Density Functional Study of Cubic to Rhombohedral Transition in $\alpha$-AlF$_3$
Extracted source text ยท arXiv cond-mat/0302272
\documentclass[aps,prb,twocolumn,groupedaddress,showpacs]{revtex4}
\usepackage[dvips]{graphics}
\begin{document}
\title{Density Functional Study of Cubic to Rhombohedral Transition
in $\alpha$-AlF$_3$}
\author{Yiing-Rei Chen$^1$, Vasili Perebeinos$^2$, and Philip B. Allen$^{1,3}$}
\affiliation{$^1$Department of Physics and Astronomy, State University
of New York, Stony Brook, New York 11794-3800\\
$^2$IBM Research Division, Thomas J. Watson Research Center, Yorktown
Heights, New York 10598 \\
$^3$Department of Applied Physics and Applied Mathematics, and the
Materials Research Science and Engineering Center, Columbia University,
New York, New York 10027}
\date{\today}
\begin{abstract}
Under heating, $\alpha$-AlF$_3$ undergoes a structural phase transition
from rhombohedral to cubic at temperature $T$ around 730 K. The density
functional method is used to examine the $T$=0 energy surface in the
structural parameter space, and finds the minimum in good agreement with
the observed rhombohedral structure. The energy surface and electronic
wave-functions at the minimum are then used to calculate properties
including density of states, $\Gamma$-point phonon modes, and the dielectric
function. The dipole formed at each fluorine ion in the low temperature
phase is also calculated, and is used in a classical electrostatic picture
to examine possible antiferroelectric aspects of this phase transition.
\end{abstract}
\pacs{64.70.Kb,71.20.-b,77.80.Bh}
% 77.80.Bh Antiferroelectricity materials phase transition and Curie point
% 64.70.Kb Solid-Solid transition
% 71.20.-b Electron density of states and band structure in crystalline solid
\maketitle
%
%
\section{Introduction}
The ionic insulator AlF$_3$ has a number of known polymorphs
\cite{phases1,phases2}, which all convert irreversibly to the stable
$\alpha$-AlF$_3$ within the temperature range approximately 730 to 920 K.
Recent interest in this material arises due to its catalytic activity for
dismutation and halogen exchange reactions \cite{catalytic}.
Above its transition temperature (about 730 K)\cite{Grey}, $\alpha$-AlF$_3$
has the cubic perovskite structure AMX$_3$, with the A cations absent (or
the ReO$_3$ structure.) Aluminum plays the role of the M cation, and is
surrounded by an octahedron of corner-shared fluorine atoms. At low
temperature, the structure becomes rhombohedral, and this symmetry lowering
can be characterized as a rotation of the fluorine octahedron about one of
the three-fold axis of the perovskite cubic cell (or the $a^-a^-a^-$ system
in Glazer's tilt system \cite{Woodward}). As every adjacent octahedron
rotates in the opposite sense, the wave-vector of this distortion is
($\pi,\pi,\pi$), and the unit cell becomes double the size of that in the
cubic phase.
We study the $\alpha$-AlF$_3$ structure by using pseudopotentials, a plane-wave
basis, and the LDA method. After being tested, the pseudopotentials
are applied in bulk $\alpha$-AlF$_3$. The total energy surface is examined
in the structural parameter space where the cubic phase is compared with
the rhombohedral phase found at the deepest nearby minimum of the energy
surface.
\section{Testing Different Pseudopotentials}
The Troullier and Martins method \cite{T&M} is used to generate the
pseudopotentials. Since there are different possibilities of choosing
cut-off radii $r_{cl}$, as well as how many valence orbitals to include
in the pseudopotentials, we test the pseudopotentials by doing LDA
self-consistent iterations for a crystal or molecule prototype and
compare the results with FP-LAPW \cite{Blaha} all-electron calculations
(full potential linearized augmented plane-wave \cite{Singh}). This
provides guidance on the choice of energy cut-off of the plane-wave basis,
and also the choice of local potential, in the $\alpha$-AlF$_3$ calculation
that follows. The results of these tests are shown in Table \ref{tab:Al}
and Table \ref{tab:F}, for aluminum and fluorine separately. For the
Ceperley-Alder exchange-correlation energy \cite{CA}, the Vosko-Wilk-Nusair
\cite{VWN} parametrization is used.
The test of the aluminum pseudopotential is performed in bulk Al metal, not
only the observed {\it fcc} structure, but also the {\it sc} and
{\it bcc} structures, where comparison can still be made between the
all-electron method and LDA. The results show that the pseudopotential
for the Al$^{3+}$ ion should be capable of describing the unoccupied
$3d$ orbital. Otherwise, even the large $r_{cl}$ ($l=0,1$ for $s$ and
$p$ orbitals) needed to fix the lattice constant in the {\it fcc} aluminum
are insufficient to remove the $r_{cl}$-dependence of the lattice
constant and bulk modulus. Due to the more attractive nature of the $d$
pseudopotential, including the $3d$ orbital reduces the dependence of
these two quantities on $r_{cl}$ and the choice of local potential, and
improves the agreement with the all-electron results. Better agreement
is also seen in comparisons for $sc$ and {\it bcc} structures of Al metal.
The fluorine pseudopotential is tested for the F$_2$ molecule. Unlike
the low plane-wave cut-off needed for aluminum (less than 40 Rydberg),
more than 90 Rydberg is needed in the F$_2$ molecule. For bulk Al metal
we use a k-point mesh of 12$\times$12$\times$12 that contains 56 special
k-points, while for F$_2$ we do a one k-point calculation.
Results in the following sections are calculated using pseudopotentials
(Al$^\dag$, F$^\dag$) as shown in the captions of the Tables.
This set of pseudopotentials gives the lattice constant ($a = 5.02 \AA$),
and bulk modulus ($B = 146.6$ GPa) of cubic $\alpha$-AlF$_3$ in
good agreement with all-electron results ($a=5.043\AA$, $B=150$ GPa.)
\begin{table}
\caption{Aluminum pseudopotential tests on bulk Al. The first column
shows the cut-off radii in units of Bohr radius ($a_0$) for the valence
orbitals (l=0,1,2). Tests with r$_{\rm c2}$ omitted do not include the
$d$ orbital. The following columns show the use of core correction, the
choice of local potential, lattice constant, bulk modulus and the total
energy per atom with respect to the {\it fcc} structure. Values are shown
in atomic units ($e=$ electron charge, $a_0=$ Bohr radius.) $\Delta$E
is defined to be zero for the {\it fcc} structure, while the total energy
differences between {\it fcc} and other structures can be compared with
all-electron results. The pseudopotential with a $^\dag$ is chosen for the
calculations of $\alpha$-AlF$_3$. \label{tab:Al} }
\begin{ruledtabular}
\begin{tabular}{cccccc}
\makebox[1.2cm]{r$_{\rm c0}$/r$_{\rm c1}$/r$_{\rm c2}$} &
\makebox[0.8cm]{core} &
\makebox[0.7cm]{local} &
\makebox[1cm]{lattice} &
\makebox[1cm]{B} &
\makebox[1cm]{$\Delta$E} \\
($a_0$) & corr. && const.($a_0$) & ($\times$10$^{-3}e^2/a_0^4$) &
($e^2/a_0$)\\
\hline
{\it fcc} structure &&&& \\
2.2/2.5/--- & yes & $s$ & 7.738 & 2.487 & 0 \\
2.4/2.5/--- & yes & $s$ & 7.712 & 2.524 & 0 \\
2.6/2.6/--- & yes & $s$ & 7.672 & 2.607 & 0 \\
1.9/2.4/2.7 & yes & $d$ & 7.507 & 2.873 & 0 \\
1.8/2.2/2.7 & yes & $s$ & 7.500 & 2.850 & 0 \\
1.8/2.2/2.7 & yes & $p$ & 7.516 & 2.896 & 0 \\
1.8/2.2/2.7 & yes & $d$ & 7.507 & 2.873 & 0 \\
1.8/2.2/2.6 & yes & $d$ & 7.505 & 2.934 & 0 \\
2.6/2.6/2.6$^\dag$ & no & $d$ & 7.481 & 2.865 & 0 \\
all-electron & --- & --- & 7.536 & 2.905 & 0 \\
experiment & --- & --- & 7.650 & 2.579 & --- \\
\hline
{\it sc} structure &&&& \\
1.9/2.4/2.7 & yes & $d$ & 5.063 & 2.176 & 0.01431 \\
1.8/2.2/2.7 & yes & $s$ & 5.061 & 2.170 & 0.01451 \\
1.8/2.2/2.7 & yes & $p$ & 5.069 & 2.195 & 0.01425 \\
1.8/2.2/2.7 & yes & $d$ & 5.063 & 2.177 & 0.01431 \\
1.8/2.2/2.6 & yes & $d$ & 5.062 & 2.175 & 0.01436 \\
2.6/2.6/2.6$^\dag$ & no & $d$ & 5.050 &2.072 & 0.01462 \\
all-electron & --- & --- & 5.065 & 2.096 & 0.01449 \\
\hline
{\it bcc} structure &&&& \\
1.9/2.4/2.7 & yes & $d$ & 6.015 & 2.546 & 0.00368 \\
1.8/2.2/2.7 & yes & $s$ & 6.009 & 2.534 & 0.00374 \\
1.8/2.2/2.7 & yes & $p$ & 6.023 & 2.581 & 0.00368 \\
1.8/2.2/2.7 & yes & $d$ & 6.015 & 2.542 & 0.00368 \\
1.8/2.2/2.6 & yes & $d$ & 6.014 & 2.538 & 0.00370 \\
2.6/2.6/2.6$^\dag$ & no & $d$ & 5.992 & 2.472 & 0.00380 \\
all-electron & --- & --- & 6.027 & 2.525 & 0.00414 \\
\end{tabular}
\end{ruledtabular}
\end{table}
%
\begin{table}
\caption{Fluorine pseudopotential tests on F$_2$ molecule. The last
column indicates the plane-wave cut-off needed to obtain the convergent
result. The pseudopotential with a $^\dag$ is chosen for the calculations
of $\alpha$-AlF$_3$. \label{tab:F}}
\begin{ruledtabular}
\begin{tabular}{lccccc}
\makebox[1.5cm]{r$_{\rm c0}$/r$_{\rm c1}$} &
\makebox[0.8cm]{core} &
\makebox[0.7cm]{local} &
\makebox[1.0cm]{bond} &
\makebox[1.0cm]{force} &
\makebox[1.0cm]{cut-off} \\
($a_0$) & corr. && ($a_0$) & const.($e^2/a_0^3$) & (Ryd.) \\
\hline
1.4/1.6 & no & $p$ & 2.780 & 0.370 & 90 \\
1.4/1.45 & yes & $p$ & 2.710 & 0.408 & 90 \\
1.3/1.3 & yes & $s$ & 2.599 & 0.384 & over 100 \\
1.3/1.3$^\dag$ & no & $s$ & 2.603 & 0.376 & 90 \\
1.2/1.2 & yes & $s$ & 2.617 & 0.387 & over 100 \\
all-electron & --- & --- & 2.632 & 0.376 & --- \\
experiment & --- & --- & 2.669 & 0.302 & --- \\
\end{tabular}
\end{ruledtabular}
\end{table}
%
%
\section{Results from LDA}
\subsection{k-point Test and Density of States}
A k-point mesh of 3$\times$3$\times$3 is used for the Brillouin zone
summation for $\alpha$-AlF$_3$. This is tested for cubic $\alpha$-AlF$_3$,
where a finer mesh of 4$\times$4$\times$4 gives a barely different energy
{\it vs} plane wave cut-off curve, with energy uncertainty less than 1 meV.
This mesh gives six special k-points after symmetrization (according to the
point group D$_{\rm 3d}$ of the crystal), over which we include at least
36 states in the truncated Hamiltonian diagonalization processes. Since in
the rhombohedral unit cell of $\alpha$-AlF$_3$ we consider the 48 valence
electrons from aluminum $3s^{2}3p^{1}$ and fluorine $2s^{2}2p^{5}$, these
electrons will occupy the 24 lowest lying bands. A test including more
states, 48 states for example, shows that in the process of iterative
diagonalization, using 36 states gives convergent answers for the eigenvalues
of the lowest 24 bands. Thus we include 36 states in the calculation of the
minimum on the energy surface. To make the density of states plot,
we include 50 states to better describe the empty states from aluminum
orbitals. As shown in Fig. \ref{fig:dos}, LDA gives an insulator
band-gap of 8 eV. We choose 90 Rydberg to be the energy cut-off for the
plane-wave basis. This choice gives the same energy difference between
the two structures as that from the choice of 140 Rydberg (with error
within $1\sim 2$ meV. The two structures are the cubic and the rhombohedral
structure at the minimum.)
%
\begin{figure}[ht]
\centerline{\scalebox{0.33}[0.33]{\includegraphics*{rhomdos.eps}}}
\caption{Density of states plot of $\alpha$-AlF$_3$ in the rhombohedral
ground state, using pseudopotential (Al$^\dag$, F$^\dag$). The
rhombohedral ground state has unit cell edge 4.76$\AA$,
$\theta$=57.5$^\circ$ and $\delta$=0.098. This calculation
includes 50 states and uses a k-point mesh 6$\times$6$\times$6.
The temperature broadening is 54 meV. \label{fig:dos}}
\end{figure}
%
%
\subsection{Crystal Structure}
We use LDA to search for the ground state in the rhombohedral symmetry
class $R\overline{3}C$ seen experimentally in $\alpha$-AlF$_3$\cite{Grey}.
There are three free parameters: (1) the lattice constant $a$, (2) the angle
$\theta$ between any two of the rhombohedral lattice vectors, and (3) the
inner parameter $\delta$, which describes the bending of the Al-F-Al bond.
To describe these three parameters, we first start from the cubic structure
with lattice vectors $(2b,0,0)$, $(0,2b,0)$ and $(0,0,2b)$, where $b$ is the
Al-F bond length. A rhombohedral unit cell is formed by picking new lattice
vectors $\vec{a}_1=(0,2b,2b)$, $\vec{a}_2=(2b,0,2b)$, and $\vec{a}_3=(2b,2b,0)$.
The angle $\theta$ defined as
$\theta = \cos^{-1}(\vec{a}_1\cdot\vec{a}_2/|\vec{a}_1||\vec{a}_2|)$
is $60^\circ$ in this case. This new cell is then elongated along the [111]
direction. Consequently, the strains
$\epsilon_{23}=\epsilon_{31}=\epsilon_{12}$ become nonzero, and
$\theta\neq 60^\circ$. Viewed from the [111] direction, in the aluminum plane
the ions form a hexagonal pattern, and the nearest Al-Al distance on this
(111) plane is the lattice constant $a$. This rhombohedral structure can also
be described in a hexagonal setting, with the following $c/a$ ratio
%
\begin{equation}
\frac{c}{a} = \sqrt{\left(\frac32\right)
\left(\frac{1+2\cos\theta}{1-\cos\theta}\right)}
\end{equation}
%
The cell shape and cell volume $V$ are determined by $a$ and $c/a$:
%
\begin{equation}
V = \frac{a^3}{2\sqrt{2}}\sqrt{\frac{1+2\cos\theta}{1-\cos\theta}}
= \frac{a^2c}{2\sqrt{3}},
\end{equation}
%
while the relation between the lattice constant $a$ and the length of
the rhombohedral lattice vectors is:
%
\begin{equation}
a / |\vec{a}_1| = \sqrt{2(1-\cos\theta)}
\end{equation}
%
There are two aluminum ions and six fluorine ions in each rhombohedral
unit cell. The fluorine ions sit at the 6e sites ($(x,\bar{x}+\frac 12,
\frac 14)$, $(x,x+\frac 12,\frac 34)$, etc.), and $\delta$=$x$-0.75 is
the deviation from $x$=0.75 where the Al-F-Al bond angle is strictly
180$^\circ$. The distortion caused by $\delta$ alone defines the octahedron
rotation angle $\omega$ about the [111] axis:
%
\begin{equation}
\tan\omega = 2\sqrt{3}\delta,
\label{eq:delta}
\end{equation}
%
On the other hand, the $c/a$ ratio and the $\omega$ have the following
relation:
%
\begin{equation}
\frac{c}{\sqrt{6}a} = \frac{1+\zeta}{\cos\omega},
\label{eq:zeta}
\end{equation}
%
where $\zeta$ describes the flattening ($\zeta<0$) and elongation ($\zeta>0$)
of the octahedron along [111].
We first assume cubic symmetry by fixing $\theta=60^\circ$
and $\delta=0$. LDA gives the lattice constant $a=a^*=5.02\AA$ for
the cubic phase. With the lattice constant kept fixed at the value $a = a^*$,
$\delta$ is then relaxed to give minimum energy
at $(a, \theta, \delta) = (a^*, 60^\circ, \pm 0.04)$. As shown in
Fig. \ref{fig:delta}, the relaxation of $\delta$ finds minima at
$\delta$=$\pm0.04$ and a maximum at $\delta=0$, with an energy difference
of 14.4 meV, showing that in our $T$=0 energy surface study, the unbent
Al-F-Al bond, and therefore the strictly cubic structure, is not
a metastable solution.
The three parameters are then relaxed in turn to approach the deepest
nearby minimum. In fact, in the parameter space, there is a path that keeps
the fluorine octahedra rigid and leads to the region near the overall
minimum. A Rigid octahedron means that the Al-F bonds are fixed at the bond
length found by LDA in the cubic phase
%
\begin{equation}
b = \frac{a^*}{2\sqrt{2}} = \frac{a}{2\sqrt{2} \cos\omega}
\end{equation}
%
and that the angles between neighboring Al-F bonds are always 90 degrees
($\zeta=0$). With these two restrictions, only one parameter $\delta$
remains and defines the rigid octahedron path. The energy along this
path is also shown in Fig. \ref{fig:delta}. The overall minimum locates
not far from the minimum on this path. It has an Al-F bond length only
$0.3\%$ longer, a pretty small octahedron elongation factor $\zeta = 0.25\%$,
and the total energy 93 meV lower than the cubic phase.
Table \ref{tab:min} shows the comparisons between LDA and experimental data
\cite{Grey}. While no $T=0$ experimental numbers are available,
one can only make reasonable comparisons for the deviation of bond length and
$\zeta$. The rhombohdral phase observed in the experiment has an Al-F bond
length $0.43\%$ longer than the value observed in the cubic phase, and
$\zeta = -0.102\%$ (meaning the octahedron is slightly flattened.) This
indicates that both the LDA and the experimentally found rhombohedral phases
are close to the rigid octahedron path.
%
\begin{figure}[htbp]
\centerline{\scalebox{0.32}[0.32]{\includegraphics*{delta.eps}}}
\caption{The relaxation of $\delta$ from the cubic structure is shown
in plot (a). The curve of total energy (per two formulas of AlF$_3$)
{\it vs} $\delta$ at $\theta=60.0^\circ$ and $a=5.02\AA$ shows that
the cubic phase is sitting at an energy maximum. In (b), the curve
follows the rigid-octahedron path, and leads to the region near the
overall minimum in the parameter space. The dotted-dashed line shows the
energy of the overall minimum.}
\label{fig:delta}
\end{figure}
%
%
\begin{table}[here]
\caption{The structure parameters found by LDA are compared with
experimental values \cite{Grey} here for different phases. The total
energy (per two formulas of AlF$_3$) difference between the two phases
is also shown here, where the total energy of the cubic phase is set
to zero. The values for the rhombohedral phase are calculated at
the minimum of the energy surface. \label{tab:min}}
\begin{ruledtabular}
\begin{tabular}{lcccc}
\makebox[2.2cm] &
\makebox[0.8cm]{lattice} &
\makebox[0.4cm]{$\theta(^\circ$)} &
\makebox[0.4cm]{$\delta$} &
\makebox[0.4cm]{$\Delta$E} \\
& const.($\AA$)&&&(meV) \\
\hline
psp(Al$^\dag$, F$^\dag$) rhom & 4.76 & 57.5 & 0.098 & -93 \\
experiment (LT) & 4.9382 & 58.82 & 0.0691 & --- \\
psp(Al$^\dag$, F$^\dag$) cubic & 5.02 & 60.0 & 0.0 & --- \\
experiment (HT) & 5.0549 & 59.94 & 0.0129 & --- \\
\end{tabular}
\end{ruledtabular}
\end{table}
%
%
\subsection{Phonon Modes}
Previous workers \cite{MF3} investigated the phonon spectrum of the
cubic phase using the generalized Gordon-Kim method\cite{Ivanov}.
Density functional theory was applied to construct the charge
distribution and polarizability of ion. An approximate crystal energy
is then calculated from the ion results, and is used to examine
the crystal lattice dynamics. However they did not find an instability
of the cubic phase, which is contradictory to the experiment. Here we use
full LDA to examine the energy surface around the rhombohedral minimum,
and extract the soft phonon modes directly. The rhombohedral minimum
described in the previous section, and the dipole aspect discussed
in Sec. \ref{sec:dipole}, both show the instability of the cubic phase.
At the $\Gamma$ point of the rhombohedral phase, the irreducible
representations are those of the point group $D_{3d}$ (note that the
space group of the rhombohedral phase is $R\overline{3}C$, which is
nonsymmorphic.) Among the 12 $\Gamma$ point modes provided by the fluorines,
namely, $A_{1{\rm g}} \oplus 2A_{2{\rm g}} \oplus 3E_{\rm g} \oplus A_{1\rm u}
\oplus 2A_{\rm u} \oplus 3E_{\rm u}$, the four modes
$A_{1{\rm g}} \oplus 3E_{\rm g}$ are Raman active. Viewed from the cubic
phase, the octahedron rotation that governs the structural transition is the
R$_5$ mode (in Kovalev labelling) at the zone boundary
($\pi,\pi,\pi$)\cite{Raman1}. After distortion, the R$_5$ mode gives rise
to the zone-center A$_{\rm 1g}$ mode and one of the three E$_{\rm g}$ modes.
Raman experiments \cite{Raman1, Raman2} showed that the A$_{\rm 1g}$ and one
E$_{\rm g}$ are the soft phonon modes below transition. The A$_{\rm 1g}$ is
the Al-F-Al bond angle bending mode. Using the fluorine atomic mass, and the
energy {\it vs} $\delta$ curve, the A$_{\rm 1g}$ frequency is calculated to be
205 cm$^{-1}$. We also calculate the other restoring coefficients of the
three E$_{g}$ modes, and the two Raman-inactive A$_{\rm 2g}$
modes. Their frequencies are listed in Table. \ref{tab:freq}.
%
\begin{table}
\caption{Some of the $\Gamma$ point phonon modes frequencies calculated by LDA.
direct comparison with experiments are not available since previous
observations are done at RT and above. However, in the
experiment it is found that the two high energy E$_{\rm g}$ modes only
weakly depend on temperature. A crude extrapolation using
$\omega^2 = A(T-T_c)$ (from Curie-Weiss law and Lyddane-Sachs-Teller relation)
and the RT experimental data\cite{Raman1} gives the frequency values at
$T=0$. \label{tab:freq}}
\begin{ruledtabular}
\begin{tabular}{lcccc}
\makebox[1.0cm] &
\makebox[0.6cm]{A$_{1g}$} &
\makebox[0.6cm]{E$_g$} &
\makebox[0.6cm]{E$_g$} &
\makebox[0.6cm]{E$_g$} \\
\hline \\
LDA & 205 & 182 & 350 & 487 \\
experiment(RT) & 158 & 98 & 383 & 481 \\
extrapolation & 190 & 118 & 388 & 481 \\
\hline\hline
& A$_{2g}$ & A$_{2g}$ & & \\
\hline
LDA & 361 & 691 & & \\
\end{tabular}
\end{ruledtabular}
\end{table}
%
\subsection{Dielectric Function}
The electronic part of the dielectric tensor
$\epsilon(\mbox{\boldmath $q$}=0,\omega)$ is also calculated.
$\epsilon_{xx} = \epsilon_{yy} \not= \epsilon_{zz}$ is expected since
we have chosen lattice vectors such that $\hat x$ and $\hat y$ are
perpendicular to the three-fold axis. The imaginary part of the dielectric
function can be directly obtained by calculating all direct inter-band
transitions \cite{Yu}:
%
\begin{eqnarray}
\epsilon_2(\omega) &=& (\frac{2\pi e}{m\omega})^2
\sum_{k}\sum_{m,n}\left|\langle\psi_{n,k}\left|\vec p
\right|\psi_{m,k}\rangle\right|^2
f_n(1-f_m) \nonumber \\
&\times& \delta(E_m-E_n-\hbar\omega),
\end{eqnarray}
%
where $\vec p$ is the momentum operator, and k runs through all k-points
in the Brillouin zone allowed in a unit volume. The band indices are
$m$ and $n$, $f_{n,k}$ and $f_{m,k}$ are the occupation number of the
$n^{\rm th}$ and $m^{\rm th}$ states at the $k^{\rm th}$ k-point. The real
part of dielectric function is obtained using the Kramers-Kronig relation.
The result at $T$=0 is given in Fig. \ref{fig:dielectric}, but we do not know
of any experiment to compare with. However this calculation does neglect
the non-local potential effect \cite{nonlocalV} and the local field effect
\cite{localE}, and follows the pseudo-wave-functions that are smoothened
in the ion core regions.
%
\begin{figure}[htbp]
\centerline{\scalebox{0.35}[0.35]{\includegraphics*{dielectric.eps}}}
\caption{Plots of dielectric functions. This calculation includes
50 states and uses a k-point mesh $6\times6\times6$.}
\label{fig:dielectric}
\end{figure}
%
%
\section{Inner Parameter $\delta$ and Dipole Formation}
\label{sec:dipole}
The formation of a dipole plays an important role in the energy difference
between the two $\alpha$-AlF$_3$ structures. When the inner parameter
$\delta$ is non-zero, each fluorine atom develops a dipole (Aluminum
atoms sit on inversion centers and therefore cannot have dipoles).
Starting from the cubic structure, the distortion from $\delta$ alone
defines the octahedron rotation angle $\omega$ about the [111] axis
(see Eq. \ref{eq:delta}) and already lowers the symmetry to $R\overline{3}C$.
The parameter $\delta$ is irrelevant to the $c/a$ ratio, but it does flatten
the octahedron (see Eq. \ref{eq:zeta}). It also elongates the Al-F bond
length, and introduces a polarization to each fluorine. The energy gain from
the fluorine dipole-dipole interaction drives the structure off-cubic. This
is shown in Fig.\ref{fig:delta}, where the cubic structure is found to be not
a metastable solution.
A naive picture of ionic solids has spherical electron charge clouds of
total charge Q$_i$ around the $i^{\rm th}$ ion. In reality, charge clouds
are distorted. For example, in high T cubic $\alpha$-AlF$_3$, the F ions
are noticeably prolate when examined in a (100)-plane charge contour
calculation \cite{Madden}. This distortion is still evident in the low T
rhombohedral phase, as shown in Fig. \ref{fig:CDP}. The value of the charge
Q$_i$, on the other hand, is not uniquely definable. By creating a sphere
centered at each fluorine atom, with a radius that only allows neighboring
spheres in the rhombohedral phase to touch each other at one point, we define
a volume to examine the charge and the dipole moment of the fluorine.
In the cubic phase, such a sphere contains charge -0.675$|e|$, while it
contains -0.68$|e|$ in the rhombohedral phase (with a slightly different
radius.) Using the same sphere, we compute the local induced dipole moment
$\mbox{{\boldmath ${p}$}}_{\rm ind}$ of the fluorine ion (which also has no
unique definition \cite{AlCl3}.) As expected from symmetry, $p_{\rm ind}=0$
in the cubic phase, whereas $p_{\rm ind}=0.103 |e|\AA$ in the rhombohedral
phase, pointing from the fluorine position of the unbent Al-F-Al bond towards
the actual distorted fluorine position. Note that the displacive dipole
is $p_{\rm dis} = 0.315 |e|\AA$, defined by the fluorine charge -0.675$|e|$
and the displacement $0.467\AA$ of the fluorine ion when the Al-F-Al bonds
bend. By symmetry, the other fluorine sitting at the opposite side of the
octahedron has its dipole pointing exactly in the opposite direction.
Therefore the material could be called antiferroelectric \cite{AFE}.
%
\begin{figure}[bpt]
\centerline{\scalebox{0.42}[0.42]{\includegraphics*{newrhom100.eps}}}
\caption{Charge density plot of the rhombohedral phase on a (100) plane.
The [100] direction is ill-defined since it is inherited from the cubic
phase, but the plane shown here does contain the aluminum ions (labelled
in solid dots) that form the (100) plane in the cubic phase. The
fluorine ions are slightly off the plane shown here, and display clearly
a prolate charge distribution.
\label{fig:CDP}}
\end{figure}
%
The dipoles and charges can as well be determined by doing integration in
one of the equivalent polyhedra that surround the F$^-$ ions and partition
the whole space (the Al$^{3+}$'s sit on shared vertices of the F$^-$
polyhedra.) In this way, the ions are charged as Al$^{3+}$ and F$^-$,
while $p_{\rm ind}=0.0963 |e|\AA$ and $p_{\rm dis} = 0.467 |e|\AA$.
It is worthwhile to see if a simplified picture with these numbers
helps to understand the energy. In LDA, the energy includes
the exact electrostatic or Hartree energy from the electron charge
clouds, as well as other quantum effects (energy of delocalization,
exchange and correlation.) Purely classical electrostatic models (plus
hard core repulsion), on the other hand, provide a simple but useful
view. We shall see how much classical electrostatic energy comes
from the charge and if this helps to explain the stability of the
structural deformation.
In a simplified picture where only electrostatic energy and dipole
formation energy ($E_1$ in Eq. \ref{eq:madel}, where $\alpha = 0.858 \AA^3$
\cite{Kittel}) are considered, the total energy is
%
\begin{eqnarray}
E &=& E_1 + E_2 + E_3 + E_4 \nonumber \\
&=& \sum_i\frac{\mbox{\boldmath $p$}^2_i}{2\alpha}
+ \sum_i\sum_{j<i}\frac{Z_iZ_j}{r_{ij}}
+ \sum_i\sum_{j<i}\frac{Z_i(\mbox{\boldmath $p$}_j
\cdot \mbox{\boldmath $r$}_{ij})}{r^3_{ij}}
\nonumber \\
&+& \sum_i\sum_{j<i}\frac{\mbox{\boldmath $p$}_i
\cdot \mbox{\boldmath $p$}_j-3(\mbox{\boldmath $\hat r$}_{ij}
\cdot \mbox{\boldmath $p$}_i)(\mbox{\boldmath $\hat r$}_{ij}
\cdot \mbox{\boldmath $p$}_j)}{r^3_{ij}}.
\label{eq:madel}
\end{eqnarray}
%
This form leaves out the quantum effects and the effect from covalent
bond angle. We find that $E_2$, which is purely ionic electrostatic
(including displacive dipoles), is less negative in the rhombohedral
structure than in the cubic structure (i.e., if the same ion charge
assignment is considered in both cases.) However, when the
induced-dipole-related terms ($E_1+E_3+E_4$) are also considered, the energy
is lowered. By assuming point charges and point dipoles, we use the results
from the polyhedron method to illustrate this picture: $E_2$ per two AlF$_3$
formulas in the rhombohedral phase is 1 eV higher than that in the cubic
phase; but $E_3$ and $E_4$ bring the energy $E$ down, with the cost of $E_1$,
to 0.97 eV lower than the cubic phase (which has only $E_2$.)
This suggests that the effect of induced dipoles is more than enough to
compensate the energy loss from the structural distortion.
%
%\begin{table}
% \caption{The comparison of electrostatic energies(per two AlF$_3$ formulas).
% In (A), for the cubic phase, each fluorine ion is assigned a charge of
% -0.675$|e|$, while a charge of -0.682$|e|$ instead is assigned for the
% rhombohedral phase. The compensating charges +2.025$|e|$ and +2.046$|e|$ are
% assigned to the aluminum ions for the two phases separately. A point dipole
% of magnitude $|\mbox{\boldmath $p$}_i|$=0.103$|e|\AA$ is assigned to each
% fluorine ion, and the dipole formation energy uses polarizability
% $\alpha=0.858\AA^3$ \cite{Kittel}. In (B), the similar calculations are
% done, but with F$^-$ and Al$^{3+}$ ions for both phases. The magnitude of
% dipoles in this case is $|\mbox{\boldmath $p$}_i|$=0.0963 $|e|\AA$, determined
% by doing integration in one of the polyhedra that surround the F$^-$ ions
% and partition the whole space. \label{tab:madel}}
% \begin{ruledtabular}
% \begin{tabular}{lcc}
% \makebox[3cm] &
% \makebox[2cm]{(A) Energy} &
% \makebox[2cm]{(B) Energy} \\
% &(meV)&(meV)\\
% \hline
% ionic electrostatic (cubic) & 0 & 0\\
% ionic electrostatic (rhom.) & -915 & +1010\\
% rhom. phase with dipole effect & -2201 & -973\\
% \end{tabular}
% \end{ruledtabular}
%\end{table}
\section{Conclusion}
We report an LDA study of bulk $\alpha$-AlF$_3$. By examining the $T$=0 energy
surface for structures of the phases on both sides of the transition, we find
the structural parameters to agree with previous experiments, and the cubic
phase not to be a metastable solution. Using the result of LDA, the density of
states plot and the dielectric function are provided. At the $\Gamma$-point,
the predicted A$_{\rm 1g}$ soft phonon mode and E$_{\rm g}$ modes are compared
with previous Raman experiment, while two other A$_{\rm 2g}$ modes are predicted.
We look at the charge and dipole moment at each fluorine ion, and use these
quantities to calculate the classical electrostatic energy. In the
antiferroelectric distortion which accompanies the structural transition,
a classical calculation shows that the electrostatic energy gain from dipoles
is more than enough to compensates the energy loss from the ion-array deformation.
\acknowledgments
We thank C. Grey and S. Chaudhuri for suggesting the project and for
helpful discussions. The LDA code (BEST) and the FT-LAPW code is provided
by Brookhaven National Laboratory. Work at Stony Brook was supported in
part by NSF grant no. DMR-0089492. Work at Columbia was supported in part
by the MRSEC Program of the National Science Foundation under Award Number
DMR-0213574.
\begin{references}
\bibitem{phases1} N. Herron, D. L. Thorn, R. L. Harlow,
G. A. Jones, J. B. Parise, J. A. Fernandez-Baca,
and T. Vogt, Chem. Mater. {\bf 7}, 75 (1995).
\bibitem{phases2} C. Alonso, A. Morato, F. Medina, F. Guirado, Y. Cesteros,
P. Salagre, and J. E. Sueiras, Chem. Mater. {\bf 12},
1148 (2000).
\bibitem{catalytic} E. Kemnitz and L. E. Manzer, Prog. Solid State Chem.,
{\bf 26} 97 (1998).
\bibitem{Grey} P. J. Chupas, M. F. Ciraolo, J. C. Hanson, and C. P. Grey,
J. Am. Chem. Soc. {\bf 123}, 1694 (2001).
\bibitem{Woodward} P. M. Woodward, Acta Cryst. {\bf B53}, 32 (1997).
\bibitem{T&M} N. Troullier and J. L. Martins,
Phys. Rev. B {\bf 43}, 1993 (1991).
\bibitem{Blaha} P. Blaha, K. Schwarz, and J. Luitz,
in {\it Proceedings of WIEN97}
(Techn. Universit\"{a}t Wien, Austria, 1999).
\bibitem{Singh} D. J. Singh, {\it Planewaves, Pseudopotentials
and the LAPW Method} (Kluwer Academic, Boston, 1994).
\bibitem{CA} D. M. Ceperley and B. J. Alder.
Phys. Rev. Lett., {\bf 45}, 566, 1980.
\bibitem{VWN} S. H. Vosko, L. Wilk, and M. Nusair,
Can. J. Phys., {\bf 58}, 1200, 1980.
L.~Wilk and S.~H. Vosko,
J. Phys. C Solid State, {\bf 15}, 2139, 1982.
\bibitem{MF3} V. I. Zinenko and M. G. Zamkova, Phys. Solid State
{\bf 42}, 1348 (2000).
\bibitem{Ivanov} O. V. Ivanov and E. G. Maksimov, JETP {\bf 81} 1008 (1995).
\bibitem{Raman1} P. Danial, A. Bulou, M. Rousseau, J. Nouet, J. L. Fourquet,
M. Leblanc, and R. Burriel, J. Phys.: Condens. Matter
{\bf 2}, 5663 (1990).
\bibitem{Raman2} P. Danial, A. Bulou, M. Rousseau, and J. Nouet, Phys. Rev.
B {\bf 42}, 10545 (1990).
\bibitem{Yu} P. Yu and M. Cardona, {\it Fundamentals of Semiconductors},
second edition, page 251. (Springer, 1999)
\bibitem{nonlocalV} B. Adolph, V. I. Gavrilenko, K. Tenelson, F. Bechstedt
and R. Del Sole, Phys. Rev. B {\bf 53} 9797 (1996).
\bibitem{localE} M. S. Hybertsen and S. G. Louie, Phys. Rev. B {\bf 35}
5585 (1987).
\bibitem{Madden} This was pointed out to us by P. Madden (private
communication), and later also appeared in our charge
contour plot.
\bibitem{AlCl3} L. Bernasconi, P. A. Madden and M. Wilson,
Phys. Chem. Comm. {\bf 5}, 1 (2002).
\bibitem{AFE} R. Blinc and B. \v{Z}e\v{k}s, {\it Soft Modes
in Ferroelectrics and Antiferroelectrics},
edited by E. P. Wohifarth
(North-Holland Publishing Co. Amsterdam, 1974)
\bibitem{Kittel} C. Kittel, {\it Introduction to Solid State Physics},
seventh edition, page 391 and references therein, John
Wiley and Son, New York.
\end{references}
\end{document}