PHYSICAL REVIE%
8
VOLUME 37, NUMBER 6 15FEBRUARY 1988-IISurface
metallization
of
silicon by
potassium
adsorption
on
Si(QQ1)-(2X 1)
S.
CiraciDepartment
of
Physics, Bilkeni University, Ankara, TurkeyInder
P.
BatraIBMAlmaden Research Center, Mail Stop@33-801, 650Harry Road, San Jose, California 95120-6099 (Received 20August 1987)
%e
present the detailed results ofself-consistent and geometry-optimized total-energy, band-structure, and charge-density calculations for a potassium-covered Si(001)-(2X1)surface, and foran unsupported potassium monolayer.
%e
found that the (2X1) reconstruction and the dimer bonds ofthe Si surface continue to be stable after the adsorption ofalkali-metal atoms. At the monolayer coverage the charge from the adsorbed potassium atoms is transferred into the empty, antibonding dangling-bond surface states, resulting in the metallization ofthe Si(001)substrate sur-face. The bonding between the overlayer and the substrate surface is ionic, and the Fermi level ispinned by the partially 611edsilicon surface states. Our theory for the metallization and the sur-face collective excitations is diferent from previous ones developed for an alkali-metal overlayer on a metal substrate which suggest that the system undergoes aMott transition, and can
success-fully account for recent experimental observations. The presence of the active dangling-bond
states prevents the alkali-metal monolayer from meiallization, and thus provides the crucial dil'erence between metal and semiconductor substrates.
I.
INTRODUCTIONGeneration
of
desired electronic properties by metal deposition on the semiconductor surfaces has been ex-ploited' in a numberof
technical applications in the areaof
microelectronics and electronic devices. Becauseof
their rectifying properties, metal-semiconductor junc-tions, and particularly the mechanismof
the Fermi-level pinning and the formationof
the Schottky barrier, have been extensively investigated. ' Metal-induced gap states,'
which propagate in the thick metal film, but be-come evanescent in the semiconductor, have been pro-posed as the states responsible for the pinningof
the Fermi level. Recent experiments, however, have cometo
difFerent conclusions, revealing the fact that the Fermi level can be pinned and thus the Schottky barrier is formed even at submonolayer coverage."
Different kindsof
states, such as intrinsic surface states, defect states, and chemisorption bond states, ' have beenpostulated for the pinning mechanism at submonolayer coverage. Batra and Ciraci' suggested coverage-dependent effects after studying a prototype, lattice-matched metal-semiconductor interface'
[i.e.
, Al on a Ge(001) surface] from submonolayer to multilayer cover-ages. At submonolayer coverage they found that the chemisorption bond states and Al-atom states dominate the energy spectrum near the band-gap region, and are responsible for the pinningof
the Fermi level. Above the monolayer coverage they showed, however, that ow-ing to the interaction among deposited Al atoms the overlayer changes intoa
(quasi-) two-dimensional (2D) metal characterized by a (modulated) ladder-type densityof
states.''
Concomitantly, the metallic overlayer re-laxes outwards away from the semiconductor surface,and the metal-semiconductor bonds are delocalized.
For
those semiconductors which have active surface states the metallizationof
the overlayer must compete with the formationof
metal-semiconductor bonds.' In fact, depending on the relative valueof
the metallic cohesive energy and the metal-substrate interaction ener-gy the overlayer metallization may be suppressed alto-gether. In this respect, the adsorptionof
alkali met-alatoms on metal and semiconductor surfaces is rather special, and presents very interesting conceptual ideas about bonding, metallization, and collective excitations.
Alkali-metal atoms are usually characterized by a sin-gle electron
of
large atomic radius. In the metallic state, the structure is open, and the charge density is feature-less and low as comparedto
other divalent and trivalent close-packed metals. ' In band-structure language the conduction band can be described by a nearly-free-electron picture. From the surface-science pointof
view alkali-metal adsorption is important becauseof
its work-function lowering effect,' which provides several technological applications.Interesting coverage-dependent features
of
alkali-metal adsorption on metal surfaces have appeared from the workof
Mac Rae et a/.' At the initial stageof
cesium adsorption on the W(001) surface, the work function4
decreases rapidly, and Cs atoms form a(2X2)
structure with a Cs-Cs nearest-neighbor distance significantly smaller than that in the Csmetal. Further Cs deposition (above the monolayer coverage) yields a close-packed hexagonal Cs layer with a nearest-neighbor distance comparable to thatof
the bulk metal. At this coverage the work function passes througha
minimum, and a loss peak grows in intensity. Subsequent studies have shown that these coverage-dependent features are common toS.CIRACI AND INDER P.BATRA
other alkali-metal-covered metal surfaces. '
Recent work by Tochihara
'
has attracted much at-tention to the adsorptionof
alkali-metal atoms on semi-conductor surfaces. His experimental observations on the potassium-covered Si(001) surface suggested that alkali-metal atoms on semiconductor surfaces lead to practically the same features as observed on transition metals. The maximum coverage obtained is one mono-layer[8=1,
i.
e., oneK
atom per(2X1)
surface unit cell] showing a clear(2
X1)
low-energy electron-difFraction(LEED)
pattern but with a diFerent intensity as compared tothe clean surface. The work function de-creases rapidly at submonolayer coverage(8«1),
and later, concomitant with the onsetof
an electron-energy loss peak at 2 eV, it decreases slowly without passing through a minimum. As for alkali-metal adsorption on metal surfaces, '8 these observations were interpretedto
imply a metal-insulator (Mott) transition.It
was pro-posed that up to8=0.
5,K
atoins donate their 4s valence electronsto
the substrate and become ionic as evidenced by a higher core excitation energy and lower work function. Continued depositionof
alkali-metal atoms causes the interatomic distanceof
the potassium atoms to decrease. Beyond a threshold adatom density the smallerK-K
distance allows the formationof
metal-lic bands which, in turn, results in a retransferof
charge to the alkali-metal overlayer. Accordingly, the growing intensityof
the peak in the electron-energy-loss spectra (EELS)with increasing8
was attributed to a collective excitationof
the metallic electronsof
the overlayer. Data for the Cs-covered Si(001)surface were interpret-ed in asimilar fashion.Aruga, Tochiara, and Murata carried out detailed analysis
of
the plasmon loss by the useof
angle-resolvedEELS
and obtained an anisotropic and positive disper-sion in contrastto
the plasmon dispersionof
an slkali-metal overlayer on a metal surface. In viewof
the inter-band and intraband (zero-sound) plasmon modes calcu-lated ' fora
simple model, these q-dependent losspeaks were attributed
to
the10
plssmons associated with theK
chains on the surface.The experimental data on the alkali-metal-covered
Si(111)
surface are unfortunately not as abundant as for the Si(001)-(2X1)
surface. However, the ultraviolet pho-toemission spectrum (UPS}taken from the Cs- coveredSi(111)-(2X1)
surface' has been interpretedto
imply the metallizationof
the alkali-metal overlayer beyond the adatom coverageof
6=1.
In this paper we present a detailed study
of
the K-coveredSi(001)-(2X1)
surface at8=1,
which is based on 6rst-principles total-energy, electroni-sstructur, and force calculations. Someof
our results were briefly re-ported earlier. The objective here isto
elucidate the interactionof
an alkali metal with the reconstructed Si surfaces, and to place emphasis on the metallization by contrasting it with that encountered in other metal over-layers. Our theory for the metallizationof
the alkali-metal overlayer and the characterof
collective excita-tions is at variance with previous suggestions, snd can successfully account for the experimental Sndings.It
is also quite diferent from the metallizationof
divalentand trivalent metal overlsyers adsorbed on semiconduc-tor surfaces. Our results indicate that the metallic char-acter attributed to the alkali-metal monolayer is, in fact, the metsllization
of
the reconstructed Si surface. More-over, our work introduces a metal-semiconductor inter-face, wherein the pinningof
the Fermi level at6=1
is almost totally determined by the intrinsic surface states.The organization
of
this paper is as follows. In the next section the model and the detailsof
the self-consistent-field (SCF}calculations are described. InSec.
III
the optimizationof
the atomic geometry for theK+Si(001)
system, yielding an equilibrium structure with a large ionic binding energy, is explained. Section IV is devotedto
discussionof
the calculated electronic structure, such as the energy band structureof
the un-supportedK
monolayer, the clean and K-covered Si(001) surfaces, and the work function. InSec.
V the analysisof
theSCF
charge density is presented. InSec.
VI our results are compared with the experimental and theoreti-cal studies, and our interpretation is presented. In addi-tion, a brief discussionof
the surface collective excita-tions relevant to the system at hand is given. Finally, our conclusions are stated inSec.
VII.
II.
METHOD OFCALCULATIONSWe performed SCF-pseudopotential calculations within the framework
of
the local-density functional theory applied in momentum space. We used nonlocal, norm-conserving ionic pseudopotentials given by Bache-let er a/. These ionic pseudopotentials were generated by using Ceperley-Alder exchange-correlation poten-tial. 'For
the sakeof
compatibility, we carried out cal-culations by using the same exchange-correlation poten-tial with a parametrized formof
Perdew and Zunger. i The calculations were done for an unsupportedK
mono-layer, and for a clean, and a K-covered Si(001)surface using a repeating slab geometry. The vacuum spacing between slabs wss takento
be 14s.
u. The silicon sub-strate with the (001)-(2X1)
reconstructed surface is simulated by a slab consistingof
eight atomic layers (i.e.
, 16Si atoms in the slab unit cell). The top surface is ei-ther clean or covered by alkali-metal atoms, but the bot-tom surface isalways saturated by hydrogen atoms.For
the positions
of
the Si atoms we used the optimized reconstruction geometry with a symmetric dimer bond model proposed by Abraham sndastra.
That the symmetric dimer bond is favorable relative to the asym-metric dimer bond was first suggested by Pandey, and subsequently was observed by scanning-tunneling rni-croscopy.Electronic states were represented by
-550
plane waves. During the self-consistency iterations the charge density was sampled at 15k
points placed uniformly in the surface Brillouin zone (BZ). This parameter set, though sufFiciently large, may not provide absolutely converged results. However, we believe that the degreeof
convergence to be achieved should be compatible with the accuracy required in the calculationof
a particular property.For
example, a small energy difFerence be-tween the symmetric snd asymmetric dimer bondrecon-37 SURFACE METALLIZATION OFSILICON BY POTASSIUM. .
.
struction certainly requires a much larger basis set, whereas the gain
of
energy from thediner
bond forma-tion in a(2X
1) reconstruction israther large (—
1.
5eV), and thus the overall propertiesof
theSi(001}-(2X1)
sur-face can be dealt with by a relatively small basis set. As will be seen in the following section the binding ener-gies resulting in these calculations are sufBciently large, and thus can be handled with reasonable accuracy by us-ing a basis set consistingof
about 550plane waves and a charge density sampled at 15k
points. Since the surface bands and the s bandof
the unsupportedK
monolayer are smooth, and are also devoidof
band crossings near the Fermi level, the coarseBZ
sampling may leadto
an errorof
-+0.
03
eV in the calculated total energies. This error is further reduced by using a thermally broadened Fermi-Dirac distribution function. As an in-dependent test for the convergence the widthof
the valence bandof
an eight-layer Si slab was closeto
the value for bulk Si calculated with a basis setof
250plane waves, which corresponds to a kinetic energy cuto8;~k+6~
=12
Ry. We have repeated calculationsof
structure optimization with a larger kinetic energy cuto8 and with a thinner
[i.e.
, five-layerSi(001)-(2X1)]
slab, and observed that the values, in particular the equilibri-um configuration and the binding energy, are not seri-ously affected. Moreover, for critical geometries (i.e.
, the equilibrium structure proposed here and the struc-ture in which the Si-Kinternuclear distance d~s;K~ isob-tained from the sum
of
the atomic radiiof
Si andK}
we used a large basis set consistingof
-1000
plane wavesto
calculate total energies.
I l I
a!a'
((
~
FIG, 1. Side and top views ofthe equilibrium structure of the potassium-covered Si(001)-(2X1)surface, Shaded large and empty small circles denote K and Si atoms, respectively. Numerals in the circles indicate the atomic layers. a
=10.
26 a.u. ,h (orz)=2.
4a.u.DI.
EQUILIBRIUM STRUCriJRE AND MNDING ENERGVTo
Snd the equilibrium structure and the positionof
the adsorbed potassium layer we carried out an extensive geometry optimization within the(2X1)
unit cell. In viewof
theLEED
data exhibiting a clear(2X1}
pat-tern"
'(1X1)
reconstruction was not considered. Bypositioning a potassium atom above ihe dimer bond, along the dangling bond, and above the hollow site with varying heights (h) from the surface, we investigated several adsorption sites.
For
each atomic configuration (corresponding to the diferent location and heightof
theK
atom} the total energies were calculated with a self-consistency toleranceof
10 Ry (root-mean-square de-viation), assuring a reasonable levelof
self-consistency. Among these atomic configurations, our calculations yielded the lowest total energy for theK
atom located at the centerof
the sixfold hollow site, between two parallel dimer bonds, and2.4
a.
u. above the surface. This leadsto
an internuclear distanceof
the substrate silicon and the adsorbed potassium,Its;
K]—
—
4.9 a.
u. The equilibri-um structure as illustrated inFig.
1 indicates that theadsorbed
K
atoms form achain alon the[110]
direction with an interchain distanceof
2a 2 (a being the lattice parameterof
Si). Becauseof
this atomic configuration theK
overlayer (andK
monolayer) is denoted as theK
chain in the text. Along the chain each
K
atom has two nearest neighbors with an internuclear distanceof 7.
26a.
u. Based on theLEED
data the formationof
chain structure from theK
overlayer was previously pro-posed.We found that
K
located above Si asif
it saturates the dangling bonds is energetically unfavorable relative to the optimized equilibrium structure. This implies thatK
behaves differently as compared to other atoms (such as H}, which form strong (covalent) chemisorption bonds with the surface Siatoms.
The stability
of
the clean, reconstructed surface upon adsorption is usually a critical point to be taken into ac-count. In the present system the dimer bond is a strong bond (almost as strong as the bulk Si—
Si bond). As far as the adsorption siteof
K
is concerned, one does not expect any bond breaking,or
any other signi6cant rear-rangementof
surface atoms uponK
adsorption. Howev-er, since the atomic structureof
the substrate was 6xed during the geometry optimizationof
the adsorption, the stabilityof
theSi(001)-(2X1)
surface has to be investi-gated to assure that this is indeed the case. We exam-ined the stabilityof
the (2 X 1)reconstruction geometry by calculating the interatomic forces in the presenceof
the
K
overlayer.%e
found the forces acting on the Si atoms were small. This suggests that the displacementof
Si atoms is insignificant, and the substrate surface is stable after the adsorptionof
potassium at8=1.
Having determined the equilibrium structure we next calculate the binding energy. This isdone in three steps.
S.CIRACI AND INDPR
P.
BATRA VETALLlC K CHAlN CD CA CD UJ CA n C CO ulk K(CA) Bulk Ktw) K+Si(001)-(2x 1)%'e first calculate the total energy
of
the unsupportedK
monolayer. Secondly, we calculate the total energy
of
the clean Sisubstrate.For
the sakeof
aconsistent com-parisonof
the total energies the parametersof
calcula-tions inSec.
II
are kept fixed. The binding energyof
theK
monolayerto
the Sisurface is foundto
be2.4
eV perK
atom. Byadding the cohesive energyof
theK
mono-layer we obtain the binding energyof a
potassium atomto
the Si(001)-(2X1)
surface to be-3
eV.To
provide an analysisof
the stability, we calculate also the total en-ergyof
the bulkK
by varying the cubic latticeparame-ter.
The energeticsof
theK+Si(001)-(2X
1)system and bulk potassium with respectto
the unsupported, metallicK
chain in registry with theK
adsorbed on Si(001)-(2 X 1)surface are presented in Fig.2.
Recently, Kendelewicz et
al.
based on surface-extended x-ray-absorption fine-structure(SEXAFS)
re-sults, estimated that the internuclear distanceof
the sur-face silicon and the absorbed potassium, d~s;~& is5.9
a.
u.(3. 14%0.
1A).
This particular structure was con-sidered in our earlier geometry optimization and was found to be energetically unfavorable. We have since re-peated the calculations for these two structures (i.e.
, d~s;Ki—
—
4.9
and5.9
a.
u.) with a much larger basis set(-1000
plane waves). Our pseudopotential calculations find the structure derived fromSEXAFS
data has-44
mRy higher energy (less stable). This is even 6 mRy higher than that calculated by using-550
plane waves. The significant point to note, however, is that the SEXAFS-derived structure with d~s; ~i—
—
5.9 a.
u. still hasa high binding energy
(-2.
5 eV).It
is not too surpris-ing since the local-density approximation underestimates the structural parameters related toK.
For
example, calculations'
with a Hedin-t. undquist exchange-correlation potential provide excellent prediction for the binding energyof
the bulk potassium, but underesti-mate the cubic lattice constant by-0.
6 a.
u. Calculated structural parameters are strongly dependent on the treatmentof
the exchange-correlation potential. The values obtained for the cubic lattice constant by using the Wigner and Ceperly-Alder ' exchange-correlation potentials are9.
2 and8.
9
a.
u., respectively. Owing to the low charge density between Si andK
atoms (and also perhaps due to the ionic pseudopotential used here) our calculations may underestimate the valueof
d~s;xi.
Nevertheless, as will be shown in
Sec.
VI, the natureof
the interaction between
K
and the Si surface is unaltered even for d~siK~ derived fromSEXAFS.
Also, the bind-ing energies and the characterof
the interaction betweenK
and the Si surface are in agreement with the indepen-dent Hartree-Fock calculations.Another possibility one considers is that
SEXAFS
measurements may correspond
to
a different adsorption site. In viewof
the fact thatK
adsorbs in the low-@barge-density regions, we have examined also the second hollow site,i.e.
,K
positioned above the third-layer Si, between two dimer bonds inFig. 1.
This configuration leads to a valueof
d~s;iti which is close to that estimated fromSEXAPS,
and may also be the lo-cation whereK
atoms are adsorbed at6~1.
This ad-sorption site is, however, less stable accordingto
our cal-culations. In TableI
the total energiesof
these struc-tures are compared.High binding energy obviously indicates that a strong interaction takes place between the
K
overlayer and the Sisurface. The calculated binding energy is three times larger than the binding energyof
the bulk metal, and also five times larger than thatof
the metallicK
chain.It
is also seen that the unsupported, metallicK
chain is unstable relative to the bulkK.
However, owing to this strong interaction both the metallicK
chain (or mono-layer) and the bulkK
become unstable relative to theK
adsorbed on the Si surface at
8=1.
We now examine the natureof
this interaction, and determine the efFectof
the Si surface, which makes the metallic state
of
theK
chain unstable.
I
6
z,d(a.u.)
FIG.
2. Calculated binding energies of K+Si(001)-|,'2~1)
and the bulk Krelative to the unsupported, metallic K chainin rey'stry with the K-adsorbed Si(001)-(2&1).
%
and CA denote signer and Ceperley-Alder exchange-correlation poten-tials used in the calculations, z the height ofthe adsorbed K above the surface, and d the nearest-neighbor distance in the bcclattice ofthe bulk potassium.TABLE
I.
Comparison oftotal energies calculated for three adsorption structures. ET',the total energy ofthe equilibriumstructure me found with d&&;K)
—
—
4.9a.u. and h=2.
4a.u. ;ET',same adsorption site but h
=4.
2 a.u. and d(s;K)—
—
5.9 a.u. asproposed in Ref. 40; ET',the hole site above the Siatom in the third layer and d(s;z)
—
—
6.2a.u. The reference ofenergy is tak-en to beET'.
Calculations are performed using-550
plane%'aves.
E"'
(mRy) -S6SURFACE METALLIZATION OFSILICON BY POTASSIUM.
. .
IV. ELECTRONIC STRUCTURE
The character
of
the interaction underlying the large binding energy is clari6ed bya
studyof
the electronic structure. The energy band structureof
the unsupportedK
monolayer with the same20
lattice registry as theK
overlayer is shown inFig.
3(a). The lowest band is formed from the 4s-valence orbital, and thus has an s symmetry. The dispersionof
this band is large along the chain direction (~~k„}but small in the directions perpen-dicular to the chain (~~k„),so that the ratioof
the effective massesm„'
/rrt'
=
6.
This implies that the alkali-metal overlayer would display a 1D metallic char-acterif
it underwent a metal-insulator transition. How-ever, as we shaB see later, this does not happen. The dimensionalityof
the upper p-like bands is not as obvi-ous asof
the lowest s-like band.The energy bands
of
the K-covered Si are shown inFig.
3(b).It
is seen that upon adsorption the metallic bandsof
theK
monolayer are almost totally discarded, and perhaps are merged in the conduction bandof
the substrate. Two bands in the gap, labeled D& and D2, originate from the dangling bondsof
the reconstructed Si(001) surface. InFig.
3(c) we have schematically traced the originof
these two bands. Dangling orbitals (rabbit ears)of
a Si atom interact with thoseof
the adja-cent Siatom to give a fully occupied0
dimer bond and a completely empty0'
band in addition to rr(D, }andrr'(D2)
bands in the regionof EF.
Earlier it was pro-posed that owing to the asymmetryof
the dimer bond, theD
t andDt
bands split, causing the cleanrecon-structed surface to have an insulating character.
It
was later shown3 that the energy difFerence between the symmetric and asymmetric dimer bonds is too small for thisto
occur, and upon adsorption the dimer bond is symmetrized leading to a slight overlapof
the surface-state bands. Therefore the overlapof
the bands in the gap arises mainly from the polarization effectsof
the ad-sorbed potassium. The Fermi level passes through the Dz band, which is normally empty when the Si surface is freeof
adsorbates.It
appears that the valence 4s elec-tronsof
theK
overlayer are accommodated by D2 (which becomes partially occupied) leading to the metall-izationof
the surface.Surface states
or
resonances in the conduction band are shown only in the energy region where the interband plasmon dispersion is measured. These states are ac-cessible by the excited electrons in theEELS
experiment, which samples mainly the surface region, and are likelyto
be responsible for collective and individual excita-tions. The resonance stateR
(which originates from0')
hybridizes with different bulk conduction-band states along the[100]
direction. At the centerof
the surfaceBZ
theR
state is located1.9
eV above the filled D&band. The individual excitation energy inferred from the energies
of
theD,
and R states is in fair agreement with the observedEELS
peak.'
The local and total densityof
states inFig. 4
shows the surface characterof
D,
and D2 states. The backbonding and dimer-bond states can be identified in the valence band, indicating that the effectof
the adsorbed alkali-metal layer on the surfaceUJ C 0 K
-1 -
[11O] 7 K IFKIf. 3. Calculated energy-band structures: (a) unsupported Kmonolayer (or
K
chain) with the same20
lattice registry as the K overlayer, {b)K+
Si(001)-(2& 1) surface withd&s;K,
—
—
4.9 a.u. For (c},see text. Surface states for k~([100] are shown by dotted lines. Surface Brillouin zone and the sur-face atomic con6guration are shown by the inset. Large shad-ed, small sohd, and small open circles describe Kand Srst- and second-layer Si atoms, respectively.states is insigni6cant. The calculated work function
of
the K-covered surface is found
to
be2.
3 eV lower than the work functionof
the clean surface.fhis
value is in fair agreement with the observed loweringof
the work function upon theK
adsorption. 'S.CIRACI AND INDBR
P.
SATRA 37 CL) lg +J CA V O CA lU (ti Qp Cg 2Si(1L)+K—.
—
2Sit4L) D)+D2 (a) -6 -4 Energy (eV)FIG.
4, Layer density ofstates calculated at the surface (1L) and at the fourth layer (4L) of the K-covered Si(001)-(2X1) slab. Pdb and Sdb are p- and s-like dimer bond states. SPb isan (s+p)-like backbonding state (see Ref. 38). D& and D& aredangling-bond surface states.
(b)
V. CHARGE
DENSE
YANAI.VSISD,
(
The origin and the localization
of
three statesD„Dz,
andR,
which are relevant for the surface metallization, and the excitation spectrum are deduced from analysisof
the state charge distribution. InFig.
.5 the contoursof
charge densitiesof
D& and D2 for the cleanSi(001)-(2X1)
surface show clearly their dangling-bond origin. The resonance state8
in the conduction band has a sur-face localization and an antibonding character. Figure 6 iBustrates the same states after the adsorptionof
potassi-um.It
is seen that except for slight polarization and changes in the valueof
charge maxima the dangling-bond characterof
D,
and D2 is unaltered. However, no charge accumulation is observed in the proximityof
the potassium atom; this would be suggestiveof
the significant contributionof
the alkali-metal atoms,or
the retransferof
charge upon the metallizationof
the over-layer. The changes in the characterof
the empty reso-nance state owingto
the contributionof
theK
atom are, on the other hand, observable [see Fig.6(c)].
While the density between the surface and the subsurface layers re-cedes, the vacuum region above the alkali-metal over-layer gains some charge. The contour plotsof R
inFig.
6(c) give an impression about the formation
of
the charge densityof
a state having some origin from the overlayer. Note that the clean surface has also a reso-nance state about 2 eV above the Fermi level. This state prevails in the presenceof
the overlayer, except that its energy shifts slightly.The contour plots
of
the total (valence) charge density in a horizontal plane2.4 a.
u. above the surface (which correspondsto
the planeof
the overlayer with the equi-librium d~s; z~—
—
4.9 a.
u. found in our study) are shown inFig.
7 for the clean and K-covered surfaces. Their shape with two maxima, and their location with respectto
the surface Si atoms, indicate that the charge density above the surface is peculiar to the substrate surface. The characteristic shapeof
the contours prevails evenFIG.
5. Charge-density contour plots ofD&,D2,and R cal-culated for the clean Si(001)-(2X1)surface atk=(0;
0.
108 a.u. )in avertical plane described in Fig.6. The atomic (001)planes ofSi substrate are shown by dash-dotted lines. Charge
density increases in the direction of small arrows. Contour spacings are (a)5X
10,
(b) 5X10,
and (c) 10 a.u. far above the overlayer. The charge densityof
the un-supportedK
monolayer, which is indicativeof
how the charge with s symmetry would be distributed upon the metallizationof
the overlayer, is distinctively diN'erent and has a single maximum (seeFig.
8). A similar com-parison can be made inFig. 9,
where the charge-density contour plots in a vertical plane passing through the di-mer bond are shown. The pronounced el'ectof
the ad-sorption is the increaseof
the densityof
contours around surface silicon atoms, but again no charge is seen above the alkali-metal overlayer.A pictorial manifestation
of
the fact that the partiallyfiilled gap states are only the antibonding dangling-bond states is exhibited in
Fig. 10.
The vertical plane chosen passes through the potassium and its twonearest-neighbor surface Si atoms. Figure 10(a) shows the charge density for the clean Si(001)-(2
X 1)
surface,pISiI.
The corresponding contours in the presenceof
the
K
overlayer,pIK+Sij,
are seen inFig.
10(b). As in previous plots there is no obvious dramatic change in the shapeof
the contours, but notably absent is any directional bond betweenK
and the nearest Siatoms or any metallic charge distribution above the overlayer. The densityof
the contours increases in the caseof
theK
overlayer in going from the surface into the vacuum with a general accumulationof
charge in the dangling-bond region. A crucial result emerges upon examining the dtfference plot,pI
K+81
I—
pISlI,
1n F1g. 10(c).Ow-ing to the core repulsion the charge is actually depleted from the region around the
K
atom and instead it has accumulated around the surface Siatoms. The shapeof
the accumulated charge is reminiscent
of
the dangling-bond surface state D2 inFig.
6(b).This analysis
of
the charge density demonstrates that the adsorptionof
the alkali-metal atom does not leadto
any new state in the gap, but the 4s valence electron has simply gone into SBing the empty dangling-bond surface state.VI, MSCUSSIONS
The total energy and force calculation in
Sec.
III
yielding an equilibrium structure, in which the adsorbed potassium atom was located at the center
of
the hollow site between two dimer bonds. As depicted byFig.
10, the total charge has the lowest density, where theK
atom is located. According
to
the equilibrium structure found in this study each potassium has four nearest-neighbor silicons, the K-Si internuclear distance beingD2
FIG.
6. Charge-density contour plots ofD&,D&,and R cal-culated for the K-covered Si(001)-(2X1)surface atk=(0;
0.108a.u. '}in a vertical plane passing through K and two nearest surface Si atoms. d&&; K»——4.9a.u. and contour spacingsare 5X 10 for(a)and (1),and 10 for (c)in a.u.
FIG.
7. Contour plots ofthe total charge density in an hor-izontal plane 2.4a.u. above the Si surface. @[[[110];ye~[110], and zI[001].
The surface unit cell is delineated by dash-dotted lines. Contour spacings are0.
002a.u.S. CMACI AND INDER P.BATRA 37
4.
9a.
u. On the other hand, the nearestK-K
distance in the overlayer chain is7.
26a.
u.,which is1.
4
a.
u. smaller than the nearest-neighbor distanceof
bulk potassium in the metallic state. The shorter distance is consistent withK
being ionic on the surface. Note that Cs atoms adsorbed on W(001) surface were observedto
have the Cs-Cs distance smaller than that in the bulk metallic state.' In viewof
the rapid decreaseof
the work func-tion an ionic state was attributedto
the Cs overlayer in the initial stageof
adsorption.The cohesive energy
of
potassium in the bulk metallic state isknownto
be—
1 eV. We calculated the cohesive energyof
the unsupportedK
monolayer with the same lattice registry as theK
overlayer to be0.
6
eV per atom. The calculated binding energyof
the adsorbedK
atomwas rather large (3 eV) and is comparable with the cohesive energy in the ionic KC1 crystal. This simple analysis based on energetic considerations suggests that the binding is ionic and the metallization
of
the adsorbedK
overlayer does not take place. Otherwise, the metalli-zation would cause the alkali-metal atoms overlayerto
relax from the surface by regaining their charge, so that they would lose 3 eV per atom but gain only
0.
6 eV through metallic cohesion.Our results obtained from the calculations
of
the elec-tronic structure andSCF
charge density corroborate our conclusion that the binding is ionic.It
becomes clear that the adsorptionof
the alkali-metal atom does not lead to any significant changes in the statesof
the clean surface. The bonding and antibonding dangling-bond bands persist with slight modifications in their disper-sions. No genuineK
states suggestiveof
the overlayer metallization appear near the Fermi level. A sizable effectof
theK
atom was seen only in the conduction band.FIG.
8. Contour plots ofthe charge density calculated for the unsupported Kchain (monolayer) with the20
lattice regis-try of the K-covered Si surface. The unit cell is shown bydash-dotted lines ofthe (a) horizontal and (1)vertical plane as
in Fig.6. Contour spacings are 4)&10 a.u.
FIG. 9.
Contour plots ofthe total charge density in a verti-cal plane passing through the dimer bond: (a) clean surface and (1)K-covered surface. Contour spacings are0.
004a.u.SURFACE METALLIZATION OFSILICON BY POTASSIUM.
.
.
(a)
(b)
(c}
FIG.
10. Contour plots of the total density in a vertical plane passing through Kand two nearest surface Si atoms (the same plane as in Figs. 5 and 6): (a) clean Si(001)-(2X1);(b)K+Si(001)-(2)&1)with d&s;K,——4.9a.u. (c) DifFerence plot of
(a) and (b) showing regions ofcharge depletion (dashed lines)
and charge accumulation (solid lines), Uncharged-density con-tours are shown by dash-dotted curves. Atoms in the plane
and atoms projected onto the plane are shown by solid and open circles, respectively. The contour spacings are 0.004 in (a)and (b), and 8)&10 a.u. in (c).
In order to explore the character
of
the empty bands, the K-coveredSi(001}-(2X1)
surface has been recently studied by using angle-resolved inverse-photoemission spectroscopy. At normal incidence two surface-sensitive features0.
25 and2.
8 eV above the Fermi level are identified, which are quenched upon exposing the surface to oxygen. The energy locationsof
these features coincide with the calculated empty surface states D&andR.
The lower surface feature exhibits verylittle dispersion around
I
and then fades away. This isconsistent with the fact that the calculated Dz band dips
belo~
the Fermi level and becomes occupied. The up-ward dispersion at larger k along the[100]
direction, however, is not accounted for in the calculations. The observed feature in the conduction band disperses up-wards along the[100]
direction, and could agree well with theR
band.These results indicate that the valence electrons
of
theK
atoms are donated to fill the empty surface states leading to the metallizationof
the Si surface. In con-trast, alkali-metal atoms adsorbed on transition-metal surfaces regain their charge at saturation coverage, and become metallized. This is shown by the variationof
the work function, which passes through a minimum, and increases towards a saturation value correspondingto
the work function
of
the alkali metal. Here the presenceof
the active dangling-bond states provides the crucial difference between a metal and a semiconductor surface at monolayer coverage. In the present system the metal-insulator transition takes place through the metall-izationof
the semiconductor surface, but not through a Mott-type transition. Stated difFerently, metallic bands do not form on theK
monolayer adsorbed on the Si(001} surface.The attractive Coulomb interaction between the posi-tively charged core and the surrounding electron density
of
the dangling bonds is responsible for the strong bind-ing. Such an interpretation is also supported by the total-energy calculations. In the courseof
structure op-timization, the valence electronsof
K
were already transferred to the surface even at a large distance (h=5.
1a.
u. ). As h decreases, the Coulomb energy predominates over other componentsof
the total energy, and pulls the ion to the equilibrium position. Strong ion-ion repulsion (represented by the ionic pseudopoten-tial), however, prevents it from entering into the sub-strate.Since the adsorbed alkali-metal atom donates its valence electron to the Si surface band, and eventually becomes positively charged, the 3p-core shift and the de-crease
of
the work function are easily understandable. The rapid decreaseof
work function at8
&g1 may im-ply a relatively larger adatom-surface distance at the ini-tial stageof
coverage. As the coverage increases more charge builds up on the surface and, consequently, higher attractive forcesact
on the adatom to pull it closer to the surface. Toward saturation coverage, the adatom-surface distance becomes smaller and results in an effective core screening. The local densityof
states above a horizontal plane bisecting the erst and second Si layers yields 8.9
electrons per surface unit cell (slightly fewer than nine electrons). In this region the surface electrons can provide ascreeningof
theK
core potential comparable to thatof
the bulk metal. This picture is consistent with the decreaseof
the work function at a lower rate near saturation coverage. At this point we comment upon recent experimental work by Gelling and Miranda, revealing for the first time the multilayer coverageof
K
on the Si(001) surface. In contrast to pre-vious data, ' they observed that the work function also passes through a minimum at8=1.
Upon furtherK
S. CIRACI AND INDER
P.
SATRAdeposition
4
saturates0.
4
eV above the minimum value. By using the Helmhotz expression they estimated a di-pole lengthof
—
1a.
u. for the caseof
complete charge transfer fromK
to dangling-bond states. Their expen-mental results can be interpreted to provide additional support for our predictions. The observationof
4(8)
beyond
8
=
1 is certainly possible, but that rangeof
cov-erage has not been examined in our calculations. How-ever, the lackof
a minimum in the work-function change,64(8)
for0&8&1
confirms the absenceof
overlayer metallization even when the
K
chain has been fully formed. As the coverage increases beyond6=1,
whereK
either adsorbs at a diferent site or forms a new layer,4(8)
undergoes a change as expected. Finally, as is clear from the charge distributions shown in Figs. 6 and 10,the average dipole length should be smaller than our calculated equilibrium height, as is in fact obtained by them. This is due to the sizable protrusionof
the danghng bond towards the vacuum, in which the valence chargeof
K
is accommodated.It
should also be noted that the change in work function cannot be related in a simple way to the de reeof
ionicity becauseof
the effectsof
polarization. In our calculations a rigorous accountof
the charge distribution led toh4
in fair agreement with the experimental value. ' There is ap-parently some confusion about the changeof
h with cov-erage.It
has been shown by us earlier' that with in-creasing8,
h increases, provided that there is overlayer metallization. In the present work, upto
8=1
only, the overlayer was not metallized up to that coverage, and hence there is no increase in h.So far we have established the type
of
the binding and the originof
the metsllizstion. In viewof
these findings let us turn tos
brief discussionof
the excitation spec-trum.2 Electron-energy-loss spectra(EELS)
show apeak with the energy position depending on the
K
cover-age. ' At low coverage a broad and weak loss peak at 4 eV could not be identified clearly becauseof
the strong bulk transitions. However, the intensityof
the energy-loss peak increases sharply for8
~0.
6, and is stabilized in energy near 2 eV. Tochihara ' attributed this loss peakto
the excitationof
the plasmonof
the overlayer. Subsequently, a positive dispersion (starting from1.
7 eV at q=0
and dispersing upwards along q~~[100j and q~~[110] directions) was observed by angle-resolvedEELS.
The collective excitations
of
2D metals have been, in the past, investigated extensively. ' Introducing the so-called box model, Newns' was able to formulate the dispersion relationsof
2D metals from a 2D degenerate electron gas.For
small momentum, he found the intra-band plasmon dispersion as~„~q
and the interband plasmon dispersion as co,~a.(1
—
q).
According to his findings, co, first disperses downwards at small q and passes through a shallow minimum and thereafter in-creases with increasing q. This behaviorof
the inter-band dispersion was confirmed experimentally from work on theK
overlayer on Ni(001) surface.' Based on this dispersion relation, which is not in agreement with the observed dispersion, snd in viewof
the structural modelof
theK
overlayer on a Si(001) surface, Arugaet
al.
attempted to deduce the dimensionalityof
the overlayer. Using the parallel-rod model and also the %'snnier representation based on the surface band struc-ture, Tsukada etal.
* calculated the dispersionof
theK
overlayer.The observed plasmon dispersion should deviate from the dispersion relations based on the 2D degenerate electron gss for the following reasons: First, as illus-trated in
Fig.
11 the surface charge density is reminis-centof
a rod, which originates from the Sisurface states, but not from the metallizationof
the overlayer. Second-ly, as revealed from the self-consistent-field microscopic theoryof
the surface plasmons, ' the formof
theinitial- and final-state wave functions plays a decisive role in determining the dispersion
of
the collective exci-tations. The initial- and the final-state wave functions inFig.
6 do not display any free-electron-like separablex
FIG.
11. (a) Contours ofthe total charge density ofthe K-covered Si(001)-(2& 1)surface in the Si surface plane with con-tour spacings of0.
004a.u. Surface and subsurface Si atoms areshown by solid and open circles. Sohd and dashed lines denote the dimer and back bonds. The (x,y) position ofthe adsorbed Katom isindicated by
x.
(b} The surface plot of(a).37 SURFACE METALLIZATION OFSILICON BY POTASSIUM. .
.
components. Therefore, one cannot attribute a literally 1D character (i.
e.
,states are confined in two dimensions, but have afree-electron-like behavior in the third dimen-sion) to the charge density arising from the D2 band. In principle, the Fourier transformof
the overlap matrix has to be incorporated. Since the dimer bond and the backbonding states are localized in the proximityof
the surface, they are easily sampled byEELS.
Moreover, the energiesof
these states occur near the gap. Conse-quently, these states should also be taken into account in the initial statesof
the polarization operator. The loga-rithmic singularitiesof
the polarization operator corre-spond to single-particle (individual) excitations. The single-particle excitation fromD,
orD2 to R isexpectedto
have a significant contribution to the loss peak at about 2 eV. As noted earlier, anEELS
peak at1.
7 eV was observed~ even for the clean Si(001)-(2)&1)surface, and was assignedto
an individual excitation from D& toan unknown surface resonance in the conduction band. Such a surface resonance (labeled
R)
is found in the present study in nearly the correct energy position, and is found to persist in the presenceof
theK
overlayer. Since the overlayer has not metallized becauseof
the presenceof
active dangling bonds, the originof
theEELS
peak can reasonably be sought in single-particle excitations.In summary, the type
of
bonding proposed in this work is based on the considerably high binding energyof
potassium, the electronic structure, and an extensive charge-density analysisof
theK+Si(100)-(2X1)
system. The consensus is that the bonding is ionic fore~gl,
and ase~
1 the surface becomes metallic. Thecontro-versy lies in, however, what is metallized (the Si surface or the
K
overlayer), and what is the characterof
the bonding. Theories claiming the metallizationof
theK
overlayer are based mainly on the formation
of
a 1D me-tallic chainof
alkali-metal atoms having a nearest-neighbor distance smaller than thatof
the bulk. ' Such a metallization is conceivable for metal substrates, and the gainof
energy would not be larger than 1 eV.However, owing to the active, empty dangling-bond states, the energy
of
metallization cannot balance the large binding energyof
-3
eV, and hence the bonding continues to be ionic even at8=
1.
As becomes clear from the above discussion the present assignment for the bonding is consistent with the experimental data, and is also in compliance with the electronegativity con-siderations. Recalling the electronegativityof
Si[X(Si)
=1.
8]and the electronegativityof K [X(K)
=0.
8]
one infers that Si is twice as likely to capture charge than
K.
Our arguments about the bonding are support-ed by similarSCF
calculations by Northrup, who has predicted an ionic bonding for sodium adsorbed gn the idealSi(111)
surface. Beyond the local-density approxi-mation, Hartree-Fock calculations indicate also a large binding energy and transferof
charge fromK
to Si.
Other theoretical results '
proposing the metalliza-tlon
of
the overlayer either lack the total-energy mBlNH-zation,or
are based on inadequate interpretationof
ex-perimental data. Previously Ishida etaI.
calculated the band structureof
the KSis H4 thin film by using anLCAO-Jo.
method. They did not carry out the total-energy calculationsto
6nd the equilibrium structure, but estimated the K-Si distance from the atomic radiito
be6.
65a.
u.It
is apparent that theK-Si
distance they used in their calculations is much larger than what we found bySCF
total-energy minimization. Their LCAO-Jo.
calculations resulted in a fundamental energy gapof
3 eV (which is much larger than the experimental energy gap) and three surface bands with dominantK
contribu-tion. Based on a Mulliken s population analysis, which yielded+0.
13 charge on the adsorbedK
atom, they concluded that the overlayer was metallized. Thesere-(g]
p =0.00002(e)
5= 0.004(b)
0.07053(f)
0.004 0189 004 0.01423 0.001 0.00354 0.00330 (b]
0.0004 z 'amass gy 0.0050 0.0004 )&'( 14.512 a.u.FIG.
12. Contours ofthe total charge density. First columnis in a vertical plane passing through Kand two nearest sur-face Si shown in the inset. (a)
K+
Si(001)-(2)& 1) with d(s;K)—
—
5.9 a.u. (3.14 A); (b) clean Si(001)-(2&1)surface; (c}dNerence plot of(a)and {b);{d) unsupported, metallic Kchain
in registry with Kon Si surface. Second column [(e)-(h)in the
same order as the first column] is in the plane ofthe K over-layer 4.2 a.u. above the surface Si. p is the maximum value
of charge, and 5 is the contour sparing. Dash-dotted curves
S.CIRACI AND INDER
P.
BATRA 37suits are at variance with our findings. Kasowski and Tsai carried out pseudofunction calculations within the local-density approximation, and concluded that
K
does not form an ionic bond with Si. They also Snd a d~s; K~value larger than ours. They compared their model with the UPS spectrum obtained from the
Cs+Si(111)-(2X
1)system. That the electronic structure
of
K+Si(ill)-(2X
1)
is diff'erent from thatof
K+Si(001)-(2X
1)and is coverage dependent, has been demonstrated recently. Their arguments based on theEELS
peak observed at-2
eV are not conclusive, because the clean Si surface exhibits anEELS
peak in ihe same energy region. These experiments do not provide any direct support for their model calculations. Using an arbitrary charge par-titioning they assign the high charge densityof
the clean Si surface toK.
Consequently, they find charge transfer from Sitoless electronegativeK.
More recently, angle-resolved photoelectron spectros-copy by Enta et
aI. '
rules out previous band models, and veri6es our theoryof
the surface metallization. By going to higherK
coverage, they have observed that the metallic surface becomes insulating ate=2,
because the surface-state band Dz becomes fully occupied.Conclusions about the nature
of
bonding based onSEXAFS
work measuring d~» K~ are not de6nite.Ac-cording
to
acriterion set by Citrin, the interatomic dis-tance measured bySEXAFS
is a perfect candidate for ionic bonding. We believe that one cannot draw definite conclusions about the natureof
the bond simply by look-ing at the bond length. Since the local-density approxi-mation underestimates the structural parametersof
K,
and thus calculated values for d~s;K~ depend on the form
of
the exchange-correlation potential, the important is-sue isthe natureof
the bonding. The pointto
emphasize is that the bonding is ionic even for the valueof
d~s;it~deduced from
SEXAFS. To
this end we show theSCF
charge distribution for d~s; z~——
5.9 a.
u.(3.
14 A). Plotsof
the valence charge-density difference,i.e.
,pIK+Sij
—
pISi),
inFig.
12 unambiguously show that the originof
the surface metallization is the dangling bond and not the metallic chain. As a result, the bond is ionic at our d~s; K~, as well as at the large d~s;K~ derivedfrom
SEXAFS.
VII. CONCLUSIONS
The alkali-metal-Si(001) surface system at monolayer coverage presents remarkable features in the adsorption phenomena, as mell as in the electronic structures
of
sur-faces.It
appears that the alkali-metal atoms, which nor-mally form simple metals, lead to ionic bonding when they are adsorbed on the Si(001) surface. The substrate with a superlattice gap changes into ametallic state. Al-though the surface charge density displays a 1D-like character, the observedEELS
dispersion cannot be asso-ciated with the metallizationof
the adsorbedK
chains. As far as the pinningof
the Fermi level isconcerned, theK
overlayer on the Si(001) surface is a unique system among metal-semiconductor interfaces investigated so far. The Fermi levelof
the present system is pinned completely by the surface states as proposed by Bar-deen, but not by chemisorption states or metal-induced states. With these results the interactionof
potassium with the Si surface and the characterof
the normal modesof
charge-density fluctuations will be elucidated, leading to signi6cant revisions in the previous under-standing. The multilayer coverage, on the other hand, seems to remain as a challenging problem both experi-mentally and theoretically.'R.
H. Williams, Contemp. Phys. 23, 329 (1982). L.J.
Brillson, Surf. Sci. Rep.2, 123(1982). 3M. Schluter, Thin Solid Films 93,3(1982). 4A. Kahn, Surf. Sci. Rep.3,193(1982).G. W. Rub)os, in Fesskorperprob1eme (Aduances in Physiesj, edited by
J.
Treusch (Pergamon, Vieweg, 1983),Vol.XXII,
p. 179.
J.
Bardeen, Phys. Rev.'71,717 {1947).~V.Heine, Phys. Rev. 138, A1689 (1965).
S.G.Louie and M.L.Cohen, Phys. Rev. Lett. 35,866(1975); Phys. Rev.
8
13,2461(1976).9J.
E. Rose,
S.B.
Christman, and G. Margaritondo, Phys. Rev. Lett. 35,1471(1975).~oG. Margaritondo,
J.
E.
Ro~e, and S.B.
Christman, Phys.Rev.
8
14, 5396 (1976)."%'.
E.
Spicer,I.
Lindau, P. Skeath, C.Y.
Su, and P.Chye, Phys. Rev. Lett. 44, 420(1980)„J.
Vac. Sci. Technol. 17,1019 (1980).
'2I.
P.
Batra,J.
Vac. Sci.Technol.8
1,558(1983);I.
P.Batra andF.
Herman,J.
Vac. Sci.Technol. A 1,1080{1983). 'I.
P. Batra and S.Ciraci, Phys. Rev.8
29, 6419 (1984);J.
Vac. Sci.Technol.
8
3, 427 (1984);Phys. Rev.8
33, 4312 (1986).'4S.Ciraci and
I.
P.Batra, Phys. Rev.8
33,4294(1986). A.Zunger, Phys. Rev.8
24, 4372{1981).'6N. D.Lang, Phys. Rev.
8
4, 4234 (1971).'
E.
%'immer, H. Krakauer, M.Weinert, and A.J.
Freeman,Phys. Rev.
8
24, 864(1981).~SA.U. MacRae,
K.
Muller,J.
J.
Lander,J.
Morrison, andJ.
C.Phillips„Phys. Rev. Lett. 22, 1048(1969). The two-layer model, proposed in this work for explaining the minimum in
hN, has been challenged on other grounds by
J.
L.
Fehts, T.J.
Lee,B.
J.
Hopkins, andR. E.
Stickney, Surf. Sci. 21,197 (1970);andJ.
P.Muscat andI.
P. Batra„Phys. Rev.8
34, 2889(1986).'9U.Jostell, Surf. Sci.S2,333(1979).
S. A.Lindgren and L.%'allden, Phys. Rev.
8
22,5967 (1980). ~'H.Tochihara, Surf.Sci. 126,523(1983).2 H. Tochihara and Y. Murata,
J.
Phys. Soc. Jpn. 51, 2920(1982).
2
T.
Aruga, H. Tochihara, andY.
Murata, Phys. Rev. Lett. 33, 372(1984).
M. Tsukada, H. Ishida, and N. Shima, Phys. Rev. Lett. 53, 376 (1984); H.Ishida, N.Shima, and M.Tsukada, Phys. Rev.
8
32, 6236 (1985).37 SURFACE METALLIZATION OFSILICON BY POTASSIUM.
.
.
2967(1985).
~6H. Tochihara, M. Kubota, M.Miyao, and Y.Murata, Surf. Sci. 158, 497 (1985).
2~S.Ciraci and
I.
P.
Batra, Phys. Rev.Lett. 56,877(1986).~8M. Schluter,
J.
R.
Chelikowsky, S. G. Louie, and M. L. Cohen, Phys. Rev.8
12, 4200 (1975).J.
Ihm, A. Zunger, and M. L. Cohen,J.
Phys. C 12, 4409 {1979); M.T.
Yin and M. L. Cohen, Phys. Rev. Lett. 45,1004(1980}.
30C.
8.
Bachelet,D. R.
Hamann, and M.Schliiter, Phys. Rev.8
26,419(1982).3'D. M. Ceperley and
B.
J.
Alder, Phys. Rev. Lett. 45, 566 (1980).3
J.
P.Perdew and A.Zunger, Phys. Rev,8
23,5048(1981}.F.
F.
Abraham andI.
P. Batra, Surf. Sci. Lett. 163,L572 (1985). The asymmetric dimer bond was proposed by D.J.
Chadi, Phys. Rev. Lett. 43, 34 (1979).
34K,C.Pandey, in Proceedings
of
the Seuenteenth InternationalConference on the Physics
of
Semiconductors, edited by D.J.
Chadi and W. A. Harrison (Springer-Verlag, New York, 1985),p. 55.
35R.M. Tromp,
R.
J.
Hamers, andJ.
E.
Demuth, Phys. Rev. Lett. 55; 1303 (1985}.36M.
T.
Yin and M.L
Cohen, Phys. Rev.8
24,2303(1981).J.
D.Levine, Surf. Scl. 34, 901 (1973).38S.Ciraci,
R.
Butz,E.
M.Oelling, and H.Wagner, Phys. Rev,8
30, 711 (1984). Many references related to the clean, reconstructed Si(001)surface can be found in this reference.3For the method and related references see
I.
P. Batra, S.Ciraci,
G.
P.Sirivastava,J.
S.Nelson, and C. Y.Fong, Phys. Rev8
34,8246(1986)T.Kendelewicz, P.Soukiassian,
R.
S.List,J.
C.Woicik, P. Pianetta,I.
Lindau, and%.
E.
Spicer (unpublished).'V. L.Moruzzi,
J.
F.
Janak, and A.R.
Wilhams, Calculated Electronic Properties ofMetals {Pergamon, New York, 1978). L.Hedin andB.
I.
Lundquist,J.
Phys. C 4,2064(1971).~3Foran extensive discussion see D.Pines, Elementary
Excita-tions in Solids {Benjamin, New York, 1964),p. 156.
~I.
P.Batra and P. S.Bagus,J.
Vac.Sci.Technol. {tobepub-lished).
45S.Ciraci and
I.
P.Batra, Phys. Rev. Lett. $8, 1982{1987}. ~6F.J.
Himpsel and D.E.
Eastman,J.
Vac. Sci.Technol. 16,1297(1979);D.
J.
Chadi, ibid. 16,1297(1979}.47J.
E.
Rowe and H. Ibach, Phys. Rev.Lett. 32,451{1976).~SI.P.Batra,
J.
M.Nicholls, andB.
Reihl,J.
Vac. Sci. Tech-nol. A5, 898 {1987).~
E.
M. Oelling andR.
Miranda, Surf. Sci.1??,
L947 {1986).In this work the coverage was defined as the atomic ratio
be-tween Kand surface Si. Consequently,
6=0.
5in this work corresponds to8=1
here.5OL.
G.
Pettersson and P.S.Bagus, Phys. Rev. Lett. 56, 500(1986);
I.
P.
Batra, Prog. Surf. Sci.(to be published). 5~F. Stern, Phys. Rev. Lett. 18,546(1967).52J.N.Gadzuk, Phys. Rev.
8
1,1267(1970).5 D.M.Newns, Phys. Rev.
8
1, 3304 (1970);Phys. Lett. 38A,341(1972).
54S. Anderson and U. Jostell, Solid State Commun. 13, 833
(1973).
M.Nakayama,
J.
Phys. Soc. Jpn. 39,265(1975).56H. Froitzheim, H. Ibach, and D. L. Mills, Phys. Rev.
8
1, 4980(1975).57For a review ofthe subject, see
T.
Ando, A.B.
Fowler, andF.
Stern, Rev. Mod. Phys. 54,437 (1982).5SM. Nakayama,
T.
Kato, andK.
Ohtomi„Solid StateCom-mun. 50, 409 (1984).
J.
E.
Northrup,J.
Vac.Sci. Technol. A4, 1404(1986).R.
V.Kasowski and M.-H.Tsai, Bull. Am. Phys. Soc.32,865{1987).