-Br/Pt(110) surface

PHYSICAL REVIEW B, VOLUME 65, 165408

Structure of the c„2Ã2…-Br ÕPt„110… surface

V. Blum, L. Hammer, and K. HeinzLehrstuhl fur Festkorperphysik, Institut fu¨r Angewandte Physik, Staudtstrasse 7-A3, D-91058 Erlangen, Germany

C. Franchini* and J. RedingerCenter for Computational Materials Science, Vienna University of Technology, Getreidemarkt 9, A-1060 Vienna, Austria

K. Swamy, C. Deisl, and E. BertelInstitut fur Physikalische Chemie, Unversita¨t Innsbruck, Innrain 52a, A-6020 Innsbruck, Austria

~Received 14 September 2001; published 4 April 2002!

We present a detailed investigation of thec(232)-Br/Pt(110) adlayer structure supplemented by the analy-sis of the (132) missing-row~MR! structure of the clean Pt~110! surface. Quantitative low energy electrondiffraction analyses and first-principles calculations are in impressive agreement in both cases. The cleansurface reconstruction is determined with unprecedented accuracy. For the adsorbate, the analysis retrieves asimple Br-adlayer structure with the Br atoms residing in every second short bridge position on the close-packed Pt rows with the MR reconstruction lifted. The Br-Pt bond lengthL52.47 Å is almost equal to thesum of the atomic radii. The substrate below the adsorbate exhibits a contraction of the first layer spacingwhich amounts to half of that calculated for an unreconstructed clean surface.

DOI: 10.1103/PhysRevB.65.165408 PACS number~s!: 68.43.Fg, 61.14.Hg, 68.43.Bc, 68.37.Ef

ollc

acra

ticdyd-oa

nn

ceh

helowon

c

ED

ad

geg

s-

e

g.,

eti-ga-

ae

sti-d ins-

ried

entf a

rre-fromtivefa-s.foron

ise,our

he

entns.

I. INTRODUCTION

Halogens interact strongly with all metal surfaces. Fthis reason they play an important role in many technicaimportant processes. For instance, halogens are stronglyrosive. By the same token they may be used for surfetching.1 As strong adsorbates, they efficiently poison sevecatalytic reactions, but are also used as promoters in parlar cases.2 Finally, they are interesting candidates to stuwithin the context of self-structuring. Due to their high asorption energy one may anticipate halogen-induced recstruction, but also major changes in the surface stressconsequently themesoscopicsurface morphology.3 In addi-tion, the well-known tendency of halogen-bridged transitiometal linear-chain compounds to form competing groustates@such as charge density waves~CDW’s! or spin densitywaves4# renders halogen adsorption on anisotropic surfaan interesting candidate to study these low-dimensional pnomena on surfaces.5

There are, however, only few studies dealing with tquantitative determination of the geometric structure of hagens on metals. In most previous studies, the structurededuced from qualitative low energy electron diffracti~LEED!6–11 or scanning tunneling microscopy~STM!.3,12–16

Adsorption site analysis was carried out mainly by surfaextended x-ray-absorption fine structure~SEXAFS!,17–23

whereas quantitative structure determinations by LE~Refs. 24–26! or first-principles calculations27,28 are sparse.

In all quantitative halogen adsorption studies a simplesorption layer~simple overlayer model in Ref. 18! is favoredover substitutional adsorption~mixed-layer model in Ref.18! for halogen coverages up to 0.5 monolayers~ML !. Forhigher coverages, however, the formation of a metal-halocorrosion layer is observed for different low-index Asurfaces.29

As an example, Cl/Cu~111! represents a well-studied sy

0163-1829/2002/65~16!/165408~13!/$20.00 65 1654

ryor-elu-

n-nd

-d

se-

-as

e

-

n

tem, investigated by, e.g., SEXAFS,19 normal incident x-raystanding wave field absorption~NIXSW!,30 and DFTcalculations.27 Cl is found to adsorb in the fcc hollow sitforming a simple overlayer. In the case of~100! surfaces,simple adsorption in the fourfold hollow site is favored, e.by a LEED structure analysis@Cl/Ag~100!,24,31 Cl/Cu~100!~Refs. 25,26!# and SEXAFS studies@I/Cu~100!,17 Cl/Cu~100!~Refs. 18,23!#. With regard to halogen adsorption on thmore open~110! surfaces of fcc metals, there are no quantative studies of the adsorption structure. In most investitions simple adsorption was assumed.3,6–9,12,15A cluster cal-culation for I/Ag~110! ~Ref. 28! favors the short bridgeposition as adsorption site.

In a previous paper16 some of the authors presentedstudy of Br/Pt~110! mainly based on STM and qualitativLEED measurements. For a Br coverage ofQ50.5 ML ac(232) structure was found. It was interpreted as a subtutional structure due to reasons discussed in Ref. 16 anSec. IV of this paper. However, a quantitative LEED invetigation, which is based on intensity measurements carout under identical preparation conditions~in the same lab!,as well as a DFT analysis, both discussed in the prespaper, clearly prove a simple adlayer structure instead osubstitutional adsorption site. The close quantitative cospondence between the structural parameters derivedboth independent methods leaves no room for alternamodels. The adsorption site found in this study is alsovored by new atom-resolved low-coverage STM image32

These new results imply that our previous conclusionsgeneral substitutional adsorption in the case of halogensmetal fcc~110! surfaces have to be revised. We emphashowever, that the new adsorption model does not changeconclusions with regard to the one dimensionality of tc(232)-Br/Pt(110) structure.

The paper is organised as follows. In Sec. II we presexperimental and computational details of the investigatio

©2002 The American Physical Society08-1

hey

e-olte

e-

a-

D

oheis

thin

tl

sit

n

.

resryee

f-in

aiteatufooron

58

chn

her

ofatyeby a

tiontialon-ortialthegy.

ich

is-sed,xi-ic

t-

f

V. BLUM et al. PHYSICAL REVIEW B 65 165408

In Sec. III the missing-row~MR! reconstruction of cleanPt~110! is reanalyzed both by means of LEED and DFT. Tmetastable Pt(110)-(131) structure was analyzed as well bmeans of DFT. The structural results for thec~232!-Br/Pt~110! surface derived from LEED and DFT are prsented and discussed in detail in Sec. IV. This section ccludes with a reinterpretation of former experimental resufrom Ref. 16 in the light of the newly determined brominadsorption structure. The paper closes with a summary.

II. EXPERIMENTAL AND COMPUTATIONAL DETAILS

The experiments were performed in a UHV chamb~base pressure,8310211 mbar) equipped with a commercial STM ~Ref. 33! and control electronics,34 a four-gridLEED system,35 and the usual facilities for sample prepartion and characterization. The preparation of the Pt~110!crystal according to the standard procedure36 resulted in aclean surface showing the well known 132 superstructureoriginating from the missing-row reconstruction. The LEEspots are rather intense, but somewhat elongated in the@001#direction, as was also reported in earlier LEED studies~e.g.,Ref. 37!. This broadening is due to small terrace widthsonly about 15 Å on average for this direction due to tmesoscopic ‘‘corrugated-iron’’ structure found for thsurface.36,38Bromine molecules were dosed atT'130 K bymeans of a solid-state electrolysis cell. Annealing ofsample at 780 K leads to partial bromine desorption leav0.5 ML of dissociated Br on the surface~for details of thepreparation see Ref. 16!. The resulting STM images@see Fig.1~a!# show a well-orderedc(232) structure with only occa-sionalp(231) regions near defects and steps. Consistenthe corresponding LEED pattern is also ofc(232) symme-try with clear superstructure spots of moderate intencompared to that of the substrate spots@Fig. 1~c!#. Addition-ally, however, there are faint streaks in@001# directionthrough the superstructure spots as well as some diffusetensities at@(2m21)/2,n# positions @corresponding to ap(231) periodicity# indicative of residual disorder. This cabe understood in terms of antiphase domains in@001# direc-tion and/or small coexistingp(231) patches at the surfaceSuch a coexistence ofc(232) and p(231) phases wasfound in STM below the optimum annealing temperatuneeded for the well-orderedc(232) structure and at defectand steps.16 Entire LEED patterns taken at normal primabeam incidence and with the sample at about 130 K wrecorded in steps of 0.5 eV in the energy range 40–500by means of a 12-bit digital CCD camera39 ~images weresampled 8 times!. They were stored on a computer for ofline evaluation of sufficiently intense beams. The beamtensities resulted by pixelwise summation within a certframe around a spot whereby the background level demined at the frame’s edges was subtracted. Spectra of besymmetrically equivalent at normal incidence were evenally averaged and slightly smoothed when necessary,lowed by normalization to the primary beam intensity. Fthe 132 missing-row structure of the clean surface this pcedure resulted in a data set of 16 integer and 11 fractioorder beams with integrated energy widths of 3761 and 2

16540

n-s

r

f

eg

y,

y

in-

reV

-nr-ms-l-

r-al7

eV, respectively, yielding a total data base width of as muas DE56348 eV. For thec(232)-Br adsorbate phase teinteger order~3297 eV! and four fractional order~1289 eV!beams were taken with a total energy width ofDE54586 eV.

LEED intensity calculations were performed using tTENSERLEED program package.40 This applies the TensoLEED perturbation scheme,41–43 extended to allow not onlyfor the variation of the surface geometry but also for thatvibrational parameters.44,45For a certain reference structurefull dynamical calculation is carried out and intensichanges due to~small! deviations from this reference arcalculated by perturbation. The structural search is madefrustrated simulated annealing procedure46 and guided by thePendryR-factorRP ~Ref. 47! for the quantitative comparisonof experimental and calculated spectra. Electron attenuawas described by an imaginary part of the optical potenV0i . It was adjusted for the clean surface resulting in a cstant value ofV0i56 eV and the same value was used fthe Br adsorbate phase. The real part of the inner potenV0r was allowed to be energy dependent according tovariation of the exchange-correlation potential with enerFollowing the literature48 the expressionV0r5V001max(210.64 eV,0.63 eV285.10 eV/AE/eV117.02) wasused with only the constant part fitted to the spectra whresulted inV00520.5 eV for the clean andV0050 eV forthe bromine covered surface. As much as 14 fully relativtically calculated and spin averaged phase shifts were usufficient to describe the atomic scattering up to the mamum energy of 500 eV. They were corrected for isotrop

FIG. 1. ~a! STM image of thec(232)-Br/Pt(110) surface (I t

50.87 nA, Vt5820 mV). The contrast is 0.25 Å.~b! CalculatedSTM image fromFLEUR-GGA-9L results: A constant current STMimage for an energy range equivalent to~a! shows Br appearing asa protrusion with a corrugation amplitude of 0.25 Å at a tip Psurface distance of;6 Å. Vertical distances~in Å) are given withrespect to the top Pt layer.~c! Corresponding LEED pattern (U5147 eV). ~d! ~1,1! and (1/2,1/2) spot intensity as a function otime atU5147 eV ~see Sec. IV C!.

8-2

ecoÅorenefroatte

rtoi

thseer-

eesn

rea

t oenin

va

ll-

y

sti

r

r

rsne

h

thi-

il-ir--

0

ent

ex-lab

n tox,ef-antryÅ.

heeuc-

aton-ns

ore,n thebybe-on-wn

g--

ureof

aseto

theac-

es

as

ersnwase-the

ge

STRUCTURE OF THEc(232)-Br/Pt(110) SURFACE PHYSICAL REVIEW B 65 165408

thermal vibrations with fixed amplitudevb50.07 Å for bulkatoms and of variable~i.e., to be fitted! amplitudes for atomsin layers in the vicinity of the surface. For the bulk of thplatinum sample the lattice parameter 3.923 Å was usedresponding to a surface parallel in-plane value of 2.774

Attention had to be paid not to leave the validity rangetensor LEED during the structural fit procedure. Therefonew reference calculations were carried out wheneverpoints reached in the parameter space were too distantthe former reference until convergence was achieved. Sttical error limits for the varied parameters were estimathrough the variance of the PendryR factor47 var(RP)5RPA8V0i /DE with RP the minimumR factor. All struc-tures withR factors below@RP1var(RP)# are supposed tobe within the limits of error. This procedure takes the diffeent sensitivities of theR factor with respect to the differenparameters into account. In principle one should also csider correlations between different parameters, but thisvolves extremely time consuming calculations and soinfluence of such correlations was neglected in the prework. This is actually the standard in LEED structure detmination. Accordingly, the error limits for a certain parameter were determined by theR factor’s crossing of the vari-ance level with variation of the parameter undconsideration but all other parameters kept fixed at their bfit values. We point out that this neglect of correlations geerally underestimates the error limits. For simple structufor which their consideration is still possible, we found ththe true error is about twice that resulting by the negleccorrelations,49 so this might give a rough idea for the truerrors. We also point out that atomic positions not varied aso possibly not correctly taken into account lead to ancreasedR factor and, as a consequence, to an increasedance level.

First principles density functional theory~DFT! calcula-tions were performed using both, the all-electron fupotential linearized augmented plane wave~FLAPW!method,50 as implemented in the packageFLEUR,51 and theVienna ab initio simulation package~VASP! ~Ref. 52! withthe projector augmented wave53 method as implemented bKresse and Joubert.54

Throughout this work two DFT potential approximationhave been used: the generalized gradient approxima~GGA! according to Perdew and Wang~PW91! ~Ref. 55! andthe local density approximation~LDA ! in the Perdew-Zunge~Ceperly-Alder! parametrization scheme.56,57 The surfacewas modeled using a slab of up to 15 layers thickness,peated along the surface normal forVASP, and a single slabwith vacuum on both sides forFLEUR. In VASP the repeatedslabs are separated by a vacuum layer of at least 8 Å.

In FLEUR calculations a symmetric~with respect to themiddle layer! nine layer slab was used, allowing four layeto relax. Well converged results were obtained for plawave cutoffkmax53.7 a.u. Inside the muffin-tin spheres, thangular momentum expansion was taken up tol max58, bothfor the full-potential and charge-density representation. Tcore electrons, including the 5p states, were treated fullyrelativistically and the valence electrons derived fromatomic 5d, 6s, and 6p orbitals are treated semirelativist

16540

r-.f,wm

is-d

-

n-n-ent-

rt--s,tf

d-ri-

on

e-

e

e

e

cally, i.e., dropping only the spin-orbit term in the Hamtonian. Thek-mesh consisted of 16 special points in thereducible wedge of the strictly two-dimensional Brillouinzone chosen according to Cunningham.58

For theVASP calculations an energy cutoff of about 23eV and a 63631 Monkhorst-Pack typek-point mesh turnedout to generate results sufficiently accurate for the prespurposes. Due to its high efficiency,VASP allows one to treatthicker slabs within a reasonable computing time. Forample, the accuracy of our calculations as a function of sthickness was checked by performingVASP calculations for a15 layers slab where 4 layers on one side had been frozebulk geometry. Allowing the remaining 11 layers to relaerrors in the final calculated geometry due to finite slabfects for all the different structure models investigated, cbe tracked easily. Finally, for both methods, the geomewas optimized until all forces were smaller than 0.01 eV/

III. REANALYSIS OF THE CLEAN RECONSTRUCTEDPT„110… SURFACE

Though the low temperature equilibrium structure of tclean Pt~110! surface is known since long to be of thmissing-row type, there is a remarkable scatter in the strtural parameters determined both in experimental37,59–63andtheoretical64–67 studies. In experiment one might argue ththese differences simply stem from varying preparation cditions. Yet, they may also be caused by computer limitatioholding at the time these studies were performed. Therefit appeared reasonable to reanalyse the clean surface ogrounds of today’s standards in structure determinationLEED and DFT. Simultaneously, a close correspondencetween the results of the two methods would give some cfidence for their successful application to the yet unknobromine adsorbate system.

A. LEED results for the 1Ã2 missing-row structure

There is no doubt concerning the validity of the missinrow model for Pt(110)-(132). Hence only parameters describing this model were varied in the course of the structdetermination. Intensities were calculated up to energies500 eV in order to use the whole experimental data bwidth of DE56348 eV. This unusually large value leadsa rather low value of the variance of theR factor and conse-quently promises a correspondingly high accuracy ofanalysis. Fit parameters were the outermost five layer spingsdi ,i 11 ( i 51 –5), vertical bucklingsb3 andb5 within thethird and fifth layer, respectively, lateral pairing amplitudp2 and p4 between neighboring atoms in@001# directionwithin the second and fourth layer, respectively, as wellthe ~isotropic! vibrational amplitudesv i for the outermostthree layers (i 51 –3). The variational grid width for fit pa-rameters was usually set to 0.01 Å. All other parametwere kept fixed at their bulk values. In order to allow for aunbiased approach to the best-fit geometry the searchstarted with bulklike interatomic distances. As a consquence, it took three reference calculations to approachbest-fit structure and to stay always within the validity ran

8-3

er.

a

der

est

c-

tural

micea-so-

orr

n,d

n,fthhe

t-fian


of tensor LEED. The quality of agreement between expmental and calculated best-fit spectra is displayed in Figfor a selection of beams. The minimumR factor amounts toRP50.22(5), whereby both subsets of integer and fractionorder beams exhibit very similar fit qualities@RP

int

50.21(6),RPfrac50.23(6)#. Additionally, the ratior between

FIG. 2. Comparison of experimental and calculated besLEED-I~E! spectra for the missing-row reconstruction of clePt~110!.

16540

i-2

l

energy averaged intensities of fractional and integer orbeamsr 5^I & frac/^I & int , fits very well, too (r exp50.46,r calc

50.45).The reader might argue that theR-factor fit level is only

modest in view of values in the range 0.1–0.2~or evenlower! reported repeatedly in the literature~including contri-butions of our own group!. We point out, however, that thevalue achieved in the present work is still among the bever reached for a fcc~110! surface~see Ref. 68! and a sig-nificant improvement when compared to earlier LEED struture determinations of Pt(110)-(132) (RP50.36 in Ref. 37;Ref. 59 uses a noncomparable multi-R-factor average!. Thereason for the generally higherR factors for this class ofsurfaces seems to come from the spectra’s unusual strucrichness~extrema, shoulders, etc.! to which the PendryRfactor is extremely sensitive as it is based on the logarithderivatives rather than the mere intensity level. Another rson might be the high step density connected with the mescopic corrugated-iron structure of the Pt(110)-(132) sur-face which has been discussed in Refs. 36,38.

All structural best fit parameters together with their errmargins are summarized in Table I~the comparison to otheresults will be discussed in Sec. IV C!. Our fit reproduces thewell-known features of the missing-row reconstructiowhich is displayed in Fig. 3. Atoms of the first half occupielayer are strongly relaxed inward byDd12520.24 Å push-ing those directly below further into the bulk. This, in turleads to an extremely buckled third and slightly buckled filayer. Additionally, by this process neighboring atoms in t

t

lk

r

85

820885

TABLE I. Compilation of experimental and theoretical results for the 132 missing-row structure of cleanPt~110!. Ddi ,i 11 denotes the changes in the~average! inter-layer spacing with respect to the ideal buinterlayer spacingd0, while pi and bi denote the lateral pairing and buckling in layeri, respectively~allvalues in Å). Error limits for the parameters derived by the present LEED analysis~neglecting parametecorrelations! are 60.02 Å for Ddi ,i 11 , 60.05 Å for bi and 60.07 Å for pi . (L: number of layers con-sidered in calculations.!

Method Ref. Dd12 Dd23 Dd34 Dd45 Dd56 b3 b5 p2 p4 p6 d0

Experiment

LEED present 20.24 20.01 0.02 0.02 20.01 0.26 0.03 0.07 0.13 1.385LEED 37 20.28 20.01 0.02 0.01 0.01 0.17 0.03 0.08 0.10 1.3LEED 59 20.26 20.18 20.12 20.01 0.32 0.13 0.24 1.385MEIS 60 20.22 0.06 0.10 1.385XRD 61 20.27 20.11 0.10 0.08 1.385RHEED 62 20.37 0.07 0.18 0.08 1.385RHEED 63 20.34 20.01 0.12 0.09 1.385Theory

VASP-LDA-15L present 20.26 20.01 0.02 0.01 0.00 0.32 0.03 0.04 0.12 0.05 1.3VASP-GGA-15L present 20.27 0.00 0.03 0.01 0.01 0.33 0.04 0.06 0.13 0.03 1.4VASP-GGA-9L present 20.25 20.02 0.03 0.01 0.37 0.01 0.18 1.40FLEUR-GGA-9L present 20.23 0.00 0.02 0.00 0.37 0.01 0.19 1.40FLAPW 67 20.24 0.01 0.25 0.04 0.11 1.385Tight binding 64 20.11 0.02 20.03 0.00 0.04 1.385Emb. atom 65 20.25 20.07 0.01 0.11 0.05 0.08 1.385Emb. atom 70 20.26 20.06 0.23 0.09 1.385LO-MD 66 20.33 20.08 0.03 0.24 0.05 0.05 1.415

8-4

sughthrdoh

t

r-al-in

lyte

ean

eb

duea

-thlsca

erthemetra

us,theet-

iv-mi-to

in

el-atic

up-ccu-hellyace

x-nsf

osethe

to

ereturalen-tedot

herlk

eng

er

ith


second and fourth layer are squeezed a bit apart. This rein a lateral pairing of atoms within those layers. Thouthere are remarkable distortions within several layers,average interlayer distances below the second layer hadeviate from the bulk value. The vibrational amplitudesatoms in the first three layers are as expected, i.e., the higvalue applies to the outermost layer (v150.11 Å), the vi-brations in the following layers (v250.10 Å;v350.08 Å)approach the bulk value (vb50.07 Å). In agreement withcommon experience, however, the fit is not very sensitivethese parameters as reflected by error limits of60.04 Å forall three amplitudes.

B. DFT results for the unreconstructed 1Ã1and the 1Ã2 missing-row structure

The results of our structural optimization for the MR suface are given in the ‘‘theory’’ part of Table I. The theoreticinterlayer spacingsd0 used were obtained from bulk calculations performed by the respective methods. It is satisfyto realize that the GGA-d0 values obtained byVASP andFLEUR are almost identical, 1.408 and 1.405 Å, respectivewith only a slight overestimation of the lattice parame~experimental value 1.385 Å). On the contrary, theVASP

LDA-d0 value 1.382 Å is in very good agreement with thexperimental value. For all other calculated spacingspairings our two theoretical approaches,FLEUR and VASP,give also very similar results and are in excellent agreemwith the experimental results of the present work, as willdiscussed in more detail in the next section. Obviously,to the openness of the mr surface, the number of lay~thickness of the slab! considered in the calculations playsnon-negligible role. In fact, for the vertical bucklingb3 onenotices a considerable difference ('13 %) betweeen the results from the 9- and 15-layer calculation. Furthermore,pairings in the second and fourth layer, albeit small, aexperience considerable modifications. So, in the 9-layerculation p2 almost vanishes andp4 is by about 50 % largerthan the 15-layer value, the latter agreeing with our expmental result. We interpret these findings as follows: Inthicker slab, which models the actual semi-infinite systbest, more atoms can relax. Consequently, since the engained is distributed over a larger set of interlayer and in

FIG. 3. Geometry of the Pt(110)2(132) MR surface.

16540

lts

ely

fest

o

g

,r

d

nteers

eol-

i-e

rgy-

layer spacings, relaxations take place more gradually. Ththe rearrangement of atoms is found to extend down tosixth layer. In a thinner slab, as, e.g., for the present symmric 9-layer slab, relaxations can only affect four layers, ging rise to larger local changes. Clearly, both slabs give silar trends, but for the 9-layer slab the pairing is restrictedthe top region only, and therefore is larger.

The results for the unreconstructed surface are listedTable II. The 9-layer slab results byVASP andFLEUR, whichare compared in the first two lines of the table, are in exclent agreement with each other. For an even more systemcomparison of the methods,FLEUR was also run in the re-peated slab mode for the present surface, which further sports the equivalence of both packages with respect to aracy. Both methods predict an inward relaxation of ttopmost surface layer and of the third layer, only partiacompensated by an outward relaxation of the subsurflayer. Going to thicker slabs, investigated only byVASP,mainly subsurface relaxations are reduced, e.g.,Dd23 by0.03 Å. Obviously, there are no significant interlayer relaations deeper in the surface. Adding all vertical relaxatioyields for VASP-GGA a total surface compression o20.13 Å for the 9-layer slab and20.16 Å for the 15-layercalculation. These values are significantly smaller than thfor the reconstructed surface which can be understood bylatter’s greater openness and more degrees of freedomrelax.

C. Comparison and discussion of structural results

Table I summarizes our experimental andab initio struc-tural results for the Pt(110)-(132) MR structure~see Fig. 3!and compares them to those of earlier work. Evidently, this an amazingly close correspondence between the strucparameter values determined in the present work experimtally by quantitative LEED and theoretically by DFT. Mosof the values agree not only within the error limits estimatfor the LEED analysis but deviate from each other by nmore than 0.01 Å. The reader should note that thoughaver-age interlayer spacings below the second layer are ratbulklike, atoms within these layers are well off their bupositions due to buckling and pairing of atomic rows.

An exception of the almost perfect agreement betweLEED and DFT seems to hold for the third layer bucklin

TABLE II. Comparison of calculated changes in inter-layspacings of the metastable unreconstructed Pt(110)2(131) sur-face. Ddi ,i 11 denote the changes in the inter-layer spacing wrespect to the ideal bulk interlayer spacingd0 in Å. (L: number oflayers considered in present calculations.!

Method Ref. Dd12 Dd23 Dd34 Dd45 Dd56 d0

FLEUR-GGA-9L present20.22 0.14 20.05 1.405VASP-GGA-9L present20.23 0.15 20.05 1.408VASP-LDA-15L present20.22 0.11 20.03 0.00 0.00 1.382VASP-GGA-15L present20.23 0.12 20.04 20.01 0.00 1.408LCAO 77 20.16 0.07 20.02 1.377Tight binding 64 20.06 0.01 0.00 1.415Emb. atom 86 20.24 0.04 1.385

8-5

eiosifr

anep

reerefetheefoghe-b

h-

asite

sg

r-

uofec

t

ou

of

ienp-an,l-

fo

e

d-

henthe

y de-

ofge-n-thisrre-

e Pt

e

yerckling-llerult

ail-

.esorp-ast

g.,

tu-

bleax-forly

n-eanduceis

sandtothe

esa

g isinge.


which, though of the same order, differs by 0.06 Å betwethe two methods. Yet, even this deviation can in our opinbe traced back to a physical origin. While the DFT analydescribes a perfectly ordered (132) reconstructed surface oinfinite extension, the real sample exhibits a vast numbesteps as known through STM.36,38 Below a step edge thebuckled third layer has to match the unbuckled secondfourth layer of the neighboring terraces. Additionally, strelaxations as known from other stepped surfaces69 will besuperimposed and the overall structural matching mayduce the average buckling amplitude. In this light the agrment between LEED and DFT is rather good also withspect to the value ofb3. Interestingly, the LEED study oRef. 37 deviates even more from the DFT result with respto b3, possibly due to a different step density. Generally,different experimental LEED results listed in Table I agrrather well with each other except for those of Ref. 59,which discrepancy we have no explanation. Reflection hienergy electron diffraction~RHEED! seems to overestimatthe first layer contraction,62,63 possibly due to nonconsideration of p4. Small discrepancies with respect to results otained by medium-energy ion scattering~MEIS! ~Ref. 60!might also arise from neglecting some of the relaxations. Tx-ray diffraction ~XRD! results61 are close to the LEED result given that XRD is more~less! accurate with respect tolateral ~vertical! parameters than LEED.

We emphasize that the huge data base of our LEED ansis provides an unprecedented structural accuracy, poswith exception ofb3 for which DFT should give the correcvalue in case of the ideal surface. In general, the agreembetween DFT and LEED exhibits an impressive closenesthe structures independently retrieved, thus demonstratinday’s achievable accuracy at a 1022 Å level.

Quite surprisingly, only a few other theoretical results aavailable for the Pt(110)-(132) MR surface. The most recent work by Leeet al.67 employs theab initio FLAPWmethod in a very similar setup to ourFLEUR calculations, i.e.,LDA and a seven-layer~7L! thick single slab. Their resultsagree very well with present LEED data and also with oresults for the much thicker 15L slab. However, in viewthe discussion in the preceding section regarding the effof slab thickness, especially their value for the bucklingb3 isquite remarkable. We like to emphasize at this point, tha7L slab is too thin for the reliable calculation ofb3. In orderto rule out any ambiguities regarding the accuracy ofmethods, we recalculated a 7L Pt(110)-(132) MR slab withVASP using LDA and the same bulk lattice constant3.92 Å. Our results confirm the findings of Lee’s FLAPWcalculations, we obtainb350.27 Å andp450.12 Å. Thuswe conclude that this apparently excellent agreement wexperiment is due to the lack of the ‘‘bulk’’ in a 7L slab: thPt atom in the third layer below the surface Pt rows doeshave enough freedom to move into the ‘‘bulk’’ which is reresented by only the center layer of the symmetric slableads to a smaller buckling (b3) in this layer. Consequentlythe pairingp4 in the fourth layer is reduced since the repusive force due to this third layer Pt atom is now smallerthe atoms in the central~fourth! layer. This trend is alsonoticeable in our other calculations: ifb3 is smaller~15L vs

16540

nns

of

d

e---

cte

r-

-

e

ly-bly

ntofto-

e

r

ts

a

r

th

ot

d

r

9L!, so is p4. Comparing with other theoretical results onfinds the largest deviations for the tight binding~TB!approach,64 whereas the two calculations using the embeded atom method~EAM!—a bulk one65 and its extension tosurfaces70—yield reasonable results. Both the EAM and tTB approach are notab initio methods, but rather depend oexternal parameters which makes them less suitable forpresent purposes. Furthermore, these studies were mainlvoted to the structural stability of the Pt~110! surface, testingseveral models for the reconstruction by a comparisontotal energies. Despite its shortcomings in predicting theometry quantitatively, both methods find the MR recostructed surface to be the stable one. We mention incontext that our 15LVASP calculations of course also favothe MR structure. It’s energy is lower than that of the unconstructed surface by 228 meV~GGA! @250 meV~LDA !#.This is actually the energy gain when the removed surfacchains are incorporated into the bulk.

The only otherab initio study among those listed in TablI, the local-orbital molecular dynamics~LOMD!66 method,overestimates the compression of the first two interlaspacings and, as a consequence, underestimates the bub3. Although the LOMD method employs LDA, the calculated lattice parameter, which one would expect to be smathan the experimental value, is closer to our GGA reswhich exceeds the experimental value by'1.5 %.

To our knowledge there are no experimental data avable for the (131) structure of Pt~110!, though it is knownsince long that it can be prepared as a metastable phase71,72

However, regarding the preparation recipe, which involvCO adsorption and subsequent electron stimulated destion, the final cleanliness of such a surface remains at lequestionable. Another possibility is epitaxial growth on, e.a Pd~110! substrate73 which has a lattice constant only 0.8 %smaller than Pt and is known not to reconstruct. Unfornately the unreconstructed (131) structure only exists up toa Pt coverage of two monolayers, then the usual (132) re-construction sets in. The relaxations for the (131) structuredetermined by LEED,Dd12520.09 Å andDd2350.06 Å,are considerably smaller than our DFT results given in TaII, but nevertheless show the onset of an oscillatory relation profile. We also note that the relaxations obtainedthe (132) reconstructed 3 ML Pt film are considerabsmaller ~e.g., Dd12520.15 Å) than other results for the~110! surface of the Pt bulk. In view of the close correspodence between LEED and DFT results achieved for the clreconstructed surface, DFT can be safely assumed to procorrect results for the (131) structure, too. This statementreinforced by work on other (131) fcc~110! surfaces suchas Ni~110!,74 Cu~110!,75 or Pd~110! ~Ref. 76, and referencetherein!, where good agreement between experimentDFT calculations is found. According to these as well asour present results the unreconstructed surface showstypical oscillatory relaxation profile for open metal surfacwith a strong contraction of the first layer spacing andsmaller expansion for the second one. The third spacinalready almost bulklike due to effective electronic screenof the distortion introduced by the existence of the surfac

8-6

is

dluorwdis

-

errea

teer

r-F

a

don

ice

ngo

redonmedvaca

thtiothsiing

oav

EDithothPter-

atry

ae in

c-of

as

-ndledrt-

ar-e

ftheis

aceheen-rex--f

on,

ely

de

are

tochilltne.

-s isonore


Concerning earlier theoretical results for Pt(110)-(131)~see Table II!, to the best of our knowledge the onlyab initioDFT calculation performed to yield an optimized geometrywithin Feibelman’s work on surface stress on fcc~110!surfaces.77 The LDA calculation employs a LCAO methoand a 13L slab, to model the surface. The published vaare systematically smaller by about 30 %, but the oscillattrend of the relaxation is nicely reproduced. In the last trows theoretical data obtained by TB and EAM are listeNot fully unexpected, the optimum geometries obtained dagree again with our presentab initio data. Despite its approximations, EAM still finds aDd12 value very similar toours. Yet it fails in predicting any reasonable further intlayer relaxation. Concerning the TB approach, we only agon the sign of the relaxations. However, their amplitudesdetermined by TB are smaller by at least a factor of 4.

Eventually, the reader should note that the average inlayer spacingd12 of the MR phase is very close to the samquantity of the unreconstructed surface. Some differencesgardingd23 andd34 obviously arise from bucklings and paiings in the reconstructed surface. In the next section the Danalysis of the clean Pt(110)-(131) surface will serve as areference to understand the structural impact of brominesorption.

IV. COVALENT BONDING OF Br ON Pt „110…:THE STRUCTURE OF THE c„2Ã2…-Br ÕPt„110… SURFACE

In a previous publication16 part of the authors combinevarious experimental observations to deduce a substitutiadsorption site of bromine on Pt~110!: First, only a minorchange of the work function (Dfmax5100 meV) and hencea small dipole moment was observed, in contrast to typhalogen adsorption induced changes in the eV region. Sond, bromine atoms very dilutely adsorbed on the missirow structure were identified as depressions in STM. Sappeared reasonable to transfer these results to thec(232)phase, where a pattern of alternating protrusions and depsions with practically the same corrugation is observThen, however, a bromine adsorption site considerably abthe platinum surface plane cannot lead to a depressioSTM without an enormous charge transfer which is incopatible with the small surface dipole moment observThird, substitutional adsorption would have allowed remoof the missing-row reconstruction by means of a purely lomass transport consistent with the observation ofc(232)structural elements already atT5420 K. A further observa-tion was the remarkable stability of Pt-Br-Pt chains onc(232) surface. Consequently, a substitutional adsorpof Br had been postulated consisting of Pt-Br-Pt rows atsurface. All other experimental features could also be contently interpreted in the framework of such an alternatchain model.

A. LEED analysis

For the reasons mentioned above, the LEED analysis ccentrated at first on the substitution model. In order to sits local C2v symmetry@see Fig. 4~a!# only vertical coordi-

16540

esyo.-

-es

r-

e-

T

d-

al

alc--it

es-.

vein-.ll

enes-

n-e

nates of atoms were varied. Using several Tensor LEreference calculations the relative position of Br atoms wrespect to the Pt atoms of the first layer was tested for binward and outward relaxations. Additionally, the firstsublayer and the next four layers were allowed to shift vtically. However, none of these configurations achievedPendryR factor better than 0.73. Also, an imagined regisshift of the first layer by half a lattice parameter@Fig. 4~b!#did not improve the result (RP50.80). The same holds formodel with the bromine substitution assumed to take placthe third instead of the first layer@Fig. 4~c!# and so inducinga buckling in the top Pt layer which could alternatively acount for the STM contrast observed. Yet, no combinationparameters produced anR factor below 0.77.

Therefore, the remaining class of adsorption models wexamined, i.e., bromine atoms adsorbedon the [email protected]~d!–4~g!#. In a preliminary structural search, the high symmetry adsorption sites on-top, hollow, as well as long- ashort-bridge were tested. While the first three models alsoto R factors not lower than 0.77, the analysis of the shobridge site instantaneously produced a PendryR factor aslow as 0.27. An additional fine-tuning of parameters, in pticular lateral shifts and vibrations within this short-bridgmodel, further reduced theR factor to a final value ofRP

50.23(1). Interestingly, theR factors for both subsets ointeger and fractional order beam spectra are exactlysame. Also, practically the same quality of agreementachieved as for the missing-row structure of the clean surfas visualized in Fig. 5. However, a peculiar feature of tLEED analysis should also be mentioned: The ratio ofergy averaged intensitiesr between fractional and integeorder beams is much higher in the calculations than inperiment (r calc50.44,r exp50.12). We attribute this to the instability of the c(232) structure against minor traces ocontaminants originating from the filament of the electrgun ~see Sec. IV C!. With regard to our structural resultshowever, we emphasize at this point that it is rather unlikto achieve a LEED fit quality as good as that above (RP50.23) with all other models failing at the level describe(RP.0.73). This makes us confident in the validity of thshort-bridge model.

The structural parameters of the best-fit geometrycompiled in Table III. From the spacingdBr between thebromine overlayer and the first platinum layer~together withthe small lateral shiftp1,2 of the Pt atom! a Br-Pt bond lengthof L52.47 Å results. This corresponds almost perfectlythe sum of the covalent radii of Pt and neutral Br, whiamounts toLatomic52.495 Å. The underlying substrate is stslightly relaxed with a 0.10 Å contraction of the first Pinterlayer spacing and a 0.03 Å expansion of the next oCompared to the clean (131) structure of Pt~110! as derivedfrom DFT ~see Sec. III! this is equivalent to a roughly halfway derelaxation induced by the bromine adsorption. Thia typical feature known from many other adsorptisystems.68 The lateral shifts found in the analysis are minand insignificant in view of the error limits involved. Thvibrational amplitudes~not included in Table III! are largestfor the Br atoms (v050.1560.03 Å) and still somewhat

8-7

e


FIG. 4. Survey of the models tested in thLEED-I~E! analysis for thec(232)-Br/Pt(110)structure.

te

ththiliagene

oth

sidni

e

-f

d-calstt to

eof

y,Br

-d

rornt

increased in the uppermost two Pt layers (v150.1160.04 Å, v250.1260.05 Å) compared to the substrabulk.

B. DFT analysis

In order to gain a deeper insight into the energetics ofadsorption model, we performed DFT calculations usingsame setup as for the clean surface in Sec. III. The stabof a particular adsorption site is determined by the averadsorption energy per Br atomEads, defined as the differencin formation energies of ac(232) Br covered and a clea(131) Pt~110! 9L slab. The adsorption energies togethwith the corresponding PendryR factors are listed in TableIV. The calculations show unambigously that adsorptionthe short bridge site is favored, in perfect agreement withLEED analysis.

The energetical ordering of the adsorption sites conered is independent of whether GGA or LDA is used, aeven the least favorable direct hollow Br adsorption sitemore stable by;0.7 eV than a substitutional model. Th

16540

eetye

r

ne

-ds

absolute value ofEadsLDA is always larger than that ofEads

GGA

reflecting the well known tendency of LDA to overbind. Another tendency of LDA clearly visible is its favoring ohighly coordinated sites.Eads

LDA is larger thanEadsGGA for the top

site by 0.58 eV, by 0.69~71! eV for the short~long! bridgeand by 0.77 eV for the hollow site.

The optimum relaxed geometry for the favorable Br asorption on the short bridge site using different theoretisetups is given in Table III. It is interesting to note that moof the relaxations take place near the surface, in contrasthe clean~reconstructed! surface where relaxations involvatoms down to the fourth layer. Therefore the thicknessthe slab~9L vs 15L! is of only minor importance. Similar tothe clean surfaces, differences betweenVASP and FLEUR re-sults are well within the experimental error limits. Actuallthe largest differences are found for the distance betweenand the Pt~110! substrate using either LDA or GGA potentials. Compared to GGA, the tendency of LDA to overbinleads to smaller values ofdBr by 0.05–0.07 Å, the differ-ence being only slightly larger than the experimental ermargin of 60.03 Å. It should be noted that the prese

8-8

llon

eo

ghaae

asPt

nalge

oftoth

gex-

s

ndbe-ess

con-nthat

tive

t-fi

he


c(232)-Br/Pt(110) overlayer and ap(231)-Br/Pt(110)overlayer with Br in the short-bridge sites are energeticadegenerate within the error limits of the present calculatithis is consistent with our previous experimental result.16

C. Discussion

In the following we first address the reliability of thpresent structure determination and compare the resultstained experimentally by quantitative LEED and throufirst-principle calculations applying DFT. Then we show ththe main experimental results of Ref. 16, namely, the smvalue of the work function change along with the observ

FIG. 5. Comparison of experimental and calculated besLEED-I (E) spectra forc(232)-Br/Pt(110).

16540

y;

b-

tlld

negative constrast of Br in STM for low coverages as wellthe nonlocal mass transport and the stability of Br-bridgedchains—which had been interpreted by a substitutiomodel—can easily be reconciled with the new short-bridmodel.

~i! Reliability of the structure determination. The resultsof the experimental and theoretical structure determinationthe c(232)-Br/Pt(110) surface not only exhibit excellenquantitative agreement as apparent from Table III. They bappear to be on very safe grounds on their own.

Concerning LEED, only the adsorption model involvinthe short bridge site produced a convincing fit betweenperiment and model calculations (RP50.23). TheR factorsof all other models~substitutional or other adsorption site!were not only much above theR-factor variance level butwith RP.0.73 approaching the range where calculated aexperimental intensity spectra become uncorrelated. Thishavior is due to the extreme and unusual structural richnof the spectra as visualized in Fig. 5~a feature applying alsoto the data of the clean surface!. Only the correct model canreproduce the many peaks over the large energy rangesidered (DE54586 eV), a width which is not routine evein today’s structure analyses. The reader should also notenot only the peak positions~on which the PendryR factorfocuses! are reproduced but to a high degree also the relaintensities within each spectrum.

t

TABLE IV. Atomic bromine adsorption energiesEads for differ-ent Br/Pt~110! adsorption models calculated withVASP using bothGGA and LDA potentials. The corresponding best-fit PendryR fac-torsRP of the LEED analysis are also included for comparison. Tfirst four models refer to direct adsorption of Bron the Pt~110!surface. The last line refers to the substitutional model@Fig. 3~a!#originally suggested in Ref. 16.

model EadsGGA~eV! Eads

LDA~eV! RP

c(232) short bridge 23.28 23.97 0.23c(232) long bridge 23.13 3.84 0.77c(232) top 22.92 23.50 0.78c(232) hollow 22.56 23.33 0.77c(232) substitution 21.80 22.62 0.73

e
TABLE III. LEED and DFT results for the Br/Pt(110)2c(232) phase.dBr denotes the Br-substratdistance,Ddi ,i 11 the changes in the interlayer spacing with respect to the ideal bulk spacingd0, andpi thelateral pairing in layeri, all values given in Å. Signs of the lateral pairing parameterspi are given withrespect to the position of the Br atoms. Error limits for the parameters derived by LEED~neglecting param-eter correlations! are60.03 Å for dBr andDdi ,i 11 , 60.08 Å for p1 and60.10 Å for p2. (L: number oflayers considered in the calculations.!
Method dBr Dd12 Dd23 Dd34 Dd45 p1 p2 p3 p4 d0

LEED 2.04 20.10 0.03 0.01 0.00 0.04 20.04 1.385VASP-LDA-15L 2.02 20.12 0.03 0.01 0.01 0.01 20.01 0.02 20.01 1.382VASP-GGA-15L 2.09 20.11 0.03 0.02 0.01 0.01 20.04 0.02 20.03 1.408VASP-LDA-9L 2.04 20.11 0.02 0.03 0.00 20.02 0.01 0.00 1.382VASP-GGA-9L 2.09 20.11 0.02 0.02 0.01 20.04 0.01 20.02 1.408FLEUR-GGA-9L 2.11 20.12 0.01 20.01 0.05 20.06 0.01 0.00 1.405

8-9


TABLE V. Metal-halogen bondlength, metal-halogen layer distancez and work function changeDF.

System Adsorption site Bond length z DF ~meV! Refs.

Br/Pt(110)2c(232) short bridge 2.47 Å 2.05 Å 100 this workCl/Pt(110)2p(231) short bridge 2.38 Å 1.93 Å 550 6,87Cl/Ag(100)2c(232) fourfold hollow 2.6122.69 Å 1.6221.75 Å 1700 24,31,88Cl/Ag(111)2A33A3R30° fcc site 2.4822.70 Å 1.8322.12 Å 1400 20,89–91Cl/Ag(110)2p(231) fourfold hollowa 2.56 Å 0.53 Å 1050 21,92Cl/Cu(100)2c(232) fourfold hollow 2.3722.41 Å 1.5421.60 Å 1100 18,25,26Cl/Cu(111)2A33A3R30° fcc site 2.39 Å 1.88 Å 900 30,27,93I/Ag~110! short bridge 3.11 Å 2.75 Å 1000 92,28I/Cu(100)2p(232) fourfold hollow 2.69 Å 1.99 Å 94,95Br/Ag(110)2p(231) 1000 8Br/Cu(100)2c(232) 900 96Cl/Ni(110)2c(232) 1100 6Cl/Pd(110)2c(232) 1200 6Cl/Ni(111)2A33A3R30° 380 97

Cl/Pd(111)2A33A3R30° 260 96Cl/Pt(111)2333 100 85Br/Pd(111)2A33A3R30° 350 98Br/Pt(111)2333 80 10

aAssumed.

edi

a

ctee

sisttiit

hee

s

od

potr

thtse

idntceD

redots.n.ined

-elling

hees

Theap-b-

erala-r

tilled

e

l-edms2n,ural-in

As mentioned in Sec. IV A the ratio of energy averagintensitiesr between fractional and integer order beamsmuch higher in the calculations than in experiment (r calc50.44, r exp50.12). The simplest explanation would be thpart of the surface exhibits (131) rather thanc(232) or-der, consistent with the fact that relative intensities of spewithin the subsets of beams are well reproduced. Yelectron-induced desorption can be ruled out from testperiments and the occurrence of (131) domains is in con-trast to the STM observations. Alternatively and also content with relative intensities measured for beam subsets,low r value can be interpreted as being due to substanintrinsic disorder. Again, this seems to be at variance wSTM investigations@see Ref. 16 and Fig. 1~a!# which findlarge and well orderedc(232) domains in agreement witthe observation that the half-width of all LEED spots is dtermined by the transfer width of the optics. On the othhand, it is also observed that adsorption of small amounthalogens, NO and CO quickly destroys thec(232) structure@eventually leading to a (331) superstructure78–80#. Yet, thisdoes not destroy the coherence between the remainingdered adatoms, so the half-width of fractional order spremains constant though their intensity decreases. Thiscrease should be more rapid than that of integer order sbecause the ordered bulk below the disordered parts ofoverlayer also contributes to the integral spots. The distuing CO molecules seem to originate from the filament ofelectron gun as checked by time dependent measurementhe intensity of an integral and a fractional spot at a fixelectron energy displayed in Fig. 1~d!. Evidently, the inten-sity of the fractional order spot decreases much more rapThis result suggests that the low intensity of the experimefractional order spots may be attributed to disorder induby traces of CO originating from the filament of the LEE

16540

s

t

rat,x-

-healh

-rof

or-tse-ts

heb-e

ofd

ly.ald

gun whereby, however, there are still large and well ordec(232) domains accounting for sharp superstructure sp

The DFT results are equally convincing on their owThere are only little differences between parameters obtaby all-electron full potential augmented plane wave~FLEUR!and projector augmented wave~VASP! methods and by comparing the 9L and 15L results the calculations appear wconverged with respect to the slab thickness. Similarly, usa free standing single slab model~FLEUR! or a repeated slabmodel ~VASP! makes no difference. Last but not least tcomparison to the LEED result is excellent, which provthat theab initio DFT packagesFLEUR and VASP representone of the most accurate approaches available today.almost quantitative agreement between LEED and DFTplies in particular to all relaxational parameters of the sustrate Pt~110! which agree well within the error limits. Eventhe change of sign and also the magnitude of the small latpairings p1 and p2 are nicely reproduced. There is onlysmall deviation (;0.05 Å) concerning the GGA calculations of dBr which is slightly outside the experimental errolimits. However, this overestimated Pt-Br bond-length is swithin the expected GGA error limits and is certainly alterby an improved GGA functional.

~ii ! Low work function change. The measured changeDFof the work function caused by the transition from thPt(110)-(132)MR to the c(232)-Br/Pt(110) surface isDF'70 meV.16 The present DFT calculations yield an amost negligible work function change too. This is in markcontrast to halogen-Cu or halogen-Ag adsorption syste~see Table V! with work function changes of the order of 1–eV.81 While, in the absence of direct structural informatiosubstitutional adsorption seemed to provide the most natexplanation,16 the DFT calculation and LEED results demand an alternative interpretation. Clearly, the alternative

8-10

tofoe

r

lor ahph

fpao

,he

thhath

in

nm

r

theif-ed

6.

rtsin

rm

ndser

-engly

lossto

r.de-edsinsPt

lckedort-ht-orlledays

di-notyo ane-on-

hede-ro-e

el,

iasntal

etaneorpo


this case is to attribute the low work function changecovalent instead of ionic bonding usually assumedhalogen-metal bonding. The covalent bond character seto be specific for Pt~and—to a lesser extent—the otherd9

transition metals! as indicated in Table V; both Cl and Bcause anomalously low work function changes on Pt~111! aswell. Figure 6 shows a plot of the dipole moment per hagen atom versus the distance between the halogen layethe first metal layer, for different halogen-metal systems. Tresulting dipole moments depend on the substrate and apto be grouped around a distinct value for each metal: Tdipole moment is larger for Ag (p50.7D) than for Cu (p50.35D), whereas for Br/Pt~110! we find an anomalouslysmall value ofp50.04D. For Ag and for Cu the variation othe dipole moment with surface orientation and halogen scies, and hence the halogen-metal layer distance, is smThe comparatively large difference between the dipole mment of Cl/Pt~110! and Br/Pt~110! seems to be not in linebut the Cl data are taken from an early measurement, wCl was dosed by exposure to ambient Cl2 atmosphere.6

Qualitatively the difference in dipole moments is in line withe electronegativity difference between Cl and Br. Note tthe dipole moments given here refer to coverages inmonolayer range. Depolarization effects are not takenaccount. In summary, for Br/Pt~110! we can exclude an ionicbonding model. This corroborates our previous argumefor covalent bonding and a covalent poisoning mechanis82

which were based on photoemission results.~iii ! Local versus nonlocal mass transport. In Ref. 16 the

substitutional adsorption model was favored, because itquired only local platinum transport to form ac(232) struc-

FIG. 6. Dipole moment per halogen atom versus halogen-mlayer distance. The symbols represent different experimentaltheoretical results for the adsorption of Cl, Br, and I on Ag, Cu, aPt surfaces~see Table V!. Error bars indicate the scatter in threported values for the bond length. Typical errors in the wfunction change measurements result in an uncertainty of the dimoment of60.05D. Br/Pt~110!: this work, Cl/Pt~110!~Refs. 6,87!,Cl/Ag(111) ~Refs. 20,89–91!, Cl/Ag(100) ~Refs. 24,31,88!,Cl/Ag(110) ~Refs. 21,92!, Br/Ag~110! ~Ref. 8! ~no structural data!,I/Ag~110! ~Refs. 92,28!, Cl/Cu(111) ~Refs. 30,27,93!, Cl/Cu(100)~Refs. 18,25,26!, Br/Cu~100! ~Ref. 96! ~no structural data!;I/Cu~100! ~Refs. 94,95!.

16540

rms

-ndeeare

e-ll.-

re

te

to

ts

e-

ture. Within the present short-bridge adsorption model,mass transport lifting the missing-row reconstruction is dferent, but still rather local and analogous to the CO induclifting of the Pt(110)-(132) missing-row reconstruction.83

It is entirely consistent with the experimental data of Ref. 1At room temperature, Br forms a (232) overlayer on the(132) missing-row reconstructed Pt~110! surface.16 Forma-tion of vacancy islands within the topmost Pt layer staafter annealing to 370 K. The displaced Pt atoms residethe troughs of the remaining missing-row areas and focrosslike structures or a square grid pattern.16 At succes-sively higher annealing temperatures the vacancy islagrow and the Pt atom density in the remaining first laytroughs increases until locally a (131) structure prevails. Atthis stage~annealing to 420 K! the surface consists of irregularly shaped (131) domains of partly the first and partly thsecond Pt layer. On a mesoscopic scale the pattern is stroreminiscent to the one formed by CO on Pt~110!.83 The ad-sorbed Br forms mixedp(231) and c(232) overlayerstructures on both, the first and the second layer (131) do-mains. AtQ50.5 ML Br desorbs only atT5800 K. Con-sequently, the surface can be annealed at 780 K withoutof Br. At this temperature the mobility is high enoughallow formation of large, almost perfectly developedc(232)-Br/Pt(110) terraces@see Fig. 1~a!#.

~iv! Stability of the Br-bridged Pt chains. By heating to800 K thec(232) structure is object to a slight loss of BAs a result, a missing-added-row reconstruction starts tovelop. The single added rows exhibit the typical Br induccorrugation.16 In Refs. 16 and 5 the stability of these rowwas interpreted as a tendency to form linear Pt-Br-Pt chain analogy to the quasi-one-dimensional halogen-bridgedlinear chain compounds.4 In the light of the present structuramodel the added rows have to be interpreted as close-paPt rows bearing an adsorbed Br atom on every second shbridge site. The implications are twofold. First, even a sligsubstoichiometry of thec(232) structure triggers an instability towards a missing-row reconstruction. Note that fevery Br atom lost, on average seven Pt atoms are expefrom the surface layer. Second, the added rows are alwfully occupied by Br atoms in every second bridge site incating a peculiar stability of these Br-bridged Pt chainsonly in thec(232) or p(231) structure, but also as solitarchains on flat terraces. Both observations clearly point tsubstantial anisotropy, that is to say a quasi-odimensionality, of the Br-bridged Pt chains as already ccluded in Refs. 16 and 5.

~v! STM contrast. Figure 1~a! shows an STM image of thec(232)-Br/Pt(110) structure. From this image neither tadsorption site nor the chemical contrast of Br can beduced. DFT calculations indicate that Br is imaged as ptrusion @see Fig. 1~b!#. Constant current STM images havbeen calculated on the basis of the Tersoff-Hamann mod84

which yields a tunneling currentI (z) proportional to thesample’s local density of states~LDOS! rS(z,E) at the posi-tion z of the tip above the surface. Only states of energyEsatisfying resonant tunneling conditions for the applied bvoltage are considered. Simulating the present experimeSTM image the LDOSrS(z,E) was integrated betweenE

alndd

kle

8-11

nt

plun

eepeinBTat

nedn

ansacnsavio

t tolso

etry

ter-

allyad-insis

b-tion

al-ngin-is-

hellysly

b-of

d.


5EF and E5EF1820 meV. The isosurface of constaLDOS ~current! displayed in Fig. 1~b! shows a corrugationof 0.25 Å close to the experimental findings. The tip-samdistance leading to this corrugation is calculated to be aro6 Å with respect to the top Pt layer (;4 Å with respect tothe Br!, quite typical for metallic surfaces. Hence, in thSTM images of thec(232) structure the bright spots havto be interpreted as Br atoms and the dark spots as emshort-bridge sites. For small Br coverages STM imagshowed depressions in the close-packed rows on the missrow reconstructed surface. They were interpreted asdimers.16 New results, of both, STM experiments and DFcalculations, however, question this interpretation. Henceditional experiments are presently underway in orderclarify the adsorption structure at low coverage.

V. SUMMARY

We have carried out a structure analysis of thec(232)-Br/Pt(110) surface by quantitative LEED analysis aby first principles DFT calculations. The latter were carriout by using a full-potential linearized augmented plawave method as implemented in theFLEUR code,51 as well asa pseudopotential projector augmented wave method~VASP

code!.52 In addition, the missing-row reconstructed clePt~110! surface was reanalyzed and compared to previoureported results. The LEED analysis was carried out withextraordinary broad database and shows a remarkablevergence with the results of the first-principles calculatioThis proofs that state-of-the-art calculational methods hmatured to a complete tool-box for structure determinat

v.

r

J.

e

16540

ed

tysg-r

d-o

d

e

lyn

on-.en

rivaling present-day experimental methods with respecprecision and reliability. Encouraged by this result, we acalculated the geometry of the unreconstructed (131)Pt~110! surface in order to assess changes in the geomfor adsorption systems where the reconstruction is lifted.

As expected, only the precision of earlier structure deminations was improved for the clean (132) MR recon-structed Pt~110! surface. For thec(232)-Br/Pt(110) sys-tem, however, the previous structural assignment16 had to berevised: Instead of substitutional adsorption with essenticolinear Pt-Br-Pt chains the present study finds simplesorption with every second short-bridge site on the Pt chabeing occupied by Br. The missing-row reconstructionlifted. As usual, the first interlayer contraction in the sustrate is seen to be significantly reduced in the adsorpsystem. Remarkably, the Br-Pt bonding distance amountsmost precisely to the sum of the covalent radii indicatiessentially covalent bonding. A covalent bonding is alsodicated by the experimental work function and photoemsion data.16,82

Finally, theoretical STM images were calculated on tbasis of the Tersoff-Hamann formalism. They unequivocashow Br imaged as protrusion in contrast to the previougiven interpretation of the STM images.16 The present studyprovides therefore a further example for the difficulty to otain reliable structure assignments solely on the basisSTM and qualitative LEED measurements.

ACKNOWLEDGMENT

This work was supported by the Austrian Science Fun

s.

f,

,

le,l.

D.

T.

v. B

ate

.

*Also at: Istituto Nazionale di Fisica della Materia~INFM!, Dipar-timento di Scienze Fisiche, Universita` degli Studi di CagliariI–09124 Cagliari, Italy.1H.F. Winter and J.W. Coburn, Surf. Sci. Rep.14, 161 ~1992!.2R.G. Jones, Prog. Surf. Sci.27, 25 ~1988!.3T.W. Fishlock, J.B. Pethica, and R.G. Egdell, Surf. Sci.445, L47

~2000!.4M. Alouani, J.M. Wilkins, R.C. Albers, and J.M. Wills, Phys. Re

Lett. 71, 1415~1993!.5K. Swamy, A. Menzel, R. Beer, and E. Bertel, Phys. Rev. Lett.86,

1299 ~2001!.6W. Erley, Surf. Sci.114, 47 ~1982!.7B. Kruger and C. Benndorf, Surf. Sci.177, 515 ~1986!.8C. Benndorf and B. Kru¨ger, Surf. Sci.151, 271 ~1985!.9H.-G. Zimmer, A. Goldmann, R. Courths, and H. Saalfeld, Su

Sci. 173, 465 ~1986!.10E. Bertel, K. Schwaha, and F. Netzer, Surf. Sci.83, 439 ~1979!.11E. Bertel and F. Netzer, Surf. Sci.97, 409 ~1980!.12T.W. Fishlock, J.B. Pethica, F.H. Jones, R.G. Egdell, and

Foord, Surf. Sci.312, 629 ~1997!.13H. Xu, R. Yuro, and I. Harrison, Surf. Sci.411, 303 ~1998!.14C.Y. Nakakura and E.I. Altman, Surf. Sci.398, 281 ~1998!.15T.W. Fishlock, J.B. Pethica, A. Oral, R.G. Egdell, and F.H. Jon

Surf. Sci.426, 212 ~1999!.16K. Swamy, P. Hanesch, P. Sandl, and E. Bertel, Surf. Sci.466, 11

~2000!.

f.

S.

s,

17P. Citrin, P. Eisenberger, and R.C. Hewitt, Phys. Rev. Lett.45,1948 ~1981!.

18P.H. Citrin, D.R. Hamann, L.F. Mattheiss, and J.E. Rowe, PhyRev. Lett.49, 1712~1982!.

19M. Crapper, C. Riley, P. Sweeney, C. McConville, D. Woodrufand R. Jones, Europhys. Lett.2, 857 ~1986!.

20G. Lamble, R. Brooks, S. Ferrer, D.A. King, and D. NormanPhys. Rev. B34, 2975~1986!.

21D.J. Holmes, N. Panagiotides, R. Dus, D. Norman, G.M. LambC.J. Barnes, F. Della Valla, and D.A. King, J. Vac. Sci. TechnoA 5, 703 ~1987!.

22G. Lamble, R.S. Brooks, J.-C. Campuzana, D.A. King, andNorman, Phys. Rev. B36, 1796~1987!.

23M. Kiguchi, T. Yokoyama, S. Terada, M. Sakano, Y. Okamoto,Ohta, Y. Kitajima, and H. Kuroda, Phys. Rev. B56, 1561~1997!.

24E. Zanazzi, F. Jona, D.W. Jepsen, and P.M. Marcus, Phys. Re14, 432 ~1976!.

25D. Westphal, A. Goldmann, F. Jona, and P.M. Marcus, Solid StCommun.44, 1712~1982!.

26F. Jona, D. Westphal, A. Goldmann, and P. Marcus, J. Phys. C16,3001 ~1983!.

27K. Doll and N.M. Harrison, Chem. Phys. Lett.317, 282 ~2000!.28H.-R. Tang, W.-N. Wang, Z.-H. Li, W.-L. Dai, K.-N. Fan, and

J.-F. Deng, Surf. Sci.450, 133 ~2000!.29K. Kleinherbers and A. Goldmann, Fresenius’ Z. Anal. Chem

329, 285 ~1987!.

8-12

.

urno

ys.

na,

.L.

hys.

em.

tel,

rtel

tion

U.

ett.

S.

urf.

nn,


30D. Woodruff, D.L. Seymour, C.F. McConville, C.E. Riley, M.DCrapper, N.P. Prince, and R.G. Jones, Phys. Rev. Lett.58, 1460~1987!.

31F. Jona and P. Marcus, Phys. Rev. Lett.50, 1823~1983!.32K. Swamy, E. Laegsgard, and F. Besenbacher~unpublished!.33Rasterscope 3000, DME, Herlev, Denmark.34SPM 100, RHK Technology, Michigan, USA.35ErLEED 3000, Specs, Berlin, Germany.36P. Hanesch and E. Bertel, Phys. Rev. Lett.79, 1523~1997!.37P. Fery, W. Moritz, and D. Wolf, Phys. Rev. B38, 7275~1988!.38K. Swamy, E. Bertel, and I. Vilfan, Surf. Sci.425, L369 ~1999!.39SensiCam, PCO Computer Optics, Kelheim, Germany.40V. Blum and K. Heinz, Comput. Phys. Commun.134, 392~2001!.41P. Rous, J. Pendry, D. Saldin, K. Heinz, K. Mu¨ller, and N. Bickel,

Phys. Rev. Lett.57, 2951~1986!.42P. Rous and J. Pendry, Surf. Sci.219, 355 ~1989!; 219, 373

~1989!.43P. Rous, Prog. Surf. Sci.39, 3 ~1992!.44K. Heinz, Rep. Prog. Phys.58, 637 ~1995!.45K. Heinz, M. Kottcke, U. Loffler, and R. Do¨ll, Surf. Sci.357-358,

1 ~1996!.46M. Kottcke and K. Heinz, Surf. Sci.376, 352 ~1996!.47J. Pendry, J. Phys. C13, 937 ~1980!.48J. Rundgren, Phys. Rev. B59, 5106~1999!.49S. Muller, Ph. D. thesis, University Erlangen-Nu¨rnberg, 1996.50M. Weinert, E. Wimmer, and A. Freeman, Phys. Rev. B26, 4571

~1982!.51URL: http://www. flapw.de.52G. Kresse and J. Furthmu¨ller, Comput. Mater. Sci.54, 11 169

~1996!.53P.E. Blochl, Phys. Rev. B50, 17 953~1994!.54G. Kresse and D. Joubert, Phys. Rev. B59, 1758~1999!.55Y. Wang and J.P. Perdew, Phys. Rev. B44, 13 298~1991!.56J.P. Perdew and A. Zunger, Phys. Rev. B23, 5048~1981!.57D.M. Ceperley and B.J. Alder, Phys. Rev. Lett.45, 566 ~1980!.58S.L. Cunningham, Phys. Rev. B10, 4988~1974!.59E.C. Sowa, M.A. van Hove, and D.L. Adams, Surf. Sci.199, 174

~1988!.60P. Fenter and T. Gustafsson, Phys. Rev. B38, 10 197~1988!.61E. Vlieg, I. Robinson, and K. Kern, Surf. Sci.233, 248 ~1990!.62M. Stock, J.R.U. Korte, and G. Meyer-Emsen, Surf. Sci.233,

L243 ~1990!.63U. Korte and G. Meyer-Ehmsen, Surf. Sci.271, 616 ~1992!.64H.-J. Brocksch and K. Bennemann, Surf. Sci.161, 321 ~1985!.65S. Foiles, Surf. Sci.191, L779 ~1987!.66M.-H. Tsai and K. Hass, Phys. Rev. B52, 16 420~1995!.67J.I. Lee, W. Mannstadt, and A.J. Freeman, Phys. Rev. B59, 1673

~1999!.68P. Watson, M. V. Hove, and K. Hermann, NIST Surface Struct

Database - Ver. 3.0, National Institute of Standards and Techogy, Gaithersburg, MD, 1999.

16540

el-

69S. Walter, H. Baier, M. Weinelt, K. Heinz, and T. Fauster, PhRev. B63, 155407~2001!.

70M. Haftel, Phys. Rev. B48, 2611~1993!.71C. Comrie and R. Lambert, J. Chem. Soc., Faraday Trans.72,

1659 ~1976!.72S. Ferrer and H. Bonzel, Surf. Sci.119, 234 ~1982!.73O.L. Warren, H. Kang, P.J. Schmitz, P.A. Thiel, P. Kaukasoi

and M. Lindroos, Phys. Rev. B47, 10 839~1993!.74F. Mittendorfer, A. Eichler, and J. Hafner, Surf. Sci.423, 1

~1999!.75J. Redinger, P. Weinberger, H. Erschbaumer, R. Podloucky, C

Fu, and A.J. Freeman, Phys. Rev. B44, 8288~1991!.76V. Ledentu, W. Dong, P. Sautet, G. Kresse, and J. Hafner, P

Rev. B57, 12 482~1998!.77P.J. Feibelman, Phys. Rev. B51, 17 867~1995!.78C. Deisl, K. Swamy, S. Penner, and E. Bertel, Phys. Chem. Ch

Phys.3, 1213~2001!.79K. Swamy, A. Menzel, R. Beer, C. Deisl, S. Penner, and E. Ber

Surf. Sci.482-485, 402 ~2001!.80K. Swamy, C. Deisl, A. Menzel, R. Beer, S. Penner, and E. Be

~unpublished!.81For Cl and Br on Pt~111! a sign reversal of the work function

change is observed at low coverages. Substitutional adsorpwas proposed as a possible explanation~Refs. 10,85!.

82A. Menzel, K. Swamy, R. Beer, P. Hanesch, E. Bertel, andBirkenheuer, Surf. Sci.454-456, 88 ~2000!.

83T. Gritsch, D. Coulman, R.J. Behm, and G. Ertl, Phys. Rev. L63, 1086~1989!.

84J. Tersoff and D.R. Hamann, Phys. Rev. B31, 805 ~1985!.85W. Erley, Surf. Sci.84, 281 ~1980!.86S. Foiles, M. Baskes, and M. Daw, Phys. Rev. B33, 7983~1986!.87C. Deisl, K. Swamy, E. Bertel, C. Franchini, J. Redinger,

Walther, L. Hammer, and K. Heinz~unpublished!.88K. Kleinherbers, E. Janssen, A. Goldmann, and H. Saalfeld, S

Sci. 215, 394 ~1989!.89A. Shard and V.R. Dhanak, J. Phys. Chem. B104, 2732~2000!.90K. Doll and N.M. Harrison, Phys. Rev. B63, 165410~2001!.91K. Kleinherbers, Ph. D. thesis, Duisburg, 1987.92H. Hinzert, K.K. Kleinherbers, E. Janssen, and A. Goldma

Appl. Phys. A: Mater. Sci. Process.49, 313 ~1989!.93K. Kleinherbers and A. Goldmann, Surf. Sci.154, 489 ~1985!.94P. Citrin, P. Eisenberger, and R.C. Hewitt, Phys. Rev. Lett.45,

1948 ~1980!.95S.B. DiCenzo, K. Wertheim, and D. Buchanan, Phys. Rev. B24,

6143 ~1981!.96K. Kleinherbers, H.-G. Zimmer, and A. Goldmann, Surf. Sci.167,

417 ~1986!.97W. Erley and H. Wagner, Surf. Sci.66, 371 ~1977!.98W.T. Tysoe and R.M. Lambert, Surf. Sci.115, 37 ~1982!.

8-13

Documents

-Br/Pt(110) surface