Metal adatoms on graphene and hexagonal boron nitride:

Towards the rational design of self-assembly templates

Oleg V. Yazyev∗ and Alfredo Pasquarello
Institute of Theoretical Physics, Ecole Polytechnique Fédérale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland and
Institut Romand de Recherche Numérique en Physique des Matériaux (IRRMA), CH-1015 Lausanne, Switzerland
(Dated: July 13, 2010)
Periodically corrugated epitaxial graphene and hexagonal boron nitride (h-BN) on metallic sub-
strates are considered as perspective templates for the self-assembly of nanoparticles arrays. By
arXiv:1007.1704v1 [cond-mat.mes-hall] 10 Jul 2010

using first-principles calculations, we determine binding energies and diffusion activation barriers of
metal adatoms on graphene and h-BN. The observed chemical trends can be understood in terms of
the interplay between charge transfer and covalent bonding involving the adatom d electrons. We
further investigate the electronic effects of the metallic substrate and find that periodically corru-
gated templates based on graphene in combination with strong interactions at the metal/graphene
interface are the most suitable for the self-assembly of highly regular nanoparticle arrays.

PACS numbers: 36.40.Sx, 61.48.De, 68.43.Bc, 68.65.Cd

I. INTRODUCTION h-BN. Then, we model the electronic effects of periodi-

cally corrugated epitaxial layers deposited on two metal-
Two-dimensional graphene and hexagonal boron ni- lic substrates representing the limiting cases of strong
tride (h-BN) are attracting considerable attention due and weak interactions at the interface.
to their extraordinary physical properties and perspec- The present paper is organized as follows. In Sec.
tive technological applications.1,2 Epitaxial single lay- II we describe our first-principles methodology and the
ers of graphene and h-BN can be grown via the chem- adopted models. Section III discusses the binding and
ical vapor deposition route on a large variety of metallic the diffusion of adatoms on free-standing graphene and
substrates.3 In many cases, the epitaxial layers of these h-BN. Section IV is devoted to the investigation of sub-
two materials are extremely well-ordered but reveal long- strate effects. The conclusions are drawn in Sec. V.
wavelength periodic corrugations, or Moiré patterns, re-
sulting from the lattice constant mismatch.4–7 Such su-
perlattices are considered as promising templates for the II. METHODS
chemical self-assembly of periodic arrays of nanoparticles,
with perspective applications in ultra-high density infor- In our approach, the electronic structure is described
mation storage, catalysis, sensing, etc.8 The potential of through the use of the Perdew-Burke-Ernzerhof (PBE)
this approach has already been affirmed by the successful exchange-correlation density functional within density
production of regular arrays of nanoclusters with narrow functional theory.13 The ultrasoft pseudopotentials14
size distributions.9–12 used in the present study treat the d-electrons of all
One attractive property of this approach to nanoscale transition metals as valence electrons. The semi-core sp-
self-assembly is the high degree of customization. Indeed, states were also treated explicitly in the case of early
the space of allowed chemical compositions involves three third row elements (K–Fe). The one-electron valence
degrees of freedom: (i) deposited metallic nanoparticles wave functions and the electron density were described
formed by virtually any metal from the periodic table; by plane-wave basis sets with kinetic energy cutoffs of
(ii) a monolayer of either graphene or h-BN; (iii) metallic 30 Ry and 300 Ry, respectively.15 All calculations were
surfaces, one of the many (111) fcc or (0001) hcp surfaces performed using the spin-unrestricted formalism. We
of forth or fifth row late transition metals. Understanding used the pwscf plane-wave pseudopotential code of the
the roles these three factors play in the self-assembly pro- quantum-espresso distribution.16 The convergence of
cess is of paramount importance for the rational design of the results with respect to the simulation parameters was
nanoparticle arrays with novel properties and functions. systematically verified.
In this work, we aim at understanding the chemical The constructed models are based on periodic two-
trends in the binding and diffusion of individual metal dimensional 3×3 supercells in combination with 4×4 k-
adatoms, the initial step which largely predetermines the point meshes. Potential energy surfaces for the free-
overall self-assembly process. In particular, by using first- standing graphene and h-BN systems were studied by
principles calculations we perform a systematic study of relaxing all atomic positions with the initial coordinates
the adatom binding and diffusion as well as of the elec- of the adatom set to the high-symmetry positions shown
tronic properties of the metal adatoms upon deposition. in Fig. 1. This procedure allowed us to determine both
First, we focus on the chemical trends observed for metal the potential energy surface minima and the transition
adatoms varying horizontally and vertically in the peri- state configurations connecting the neighboring equiva-
odic table, when absorbed on free-standing graphene and lent minima, since symmetry was retained during the
 2

relaxation. The validity of this approach was verified (a) E (eV)

h b h’
through nudged elastic band calculations.17 For specific E (eV) -0.8
cases, we could compare our results with those of other -0.8
-1.2
investigations,18,19 finding good agreement. -0.9
-1.6
We investigated the effects of local electronic-structure h
variations due to the presence of metallic substrates by b t -1.0
considering the case of Co adatoms on epitaxial graphene C -1.1
and h-BN supported by lattice-matched Ni(111) and
Cu(111) substrates. These metallic substrates corre- tB -1.2
(b) h h’
spond to regimes of strong and weak monolayer-substrate
E (eV)
interactions, respectively.20 Under realistic conditions -0.8
-1.3
these metals do not produce Moiré pattern due to the -1.2
small lattice mismatch.3,21–23 However, through models -1.4
-1.6
involving such structures, we could study electronic ef- h
b tN tB -1.5
fects induced by the Moiré superlattice using sufficiently
small simulations cells and without introducing large lat- B N
-1.6
eral strains. A very similar methodology was also used
in Ref. 4. The considered model calculations consist of
3×3 two-dimensional slabs composed of 4 atomic planes FIG. 1: (Color online) Potential energy surfaces of the Co
adatom on freestanding (a) graphene and (b) h-BN obtained
of metal with both surfaces covered by graphene or h-
by constraining the in-plane position of the Co adatom on a
BN in order to avoid spurious dipole-dipole interactions. 6×6 points mesh in the unit cell. The same energy scale is
The slab configurations were fully optimized, including used in both plots and the energy is referred with respect to
the distance between the metal layers.24 that of an isolated Co atom. The high-symmetry positions
are indicated with labels. The arrows show the lowest en-
ergy diffusion pathways connecting the local minima, h sites,
III. ADATOMS ON SUSPENDED GRAPHENE through the transition state configurations, the b sites and tB
AND h-BN sites in the case of graphene and h-BN, respectively.

First, let us compare the potential energy surfaces

(PES) for a representative case of a Co adatom on free- tively. The diffusion activation barriers are again higher
standing graphene and h-BN (Fig. 1). For both monolay- for graphene, 0.40 eV vs. 0.13 eV for h-BN. However, this
ers the PES minima correspond to the hollow (h) sites. trend is not systematic across the periodic table and very
Interestingly, in the case of Co as well as for most of often the diffusion on h-BN is characterized by higher ac-
the other metal adatoms the distortions of the graphene tivation barriers [cf. Fig. 2(a,b)]. Notably, the calculated
and h-BN lattices are very weak. Although the two activation barrier in the case of h-BN agrees well with
PES involving the Co adatom are qualitatively similar, the experimental value of 0.14±0.03 eV for the diffusion
the binding energy on graphene are substantially higher of Co adatoms on h-BN deposited on Ni(111).26
(1.60 eV vs. 1.03 eV). The stronger binding to graphene Figure 2(a) shows the horizontal trend in the binding
is a systematic feature which can be understood consider- energies and the diffusion activation barriers across the
ing that the chemisorption of metal adatoms is governed third row of the periodic table. Both graphene and h-
by the following attractive contributions: (i) strong cova- BN show double-peaked features with maximum binding
lent bonding, and (ii) moderate electrostatic interaction energies at Ti and Ni which have d2 and d8 electronic
as a result of charge transfer between the adatom and the configurations, respectively, while the very weak binding
monolayer. While graphene is a semimetal, h-BN is an of Cr and Mn corresponds to the situation of half-filled
insulator with a band gap of ≈6 eV.25 The availability d-shells. Strong binding energies with magnitudes up
of low-energy electronic states in graphene results in a to 2.08 eV (1.14 eV) for graphene (h-BN) indicate the
more efficient interaction with the adatom states and al- contribution from covalent binding involving d electrons.
lows for charge transfer between the graphene layer and However, the fact that the atomic magnetic moments due
the adatom, thus introducing an electrostatic component. to the partially filled d-shells are largely preserved upon
The preferential binding to the h sites is very common, al- binding [Fig. 2(c)] shows that graphene and h-BN act
though we find a number of exceptions, especially among as weak ligand fields with respect to the metal adatoms.
the heavy transition metals. In particular, for Pd, Ir and Large magnetic moments associated to half-filled d-shell
Pt, the bridge (b) site is the lowest energy position on elements correspond to large values of spin splitting of
graphene. The nitrogen on-top sites (tN ) correspond to d-electron states and, thus, to their reduced participa-
the PES minima on h-BN for V, Ni, Pd, Ir and Pt. tion to the covalent binding. The other elements in the
The lowest energy pathways connecting the neighbor- first (second) half of the transition metal series tend to
ing local minima involve transition states at b and tB give enhanced (reduced) magnetic moments upon bind-
sites for the Co adatom on graphene and h-BN, respec- ing. To elucidate the origin of this behavior, we per-
 3

(a) (b) QM of metal adatoms on the two monolayers [Fig. 2(c)].

2.0 Vertical trends in the periodic table are illustrated
in Fig. 1(b) for the late transition and coinage metal
adatoms. In general, the binding energies tend to in-
1.5 crease when moving down the columns of the periodic
table. The reductions can be associated with the changes
E (eV)

Eb : gr.
Eb : BN of the lowest energy adsorption sites discussed above.
1.0 Ea : gr. The largest binding energy of 2.02 eV is found for Rh on
Ea : BN
graphene while the highest activation barrier of 0.77 eV
corresponds to Pt on h-BN. At variance, the diffusion of
0.5 Pd on graphene and of Rh and Ir on h-BN are character-
ized by very low diffusion barriers (<0.05 eV) combined
with high binding energies (>1 eV). Inert coinage met-
0 als have very low binding energies and diffusion barrier,
K Sc V Mn Co Cu Ga Co Ir Ni Pt Cu Au
Ca Ti Cr Fe Ni Zn Rh Pd Ag with minimum values for Ag. Similar trends were also ob-
(c) (d) served for carbon on coinage metal surfaces.32 Even on
6 0.5
graphene, Ag shows practically no charge transfer as this
M : gr.
5 M : BN
element is located at the point of crossover between elec-
M : at. tron donating (Cu) and electron accepting (Au) behavior
QM : gr. [Fig. 2(d)]. The hole-doping of graphene which is other-
M (µB)

4 0.25
QM : BN QM (e-) wise difficult to achieve by chemical means was recently
3 demonstrated in Au deposition experiments.33 Interest-
ingly, we find that also strongly binding Ir adatoms act
2 0 as electron acceptors.
It is worth stressing that the presently used theory
1
does not provide a correct description of weak van der
0 -0.25
Waals interactions. This might quantitatively affect the
results, especially when the considered adatom does not
4s1
4s2
3d14s2
2 2

3d34s2
3d54s1
3d54s2
3d64s2
3d74s2
3d84s2
3d104s1
3d104s2
4s24p1

3d74s2
4d85s1
5d76s2
3d84s2
4d105s0
5d96s1
3d104s1
4d105s1
5d106s1
3d 4s

give rise to strong covalent bonding and/or electrostatic

interactions.
FIG. 2: (Color online) (a) Horizontal and (b) vertical chemical
trends in the binding energies Eb and the diffusion activation
barriers Ea of metal adatoms on graphene and h-BN. Insets IV. EFFECT OF METALLIC SUBSTRATE
show the concerned parts of the periodic table: K–Ga in (a)
and Co–Ir, Ni–Pt, Cu–Au in (b). (c) Horizontal and (d) ver-
tical chemical trends in the magnetic moments M and the On the surfaces of forth and fifth row transition met-
atomic charges QM of the adsorbed metal adatoms in their als, epitaxial graphene and h-BN produce Moiré patterns
lowest energy configurations. Crosses refer to the magnetic due to the mismatch between the lattice constants of the
moments of isolated atoms. Electronic configurations of the substrate and of the monolayer [cf. Fig. 3(a)]. The lo-
isolated atoms are given at the bottom of the figure for refer- cal shifts of the monolayer lattice with respect to the
ence. substrate lattice across the Moiré unit cell result in long-
wavelength modulations of the PES which drive the self-
assembly of periodic arrays of nanoparticles. These mod-
formed the Löwdin population analysis27 and found that ulations are due to variations in the monolayer-substrate
the adsorption of metal adatoms leads to the partial pro- distance and to changes in the local electronic structure.6
motion of electrons from the s- to the d-shell thus leading Below, we focus on the role of local electronic structure
to the observed changes of the magnetic moments. variations in the diffusion of individual metal adatoms on
Such trends depending on the partial d-shell fill- metal-supported graphene and h-BN. Collective effects
ing appear to be generic since a very similar behav- due to the local rehybridization induced by few-atom
ior was found for metal-benzene molecular complexes28 clusters have been considered elsewhere.34,35 In addition,
and even for strongly bound substitutional impurities in the distance modulations alone can lead to only minor
graphene.29–31 The binding energies on h-BN as well as curvature-induced effect on the adatom binding energies
the activation barriers on both monolayers follow exactly since the estimated height variations are actually rather
the same qualitative trend within the transition metal se- small (∆h<1.5 Å4,7 ) compared to the size of the Moiré
ries (Sc–Zn). Outside of this region (K, Ca and Ga) only unit cell (a∼3 nm).36
graphene is able to bind adatoms strongly since the inter- In order to understand the effect of the local electronic
actions are then governed by the sole electrostatic con- structure variations, we study the binding of Co adatoms
tribution resulting from charge transfer. This contrast is on epitaxial graphene and h-BN supported by lattice-
well illustrated by comparing the Löwdin atomic charges matched Ni(111) and Cu(111) substrates. These two
 4

(a) C (N)
B (N)
metal (layer 1)
metal (layer 2)

I: C(N) - top
C(B) - fcc
II: C(N) - fcc
C(B) - hcp
III: C(N) - hcp
I II III I C(B) - top
(b) (c) 1.7
-1.2 II
1.6

Eb (eV)
III
E (eV)

-1.4
1.5
-1.6 I
1.4
-6 -4 -2 0 2
x (arb. units) QC (x10-3 e-)

FIG. 3: (Color online) (a) Schematic illustration of the Moiré unit cell with three binding regions indicated. (b) Schematic
representation of the potential energy surface (solid line) for the diffusion of the adatom moving along the horizontal direction
of the unit cell shown in (a), based on calculations for Moiré domains I–III. The actual local binding energies (dashed line) and
diffusion barriers (oscillation amplitudes) correspond to the case of the Co adatom on graphene supported by Ni(111). The
dotted line refers to the case of the Co adatom on free-standing graphene. (c) Binding energy Eb of the Co adatom versus the
charge transfer per carbon atom QC from the metal to graphene for the three Moiré domains.

cases model the regimes of strong and weak monolayer- thin oxide films.40 For Co adatoms on h-BN|Ni(111), we
substrate interactions, respectively.20 We stress that un- found the same tendency, although the local variations of
der realistic conditions these metals do not produce Moiré the binding energy are smaller in this case. On a weakly
pattern due to the small lattice mismatch.3,21–23 The binding substrate such as the Cu(111) surface, the PES
three principal Moiré domains [I–III in Fig. 3(a)] are remains practically unchanged across the Moiré unit cell.
modeled by introducing an artificial lateral shift between The Moiré domains associated with higher binding en-
the monolayer and the substrate. Further details of this ergies are expected to show larger thermal populations
procedure are given in Sec. II. of adatoms, and will thus act as nucleation centers for
The calculated binding energies and activation barri- nanoparticles. Therefore, larger variations of the bind-
ers for Co adatoms on graphene|Ni(111) show apprecia- ing energies across the Moiré template will favor more
ble variation across the three Moiré domains [cf. Table I]. ordered nanoparticle arrays with narrower size distribu-
The resulting PES [schematically depicted in Fig. 3(b)] tions. We conclude that the templates made of graphene
can be viewed as a modulation of the PES of free- in combination with strongly binding metal surfaces (e.g.
standing graphene [Fig. 1(a)] by the long-wavelength per- Ir, Rh, Ru) are more promising for the self-assembly of
turbations resulting from the periodicity of the Moiré su- metal nanoparticle arrays.
perlattice. The Moiré domain I, which is characterized by
the shortest graphene-substrate distance (2.23 Å),37–39
shows the lowest binding energy while the opposite is
true for region II showing the largest graphene-substrate TABLE I: Comparison of binding energies Eb (in eV) and
distance (4.11 Å). However, Moiré domain III presents a diffusion activation barriers Ea (in eV) of the Co adatom on
free-standing graphene (gr.) and h-BN with corresponding
short graphene-substrate distance (2.49 Å) in combina-
epitaxial monolayers deposited on Ni(111) and Cu(111). The
tion with a high binding energy [Table I], indicating that different positions of the monolayer atoms refer to three dis-
there is no clear correlation between these quantities. At tinct local regions of the Moiré pattern shown in Fig. 3.
variance, a clear correlation can be established between
the binding energies of the Co adatoms and the local Eb Ea
charge transfer per carbon atom (QC ) from the metal to Moiré domain I II III I II III
graphene [cf. Fig. 3(c)]. This result can be intuitively un- gr. 1.60 0.40
gr.|Ni(111) 1.42 1.67 1.62 0.23 0.47 0.32
derstood: the charge transfer from the metal to graphene
gr.|Cu(111) 1.66 1.67 1.67 0.45 0.45 0.45
reduces the charge transfer from the Co adatom to the h-BN 1.03 0.13
template surface and, thus, the electrostatic component h-BN|Ni(111) 0.86 0.89 0.88 0.10 0.14 0.14
of the binding energy. Very recently a similar electro- h-BN|Cu(111) 1.17 1.17 1.17 0.16 0.20 0.20
static effect has also been found for adatoms on ultra-
 5

V. CONCLUSIONS BN which can find applications in ultra-high density in-

formation storage, catalysis and sensing, and so on.
We systematically studied the potential energy sur-
faces of metal adatoms on graphene and h-BN across
the periodic table, and clarified the role of the metallic
ACKNOWLEDGMENTS
substrate in adatom diffusion. The present results are
well understood within a simple picture involving cova-
lent and electrostatic interactions. Our work formulates We acknowledge fruitful discussions with H. Brune,
general principles required for the rational design of self- P. Buluschek, R. Decker, A. Lehnert, and S. Rusponi.
assembly templates based on epitaxial graphene and h- The calculations were performed at the CSCS.

∗
Present address: Department of Physics, Univer- Lett. 101, 026803 (2008).
21
sity of California, Berkeley, CA 94720, USA; E-mail: A. Nagashima, N. Tejima, and C. Oshima, Phys. Rev. B
[email protected] 50, 17487 (1994).
1 22
A. K. Geim and K. S. Novoselov, Nature Mater. 6, 183 Y. Gamo, A. Nagashima, M. Wakabayashi, M. Terai, and
(2007). C. Oshima, Surf. Sci. 374, 61 (1997).
2 23
M. I. Katsnelson, Materials Today 10, 20 (2007). X. Li, W. Cai, J. An, S. Kim, J. Nah, D. Yang, R. Piner,
3
C. Oshima and A. Nagashima, J. Phys.: Condens. Matter A. Velamakanni, I. Jung, E. Tutuc, S. K. Banerjee, L.
9, 1 (1997). Colombo, and R. S. Ruoff, Science 324, 1312 (2009).
4 24
R. Laskowski, P. Blaha, T. Gallauner, and K. Schwarz, K. Kádas, Z. Nabi, S.K. Kwon, L. Vitos, R. Ahuja, B.
Phys. Rev. Lett. 98, 106802 (2007). Johansson, and J. Kollár, Surf. Sci. 600, 395 (2006).
5 25
J. Coraux, A. T. N’Diaye, C. Busse, and T. Michely, Nano B. Arnaud, S. Lebegue, P. Rabiller, and M. Alouani, Phys.
Lett. 8, 565 (2008). Rev. Lett. 96, 026402 (2006).
6 26
A. L. Vázquez de Parga, F. Calleja, B. Borca, M. C. G. W. Auwärter, M. Muntwiler, T. Greber, and J. Oster-
Passeggi, J. J. Hinarejos, F. Guinea, and R. Miranda, walder, Surf. Sci. 511, 379 (2002).
27
Phys. Rev. Lett. 100, 056807 (2008). P.-O. Löwdin, J. Chem. Phys. 18, 365 (1950).
7 28
D. Martoccia, P. R. Willmott, T. Brugger, M. Bjorck, R. Pandey, B. K. Rao, P. Jena, and M. A. Blanco, J. Am.
S. Gunther, C. M. Schleputz, A. Cervellino, S. A. Pauli, Chem. Soc. 123, 3799 (2001).
29
B. D. Patterson, S. Marchini, J. Wintterlin, W. Moritz, A. V. Krasheninnikov, P. O. Lehtinen, A. S. Foster,
and T. Greber, Phys. Rev. Lett. 101, 126102 (2008). P. Pyykko, and R. M. Nieminen, Phys. Rev. Lett. 102,
8
S. Berner, M. Corso, R. Widmer, O. Groening, 126807 (2009).
30
R. Laskowski, P. Blaha, K. Schwarz, A. Goriachko, E. J. G. Santos, A. Ayuela, and D. Sánchez-Portal, New
H. Over, S. Gsell, M. Schreck, H. Sachdev, T. Greber, and J. Phys. 12, 053012 (2010).
31
J. Osterwalder, Angew. Chem. Int. Ed. 46, 5115 (2007). D. W. Boukhvalov and M. I. Katsnelson, Appl. Phys. Lett.
9
A. T. N’Diaye, S. Bleikamp, P. J. Feibelman, and 95, 023109 (2009).
32
T. Michely, Phys. Rev. Lett. 97, 215501 (2006). O. V. Yazyev and A. Pasquarello, Phys. Rev. Lett. 100,
10
J. Zhang, V. Sessi, C. H. Michaelis, I. Brihuega, J. Honolka, 156102 (2008).
33
K. Kern, R. Skomski, X. Chen, G. Rojas, and A. Enders, I. Gierz, C. Riedl, U. Starke, C. R. Ast, and K. Kern, Nano
Phys. Rev. B 78, 165430 (2008). Lett. 8, 4603 (2008).
11 34
A. T. N’Diaye, T. Gerber, C. Busse, J. Mysliveček, J. P. J. Feibelman, Phys. Rev. B 77, 165419 (2008).
35
Coraux, and T. Michely, New J. Phys. 11, 103045 (2009). P. J. Feibelman, Phys. Rev. B 80, 085412 (2009).
12 36
K. Donner and P. Jakob, J. Chem. Phys. 131, 164701 We assume spherically curved regions of diameter d=a/2.
(2009). The estimated curvature radius R≈a2 /(32∆h)≥2 nm is
13
J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. too large to produce any sizable effect on the local binding
Lett. 77, 3865 (1996). energies.41
14 37
D. Vanderbilt, Phys. Rev. B 41, 7892 (1990). G. Bertoni, L. Calmels, A. Altibelli, and V. Serin, Phys.
15
A. Pasquarello, K. Laasonen, R. Car, C. Lee, and D. Van- Rev. B 71, 075402 (2005).
38
derbilt, Phys. Rev. Lett. 69, 1982 (1992); K. Laasonen, A. V. M. Karpan, G. Giovannetti, P. A. Khomyakov, M. Ta-
Pasquarello, R. Car, C. Lee, and D. Vanderbilt, Phys. Rev. lanana, A. A. Starikov, M. Zwierzycki, J. van den Brink,
B 47, 10142 (1993). G. Brocks, and P. J. Kelly, Phys. Rev. Lett. 99, 176602
16
P. Giannozzi et al., J. Phys.: Condens. Matter 21, 395502 (2007).
39
(2009); http://www.quantum-espresso.org/. O. V. Yazyev and A. Pasquarello, Phys. Rev. B 80, 035408
17
G. Mills, H. Jónsson, and G. K. Schenter, Surf. Sci. 324, (2009).
40
305 (1995). L. Giordano, G. Pacchioni, J. Goniakowski, N. Nilius,
18
K. T. Chan, J. B. Neaton, and M. L. Cohen, Phys. Rev. B E. D. L. Rienks, and H.-J. Freund, Phys. Rev. Lett. 101,
77, 235430 (2008). 026102 (2008).
19 41
H. Sevinçli, M. Topsakal, E. Durgun, and S. Ciraci, Phys. F. J. Ribeiro, J. B. Neaton, S. G. Louie, and M. L. Cohen,
Rev. B 77, 195434 (2008). Phys. Rev. B 72, 075302 (2005).
20
G. Giovannetti, P. A. Khomyakov, G. Brocks, V. M.
Karpan, J. van den Brink, and P. J. Kelly, Phys. Rev.