How Carbon Monoxide Bonds to Metal Surfaces

7/29/2019 How Carbon Monoxide Bonds to Metal Surfaces

1/7

578 J . Am. Chem. SOC.chains wind wildly and exhibit a broad range of hydrogen bondgeometries. There is certainly no preference for coplanar dimers.The same description applies to liquid N M A except that fewmonomers participate in more than two hydrogen bonds so thereis little branching of the chains. Local order is also evident inliquid DM F, although it is not as easy to characterize. Nearneighbors definitely organize to provide favorable Coulombicinteractions. Short contacts between oxygen and the methyl groupsare particularly common. Some neighbors with roughly anti-parallel dipoles are apparent. However, this is not a strikingfeature and seems to be hidden by the apparent medium andlong-range disorder.Conclusion

Optimized intermolecular potential functions have been derivedfor use in f luid simulations of amides and peptides. The functionshave been demonstrated to yield reasonable energetic andstructural descriptions of amide dimers and amide-water com-plexes. Moreover, the principal support for their validity comesfrom results of M onte Carlo simulations of liquid formamide,

1985, 107, 578-584NM A , and DM F. The computed heats of vaporization anddensities of the liquids are in essentially exact accord with ex-perimental data. A greement with available dif fraction data onpeak positions for radial distribution functions is also excellent.In addition, the simulations yielded thorough descriptions of thestructure and hydrogen bonding in the liquid amides. The dom-inance of hydrogen-bonded chains in liquid formamide and NM Awas confirmed, while more subtle local order is apparent in D M Fas revealed by dipole correlation functions. The simplicity, form,and demonstrated util ity of the OPL S functions make themsuitable for study of a wide range of organic and biochemicalsystems.

Acknowledgment. Grati tude is expressed to the NationalScience Foundation and National Institutes of Health for supportof this work. Dr. Phill ip Cheeseman kindly provided the programfor making the stereo plots.Registry No. NMA, 79-16-3;DMF, 68-12-2;NM A dimer, 91385-81-8;DMF dimer, 93984-95-3; ormamide,75-12-7; formamide dimer,28704-51-0.

How Carbon Monoxide Bonds to Metal SurfacesShen-Shu Sung and Roald Hoffmann*Contribution fr om the Department of Chemistry, Cornell University, I thaca, New Y ork 14853.Received M ay 9, 1984

Abstract: An analysis is presentedof the similarities and differences in the bonding of CO to Ni(100), N i(l 1 ), Co(OOOl),Fe(1 lo), C r(l lO), and Ti(0001). The primary interactions in every case are the expected forward and back donation, themixing of CO 5u with surface dzz and s states, and the mixing of C O 2a with metal dxryz. The latter interaction is dominant.Projections of the density of states and crystal orbital overlap population curves show these interactions clearly. As the metalchanges from Ni to Ti, the effects of the rise of the Fermi level and the dif fusenessof the d orbitals combine to put moreand more electron density into 2a of CO. This is why CO dissociates on the earlier transition-metal surfaces. Our studiesof surface coverage show little effect on the diatomic dissociation. There is some indication in the calculations that higherindex surfaces should dissociate CO more readily than the lower index ones, but the effect is smaller than that of changingthe metal.

The chemisorption of C O ontransition-metal surfaces has beenthe subject of numerous experimental and theoretical studies overthe past few decades. BrodEn, Rhodin, et al. have succeededin establishing a criterion for dissociative chemisorption behaviorof carbon monoxide on a number of transition-metal surfaces asa function of the position of the metal in the periodic table. Onthe left side of the first-row transition metals, up to Fe, the ad-sorption is likely to be dissociative. On the right side from Co,it is molecular. Experimental results are available for T i,* Fe,3CO,~nd N i lb n the first row.In this report we analyze and discuss the electronic consequencesof the chemisorption of C O on several transition metals, by usingtight binding calculations of the extended Huckel type. M ono-layers of CO, C O on Ni(100), comparisons of different metals,dif ferent coverages, and different surfaces of the same metal willbe studied in this contribution.

(1) (a) BrodEn, G.; Rhodin, T.N.; Brucker,C.;Benbow, R.; Hurych, Z.Surf . Sci. 1976,59, 593. (b) Engel, T.; Ertl, G. Ado. Catal. 1979,28, 1.(2) Eastman, D. E. Solid State Commun. 1972, 10, 933.(3) BrodEn, G.;Gafner, G.;Bonzel, H. Appl. P hys. 1977,13,33; ensen,E. S. ;Rhodin, T. N. Phys. Rev. B 1983,27, 3338.(4) Greuter, F.;Heskett, D.; Plummer, E. W.;Freund, H. J . Phys. Rev.B 1983,27, 7117.(5) (a) Blyholder, G. J . Phys. Chem. 1964, 68, 2772. (b) Anderson, A .B. J . Chem. Phys. 1976,64, 4046. (c)van Santen, R. A. Proc. In?. Cong.Catal., 8th 1984.(6) Andreoni, W.; Varma, C.M. Ph ys. Rev. B 1981,23,437. Allison, J .N.; Goddard, W. A,, 111Surf. Sci. 1981,110,L615. Doyen, G.;Ertl,G. Surf .Sci. 1977,B5, 641. Doyen, G.; Ertl, G. Surf .Sci. 1977,69, 157. Bagus, P.S.;Hermann, K. Phys. Rev. B. 1977,16,4195. Davenport, J . W . Phys. Rev.L et?.1976,36,945. Bullett, D. W.; Cohen, M. L. J . Phys. 1977,CI O, 2101.0002-7863/85/1507-0578$01.50/0

M onolayers of COThe molecular orbitals of an i solated CO molecule are well-known. The highest occupied molecular orbital (HOM O) ismainly a carbon lone pair, as indicated schematically in l a.K

n7METAL

z7zzzV7-METAL METALa b

1Its energy (ca. -14 eV)* is lower than the d states of transitionmetals. In the generally accepted end-on CO chemisorptionconfiguration the 5u is pushed down by surface states, relativeto other C O levels. The lowest unoccupied molecular orbital set(L UM O) of the free CO molecule,2?r,consists of two antibonding(7) (a) J orgensen, W. L .; Salem, L. TheOrganic Chemists Book ofOrbitals;Academic Press: NewY ork 1973;p 78. (b) Huo,W.M. J . Chem.Phys. 1964,43,624.McLean,A. D.;Y oshimine, M. BM J . Res. Dev. Suppl .1968,3, 206.(8) Gelius, U.; Basiler, E.; Svensson, S.; Bergmark, T.;Siegbahn, K. J .Electron Spectrosc. Relat. Phenom. 1973, 2, 405. Turner, D. ; Baker, C.;Baker A . D.; Brunkle, C. R. Molecular Photoelectron Spectroscopy; Wi -ley-Interscience: New Y ork, 1970

0 1985 A merican C hemical Society


2/7

How Carbon Monoxide Bonds to M etal Surfaces- 401 l r I I

II [-2 0 0r M K r r M K r r M K r

dco.co - 4 35 H dco.co = 3 29 d dCO.CO 2 510 b C

Figure 1. Band structures of monolayer CO along the symmetry linesindicated by I'M, M K , andKr for CO-CO distances 4.35, 3.29 and 2.51A .orbitals with larger coefficients at the carbon atom, as shown inlb. In the absorption the 27r derived orbitals are partiallv oop-ulated and the CO bond weakened. That the 27r orbi tals. y acrucial role in C O dissociation is a general conclusion previouslyreached by others, and it will be supported by our calculations.Below these frontier orbitals there are the other 17r, 4a, 3a, ...orbitals, which are involved very little in the chemisorption.

Let us first see what the band structure of a C O monolayerlooks like. Plummer and co-workers have measured the bandstructure for C O adsorbed on a Co(OOO1) surface at (v'3Xv'3)R30 (coverage 0 = monolayer) and (2v'3X2v'3)R3Oo (0=7/12 monolayer) and compared with their tight-binding calcu-lat ion~.~hese assumed only nearest neighbor interactions withab initio wave functions or CNDO-type wave functions. Corre-sponding to the above 0 = 0 = ' I l2coverages and a hypo-thetical 0 =1 coverage we carried out tight-binding calculationsof the extended Hii ckel type for hexagonal monolayer C O withnearest neighbor distances of 4.35, 3.29, and 2.51 A (the nearestneighbor C o and Ni contacts are 2.51 and 2.49A , respectively).From the calculated band structure, Figure 1, we see that at thedistance of 4.35 A , Figure la, the CO-CO interaction is negligible;at 3.29 A , Figure l b, the interaction is turned on; and at 2.51 A,Figure IC, he band widths attain several electron volts. Especiallythe27r band dispersion is large, over 4 eV. CO-CO interactionhas been discussed elsewhere, notably by Plummer and co-workers4and by A nder~on. ~n our subsequent computations of C O onmetal surfaces we will often choose a high coverage, for reasonsof computational economy. The resulting CO band widths willbe inherently high, due to CO-CO interactions. They will oftenbe higher than the experimental results, which are usually for lowercoverage.CO Chemisorption on the Ni(100) Surface

We eventually will compare the behavior of C O on varioussurfaces-Ni(100), N i ( l l l ) , Co(OOOl), Fe(llO), Cr(l lO ), andTi(0001). But let us begin with the best known system the c-(2X2)CO-Ni(100). The geometry of this surface, in two views,parallel and perpendicular to the surface, is shown in 2.

From top0 N I atom0 C atom0 0 atom

2

(9) Anderson, A. B.Surf . Sci. 1977, 62, 119.

J . Am. Chem. Soc., Vol. 107, No. 3, 1985 519

t t0 b

Figure 2. Projected DOS of a Ni(100) four-layer slab. The solid lineindicates s states in part a and p states in part b. The broken lines arethe total DOS in both parts a and b. The peak between -8 and -12 eVindicates mainly d states penetrated by s and p states.It is known experimentally that the CO molecule is chemisorbedwith its molecular axis normal to the surface and its carbon endcloser to the N i atom.I02l1 The C -0 bond length is 1.1 f 0.1

' 4 . 1 1In order to achieve the best compromise between time ofcomputation and accuracy of the model, we chose to use afour- layer-slab model of the metal surface. From previous workin our group,I2 reasonable convergence is reached for such afour-layer-slab model. A 15K point set13was used for the squarelattice. The CO molecules are adsorbed on one side of the slab.When CO molecules are adsorbed on both sides of the metal slab,the calculated electron densities on the inner-layer atoms aredifferent from those when C O molecules are adsorbed onone side.However, the electron densities on CO and on the surface metalatoms bearing CO, which we focus on in this paper, are verysimilar to those in the case of single-face adsorption. The Ni-Cdistance is taken as 1.8 &1 Further details of the computationsare given in the Appendix.Without adsorbate the clean Ni( 100) slab is characterized bythe calculated density of states (DOS) of F igure 2. The d statesform a band of -4 eV width; the s and p bands are much wider.Note the substantial penetration of the d band by the s, p bands.If we look at the detailed composition of, say, the surface layerof the slab, we obtain the orbital populations d9.370s0,618o.Ix6, otvery dif ferent from the usual assumption of d9s'. Note the surfacelayer is negatively charged. This has been given a simple ex-planation in a previous paper from our group'* and in the workof others.I4J5 Simply speaking, the surface states are less dispersedbecause the surface atoms have fewer nearest neighbors than bulkatoms do. I t fol lows that for d electron counts corresponding tomore than half-filling the d band, the surface is more negativethan the bulk, while for low-electron counts the surface shouldbe positive. The smaller dispersion of the surface states may beseen in Figure 3, which shows the fraction of the total slab DOSwhich is on the surface layers and that which is on the inner,~ ~(10) (a) Allyn, C.; Gustafsson, T.; Plummer, E. Solid Store Commun.1978, 28, 85 . (b) Passler, M.; Ignatiev,A,; Jona, F.; Jepsen, D.; Marcus,P.Phys. Rev. Lett. 1979, 43, 360.( 1 1 ) Anderson, S .; Pendry, J . Phys. Reu. Lett. 1979, 43, 363.(12) Saillard, J .-Y .; Hoffmann, R. J . Am . Chem. SOC.1984, 106, 2006.(13) Pack, J . D.; Monkhorst,H. J . Phys. Rev. E 1977, 16, 1748.(14) (a) Shustorovich,E.; Baetzold,R. C. J . A m . Chem.SOC. 980, 102,5989. (b) Shustorovich,E. J . Phys. Chem. 1982, 86 , 3114. (c) Shustorovich,E. Solid State Commun. 1982, 44, 567.(15) Feibelman, P. J .; Hamann, D. R. Solid State Commun. 1979, 31,413. Desjonqueres,M.C.;Spanjaard, D.; Lassailly, Y.; Guillot,C. Solid StareCommun. 1980, 34, 807.


3/7

580 J . Am. Chem.SOC.,Vol. 107, No. 3, 1985 Sung and Hoffmann

Projection of \surface otoms1

a

t

Projoclion ofinnor atoms

bFigure 3. Projected DOS of surface and inner layers of a Ni(100)four-layer slab. I n better, wlf-consistent calculations similar dispersingeffects are observed, but the surface states are more skewed toward theFermi level.

+== 53+== 1 -t----dL- - 2 IN I 11001 slob Ci2r21CO - N 11001 mono1oyer Of c0

Figure4. Total DOS of a Ni(100) slab, a c(2X2)CO-Ni(100) system,and a monolayer of CO.Table I . Calculation Results on c(2X2)CO-Ni(100)

Overlap PopulationsM-C 0.84c-0 1.04 (1.21 in free CO)

1.62 (2 in free CO)0.37' (0 in free CO)CO Electron Densities5a2R

Electron Density Changes on Surface Atoms"As -0.05API 0.17APT 0.04Ad, -0.50Ad, -0.500.03total -0.80energy change,bAE , eV -2.43

Electron density changes of those surface atoms having adsorbedCO on it. "E (N i slab and adsorbed CO) - E(N i slab and separatedCO) for one unit cell, Le., eight Ni atoms and oneCO. 'This is theelectron density in each of the 2x orbitals, that is, the two degenerate2 a orbitals gain 0.74 e altogether.bulk-l ike layers. In better, self-consistent calculations simil ardispersion effects are observed, but the surface states are moreskewed toward the Fermi level.Now let us see what happens when the c(2X2)CO-Ni(100)system is assembled. A t left in Figure 4 is the DOS of the nakedslab, at right the isolated CO layer, and in the center the compositesurface plus overlayer. The electron-density changes are bestindicated in tabular form (Table I).The density of states curve clearly shows that the major sur-face-adsorbate interactions involve the 5 0 and 2a orbitals of thecarbon monoxide. The surface N i d electron densities decrease

0 I bFigure5. Projected DOS of dz2 of (a) a clean surface layer and (b) thosesurface atoms having adsorbedCO. The increase of DOS above theFermi level in part b, compared to that in part a, corresponds to the0.50electron density decrease in dz2(see Table I ).-14.01' I I I

a bFigure 6. Projected DOS of (a) d, of those surface atoms which haveCO adsorbed on them and (b) 5a of CO in the c(2X2)CO-Ni(100)system. T he dotted line is an integration. Those states above the Fermilevel in part b illustrate the electron density decrease from 2 to 1.62 (seeTable I ).dramatically, in both the u and a components.The electron density in d,, or dZ2, f we take the normal of Nisurface as the Z direction, changes substantially. The dZ2orbitalsare pointed directly toward theCO 5 0 orbitals,3,and are at higher

U3

energy than C O 5 u . The interaction pushes the d,2 orbitals up,and consequently a greater portion of the d22 band ri ses above theFermi level. Since the d states are nearly completely fil led forthe clean surface, this interaction decreases the d22 electron density.The projected d,z DOS in the chemisorption system is shown inFigure 5 . Comparison with the naked surface clearly shows theshif t in density that is described above. We wil l see later thatas one goes from Ni to T i, the d orbitals are less populated forthe surface and the d, density changes consequently becomesmaller.There is no contradiction between the concept of strong met-a1420 5u orbital interaction, with consequent d2 (and to a smallerextent s) depletion, and the generally accepted notion of CO- to-metal electron transfer. The 50 loses electron density, its popu-lation going from 2.0 in the free molecule to 1.62 in the chemi-sorbed system. The states which are pushed up above the Fermilevel, and which we said originated from d,z, are in reality anantibonding mixture of dZ2 and 50 orbitals, and vacating them


4/7

How Carbon Monoxide Bonds to Metal SurfacesSurface + co Surface +coct- -401 i i,

1001-12 0 2lT t 2 T d W t d W

a b C dFigure 7 ProjectedDOS for (a)27r of a monolayerof CO, (b) 2~of thesameCO, adsorbed, (c) d, levels of those surface atoms having adsorbedCO, and (d) d, of the clean metal surface.depletes both metal d, and 5u of carbonyl. Figure 6 shows howthe metal d, and carbonyl 5u electron densities are correlated.A nother significant change in the electron densities of the N isurface is in the d, or dxlg2orbitals. These orbitals are pointeddirectly toward the CO 2a orbitals (with overlap 0.12 in afragment) and the l a orbitals (wi th overlap0.04). The laorbitalsof C O are bonding between C and0,with the larger componenton the oxygen atom and major electron density in the regionbetween C and 0. Their overlap with d, is much smaller thanthat of the 2a set. The energy of the la orbitals is several electronvolts below the corresponding value of the metal d electrons.Therefore, the 1~ orbitals are essentially not involved in metal 40interaction. The 2a set is quite different. These orbitals areantibonding between C and 0 with larger components on thecarbon atom and an energy very close to transition metal dxlyr,4. They interact strongly with metal d, or dxr,y2 rbitals.

4We can follow this a interaction in a number of ways. Firstwe see that 2a is populated by 0.37and d, depleted by 0.495 e.This is once again the result of interaction between the two leveltypes shifting some of the d, density above the Fermi level. Theprojections of the DOS before and after interaction (Figure 7)clearly show the buildup of density in one orbital in the energyregion of the other.A nother way to follow the involvement of the various orbitalsin bond formation is through crystal orbital overlap population(COO P) curves.I6 The COOP is really an overlap populationweighted density of states, Le., it weights the DOS in each energyinterval by its contribution to the overlap population. In Figure7 we saw in the d, and 2a projections a buildup of electron densityin two regions, at -10, -1 1and around -7 eV. We would suspectthat these regions correspond to metal-CO a bonding and an-tibonding combinations, respectively. Schematic drawings aregiven in 5 and 6. Note that whereas 5 and 6 differ in M-CO

5 6bonding, they are both C -0 antibonding. Figure 8 shows the

(16) Hughbanks, T.; Hoffmann, R. J . Am. Chem.SOC. 983, 105, 1150.

J . Am. Chem. Soc., Vol. 107, No. 3, 1985-anfibonding bondinq-

W

-10.0

112.0 .....I--....-14.01 I

S81

tFigure8. Crystal orbital overlap population in the c(2X2)CO-Ni( 100)chemisorption system.computed COOP curve for the surface plus adsorbate. Note thatthe curves, which are sums over all orbitals, nevertheless pick upthe features of the 2a and d, DOS and that signs of the overlappopulations follow the above simple picture. Thus the regionimmediately below c F is M-C bonding and C-0 antibonding, andthat above the Fermi level is antibonding M-C and C-0.Our general point of view, not at all different from that of otherworker^ *^ ^ is that 5 0 and 27 bind the molecule to the surface.These are the same forward- and back-donating interactions thatoperate in discrete transition-metal complexes. How is the C Obond then actually broken, following chemisorption or bonding?We do not trust the extended Hi ickel calculations sufficiently tocompute a reaction path. Instead we choose to concentrate onthe population of 2a, as others have d ~n e . ' ~~' ~resumably whenthe bond is sufficiently weakened by population of this CO an-tibonding orbital, at some point a further sideways motion ensues,with spli tting of C O, and coordination of separate C and 0 tothe surface. We will see that as we move from N i to Fe, Cr, andTi that population of 2a will rise significantly.The electron density change of dsfor d,2~~2nd dxy8 rbitals isvery small in this system. These orbitals have locally two per-pendicular nodal planes which cannot be found in any CO orbitals.Therefore they do not interact with their proximal CO. However,the 2a orbitals may interact wi th d6 sets of neighboring metalatoms, as indicated in 7. Of course, these interactions are weak.z

7The overlap between one C O 27 and one dX2-y2of neighboringmetal is 0.011 for Ni, and slightly l arger (0.020) for Ti.I t would be nice to have a simple interaction diagram for COon a surface, just the way we draw them for discrete transition-metal complexes. We do understand all the interactions, as theabove discussion shows, but still it is difficult to draw a simple

(17) Anderson,A. B.; Hoffman, R. J . Chem. Phys. 1974, 61, 4545.(18) (a) Kobayashi, H.; Y oshiba, S. ; amaguchi, M. Surf: Sci. 1981, 107,321. (b) Kobayashi, H.; Y amaguchi, M .; Y oshida, S.;Y onezawa,T. J . Mol.Catal. 1983, 22, 205.(19) Ray, N. K.; Anderson, A. B. Surf. Sci. 1982, 119. 3 5 .(20) Bagus,P.S.;Nelin,C.J .;Bauschlicher, C.W., J r.Phys. Reu.5. 1983,28 , 5423.


5/7

Chem.SOC.,Vol.107, No. 3, 1985l - l

I

META L SURFACE SURFACE t CO coFigure9. Principal interactions n the c(2X2)C&Ni(lOO) chemisorptionsystem. The solid lines indicate major contributions while dashed linespoint to weaker mixing. The open bars are meant to indicate that thereis a substantial number of states at stil l higher energy.

CO lC001) Fe Ill01 C , ( tho1 1 100011Figure 10. Calculated DOS of the surface layer of Co, Fe, Cr, andTifour-layer slabs. Solid lines are the projected d states and broken linesthe projected s states; the arrow indicates the Fermi level.diagram, one that is easily portable. Figure 9 is a try at one. Thereader is directed to the work of van SantenSc or an analysis thatbears substantial similarities to ours and also attempts to drawa generalized interaction diagram for C O chemisorption.Different Transition-M etal Surfaces

To compare the chemisorption behavior of different surfaceswe did calculations of C O on Ni ( oo) , Ni ( 11 ) , Co(OOOl), Fe-(1 lo), Cr(1 O) , and Ti(0001) surfaces. The first layer of thesesurfaces is shown in 8.

0 b C

88a is a two-dimensional hexagonal lattice, which represents thefi rst layer of N i (l 1 ), Co(OOOl), and Ti(0001), because the ge-ometry of the first two layers of fcc( l l l ) and hcp(0001) surfacesis the same. 8b is a two-dimensional oblique lattice representingthe bcc( 110) surface for Fe( 110) and C r( 110). 8c is a two-di-mensional square lattice representing the Ni( 100)surface. Theangles between the two lattice vectors in 8a, 8b, and 8c are 120'(or 60'), less than 90, and 90, respectively.In our calculations, a 14K point setI3 was used for the hexagonallattice of N i (l 1 ), Co(OOOl), and Ti(0001), a 12K point set13 orthe oblique lattice of F e(l l0) and Cr( l lo), and a 15K point setI3for the square lattice of the Ni(100) surface. The slab modelcontains four layers of metal atoms in each case. Figure 10 isthe calculated density of states curve of the surface layer of theCo, Fe, Cr , and Ti slabs. The solid lines are the DOS of the dstates and the broken lines those of s states. Note first that thed bands goup from N i to Ti, as do the Fermi levels. Meanwhilethe s band energies change very little. This isbecause the d orbitalHi:s go up while the s Hits change much less (see Appendix).Second, from Ni toTi the d band widths increase. This is because

Sung and HoffmannTable 11. Calculated Metal Surface Electronic Structure

surfacesTi(0001) Cr(l l0) Fe(ll0) Co(OOO1) Ni(100)

Electron Densities on Surface LayerS 0.80 0.78 0.71 0.68 0.62PO 0.06 0.07 0.07 3.06 0.04P1 0.14 0.28 0.21 0.20 0.15do 0.50 1.06 1.61 1.81 1.934 1.20 2.05 3.14 3.57 3.81db 1.21 2.12 2.8 1 3.20 3.63total 3.91 6.35 8.55 9.52 10.17

Fermi Levelstr, eV -6.47 -7.35 -8.17 -8.42 -8.59

the d orbitals are becoming more and more diffuse. The computedelectronic distribution of these surfaces is presented in Table 11.What are the differences when carbon monoxide adsorbed onthese surfaces? As in the c(2X 2)CO-Ni(100) system, we use afour-layer metal sl ab with C O adsorbed on one side, on top.Bridging adsorption will be discussed later.In order to save computing time a smaller unit cell containingfour metal atoms and one carbon monoxide molecule is used inthe calculations. This corresponds to full coverage, 0 =1, whilein the c(2X 2)CO-Ni(100) system we worked with 0 =0.5. Itis obvious that in the smaller unit cell the CO-CO interactionis overemphasized. The CO band widths are exaggerated. Thecalculational results are listed in Table 111. A comparison of theelectron densities and overlap populations of the small cell Ni-(100)-CO system in Table I11 with those for c(2X 2)C GNi(100)in T able I shows that the difference is very small, comparing withthe difference between dif ferent metals. This tells us the smallunit cell simplification is adequate, and it also informs us thatcoverage has much less effect on surface-adsorbate interactionsthan the electronic nature of the surface.Let us analyze the trends as one moves from a N i to a Ti surfaceinteracting with carbon monoxide. The d bands move up in energyalong that series, as Figure 10 showed. Thus, the interactionbetween d, and 5a of C O should decrease, because 5 0 is belowthe metal d bands. A lso the d electron count decreases, thereforeAd, gets smaller in absolute value. Meanwhile, thes bands changevery li ttle in energy and the s electron number is even larger (seeTable 11), therefore As increases in absolute value. However, theAd, remains almost unchanged at about -0.3. Unlike d,i and sorbitals, which interact mainly wi th CO 5a orbitals, d, or dxr,yrorbitals interact mainly with CO 27r orbitals. 5a orbitals are lowerin energy than metal orbitals, but 27r orbitals are higher and closeto the metal orbitals, especially to the d bands.lh4s the metal dbands goup from N i to T i, their interaction with 50 gets smaller,but their interaction with 27r increases. A nother factor is the bandwidth of the d states, which increases from N i to Ti, and thusallows mixing with 277 orbitals over a wide range. Furthermore,the Fermi level is rising from N i to making electron transferto 2 r of CO easier. A ll of these factors reinforce in a naturalway.It should be noted that this discussion of the special role of C O27r and its energy relative to the metal band is not original to us.It has been made by others, particularly in the work of Anderson.22Let us look at the extreme case, the Ti (0001) surface, in somedetail. Figure 11 shows the part of theDOS that is on CO 27ron that surface. The lower part of the d band contains most ofthe 27r density. In contrast in Ni(100) we saw that the major partof the DOS of 27r is above the Fermi level.

(21) It should be noted that our calculations do not reproduce wel theabsolute value of the metal work functions. These are 4.3 eV (Ti), 4.5 eV(Cr),4.5eV (Fe), 5.0eV (Co), 5.2 eV (Ni) for polycrystaline films: Eastman,D. E. Ph ys. Rev. 1970, 82 , 1.(22) Ray, N . K.; Anderson, A. B. J . Phys. Chem. 1982, 86 , 4851; Surf.Sci. 1983, 125, 803. See also: Russell, J . W.; Overend, J .; Scanlon, K .;Severson, M.; Bewik, A. J . Phys. Chem. 1982, 86 , 3066. Russell, J . W.;Severson,M.; Scanlon,K .; Overend, .; Bewik,A. J .Phys. Chem. 1983,87,293. BrodSn,G.; Gafner, G.; Bonzel, H. P. Sur5 Sci. 1979, 84 , 295.


6/7

How Carbon Monoxide Bonds to Metal Surfaces J . Am. Chem.Soc., Vol. 107, No. 3, 1985 583Table111. Some Computational Results of CO Chemisorption Systems

substratesTi(0001) Cr(1 O) Fe(110) co(ooo1) Ni( 100) N i(l l1)

Overlap PopulationsM-C 1.11 0.93 0.91 0.83 0.78 0.75C-O 0.43 0.87 0.96 1.01 1.03 1.025u277

CO Electron Densities1.73 1.67 1.62 1.60 1.60 1.591.61 0.74 0.54 0.43 0.39 0.40Electron Density Changes on the Surface Layer"As -0.18 -0.18 -0.14 -0.13 -0.07 -0.08APo 0.15 0.17 0.15 0.15 0.15 0.14APT 0.03 -0.05 0.01 0.00 0.03 0.01Adc -0.01 -0.16 -0.57 -0.66 -0.53 -0.65Ad, -0.31 -0.39 -0.31 -0.33 -0.43 -0.36Ad, -0.26 -0.23 -0.07 -0.09 -0.00 -0.03total -0.57 -0.84 -0.92 -1.06 -0.85 -0.96

Other Layerstotal -2.29 -0.17 0.37 0.75 0.60 0.72AE,b eV -6.11 -3.44 -2.64 -1.98 -1.97 -1.66

'Electron density changes of the surface layer on which the CO's are adsorbed. These numbers are the differences from those of Table 11.*E(adsorptionsystem) - E,,,, +Eco) per unit cell (four metal atoms and one CO).100%

-12.0 DOS-igure 11. Projected DOS of 2 H of CO on the Ti(0001) surface. Thedotted line is the integrated DOS of 2H. The broken line is the totalDOSof the chemisorption system.I n summary, the population of carbon monoxide 2a i ncreasessteadily from N i to Ti . The C O bond is much weakened, itsoverlap population down to 0.43onTi(0001) from 1.02onNi(100)and 1.21 in free CO. W e would anticipate a greater ease of C Ocleavage as one moves to the left in the transition series, and indeedthis is what is observed.

Different Surfaces of the Same M etalTable I11 allows a comparison of the adsorption of C O onNi( 11 1) vs. N i( 100). We find that the (1 11) surface weakensthe C-0 bond a little more than the (100) surface. This is inagreement with some ex ~er i men tal ~~,~~nd theoreticallsb studies.However, the matter is complicated because different adsorptionsites are involved in the experimentally observed species. One ofthe reviewers of this paper pointed out that the trend in C -O bondweakening might have been expected to be (100) >(11 ), sincethe work function of low-index surfaces is smaller than that ofhigh-index ones.25 A lso in the case of a stepped surface, theextreme case of a low-index surface, there are indications ofsubstantial CO weakeningz6 We are not certain here of the(23) Iwasawa, Y. ;Mason, R.; Texter, M.; Somorjai, G .A. Chem. Phys.L ett. 1976, 44, 468.(24) (a ) Brodtn, G.; Pirug, G.; Bonzel, H. P.Chem.Phys. Lett. 1977,51,250. (b) Andersson,S . Solid State Commun. 1977,21, 75. (c) Bertolini, J .C.;Dalmai-Imelik, G.; Rousseau, J . Surf: Sci. 1977,68,539. (b) Compuzano,J . C.;Greenler, R. G. Surf. Sci. 1979, 83, 301.(25) Holzl, J . I n 'Solid Surface Physcs" ,Springer-V erlag: Berlin, 1979;p 75.

overall strength of either the experimental conclusions or ourtheoretical results. What is clear is that the difference betweendifferent surfaces of a given metal is less than that arising fromchanging the metal.Different Adsorption Sites

For CO adsorption onNi( 111) it is reported that CO is boundmainly in a bridging po ~i ti o n , ~~~,~hile for C O on Ni( 100) it isknown that the relative proportion of bridge-bonded and on-topC O is strongly dependent on adsorption temperature and cover-age.24b The extended-Huckel method is not very good at deter-mining bond distances, and the comparison by this method of twoisomers which differ substantially in geometry is problematic.Nevertheless, we think the topology of the interactions and theirmagnitude is correctly given by this approximate procedure. Thuswe proceeded to carry out calculations for twofold bridge andthreefold bridge site adsorption on Ni( 11 ) , 9 and 10.

Two-fold bridge Three-fold bridge9 10

0 co0 "The computational results are shown in Table IV , which includethe on-top adsorption result for comparison. I n each bridge site

we did two calculations, one at a Ni-C distance of 1.80A theother at a perpendicular carbon-surface layer distance of 1.80A. Compared to on-top adsorption, the metal surface d,2 electrondensity changes are much smaller, while those of the dxly, set arelarger. This is because on bridge-site adsorption the metal d22orbitals are no longer pointed directly at CO. Instead d,,, dy,orbitals are now oriented for interaction. Meanwhile Ads-?, Ad,become larger in absolute value because of larger overlap withCO orbitals.In the C-surface perpendicular distance 1.80-A case, it hasalmost the same 2a electron density and the C-0 overlap popu-lation as in the case of on-top adsorption. Unfortunately, the result(26) Erley, W.; Ibach, H .; Lehwald, S.; Wagner, H.Surf. Sei. 1979,83,585.


7/7

584 J .Am. Chem. SOC., ol. 107, No. 3, 1985 Sung and Hoffmannfor terminal and bridged carbonyl ^ Our feeling is that thebridge-adsorbed C O should have a Ni-C distance between thetwo cases li sted in Table IV . Consequently, the bridge-adsorbedcarbonyls should have a higher 27r occupation and a weaker C Obond.Conclusion

When CO molecules are adsorbed in on-top sites on transi-tion-metal surfaces, the CO 5 0 orbitals interact with metal d,zandsstates and C O 2x orbitals interact with metal dxryz. Theseare the same interactions that occur in discrete transition-metalcarbonyls. There is li ttle mixing of these two types of orbitals,and indeed CO low-lying orbitals and metal p orbitals do not getmuch involved. The metal-CO 27r back-bonding determines thedissociative behavior of the carbonyl. From Ni to Ti the Fermilevel rises, the d orbitals become more diffuse, and as a conse-quence the CO 2 r evels gain more and more electron density.This is why C O dissociates on the early transition metals.The coverage of C O has li ttle effect on CO dissociation. Inour calculations the higher index surface would seem to have atendency to dissociate C O more easily than the lower index one.However, this is a secondary effect compared to the electronicnature of the metal itself.

Acknowledgment. We are grateful to Sunil Wijeyesekera,J ean-Y ves Sail lard, ThorN. Rhodin, John Wi lkins, M.-H. Tsai,Gayanath Fernando, and Cyrus Umrigar for suggestions anddiscussions, to J ane Jorgensen for the drawings, to Eleanor Staggfor the typing, and to the reviewers of this paper for their com-ments. Our research was supported by the National Sci enceFoundation through Research Grant C H E 7828048 and the Officeof Naval Research.Appendix

All the calculations were of the tight-binding extended-Huckeltype. The Hi is for transition metals were obtained from earlierwork.12 They were determined by charge iteration onbulk metals,assuming the charge dependence of metal Hii)sgiven by Gray'sequations.28 The A , B, andC iteration parameters were takenfrom ref 29. Experimental hcp, fcc, and bcc structures were used.TheH,'s of C and0were determined by charge iteration on COadsorbed on an Fe( 110) slab of four layers, iterated three cycles.The iteration parameters were taken from ref 29 also. Extend-ed-Huckel parameters for all atoms used are listed in T able V.The geometrical parameters for chemisorption systems are thefollowing: C -0 distance 1.15A; C-metal atom distances for Ni ,Co, Fe, Cr, and Ti are 1.80, 1.82, 1.84, 1.92, and 2.03 A, re-spectively.

RegistryNo. CO, 630-08-0; Ni , 7440-02-0; Co, 7440-48-4; Fe, 7439-89-6; Cr, 7440-47-3; Ti, 7440-32-6.

Table IV. Comparison of Different Adsorption Sites on the Ni( 111)Surfacebridge adsorption

d, =1.80& dNix = 1.80Adgeometry on top 2-fold 3-fold 2-fold 3-fold

Overlap PopulationsNi-C 0.76 0.31 0.16 0.57 0.41C-O 1.02 1.02 1.02 0.97 0.93CO Electron Densitysa 1.59 1.57 1.54 1.57 1.562n 0.40 0.41 0.41 0.54 0.64

Electron Density Changes in the Surface Layer'As -0.08 -0.14 -0.14 -0.15 -0.16APZ 0.14 0.11 0.07 0.12 0.07APX, 0.01 0.01 0.04 0.02 0.08Adrz -0.65 -0.39 -0.33 -0.29 -0.28AdxyT2-yz -0.03 -0.13 -0.11 -0.43 -0.47

Other Layerstotal 0.72 0.61 0.64 0.94 0.78AE,beV -1.66 -0.89 -0.71 -2.72 -3.24

",*As n Table 111. CThedistance between the carbon atom and themetal surface is 1.80 A. In these geometriesdNix =2.19 and 2.3Afor the 2-fold and 3-fold bridge case, respectively. dThe distance be-tween the carbon atom and the nearest metal atom is 1.80A. In thesegeometriesd, =1.30 and 1.08A for the 2-fold and 3-fold bridge case,respectively.Table V. Extended Hiickel Parameters

Ad,$ -0.36 -0.33 -0.44 -0.66 -0.66total -0.96 -0.87 -0.89 -1.39 -1.43

orbital Hli ,eV CZ CI' c2.Ti 4s -6.3 1.50Ti 4p -3.2 1.50Ti 3d -5.9 4.55 1.40 0.4206 0.7839Cr 4s -7.3 1.70Cr 3d -7.9 4.95 1.60 0.4876 0.7205Fe 4s -7.6 1.90Fe 3d -9.2 5.35 1.80 0.5366 0.6678c o 4s -7.8 2.00CO4p -3.8 2.00,Co 3d -9.7 5.55 1.90 0.5550 0.6460Ni 4s -7.8 2.1Ni 3d -9.9 5.75 2.00 0.5683 0.6292c2s -18.2 1.63c2P -9.5 1.630 2s -29.6 2.270 2p -13.6 2.27

Cr 4p -3.6 1.70Fe 4p -3.8 1.90

Ni 4p -3.7 2.1

a Contraction coefficients used in the double-{ expansion.depends on the C-surface distance. A t C-Ni =1.808,the bridgesite properties are quite different from those of the on-top site.For one distance the bridge adsorption energy is small er and forthe other distance larger than for on-top binding. This is dis-couraging, for we cannot decide if bridge bonding is favored. Innickel carbonyl complexes the Ni-C distances are very similar

(27) Ruff, J . K.; White, R. P., J r.; Dahl,L . F. J . Am. Chem. SOC. 971,93, 2159. Longoni, G.; Chini,P. ;Lower, L. D.; Dahl, L. F. J . Am. Chem.SOC. 975, 97, 5034.(28) Ballhausen, C. J .; Gray, H. B. "Molecular Orbital Theory"; W. A.Benjamin, Inc.: New Yrok 1965;p 125.(29) McGlynn,S.P.;Van Quickenborne, L. G.; Kinoshita, M.; Caroll, D.G. "Introduction to Applied Quantum Chemistry": Holt. Rinehart and Win-ston, Inc.: New York, 1972.

Documents

How Carbon Monoxide Bonds to Metal Surfaces