Surface Science 460 (2000) 162–169

www.elsevier.nl/locate/susc

Structure and stability of the anatase TiO

2
(101) and (001) surfaces
R. Hengerer a, *, B. Bolliger b, M. Erbudak b, M. Grätzel a
a Laboratoire de photonique et interfaces, École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland
b Laboratorium für Festkörperphysik, Eidgenössische Technische Hochschule Zürich, CH-8093 Zurich, Switzerland

Received 7 January 2000; accepted for publication 13 April 2000

Abstract

Well-defined (101) and (001) surfaces of anatase TiO were studied for the first time by secondary-electron imaging
2
and low-energy electron diffraction. Initially, both surfaces show a crystalline structure corresponding to the bulk
anatase phase, buried under an atomically thin contamination layer. After mild sputtering with 500 eV Ne+ ions, we
have observed a surface phase transition from tetragonal anatase to face-centered cubic titanium monoxide TiO.
Subsequent annealing at 900 K restores the (1×1) anatase structure at the (101) surface. On the (001) surface,
however, a (1×4) reconstruction appears. The unreconstructed structure can be recovered by exposing the surface to
oxygen. These observations demonstrate the stability of the anatase surfaces and illustrate the feasibility of preparing
and investigating clean surfaces of this technological important form of TiO . © 2000 Elsevier Science B.V. All
2
rights reserved.

Keywords: Electron–solid interactions; Low energy electron diffraction (LEED); Secondary electron emission measurements;
Sputtering; Surface relaxation and reconstruction; Surface structure, morphology, roughness, and topography; Titanium oxide

1. Introduction dynamic phase stability of TiO , however, is actu-

The wide-bandgap semiconductor TiO mainly
2
exists in two polymorphic phases, anatase and
rutile, which crystallize both in a tetragonal lattice.
Their structure can be described in terms of chains
of TiO octahedra that are more distorted and
6
differently interconnected in anatase than in rutile
[1], resulting in significantly different unit cells and
distinct physical and chemical behavior [2,3]. So
far, only the surface properties of rutile have been
investigated [4,5], because it is considered to be
the more stable form of TiO [6 ]. The thermo-
2
* Corresponding author. Fax: +41-21-693-4111.
E-mail address: roland.hengerer@epfl.ch (R. Hengerer)
2
ally a function of particle size [7], and it is anatase
that plays a crucial role in a number of charge-
separating devices, like dye-sensitized solar cells
[8] or rocking-chair lithium batteries [9]. These
devices are based on highly porous films with very
large surface areas that consist of nano-crystalline
anatase. Although they show already a remarkable
performance, a fundamental understanding of sur-
face processes, like dye adsorption or charge
transfer, is still deficient. Due to the lack of suitably
large and pure single crystals, systematic investiga-
tions on specific surfaces are still very rare [10,11].
The different behavior of anatase surfaces com-
pared to that of rutile, e.g. in terms of molecular
adsorption [12], could hence only been deduced,

0039-6028/00/$ - see front matter © 2000 Elsevier Science B.V. All rights reserved.
PII: S0 0 39 - 6 0 28 ( 00 ) 0 05 2 7 -6
 R. Hengerer et al. / Surface Science 460 (2000) 162–169
making the assumption of a perfect surface ter-
mination of the bulk anatase lattice. In this paper,
we report the first successful imaging of the surface
lattice of TiO single crystals in the anatase struc-
2
ture by secondary-electron imaging (SEI ) and low-
energy electron diffraction (LEED)1. The measure-
ments were made on samples that were prepared
by a chemical transport reaction [10,11], which
proved to be an appropriate method by which to
obtain high-quality anatase single crystals. The
crystals exhibited mm2-sized (101) and (001) sur-
faces; these are also the main surfaces in nanocrys-
talline films [14]. On both faces, we have observed
reversible structural transitions, induced by sput-
tering and subsequent annealing. These cycles can
be repeated reproducibly, demonstrating the sta-
bility of the anatase structure, which could per-
fectly be recovered even after its complete
destruction at the surface. The sharp LEED pat-
terns presented here for both surfaces are, to our
knowledge, the first patterns reported for anatase
TiO . Earlier attempts to investigate naturally
2
grown anatase using LEED have failed due to
surface contaminations [15].
2. Experimental
Anatase single crystals were grown from the
gas phase by a chemical transport reaction as
described previously [10]. The transparent and
colorless crystals ordinarily have a truncated bipyr-
amidal habitus exhibiting mainly the (101) and
(001) faces, as confirmed by Laue X-ray diffrac-
tion. They have surface areas of 1–4 mm2, and
atomic-force microscopy revealed that the corruga-
tion of the working surfaces was typically below
1 nm. Prior to insertion into a vacuum, all crystals
were rinsed in water, ethanol, and acetone and
contacted with silver glue to a copper plate in
direct contact with an oven. Initially, we investi-
gated a crystal that was reductively doped in pure
hydrogen at 800 K for 24 h in order to prevent
electrostatic charging of the samples. However, we
1 During the revision of this paper, another publication
appeared in Ref. [13] on X-ray photoelectron diffraction of an
anatase (001) thin film.
163
found out that, due to the high electron mobility
in the conduction band, anatase samples appa-
rently do not need to be doped. We therefore used
undoped samples in our subsequent work in order
to preserve the high degree of stoichiometry of the
crystals. Heating up to 900 K in UHV did not
dope the anatase crystal either, contrary to rutile,
which generally seems to be more easily doped
than anatase2.
The experiments were performed in an ultra-
high-vacuum chamber with a total pressure in the
lower 10−10 Torr range. Chemical information
about the surface was obtained by Auger electron
spectroscopy (AES ). The electron energy was
2 keV for SEI as well as for AES. The same device,
a display-type, retarding-grid spherical collector
system, is used for LEED and SEI, the only
difference being the energy of the primary
electrons [16 ].
SEI records the backscattered (secondary)
electrons generated with primary electrons having
an energy in the keV range. SEI carries real-space
information about the local geometric arrangement
of atoms near a surface and is therefore comple-
mentary to the reciprocal-space information acces-
sible in LEED. Direct-projection imaging of the
interatomic directions, produced by the forward-
focusing effect in electron-atom scattering, is
responsible for the formation of the SEI pattern.
In the patterns presented here, the polar angle, h,
is defined by the direction of the primary-electron
incidence and the particular direction of the
secondary electrons. h can be determined within
an accuracy of ±1°, and its maximum value is
limited by approximately 52°, which corresponds
to the outer rim of the collector screen. Images
can be recorded while the sample is rotated, which
provides three-dimensional views of the structure.
For a structural transition, as reported here, SEI
allows the correspondence of the high-symmetry
axes of these structures to be identified, from which
the relative orientation of the initial and final
structures can be readily determined [17]. While
LEED is extremely sensitive to changes in the
surface atomic structure owing to the short mean
2 This is also evident by the fact that reductive doping takes
much longer for anatase than for rutile.
164 R. Hengerer et al. / Surface Science 460 (2000) 162–169
free path of the 30–400 eV electrons, SEI provides
information on the first 10–20 atomic layers of the
surface. Thus successful imaging is possible with
SEI even when the surface atomic layer is disor-
dered or contaminated [15].
3. Results
3.1. The (101) surface
210 and the
110 directions, respec-
Fig. 1a shows a secondary-electron pattern of
the (101) surface of an anatase crystal recorded
directly before any treatment of the surface. The
primary-electron beam is incident along the [100]
direction of the crystal, which coincides with the
center of the display unit. This is achieved by
Fig. 1. Secondary-electron patterns obtained from the (101)
anatase surface: (a) before and (b) after sputtering with 500 eV
Ne+ ions. The sample is tilted by 23° so that the primary
electrons are incident along the [100] direction. The central
portion of the pattern (polar angle h=0°) is covered by the
shadow of the electron gun used for the excitation.
101
tilting the (101) surface normal from the display
axis by 23°. The twofold symmetry of the (100)
surface of the tetragonal anatase is clearly detecta-
ble. The horizontal bright band corresponds to the
(010) lattice plane of the anatase structure. The
central vertical band contains directions in the
(001) plane that are above or below the (010)
plane at a distance of approximately ±27 and
±45° from [100]. They can be identified as belong-
ing to the
tively. The bright spot, 23° to the left of the
electron gun, corresponds to the (101) surface
normal of the crystal. Other salient features can
be assigned to different symmetry directions of the
anatase lattice. These results are in good agreement
with earlier investigations [15] and imply that the
positions of the atoms in a near-surface region do
not deviate from the bulk values, i.e. the crystalline
structure is flawless and not reconstructed within
the inelastic mean free path of 2 keV electrons.
Note that this untreated surface is still covered by
a thin layer of contamination and/or disordered
atoms so that no LEED observation could be
made.
Attempts to produce clean and atomically
ordered surfaces on naturally grown anatase by
only sputtering (500 eV He+ ions for 15 min) failed
in a previous experiment due to the destruction of
the surface atomic order [15]. The SEI patterns
smeared out and could not be restored even after
a heat treatment at 500 K and exposure of the
surface to oxygen up to 10−3 Torr · s. A similar
problem was encountered with the present vapor-
phase-grown anatase already after sputtering for
only 30 s with 500 eV Ne+ ions. The SEI pattern
became diffuse, but still showed twofold symmetry
around the (010) plane. However, after continuing
the sputtering for 5 min, this symmetry changed
to fourfold (cf. Fig. 1b). Viewed along the [100]
direction, one recognizes a highly symmetrical
cross with the horizontal band still being the (010)
plane and the vertical band the (001) plane. Two
other bands corresponding to the (011) and (011: )
planes can now be identified at 45° to these bands.
Tilting the crystal around the [010] direction, one
detects bright patches at angles of 45 and 90°
relative to the [100] direction, signalling the
 R. Hengerer et al. / Surface Science 460 (2000) 162–169
Fig. 2. Secondary-electron patterns of the ion-bombarded (101)
surface with electrons incidence normal to the (101) surface of
anatase. Image (a) was taken before annealing, (b)–(e) during
annealing at 900 K at subsequent 2 min intervals, and (f ) after
annealing and cooling down of the sample.
111.
and
001 directions, respectively. All these fea-
tures are characteristic of a cubic structure.
Since the structural order in the top surface
layers of the crystal was obviously disturbed by
sputtering, broad bands rather than distinct sharp
patches appeared on the SEI pattern (cf. Fig. 1b).
To restore structural perfection, the crystal was
heated up to 900 K. The influence of this subse-
quent annealing of the ion-bombarded surface is
shown by the sequence of secondary-electron pat-
terns in Fig. 2, which illustrates the kinetic evolu-
tion of the structural phase transition at the
surface. The crystal is tilted by approximately 23°
with respect to its position in Fig. 1, so that the
165
Fig. 3. LEED pattern of the sputtered and annealed (101) sur-
face of anatase, recorded at 102 eV.
primary-electron beam is now incident perpendicu-
lar to the sample surface. The pattern in Fig. 2a
was taken directly after sputtering. Each successive
pattern was obtained after annealing for additional
2 min increments. The differences between succes-
sive patterns are quite small, but one can still
recognize the change in fourfold symmetry around
the [100] axis ( located at the right-hand side of
the display center) to a more complex pattern
possessing a twofold symmetry of the original
anatase surface (cf. Fig. 1a). More clearly detecta-
ble is the transition by observing the disappearance
of the symmetric [101] pattern of the cubic crystal,
approximately 22° to the left of the center in
Fig. 2a. Similarly, two bright patches above and
below the [101] direction on the left-hand side of
the image are identified, both belonging to
In Fig. 2a, there is a threefold symmetry in the
distribution of electron intensity around these
directions that gradually fades away as annealing
proceeds. The surface of the crystal eventually
coincides again with a (101) face of anatase.
Along with the structural transition, also, the
definition of contrast in Fig. 2 improves continu-
ously with annealing, indicating the gradual heal-
ing of structural defects. In fact, after the complete
sputtering–annealing cycle, the entire surface,
including the top atomic layer, was so perfect and
clean that very sharp LEED patterns could be
observed from the anatase surface, as shown in
Fig. 3. Besides the definition of the spots, their
symmetry shows that the surface structure seems
166 R. Hengerer et al. / Surface Science 460 (2000) 162–169
to be bulk-terminated, i.e. the surface is obviously
not reconstructed3. The surface lattice constants,
calculated from the LEED pattern, correspond to
the bulk values within an error limit of ≤1%. The
(101) surface is thus undistorted, similar to (110),
the most stable face of rutile [18].
201 directions. Other
3.2. (001) surface
The next common surface of anatase is the
(001) face, for which a secondary-electron pattern
is presented in Fig. 4a, recorded before any surface
treatment. The primary-electron beam is incident
along the [001] direction of the crystal. The quasi-
fourfold symmetry in the atomic arrangement at
Fig. 4. Secondary-electron patterns obtained from the (001)
anatase surface: (a) before and (b) after sputtering with 500 eV
Ne+ ions. The primary-electron beam is incident along the [001]
direction of the crystal.
3 It should be mentioned, however, that a (1×1) symmetry
found by LEED cannot exclude a surface unit cell with a sym-
metric defect.
this surface is clearly detectable. The horizontal
bright band corresponds to the (010) lattice plane
and the vertical band to the (100) plane of the
anatase structure. The bright spots, approximately
38.5° to the left and right of the electron gun (as
well as above and below it), all marked with a
cross, correspond to the
symmetry directions of the anatase lattice produce
the remaining salient features observed in the
pattern. These results show again that the atoms
near the surface have positions given by their bulk
values. It is remarkable that, similar to the (101)
surface, the crystal structure of the untreated (001)
surface has an excellent quality, observed using
SEI.
An electron-diffraction pattern obtained from
the (001) surface at an energy of 191 eV is pre-
sented in Fig. 5a. This pattern is observed after
heating the sample to 900 K. Apparently, sputter-
Fig. 5. LEED pattern of (001) surface of anatase recorded after
annealing the surface at 900 K (a) without and (b) with
sputtering.
 R. Hengerer et al. / Surface Science 460 (2000) 162–169
ing is not a prerequisite for cleaning the anatase
samples: just heating to an elevated temperature
causes desorption of the contamination, which has
prevented successful imaging of the surface using
LEED. In the pattern, fourfold symmetry is found
without any indication of a superstructure. The
wide spot of the primary-electron beam partly
illuminates the (101) surface of the tiny crystal
face, causing the appearance of the few additional
spots. The surface lattice constants, calculated
from the LEED pattern, equal the bulk values
within an error limit of ≤1%.
Subsequently, the surface has been sputtered
with Ne+ ions at 500 eV for 5 min. We observe
that sputtering of the surface only blurred the
pattern, while the overall symmetry remained
unchanged, as can be seen in Fig. 4b. However,
the bright spots at 38.5° disappeared, and new
spots showed up further outside at 45°. Essentially,
the pattern of Fig. 4b is the same as that shown
in Fig. 1b. This is an additional indication that the
surface structure has changed to a face-centered
cubic (fcc) phase for which the (100) and the (001)
surfaces are indeed indistinguishable. After sput-
tering, the crystal was briefly heated once more to
900 K. Sharp LEED spots reappeared, but a new
pattern came into view (cf. Fig. 5b). Here, the
primary-electron energy was 122 eV. The symme-
try and the lattice constants did not change, but
three additional minor spots could be detected
between the original main spots in the [100] as
well as in the [010] direction. These indicate a
(1×4) superstructure. We have to consider that
there are two equivalent orientations for the (1×4)
reconstructed surface unit cell on the fourfold
symmetrical (001) surface. Hence, the electron
beam can illuminate different domains aligned
perpendicular to each other, thus leading to the
appearance of the fourfold superstructure on the
surface. Successive heating for longer times did
not restore the original surface structure. No
attempt was made to anneal at higher temperatures
because the phase transition from anatase to rutile
would occur above approximately 900 K [6 ]. Only
when the sample is exposed to oxygen or after
introducing air into the experimental chamber is
the (1×1) surface order recovered.
167
4. Discussion
We have observed that the sputtering of anatase
surfaces results in a transformation into a cubic
surface phase. This is evident at the (101) as well
as the (001) face. The only cubic phase in the
binary titanium–oxygen system is titanium(II )
monoxide, which has a defective rocksalt structure
with up to 15% random Ti and O vacancies [19].
Cleaning of transition-metal oxide surfaces by ion
bombardment is known to lead to the preferential
removal of oxygen and, therefore, to a modifica-
tion of the surface composition [20]. Apparently,
the preferential sputtering is in our case so effective
that it leads to a phase transformation into a fcc
structure with a fairly different stochiometry. AES
confirmed this conclusion by showing a significant
decrease of the oxygen signal. Still, the new struc-
ture is remarkably well ordered and epitaxial with
the anatase. Considering the information depth
accessible in SEI [16 ], we estimate the thickness
of the cubic layer to be at least 20 Å. For rutile,
preferential sputtering is less pronounced. In con-
trast to our findings, even prolonged sputtering of
this polymorph with 500 eV Ar+ results only in a
transformation to titanium(III ) oxide, Ti O [20].
2 3
A photoemission investigation revealed that
intense sputtering of rutile partially changes the
chemical state of the titanium atoms from Ti(IV )
to Ti(III ) or to Ti(II ), but even after bombard-
ment with 10 keV Ar+, the fraction of Ti(II ) is
only about one-third [21]. This observation is in
agreement with the stability of the rutile phase.
A comparison of the anatase and the rocksalt
structure of titanium monoxide is shown in Fig. 6.
The structure of anatase is conceivable as a closed
packing of large oxygen atoms with titanium atoms
in every second octahedral interstitial site in a
zigzag alignment. Every titanium atom is thus
surrounded by an oxygen octahedron. The atoms
have a pronounced ionic character, resulting in a
strong crystal field that distorts the octahedra [1].
The resulting structure of the unit cell is shown in
Fig. 6a. Removal of every second oxygen atom or,
equivalently, filling every interstitial site with a
titanium atom leads directly to the structure shown
in Fig. 6b. Again, each titanium atom is sur-
rounded by an oxygen octahedron, but in this
168 R. Hengerer et al. / Surface Science 460 (2000) 162–169
Fig. 6. (a) Unit cell of anatase TiO with octahedral interstitial
2 in this direction [11]. The (1×4) superstructure
sites indicated in black and (b) twice the unit cell of titanium
monoxide TiO. Titanium: small spheres; oxygen: large spheres.
case, because of the highly symmetric distribution
of titanium atoms, no distortion takes place. This
leads to an fcc lattice. There is a change in the
interatomic distances during this transition. The
value for the a axis is 3.78 Å for anatase and
4.17 Å for titanium monoxide, the latter value
decreasing with increasing oxygen content, i.e. with
the stoichiometry closer to anatase [22]. For the c
axis, which corresponds to twice the a axis in the
cubic lattice, the values are 9.51 and 2×4.17=
8.34 Å, respectively. Presumably, this close sim-
ilarity between the two structures facilitates the
smooth phase transition, as observed in Fig. 2,
where, in addition, the bulk anatase crystal can
act as a template.
Annealing proved to be a successful process by
which to restore the original (101) surface. Upon
completing the structural recovery, AES shows
that the chemical composition at the surface
returns to its bulk value. Since no gas-phase oxi-
dant was introduced to the UHV chamber, this
surface re-oxidation has to originate from point-
defect diffusion into the bulk. This may function
either by oxygen vacancy diffusion or by titanium
interstitial diffusion. Recent experiments with iso-
topically enriched rutile indicate that the
re-oxidation of the surface layers proceeds via
diffusion of Ti cations from the surface to the bulk
at elevated temperatures [23]. Anatase, too, is
known to be structurally transparent for cations
[10], so we adapt this mechanism of bulk-assisted
re-oxidation as an explanation for the restoration
of the original structure.
In contrast to the (101) face, the destruction of
the (001) surface lattice cannot be healed com-
pletely by merely heating up the sample. Even
after prolonged annealing, the reconstructed
(1×4) pattern remained, so a less efficient diffusion
of defects along the c axis cannot account for the
incomplete recovery. Furthermore, there is no
structural indication that diffusion should be
slower along the c direction, especially since elec-
trochemical experiments showed quite the oppo-
site, i.e. a higher structural transparency for ions
hence seems to be a metastable phase of the (001)
surface that needs a thermodynamical rationaliza-
tion. A recent theoretical calculation [24] indicates
a somewhat lower energy for the (001) surface
(1.28 vs. 1.40 J m−2). This should in fact make the
(001) face the most stable surface of anatase.
Facetting of this face is therefore rather unlikely
and cannot provide an explanation for the
observed superstructure. However, in the same
study, a lower attachment energy for the (001)
than for the (101) surface was calculated, so the
outermost atoms should be less strongly bonded
on this face. In addition, there is experimental
evidence that externally introduced oxygen can
recover the original unreconstructed surface lattice.
This implies that an oxygen deficiency causes the
(1×4) reconstruction. Apparently bulk-assisted
reoxidation for the (001) surface works up to the
point of restoring the old anatase structure, but
cannot completely remove all the oxygen vacancies
on the surface. A simple missing oxygen-row model
for the (001) surface could elucidate the origin of
the observed (1×4) pattern. However, further
careful structural investigations (e.g. with AFM or
STM ) are necessary to either confirm or modify
this model, especially in view of the history of the
long-debated various reconstructed rutile surfaces.
It should also be mentioned that the above-
cited theoretical investigations of the anatase sur-
 R. Hengerer et al. / Surface Science 460 (2000) 162–169
face [24] predict a small relaxation of the (101)
and the (001) surfaces. These shifts in the atomic
positions are, however, only in the order of pm
and therefore beyond the resolution limit of the
LEED measurements. Strictly speaking, our state-
ment of perfectly bulk-terminated surfaces does
not consider these minor, but nevertheless pos-
sible, effects.
5. Conclusion
This work aimed to clarify the surface geometry
of the anatase (101) and (001) surfaces for a better
understanding of adsorption and charge-transfer
processes on anatase TiO . Therefore, both sur-
2 [8] B. O’Regan, M. Grätzel, Nature 353 (1991) 737.
faces were studied by SEI and LEED. We have
observed a structural transition to titanium mon-
oxide on the surfaces, induced by preferential
sputtering. Upon annealing, the bulk acts as a
thermodynamic and structural template, and the
stoichiometry and original anatase structure at
the (101) face is restored. The (001) surface, on
the other hand, shows a (1×4) anatase superstruc-
ture after sputtering and annealing, which is resolv-
able by missing oxygen rows. The original (001)
structure, which has a (1×1) symmetry as the
(101) face, can be restored by exposure to oxygen.
By demonstrating the possibility of obtaining clean
and structurally perfect anatase surfaces, this work
opens the way for more surface-specific investiga-
tions on anatase, which has been studied consider-
ably less than the rutile phase of TiO .
2 36 (1976) 1335.
Acknowledgements
This work was financially supported in part by
the Swiss National Science Foundation. Thanks
169
are due to H. Berger for technical help during the
crystal growth.
References
[1] A. Fahmi, C. Minot, B. Silvi, M. Causa, Phys. Rev. B 47
(1993) 11717.
[2] M. Grätzel, F.P. Rotzinger, Chem. Phys. Lett. 118
(1985) 474.
[3] J. Augustynki, J. Electrochim. Acta 38 (1993) 43.
[4] V.E. Henrich, P.A. Cox, The Surface Science of Metal
Oxides, Cambridge University Press, Cambridge, 1994.
[5] D.A. Bonnell, Prog. Surf. Sci. 57 (1998) 187 and refer-
ences therein.
[6 ] H.O. Finklea, Semiconductor Electrodes, Elsevier,
Amsterdam, 1988.
[7] H. Zhang, J.F. Banfield, J. Mater. Chem. 8 (1998) 2073.
[9] S.Y. Huang, L. Kavan, I. Exnar, M. Grätzel,
J. Electrochem. Soc. 142 (1995) L142.
[10] L. Kavan, M. Grätzel, S.E. Gilbert, C. Klemenz,
H.J. Scheel, J. Am. Chem. Soc. 118 (1996) 6716.
[11] R. Hengerer, L. Kavan, P. Krtil, M. Grätzel,
J. Electrochem. Soc. 147 (2000) 4.
[12] A. Vittadini, A. Selloni, F. Rotzinger, M. Grätzel, Phys.
Rev. Lett. 81 (1998) 2954.
[13] S. Herman et al., Surf.Sci. 441 (2000) 201.
[14] S. Burnside et al., Chem. Mater. 10 (1998) 2419.
[15] V. Shklover et al., J. Solid State Chem. 132 (1997) 60.
[16 ] M. Erbudak, M. Hochstrasser, E. Wetli, M. Zurkirch, Surf.
Rev. Lett. 4 (1997) 179.
[17] M. Erbudak, E. Wetli, M. Hochstrasser, D. Pescia, D.D.
Vvedensky, Phys. Rev. Lett. 79 (1997) 1893.
[18] Y.W. Chung, W.J. Lo, G.A. Somorjai, Surf. Sci. 64
(1977) 588.
[19] A.F. Wells, Structural Inorganic Chemistry, Clarendon
Press, Oxford, 1975.
[20] V. Henrich, G. Dresselhaus, H.J. Zeiger, Phys. Rev. Lett.
[21] D. Leinen, A. Fernandez, J.P. Espinos, A.R. Gonzales-
Elipe, Appl. Phys. A 63 (1996) 237.
[22] J. Whitehead, in: Encyclopedia of Chemical Technology
Vol. 23, Wiley, New York, 1983, p. 137.
[23] M.A. Henderson, Surf. Sci. 419 (1999) 174.
[24] P. Oliver, G. Watson, E. Kelsey, S. Parker, J. Mater. Chem.
73 (1997) 56.