Surface Science 603 (2009) 2734–2741

Contents lists available at ScienceDirect

Surface Science

journal homepage: www.elsevier .com/ locate /susc

DFT study of CO and NO adsorption on low index and stepped surfaces of gold

A. Hussain a, D. Curulla Ferré a,1, J. Gracia a, B.E. Nieuwenhuys a,b, J.W. Niemantsverdriet a,*

a Schuit Institute of Catalysis, Eindhoven University of Technology, P.O. Box 513, 5600 MB, Eindhoven, The Netherlandsb Leiden Institute of Chemistry, Leiden University, P.O. Box 9502, 2300 RA Leiden, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history:Received 29 April 2009Accepted for publication 9 July 2009Available online 24 July 2009

Keywords:Adsorption on GoldCO adsorptionNO adsorptionDFTCatalysis

0039-6028/$ - see front matter � 2009 Elsevier B.V. Adoi:10.1016/j.susc.2009.07.023

* Corresponding author.E-mail address: j.w.niemantsverdriet@tue.nl (J.W.

1 Present address: Total S.A., Gas and Power, R&DFrance.

Adsorption energies and vibrational frequencies of CO and NO adsorbed on gold (1 1 1), (1 0 0), (1 1 0)and (3 1 0) surfaces, as well as on adatoms on Au(1 0 0) have been calculated using density functionaltheory. The results clearly show that the adsorption energy of the molecules increases considerably withincreasing the degree of coordinative unsaturation of the gold atoms to which the molecules bind, andthus support the view that defects, steps and kinks on the surface determine the activity of gold catalysts.

� 2009 Elsevier B.V. All rights reserved.

1. Introduction

The pioneering work of Haruta et al. [1,2], who showed thatcatalysis based on nanoparticles of gold are highly active whereasextended surfaces are chemically almost inert, has led to greatinterest and much activity in gold nanocatalysis. The origin ofthe high activity of gold nanoparticles has been sought in the elec-tronic properties of nanoparticles and the influence of the support-ing oxides [2–7]. The presence of sites in which the metal atomshave low coordination has been identified as the most importantreason why small gold particles are active [8] and it is well con-ceivable that most of the effects of supports can be explained byconsidering how these affect the morphology of the supportedcrystallites [9].

Both computational and experimental studies of adsorption ongold surfaces have shown that the effect of the surface structure islarge. For example, Vinod et al. [10,11] showed experimentally thatthe OH bond in methanol dissociates on the stepped Au(3 1 0) sur-face, and that NO decomposes producing N2O and O-atoms at lowtemperatures, whereas the (1 1 1) and (1 1 0) surfaces are inactivein these bond breaking processes. Adsorption studies of CO revealthe importance of surface structure very clearly as well; Table 1presents an overview. A number of experimental studies confirmCO adsorption on the more corrugated surfaces of gold, with a heatof adsorption on the order of 60 kJ/mol [12–14]. The adsorptionmode is linear, as indicated by the CO stretch frequency in the

ll rights reserved.

Niemantsverdriet).Carbochemistry, 92078 Paris,

range of 2110–2120 cm�1 [14–16]. Computational works confirmthat CO bonding to the closely packed (1 1 1) surface is weak withadsorption energies between 15 and 40 kJ/mol, attributed for phys-isorption rather than chemisorption. For CO on the (1 0 0) and(1 1 0) surfaces, heats of adsorption between 55 and 70 kJ/molhave been computed, while numbers well above 100 kJ/mol werepredicted for adsorption on stepped surfaces, Au adatoms and clus-ters. Studies on the adsorption of NO are rare [11,17,18]; the avail-able results are shown in Table 2.

The purpose of this paper is to report adsorption energies andvibrational frequencies of NO and CO on a number of gold surfaceson the basis of Density Functional Theory (DFT). Included are thelow index surfaces (1 1 1), (1 0 0), and (1 1 0), the stepped (3 1 0)surface and an even more corrugated surface, in which an addi-tional gold atom has been placed on the hollow sites of aAu(1 0 0) surface, as done before by Liu et al. [19]. We confirmmethodically that NO and CO adsorb with considerable adsorptionstrength on low coordinated gold atoms.

2. Computational details

We used the Vienna ab-initio simulation package (VASP)[20,21], which performs an iterative solution of the Kohn–Shamequations in a plane-wave basis set. Plane-waves with a kinetic en-ergy below or equal to 400 eV were included in the calculation. Theexchange–correlation energy was calculated within the general-ized gradient approximation (GGA) proposed by Perdew and Wang(PW91) [22]. The electron–ion interactions for C, N, O and Au weredescribed by the projector-augmented wave (PAW) method devel-oped by Blöchl [23]. This is essentially a scheme combining the

Table 1Literature data: CO adsorption on various gold surfaces.

Surface Site Coverage (ML) Eads (eV) Frequency (cm�1) DFT Reference

ExperimentalAu(1 1 1) Top 2070 [45]Au(1 0 0) Top 2080 [45]Au(1 0 0) �0.60 [11]Au(1 1 0)–(1 � 2) �0.61 [13]Au(1 1 0) 2110 [15]Au(1 1 0) �0.34 [37]Au(3 3 2) 2120 [16]Au(3 1 1) Top 2117 [45]Gold films �0.57 2115–2120 [14]

ComputationalAu(1 1 1) Fcc-hollow �0.17 PBE [19]Au(1 1 1) Fcc-hollow �0.40 PW91 [38]Au(1 1 1) Top 0.25 �0.30 PW91 [8]Au(1 1 1) Top – �0.04 PW91 [39]Au(1 1 1) Top 0.25 �0.32 PW91 [40]Au(1 1 1) �0.26 PW91 [33]Au(1 1 1) Top 0.33 �0.28 2046 PW91 [44]Au(1 0 0) Bridge 0.5 �0.46 1902 PW91 [44]Au(1 0 0) �0.60 PW91 [33]Au(1 1 0) Top 0.25 �0.69 2042 PW91 [35]Au(1 1 0) Short bridge 0.25 �0.67 1962 PW91 [35]Au(1 1 0) Top 0.125–1.0 �0.69 to �0.56 2102–2116 PW91 [34]Au(1 1 0)–(1 � 2) Top 0.125–0.5 �0.73 to �0.52 2095–2100 PW91 [34]Au(2 1 1) Step-edge bridge 0.11–0.16 �1.40 to �1.05 – PBE [19]Au(2 1 1) Step-edge bridge 0.16 �0.66 – PW91 [8]Au(221) Step-edge bridge 0.08–0.125 �1.31 to �1.00 – PBE [19]Auð23 0Þ 0.33–1.0 �0.73 to �0.70 2102–2103 PW91 [34]

Au(3 2 1) Edge top �0.77 PW91 [41]Au(3 2 2) Top-step-edge 0.5 �0.63 2072 PBE [46]Au/Au(1 1 1) Top 0.25 �1.03 – PBE [19]Au-10 cluster Top – �0.95 – RPBE [42]Au-12 cluster – – �1.25 – PBE [19]

A. Hussain et al. / Surface Science 603 (2009) 2734–2741 2735

accuracy of all-electron methods and the computational simplicityof the pseudo potential approach [24].

Gold exhibits a face-centred cubic (fcc) structure. The relativepositions of the metal atoms were fixed initially as those in thebulk, with an optimized lattice parameter of 4.18 Å (the experi-mental value is 4.08 Å) [25]. The optimized lattice parameter wascalculated using the fcc unit cell and its reciprocal space was sam-pled with a (15 � 15 � 15) k-point grid generated automaticallyusing the Monkhorst–Pack method [26].

The CO molecule was calculated using a 10 � 10 � 10 Å3 cubicunit cell while for NO, a 10 � 12 � 14 Å3 orthorhombic unit cellwas used. Non-spin polarized calculations for the CO moleculeand spin-polarized calculations for NO were done at the C point.The Au(1 0 0), Au(1 1 0), Au(1 1 1) and Au(3 1 0) surfaces were rep-resented within the slab model approximation using a five-metallayer model and seven vacuum layers (>10 Å) except forAu(3 1 0), in which we used the slab model in Fig. 1. The depen-dence of the CO adsorption energy on the number of layers is stud-

Table 2Literature data: NO adsorption on various gold surfaces.

Surface Site Coverage(ML)

Eads(eV)

Frequency(cm�1)

DFT Reference

ExperimentalAu(1 0 0) �0.59 [17]Au(3 1 0) �0.75 [11]

ComputationalAu(1 1 1) Fcc-

hollow�0.21 PW91 [18]

Au(1 1 1) Top 0.25 �0.41 1903 PW91 [27]Au(3 2 1) Top �0.73 PW91 [43]

ied by Mavrikakis et al. [8]. The authors have reported thatadsorption does not depend on the number of Au layers if theseare more than 2. Moreover, in study [27], authors have concludedenergies to be well converged when used 4 layers of Au for NOadsorption. But the differences for NO using 4, 6 and 8 Au layersare not significant; we used 5 layers which with reference to study[27] seem quite reasonable.

Different slab models were used throughout this work: ap(2 � 2) unit cell to study surface coverages as low as 1/4 ML, a(p

3 � p3)R30� to study surface coverage of 1/3 ML, a c(2 � 2)and p(2 � 1) to study surface coverages of 1/2 ML and a p(1 � 1)to study surface coverages as large as 1 ML. The reciprocal spacefor p(3 � 3), p(2 � 2) and c(2 � 2) Au(1 0 0) unit cells were sam-pled with (3 � 3 � 1), (5 � 5 � 1) and (7 � 7 � 1) k-point meshes,respectively; p(2 � 2), ½(

p3 � p3)R55�and p(2 � 1) Au(1 1 0) unit

cells with (5 � 3 � 1), (6 � 6 � 1), (5 � 6 � 1) k-point meshes;p(2 � 2), (

p3 � p3)R30�Au(1 1 1) unit cells with (5 � 5 � 1) and

(7 � 7 � 1) k-point meshes; p(2 � 2), p(2 � 1), p(1 � 2) andp(1 � 1) Au(3 1 0) unit cells with (3 � 5 � 1), (3 � 9 � 1),(5 � 4 � 1) and (5 � 8 � 1) k-point meshes. In all cases, k-pointmeshes were automatically generated using the Monkhorst–Packmethod. A first-order Methfessel–Paxton smearing-function witha width 60.1 eV was used to account for fractional occupancies[28] (see Fig. 2).

Partial geometry optimizations were performed includingrelaxation of the two topmost metal layers for low index and fourtopmost layers for the stepped surface, using the RMM-DIIS algo-rithm [29]. In this method, the forces on the atoms and the stresstensor are used to determine the search directions for finding theequilibrium positions. Geometry optimizations were stoppedwhen all the forces (of the degrees of freedom set in the calcula-tion) were smaller than 0.01 eV/Å. Vibrational frequencies were

NO

AuAu Au

1.19 Å

2.08 Å96.9°121.3°

C

O

AuAu Au

1.16 Å1.98 Å102.6°

115.8°

122.7°

173.1°

Fig. 1. Distributions of electron densities for the most stable adsorption geometryof NO (up) and CO (down) on the top of step of the Au (3 1 0) surface. Figuregenerated with VESTA [36], isosurface levels were set at 0.1 ao

�3 where ao is theBohr radius, including the (0 0 1) cutoff plane. In the images isosurface cuts arecolored according to variations in the electron-density. Important geometricparameters are also illustrated. (For interpretation of the references to colour inthis figure legend, the reader is referred to the web version of this article.)

2736 A. Hussain et al. / Surface Science 603 (2009) 2734–2741

calculated within the harmonic approximation. The second-deriv-ative matrix (or Hessian matrix) is usually calculated numericallyby displacing every atom independently out of its equilibrium po-sition twice (e.g. ±0.02 Å). The adsorbate-surface coupling was ne-glected and only the Hessian matrix of the adsorbate wascalculated [30]. Zero point energy corrections were calculatedbut were small (�0.02–0.07 eV) and have not been included inthe results. We have performed several spin polarized calculationsto re-optimize NO adsorption. We do not observe significant differ-ences in adsorption energies (up to two decimal places) andgeometries.

110 100

-0e

-0.55eV

honot

-0.41eV

-0.53eV

-0.73eV

short bridge

hollow orlong bridge not stable

hollow not stable

topinside stepnot stable

-0.15 to -0.25 eVno clear site preference

-0e

CO / Au(111) CO / A

CO / Au(310) CO / Au

top

not stable

100110

Fig. 2. Overview of adsorbate structures of CO on different gold surfa

Analysis of the density of states (DOS) has been done for the COand NO adsorption on Au(1 0 0) and Au(3 1 0) surfaces. Interestedreader is referred to the supplementary material in Fig. 2s. We havenot included DOS analysis in the text because the informationseems unneeded, rationally they show increasing overlapping withthe adsorption energy.

3. Results

We calculated the adsorption energies, geometries and vibra-tional frequencies for CO and NO on the different gold surfaceswith several unit cell sizes, corresponding to varying adsorbatecoverages. As the results in general differ only marginally, we havechosen to present only one structure per surface in this paper, un-less differences were significant. Also, we only give the internalstretch frequency of the adsorbed CO and NO. The adsorption ener-gies were calculated by subtracting the energies of a CO or NO mol-ecule in the gas phase and a clean Au surface from the total energyof CO(NO)/Au system:

Eads ¼ ECOðNOÞ=Au � ECOðNOÞ � Eslab

The complete set of results is available as Supplementaryinformation.

3.1. Carbon monoxide adsorption on gold surfaces

Table 3 shows the most important data for the adsorption of COon five different Au surfaces. In all cases CO binds via the carbonatom to the metal, with the C–O bond perpendicular to the surfaceplane, unless indicated otherwise.

3.1.1. Au(1 1 1)CO adsorbs weakly, without clear preference for a specific

geometry on the (1 1 1) surface, with adsorption energies in the�0.12 to �0.25 eV range. Our values are in the middle of theadsorption energy range reported for CO on Au(1 1 1) in literature,which varies from �0.04 to �0.40 eV (see Table 1). The analysis ofthe topology of the electron density and of its Laplacian [31] doesnot let us discern between chemisorption and physisorption in anyof the models presented in this study. Taking as example Fig. 1,each single case included here shows a variable accumulation of

-0.53eV

-0.58eV

long bridgeand

hollownot stable

.46V

llowstable

.88V

u(100) CO / Au(110)

/Au(100)

CO stretch frequencies

All top sites:2062 ± 7 cm-1

All bridge sites:1910 ± 15 cm-1

ces, along with adsorption energies and CO stretch frequencies.

Table 3CO adsorption energy, metal carbon and carbon to oxygen distance, and CO stretchfrequency for adsorption on different gold surfaces.

Surface Unit cell Adsorption site Eads

(eV)dM–C

(Å)dC–O

(Å)vC–O

(cm�1)avisb

Au(1 1 1) p(2 � 2) Top �0.16 2.058 1.149 2062 2c

Bridge �0.15 1.481 1.170 1899 0Fcc hollow �0.16 1.341 1.181 1819 0Hcp hollow �0.12 1.366 1.178 1836 0

(p

3 � p3)R30�

Top �0.24 2.233 1.151 2056 2

Bridge �0.25 1.466 1.169 1909 0Fcc-hollow �0.15 1.446 1.176 1857 0Hcp-hollow �0.17 1.424 1.175 1867 0

Au(1 0 0) p(2 � 2) Top �0.46 2.020 1.150 2064 1c

Bridge �0.55 1.540 1.168 1898 2c

Hollow �0.07 1.264 1.180 1812 2

Au(1 1 0) p(2 � 2) Top �0.53 2.000 1.150 2060 2c

Short bridge �0.58 1.467 1.169 1907 1c

Long bridge �0.09 1.323 1.166 1913 1

Au(3 1 0) p(2 � 1) Top at step �0.73 1.98 1.13 2062 1c

(1 0 0) Bridgeat outer step

�0.53 1.45 1.15 1894 0

(1 0 0) Bridgeat inside step

�0.41 1.48 1.15 1894 0

(1 1 0) Shortbridge

�0.54 1.50 1.15 1900 1

(1 0 0) Top �0.41 2.01 1.15 2047 1

Au/ Au(1 0 0)p(2 � 2) Top �0.88 1.96 1.15 2060 2c

a Stretch frequency for CO molecule in gas phase is calculated as 2130 cm�1.b Number of imaginary frequencies at displacement ±0.02 Å.c Long displacements around the mean position have been used for the most

favorable configurations yielding in all cases to real values (see Supplementarymaterial).

A. Hussain et al. / Surface Science 603 (2009) 2734–2741 2737

electron density in the bonding region consistent with the lower-ing of the potential energy and decrease of the total energy. Mostlikely, caused by the shallow nature of the potentials associatedwith the relatively weakly adsorbed CO molecule on gold, we can-not confirm the nature of all possible minima with real vibrationfrequencies for all modes (see Table 3). We discuss this separatelylater in the article because of its recurrent appearance in all slabmodels.

3.1.2. Au(1 0 0)Adsorption of CO on the (1 0 0) surface of gold is markedly

stronger than on Au(1 1 1), with energies of �0.55 eV on the bridgesite and �0.46 eV on the top site, which seem to the authors bothstable adsorption geometries. The CO stretch frequencies for linearand bridged CO in Table 3 fall within the ranges commonly ob-served for these bonding geometries of CO on other metals [32]and that they fall at the higher end of the ranges, in agreementwith the relatively weak interaction with gold. The metal carbonand carbon–oxygen distance changes between adsorption on topand on the bridge site are consistent with the changes in the vibra-tional frequencies. Our value of �0.55 eV for the adsorption energyof CO on Au(1 0 0) is in good agreement with the �0.60 eV com-puted by Molina and Hammer [33] and also with the experimen-tally determined �0.6 eV reported for CO adsorbed at lowcoverage by Elhiney and Prichard [12].

3.1.3. Au(1 1 0)Two particularly stable adsorption modes for CO appeared,

namely on the short bridge (i.e. between two near neighbor Auatoms on a row) and on top. Adsorption states on the long bridge

(i.e. between two adjacent rows) and in the hollow site were notstable. The adsorption energies for CO on the short bridge(�0.58 eV) and on top (�0.53 eV) are on the order of 0.1 eV less(i.e. weaker) than calculated by Loffreda and Sautet [34] and byShubina and Koper [35]. Our values are also within 0.1 eV of theexperimental value of �0.61 eV reported by Gottfried et al. [13],but disagree with Outka and Madix who reported an adsorptionenergy of �0.34 eV. It is likely that the experimental data are asso-ciated with linear CO, as the experimental frequency of 2110 cm�1

reported by Jugnet et al. [15] indicates. The frequency we found forlinear CO is substantially lower at 2060 cm�1.

We also performed calculations for other unit cells, ½(p

3 � p3)R55�and p(2 � 1) to investigate the effect of different CO coverages(1/2 ML) and unit cell structure. CO is equally stable on shortbridge sites of both the unit cells with an adsorption energy of�0.58 eV. The top configuration is slightly less favored than shortbridge for CO adsorption; however, the interaction is very weakon long-bridge sites. We did not observe the CO adsoption on hol-low sites at all the studied unit cells of this surface in agreementwith DFT study reported in Ref. [35].

3.1.4. Au(3 1 0)The Au(3 1 0) surface was modeled using the construction in

Fig. 1. The structure is more or less equivalent to a 4 gold layeron low index Au(1 0 0) or Au(1 1 1) slab models. Gold atoms ap-pear in 11 different layers according to their height; in geometryoptimizations those gold atoms at the four topmost layers arerelaxed.

This surface consists of (1 0 0) terraces and (1 1 0) steps, and ex-poses Au atoms of lower coordination than the (1 1 0) surface does.Indeed, we observe that CO bound linearly on these low coordi-nated Au atoms at the step yields the highest adsorption energy,namely �0.73 eV. The molecule tilts to a direction between thenormals of the facets forming the step (see Fig. 1). The metal to car-bon distance for this linear CO species is about 1% smaller than onthe (1 1 0) surface. The two bridge positions on the (1 0 0) terraceyield stable adsorption geometries, with the position near the stepedge being preferred over the one near the inner side of the step.CO on the top site of the (1 1 0) plane and perpendicular to it showsa substantial interaction, but CO moves away to the most stableadsorption geometry described above, which involves the samegold atom. Fig. 1 shows the structure of the most relevant COadsorption geometry on the (3 1 0) surface.

3.1.5. Au/Au(1 0 0)To explore whether a further decrease in coordination of gold

leads to even stronger adsorption than on the step edge as inFig. 1, we also investigated CO on top of an additional gold atomplaced on the hollow site of the Au(1 0 0) surface. We did this forp(2 � 2) and p(3 � 3) geometries but found no essential differ-ences. One apparent minimum for CO on this surface appeared,namely with CO linearly on top of the adatom, perpendicular tothe underlying (1 0 0) substrate. The adsorption energy of�0.88 eV is the highest observed in this study. The CO stretch fre-quency equals that of linear CO on the other surfaces considered,but the metal – carbon bond is some 2% shorter than on the uncor-rugated surfaces, and 1% lower than for linear CO on the (3 1 0)step. The situation may be compared to the somewhat similaradsorption of CO on the Au/Au(1 1 1) system as reported by Liuet al. [19], who reported an adsorption energy of �1.03 eV, i.e.somewhat higher than the �0.88 eV we find here. This would bein agreement with the lower coordination of an adatom on anfcc(1 1 1) surface, but we should also take into account that a dif-ferent functional (PBE) was used in the study of Liu et al.

2738 A. Hussain et al. / Surface Science 603 (2009) 2734–2741

3.2. Nitrogen oxide adsorption on gold

Table 4 shows the most important data for NO adsorption onthe Au surfaces. As with CO, bonding is between the nitrogen atomand the metal, with the N–O bond perpendicular to the surfaceplane in most cases.

3.2.1. Au(1 1 1)NO adsorbs weakly, and is most stable in the hollow sites. Top

coordination do not yield stable adsorption sites. If we usea(p

3 � p3) R30�unit cell instead of the p(2 � 2), both the bridgeand the hollow sites appear as actual minima and the adsorptionenergies are slightly higher at �0.27 eV. The energy differences be-tween the sites are negligible indicating that unhindered diffusionis possible at virtually all temperatures. Torres et al. [18] reportadsorption energies of �0.21 eV, which is in agreement with ourresult.

3.2.2. Au(1 0 0)Adsorption of NO on Au (1 0 0) is much stronger than on

Au(1 1 1). The bridge site yields to the preferred adsorption, withan energy of �0.57 eV. Our result is in agreement with TPD obser-vations (57 ± 4 kJ/mol) by Rienks et al. [17].

3.2.3. Au(1 1 0)NO appears as stable at the short bridge of this surface, and the

adsorption energy depends on the coverage of NO. For the p(2 � 2)geometry we find an adsorption energy of �0.50 eV while thep(2 � 1) structure yields a stronger adsorption bond of �0.61 eV.

Table 4NO adsorption energy, metal nitrogen and nitrogen to oxygen distance, and NOstretch frequency for adsorption on different gold surfaces.

Surface Unit cell Site Eads

(eV)dM–N

(Å)dN–O

(Å)vC–O

(cm�1)avisb

Au(1 1 1) p(2 � 2) Fcc-hollow �0.15 1.572 1.189 1629 2c

Hcp-hollow �0.10 1.594 1.187 1646 2c

Bridge �0.13 1.607 1.181 1671 1Top 0.17 2.613 1.163 1839 4

(p

3 � p3)R30� Bridge �0.27 1.444 1.187 1651 0Fcc-hollow �0.27 1.371 1.196 1600 0Hcp-hollow �0.22 1.483 1.190 1635 0Top 0.15 2.597 1.166 1834 4

Au(1 0 0) p(2 � 2) Bridge �0.57 1.412 1.190 1638 1c

Hollow �0.45 1.460 1.185 1649 2Top �0.04 2.120 1.170 1771 4

Au(1 1 0) p(2 � 2) Short bridge �0.50 1.584 1.193 1636 2c

Long bridge �0.38 1.330 1.181 1680 1Hollow �0.23 1.649 1.168 1765 1Top �0.07 2.120 1.170 1762 4

p(2 � 1) Short bridge �0.61 1.545 1.191 1669 0Long bridge �0.43 1.367 1.176 1736 1Hollow �0.22 1.690 1.167 1796 2Top �0.10 2.066 1.177 1778 4

Au(3 1 0) p(2 � 1) Top at step �0.71 2.080 1.190 1726 1c

(1 0 0) Bridgeat outer step

�0.65 1.391 1.170 1631 0

(1 0 0) Bridgeat inside step

�0.48 1.788 1.170 1648 0

(1 1 0) Shortbridge

�0.55 1.487 1.180 1640 1

(1 0 0) Top 0.03 2.150 1.070 1707 3

Au/Au(1 0 0)

p(2 � 2) Top �0.47 1.96 1.18 1792 4

a Stretch frequency for NO molecule in gas phase is calculated as 1904 cm�1.b Number of imaginary frequencies at displacement ±0.02 Å.c Long displacements around the mean position have been used for the most

favorable configurations yielding in all cases to real values (see Supplementarymaterial).

All other adsorbate configurations are less favorable with the lineargeometry being the less likely. No literature data are available forcomparison.

3.2.4. Au(3 1 0)Au(3 1 0) surface with (1 0 0) terraces and (1 1 0) steps enables

adsorption geometries for the NO which are to a large extent sim-ilar as found for CO adsorption. The preferred adsorption site forNO on Au(3 1 0) is on a single Au atom at the step, with the Au–N bond perpendicular to the (3 1 0) plane, however with the N–Obond at an angle of 122.7� with respect to the Au–N bond, whereO points towards upper terrace. The Au–N–O plane is perpendicu-lar to the step, see Fig. 1. The adsorption energy is �0.71 eV.

The second most favorable site for NO adsorption is the (1 0 0)bridge site at the outer step with an adsorption energy of �0.65 eV.The other bridge site on the (1 0 0) terrace, adjacent to the inside ofthe step, also provides a stable adsorption state at �0.48 eV.

Vinod et al. [11] reported a desorption temperature of about300 K for NO from Au(3 1 0) in their XPS and TPD studies. Assum-ing a standard pre-exponential factor of 1013 s�1 the adsorption en-ergy is on the order of �72 kJ/mol (0.75 eV), which agrees very wellwith adsorption energy of �0.71 eV that we calculated.

3.2.5. Au/Au(1 0 0)At first no stable adsorption states for NO could be found on Au/

Au(1 0 0). However, NO on the top site gives considerable adsorp-tion energy of �0.47 eV. Two degenerate pairs of imaginary fre-quencies (values given in the Supplementary information) werefound, resulting in a movement of the NO molecule away fromthe adatom towards a stable adsorption mode on the nearby(1 0 0) terrace, which according to the discussion above will be abridge site.

3.3. Frequency analysis

Plane-wave based codes, such as VASP, do not implement ana-lytical calculation of vibrational frequencies. Numerical frequencycalculations are usually carried out by displacing the adsorbate nu-clei out of their equilibrium position twice in x, y and/or z direction.The displacement needs to be large enough to ensure an accuratecalculation of the forces acting on the nuclei (actually, one needsto impose the requirement that the order of magnitude of the dif-ference between the forces on the nuclei by displacing a given nu-cleus in a given direction and its corresponding negativedisplacement is larger than the precision in the calculation of suchforces) and short enough to prevent exiting the harmonic region ofthe vibrational motion. Most likely a standard displacement of±0.02 Å for all nuclei balances these two requirements. In addition,as the curvature of the various normal modes sub-spaces is notequal one can think of using adapted displacements for each nu-cleus in order to optimize the accuracy in the calculation of allvibrational frequencies.

For weakly interacting systems when the mixed surface-adsor-bate sub-spaces are very flat, inaccuracies inherent to the selectionof computational parameters can result into the calculation ofimaginary frequencies because small displacements are in the pre-cision limit for the calculation of forces; or due to optimizations re-sult into geometries close to but not at the energy bottom of thelocal minimum. For such cases the best selection but most costlyseems to decrease the tolerance of the calculations. Due to theextension of current work this has not been done here, and wekeep the door open for future works; nevertheless the problemhas been approached at other level. Under the condition that theoptimized geometry is almost indistinguishable of the actual min-imum, one can increase the displacements (crossing any immedi-ate energy bottom or precision limit in the calculation of forces)

A. Hussain et al. / Surface Science 603 (2009) 2734–2741 2739

and so avoiding imaginary frequencies while keeping the localstructure on the potential energy surface. This is indeed what wehave done in this study for the CO/Au and NO/Au systems. We havemoved the C and O atoms (or N and O atoms) out of their equilib-rium positions in the x, y-plane (assuming that the z direction isthe normal to the surface), with displacements as large as 0.16 Åin order to ensure the balance between accuracy and harmonicity.One must double-check when using such large displacements thatthe positive and negative-displacement result in geometries thatare isoenergetic and that the calculated forces acting on the dis-placed nucleus are opposite in the direction of the displacement;thus proving that one is within the harmonic vibrational motion.Because large displacements may also falsely convert in somecases imaginary frequencies into real frequencies, thus changingthe qualitative nature of the configuration from a transition state(or higher-order saddle point) to a stable minimum; we cannotguarantee the nature of the adsorption sites, unless conformationsrepresent a global minimum. Under these conditions, as it is indi-cated in Tables 3 and 4, all imaginaries frequencies change into realvalues for the preferred configurations.

Table 5 illustrates the procedure with CO and NO adsorbing onthe bridge site of the Au(1 0 0) surface. Both cases represent themost stable adsorption configuration on this surface, but the usu-ally applied displacements of 0.02–0.04 Å around the equilibriumbond length for exploring the shape of the potential energy surfacearound the minimum had to be increased to at least 0.10 Å to ob-tain real values for all the vibrational frequencies. Analysis of theenergy and the forces acting on the nuclei in the v6 vibrationalmode at the displacement of ±0.10–0.14 Å shows that the systemscan still be considered harmonic, thus validating such large dis-placements. Table 5 also displays the effect of the displacementon the vibrational frequencies of NO adsorbed on top onAu(1 0 0). With a displacement of ±0.02 Å, there are four imaginaryfrequencies. Larger displacements of ±0.04 and ±0.08 Å still showsfour imaginary frequencies. However, even larger displacements of±0.10, ±0.14 and ±0.16 Å convert two imaginary frequencies into

Table 5Selected examples showing how imaginary frequencies disappear if the displace-ments dd in the frequency calculation are increased; v1 is the internal stretchfrequency of the CO and NO, v2 is metal carbon or metal nitrogen stretch, v3–v6 are thefrustrated rotations and librations of the adsorbed molecules.

dd (Å) v1

(cm�1)v2

(cm�1)v3

(cm�1)v4

(cm�1)v5

(cm�1)v6

(cm�1)

Au(1 0 0)–p(2 � 2)-CO-bridged0.02 1898 249 346 246 177 63i0.04 a a 348 299 170 62i0.06 a a 352 303 172 51i0.08 a a 357 308 176 15i0.10 a a 362 314 178 380.14 a a 375 330 177 65Au(1 0 0)–p(2 � 2) NO-

bridged0.02 1638 242 498 140 78 105i0.06 a a 498 157 79 68i0.10 a a 499 163 89 460.13 a a 499 160 113 67

Au(1 0 0)–p2 � 2-NO-top0.02 1771 204 86i 88i 301i 302i0.04 a a 73i 74i 296i 296i0.08 a a 27i 27i 280i 280i0.10 a a 23 22 270i 270i0.14 a a 49 49 245i 245i0.16 a a 50 50 230i 230i

a The internal stretch frequency of CO and NO, metal carbon and metal nitrogenstretch, are accurate enough when smaller displacements such as ±0.02 Å are used.Therefore, calculations were carried out only along x and y-axes for largerdisplacements.

real frequencies and keeps the other two imaginary indicating thatNO at on top sites on Au(1 0 0) is a second-order saddle point,which is consistent with the chemical perception. Within this pro-cedure, we can avoid alternative solutions orientated to use expen-sive computational parameters to deal with the flatness of thepotential energy surface while keeping physical meaning and thestandard accuracy of the initial calculations. Nevertheless, withinthis scheme we cannot classify with guarantees a conformation,but for those configurations predicted as global energy minima,where long displacements make full sense in order to get rid ofimaginary values.

4. Discussion

Fig. 2 summarizes the key results for CO adsorption on the fivegold surfaces considered in this paper. The adsorption energiesclearly reveal how the Au–CO interaction energy becomes strongeras the coordinative unsaturation of the gold atoms involved in thebonding increases. Fig. 3 illustrates this clearly for linear CO ongold: decreasing coordination of gold from N = 9 on Au(1 1 1) toN = 4 in the Au adatom on the Au(1 0 0) surface is associated withan increase in adsorption energy from �0.15 to �0.88 eV, and a de-crease in metal to carbon distance from 2.06 to 1.96 Å. Interest-ingly, the C–O stretch frequency is not much affected, it remainswithin 2062 ± 7 cm�1 for all situations considered here, indicatingthat back donation from the Au into the CO 2p* orbitals is very lim-ited as is expected from an s-metal.

According to the DFT calculations CO generally prefers thebridge mode on the flat surfaces, which is not in agreement withthe few vibrational studies that are available in the literature;these indicate frequencies characteristic for linear adsorption[14–16]. This is a known problem with DFT [40].

We observe no significance influence of coverage on adsorptionenergies. On Au(3 1 0) even at 1 ML, CO molecules are far apart,6.61 � 4.18 Å, and interactions between the molecules in the cal-culations seem irrelevant. The effect of coverage is higher thanour one in the study done by Mehmood et al. [44]. Coverages upto 1 ML, here up to ½ ML, on low miller indexes cuts reflect theinfluence of side interactions from close CO molecules. At 1 MLcoverage, the distance between CO molecules is 2.96 � 2.96 Åand 2.79 � 2.79 Å for Au(1 0 0) and Au(1 1 1), respectively.

Fig. 4 summarizes the adsorption of NO on gold in a similar for-mat as for CO. Clearly, NO adsorbs on the bridge sites, and the en-ergy varies only slightly with the degree of coordination. Only on

1.90

1.94

1.98

2.02

2.06

2.10

0 0.2 0.4 0.6 0.8 1CO Adsorption Energy (eV)

Met

al –

Car

bon

dist

ance

(A) Linear CO on Au Surfaces

Au(111)N = 9 Au(100)

N = 8 Au(110)N = 7

Au(310)N = 6

Au/Au(100)N = 4

- - - - -

Fig. 3. Au–C bond length versus CO adsorption energy for linearly bound CO ondifferent gold surfaces; N is the coordination number of the Au atom involved in thebonding.

100

-0.55±0.06eV

long bridgeand

hollownot stable

-0.57eVtop

and

hollownot stable

-0.48eV

-0.65eV

-0.71eV

short bridge,

hollow, orlong bridge not stable

hollow not stable

topinside stepnot stable

-0.13 to -0.27 eVno clear site preference

no stable

adsorption

sites

110

NO / Au(111) NO / Au(100) NO / Au(110)

NO / Au(310) NO / Au/Au(100)

NO stretch frequencies

All bridge sites:1635 ± 10 cm-1

(310) top site at the step:1730 ± 10 cm-1

topnot

stable

top (100)

not stable

100110

Fig. 4. Overview of adsorbate structures of NO on different gold surfaces, along with adsorption energies and NO stretch frequencies.

2740 A. Hussain et al. / Surface Science 603 (2009) 2734–2741

the stepped surface, NO binds to a single Au atom of the edge, butin a different way than CO does, as the NO bond tilts towards theupper terrace. NO does clearly not adsorb in linear modes, thismode is repulsive on the (1 1 1) surface, almost non-bonding on(1 0 0), very weak on (1 1 0) and although on the lowly coordinatedAu atom of the Au/Au(1 0 0) adatom system the interaction energyhas increased to �0.47 eV, the imaginary frequencies in the normalmode analysis indicate that this state is unstable.

On Au(1 1 1) and Au(1 1 0) the adsorption energy of NO in-creases with coverage. Differences in the structure between unitcells do not seem to be of relevance, and we do not provide a rea-sonable explanation out of possible intrinsic errors in the method.

5. Conclusions

NO and CO adsorb with moderately strong adsorption energiesbetween �0.5 and �0.9 eV on gold surfaces, provided the Au atomsare more uncoordinated than on the close packed (1 1 1) surface.The adsorption energy of CO decreases in the order Au/Au(1 0 0) > Au(3 1 0) > Au(1 1 0) � Au(1 0 0) > Au(1 1 1). Similartrend is followed for NO adsorption energies. Au atoms at stepedges or Au adatoms on terraces show the strongest interactionwith CO and NO. These results suggest that the catalytic activityof small gold particles with diameters of a few nanometers canbe attributed to the presence of low coordinated gold atoms.

Mostly, a slight decrease in adsorption energies with an in-crease in coverage has been predicted due to lateral interactions,however on Au(1 1 1) and Au(1 1 0) the adsorption energy of NOincreases with coverage. Bridge site is favored for both moleculesat low index surfaces but at the stepped (3 1 0), top coordinationis preferred. In all adsorption configurations, C–O and N–O molec-ular axes are perpendicular to the surfaces except for NO onAu(3 1 0). NO shows a tilted geometry with the nitrogen atom al-most on top of a gold atom at the edge of the step, while the oxy-gen atom points towards the next layer of gold atoms belonging tothe terrace.

Acknowledgements

We thank NCF (Grant SG-06-2-202) for generously grantingcomputer time at the Huygens Super Computer. Akhtar Hussain

acknowledges financial support from the Pakistan Higher Educa-tion Commission to enable his stay at Eindhoven University ofTechnology.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.susc.2009.07.023.

References

[1] M. Haruta, Catal. Today 36 (1997) 153.[2] M. Haruta, T. Kobayashi, N. Yamada, Chem. Lett. 2 (1987) 405.[3] G.C. Bond, D.T. Thompson, Gold Bull. 33 (2000) 41.[4] G.C. Bond, C. Louis, D.T. Thompson, Catalysis by gold Science Series, vol. 6, Imp.

College Press, 2006.[5] B.E. Nieuwenhuys, A.C. Gluhoi, E.D.L. Rienks, C.J. Weststrate, C.P. Vinod, Catal.

Today 100 (2005) 49.[6] B. Hammer, J.K. Nørskov, Nature 376 (1995) 238.[7] M. Valden, X. Lai, D.W. Goodman, Science 281 (1998) 1647.[8] M. Mavrikakis, P. Stoltze, J.K. Nørskov, Catal. Lett. 64 (2–4) (2000) 101.[9] N. Lopez, T.V.W. Janssens, B.S. Clausen, Y. Xu, M. Mavrikakis, T. Bligaard, J.K.

Nørskov, J. Catal. 223 (2004) 232.[10] C.P. Vinod, J.W. Niemantsverdriet, B.E. Nieuwenhuys, Phys. Chem. Chem. Phys.

7 (2005) 1.[11] C.P. Vinod, J.W. Niemantsverdriet, B.E. Nieuwenhuys, Appl. Catal. A: General

291 (2005) 93.[12] G. McElhiney, J. Prichard, Surf. Sci. 60 (1976) 397.[13] J.M. Gottfried, K.J. Schmidt, S.L.M. Schroeder, K. Christmann, Surf. Sci. 536

(2003) 206.[14] M.L. Kottke, R.G. Greenler, H.G. Tompkins, Surf. Sci. 32 (1972) 231.[15] Y. Jugnet, F.J.C.S. Aires, C. Deranlot, L. Piccolo, J.C. Bertolini, Surf. Sci. 521 (2002)

L639.[16] C. Ruggiero, P. Hollins, Surf. Sci. 377–379 (1997) 583.[17] E.D.L. Rienks, G.P. van Berkel, J.W. Bakker, B.E. Nieuwenhuys, Surf. Sci. 571

(2004) 187.[18] D. Torres, S. Gonzalez, K.M. Neyman, F. Illas, Chem. Phys. Lett. 422 (2006) 412.[19] Z.P. Liu, P. Hu, A. Alavi, J. Am. Chem. Soc. 124 (2002) 14770.[20] G. Kresse, J. Furthmüller, Phys. Rev. B 44 (1991) 13298.[21] G. Kresse, J. Hafner, Phys. Rev. B 47 (1993) 558.[22] J. Perdew, Y. Wang, Phys. Rev. B 45 (1992) 13244.[23] P.E. Blöchl, Phys. Rev. B 50 (1994) 17953.[24] G. Kresse, D. Joubert, Phys. Rev. B 59 (1999) 1758.[25] Structure Data of Elements and Intermetallic Phases, Landolt-Bornstein, New

Series, vol. IIIb, Springer, Berlin, 1971, <www.webelements.com>.[26] H.J. Monkhorst, J.D. Pack, Phys. Rev. B 13 (1976) 5188.[27] W. Zhang, Z. Li, Y. Luo, J. Yang, J. Chem. Phys. 129 (2008) 134708.[28] M. Methfessel, A. Paxton, Phys. Rev. B 40 (1989) 3616.[29] P. Pulay, Chem. Phys. Lett. 73 (1980) 393.[30] J.D. Head, Int. J. Quantum Chem. 65 (1997) 827.

A. Hussain et al. / Surface Science 603 (2009) 2734–2741 2741

[31] R.F.W. Bader, Atoms in Molecules: A Quantum Theory, Oxford University Press,Oxford, UK, 1990.

[32] N. Sheppard, T.T. Nguyen, in: R.J.H. Clark, R. Hester (Eds.), Advances in Infraredand Raman Spectroscopy, vol. 5, Heyden, London, 1978, p. 67.

[33] L.M. Molina, B. Hammer, Phys. Rev. B 69 (2004) 155424.[34] D. Loffreda, P. Sautet, J. Phys. Chem. B 109 (2005) 9596.[35] T.E. Shubina, C. Harting, M.T.M. Koper, Phys. Chem. Chem. Phys. 6 (2004) 4215.[36] K. Momma, F. Izumi, J. Appl. Crystallogr. 41 (2008) 653.[37] D.A. Outka, R.J. Madix, Surf. Sci. 179 (1987) 351.[38] S. Kandoi, A.A. Gokhale, L.C. Grabow, J.A. Dumesic, M. Mavrikakis, Catal. Lett.

93 (2004) 93.

[39] B. Hammer, Y. Morikawa, J.K. Nørskov, Phys. Rev. Lett. 76 (1996) 2141.[40] M. Gajdos, A. Eichler, J. Hafner, J. Phys.: Condens. Matter 16 (2004) 1141.[41] J.L.C. Fajin, M.D.S. Cordeiro, J.R.B. Gomes, J. Phys. Chem. C 112 (2008) 17291.[42] I.N. Remediakis, N. Lopez, J.K. Nørskov, Appl. Catal. A: General 291 (2005) 13.[43] J.L.C. Fajin, M. Natalia, D.S. Cordeiro, J.R.B. Gomes, J. Phys. Chem. C 113 (2009)

8864.[44] F. Mehmood, A. Kara, T.S. Rahman, C.R. Henry, Phys. Rev. B 79 (2009) 075422.[45] I. Nakamura, A. Takahashi, T. Fujitani, Catal. Lett. 129 (2009) 400.[46] W. Yim, T. Nowitzki, M. Necke, H. Schnars, P. Nickut, J. Biener, M.M. Biener, V.

Zielasek, K. Al-Shamery, T. Klüner, M. Bäumer, J. Phys. Chem. C 111 (2007) 445.