Analysis of energy level shifts in the photoemission spectra of adsorbed molecules: CO on Ni

Solid State Communications, Vol. 20, pp. 5—8, 1976.

Pergamon Press.

Printed in Great Britain

ANALYSIS OF ENERGY LEVEL SHIFTS IN THE PHOTOEMISSION SPECTRA OF ADSORBED MOLECULES: CO ON Ni P.S. Bagus and K. Hermann IBM Research Laboratory, San Jose, CA 95193, U.S.A. (Received 24 December 1975; in revised form 13 May 1976 by A.A. Maradudin) Self-consistent Hartree—Fock calculations on a linear NICO cluster have been used to analyze and interpret the experimental relative binding energy shifts of CO upon adsorption on a Ni surface. Chemical shifts due to the changed environment of the adsorbed molecule are found to contribute significantly to the initial state shifts of nonbonding levels. The final state or relaxation shifts are quite different among groups of levels;bonding, valence non-bonding, and core. THE ASSIGNMENT of the molecular origin of the photoemission peaks of chemisorbed molecules is important for the understanding of the nature of surface bonds formed. A partial survey of analyses for CO adsorbed on Ni indicates how difficult such an assignment may be. The first assignment of the UPS (ultra. violet photoemission spectroscopy) observations14 indicated that the adsorbate So and lv derived levels were shifted nearly uniformly from their gas phase positions and that the 4o level was not detected. Several theoretical studies~8of CO adsorbed on Ni showed that the bonding 5 a level was shifted downward by several eV with respect to the nonbonding (or wealdy bonding) lv and 4o levels. This relative shift was confirmed from variable photon studies9 which clearly supported a reassignment of the experimental levels, Several distinct origins of the observed shifts have been considered. The first is a relaxation shift, E~ER= ER (adsorbate) ER (gaseous). This shift isdue to final state effects and is normally positive (i.e. lowers the observed binding energy). It arises from the fact that charge flow toward the positive hole, polarizabiity of the surrounding medium, etc. are all larger for the adsorbed molecule than for the gaseous system. The second shift arises from the factthat certain of the molecular orbitals (MO’s) may form chemical bonds with the substrate. This bonding shift, AEB, is essentially an initial state effect. For several hydrocarbons adsorbed on Ni, Demuth and Eastman10 assumed (for the valence levels observed in UPS spectra) that I~ERis constant and

It is well known that the potential associated with this change can cause significant shifts in the levels of core orbitals of atoms in free molecules’~and condensed matter1~(this is sometimes referred to as initial state screening). The importance of this effect in the case of adsorbed molecules has been pointed out in a theoretical study by Madhukar and Bell.’2~’ In order to study the origins and behavior of the adsorbate shifts, we have performed ab initio Hartree— Fock calculations on a linear NiCO “cluster”. We have obtained fully self-consistent wave functions and energies for the relaxed final ionized states as well as for the initial states. Clearly, NiCO is too simple a model system to provide quantitative results for the actual adsorption of CO on a Ni surface. However, we can obtain rigorous and detailed information about the contributions to the level shifts in this model system. In particular we find that the size of the chemical shift for the nonbonding core and valence levels of CO is comparable to the binding shift of the So level. Further, the initial state and final state shifts show significant variations among different levels. We expect the qualitative conclusions to be valid for the real situation. In the following the computational details of the model are described before the conclusions drawn from the numerical results are presented. The interatomic distances for the linear NiCO cluster model are chosen to be the experimental values for the Ni(CO) 13 4 molecule, dNj_c = 3.477 a.u. and d~_ø= 2.173 a.u.,respectively. This dc,.. 0 was also used for calculations on the free CO molecule.5”~that It is known from other cluster calcuthe general features of the Ni—CO bond 1ations are relatively insensitive to the bond distance for a reasonable range of Ni—C distances. Further, we find that the qualitative behavior of the initial state shifts does not change when dNj_c is varied for a reasonable range. Hartree—Fock calculations were performed using

—

E~EBis

non-zero only for the bonding v-levels. In this way, they were to interpret the spectra deduce reasonable valuesable of the chemisorption bondand energies. Brundle” has given a critical review of the application of these shifts in a variety of cases. Finally, there is an additional initial state shift namely that due to the changed chemical environment of the adsorbed species, —

5

6

ENERGY LEVEL SHIFTS IN Co ON Ni

Table 1. Computed ionization potentials of the CO like states of the NiCO cluster. The notation ~ refers to the ftvzen orbital IF;~ is obtained by taking differences of theHartree—Fock total energies of the initial andfinal states State

~

L0(o~) 2ci(CLc)

563.54 310.81 42.11 23.01 18.64 18.28

30 40

5a Fir

(cv)

,

~

541.99 297.69 38.50 20.99 16.40 16.01

lo (OIs) 562.42 2o (CIs) 309.39 3a 41.08 4a 21.80 So 15.15 hr 17.12 a See reference 17

~r’

~evj

541.95 297.55 38.31 19.87 13.52 14.96

____________________________________________

State

i~E~• (cv)

i~E1(eV)

I~iER(eV

la

—0.04 —0.13 —0.18 —1.12 2.87 —1.05

—1.11 —1.41 —1.03 —1.21 3.49 —1.15

1.07 1.28 0.85 0.09 0.62 0.10

—

,-

rpAa~

Table 3. Calculated level shifts of the CO states due to the interaction with the Ni atom. ~E1 and ~ER are, respectively, initial and final state contributions to the total shift ~E~0t

IP~~ (eV)

Table 2. Computed tials of and theexperimental free CO molecule ionization poten~,tate

Vol. 20, No. 1

~eV)

Irexp

~r-~a

~ev~,

542.3 296.2 38.9 19.8 14.0 17.2

extended Gaussian basis sets.14 It is unlikely that the results will change significantly if larger basis sets are used, From calculations on possible configurations for the NiCO cluster we found that the lowest energy state is ~ with open shell structure 12o’(Ni4s) 163(Ni3d) resulting in a 3d94s1 like electronic structure of the Ni atom in the cluster. The N1CO orbitals were compared with those obtained (with the same C and 0 basis functions) for free CO. We find that the CO orbitals retain much of their molecular character in the NiCO cluster. Thus, we

—

15 usual experimental interpretation of work function changes. The calculated ionization potentials (IP) of the CO like states in the model cluster are shown in Table 1. Here ~ refers to the frozen orbital value as given by the orbital energy of the respective one electron state: IPp~~~ has been determined from the difference of the total energies of the cluster in the ground state and the respective ionized state.’6 The IP’s of the free CO molecule are given in Table 2. This table also includes experimental values for the IP’s determined by XPS17 (X4~iiyPhotoelectron Spectroscopy). The We,~,. agree fairly well with the calculated values including relaxation. The vertical IP’s reported in Tables 1 and 2 are the appropriate quantities to compare with the maxima of the observed peaks or (in the case of free CO) bands. Geometry changes in the final state lead, through Franck—Condon overlap, to vibrational (phonon) broadening of these peaks. Table 3 shows the calculated energy shifts of the CO levels when the molecule interacts with nickel: Co ,~iCO = 1Ppeiaxe~,j Iri~x~,j I ~E 1 = ~ IPf~J~lIWIS (2) 3 —

—

denote these cluster levels by their CO notation with a tilde added: ~i, fl~r,etc. A Mulliken population analysis shows that the CO core states f~j to S~rremain nearly unchanged in the presence of the Ni atom. The NiCO hr state shows no bonding tendency and the ~i state becomes slightly antibonding to the Ni atom. In contrast, the ~, state contains 13% Ni 3d character and therefore is the most bonding of the CO states. The singly occupied Ni 4s-like MO has a large, 27%, admixture of Nip character such that this MO is directed away from CO, and in effect, removes charge from thebond, regioninbetween Ni and C. is Altogether, the Ni—CO the present model, formed by the interaction of the Ni d orbital with the So orbital of the CO molecule. The gross atomic popu. lation analysis of the NiCO cluster gives an essentially neutral CO molecule. This is in agreement with the

L~ER

=

L~Etht

—

L~E~0~ is the only quantity which may be actually observed in the adsorption experiment. The initial state contribution, z~E1,has two origins; a bonding shift and a chemical or environmental shift as discussed above. The final state contributions to ~ are denoted L~lE~ since they are due to changed relaxation between the free molecule and the cluster. The comparison of~~E1for the different levels gives negative values of 1 eV for all18but Forthe thebonding CO coreSo where ~E1 considerably larger. states, f& toisS~, which do not participate in the Ni—CO bond t~~E is1exclusively a chemical shift. For the valence states, there is no rigorous way to make a quantitative separation between the bonding shift and the chemical shift. —

Vol. 20, No. 1

ENERGY LEVEL SHIFTS IN CO ON Ni

Ionization from a nonbonding valence level, 4~,or

f~,leads to a small i~ERof 0.1 eV, ionization from the bonding ~, level leads to a larger 1~ER= 0.6 eV. The physical origin of L~E~ lies in the response of the Ni electrons to the removal of a (dominantly) CO electron, For the non-bonding valence levels, the response will be small since these orbitals are quite diffuse and reasonably localized on the oxygen atom. The electronic reorganization upon ionization will be localized about 0 and the Ni electrons, screened by C, will not participate to a significant extent. Since ~r is the bonding orbital with Ni, the Ni electrons will participate in the electronic reorganization when this level is ionized; thus the larger value of L~ERis quite reasonable. The values of L~ERfor core level ionization are relatively large, about 1 eV, since the core levels are quite contracted. In fact, the removal of a f~i(2~r)electron may be regarded as increasing the effective charge of the oxygen (carbon) nucleus by one. So the removal of a core electron leads to a reorganization of charge over the entire cluster including the Ni electrons and therefore yields large values of I~ER. It may be seen from Table 2 that ER (ER = ~ IP~~) for the So level of the CO is 0.5 eV smaller than the nearly equal ER for the 4a and lv levels. The larger value of I~ERfor ~‘~r has the effect of making the E~for ~r and fr nearly equal in NiCO (c.f., Table 1). Thus it might seem that the nonuniform values of L~ERhave their origin in the CO molecule rather than the NiCO cluster. This, however, is not the case. part, The variation of ERthat in ER the for valence levels of CO arises,in from the fact the 2P ionization of the C atom is smaller than that for the 0 atom.19 The —

‘~,

variation of L~ERin the cluster is due, as noted above, to the greater participation of ~theionization Ni electrons in upon the electronic reorganization upon than 4u or laAionization, quantitative comparison of the total level shifts ~ calculated in the present model with data obtained from photoemission experiments is not possible. First, the relaxation contribution 1~ERmust be strongly underestimated since the Ni surface is repre. sented by only a single atom. Even the binding contn bution to the shifts can be somewhat different due to a different binding situation of the CO molecule on a real Ni surface. In addition, correlation effects are not included in our Hartree—Fock calculations. However, we believe thatseen the in essential mechanisms which the level shifts photoemission spectra arecause contained in this model, The present model calculation for the interaction of CO with Ni suggests that the level shifts of the CO states mainly consist of th~eecontributions: (a) an environmental or chemical shift, (b) a binding shift, and (c) a

7

Table 4. Experimentallevel shifts of the CO states due to the interaction with a polycrystalline Ni surface. The vahes have been detennined by comparing the IFS data offree CO. reference 17, to those for CO adsorbed, reference 19. To obtain a reference to vacuum, 5.5 eV has been added to the adsorbedresults to account for the workfunction State f~ ~a S& ~ ~a 1 ___________________________________________ ~ (eV) 5.4 5.1 3.2 2.2 3.6 —

relaxation shift. Further, the data in Table 3 suggests that it may be possible to assign the adsorbate CO derived levels to three groups with the levels in each group behaving in a similar way. The first group contains the core f~r to ~r levels. These non-bonding levels have significant L~E1due to chemical shifts and large dAER The second group contains the nonbonding ‘~&and lv levels. The i~E1for this group are comparable to those for the core levels but the I~ERare much smaller. The third group contains only the bonding ~i level, which behaves differently from all others for both z~E1and L~&ER.This grouping is also consistent with the observed ~ for CO on 20 given in Table 4. However, our values of i~E~,t are too small by 5 eV. As noted above, this is probably due in large part to the strong underestimate of I~ERin our model. The ~i level has not been measured either by XPS or with synchrotron radiation. An experimental value for the ~r IP would be most useful for testing the description of the origin of level shifts proposed here. 21 have used Demuthand that /.~ER ~was and Yu et aLfor the valence the assumptions constant levels and i~E 1(in our notation) was zero for the nonbonding valence levels to estimate heats of adsorption of 10 hydrocarbons on metal surfaces. reported reasonable results for C Demuth and Eastman 2H2there and was C2H4 on or Ni;no 2’ reported that little however, Yubetween et al. the results obtained in this way for correlation C 2H2 and C2H4 on Fe, Ni, and Cu with other values for the heats of adsorption. If we generalize the conclusions drawn from our calculations for CO on Ni, there are two important points to be made in connection with this and other interpretations of the UPS spectra of chemisorbed molecules. First, L~ERmay not be constant for the valence levels. Second, chemical shift effects may be significarit contributions to E~E1leading to non-zero values for the nonbonding valence levels. Although these points 2O,2l this work provides havefirst been raised before,lO~t~, the direct theoretical evidence that they represent significant effects which must be taken into account in the analysis of photo-emission spectra ~

Acknowledgement We are grateful to C.R. Brundle for many helpful discussions. —

8

ENERGY LEVEL SHIFTS IN CO ON Ni

Vol. 20, No. 1

REFERENCES 1. 2.

EASTMAN D.E. & CASHION J.K.,Phys. Rev. Lett. 27, 1520 (1971). PAGE P.J., TRIMM D.L. & WILLIAMS P.M., .1 Chem. Soc., Faraday Tran& 170, 1769 (1974).

3.

JOYNER R.W. & ROBERTS M.W., J. Chem. Soc., Faraday Tran& 170, 1819 (1974).

4. 5.

BECKER G.E. & HAGSTRUM H.D., .1. Vac. Sci. Technol. 10,31(1973). WABER I.T., ADACHI H., AVERILL F.W. & ELLIS D.E., Proc. 2nd mt. Conf ofSolid Surfaces, p.695. Kyoto, Japan (1974). BLYHOLDERG.,J~Vac. Sc. TechnoL 11,865(1974).

6. 7. 8.

BATRA LP. & ROBAUX 0.,.!. Vac. Sci. Technol. 12, 242 (1975); BATRA I.P. & BAGUS P.S., Solid State Commun. 16, 1097 (1975). CEDERBAIJM L.S., DOMCKE W., VON NIESSEN W. & BRENIG W., Z Phys. B21, 381 (1975).

9.

GUSTAFSSON T., PLUMMER E.W., EASTMAN D.E. & FREEOUF J.L, Solid State Commun. 17,391 (1975). 10. DEMUTH J.E. & EASTMAN D.E., Phys. Rev. Lett. 32, 1123 (1974). These authors include certain initial state effects in their defmition of ~ER. 11. BRUNDLE C.R., Pmc. ofthe NA TO Advanced Study Institute on Electronic Structure and Reactivity of Metal Surfaces, Namur, Belgium (1975) (to be published). 12. (a) SIEGBAHN K. et aL ESCA Atomic, Molecular and Solid State Structure Studied by Means ofElectron Spectroscopy. Uppsala, (1967);(b) CITRIN P.H. & HAMANN D.R., Phys. Rev. 10,4948 (1974): (c) MADHUKAR A. & BELL B., Phys. Rev. Lett. 34, 1631 (1975). 13, WYCKOFF R.W.G., CrystalStructures, 2nd Edn,, Vol. II, p. 143. Interscience, New York (1964). ,

—

14.

The basis sets for C and 0 were taken from VAN DUIJNEVELDT F.B.,IBMRes. Rept. RI 945 (1971) to which ad exponent of a = 1 [followingROOS B. & SIEGBAHN P., Theor. Chim. Acta 17, 199 (1970)] has been added. The 8s, 6p, and ld functions were contracted to (4, 3, 1>. The basis set for Ni was taken from WACHTERS A.J.H., J. Chem. Phys. 52, 1033 (1970), to which p exponents a = 0.228 and 0.08 were added to represent 4p character: 14s, lip and Sd functions were contracted to (8, 6, 3).

15. 16.

17.

TRACY J.C.,J. Chem. Phys. 56, 2736 (1972). Since the initial state is a open shell state, IP~opm~ refers to a final state which is the weighted avera~eof the possible fmal doublets and quartets. The final ionic Hartree—Fock calculations were all performed for ~ states. However, the exchange integral between the open shell Ni MO’s and the CO like MO’s is likely to be small. Thus, the doublet—quartet multiplet splitting will also be small. THOMAS T.D.,J. Chem. Phys. 53, 1744 (1970).

18.

CEDERBAUM et al., reference 8, obtained similar L~E1for the

19. 20.

BAGUS P.S., BATRA I.P. & CLEMENTI E., Chem. Phys. Lett. 23,305 (1973). BRUNDLE C.R. & CARLEY A.F., Faraday Disc. 60 (to be published).

21.

YU K.Y., SPICER W.E., UNDAU I., PIANETTA P. & UU S.F.,.!. Vac. Sci. Technol. 13,277 (1976); Surf Sci. (to be published).

~

to i~rlevels.