Van der Waals Density Functional for Layered Structures
Extracted source text ยท arXiv cond-mat/0306033
\documentclass[prl,aps,showpacs,superscriptaddress,amsmath,twocolumn,floatfix]{revtex4}
%\documentclass[preprint,prl,aps,showpacs,superscriptaddress,prl,amsmath]{revtex4}
\usepackage{graphicx}
\begin{document}
\title{Van der Waals Density Functional for Layered Structures}
\author{H.~Rydberg}%,$^{1}$
\affiliation{
Department of Applied Physics, Chalmers University of Technology and G{\"o}teborg University,
SE-412 96 G\"{o}teborg, Sweden}
\author{M.~Dion}%$^{2}$
\affiliation{Center for Materials Theory,
Department of Physics and Astronomy, Rutgers University, Piscataway, New Jersey 08854-8019}
\author{N.~Jacobson}%,$^{1}$
\affiliation{
Department of Applied Physics, Chalmers University of Technology and G{\"o}teborg University,
SE-412 96 G\"{o}teborg, Sweden}
\author{E.~Schr\"oder}%,$^{1}$
\affiliation{
Department of Applied Physics, Chalmers University of Technology and G{\"o}teborg University,
SE-412 96 G\"{o}teborg, Sweden}
\author{P.~Hyldgaard}%,$^{1}$
\affiliation{
Department of Applied Physics, Chalmers University of Technology and G{\"o}teborg University,
SE-412 96 G\"{o}teborg, Sweden}
\author{S.I.~Simak}%,$^{1}$
\affiliation{
Department of Applied Physics, Chalmers University of Technology and G{\"o}teborg University,
SE-412 96 G\"{o}teborg, Sweden}
\author{D.C.~Langreth}%,$^{2}$
\affiliation{Center for Materials Theory,
Department of Physics and Astronomy, Rutgers University, Piscataway, New Jersey 08854-8019}
\author{B.I.~Lundqvist}\thanks{To whom
correspondence should be addressed;
E-mail: lundqvist@fy.chalmers.se.}%$^{1}
\affiliation{
Department of Applied Physics, Chalmers University of Technology and G{\"o}teborg University,
SE-412 96 G\"{o}teborg, Sweden}
\date{August 6, 2003}
%%%%%%%%%%%
\begin{abstract}
To understand sparse systems we must account for both strong local atom bonds and
weak nonlocal van der Waals forces between atoms separated by empty space. A fully
nonlocal functional form [H. Rydberg, B.I. Lundqvist, D.C. Langreth, and M. Dion,
Phys.\ Rev.~B {\bf 62}, 6997 (2000)] of density-functional theory
(DFT) is applied here to the layered systems graphite, boron nitride, and molybdenum
sulfide to compute bond lengths, binding energies, and compressibilities. These
key examples show that the DFT with the generalized-gradient approximation does not
apply for calculating properties of sparse matter, while use of the fully nonlocal
version appears to be one way to proceed.
\end{abstract}
\pacs{71.15.Mb, 61.50.Lt, 31.15.Ew, 73.21.-b}
\maketitle
Calculations of structure and other properties of
sparse systems must account for {\it both} strong local atom bonds {\it and}
weak nonlocal van der Waals (vdW) forces between atoms separated by empty space.
Present methods are unable to describe the true interactions
of sparse systems, abundant among materials and molecules.
Key systems, like graphite, BN, and MoS$_2$, have layered structures. While
today's standard tool, density-functional theory (DFT), has broad application,
the common local (LDA) and semilocal density approximations (GGA)
\cite{PW91,PBE,ZhYa98,RPBE} for exchange and correlation, $E_\text{xc}[n]$,
fail to describe the interactions at sparse electron densities. Here we show
that the recently proposed density functional~\cite{RyLuLaDi00} with nonlocal
correlations, $E^\text{nl}_\text{c}[n]$, gives separations, binding energies,
and compressibilities of these layered systems in fair agreement with
experiment. This planar case bears on the development of
vdW density functionals for general geometries~\cite{RyThesis,DiRyLuLa03},
as do asymptotic vdW functionals \cite{everything}.
Figure 1 with its `inner surfaces' defines the problem: voids of
ultra-low density, across which electrodynamics leads to vdW coupling.
This coupling depends on the polarization properties of the layers themselves,
and {\it not} on small regions of density overlap between the layers,
excluding proper account in LDA or GGA. For large interplanar separation $d$
the vdW interaction energy between planes behaves as $-c_4/d^4$, while LDA or
GGA necessarily predicts an exponential falloff. Layers rolled up to form two
(i) nanotubes with parallel axes a distance $l$ apart interact as $-c_5/l^5$,
or (ii) balls ({\it e.g.}, C$_{60}$), a distance $r$ apart, as $-c_6/r^6$.
If by fluke an LDA or GGA were to give the correct equilibrium for one
shape, it would necessarily fail for the others. The simple expedient
of adding the standard asymptotic vdW energies as corrections to the
correlation energy of LDA or GGA also fails. The true vdW interaction
between two close sheets must be (i) substantially stronger (Fig.~1),
(ii) {\it seamless}, and (iii) saturate as $d$ shrinks (Fig.~2).
Like earlier work directly calculating nonlocal correlations between two
jellium slabs~\cite{DoWa99}, the vdW density functional (vdW-DF)
theory~\cite{RyLuLaDi00} used here exploits assumed planar symmetry. It
divides the correlation energy functional into two pieces, $E_\text{c}[n] =
E_\text{c}^0[n]+E_\text{c}^\text{nl}[n]$, where $E_\text{c}^\text{nl}[n]$
is defined to include the longest ranged or most nonlocal terms that give
the vdW interaction and to approach zero in the limit of a slowly varying
density. The term $E_{\rm c}^0[n]$ is also nonlocal, but approaches the LDA
in this limit. The two terms are approximated differently. Here the LDA is
used for $ E_\text{c}^0[n]$, as the LDA should be much more accurate with
the principal longest range terms separated off. Ultimately we plan to
derive a gradient or GGA expansion appropriate to $ E_\text{c}^0[n]$, but
the resulting corrections are expected to be small. Thus we use
%
\begin{equation}
%
E_\text{c}[n]\approx E_\text{c}^\text{LDA}[n]+E_\text{c}^\text{nl}[n].
%
\label{eq:ec1}
\end{equation}
%
Long range terms are less sensitive to details of the system's
dielectric response. Thus very simple approximations for the dielectric
function are made for $ E_\text{c}^\text{nl}[n]$. We care to make
the polarization properties of a single layer come out reasonably.
For systems with planar symmetry, the predominant component of the
nonlocal correlation energy $E^\text{nl}_\text{c}$ giving the vdW forces can
be determined simply by comparing the solutions of the Poisson equation
$\nabla\cdot(\epsilon\nabla\Phi) = 0$ and the Laplace equation
$\nabla^2 \Phi = 0$~\cite{RyLuLaDi00}. The crux of the approximation is
the use of a simple plasmon-pole model for the dielectric function,
%
\begin{equation}
\label{eq:epsilon}
\epsilon_k(z,iu) =
1+\frac{\omega_{\rm p}^2(z)}{u^2+(v_{\rm F}(z) q_k)^2/3+q^4_k/4},
\end{equation}
%
where $\omega_{\rm p}^2(z) = 4 \pi n(z)e^2/m$, and
$mv_{\rm F}(z)=(3 \pi^2 n(z))^{1/3}$ are functions of the local density $n(z)$.
The 2D wave vector perpendicular to the $z$ direction is $\mathbf{k}$, $iu$
is the imaginary frequency and $q_k^2=k^2 + q_\perp^2$ mimicks the 3D wave
vector, with the $z$ direction accounted for by the $n(z)$ dependence and
$q_\perp$ taken as a constant.
%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=.4\textwidth]{Fig1Fin.eps}}
\caption{\label{fig1}
Schematics of the vdW forces in sparse, layered materials.
{\it Top panel\/}:
Laterally averaged electron-density profile of two graphene layers at
equilibrium separation $d\approx$ 3.5 {\AA}. The dotted ellipse indicates the
region determining the interlayer interaction within LDA or GGA. It is
unable to describe the true long range interaction (horizontal arrow)
between correlated charge fluctuations (extrema illustrated by those of the
finite-frequency charge response, and labelled by
vertical arrows). {\it Bottom panel}: Full $E_\text{c}^\text{nl}$ (solid curve) vs.\
the asymptotic form of $E_\text{c}^\text{nl}\rightarrow-c_4/d^4$ (dashed line).
Also shown: full ground state interaction energy (dash-dotted line).
{\it Insets}: Sketches of the correlated charge fluctuations for different layer
separations, explaining why $ |E_\text{c}^\text{nl}| >c _4/d^4$. In the asymptotic
region (top inset), attractive interactions BC and AD are nearly cancelled by the
repulsive interactions AC and BD, leaving little more than the weak dipolar
interaction. Near equilibrium separation (bottom inset), the repulsive
interactions AC and BD remain specified by the layer separation $d$, but the
relative strengthening of attractive interaction BC %(distance $\overline{BC} < d$)
substantially overcompensates the relative weakening of the attractive
interaction AD. At still smaller $d$, $E_\text{c}^\text{nl}$ becomes weaker and
saturates (Fig. 2, inset).
}
%
\end{figure}
%%%
Our scheme, applied to jellium surfaces in Ref.~\onlinecite{RyLuLaDi00}, is
used here for layered systems, for simplicity a two-layered system
({\it e.g.,} two parallel graphene sheets) arranged perpendicular to the
$z$ axis, at positions $z=0$ and $z=d$, respectively. The first step is to
calculate the density in DFT with a suitable approximation, which we take
to be an appropriate flavor of the GGA, and to average it in the lateral
direction to yield $n(z)$. The next two steps represent the key to the
success of our approximation. First one calculates from first principles
the perpendicular static polarizability $\alpha$ of a {\it single} layer
of the material (a graphene layer in this example) in the ground-state DFT
scheme with the above GGA flavor. Then one fixes the constant $q_\perp$ so
that dielectric model (\ref{eq:epsilon}) gives precisely the same
polarizability, that is
%
\begin{equation}
%
\alpha = \frac{1}{4\pi} \int \! dz \left[1-\frac{1}{\epsilon_0(z,0)}\right].
\label{alpha}
\end{equation}
%
This step mitigates the errors introduced both by the lateral averaging and
the dielectric model. The leading vdW interaction is proportional to an
integral over the square of the dynamic polarizabilities. Previous work in
the asymptotic limit \cite{everything} shows that simple, properly scaled
dielectric functions giving the correct static polarizabilities form good
approximations to the dynamic ones, hence giving accurate vdW interactions.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=.4\textwidth]{Fig2Fin.eps}}
\caption{\label{fig2}
Calculated vdW-DF results for the interlayer binding (adhesion)
between two sheets of graphene ({\it upper panel}) and MoS$_2$
({\it lower panel}) as functions of the interlayer separation $d$. In each
panel it is compared (solid curves) with the underlying GGA (dashed curves),
and with the adhesion {\it in the absence of} the vdW-DF term $E_\text{c}^\text{nl}$
(dash-dotted curves) calculated using only the first two terms of Eq.~(\ref{eq:exc1})
for $E_\text{xc}$. The nonlocal-correlation vdW contribution $E_{c}^\text{nl}$ alone
is shown separately (dotted curves). The figure shows that no binding results
from exchange or local correlation; the vdW interactions provides adhesion in
fair agreement with experiment (Table I). The {\it inset} shows that
our calculated value of $E_{c}^\text{nl}$
saturates for small $d$ values.
}
\end{figure}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
The correlation energy can be calculated from the charge response%
~\cite{couplingconstantintegration}. With planar symmetry only the
response to an arbitrarily charged sheet $\rho \propto\exp(i\mathbf{k} \cdot
\mathbf{r})\delta (z-z_-)$ placed at an offset $z_-$ ($z_- \ll 0$) from one
end of the sample needs to be calculated. Integrating Poisson's equation
with $\epsilon$ as in (\ref{eq:epsilon}) one finds the $z$ component of the
electric field $E_z(z_+)$ at $z=z_+\gg d$. Then $E_{\rm c}^\text{nl}$ is given
by \cite{RyLuLaDi00}
\begin{equation}
%
\label{eq:enl}
E_c^\text{nl} = - A
\int_0^\infty \! \frac{du}{2\pi} \,
\int\! \frac{d^2k}{(2\pi)^2}
\ln
\frac{E_z(z_+)}{E^0_z(z_+)},
%
\end{equation}
where $A$ is the lateral area and $E^0_z(z_+)$ is the electric field
component from the charge sheet at $z=z_-$ in the absence of the sample. In
Eq.~(\ref{eq:enl}) the spatial dependences of the two $E$'s cancel in the
ratio, leaving only the $k$ dependence. For each of the bulk solids considered
here we apply this method to 32 and 30 layer slabs to give, by subtraction,
a well converged per layer value of $E_{\rm c}^{\rm nl}$.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\includegraphics[width=.4\textwidth]{Fig3Fin.eps}
\caption{\label{fig3}
Comparison of the vdW-DF layer-to-layer binding in the staggered
bulk hexagonal-BN system (solid lines) and adhesion between two
sheets of BN (broken lines) shown as functions of the interlayer
separation $d = c/2$. In both descriptions the GGA gives
unphysical results for bond and equilibrium separation.
The total
adhesion is dominated by the contribution of the vdW interaction
$E_\text{c}^{\text nl}$. The nearest-neighbor interlayer binding
(broken lines) is seen to strongly dominate the bulk adhesion
(solid lines) for layered materials.
}
\end{figure}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
The approximation for the exchange energy functional $E_{\rm x}[n]$
should be consistent with our approximation for $E_{\rm c}^0[n]$,
that is a local or semilocal one. We use a version that best
approximates exact density functional exchange,
namely Zhang and Yang's~\cite{ZhYa98} (ZY) ``revPBE'' exchange functional
$E_\text{x}^\text{ZY}[n]$, where
the parameter controlling
the large gradient limit in the PBE exchange functional~\cite{PBE} is fitted
to exact density-functional exchange calculations. Unlike exchange
\cite{ScolesPlus} in PW91 \cite{PW91}, PBE \cite{PBE}, and RPBE~\cite{RPBE},
$E_\text{x}^\text{ZY}$ shows no tendency to bind any of the vdW systems we
have tried it on, unless the vdW correlation is actually included in the
calculation. We thus take
\begin{equation}
E_{\rm xc}[n] = E_\text{x}^\text{ZY}[n]+ E_\text{c}^\text{LDA}[n]+E_\text{c}^\text{nl}[n].
%
\label{eq:exc1}
\end{equation}
The first two terms form an important and recommended starting point
for testing nonlocal vdW functionals. They assure a vdW attraction that
actually comes from terms that in principle are capable of giving it.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{table}
\caption{\label{table1}
Properties of graphite,
BN, and MoS$_2$, calculated with vdW-DF and compared to experimental data
in parentheses, when available. The table shows the geometry ($a,c$),
binding energy ($E_0$), bulk modulus ($B_0$), and elastic constant ($C_{33}$).
GGA (not shown) binds very weakly at unphysically large $c$ or not at all,
depending on substance and GGA flavor.}
\begin{ruledtabular}
\begin{tabular}{lccc}
& Graphite& BN& MoS$_2$ \\
\hline
$a$ [\AA]& 2.47 (2.46)\footnotemark[1] & 2.51 (2.50)\footnotemark[2] & 3.23 (3.160)\footnotemark[3]\\
$c$ [\AA]& 7.52 (6.70)\footnotemark[4]& 7.26 (6.66)\footnotemark[2] & 12.6 (12.29)\footnotemark[3] \\
$E_0$ [meV/at]& 24 ($35 \pm 10$)\footnotemark[5] & 26& 60\\
$B_0$ [GPa]& 12 ($\sim$$33$)\footnotemark[1] & 11 & 39\\
$C_{33}$ [GPa]& 13 (37-41)\footnotemark[1] & 11 & 49\\
\end{tabular}
\end{ruledtabular}
\footnotetext[1]{Ref.~\onlinecite{Bornstein80}.}
\footnotetext[2]{Ref.~\onlinecite{KeKrHa99}.}
\footnotetext[3]{Ref.~\onlinecite{MoS2exp}.}
\footnotetext[4]{Ref.~\onlinecite{BaMa55}.}
\footnotetext[5]{Ref.~\onlinecite{BCCZLC98}.}
\end{table}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
The GGA is used in order to have a good account of strong valence bonds and
densities of individual layers, and the above scheme is implemented by
self-consistently calculating the energy $E^\text{GGA}$ and density
$n^\text{GGA}$, typically in the revPBE flavor
of GGA. The total energy, as a function of the lattice parameters, is then
calculated as $E\approx E^{\rm GGA}[n^{\rm GGA}] +\Delta[n^{\rm GGA}]$,
where $\Delta[n]= E_{\rm c}^{\rm nl}[n] - E_{\rm c}^{\rm grad}[n]$ and
$ E_{\rm c}^\text{grad}[n] = E_{\rm c}^\text{GGA}[n] - E_{\rm c}^\text{LDA}[n]$.
This approximation treats $\Delta$ as a post GGA perturbation. The GGA calculations
are done using the plane-wave pseudopotential~\cite{corecomment} DFT code
dacapo~\cite{DACAPO}.
The three materials considered here have layered structures
with a soft direction perpendicular to the layers
\cite{separation}. Strong covalent bonds occur between the atoms within
the hexagonal layers (lattice constant $a$), while weak vdW interactions
occur between the layers. Graphite has a staggered or A-B type stacking
with the intralayer C atoms in regular, planar hexagons stacked with
corners on top of centers. Boron nitride, isoelectronic with graphite,
has no shift in hexagon locations but B and N atoms in alternate positions
along the $c$-direction. A similar alternation occurs between Mo and S
atoms in the A-B type structure of MoS$_2$ \cite{MoS2 footnote}.
Calculations are also performed for bilayers of the three systems. The
binding-energy curve of two parallel graphene sheets~\cite{DDSSG88},
{\it i.e.} the total-energy difference at separation $d$ and at infinite
separation, respectively, for varying $d$ (Fig.~2) gives in the
approximation~\cite{RyLuLaDi00} with vdW-DF, Eq.~(\ref{eq:exc1}), a close
relation to experimental findings for binding energy and equilibrium
distance. The GGA curve, on the other hand, is completely wrong, which
we blame on absence of vdW effects. The inset shows that our
$E_c^\text{nl}$ contribution gives stronger binding than the
traditional asymptotic vdW interaction. The binding-energy curve for
two parallel BN sheets (Fig.~3) differs little from that of bulk BN
as a function of lattice constant $c$. At equilibrium the calculated
$E_\text{c}^\text{nl}$ difference is $\sim 4$\%, attributable mostly to
2nd nearest layers in the bulk material \cite{fourpercent}.
Bulk-modulus values are computed together with other structure ($a$,$c$)
and bonding properties \cite{subtraction} (Table I). It is crucial to
densely sample the region of ($a$,$c$) values around the optimal structure
($a_0$, $c_0$), and we use a new method \cite{ZiSc03} for direct evaluation
of both structure, binding energy, and bulk modulus $B_0$ in the relevant
range of $a$ and $c$ values \cite{B0comment}. GGA values were also
computed but not shown. For all three materials the various GGA flavors
give no binding or bind very weakly at unreasonably large separation.
The vdW-DF, on the other hand, gives values for lattice parameters and
cohesive energy in good agreement and for bulk modulus in fair agreement
with experimental values, when available.
In conclusion, we recommend the replacement of GGA as a standard
method in total-energy calculations with vdW-DF as given in
Eq.~(\ref{eq:exc1}) for descriptions of layered systems that contain
sparse electron distributions. This will give the right qualitative
character of soft bonds, including saturation of the vdW potential
at small separations, and even quantitative ones, like physical
values of bond lengths, binding energies, and compressibilities
\cite{improvements}.
We are grateful to K.~W.~Jacobsen for suggesting tests on MoS$_2$
layers \cite{Boll}. Financial support from the Swedish Foundation for
Strategic Research via Materials Consortia \#9 and ATOMICS, STINT, and
the Swedish Scientific Council are gratefully acknowledged. Work by
M.D.~and D.C.L.~supported in part by NSF Grant DMR 00--93070.
%%%%
\begin{thebibliography}{99}
\bibitem{PW91}
J. P. Perdew
{\em et al.},
Phys.\ Rev.\ B {\bf 46}, 6671 (1992);
{\bf 48}, 4978(E) (1993).
\bibitem{PBE}
J. P. Perdew, K. Burke, and M. Ernzerhof,
Phys.\ Rev.\ Lett.\
\textbf{77}, 3865 (1996).
\bibitem{ZhYa98}
Y. Zhang and W. Yang,
Phys.\ Rev.\ Lett.\ \textbf{80}, 890 (1998).
\bibitem{RPBE}
B. Hammer, L. B. Hansen, and J. K. N{\o}rskov,
Phys.\ Rev.\ B \textbf{59}, 7413 (1999).
\bibitem{RyLuLaDi00}
H.~Rydberg, B.~I. Lundqvist, D.~C. Langreth, M.~Dion,
Phys.\ Rev.\ B {\bf 62}, 6997 (2000).
\bibitem{RyThesis}
H.~Rydberg,
Ph D. Thesis (2001), Chalmers ISBN 91-7291-068-2.
\bibitem{DiRyLuLa03}
M. Dion, H. Rydberg, B. I. Lundqvist, and D. C. Langreth,
to be published.
\bibitem{everything}
See E. Hult, P. Hyldgaard, J. Rossmeisl, and B. I. Lundqvist,
Phys. Rev. B \textbf{64}, 195414 (2001) and
E.~Hult, H.~Rydberg, B.~I. Lundqvist, D.~C. Langreth,
Phys. Rev. B {\bf 59}, 4708 (1998), and references therein.
\bibitem{DoWa99}
J. F. Dobson and J.~Wang,
Phys.\ Rev.\ Lett.\ {\bf 82}, 2123 (1999).
\bibitem{couplingconstantintegration}
The coupling constant integration in $E^{\rm nl}$ is performed
via the following approximation:
$\epsilon -1 \approx \lambda [(\epsilon - 1)/\lambda]_{\lambda
=1}$. This approximation gives the long range van der Waals
interaction exactly \cite{DiRyLuLa03} for dielectric functions
of the type used here. The $\lambda$ dependence comes
via $q_\perp$.
\bibitem{ScolesPlus}
Y. Zhang, W. Pan, and Y. Yang,
J. Chem.\ Phys.~{\bf 107}, 7921 (1997);
X. Wu {\em et al.}, J. Chem.\ Phys.~{\bf 115}, 8748 (2001).
\bibitem{corecomment}
Only the valence density was used in the functional, because the lateral
averaging unphysically lowers the core density and increases the size
of the core, yielding an unphysically large vdW contribution. Nevertheless,
tests with the full density showed negligible effect on the geometry,
and at most a 30\% increase in well depth.
\bibitem{DACAPO}
\texttt{http://www.fysik.dtu.dk/CAMPOS/}.
\bibitem{separation}
Separation $d=c/2$, except for MoS$_2$, where $d$ is defined as
the distance from the perpendicular
position of the top-most sulfur atom of one layer to the position of
the bottom-most sulfur atom of the next.
\bibitem {MoS2 footnote}
In the MoS$_2$ interlayer calculations, each MoS$_2$ layer
is treated as a composite material (the Mo-S perpendicular separation
is $\sim$1.58 {\AA}). A composite $q_{\perp}$ value ($\sim$$0.58$),
identical in both MoS$_2$ layers, is obtained by applying an
external field across a single MoS$_2$ composite and using GGA to
calculate the static polarization.
In the interaction study
the vdW-DF energy is then calculated as a function of the separation
between the two MoS$_2$-layer composite materials, the atom positions
within each of the MoS$_2$-layer composites being kept fixed.
\bibitem{DDSSG88}
M.~S. Dresselhaus {\em et al.},
{\em Graphite Fibers and Filaments\/} (Springer, Berlin, 1988).
\bibitem{fourpercent}
The 4\% figure may be estimated as $1/(1.5\times 2^4)$, with the divisor of 1.5
occurring because the 50\% enhancement (Fig.~1) is not applicable at the 2nd
layer distance.
\bibitem{subtraction}
Using $E_\text{vdW-DF}(c,a) = E_\text{GGA}(c,a)
+ \Delta(c,a) - \Delta(\infty,a)$.
\bibitem{ZiSc03}
E. Ziambaras and E. Schr{\"o}der,
Phys. Rev. B, in press, \texttt{cond-mat/0304075}.
\bibitem{B0comment}
The value of 33 GPa reported
earlier~\protect\cite{SurfSci03} was incorrect.
\bibitem{Boll}
M. V. Bollinger {\em et al.}, Phys.~Rev.~Lett.~{\bf 87}, 196803 (2001);
M. V. Bollinger, K.~W.~Jacobsen, and J.~K.~N{\o}rskov,
Phys.~Rev.~B {\bf 67}, 085410 (2003).
\bibitem{SurfSci03}
H.~Rydberg,
{\em et al.},
Surf.\ Sci.\ {\bf 532--535}, 606 (2003).
\bibitem{Bornstein80}
{\it Landolt-B{\"o}rnstein search} (Springer, Berlin, 2003),
http://link.springer.de.
\bibitem{KeKrHa99}
G. Kern, G. Kresse, and J. Hafner,
Phys.\ Rev.\ B {\bf 59}, 8551 (1999).
\bibitem{MoS2exp}
Th. B\"oker {\em et al.}, Phys.\ Rev.\ B {\bf 64}, 235305 (2001).
\bibitem{BaMa55}
Y.~Baskin and L.~Mayer,
Phys.\ Rev.\ {\bf 100}, 544 (1955).
\bibitem{BCCZLC98}
L.~X. Benedict {\em et al.},
Chem.\ Phys.\ Lett.\ {\bf 286}, 490 (1998).
\bibitem{improvements}
Future work that may remove the remaining differences includes
(i) time dependent DFT and other calculations of the positions
of the polarization peaks (Fig.~\ref{fig1}) followed by
improvements to the approximation Eq.~(\ref{eq:epsilon}),
(ii) derivation of the gradient corrections appropriate to
$E_c^0$ with corresponding modifications to Eq.~(\ref{eq:ec1}), and
(iii) implementation of exact exchange with corresponding
modifications to Eq.~(\ref{eq:exc1}).
\end{thebibliography}
\end{document}
%%%%%%%%%%%%%%%%%