ is not unique,
and each particular case requires a physical intuition.
But in any case it has to be such that the thermal
averages which appear in the procedure can be evaluated.
For the Falicov-Kimball model, the choice which emphasizes the
weak coupling limit, is
\begin{eqnarray}
\label{G17}
{\cal H}_{0,\lambda} &=& \varepsilon_{f,\lambda} \sum_i \delta n_i
+
\sum_{i \neq j} g_{ij, \lambda} \; \delta n_i \; \delta n_j
+ \sum_k \varepsilon_{k,\lambda} \; c_k^\dagger c_k
+ E_\lambda ,\\
&& \nonumber \\
{\cal H}_{1,\lambda} &=&
\frac{U}{N} \sum_{i k k'}
\Theta (\lambda -|\varepsilon_{k, \lambda} -\varepsilon_{k', \lambda}|) \;
e^{i(k - k'){R}_i}\;
\big( \delta n_i\; \delta (c_k^\dagger c_{k'})
\big) ,
\label{G18}
\end{eqnarray}
where
$ \varepsilon_{f,\lambda}$, $ g_{ij, \lambda}$, and
$\varepsilon_{k,\lambda}$
are the renormalized parameters which are calculated below.
Note that the interaction parameter $U$ will not
change in the renormalization procedure below
so that a $\lambda$-dependence of $U$ is not considered
from the beginning.
The second term in \eqref{G17} is a new density-density interaction between
the localized electrons which is generated during the renormalization
procedure.
Additional higher order operator terms, which are also generated
by the renormalization, are neglected.
The original model ($\lambda = \Lambda$) is defined by the
initial condition, according to \eqref{G13}, \eqref{G14}
\begin{eqnarray}
\label{G19}
&& \varepsilon_{f, \Lambda} = \varepsilon_f + U \big< n_i^c \big>
\hspace*{2cm}
\varepsilon_{k, \Lambda} =
\varepsilon_k + U \big< n_i\big> \\
&& g_{ij, \Lambda} =0 \hspace*{3.7cm}
E_{\Lambda} = -N\big<n_i \big>\; ( \varepsilon_f +
U \; \big< n_i^c \big>) . \nonumber
\end{eqnarray}
Once we have made the ansatz \eqref{G17} and \eqref{G18} for the
operator structure of ${\cal H}_{0,\lambda} $ and
${\cal H}_{1,\lambda}$, we
reduce the cutoff to $\lambda -\Delta \lambda$
to find the new effective Hamiltonian
\begin{eqnarray}
\label{G20}
{\cal H}_{\lambda -\Delta \lambda} &=&
e^{X_{\lambda, \Delta \lambda}} {\cal H}_\lambda
e^{-X_{\lambda, \Delta \lambda}}
={\cal H}_{0,\lambda -\Delta \lambda}
+{\cal H}_{1,\lambda -\Delta \lambda }
\end{eqnarray}
with renormalized parameters. The new Hamiltonian is such
that ${\cal H}_{1,\lambda -\Delta \lambda }$
does not give rise to transitions with energy transfers
larger than $(\lambda -\Delta \lambda)$ (with respect to
the new unperturbed part ${\cal H}_{0,\lambda -\Delta \lambda }$).
${\cal H}_{\lambda - \Delta \lambda}$ should have the same operator
structure as ${\cal H}_{\lambda}$, however with renormalized
parameters
\begin{eqnarray}
\label{G31}
{\cal H}_{\lambda- \Delta \lambda} &=&
\varepsilon_{f,(\lambda- \Delta \lambda)} \sum_i \delta n_i
+
\sum_{i \neq j} g_{ij, (\lambda - \Delta \lambda)} \;
\delta n_i \; \delta n_j
+
\sum_k \varepsilon_{k,(\lambda - \Delta \lambda)}
\; c_k^\dagger c_k + E_{(\lambda - \Delta \lambda)}
\\
&+&
\frac{U}{N} \sum_{i k k'} \Theta(\lambda - \Delta \lambda
-|\varepsilon_{k, \lambda
- \Delta \lambda} - \varepsilon_{k', \lambda - \Delta \lambda}|) \;
e^{i(k - k'){R}_i}\;
\big( \delta n_i\; \delta (c_k^\dagger c_{k'})
\big) .
\nonumber
\end{eqnarray}
%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Unitary transformation\label{generator}}
%%%%%%%%%%%%%%%%%%%%%%%%
For the explicit evaluation of the unitary transformation \eqref{G20}
we need the generator $X_{\lambda, \Delta \lambda}$. We make the
following ansatz
\begin{eqnarray}
\label{G22}%
&& X_{\lambda, \Delta \lambda} = \frac{1}{N} \sum_{k, k',i}
A_{k, k'}^{\lambda,\Delta\lambda} \;
%
\Theta_{kk' }^{\lambda,\Delta\lambda}
\;
e^{-i(k- k'){R}_i }\;
\delta n_i \;
\delta( c_k^\dagger c_{k'}) .
\end{eqnarray}
Here $\Theta_{kk' }^{\lambda,\Delta\lambda}$ is a product
of $\Theta$-functions
\begin{eqnarray}
\label{G23}
\Theta_{kk' }^{\lambda,\Delta\lambda} &=&
\Theta(\lambda - |\varepsilon_{k, \lambda} -
\varepsilon_{k', \lambda}|)
\times
\Theta \big[|\varepsilon_{k, (\lambda- \Delta \lambda)} -
\varepsilon_{k', (\lambda - \Delta \lambda)}| -
(\lambda - \Delta \lambda) \big] ,
\end{eqnarray}
which project all the operators in $X_{\lambda, \Delta \lambda}$
'on the shell' between $\lambda$ and $\lambda- \Delta \lambda$.
More precisely, the energy differences
$|\varepsilon_{k, \lambda} -\varepsilon_{k', \lambda}|$
and $|\varepsilon_{k, (\lambda- \Delta \lambda)} -
\varepsilon_{k', (\lambda - \Delta \lambda)}|$ in \eqref{G23}
refer to the two different Hamiltonians
${\cal H}_{\lambda}$ and ${\cal H}_{\lambda- \Delta \lambda}$.
Therefore, the two $\Theta$-functions in \eqref{G23} take into account that
${\cal H}_\lambda$ possesses only
transition elements with energy transfer $|\varepsilon_{k, \lambda} -
\varepsilon_{k', \lambda}| < \lambda$, whereas in
${\cal H}_{\lambda- \Delta \lambda}$ no transitions with
$|\varepsilon_{k, (\lambda- \Delta \lambda)} -
\varepsilon_{k', (\lambda - \Delta \lambda)}| >
\lambda - \Delta \lambda$ are allowed.
The coefficients $A_{k, k'}^{\lambda,\Delta\lambda}$
are determined below, using the condition
${\bf Q}_{\lambda-\Delta\lambda} {\cal H}_{\lambda-\Delta\lambda} = 0$.
Again, the form of $X_{\lambda, \Delta \lambda}$
is suggested by its lowest order expression according to
\eqref{G16} (compare (\onlinecite{becker})).
The unitarity of the transformation requires
$X_{\lambda,\Delta\lambda}=-X_{\lambda,\Delta\lambda}^\dagger$,
so that $A_{k, k'}^{\lambda,\Delta\lambda}$ has to be
antisymmetric with respect to the interchange of $k$ and $k'$.
Expanding the exponentials in \eqref{G20}
the transformation can be written as,
\begin{eqnarray}
\label{G25}
{\cal H}_{\lambda - \Delta \lambda} &=&
e^{X_{\lambda, \Delta \lambda}} {\cal H}_{\lambda}
e^{-X_{\lambda, \Delta \lambda}}
= {\cal H}_{\lambda}
+
\sum_{n=1}^\infty \frac{1}{n!}\;
{\bf X}_{\lambda, \Delta \lambda}^n {\cal H}_{\lambda}
\end{eqnarray}
where
${\bf X}_{\lambda, \Delta \lambda} {\cal H}_{\lambda} =
[X_{\lambda, \Delta \lambda}, {\cal H}_{\lambda}]$,
${\bf X}_{\lambda, \Delta \lambda}^2 {\cal H}_{\lambda} =
[X_{\lambda, \Delta \lambda}, [X_{\lambda, \Delta \lambda},
{\cal H}_{\lambda}]]$, etc. The commutators are evaluated
in Appendix A.
The basic approximation of the PRM approach is to replace
some of the operators generated by $X_{\lambda,\Delta\lambda}$
by ensemble averages, so as to keep the structure of
${\cal H}_\lambda$ invariant during the renormalization procedure.
This additional factorization enables us
to resum the series in \eqref{G25}.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Renormalization of the coupling constants\label{couplings}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
By comparing the respective operators in \eqref{G31} with those
obtained from \eqref{G25} one finds renormalization equations
for the parameters of ${\cal H}_\lambda$. First, from
the prefactors of $\delta n_i$ one obtains
\begin{eqnarray}
\label{G33}
\varepsilon_{f, (\lambda - \Delta \lambda)}
&=& \varepsilon_{f, \lambda} +
\frac{(1- 2\big< n_i\big>)}{2N} \sum_{k k'}
\Theta_{k, k'}^{\lambda, \Delta \lambda} \;
(\varepsilon_{k', \lambda}-
\varepsilon_{k, \lambda})\;
\frac{(1- \cos v_{k k'}^{\lambda})}{C_\rho(k -k')} \;
\big< c_k^\dagger c_k \big> \\
&+& \frac{U(1-2\big< n_i \big>)}{N \sqrt{N}}
\sum_{k k'}
\Theta_{k, k'}^{\lambda, \Delta \lambda}
\frac{\sin v_{k k'}^{\lambda}}{\sqrt{C_\rho(k-k')}}
\; \big< c_k^\dagger c_k \big> .
\nonumber
\end{eqnarray}
Similarly, by comparing
the coefficients of $c_k^\dagger c_k$ one finds
\begin{eqnarray}
\label{G36}
\varepsilon_{k,(\lambda - \Delta \lambda)} &=&
\varepsilon_{k,\lambda}
+ \frac{1}{2} \sum_{k'}
\Theta_{k, k'}^{\lambda, \Delta \lambda} \;
(\varepsilon_{k',\lambda}- \varepsilon_{k, \lambda} )\;
( 1- \cos v_{k k'}^{\lambda}) \\
&+& \frac{U}{\sqrt{N}} \sum_{k'}
\Theta_{k, k'}^{\lambda, \Delta \lambda} \;
\sqrt{C_\rho(k- k')}\; \sin v_{k k'}^{\lambda} .
\nonumber
\end{eqnarray}
The renormalization equation for the interaction parameter
$g_{ij,\lambda}$ between $f$-electrons is best expressed
in terms of its Fourier transform $g_{{q}, \lambda}$,
\begin{eqnarray}
\label{G39}
g_{{q}, (\lambda - \Delta \lambda)} &=&
g_{{q}, \lambda }
+ \frac{1}{4}
\sum_k \frac{\big<c_k^\dagger c_k\big>
-\big<c_{k+{q}}^\dagger c_{k+{q}} \big>}
{C_\rho({q})}\;
\Theta_{k, k+{q}}^{\lambda, \Delta \lambda}\;
(\varepsilon_{k+{q}, \lambda} -
\varepsilon_{k, \lambda})\;
(1- \cos v_{k, k+{q} }^\lambda )
\nonumber \\
&+& \frac{U}{2\sqrt{N}} \sum_k
\frac{\big<c_k^\dagger c_k\big> -
\big<c_{k+{q}}^\dagger c_{k+{q}}\big>}
{\sqrt{C_\rho({q})}}\;
%\Big(
\Theta_{k, k +{q}}^{\lambda,
\Delta \lambda}\;
\sin v_{k, k+{q}}^\lambda \\
&-& \frac{1}{N} \sum_{q'} (\cdots) \nonumber
\end{eqnarray}
where $(\cdots)$ denotes the second and third term on the r.h.s.~with
the wave vector ${q}$ replaced by ${q'}$. In this way,
the exact sum rule $\sum_{q} g_{{q},\lambda}
=0$ is fulfilled which guarantees that only sites
with $i \neq j$ contribute to $g_{ij}$.
Finally, also the renormalization equation for
the energy shift $E_\lambda$ can be obtained which will not be given
explicitly. The other new quantities in Eqs.~\eqref{G33} to \eqref{G39}
are defined as follows
\begin{eqnarray}
\label{G28}
&& v_{k k'}^{\lambda} = 2 \sqrt{\frac{C_\rho(k- k')}{N}}
\; A_{k, k'}^{\lambda,\Delta\lambda},
\hspace*{1cm}
C_\rho({q}) = \frac{1}{N} \sum_{ij} e^{i{q}({R}_i-
{R}_j)} \big< \delta n_i \delta n_j \big>. \\
&& \nonumber
\end{eqnarray}
Finally we have to determine the coefficients
$A_{k, k'}^{\lambda,\Delta\lambda}$ of the unitary
transformation. They follow from the condition
${\bf Q}_{\lambda-\Delta\lambda} {\cal H}_{\lambda-\Delta\lambda}=0$,
which removes from ${\cal H}_{\lambda-\Delta\lambda}$
all the operators giving rise to the transitions with energy
transfers larger than $\lambda-\Delta\lambda$. One finds
\begin{eqnarray}
\label{G30}
\Theta_{kk' }^{\lambda,\Delta\lambda}
\; A_{k, k'}^{\lambda,\Delta\lambda}&=&
\Theta_{kk' }^{\lambda,\Delta\lambda} \; \frac{1}{2} \;
\sqrt{\frac{N}{C_\rho(k - k')}}\;
\arctan\Big( 2 \sqrt{\frac{C_\rho(k- k')}{N}}\;
\frac{U}{\varepsilon_{k, \lambda}
-\varepsilon_{k', \lambda}}
\Big) .
\end{eqnarray}
Note that the $A_{k, k'}^{\lambda,\Delta\lambda}$
depend on the parameters at cutoff $\lambda$ as well as from the reduced
cutoff ${\lambda-\Delta\lambda}$.
The presence of $\Theta_{kk' }^{\lambda,\Delta\lambda} $
on both sides of Eq.\eqref{G30}
means that $A_{k, k'}^{\lambda,\Delta\lambda}$ has to be
specified only for those values of $k$ and $ k'$
which are 'on the shell' defined by the condition
$\Theta_{kk'}^{\lambda, \Delta \lambda}=1$.
'Off-the-shell' coefficients
can take any value and, in what follows, we use
$A_{k, k'}^{\lambda,\Delta\lambda}=0$ for
$\Theta_{kk'}^{\lambda, \Delta \lambda}=0$.
Note that
expression \eqref{G30} includes $U$ to all orders. However,
in the thermodynamic limit
$N \rightarrow \infty$, the coefficients
$A_{k, k'}^{\lambda,\Delta \lambda}$ become
linear in $U$ since all higher order terms vanish for $N\rightarrow \infty$.
In the actual numerical evaluation of the renormalization equations
on a lattice of finite
size $N$ the excitation energies
($\varepsilon_{k',\lambda}-\varepsilon_{k,\lambda}$)
may become very small for $\lambda \rightarrow 0$ so that the
expansion of $A_{k, k'}^{\lambda,\Delta \lambda}$ to
linear order in $U$ breaks down.
In this case, the full expression \eqref{G30} has to be taken.
By help of \eqref{G30} one finds for the parameters $v_{k k'}^\lambda$
of \eqref{G28}
\begin{eqnarray}
\label{G35}
\sin v_{k k'}^\lambda &=& 2 U \sqrt{\frac{C_\rho(k- k')}{N}}
\frac{\mbox{sign} \ (\varepsilon_{k,\lambda}
- \varepsilon_{k',\lambda}) }
{ \sqrt{(\varepsilon_{k, \lambda} -
\varepsilon_{k', \lambda})^2
+4U^2 C_\rho(k- k')/{N} }}
\nonumber \\
\cos v_{k k'}^\lambda &=&
\frac{|\varepsilon_{k, \lambda} -\varepsilon_{k', \lambda}|}
{ \sqrt{(\varepsilon_{k, \lambda} -\varepsilon_{k', \lambda})^2
+4U^2C_\rho(k- k')/{N}}} \ .
\end{eqnarray}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\vspace*{1cm}
Equations \eqref{G33} to \eqref{G39} represent the final
renormalization equations for the parameters of the Hamiltonian
as the energy cutoff is reduced from
$\lambda$ to $\lambda - \Delta \lambda$.
The overall renormalization starts from the original
cutoff $\lambda= \Lambda$,
where the initial parameters are given by \eqref{G19}. For the
parameters $ \sin v_{k k'}^\lambda$, $ \cos v_{k k'}^\lambda$,
and $C_\rho(k- k')$, which contain expectation values,
an appropriate choice is taken at cutoff $\Lambda$.
Using these quantities we obtain from the renormalization
equations \eqref{G33} to \eqref{G39}
the renormalized parameters at the smaller cutoff $\Lambda-\Delta\lambda$.
The procedure is repeated until the cutoff is reduced to zero.
At $\lambda =0$, we obtain the completely renormalized Hamiltonian
$\tilde{\cal H}:= {\cal H}_{(\lambda=0)}$, which describes an effectively
decoupled system of free conduction electrons and interacting
localized electrons,
\begin{eqnarray}
\label{G40}
\tilde{\cal H} &=&
\tilde{\cal H}_c + \tilde{\cal H}_{f} + \tilde{E} ,
\end{eqnarray}
where
\begin{eqnarray}
\label{G41}
\tilde{\cal H}_c &=&
\sum_k \tilde{\varepsilon}_k\;
c_k^\dagger c_k,
\hspace*{1cm} \mbox{and} \hspace*{1cm}
\tilde{\cal H}_{f}
=\tilde{\varepsilon}_f \sum_i \delta n_i + \sum_{i \neq j} \tilde{g}_{ij}\;
\delta n_i \delta n_j .
\end{eqnarray}
The fully renormalized parameters are
defined as $\tilde{\varepsilon}_k=
\varepsilon_{k, (\lambda=0)}, \tilde{\varepsilon}_f=
\varepsilon_{f,(\lambda=0)}, \tilde{g}_{ij} = g_{ij,(\lambda=0)}$
and $\tilde{E}= E_{(\lambda=0)}$.
Note that for $\lambda \rightarrow 0$
the interaction ${\cal H}_{1,\lambda}$ has completely
vanished due to the $\Theta$-function in
\eqref{G18}.
Using $\tilde{\cal H}$ we next can recalculate the statistical averages
$\big< c_k^\dagger c_k\big>$, $C_\rho(k- k')$,
etc., which appear in Eqs. \eqref{G33} -- \eqref{G39}, to find new
values for the renormalized coupling parameters. This procedure is repeated
until the self-consistency is reached.
Note that the Hamiltonian $\tilde{\cal H}_f$ for the localized
electrons is usually denoted as the lattice gas model.
Due to the dependence of $\tilde{g}_{ij}$ on
$i-j$, the interaction in $\tilde{\cal H}_f$
is not necessarily restricted to nearest neighbors.
Let us add two side remarks:
The first one concerns the validity of the present approach. As is obvious
from the renormalization procedure in Sect.~II we are here dealing with a
weak-coupling approach. The Coulomb interaction $U$ is taken as the
'small' quantity which is successively integrated out in the procedure.
However in each renormalization step not only the lowest order in $U$ is
included. Instead, certain renormalization
contributions up to infinite order in $U$ are taken into account.
Therefore, we expect the present approach to be valid for
$U$ values which are less or at most of the order of the hopping term $t$.
For larger values of $U$ either newly generated renormalization terms would
have to be added. Or, more appropriate for large $U$
would be to take the dominant $U$ term as part of a new 'unperturbed'
Hamiltonian ${\cal H}_0$ and the hopping term as part of the new perturbation
${\cal H}_1$.
The second remark applies to the optimal choice of the
renormalization interval
$\Delta\lambda$. It is best determined by the size of the system, i.e.,
by the requirement that the 'energy shell' contains only a few
$k$-points. Thus, the number of $\Delta \lambda$ intervals should be
of the order of the lattice size $N$. For a larger number of intervals the
result would be the same since more steps occur
without renormalization since no transition
energies fit into the interval. On the other hand a smaller
number of $\Delta \lambda$ intervals leads to a quite large
renormalization in some steps since many $k$-points could contribute
in these intervals.
In practise, for a given $N$ we have increased the number of intervals until
the renormalization does not change any more.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Evaluation of Static Expectation values}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
The renormalization equations depend on expectation values
$\big< n_i\big>$, $\big< n_i n_j \big>$ and $\big<c_k^\dagger
c_k \big>$ which have been defined with
the full Hamiltonian ${\cal H}$. Since thermal averages are invariant with
respect to a unitary transformations, we can do the averaging
with the renormalized Hamiltonian $\tilde{\cal H}$, provided we
also transform the operators,
\begin{eqnarray}
\label{G42}
\big< {\cal A} \big> &=& \big< {\cal A}(\lambda) \big>_{{\cal H}_\lambda}
= \big< {\cal A}(\lambda \rightarrow 0) \big>_{\tilde{{\cal H}}} ,
\hspace*{1cm}
{\cal A}(\lambda)= e^{X_{\lambda}} {\cal A} e^{-X_{\lambda}}. \\
&& \nonumber
\end{eqnarray}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%\subsection{Conduction electrons}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
To evaluate $\big< c_k^\dagger c_k \big>$ one
needs the renormalization equations for the operators
$(c_k^\dagger c_k)(\lambda)= c_k^\dagger(\lambda) \;
c_k(\lambda)$, where
$c_k(\lambda)
=
e^{X_{\lambda, \Delta \lambda}}c_k
e^{-X_{\lambda, \Delta \lambda}}$.
For $c_k(\lambda)$ we use the following ansatz,
\begin{eqnarray}
\label{G45}
c_k(\lambda) &=& \alpha_{k,\lambda} c_k
+ \frac{1}{N} \sum_{k',i} \beta_{k,k',\lambda}\;
e^{i(k -k'){R}_i} \; \delta n_i c_{k'} ,
\end{eqnarray}
where $\alpha_{k,\lambda}$ and $\beta_{k,k',\lambda}$
are $\lambda$-dependent coefficients. The operator structure
of $c_k(\lambda)$ in \eqref{G45} is again taken from the
first order expansion in $U$.
All higher order terms are neglected.
The renormalization equations for the new parameters
can be derived in close analogy to the renormalization
of ${\cal H}_\lambda$. One finds,
\begin{eqnarray}
\label{G46}
\alpha_{k, (\lambda -\Delta \lambda)} &=&
\alpha_{k, \lambda}
- \alpha_{k, \lambda}
\sum_{k'} \big[ 1 - \cos\big(A_{k k'}^{\lambda,\Delta \lambda}
\sqrt\frac{C_\rho(k-k')}{N} \big)\big] \;
\Theta_{kk'}^{\lambda, \Delta \lambda} \\
&+&
\frac{1}{\sqrt{N}} \sum_{k'}
\beta_{k k', \lambda} \;
\Theta_{kk'}^{\lambda, \Delta \lambda}
\sqrt{C_\rho(k -k')}
\sin\big(A_{k k'}^{\lambda,\Delta \lambda}
\sqrt\frac{C_\rho(k-k')}{N} \big)
\nonumber
\end{eqnarray}
and
\begin{eqnarray}
\label{G47}
\beta_{kk', (\lambda -\Delta \lambda)} &=&
\beta_{kk', \lambda}
- \alpha_{k, \lambda} \;
\sqrt{\frac{N}{C_\rho({k-k'})}}
\sin\big(A_{k k'}^{\lambda,\Delta \lambda}
\sqrt\frac{C_\rho(k-k')}{N} \big)\big] \;
\Theta_{kk'}^{\lambda, \Delta \lambda}
\nonumber \\
&-&
\beta_{kk', \lambda} \; \big( 1 -
\cos\big[A_{k k'}^{\lambda,\Delta \lambda}
\sqrt\frac{C_\rho(k-k')}{N} \big] \big) \;
\Theta_{kk'}^{\lambda, \Delta \lambda} ,
\end{eqnarray}
where
$A_{k, k'}^{\lambda,\Delta\lambda}$
and
$C_\rho(k-k')$ are defined by Eq.~\eqref{G30}
and \eqref{G28}, respectively.
The initial values are again those of the original
model ($\lambda = \Lambda$), $ \alpha_{k, \Lambda} = 1,
\beta_{k k', \Lambda} =0$.
Eqs.~\eqref{G45} -\eqref{G47}
determine the transformation behavior of $c_k(\lambda)$.
By the use of \eqref{G45} we finally obtain
\begin{eqnarray}
\label{G49}
n_k= \big< c_k^\dagger c_k\big> &=&
|\tilde{\alpha}_{k}|^2
\big< c_k^\dagger c_k\big>_{\tilde{\cal H}_c}
+
\frac{1}{N} \sum_{k'} |\tilde{\beta}_{k,k'}|^2
C_\rho(k-k')
\big< c_{k'}^\dagger c_{k'}\big>_{\tilde{\cal H}_c} ,
\end{eqnarray}
where
$ \tilde{\alpha}_{k} = \alpha_{k, (\lambda=0)}$
and
$\tilde{\beta}_{k,k'} =
\beta_{k k', (\lambda=0)}$,
and $\big< c_k^\dagger c_k\big>_{\tilde{\cal H}_c}$
is the Fermi function formed with
$\tilde{\cal H}_c$,
$ \big< c_k^\dagger c_k\big>_{\tilde{\cal H}_c}
= ({e^{\beta \tilde{\varepsilon}_k}+1})^{-1}$. \\
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% \subsection{Localized electrons}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
The occupation number operators $n_i$
of the localized electrons commute with $X_\lambda$.
Thus $n_i$ will not be renormalized, $n_i(\lambda) = n_i$,
$(n_i n_j) (\lambda) = n_i n_j$, i.e.,
\begin{eqnarray}
\label{G52}
n_f= \big< n_i \big> &=& \big< n_i \big>_{\tilde{{\cal H}_f}} \hspace*{1cm}
\big< \delta n_i \delta n_j\big> =
\big< \delta n_i \delta n_j \big>_{\tilde{{\cal H}_f}}.
\end{eqnarray}
Here the expectation values are formed with
the lattice gas model $\tilde{\cal H}_f$ since the conduction
and the localized part of the renormalized
Hamiltonian $\tilde{\cal H}$ are decoupled.
The expectation values in \eqref{G52} can be found by direct numerical
evaluation in the local occupation number representation.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Dynamic correlation functions\label{dynamical properties}}
Dynamical quantities can be evaluated by the PRM approach along the same
lines. In what follows we shall
discuss the spectral functions of the conduction and
localized electrons.
%\subsection{Conduction electrons}
The one-particle spectral functions of conduction electrons
$A_k^+(\omega)$ and $A_k^-(\omega)$
are defined as
\begin{eqnarray}
\label{G56}
A_k^+(\omega) &=& \frac{1}{2\pi} \int_{-\infty}^{\infty}
\big< c_{k } (t) \;c_{k}^\dagger
\big> \; e^{i\omega t} dt\,, \hspace*{1cm}
A_k^-(\omega) = \frac{1}{2\pi} \int_{-\infty}^{\infty}
\big< c_{k }^\dagger \;c_{k}(t)
\big> \; e^{i\omega t} dt\, ,
\end{eqnarray}
where $A_k^+(\omega)$ describes the creation of an electron
$k$ at time zero and its annihilation at time $t$ whereas in
$A_k^-(\omega)$ first an electron is annihilated.
It is well-known that $A_k^+(\omega)$
and $A_k^-(\omega)$ can be measured
by inverse photoemission (IPE) and by photoemission (PE) experiments.
To evaluate $A_k^+(\omega)$ and $A_k^-(\omega)$ we again
exploit the property \eqref{G42} of expectation values and the
ansatz \eqref{G45} for $c_k(\lambda)$ and
$c_k^\dagger(\lambda)$. We find
\begin{eqnarray}
\label{G58}
A_k^+(\omega) &=& |\tilde{\alpha}_k|^2 \;
(1 - \big< c_k^\dagger c_k\big>_{\tilde{\cal H}_c})\;
\delta(\omega -\tilde{\varepsilon}_k)
+ \frac{1}{N} \sum_{k'} |\tilde{\beta}_{k k'}|^2\;
C_\rho(k-k')(1-
\big< c_{k'}^\dagger c_{k'} \big>_{\tilde{\cal H}_c})\;
\delta (\omega - \tilde{\varepsilon}_{k'}) \nonumber \\
&& \\
A_k^-(\omega) &=& |\tilde{\alpha}_k|^2 \;
\big< c_k^\dagger c_k\big>_{\tilde{\cal H}_c}\;
\delta(\omega -\tilde{\varepsilon}_k)
+ \frac{1}{N} \sum_{k'} |\tilde{\beta}_{k k'}|^2\;
C_\rho(k-k')\;
\big< c_{k'}^\dagger c_{k'} \big>_{\tilde{\cal H}_c}\;
\delta (\omega - \tilde{\varepsilon}_{k'}) \nonumber
\end{eqnarray}
where $\big< c_{k'}^\dagger c_{k'} \big>_{\tilde{\cal H}_c}$
is the Fermi distribution. Note that in \eqref{G58}
we have already reduced the cutoff to $\lambda \rightarrow 0$.
The
first term in $A_k^+(\omega)$ and $A_k^-(\omega)$ describes
the coherent one-electron excitation
which is the only excitation in a free electron gas model.
The remaining contributions are incoherent excitations
due to the coupling of $c_k$
to different conduction electrons with wave
vectors $k'$. The coupling is mediated by density
fluctuations of local electrons which lead to the appearance of the
charge susceptibility $C_\rho(k -k')$ of the localized
electrons in \eqref{G58}.
The spectral sum rule can be found by integrating
($A_k^+(\omega)+ A_k^-(\omega)$) over $\omega$
and using $[c_k^\dagger,c_k]_+=1$,
\begin{eqnarray}
\label{G59}
1= \int_{-\infty}^{\infty} d\omega\; (A_k^+(\omega) +
A_k^-(\omega)) &=&
|\tilde{\alpha}_k|^2 + \frac{1}{N}\sum_{k'}
|\tilde{\beta}_{k k'}|^2 C_\rho(k -k') .
\end{eqnarray}
This relation can also be derived directly from \eqref{G45}.\\
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%\subsection{Localized electrons}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Next, we evaluate the averaged one-particle spectral functions
$B^+(\omega)$ and $B^-(\omega)$ of localized electrons,
\begin{eqnarray}
\label{G60}
B^+(\omega) &=& \frac{1}{2\pi} \int_{-\infty}^{\infty}
\big< f_i (t) \;f_{i}^\dagger
\big> \; e^{i\omega t} dt\, \hspace*{0.8cm}
B^-(\omega) =
\frac{1}{2\pi} \int_{-\infty}^{\infty}
\big< f_{i}^\dagger \;f_{i}(t) \big>
\; e^{i\omega t} dt .
\end{eqnarray}
Since the local $f$-charge is a constant of motion,
all correlation functions between different sites $i$ and $j$ vanish.
For the evaluation we again use \eqref{G42} and the following ansatz
for the transformed $f$-operator
\begin{eqnarray}
\label{G61}
f_{i}(\lambda) &=& \nu_{\lambda} f_{i}
+ \frac{1}{N} \sum_{{k k'}} \mu_{k,k',\lambda}\;
e^{i(k -k'){R}_i} \; \delta (c_k^\dagger
c_{k'}) f_i ,
\end{eqnarray}
where $\nu_{\lambda}$ and $\mu_{k,k',\lambda}$
are the $\lambda$-dependent coefficients. As before, the operator structure
of the second term is assumed to agree with the one obtained from the
first order contribution in $U$ and higher order terms have been neglected.
The initial parameter values are ($\lambda = \Lambda$): $ \nu_{\Lambda} = 1,
\mu_{k k', \Lambda} =0$. The renormalization equations for
$\nu_{\lambda}$ and $\mu_{k,k',\lambda}$ can easily be derived.
In the limit $\lambda\to 0$ we find the following result
%%%
\begin{eqnarray}
\label{G65}
B^+(\omega) &=& \tilde{\nu}^2 B_0^+(\omega)
+
\frac{1}{N^2} \sum_{k k'} |\tilde{\mu}_{k k'}|^2
f(\tilde{\varepsilon}_k) (1- f(\tilde{\varepsilon}_{k'}))
B_0^+(\omega + \tilde{\varepsilon}_k -
\tilde{\varepsilon}_{k'})\\
&& \nonumber \\
\label{G66}
B^-(\omega) &=& \tilde{\nu}^2 B_0^-(\omega)
+
\frac{1}{N^2} \sum_{k k'} |\tilde{\mu}_{k'k}|^2
f(\tilde{\varepsilon}_k) (1- f(\tilde{\varepsilon}_{k'}))
B_0^-(\omega -( \tilde{\varepsilon}_k -
\tilde{\varepsilon}_{k'}) ) .
\end{eqnarray}
We have introduced new correlation functions
\begin{eqnarray}
\label{G67}
B^+_0(\omega) &=& \frac{1}{2\pi}
\int_{-\infty}^{\infty} dt e^{i\omega t}\;
\big< f_i(t) f_i^\dagger \big>_{\tilde{\cal H}_f}
\hspace*{2cm}
B^{-}_0(\omega) = \frac{1}{2\pi}
\int_{-\infty}^{\infty} dt e^{i\omega t}\;
\big< f_i^\dagger f_i(t) \big>_{\tilde{\cal H}_f}. \nonumber \\
&&
\end{eqnarray}
which are defined with the lattice gas model $\tilde{\cal H}_f$.
In Appendix B the functions $B_0^{\pm}(\omega)$
are evaluated by use of the retarded one-$f$ electron
Green function ${\cal G}_0(\omega)$, defined with $\tilde{\cal H}_f$,
\begin{eqnarray}
\label{G68}
{\cal G}_0(\omega) &=& i
\int_0^\infty dt e^{i(\omega +i \eta) t} \;
\big< [f_i(t), f_i^\dagger]_+ \big>_{\tilde{\cal H}_f}
\hspace*{2cm} (\eta =0^+).
\end{eqnarray}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Results}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Static properties}
We now discuss the results obtained by numerical evaluation
of the renormalization equations for the one-dimensional lattice. Note
that in the following all energy values are given in
units of the hopping matrix element $t$. One first has to
calculate the fully renormalized parameters
$\tilde{\varepsilon}_k$, $\tilde{\varepsilon}_f$, and
$\tilde{g}_{ij}$ which determine the effective Hamiltonian
$\tilde{\cal H}= \tilde{\cal H}_c + \tilde{\cal H}_f + \tilde{E}$
of \eqref{G41}.
All static expectation values which enter the renormalization equations
are evaluated self-consistently with the full Hamiltonian,
following the procedure
outlined in Sections III and IV.
The effective Hamiltonian $ \tilde{\cal H}_{f} $ describes the
lattice gas model of ions which interact via a possibly long range
interaction $\tilde{g}_{ij} $.
Thus, we use an additional Monte Carlo simulation to evaluate the expectation values
$n^{f} = \left\langle n_{i} \right\rangle_{\tilde{\cal H}_{f}} $ and
$C_{ff}(\kappa)= \left\langle \delta n_{i} \delta n_{j} \right\rangle =
\left\langle \delta n_{i} \delta n_{j}\right\rangle_{\tilde{\cal H}_{f}}$,
which enter the renormalization equations together with
$\left\langle c_k^{\dag} c_k \right \rangle_{\tilde{\cal H}_{c}}
=[1+ \displaystyle e^{\beta \tilde{\varepsilon}_k }]^{-1}$.
As already mentioned, we start by making a guess for the initial values of $n_f $,
$C_\rho(k- k')$ and $\big< c_k^\dagger c_k \big>$,
and calculate the renormalized parameters for $\lambda \rightarrow 0$.
That is, we find $\tilde{\varepsilon}_f=\varepsilon_{f,\lambda=0}$,
$\tilde{\varepsilon}_k= \varepsilon_{k, \lambda=0}$,
and $\tilde{g}_{q}= g_{{q}, \lambda=0}$
by solving numerically the renormalization equations \eqref{G33}--\eqref{G39}.
Next, using the effective coupling $\tilde{g}_{q}$ we find the new values
of the spatial interactions $\tilde{g}_{ij}$ and then recalculate
the localized electron expectation values $n_f$ and $C_{ff}(\kappa)$
using a classical Monte Carlo simulation.
The Fourier transform of $C_{ff}(\kappa)$ gives an approximate
expression for $C_\rho(k-k')$.
Using new values of $n_f $, $C_\rho(k-k')$ and
$\big< c_k^\dagger c_k \big>$
we iterate \eqref{G33} -- \eqref{G39} until a self-consistent solution is reached.
In the following we consider only the one-dimensional
Falicov-Kimball model at half-filling,
\begin{equation}
\label{G75}
n_{f} + n_{c} = 1 ,
\end{equation}
where
$n_f = \langle n_i \rangle$ and $n_c= 1/N \sum_k\langle
c_k^\dagger c_k\rangle $
is the averaged local occupation numbers of conduction electrons.
Condition \eqref{G75} represents a point of special interest for
valence and possibly metal-insulator transitions, caused by the promotion
of electrons from localized $f$ orbitals ( $f^n \rightarrow f^{n+1}$ )
to the conduction band states\cite{Ott}.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Ergebnisse.tex (Sykora):
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsubsection{Temperature dependence}
In Fig.~\ref{fig:1} the averaged local $f$-occupation number
$n_f$ and the renormalized
$f$-energy $\tilde{\varepsilon}_f$ from the PRM approach
are shown as function of temperature for $U=1.0$
and several values of unrenormalized
{\it f}-level position $\varepsilon_f$.
Note, the chemical potential $\mu$ has to be adjusted
at each temperature $T$, so as to fulfill condition \eqref{G75}.
In the evaluation procedure this requires extra numerical cycles,
in addition to those from the self-consistent determination of the expectation values.
Therefore, only a rather small system of 20 lattice sites has been used
to generate Fig.~\ref{fig:1}.
%
Also the considerably
increased effort in the MC sampling
for the lattice gas part in ${\cal H}_{0,\lambda}$
at higher temperatures suggest to restrict oneself to a rather
small system. This will be discussed below in some more detail.
%
As expected, for small values of the bare $f$-level $\varepsilon_f$
(dashed curve in Fig.~\ref{fig:1}a),
the average $f$-occupation $n_f$ increases with decreasing temperature
and tends to $1$ at $T=0$. The corresponding values of the renormalized
$f$-level remain below the chemical potential ($\tilde{\varepsilon}_f <0$
in Fig.~\ref{fig:1}b), so that all the sites are occupied with one
$f$-electron.
In contrast, for large values of $\varepsilon_f$ (solid curves) $n_f$ tends
to zero for $T \rightarrow 0$ and $\tilde{\varepsilon}_f$ stays above the
chemical potential ($\tilde{\varepsilon}_f >0$). Thus all the $f$-sites are
empty and the renormalized conduction electron band is completely filled.
Finally, at intermediate values of the bare $\varepsilon_f$-value (dashed
dotted curves)
the $f$ occupation at $T=0$ tends to some value between $0$ and $1$,
signalizing an intermediate $f$-valence.
In that case, the renormalized $f$-level approaches the chemical potential
($\tilde{\varepsilon}_f =0$) at $T=0$. Below the chemical potential
all the conduction electron states are occupied and $f$-states only partially,
due to the condition $n_f + n_c=1$.
%%%%%%%%%%%%%%%%%%%%%%% Fig. 1 %%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\begin{center}
\includegraphics[width=8cm,height=6cm]{Erg_nf_T_2a.eps}
\includegraphics[width=8cm,height=6cm]{Erg_epsf_T_2b.eps}
\end{center}
\caption{(a) $f$-electron occupation number $n_f$
as function of temperature for $U=1.0$ and for three different
values of the bare $f$-level $\varepsilon_{f} = 0.5$ (dashed),
$\varepsilon_{f} = 1.2$ (dashed dotted), and $\varepsilon_f=3.5$ (solid).
The open circles ($\varepsilon_{f} = 0.5$)
and open squares ($\varepsilon_{f} = 1.2$)
refer to ED results of reference \cite{QMC}. Note the
good agreement with the PRM results. The values of $\varepsilon_{f}$
are measured from the lower edge of the conduction band. (b) Renormalized
$f$-electron energy $\tilde{\varepsilon}_f$
as function of temperature for the same values of $U$ and
$\varepsilon_{f}$-values as in panel (a). }
\label{fig:1}
\end{figure}
\subsubsection{Low temperature properties at fillings $n_{f} = 1/2$ and
$n_{f} = 1/3$}
%%%%%%%%%%%%
Next, we discuss static properties for $n_f= 1/2$ and $1/3$ at low temperatures.
The parameters $\mu$ and $\varepsilon_{f}$ have to be adjusted in such a
way that the condition \eqref{G75} is fulfilled for a given value of $n_f$.
Note that in a Monte Carlo simulation
a nonzero temperature has to be used, which is set equal to
$k_{B} T= 0.1$. Also the most dominant
ground-state configurations have been incorporated
in the Monte-Carlo simulation
so that the system size could be extended to $N = 150$ sites.
For $n_{f} = 1/2$ the chemical potential $\mu$ is in the middle of
the renormalized conduction band $\tilde{\varepsilon}_k$.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Fig 2 %%%%%%%%%%%%%%%%%%
\begin{figure}[h]
\begin{center}
\includegraphics[width=8cm,height=6cm]{Erg_ren_e_c_nf05a.eps} \vspace*{0.9cm} \hspace*{1mm}
\includegraphics[width=8cm,height=6cm]{Erg_ren_e_c_nf03b.eps}
\includegraphics[width=8cm,height=6cm]{Erg_erw_e_c_nf05c.eps} \hspace*{1mm}
\includegraphics[width=8cm,height=6cm]{Erg_ren_e_c_T.eps}
\end{center}
\caption{(a) Renormalized conduction electron energy
$\tilde{\varepsilon}_k$ for different Coulomb
repulsions $U$ ($= 0.1,1.0,2.0,3.0$) and $n_{c} = n_{f}= 1/2$.
Note the appearance of a gap of magnitude $U$ at the Fermi momentum $k_F
= \pi/2$.
(b) Same quantity
$\tilde{\varepsilon}_k$ for $n_{c} = 2/3$ ($n_f= 1/3$) for
$U= 0.5, 1.0, 2.0, 2.5$. $\tilde{\varepsilon}_k$
shows two gaps where the one at the
Fermi momentum $k_F=(2/3) \pi$ is now smaller than $U$.(c) Momentum
distribution $n_k= \langle c_{k}^{\dag} c_k\rangle $ for the same
parameters as in panel (a) and $n_{f} = n_{c}= 1/2$.
The smooth behavior of $n_k$
at the Fermi momentum follows from the gap in
$\tilde{\varepsilon}_k$.
(d) $\tilde{\varepsilon}_k$ for two different
temperatures $T=0.1$ (solid line) and
$T=0.2$ (dashed dotted line) and $U=2.0$, $n_f=1/2$.
Note that the gap at $k_F$ for $T=0.1$
is completely smeared out for the higher temperature $T=0.2$.}
\label{fig:5}
\end{figure}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
In Fig.~\ref{fig:5}c the result for $n_k$ obtained from the PRM
is plotted as function of $k$ for $n_c=1/2$
and for different values of the Coulomb repulsion $U$.
The results for $n_k$ from the PRM approach are very similar to
those from Ref.~\cite{farkasovsky.03} (for the same
values of $U$).
In Fig.~\ref{fig:5}a
the renormalized one-particle energy
$\tilde{\varepsilon}_k$ is shown as function of momentum $k$
for $n_c=1/2$ ($n_f =1/2$), and
for the same interaction values $U$ as in Fig.~\ref{fig:5}c.
Note, that $\tilde{\varepsilon}_k$ can be considered as
the excitation energy of a quasiparticle of the original interacting model
\eqref{G1}. This follows from the diagonal form of
$\tilde{\cal H}_c$ in \eqref{G41} and the fact that $\tilde{\cal H}$
is obtained from the original model by a unitary transformation.
The main feature of $\tilde{\varepsilon}_k$
is the appearance of
a gap at the Fermi level, which opens for any, even small, value
of the repulsion energy $U$. Such a gap was recently anticipated by
Farkasovsky \cite{farkasovsky.03} from the correspondence
of the Falicov-Kimball model to
the Tomonaga-Luttinger fermions. There, the smooth behavior of
$n_k$ with $k$ also gives rise to a
gap in the charge excitations.
Note that for $n_c=2/3$ ($n_f=1/3$) (Fig.~\ref{fig:5}b)
again a gap opens at the Fermi
energy for $k_F = (2/3) \pi$ for all values of $U$. Thus, the
Luttinger theorem is fulfilled even for the present case of an insulating
state. By closer inspection of Figs.~\ref{fig:5}a,b one finds that the gaps
at $k_F$ scale linearly with $U$ both for $n_c=1/2$ and $n_c=2/3$. The
proportionality constant is almost $1$ for $n_c=1/2$ whereas for $n_c=2/3$
it is somewhat smaller than $1$. \vspace{0.5cm}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Fig 3 %%%%%%%%%%%%%%%%%%%%
\begin{figure}[h]
\begin{center}
\includegraphics[width=8cm,height=6cm]{Erg_erw_nij_nf05a.eps}
\includegraphics[width=8cm,height=6cm]{nij.eps}
\end{center}
\caption{(a) Correlation function $C_{ff}(\kappa)= \langle \delta n_{i}
\delta n_{i+\kappa} \rangle$ of localized electrons for
$n_f =1/2$. For $n_f =1/3$ we obtain an analogous result. The length
$\kappa$ characterizes the distance between the sites. The results
correspond to averages $(1/N)\sum_i \langle \delta n_{i}
\delta n_{i+\kappa} \rangle$ over all lattice sites (see text). (b)
Correlation function $C_{ff}(\kappa) =\langle \delta n_{i}
\delta n_{i+\kappa} \rangle$ of localized electrons for a chain with
20 lattice sites for three different temperatures $T=0.20,
0.13, 0.07$ and $U = 1.0$ and
$n_{f} = 1/2$.}
\label{fig:7}
\end{figure}
%%%%%%%%%%%%
Fig.~\ref{fig:7}a shows the spatial correlation functions
$C_{ff}(\kappa)= \langle \delta n_{i}
\delta n_{i+\kappa} \rangle$ between the localized electrons for the same
parameters as in Fig.~\ref{fig:5}a.
The index $\kappa$ denotes the distance
between the two involved lattice sites. Note that for
$n_f=1/2$ the values of the correlation function
are always the same at every second site ($\kappa= 2,4,6, \cdots$) whereas
for all sites in between ($\kappa= 1, 3, 5, \cdots $) they differ only by a
minus sign.
\subsubsection{Discussion}
Note that the long-range correlations develop
already for arbitrarily small values of $U$. This behavior should
be related to the long range order of the
one-dimensional Falicov-Kimball model at zero temperature~\cite{kennedy.86}.
Note that in the present calculations the
temperature is chosen small but
finite ($k_BT= 0.1$), so that in the strict sense no infinitely long
range order exists. The origin of the long-range correlations
can easily be understood from a low
order $U$ expansion of the effective $f$-coupling.
For the Fourier transform $g_{q}$ of the local quantity
$g_{ij}$ one finds by use of \eqref{G39}, \eqref{G28} and \eqref{G30}
\begin{eqnarray}
\label{G75a}
g_{{q}, (\lambda - \Delta \lambda)} &=& g_{{q}, \lambda}
- \frac{U^2}{2N} \sum_k
\frac{\big< c_k^\dagger c_k\big> -
\big<c_{k+{q}}^\dagger c_{k+{q}}\big>}
{\varepsilon_{k+{q},\lambda} - \varepsilon_{k,
\lambda}} \;
\Theta_{k, k+{q}}^{\lambda, \Delta \lambda} \\
&& \hspace* {1cm} +
\frac{U^2}{2N^2} \sum_{k, {q'}}
\frac{\big< c_k^\dagger c_k\big> -
\big<c_{k+{q'}}^\dagger c_{k+{q'}}\big>}
{\varepsilon_{k+{q'},\lambda} -
\varepsilon_{k, \lambda}} \;
\Theta_{k, k+{q'}}^{\lambda, \Delta \lambda} \nonumber
\end{eqnarray}
where the last term excludes equal sites $i=j$ in $g_{ij}$.
As follows from \eqref{G75a}, the interaction between the localized electrons
is caused from the coupling to particle-hole excitations
${\varepsilon_{k+{q},\lambda} - \varepsilon_{k,
\lambda}}$ where the wavevector runs over the whole Brillouin zone. Here,
the energies $\varepsilon_{k,\lambda}$ and
$\varepsilon_{k+{q},\lambda}$ have to be either below or above the Fermi
level, i.e. $|k|< k_F$ and $|k+q|> k_F$ or $|k|>k_F$ and $|k+q|< k_F$.
Note that for $q=\pi$ and $n_f=1/2$
this condition is fulfilled for all $k$-values
of the Brillouin zone so that the most dominant renormalization of
$g_{{q}}$ occurs for $q=\pi$.
In contrast, for ${q}$-values different from $\pi$ always less
$k$ terms contribute.
For instance, for $q\approx 0$ only $k$ points from the sum in
\eqref{G75a} can contribute which are located in a small
region around the Fermi momentum $k_F$. As a result of
the dominant coupling $g_{q}$ at $q=\pi$ one is immediately lead
to a infinitely ranged spatial correlation between the
$f$-electrons having an alternating sign from site to site.
This explains the long-range behavior in Fig.~\ref{fig:7}a.
Note again that this feature is not based
on a strong-coupling argument. Instead,
the long-ranged correlations are observed already for
small values for $U$. \\
The alternating behavior of $C_{ff}(\kappa)= \big< \delta n_i
\delta n_{i+\kappa} \big>$ with $\kappa$ can be interpreted as follows:
At zero temperatures the model has two equivalent
degenerate ground states of period two (for $n_f=1/2$),
separated by a potential barrier, with the arrays $\{1010 . . . 10\}$ and
$\{0101 ... 01\}$ of the localized ions. Therefore, we might use
the following ansatz for the state vector of the system
at very low but finite temperatures,
\begin{eqnarray}
\label{G76}
|\Phi_g^{(2)}\rangle = \frac{1}{\sqrt{2}} \big(
|10101 ... 10 \rangle + |0101 ... 01 \rangle
\big) \ .
\end{eqnarray}
In order to obey translation invariance the sum of the
two generated ground states have been taken and
ground-state fluctuations have been neglected. Note that the
two states are never connected by nonvanishing matrix elements when physical
quantities are evaluated.
To proceed, we use \eqref{G76} to evaluate the correlation function
$\langle n_i n_{i+\kappa} \rangle$ of Fig.~\ref{fig:7}. As explained,
the result corresponds to the average $(1/N)\sum_i
\langle n_i n_{i+\kappa} \rangle$ over all lattice sites and is independent of
$i$. From ansatz \eqref{G76} one finds
\begin{equation}
\label{G77}
\langle n_i n_{i+\kappa} \rangle = \left\{
\begin{array}{lll}
\frac{1}{2} & \mbox{:} & \kappa= 2, 4, 6, ... \\
0 & \mbox{:} & \kappa= 1, 3, 5, ...
\end{array}
\right.
\end{equation}
With $\big< n_i \big> =1/2$ one obtains for
$C_{ff}(\kappa) = \langle \delta n_i \;
\delta n_{i+\kappa} \rangle$
\begin{equation}
\label{G78}
C_{ff}(\kappa) = \left\{
\begin{array}{rll}
\frac{1}{4} & \mbox{:} & \kappa= 2, 4, 6, ... \\
- \frac{1}{4} & \mbox{:} & \kappa= 1, 3, 5, ...
\end{array}
\right.
\end{equation}
Obviously the result \eqref{G78} agrees very well with the
outcome from the PRM approach according to Fig.~\ref{fig:7}a
for small but also for rather large values of $U$. A similar reasoning may
also be given for $n_f =1/3$ ($n_c=2/3$).
Here, the degenerate ground-states have period three with arrays $\{100100 ... 100 \}$,
$\{010010 ... 010 \}$, and $\{001001 ... 001 \}$ in analogy to \eqref{G76}.
Note that the amplitude of the $f$-correlation function
$\langle \delta n_i \; \delta n_{i+\kappa} \rangle$ is almost independent of
$U$. Already for small $U$-values the
$f$-electrons are ordered according to the qualitative description
of Fig.~\ref{fig:7}a. The behavior of the conduction
electrons is quite different.
The correlation function $C_{cf}(\kappa)=
\big < n_i^c n_{i+\kappa}\big>$ describes the spatial correlation
between a conduction electron at site $i$ and an $f$ electron at
site $i+\kappa$. An evaluation for $C_{cf}(\kappa)$
correct up to first order in $U$ gives
\begin{eqnarray}
\label{G78a}
C_{cf}(\kappa) &=& \frac{n_f}{N} \sum_{k}
|\tilde{\alpha}_k|^2 \big< c_k^\dagger c_k \big>_{\tilde{\cal H}_c}
+
\frac{2}{N} (-1)^\kappa \sum_k \tilde{\alpha}_k
\tilde{\beta}_{k+ {Q}}
\big< c_k^\dagger c_k \big>_{\tilde{\cal H}_c}
+ O(U^2)
\end{eqnarray}
$(Q=\pi)$, where the renormalized parameter $\tilde{\beta}_{k+ {Q}}
\sim U$ is
negative (compare \eqref{G47}). Obviously, the second term in \eqref{G78a}
describes an oscillating behavior with $\kappa$ of
$cf$-correlations which favour the presence of conduction
electrons in between the $f$ sites. However,
in contrast to the oscillating behavior of $C_{ff}(\kappa)$,
the amplitude of this charge-density
wave-like behavior of the correlation function is proportional to $U$. \\
For an interpretation of the temperature dependence of
$C_{ff}(\kappa)$ note that the two
arrays $\{1010 . . . 10\}$ and
$\{0101 ... 01\}$ (for $n_f=1/2$) of the localized ions
are equivalent to the two N\'eel states $|\uparrow \downarrow
\uparrow \downarrow \cdots \uparrow \downarrow \rangle$ and
$|\downarrow \uparrow \downarrow \uparrow \cdots
\downarrow \uparrow \rangle$
of the one-dimensional Ising antiferromagnet. For the latter model
the spatial spin correlation function $\langle S_i^z S_{i+\kappa}^z\rangle$
decays exponentially at low temperatures
as $\exp{(-\kappa/\xi)}$ due to excitation of kinks or
domain walls. The correlation length is given by
$ \xi \sim \exp{(\beta |J_z|/2)}$, which shows that it becomes infinite
only exactly at zero temperature. $J_z/2$ is the characteristic
energy for the creation of domain walls \cite{nagaosa.98}.
For the one-dimensional Falicov-Kimball model a quite similar behavior is
expected for the correlation function $C_{ff}(\kappa)$. Here,
also domain wall states can be thermally excited.
They are obtained by interchanging the positions of neighboring $f$- and
conduction electrons. In order to investigate the
temperature dependence of the correlation length
we have evaluated
$C_{ff}(\kappa)= \langle \delta n_{i} \delta n_{i+\kappa} \rangle$
for different temperatures $T$ (Fig.~\ref{fig:7}b).
Note that for higher temperatures excited states would have to be included
in the MC-sampling which leads to considerable numerical increase.
Therefore, instead of taking a system with 150 lattice sites as before
a rather small system with only 20 sites was investigated where an
exact diagonalization was performed.
Fig.~\ref{fig:7}b shows that
for the two higher temperatures $T=0.20$ and $T=0.13$
the correlation function decays roughly exponentially
to zero within a few lattice sites whereas for the lowest
temperature $T=0.07$ no decay is observed. Thus, in the
latter case the correlation length $\xi$ obviously exceeds the length of
the lattice.
The same feature was already found in Fig.~\ref{fig:7}a, where no
decay was observed up to $\kappa_{max} \approx 70$.
We can conclude that for the lowest temperature
no excited states
contribute to the correlation function for the given lattice sizes.
Note that this result is not a consequence of approximations
used in the present PRM approach. Instead at the lowest temperature
the correlation length is larger than the
lattice extension but should be still finite.
Only exactly at $T=0$ the correlation
length $\xi$ should be infinite. An equivalent behavior is also found for the
itinerant electron-dispersion $\tilde{\varepsilon}_k$ as function of
temperature.
In Fig.~\ref{fig:5}d the dispersion of $\tilde{\varepsilon}_k$
is shown for two
temperatures $T=0.1$ and $T=0.2$ for $U=2.0$ and $n_f = 1/2$.
Whereas at $T=0.1$ the gap at $k_F$ is clearly seen it is
interesting to note that it is already smeared out for $T=0.2$. \\
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Note that an infinitely large correlation length $\xi$ leads for the
Fourier transform $C_\rho({q})$ of $C_{ff}(\kappa)$ to
$C_\rho({q}) \sim N \delta_{q, \pi}$.
Instead, for a finite correlation length $\xi$ one would find a 'softened'
Kronneker function around ${q}=\pi$ within a range of the
order of the inverse correlation length $\xi^{-1}$.
Thus the gap in the one-particle
excitation energy $\tilde{\varepsilon}_k$ of Figs.~\ref{fig:5}a,b
also softens over a range in $k$-space of the same order $\xi^{-1}$.
It folllows that the system is metallic at arbitrarily small
temperature. The softening of the gap could hardly be seen
in Figs.~\ref{fig:5}a,b for a correlation length larger than the lattice
extension of, may be, 100 times the lattice constant.
Thus, the results for $\tilde{\varepsilon}_k$ and other quantities
are extremely good approximations also for the $T=0$ limit. \\
%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Dynamical properties at low $T$}
Next, we evaluate the dynamical properties using the PRM approach.
First we consider the electronic density of states (DOS)
of conduction electrons
\begin{eqnarray}
\label{G79}
\rho_c(\omega) &=& \frac{1}{N} \sum_k (A_k^+(\omega)
+ A_k^-(\omega)) = \frac{1}{N} \sum_k
\frac{1}{2\pi} \int_{-\infty}^\infty
\langle \; [\; c_k(t), c_k^\dagger\; ]_+ \;\rangle e^{i\omega t}dt.
\end{eqnarray}
where $A_k^+(\omega)$ and $A_k^-(\omega)$ were defined in
\eqref{G56}.
The results from the PRM approach for the neutral case $n_f=n_c=1/2$
are shown in Figs.~\ref{fig:8}a and \ref{fig:8}c for three
different values of $U$. For small $U$ the
DOS resembles that of free conduction electrons in one dimension
with a divergence at the lower and upper band edge (dotted line of
Fig.~\ref{fig:8}a). For larger $U$ the opening of a charge gap of
order $U$
around the Fermi level at $\omega=0$ can be observed
(Fig.~\ref{fig:8}a and Fig.~\ref{fig:8}c). For still
larger $U$ only charge excitations away from the Fermi level are left.
This behavior can easily be understood from the former expressions
\eqref{G58}
for the two spectral functions. Note that for rather
small values of $U$ the first terms in both equations
are the dominant ones since the prefactors
$|\tilde{\beta}_{\bf k k'}|^2$ of the second terms
are of order $U^2$. From the gap in the renormalized
quasiparticle energy $\varepsilon_k$ of Fig.~\ref{fig:5}a
the occurrence of a gap in $\rho_c(\omega)$
immediately follows. Moreover, since the dispersion of
$\varepsilon_k$ around the Fermi
momentum flattens with increasing $U$ also the peaks in the DOS
at the edges of the gap can be explained. The result for the larger
$U$ value in Fig.~\ref{fig:8}c show a similar behavior as for the lower
$U$'s. Note however, for these cases our weak-coupling approach might be at
the limit of its validity.
In Fig.~\ref{fig:8}a also the exact result for the
DOS from Ref.~\cite{Lyzwa.93} at temperature $T =0$
is shown (dashed line). There, the divergencies
at the band edges appear since an infinitely large system size was
considered in that approach. In contrast the renormalization equations
in the PRM approach were solved on a finite lattice by which the
divergencies become smooth. On the other hand, in order to satisfy
the sum rule the spectral function within the bands is enlarged
in the PRM result. In a recent Monte-Carlo study \cite{maska.05}
the two-dimensional Falicov-Kimball model for the same concentrations
$n_f=n_c=1/2$ was discussed. It shows
quite similar features for the DOS as our PRM result for dimension $d=1$.
In addition in Ref.~\cite{maska.05} also temperature-dependent
effects have been discussed.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Fig 4 %%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\begin{center}
\includegraphics[width=5.4cm,height=4.2cm]{Erg_dos_c05_nf05_vergla.eps}
\includegraphics[width=5.4cm,height=4.2cm]{Erg_dos_c_nf03c.eps}
\includegraphics[width=5.4cm,height=4.2cm]{Erg_dos_c10_nf05b.eps}
\end{center}
\caption{(a) DOS $\rho_{c}(\omega)$
as function of $\omega$
for $U= 0.1$ (dotted line) and $U=1.0$ (solid line)
and $n_f=n_c=1/2$. Note the opening of a gap in the
excitation spectrum at the Fermi level $\omega=0$. The dashed line is
the exact result for $T=0$ as taken from
Ref.~\cite{Lyzwa.93}.
(b) DOS
$\rho_{c}(\omega)$ for a different filling $n_{f} = 1/3$ ($n_{c} = 2/3$)
for two values $U = 1.0$ (solid line) and $U = 2.0$ (dashed line).
There are now two gaps found in the charge spectrum.
(c) Same quantity as in panel (a) with $n_f=n_c=1/2$
for a larger $U$ value $U=2.0$ (solid line). Besides the low temperature
result the DOS is also shown for a higher temperature $T=0.2$
(dashed dotted line).
Note that the low-temperature gap at $\omega =0$ is smeared out for
$T=0.2$.}
\label{fig:8}
\end{figure}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
In Fig.~\ref{fig:8}b the DOS $\rho_c(\omega)$ is shown for
$n_f=1/3$ ($n_c=2/3$) for our lowest temperature
$T= 0.1$. As expected from the previous
results for $\tilde{\varepsilon}_k$ in Fig.~\ref{fig:5}b,
there are now two gaps found in the charge excitation spectrum,
one of which is again located at the Fermi level. Note that
in Fig.~\ref{fig:8}b the density of states is symmetric
with respect to the band center. This symmetry
can easily be understood as follows: First,
at low temperatures the correlation function
$\big< \delta n_i \delta n_{i+\kappa}\big>$
is invariant with respect to translations by
three lattice sites (compare the lower panel of Fig.~\ref{fig:7}).
Therefore, $C_\rho(k- k')
\sim N \; \delta_{k -k', \pm 2\pi/3}$
at low enough temperatures. By exploiting this relation
and Eq.~\eqref{G36} one finds that all $k$ points which are
symmetric to $\pm \pi/2$ lead to the same renormalization
of the quasiparticle energy $\varepsilon_{k, \lambda}$
but with an opposite sign. Thus, the renormalized quasiparticle energy
$\tilde{\varepsilon}_k$ becomes antisymmetric,
since also the original $\varepsilon_k$
was antisymmetric relative to $\pm \pi/2$ and the
DOS becomes symmetric relative to the band center
in the present approach. This result is in contrast to the expectation
that for occupation $n_c$ away from half-filling
the DOS should be asymmetric. In fact,
one can assure oneself that
higher order terms to $X_{\lambda,\Delta \lambda}$ would give
rise to a small additional asymmetric contribution to the DOS
in our approach.
Note that such a small asymmetry is found
in an exact result \cite{Lyzwa.93} for the DOS in one dimension
at $T=0$ for $n_c=2/3$.
Finally, in Fig.~\ref{fig:8}c also the influence of temperature on
$\rho_c(\omega)$ is shown for
$n_f=n_c=1/2$. Note that at $T=0.1$ the gap in the DOS is clearly seen.
However, by increasing the
temperature by a factor of 2 the gap becomes completely smeared out.
\\
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Fig 5 (Fig.7 alt) %%%%%%%%%%%%%%%%%%%
\begin{figure}[h]
\begin{center}
\includegraphics[width=5.2cm,height=4.2cm]{Apm_U10nf05k06.eps} \hspace{0.1cm}
\includegraphics[width=5.2cm,height=4.2cm]{Apm_U10nf05k0507.eps} \hspace{0.1cm}
\includegraphics[width=5.2cm,height=4.2cm]{Apm_U10nf05k04.eps}
\end{center}
\caption{One-particle spectral functions $A_k^+(\omega)$ (solid line) and
$A_k^-(\omega)$ (dashed line) at low temperature ($T=0.1$) as function of
$\omega$ for different $k$ vectors,
(a) $k>k_F$, (b) $k \ge k_F$, and (c) $k<k_F$. ($U=1.0$, $n_f=n_c=
1/2$.)}
\label{fig:14}
\end{figure}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
In Figs.~\ref{fig:14}a-c the one-particle spectral functions
$A_k^+(\omega)$ and $A_k^-(\omega)$ for different $k$ values
are shown for the lowest possible temperature
of $T=0.1$ and $n_f=n_c=1/2$. For $k > k_F$ (Fig.~\ref{fig:14}a) a
coherent excitation is found in $A_k^+(\omega)$
and a small incoherent excitation in $A_k^-(\omega)$.
To understand
the latter feature, note that only a single $k'$ value,
$k'=k - \pi$, contributes to the incoherent part in \eqref{G58} due to
$C_\rho(k-k') \sim N \delta_{k-k',\pi}$. The same remains true
when $k$ comes closer to $k_F$ (Fig.~\ref{fig:14}b). In this case, however,
the incoherent contribution from $A_k^-(\omega)$ has gained
weight. The reason is the renormalization factor $|\tilde{\beta}_{kk'}|^2$
in equation \eqref{G58}. It becomes stronger when $k$ and also
$k'$ approaches the Fermi momentum as follows from \eqref{G47}
and \eqref{G30}.
Finally, for $k<k_F$ (Fig.~\ref{fig:14}c)
a coherent contribution is found in $A_k^-(\omega)$ and an
incoherent excitation in $A_k^+(\omega)$.
In Figs.~\ref{fig:11}a-c the one-particle
spectral functions $B^{+}(\omega)$ (black) and $B^{-}(\omega)$ (red)
of the localized electrons averaged over all
lattice sites from section IV are shown for different values of
$U$ and $n_f=n_c=1/2$. For the two lower $U$-values ($U=0.1, 1.0$)
at a rather low temperature $T=0.1$ the coherent parts (of weight
$\tilde{\nu}^2$) dominate in Fig.~\ref{fig:11}a
the spectrum (compare \eqref{G61}).
In contrast, for the two larger $U$-values ($U=2.0, 3.0$)
the incoherent excitations
have gained considerable weight (Fig.~\ref{fig:11}b and Fig.~\ref{fig:11}c).
Note that at a higher temperature $T=0.2$ the
incoherent part is smeared out over a wide frequency range
for $U=2.0$ (dashed dotted curve in
Fig.~\ref{fig:11}b). The same feature would also be expected for $U=3.0$.
However, no stable solution of the renormalization equations
was found for this temperature. The discussed behavior can again be
understood from the former PRM expressions \eqref{G65}, \eqref{G66}
for the spectral functions $B^{+}(\omega)$ and $B^{-}(\omega)$.
The first term in both equations describes the coherent excitation
of weight $\tilde{\nu}^2$ at $\omega \approx \tilde{\varepsilon}_f =0$
whereas the second term follows from the coupling of $f_i$ to
electronic particle-hole excitations (compare \eqref{G61}). Their weight
$\tilde{\mu}_{k,k'}^2$ is at least of order $U^2$. The gaps
in Figs.~\ref{fig:11}a-c of order $U$ immediately follow
from the gap in the quasiparticle energy $\tilde{\varepsilon}_k$.
The same quantities
$B^{+}(\omega)$ (black) and $B^{-}(\omega)$ (red) are also shown in
Fig.~\ref{fig:13} for $f$-filling $n_f=1/3$ $(n_c=2/3)$ and $U= 1.0,
2.0$. The gaps between the respective coherent and incoherent
parts again correspond to the gap
of the quasiparticle excitation in Fig.~\ref{fig:5}b
at the Fermi level. Note that the spectrum is no
longer symmetric around $\omega = 0$. The different
weights of $B^{+}(\omega)$ and $B^{-}(\omega)$ in Fig.~\ref{fig:13}
follow from the Fermi functions in \eqref{G65} and \eqref{G66}.
We have also checked the $f$-sum rule
\begin{eqnarray}
\label{G80}
\int_\infty^\infty d \omega \;
(B^+(\omega) + B^-(\omega)) &=& \big< [f_i^\dagger, f_i ]_+ \big> =1
\end{eqnarray}
for the spectral functions in Figs.~\ref{fig:11} and
\ref{fig:13}. We found that for the two
small $U$ values in Fig.~\ref{fig:11}a
the sum rule was perfectly fulfilled whereas for the spectral
functions with $U=2.0$ and
$3.0$ (Fig.~\ref{fig:11}b and Fig.~\ref{fig:11}c) the agreement was less satisfactory.
For instance, for $U=3.0$ a value of $0.9$ instead of $1$ was found.
The origin of these deviations
can be understood as follows. First, from Eqs.~\eqref{G65} and
\eqref{G66} one finds that \eqref{G80} is fulfilled if
\begin{eqnarray}
\label{G81}
|\nu_\lambda|^2 + \frac{1}{N^2} \sum_{k k'}
|\mu_{k k',\lambda}|^2 f({\varepsilon}_{k, \lambda})
(1- f(\varepsilon_{k',\lambda})) =1
\end{eqnarray}
is satisfied for all values of $\lambda$. For the original model
relation \eqref{G81} is trivially valid. By
restricting oneself to the lowest
order processes in the renormalization equations
for the $\lambda$-dependent coefficients in \eqref{G61}
one can show that the sum rule is fulfilled for $(\lambda -
\Delta \lambda)$ if it was fulfilled for $\lambda$. However,
there are deviations due to higher order renormalization
contributions for larger $U$ and small values of $\lambda$
as follows from expression \eqref{G30} for $A_{k,k'}^{\lambda,
\Delta \lambda}$. Also important for the deviations
is the chosen simple operator structure of the
generator $X_{\lambda, \Delta \lambda}$
from \eqref{G16}. Note however that higher
order processes in $U$ are suppressed for larger system sizes $N$.
Thus, in order to control the influence of higher order terms
a finite size scaling in $N$ should be done. Instead, in the present
approach a simpler truncation procedure has been used. The renormalization
was stopped at the smallest $\lambda$ value at which the influence of
higher order terms was still negligible.
This leads to the observed difference in the sum rule as compared
to the exact result.
\vspace{0.5cm}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Fig 6 (Fig 5 alt) %%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}[h]
\begin{center}
\includegraphics[width=5.4cm,height=4.2cm]{Erg_dos_f05_nf05a.eps}
\includegraphics[width=5.4cm,height=4.2cm]{Erg_dos_f10_nf05b.eps}
\includegraphics[width=5.4cm,height=4.2cm]{Erg_dos_f15_nf05_2c.eps}
\end{center}
\caption{(a) One-particle spectral functions
$B^{+}(\omega)$ (black) and $B^{-}(\omega)$ (red) of
$f$-electrons as function of $\omega$
for $U=0.1$ (dashed line) and $U=1.0$ (solid line),
and $n_f=n_c=1/2$, $T=0.1$. For both $U$ values the coherent
excitation dominates the spectrum. (b),(c) Same quantity as in
Fig.~\ref{fig:11}a
for two larger values of $U$, $U=2.0$ (b) and $U=3.0$ (c)
and $T=0.1$ (solid lines). Note that now the incoherent
excitations have gained considerable weight
as compared to the case of panel (a).
For $U=2.0$ (b) the
spectral function is also shown for a higher temperature $T=0.2$
(dashed dotted line).
\vspace*{1cm}}
\label{fig:11}
\end{figure}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Fig. 7 (Fig.6 alt)%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\begin{center}
\includegraphics[width=8cm,height=6cm]{Erg_dos_f05_nf03_2a.eps}
\includegraphics[width=8cm,height=6cm]{Erg_dos_f10_nf03_2b.eps}
\end{center}
\caption{Same quantity as in Fig.~\ref{fig:11}
for $U=1.0 $ (a) and $U=2.0 $ (b)
and $n_f=1/3$.}
\label{fig:13}
\end{figure}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Conclusions
\label{Conclusions}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
In summary, in this paper we have applied a recently developed
projector-based renormalization method (PRM) to the spinless
Falicov-Kimball model in one dimension and at half-filling
$n_f +n_c=1$.
As a starting point in the formalism, we have taken
the one-particle part of the initial Hamiltonian \eqref{G1} as
unperturbed Hamiltonian ${\cal H}_0$ and
the local Coulomb interaction $U$
between conduction and localized electrons as interaction ${\cal H}_1$.
The elimination of the latter part leads to the
renormalization of the Hamiltonian. Thereby a new density-density
interaction between the localized electrons is generated. Finally, a
number of static and dynamic quantities has been calculated.
Note that the present formuation of the PRM is a weak-coupling
approach which however should lead to reliable results for
$U$-values up to the order of $t$. It is shown that
the investigation of the small temperature behavior
enables us also to extract valuable information
about the zero temperature properties of static and dynamic
quantities.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section*{Acknowledgments}
We would like to acknowledge fruitful discussions with J.~Freericks,
Ch.~Geibel and
A.~H\"ubsch. This work was supported by the
DFG through the research program SFB 463, and by DFG grant
No: Dr 274/10-1. One of us (VZ) would like to acknowledge the support by the
National Science Foundation under grant number DMR-0210717.
\newpage
\begin{appendix}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Operator transformation of ${\cal H}_\lambda$}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\label{trafos}
In this appendix we evaluate the renormalized Falicov-Kimball Hamiltonian
obtained by the unitary transformation
${\cal H}_{(\lambda -\Delta\lambda)}
=
\exp\{X_{\lambda, \Delta \lambda}\} \ {\cal H}_{\lambda} \
\exp\{-X_{\lambda, \Delta \lambda}\},
$
where $X_{\lambda, \Delta \lambda}$ is given by \eqref{G22}
The transformed Hamiltonian ${\cal H}_{\lambda -\Delta \lambda}$
has no excitations with transfer energies
$|\varepsilon_{k, \lambda - \Delta \lambda} -
\varepsilon_{k', \lambda - \Delta \lambda}| >
\lambda - \Delta \lambda $. Let us first consider
the transformation of ${\cal H}_{0,\lambda}$
\begin{eqnarray}
\label{A1}
e^{X_{\lambda, \Delta \lambda}} {\cal H}_{0,\lambda}
e^{-X_{\lambda, \Delta \lambda}} - {\cal H}_{0,\lambda}
&=& \sum_k \varepsilon_{k, \lambda}
\sum_{n=1}^\infty \frac{1}{n!}\;
{\bf X}_{\lambda, \Delta \lambda}^n c_k^\dagger c_k
\end{eqnarray}
where we have introduced a new superoperator
${\bf X}_{\lambda, \Delta \lambda}$ which is defined by the commutator of the
generator $X_{\lambda, \Delta \lambda}$ with operators ${\cal A}$ on
which ${\bf X}_{\lambda, \Delta \lambda}$ is applied,
${\bf X}_{\lambda, \Delta \lambda} {\cal A} =
[X_{\lambda, \Delta \lambda}, {\cal A}]$. Note that the operators
$c_k^\dagger c_k$ are the only part of ${\cal H}_{0,\lambda}$
which do not commute with $X_{\lambda, \Delta \lambda}$. We
successively evaluate the commutators on the right-hand side of
\eqref{A1}. By use of
\begin{eqnarray}
\label{A2}
{\bf X}_{\lambda, \Delta \lambda} c_k^\dagger c_k
&=&
- \frac{1}{N} \sum_{\tilde{k},i} A_{k \tilde k}
\; \Theta_{k \tilde k}^{\lambda, \Delta \lambda} \;
\Big( e^{i(\tilde k- k){R}_i}\; \delta n_i \;
c_{\tilde k}^\dagger c_k +
e^{i(k- \tilde k){R}_i}\; \delta n_i \;
c_{k}^\dagger c_{\tilde k}
\Big)
\end{eqnarray}
we first obtain
\begin{eqnarray}
\label{A3}
{\bf X}_{\lambda, \Delta \lambda} {\cal H}_{0,\lambda}
&=&
\frac{1}{N} \sum_{kk'i} A_{k k'}
\; \Theta_{k k'}^{\lambda. \Delta \lambda} \;
(\varepsilon_{k',\lambda} - \varepsilon_{k, \lambda}) \;
e^{i(k- k'){R}_i}\; \delta n_i \;
c_{k}^\dagger c_{k'}
\end{eqnarray}
Next by applying ${\bf X}_{\lambda, \Delta \lambda}$ on
${\bf X}_{\lambda,\Delta \lambda}{\cal H}_{0,\lambda}$
where products of two fermions and of two
local density operators $\delta n_j \; \delta n_i$ occur. In order to keep
only operators which are also present in
${\cal H}_{(\lambda - \Delta \lambda)}$ a factorization approximation
has to be performed. With
\begin{eqnarray}
\label{A4}
{\bf X}_{\lambda, \Delta \lambda} \;
\delta n_i \; c_{k}^\dagger c_{k'} &=&
\frac{1}{N} \sum_{\tilde{k},j}
A_{\tilde{k} k} \Theta_{\tilde{k} k}^{\lambda, \Delta \lambda} \;
e^{i(\tilde{k}- k){R}_j}\;
\Big( \big< \delta n_i \delta n_j \big> c_{\tilde k}^\dagger c_{k'}
\\
&& \hspace*{2cm}
+ \delta_{k', \tilde k} \;
\delta n_i \delta n_j \big< c_{k'}^\dagger c_{k'} \big>
- \delta_{k', \tilde k}
\big< \delta n_i \delta n_j \big> \big<c_{\tilde k}^\dagger
c_{k'} \big>
\Big) \nonumber \\
%%
&-& \frac{1}{N} \sum_{\tilde k,j}
A_{\tilde{k} k'} \Theta_{\tilde{k} k'}^{\lambda, \Delta \lambda} \;
e^{-i(\tilde{k}- k'){R}_j}\;
\Big( \big< \delta n_i \delta n_j \big> c_{k}^\dagger c_{\tilde{k}}
\nonumber \\
&& \hspace*{2cm}
+ \delta_{k, \tilde{k}} \;
\delta n_i \delta n_j \big< c_{k}^\dagger c_k \big>
- \delta_{k, \tilde{k}}
\big< \delta n_i \delta n_j \big> \big<c_{k}^\dagger
c_{\tilde{k}} \big>
\Big) \nonumber
\end{eqnarray}
one finds
\begin{eqnarray}
\label{A5}
{\bf X}_{\lambda, \Delta \lambda}^2 {\cal H}_{0,\lambda}
&=&
\frac{2}{N^2}
\sum_{k k'ij} A_{k k'}^2 \Theta_{kk'}^{\lambda, \Delta \lambda}
(\varepsilon_{k',\lambda} - \varepsilon_{k,\lambda})
\cos[i(k'-k)({R}_i -{R}_j)] \\
&& \times \Big( \big< \delta n_i \delta n_j \big>
c_{\tilde{k}}^\dagger c_{k'} \nonumber
+ \delta_{k', \tilde{k}} \;
\delta n_i \delta n_j \big< c_{k'}^\dagger c_{k'} \big>
- \delta_{k', \tilde{k}}
\big< \delta n_i \delta n_j \big> \big<c_{\tilde{k}}^\dagger
c_{k'} \big>
\Big) \nonumber
\end{eqnarray}
or by using translational invariance
\begin{eqnarray}
\label{A6}
{\bf X}_{\lambda, \Delta \lambda}^2 {\cal H}_{0,\lambda}
&=&
\frac{2}{N}
\sum_{k k'} A_{k k'}^2 \Theta_{kk'}^{\lambda, \Delta \lambda}
(\varepsilon_{k',\lambda} - \varepsilon_{k,\lambda}) \\
&\times& \Big( C_\rho(k' -k)
c_{k}^\dagger c_k \nonumber
+ \big< c_{k}^\dagger c_k \big>
\frac{1}{N} \sum_{ij} \delta n_i \delta n_j\; e^{i(k- k')
({R}_i - {R}_j)}
- \big<c_k^\dagger c_k \big> C_\rho(k'-k)
\Big) \nonumber
\end{eqnarray}
Note that in deriving \eqref{A5} and \eqref{A6} it was assumed that the number
of $k$ points which are integrated out in the step from
$\lambda$ to $(\lambda - \Delta \Lambda)$ is small compared to the total
number of $k$ points. Therefore, the product of two
product $\delta$ functions
$\Theta_k k'$ and $\Theta_{\tilde{k} \tilde{k}'}$ is only
nonzero when the wavevectors are pairwise equal to each other
(shell condition).
In the next step, ${\bf X}_{\lambda, \Delta \lambda}$ is again
applied on $c_k^\dagger c_k$. Thus by use of \eqref{A2}
one finds
\begin{eqnarray}
\label{A7}
{\bf X}_{\lambda, \Delta \lambda}^3 {\cal H}_{0,\lambda}
&=&
\frac{1}{N} \sum_{kk'i}
\big[ -\frac{2^2 C_\rho(k -k')A_{k k'}^2}{N} \big]
A_{k k'}
\; \Theta_{k k'}^{\lambda, \Delta \lambda} \;
(\varepsilon_{k',\lambda} - \varepsilon_{k, \lambda}) \;
e^{i(k- k'){R}_i}\; \delta n_i \;
c_{k}^\dagger c_{k'} \nonumber \\
&&
\end{eqnarray}
which has again the structure of ${\bf X}_{\lambda, \Delta \lambda}
{\cal H}_{0,\lambda}$ except of the additional prefactor in the bracket
$ [ ... ]$. thus one may sum up all commutator terms from \eqref{A1}
in order to obtain a compact expression for \eqref{A1}.
In the same way one may proceed
with the transformation of ${\cal H}_{1,\lambda}$.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Green function ${\cal G}_0(\omega)$
of the lattice gas model}
To evaluate the Green function ${\cal G}_0(\omega)$
\begin{eqnarray}
\label{C1}
{\cal G}_0(\omega) &=& i \int_0^\infty dt \; e^{i(\omega + i\eta)t}\;
\big< [f_i(t), f_i^\dagger]_+ \big>_{\tilde{\cal H}_f}
\end{eqnarray}
we best use the Mori-Zwanzig projection formalism. First, let us define
an anticommutator scalar product for
operators ${\cal A}$, ${\cal B}$ by
%\begin{eqnarray}
%\label{C2}
$ ({\cal A} | {\cal B} ) = \big< [{\cal A}^\dagger , {\cal B} ]_+
\big>_{\tilde{\cal H}_f}$.
%\end{eqnarray}
In addition, we introduce the Liouville operator $\tilde{\bf L}_f$
belonging to the lattice gas Hamiltonian $\tilde{\cal H}_f$. It acts
on any operator ${\cal A}$ of the unitary space as
$\tilde{\bf L}_f{\cal A} = [{\cal H}_f, {\cal A}]$.
Thus \eqref{C1} can be rewritten as
\begin{eqnarray}
\label{C3}
{\cal G}_0(\omega) &=& i \int_0^\infty dt \;
e^{i(\omega +i \eta)t}\; \big(f_i^\dagger| e^{-i \tilde{\bf L}_f\;t } \;
f_i^\dagger \big) =
\big( f_i^\dagger | \frac{1}{\tilde{\bf L}_f - \omega - i\eta}\;
f_i^\dagger \big)
\end{eqnarray}
The well-known operator identity of the projection formalism \cite{fick.90}
can be used to transform ${\cal G}_0(\omega)$ into
\begin{eqnarray}
\label{C4}
{\cal G}_0(\omega) &=& \frac{1}{\Omega -\omega - \Sigma(\omega)}
\end{eqnarray}
where the frequency term $\Omega$ and the selfenergy $\Sigma(\omega)$
are given by
\begin{eqnarray}
\label{C5}
\Omega &=& \big( f_i^\dagger| \tilde{\bf L}_f f_i^\dagger \Big)
\hspace*{2cm}
\Sigma(\omega) = \big( \tilde{\bf L}_f f_i^\dagger| {q}
\frac{1}{ {\bf Q} \tilde{\bf L}_f {\bf Q} - \omega - i \eta}\;
{\bf Q} \tilde{\bf L}_f f_i^\dagger \big)
\end{eqnarray}
Note that in \eqref{C4} and \eqref{C5} we have already used that
$\big( f_i^\dagger| f_i^\dagger \Big)
= \big< [ f_i, f_i^\dagger]_+ \big>_{\tilde{\cal H}_f} =1 $.
The quantity ${\bf Q}$ is like ${\bf L}_f$ a superoperator which acts on
usual operators of the unitary space. ${\bf Q}$ is a projection operator
and projects on the subspace of the Liouville space
perpendicular to $|f_i^\dagger\big)$, i.e., it is defined by
${\bf Q} |{\cal A}\big) = |{\cal A} \big)
- | f_i^\dagger \big) \big( f_i^\dagger|{\cal A}\big)$. To evaluate
$\Omega$ and $\Sigma(\omega)$ we need
\begin{eqnarray}
\label{C6}
\tilde{\bf L}_f |f_i^\dagger \big) &=&
\tilde{\varepsilon}_f |f_i^\dagger \big) +
2 \sum_{j(\neq i)} \tilde{g}_{ij} |\delta n_j f_i^\dagger \big)
\hspace*{0.5cm} \mbox{and} \hspace*{0.5cm}
{\bf Q} \tilde{\bf L}_f |f_i^\dagger \big) =
2 \sum_{j(\neq i)} \tilde{g}_{ij} {\bf Q}|\delta n_j f_i^\dagger \big)
\end{eqnarray}
where in the second relation the first term of
$\tilde{\bf L}_f |f_i^\dagger \big)$ drops due
to the presence of ${\bf Q}$. As is easily seen, we obtain
$\Omega = \tilde{\varepsilon}_f$ and
\begin{eqnarray}
\label{C7}
\Sigma(\omega) &=&
4 \sum_{j(\neq i)} \sum_{l (\neq i)} \tilde{g}_{il}^* \; \tilde{g}_{ij} \;
\big( \delta n_l f_i^\dagger |{\bf Q}
\frac{1}{ {\bf Q} \tilde{\bf L}_f {\bf Q} - \omega - i \eta}\;
{\bf Q} |\delta n_j f_i^\dagger \Big)
\end{eqnarray}
Note that
the local density operators $\delta n_l$ and $\delta n_j$
in the bra- and ket vector of \eqref{C7}
commute with $\tilde{\cal H}_f$. Therefore, we can
approximate $\Sigma(\omega)$ by an expression proportional to
${\cal G}_0$
\begin{eqnarray}
\label{C8}
\Sigma(\omega) &=& \kappa^2 \; {\cal G}_0(\omega)
\hspace*{2cm}
\kappa^2 =
4 \sum_{j(\neq i)} \sum_{l (\neq i)} \tilde{g}_{il}^* \; \tilde{g}_{ij} \;
\big< \delta n_l \delta n_j \big>_{\tilde{\cal H}_f}
\end{eqnarray}
which is identical to \eqref{G67}. Note that
$\kappa^2$ in \eqref{C8} is independent of the lattice site $i$.
$\Sigma(\omega)$ together with \eqref{C4} forms a quadratic
equation for ${\cal G}_0(\omega)$ which can be solved.
It is easily seen that the imaginary part
is given by
\begin{eqnarray}
\label{C9}
\Im {\cal G}_0(\omega) &=& \frac{1}{\kappa}
\sqrt{1 - \frac{(\omega - \tilde{\varepsilon}_f)^2}{4 \kappa^2}}
\end{eqnarray}
for the frequency range $|\omega - \tilde{\varepsilon}_f| \leq 2 \kappa$
and ${\cal G}_0(\omega)=0$ elsewhere.
Thus,
$\Im {\cal G}_0(\omega)$ is nonzero
only in a small frequency region around $\omega =\tilde{\varepsilon}_f$ of
width $\pm 2\kappa$ where $\kappa$ is determined by static density-density
correlations of the lattice gas model.
The maximum value of $\Im {\cal G}_0(\omega)$ is located
at $\omega= \tilde{\varepsilon}_f$ and is given by $1/\kappa$.
From $\Im{\cal G}_o(\omega)$ the functions $B_0^+(\omega)$
and $B_0^-(\omega)$ are determined by help of the
fluctuation-dissipation theorem
\begin{eqnarray}
\label{C10}
B_0^+(\omega) &=& (1 -f(\omega)) \frac{1}{\pi}
\Im {\cal G}_0(\omega)
\hspace*{2cm}
B_0^-(\omega) = f(\omega) \frac{1}{\pi}
\Im {\cal G}_0(\omega)
\end{eqnarray}
where $f(\omega)$ is again the Fermi function. From
\eqref{C10} and \eqref{G65},\eqref{G66}
the $f$-electron Green functions
of the original Falicov-Kimball model can be found.
\end{appendix}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{thebibliography}{100}
\bibitem{falicov_kimball} L.~M.~Falicov and J.~C.~Kimball, Phys. Rev. Lett.
{\bf 22}, 997 (1969)
\bibitem{freericks.98}
J.~K.~Freericks and V.~Zlati\'c, Phys. Rev. B {\bf 58}, 322 (1998).
\bibitem{freericks.03} J.~K.~Freericks and V.~Zlatic, Rev. Mod. Phys. \textbf{75},
1333 (2003).
\bibitem{brandt.89} U.~Brandt and C.~Mielsch, Z.~Phys.~B: Condens.~Matter
\textbf{75}, 365 (1989).
\bibitem{dongen.90} P.~G.~J.~van Dongen and D.~Vollhardt,
Phys.~Rev.~Lett. \textbf{65},
1663 (1990).
\bibitem{kennedy.86} T.~Kennedy and E.~H.~Lieb, Physica A \textbf{138},
320 (1986).
\bibitem{lieb.86} E.~H.~Lieb, Physica A \textbf{140}, 240 (1986).
\bibitem{sykora.05} S.~Sykora, A.~H\"ubsch, and K.~W.~Becker,
Eur. Phys. J. B \textbf{51}, 181 (2006).
\bibitem{becker} K.~W.~Becker, A.~H\"ubsch, and T.~Sommer, Phys. Rev. B
{\bf 66}, 235115 (2002).
\bibitem{kohn.64} W.~Kohn, Phys. Rev. A \textbf{133}, 171 (1964).
\bibitem{schrieffer-wolff} J.~R.~Schrieffer, P.~A.~Wolff,
Phys. Rev. \textbf{149}, 491 (1966).
\bibitem{anderson} P.~W.~Anderson, J. Phys. C: {\it Solid State Phys.}
\textbf{3}, 2436 (1970).
\bibitem{wegner} F.~Wegner, Ann. Phys. (Leipzig) {\bf 3}, 77 (1994).
\bibitem{wilson} S.~D.~G{\l}azek and K.G.~Wilson, Phys. Rev. D {\bf 48}, 5863
(1993); S.~D.~G{\l}azek and K.G.~Wilson, Phys. Rev. D {\bf 49}, 4214 (1994).
\bibitem{Ott} H.~R.~Ott and Z.~Fisk, in \textit{Handbook on the Physics and
Chemistry of the Actinides}, edited by A: J.~Freeman and G.~H.~Lander
(Elsevier, Amsterdam, 1987).
\bibitem{QMC} P.~Farkasovsky, Phys. Rev. B \textbf{54}, 7865 (1996).
\bibitem{farkasovsky.03} P.~Farkasovsky, Int. J. Mod. Phys. B \textbf{17},
4897 (2003).
\bibitem{nagaosa.98} in N.~Nagaosa, {\it Quantum Field Theory in Strongly
Correlated Electronic Systems} Springer-Verlag Berlin
Heidelberg New York, 1998.
\bibitem{maska.05} M.~M.~Maska and K.~Czajka, Phys.~Rev.~B \textbf{74}, 035109 (2006).
\bibitem{fick.90} in {\it The Quantum Statistics of Dynamic Processes},
E.~Fick and G.~Sauermann, Springer series in Solid-State Sciences 86,
1990.
\bibitem{Lyzwa.93} R.~Lyzwa, Physica A \textbf{192},
231 (1993).
\end{thebibliography}
\end{document}