Density matrix renormalisation group for a quantum spin chain at non-zero temperature
Extracted source text ยท arXiv cond-mat/9609001
%Bursill et al cond-mat/9609001
%Standard latex, 6 pages
\documentstyle{article}
\newcommand {\tsub}[1]{_{\mbox{\protect\scriptsize #1}}}
\newcommand {\tsup}[1]{^{\mbox{\protect\scriptsize #1}}}
%\newcommand {\tsupl}[1]{{\mbox{\protect\scriptsize #1}}^}
\newcommand {\Ref}[1]{(\ref{#1})}
\newcommand
{\Stretch}[1]{\renewcommand{\baselinestretch}{#1}\large\normalsize}
\newcommand {\unStretch}{\Stretch{2}}
\newcommand {\bm}[1]{{\mbox{\boldmath $#1$}}}
\newcommand {\ket}[1]{\left|#1\right\rangle}
\newcommand {\bra}[1]{\left\langle#1\right|}
\newcommand {\inner}[2]{ \left\langle #1 \right|
\left. #2 \right\rangle }
\hoffset=-1.8cm
\textwidth=15.5cm
\textheight=22cm
\voffset=-1.5cm
\begin{document}
%\pagestyle{empty}
\Stretch{3}
\begin{center}
{\large 17/7/96}
\vspace*{2em}
{\huge \bf Density Matrix Renormalisation Group for a Quantum Spin Chain
at Non-Zero Temperature}
\vspace*{5em}
\Stretch{1.5}
{\large
R.\ J.\ Bursill$^{1\dagger}$,
T. Xiang$^{2}$ and G.\ A.\ Gehring$^{1}$
}
\vspace*{2em}
$^{1}$Department of Physics,
The University of Sheffield,
S3 7RH, Sheffield, United Kingdom.\\
$^{2}$Interdisciplinary Research Centre in Superconductivity,
The University of Cambridge, Cambridge, CB3 0HE, UK.\\
$\dagger$ Present address: School of Physics,
UNSW, Sydney, 2052, Australia.
\end{center}
%\vfill
%\eject
%\Stretch{2}
\Stretch{1.2}
\begin{abstract}
We apply a recent adaptation of White's density matrix renormalisation
group (DMRG) method to a simple quantum spin model, the dimerised $XY$
chain, in order to assess the applicabilty of the DMRG to quantum
systems at non-zero temperature. We find that very reasonable results
can be obtained for the thermodynamic functions down to low temperatures
using a very small basis set. Low temperature results are found to be
most accurate in the case when there is a substantial energy gap.
\end{abstract}
\setcounter{page}{0}
%\vfill
%\eject
%\pagestyle{plain}
Since it's recent inception, White's denstiy matrix renormalisation
group (DMRG) method has been established as the method of choice for
determining static, low energy properties of one-dimensional quantum
lattice systems \cite{development}--\cite{bursill}. Extensions
% to models with impurities \cite{impurities} and disorder
%\cite{disorder},
to the calculation of
dynamical properaties \cite{dynamics} and even to the study of low
temperature properties of two dimensional systems \cite{two_dimensions}
have been forthcoming. Moreover, Nishino's formulation of the DMRG for
two dimensional classical systems \cite{nishino} has paved the way for
the study of one dimensional quantum systems at non-zero temperature. In
this letter we present what is, to the best of our knowledge, the first
application of the DMRG to the thermodynamics of a quantum system.
The system that we consider is a simple spin chain model---the
dimerised, $S=1/2$, $XY$
model
\begin{equation}
{\cal H} = -\sum_{i=1}^{N/2}
\left[
J_{1} \left( S_{2i-1}^{x} S_{2i}^{x} + S_{2i-1}^{y} S_{2i}^{y} \right)
+
J_{2} \left( S_{2i}^{x} S_{2i+1}^{x} + S_{2i}^{y} S_{2i+1}^{y} \right)
\right]
\label{hamiltonian}
\end{equation}
where $S_{i}$ is a spin-1/2 operator for site $i$ on an (even) chain of
$N$ sites, with periodic boundary conditions.
This model is exactly solvable \cite{exact_solution}; the Helmholtz free
energy $\psi$ is given by
\begin{equation}
-\beta\psi = \lim_{N \rightarrow \infty} \frac{ \log Z_{N} }{ N }
= \frac{1}{2 \pi} \int_{0}^{2 \pi} \log \left[
2 \cosh \frac{\beta \phi(\theta)}{4}
\right] \, d \theta
\label{psi}
\end{equation}
where $Z_{N} = {\mbox Tr}\, e^{-\beta{\cal H}}$
is the partition function, $\beta \equiv 1/T$ is the inverse
temperature,
$\phi(\theta) =
\cos \xi(\theta) + \gamma \cos ( \theta + \xi(\theta) )$,
$\xi(\theta) =
-\tan^{-1} \frac{\gamma \sin \theta}{1 + \gamma \cos \theta}$,
$\gamma \equiv J_{2}/J_{1}$ and we have set $J_{1} = 1$.
Thermodynamic properties such as the internal energy
$u = - \frac{\partial\,\beta\psi}{\partial\beta}$ and the specific heat
$c_{V} = -\beta^{2} \frac{\partial u}{\partial\beta}$ are readily
obtainable from \Ref{psi}.
Because of its solvability and the fact that it possesses both gapless
$(\gamma = 1)$ and gapped $(\gamma \neq 1)$ excitations,
\Ref{hamiltonian} presents a useful test model for the extension of the
DMRG to quantum systems at non-zero temperature.
Now Nishino's formulation of the DMRG applies directly to classical, two
dimensional spin systems and so in order to apply it to the quantum
system
\Ref{hamiltonian}, we must first invoke the Trotter-Suzuki method
\cite{suzuki}. That is, we make the decomposition ${\cal H} =
{\cal H}_{1} + {\cal H}_{2}$ where
\begin{equation}
{\cal H}_{1} = -J_{1} \sum_{i=1}^{N/2} h_{2i-1,2i},
\hspace*{2em}
{\cal H}_{2} = -J_{2} \sum_{i=1}^{N/2} h_{2i,2i+1}
\label{H1H2}
\end{equation}
and $h_{i,j} \equiv S_{i}^{x} S_{j}^{x} + S_{i}^{y} S_{j}^{y}$.
We then apply the formula
\begin{equation}
Z_{N} = \lim_{M \rightarrow \infty} Z_{MN}
\equiv
\lim_{M \rightarrow \infty}
{\mbox Tr}\,
\left[
e^{-\beta{\cal H}_{1}/M}
e^{-\beta{\cal H}_{2}/M}
\right]^{M}
\label{ZMN}
\end{equation}
Inserting a collection ${\sigma}$ of $2M$ complete sets of states into
\Ref{ZMN} then yields
\begin{equation}
Z_{MN}
=
\sum_{{\sigma}}
\prod_{j=1}^{M}
\bra{ \sigma_{1}^{2j-1}\ldots\sigma_{N}^{2j-1} }
e^{-\beta{\cal H}_{1}/M}
\ket{ \sigma_{1}^{2j}\ldots\sigma_{N}^{2j} }
\bra{ \sigma_{1}^{2j}\ldots\sigma_{N}^{2j} }
e^{-\beta{\cal H}_{2}/M}
\ket{ \sigma_{1}^{2j+1}\ldots\sigma_{N}^{2j+1} }
\end{equation}
where periodic boundary conditions
$\sigma_{i}^{2M+1} \equiv \sigma_{i}^{1}$,
$\sigma_{N+1}^{j} \equiv \sigma_{1}^{j}$ are assumed.
Now, because the $N/2$ terms in the sums \Ref{H1H2} commute and act on
different pairs of sites, we have
\begin{equation}
Z_{MN}
=
\sum_{{\sigma}}
\prod_{i=1}^{N/2}
\prod_{j=1}^{M}
\tau_{1}( \sigma_{2i-1}^{2j-1} \sigma_{2i}^{2j-1} | \sigma_{2i-1}^{2j}
\sigma_{2i}^{2j} )
\tau_{2}( \sigma_{2i}^{2j} \sigma_{2i+1}^{2j} | \sigma_{2i}^{2j+1}
\sigma_{2i+1}^{2j+1} )
\end{equation}
where
\begin{equation}
\tau_{i}(\sigma'\mu'|\sigma\mu)
\equiv
\bra{\sigma'\mu'} e^{\beta J_{i}h_{1,2}/M} \ket{\sigma\mu}
\label{taui}
\end{equation}
for $i=1,2$. Moreover,
\begin{eqnarray}
Z_{MN}
& = &
\sum_{{\sigma}}
\prod_{i=1}^{N/2}
{\cal T}_{1}( \sigma_{2i-1}^{1}\ldots\sigma_{2i-1}^{2M} |
\sigma_{2i}^{1}\ldots\sigma_{2i}^{2M} )
{\cal T}_{2}( \sigma_{2i}^{1}\ldots\sigma_{2i}^{2M} |
\sigma_{2i+1}^{1}\ldots\sigma_{2i+1}^{2M} )
\\
& = &
\mbox{Tr}\,{\cal T}^{N/2}
\end{eqnarray}
where ${\cal T} \equiv {\cal T}_{1}{\cal T}_{2}$ is the {\em virtual
transfer matrix} and
\begin{eqnarray}
{\cal T}_{1}
(\mu^{1}\ldots\mu^{2M} | \sigma^{1}\ldots\sigma^{2M})
& \equiv &
\prod_{j=1}^{M}
\tau_{1}( \mu^{2j-1}\sigma^{2j-1} |
\mu^{2j}\sigma^{2j} )
\\
{\cal T}_{2}
(\mu^{1}\ldots\mu^{2M} | \sigma^{1}\ldots\sigma^{2M})
& \equiv &
\prod_{j=1}^{M}
\tau_{2}( \mu^{2j}\sigma^{2j} |
\mu^{2j+1}\sigma^{2j+1} )
\end{eqnarray}
The matrices ${\cal T}_{1}$ and ${\cal T}_{2}$ are depicted graphically
in Fig.\ 1.
It follows \cite{suzuki} that
\begin{equation}
\psi = \psi^{(M)}
\equiv
\lim_{M\rightarrow\infty}
-\frac{ \log\lambda\tsub{max} }{ 2\beta }
\label{psiM}
\end{equation}
where $\lambda\tsub{max}$ is the eigenvalue of ${\cal T}$ with maximal
modulus and generally depends on the Trotter dimension $M$.
We may now apply the formulation of the DMRG for transfer matrices to
the calculation of $\lambda\tsub{max}$. We commence by defining an
initial transfer matrix ${\cal T}\tsub{s}$ for a single-site {\em system
block}, which connects it to adjacent sites to the left and the right
viz
\begin{equation}
{\cal T}\tsub{s}( \sigma' n' \mu' | \sigma\, n\, \mu ; \sigma'' \mu'' )
=
\sum_{n''}
\tau_{1}( \sigma' \sigma'' | n' n'' )
\tau_{2}( n'' n | \mu'' \mu )
\end{equation}
The initial block, having just one site, has $m=2$ states, $n=1$ or 2
($\uparrow$ or $\downarrow$) (see Fig.\ 2).
We next define an initial {\em environment block} and associated
transfer matrix ${\cal T}\tsub{e}$ in precisely the same way:
${\cal T}\tsub{e} \equiv {\cal T}\tsub{s}$. We define a {\em superblock}
using the system and environment blocks in addition to two added sites,
arranged in a periodic fashion (see Fig.\ 3). That is, the superblock
transfer matrix is given by
\begin{eqnarray}
%& &
{\cal T} ( \sigma_{1}' n_{1}' \sigma_{2}' n_{2}' |
\sigma_{1} n_{1} \sigma_{2} n_{2} )
& = &
( {\cal T}_{1} {\cal T}_{2} )
( \sigma_{1}' n_{1}' \sigma_{2}' n_{2}' |
\sigma_{1} n_{1} \sigma_{2} n_{2} )
\\
& = &
\sum_{\sigma_{1}'' \sigma_{2}''}
{\cal T}\tsub{s}
( \sigma_{1}' n_{1}' \sigma_{2}' | \sigma_{1} n_{1} \sigma_{2} ;
\sigma_{1}'' \sigma_{2}'' )
{\cal T}\tsub{e}
( \sigma_{2}' n_{2}' \sigma_{1}' | \sigma_{2} n_{2} \sigma_{1} ;
\sigma_{2}'' \sigma_{1}'' )
\label{superblock_transfer}
\end{eqnarray}
At this point ${\cal T}$ may be diagonalised to determine
$\lambda\tsub{max}$ and hence the $M=2$ approximation \Ref{psiM} to
\Ref{psi}.
In order to proceed to larger lattices, we must augment (expand) the
system and environment blocks. We let $n \longleftrightarrow (p,\nu)$,
$p=1,\ldots,m$, $\nu=\uparrow,\downarrow$ denote a state for an
augmented system block, consisting of the initial system block and one
site added to the right. There are now $m' = 2m = 4$ states:
$n = 1 \longleftrightarrow (1,\uparrow)$,$\ldots$,
$n = 4 \longleftrightarrow (2,\downarrow)$.
The transfer matrix ${\cal T}\tsub{s}'$ for the augmented system block
is defined (see Fig.\ 4) by
\begin{equation}
{\cal T}\tsub{s}'
( \sigma' n' \mu' | \sigma\, n\, \mu ; \sigma'' \mu'' )
\equiv
\sum_{\nu''}
{\cal T}\tsub{s}
( \sigma' p' \nu' | \sigma\, p\, \nu ;
\sigma'' \nu'' )
\tau_{1} ( \nu' \nu'' | \mu' \mu'' )
\end{equation}
An augmented environment block is defined in a similar way, this time
adding a site to the left viz
\begin{equation}
{\cal T}\tsub{e}'
( \sigma' n' \mu' | \sigma\, n\, \mu ; \sigma'' \mu'' )
\equiv
\sum_{\nu''}
{\cal T}\tsub{e}
( \sigma' p' \nu' | \sigma\, p\, \nu ;
\sigma'' \nu'' )
\tau_{2} ( \sigma'' \sigma | \nu'' \nu )
\end{equation}
Now, the superblock and its associated transfer matrix
\Ref{superblock_transfer} can once again be formed using the augmented
blocks with $m' \longmapsto m$,
${\cal T}\tsub{s}' \longmapsto {\cal T}\tsub{s}$ and
${\cal T}\tsub{e}' \longmapsto {\cal T}\tsub{e}$. Moreover, the process
can be iterated, each time augmenting the system and environment blocks
to the right and the left respectively.
However, in order to prevent the superblock basis from becoming too
large, we must truncate it by capping the number of system (and
environment) block states, $m$, which in principle doubles every
iteration. To do so, we form reduced density matrices for the augmented
blocks by performing an appropriate partial trace on the projection
operator $\ket{\psi\tsub{max}}\bra{\psi\tsub{max}}$ formed from the
eigenstate $\ket{\psi\tsub{max}}$ of ${\cal T}$ corresponding to the
eigenvalue $\lambda\tsub{max}$.
That is, for the augmented system block, the density matrix
$\rho\tsub{s}$ is defined by
\begin{equation}
\rho\tsub{s}(n'|n)
\equiv
\sum_{\sigma_{1} n_{2}}
\inner{\sigma_{1} n_{1}' \sigma_{2}' n_{2}}{\psi\tsub{max}}
\inner{\psi\tsub{max}}{\sigma_{1} n_{1} \sigma_{2} n_{2}}
\end{equation}
where $n \longleftrightarrow (n_{1},\sigma_{2})$ and
$n' \longleftrightarrow (n_{1}',\sigma_{2}')$. For the augmented
environment block we have
\begin{equation}
\rho\tsub{e}(n'|n)
\equiv
\sum_{\sigma_{1} n_{1}}
\inner{\sigma_{1} n_{1} \sigma_{2}' n_{2}'}{\psi\tsub{max}}
\inner{\psi\tsub{max}}{\sigma_{1} n_{1} \sigma_{2} n_{2}}
\end{equation}
with $n \longleftrightarrow (n_{2},\sigma_{2})$ and
$n' \longleftrightarrow (n_{2}',\sigma_{2}')$.
Eigenvalues and eigenvectors of the density matrices are then found viz
\[
\{
\omega_{n}\tsup{(s)}, \ket{n}\tsup{(s)}:\;n=1,\ldots,m'
\}
\mbox{ and }
\{
\omega_{n}\tsup{(e)}, \ket{n}\tsup{(e)}:\;n=1,\ldots,m'
\}
\]
for $\rho\tsub{s}$ and $\rho\tsub{e}$ respectively where
$1 \geq \omega_{1}\tsup{(i)} \geq \omega_{2}\tsup{(i)}
\geq \ldots \geq \omega_{m'}\tsup{(i)}$ for
$\mbox{i}=\mbox{s}$
or e.
In proceeding to the next iteration then, we represent the augmented
block transfer matrices ${\cal T}\tsub{s}'$ and ${\cal T}\tsub{e}'$ in
terms of the density matrix eigenvectors and truncate the block Hilbert
spaces so as to retain only the $m$ most important states viz
\begin{equation}
\sum_{n'',n''' = 1}^{m'}
\inner{n'}{n''}\tsup{(i)}
{\cal T}\tsub{i}'
(\sigma' n'' \mu' | \sigma\, n''' \mu ; \sigma'' \mu'')
\;\tsup{(i)}
\inner{n'''}{n}
\longmapsto
{\cal T}\tsub{i}
(\sigma' n' \mu' | \sigma\, n\, \mu ; \sigma'' \mu'')
\end{equation}
for
$\mbox{i}=\mbox{s}$
and e, $n,n'=1,\ldots,m$ and where $m$ is the chosen cap on
the number of states per block.
Now, ${\cal T}$ is a large, sparse, non-symmetric matrix, for which
efficient algorithms for the determination of $\lambda\tsub{max}$ and
$\ket{\psi\tsub{max}}$ are becoming available. So far we have simply
used
Schur decomposition and the inverse power method to perform this task.
We have thus only been able to work with $m=16$.
Results for $\psi$, $u$ and $c_{V}$ for $\gamma = 1$ (pure $XY$) and
$\gamma = 2$ (dimerised $XY$) are given in
Figs 5--7.
Finite-$M$ results $\psi^{(M)}$ are obtained by fixing the inverse
temperature $\beta=\beta_{0}$ in the expressions \Ref{taui} and
increasing the superblock size $2M$, beginning with $M=2$, so at each
iteration the temperature is identified as $T = 1/M\beta_{0}$. For the
$\beta_{0}$ values we have considered (0.05--0.2) there is a small error
due to the finiteness of the Trotter decomposition but this is
negligible compared with the error due to Hilbert space truncation, a
result which is easily verified by considering the exact solution for
$\psi^{(M)}$ \cite{suzuki}. A typical calculation, using NAG library
routines on a 333 MHz DEC Alpha machine, takes 6 hours to generate a
superblock size of $2M = 300$.
We note that convergence at low temperatures is far better for the
gapped system ($\gamma = 2$). However, even in the gapless case, the
peak position and peak height in the specific heat are afflicted by
errors of only ca.\ 3\%, and results may be improved markedly by using
larger values of $m$ and the finite lattice method \cite{development}.
Also, results for $u$ and $c_{V}$ are obtained by numerically
differentiating spline fits of the computed $\psi^{(M)}$ values, and so
improvements may be obtained by combining results from a number of
different $\beta_{0}$ values.
Note that, at any given iteration, the DMRG usually works with a chain
which is spatially finite. Here the chain is infinite and the finiteness
is in the level of the Trotter approximation. Another difference is that
the DMRG usually produces its best results for the ground state energy
and less accurate results for higher excitations. A different situation
occurs here---the lower the temperature, the less accurate the result.
Even so, as can be seen in Fig.\ 5, the exact ground state energy
$\lim_{T\rightarrow 0}\psi$, is recovered with reasonable accuracy,
especially in the gapped case $\gamma=2$.
Finally, we note that, as ${\cal T}$ is non-symmetric, the use of a
truncated basis will not in general lead to a variational lower bound on
$\lambda\tsub{max}$. However, in practice it can be seen, for instance,
that the DMRG result for the internal energy $u$ is always bounded above
the exact value.
To conclude, we have applied the DMRG to a quantum spin chain at
non-zero temperature, making use of Nishimo's adaptation of the method to 2D
classical systems. Reasonable results are obtained for the specific heat
down to low temperatures in calculations involving an extremely small
basis set, agreement with the exact solution being markedly better in
the case where the system has a substantial gap. The approach may prove
useful in determining thermodynamic properties of models of
experimentally realisable systems such as coupled chains and models with
anisotropy, dimerisation and frustration.
R.\ J.\ B.\ gratefully acknowledges the support of SERC grant no.\
GR/J26748. Computations were performed on the DEC 8400
facility at CLRC.
%\vfill
%\eject
\begin{thebibliography}{99}
\bibitem{development}
S.\ R.\ White and R.\ M.\ Noack, Phys.\ Rev.\ Lett.\ 68 (1992) 3487;
S.\ R.\ White, Phys.\ Rev.\ Lett.\ 69 (1992) 2863,
Phys.\ Rev.\ B 48 (1993) 10 345.
\bibitem{bursill}
S.\ R.\ White and D.\ A.\ Huse, Phys.\ Rev.\ B 48 (1993) 3844;
E.\ S.\ S\o rensen and I.\ Affleck, Phys.\ Rev.\ Lett.\ 71 (1993) 1633;
R.\ J.\ Bursill, T.\ Xiang and G.\ A.\ Gehring, J.\ Phys.\ A 28 (1994)
2109;
R.\ J.\ Bursill, G.\ A.\ Gehring, D.\ J.\ J.\ Farnell, J.\ B.\
Parkinson, Tao Xiang and Chen Zeng, J.\ Phys.\ C 7 (1995) 8605;
S.\ R.\ White, R.\ M.\ Noack and D.\ J.\ Scalapino,
Phys.\ Rev.\ Lett.\ 73 (1994) 886;
R.\ M.\ Noack, S.\ R.\ White and D.\ J.\ Scalapino,
Phys.\ Rev.\ Lett.\ 73 (1994) 882;
L.\ G.\ Caron and S.\ Moukouri, Phys.\ Rev.\ Lett.\ 76 (1996) 4050;
G.\ A.\ Gehring, R.\ J.\ Bursill and T.\ Xiang, {\em Applications of the
density matrix renormalisation group to problems in magnetism}, To
appear in Acta Physica Polonica (cond-mat/9608127).
\bibitem{dynamics}
H.\ B.\ Pang, H.\ Akhlaghpour and M.\ Jarrell, Phys.\ Rev.\ B 53 (1996)
5086;
K.\ A.\ Hallberg, Phys.\ Rev.\ B 52 (1995) R9827.
\bibitem{two_dimensions}
S.\ Liang and H.\ Pang, Europhys.\ Lett.\ 32 (1995) 173;
T.\ Xiang, Phys.\ Rev.\ B 53 (1996) 10445;
S.\ R.\ White, {\em Spin Gaps in a Frustrated Heisenberg model for
CaV$_4$O$_9$}, Preprint;
S.\ R.\ White and D.\ J.\ Scalapino, {\em Hole and Pair Structures in
the $t$-$J$ model}, Preprint.
\bibitem{nishino}
T.\ Nishino, J.\ Phys.\ Soc.\ Jap.\ 64 (1995) 3598;
T.\ Nishino, K.\ Okunishi and M.\ Kikuchi, Phys.\ Lett.\ A, 213 (1996)
69.
\bibitem{exact_solution}
L.\ N.\ Bulaevskii, Sov.\ Phys.\ JETP 17 (1963) 1008.
\bibitem{suzuki}
M.\ Suzuki and M.\ Inoue, Prog.\ Theor.\ Phys.\ 78 (1987) 787;
M.\ Inoue and M.\ Suzuki, Prog.\ Theor.\ Phys.\ 79 (1988) 645.
\end{thebibliography}
%\vfill
%\eject
\section*{Figure Captions}
\noindent
1. Pictorial representation of matrices a)
${\cal T}_{1}
(\mu^{1}\ldots\mu^{2M} | \sigma^{1}\ldots\sigma^{2M})$ and b)
${\cal T}_{2}
(\mu^{1}\ldots\mu^{2M} | \sigma^{1}\ldots\sigma^{2M})$.
\noindent
2. Pictorial representation of the initial system block transfer
matrix
${\cal T}\tsub{s}
( \sigma' n' \mu' | \sigma\, n\, \mu ; \sigma'' \mu'' )$.
$n$ and $n'=1,\ldots,m=2$ are initial and final states of the
system block which consists of a single site.
$\sigma$, $\sigma''\mbox{ and }\sigma'=\;\uparrow\mbox{ or }\downarrow$ and
$\mu$, $\mu''\mbox{ and }\mu'=\;\uparrow\mbox{ or }\downarrow$ are initial,
intermediate and final states for the adjeacent sites to the
left and right of the block respectively. The intermediate
state $n''$ of the system block is summed over to produce
the matrix product.
\noindent
3. Pictorial representation of the superblock, which consists of
system and environment blocks connected by two added sites to form a
periodic chain. $n_{1}\mbox{ and } n_{1}'$ and $n_{2}\mbox{ and }
n_{2}'$ are initial and final state indices for the system and
environment blocks respectively.
$\sigma_{1}\mbox{ and }\sigma_{1}'$ and
$\sigma_{2}\mbox{ and }\sigma_{2}'$
are initial and final states for the two added sites. The intermediate
states $\sigma_{1}''$ and $\sigma_{2}''$ are summed over to form the
matrix product.
\noindent
4. Pictorial representation of the augmented (a) system and (b)
environment blocks, which consist of the old blocks augmented by adding
sites to the right and the left respectively. $n\mbox{ and }
n'=1,\ldots,m'$ are the initial and final states of the new (augmented)
block which consists of the old block ($p$ and $p'$) augmented with an
added site ($\nu$ and $\nu'$). $\sigma$, $\sigma''\mbox{ and }\sigma'$ and
$\mu$, $\mu''\mbox{ and }\mu'$ are state indices for the adjacent sites.
The intermediate state $\nu''$ of the added site is summed over to
produce the matrix product.
\noindent
5. Exact and DMRG results for the Helmholtz free energy $\psi$ as
a function of temperature $T$ for two cases: $\gamma = 1$ and
$\gamma = 2$.
\noindent
6. Exact and DMRG results for the internal energy $u$ as a
function of temperature $T$ for two cases: $\gamma = 1$ and
$\gamma = 2$.
\noindent
7. Exact and DMRG results for the specific heat $c_{V}$ as a
function of temperature $T$ for two cases: $\gamma = 1$ and
$\gamma = 2$.
%\begin{figure}[htbp]
%\epsfboxtw{fig1.ill}
%\caption{Pictorial representation of matrices a) ${\cal T}_{1}$ and b)
%${\cal T}_{2}$.}
%\label{T12}
%\end{figure}
%
%\begin{figure}[htbp]
%\epsfboxtw{fig2.ill}
%\caption{Pictorial representation of the initial system block tranfer
%matrix ${\cal T}\tsub{s}$.}
%\label{initial_transfer}
%\end{figure}
%
%\begin{figure}[htbp]
%\epsfboxtw{fig3.ill}
%\caption{Pictorial representation of the superblock, which consists of
%system and environment blocks connected by two added sites to form a
%periodic chain.}
%\label{superblock}
%\end{figure}
%
%\begin{figure}[htbp]
%\epsfboxtw{fig4.ill}
%\caption{Pictorial representation of the augmented (a) system and (b)
%environment blocks, which consist of the old blocks augmented by adding
%sites to the right and the left respectively.}
%\label{augmented_blocks}
%\end{figure}
%
%\begin{figure}[htbp]
%\epsfboxtw{fig5.ill}
%\caption{Exact and DMRG results for the Helmholtz free energy $\psi$ as
%a function of temperature $T$ for two cases: $\gamma = 1$ and
%$\gamma = 2$.}
%\label{free_energy}
%\end{figure}
%
%\begin{figure}[htbp]
%\epsfboxtw{fig6.ill}
%\caption{Exact and DMRG results for the internal energy $u$ as a
%function of temperature $T$ for two cases: $\gamma = 1$ and
%$\gamma = 2$.}
%\label{internal_energy}
%\end{figure}
%
%\begin{figure}[htbp]
%\epsfboxtw{fig7.ill}
%\caption{Exact and DMRG results for the specific heat $c_{V}$ as a
%function of temperature $T$ for two cases: $\gamma = 1$ and
%$\gamma = 2$.}
%% Put beta_0, m and gamma values on figures
%\label{specific_heat}
%\end{figure}
%
\end{document}