Different Adsorption States on Single SubstrateBY ROBERT GOMERDept. of Chemistry and Institute for the Study of Metals,The University of Chicago, Chicago, Illinois 60637Receiued 3 1st Jdnuary, 1966The adsorption of potassium and of CO on tungsten is examined with a view to defining thesimilarities and differences between metallic and covalent adsorption on the same substrate. Itis concluded that metallic adsorption shows no distinctly different adsorption states, variationsresulting from differences in substrate geometry and work function being quantitative rather thanqualitative, while covalent adsorption shows definite, discontinuous changes in adsorption type,which can be correlated with differences in geometry and binding mode. In both cases, however,substrate-adsorbate geometry appears to be important even though the effect in metallic adsorptionbecomes obvious only after some dissection of the factors entering the heat of adsorption.It is the purpose of this paper to examine two systems more or less at the oppositeend of the chemisorption spectrum, CO on W, and K on W, in order to see whatcan be learned from the different ways in which the same substrate is able to interactwith these different adsorbates.In particular, we shall attempt to see how andin what way substrate structure appears to affect adsorption in these systems, andwhether any generalizations may be drawn from this. This task is complicated bythe fact that only averages obtained on polycrystalline substrates are available forthe COfW system, so that interpretation must proceed with caution.In thefollowing we shall attempt first, to summarize briefly the salient results for eachsystem, how they were obtained, and their interpretation in terms of adsorptionstates. Finally, we shall turn in a qualitative way to the question of a coherenttheoretical interpretation.POTASSIUM ON TUNGSTENWe discuss first the K on W system because the contrast between measurementsof averages on polycrystalline or poly-faceted substrates with measurements onsingle crystal planes emphasizes where and to what extent caution is required ininterpreting the CO results. The average contact potential, heat of adsorption andsurface diffusion behaviour of K on W were investigated as a function of (absolute)average coverage by Schmidt and Gomer,l using field emission.The method con-sisted of evaporating K on to a tungsten field emitter, determining the absoluteflux by means of an auxiliary surface ionization detector, and then carrying outwork function and Arrhenius measurements in the usual manner. More recentlythese authors have looked at K adsorption on the 110, 211, 100 and 1 1 1 regions ofa field emitter using a probe hole, Faraday cage and rotatable emitter, which per-mitted measurements of a small portion of the total beam, corresponding to emissionfrom single crystal planes. This work is discussed in detail elsewhere?Measurements were carried out both for immobile adsorption (substrate at78°K during and after adsorption) and after thermal equilibration. After deter-mining the impinged K flux, the former measurements yielded work function againstatom density on individual planes.The flux calibration was carried out by de-1R. GOMER 15positing a given number of standard doses, equilibrating this layer by heating to400"K, obtaining its average work function 6 from the total emission, and finallycomparing the latter with the 6 against Pt (average absolute atom density) curveestablished in ref. (1). The value of E per dose thus found could then be con-verted by a simple calculation into the actual flux per dose impinged on the regionunder study. Measurements of 6 and of # on single planes after thermal equi-libration yielded values of # against i, so that combination with the # against ndata yielded values of n/Z against i.For convenience, we also use a non-dimen-sional coverage given in terms of an arbitrarily defined monolayer density,l no =3.9 x 1014 atoms/cm2, as 8 = n/no. Fig. 1 shows curves of # against 8 ; aboveFIG. 1.-Work function 4 againstK atom density n and fractionalcoverage 8 (defhed as 4 3 . 9 ~ 1014)for various planes of thermallyannealed tungsten field emitter.1.20 (monolayers)LLl.*III""n x 10-14 (atoms cm-2)1.0 20 30 40 508 = 0.6-0.8 immobile adsorption leads first to plateaus and then to monotonicdecreases in the apparent work function. This suggests incipient multilayer forma-tion with a very disordered adsorbate structure and consequent microscopic fieldenhancement as well as the possibility of genuinely decreased work functions.Inany case, the results are not considered trustworthy in this region. However, then/% against 8 values shown in fig. 2 indicate that the n/Z values for all regions except100 approach unity above 8-0.6. Consequently the 4 against 6 curves shownin fig. 3 must be close approximations to the correct # against n curves for 050.6,and have been so interpreted in computing dipole moments. These are definedin terms of the contact potential A#, the adsorbate charge q and its distance dofrom the image plane byA 4 = 2npn = LCnqdon, (1)i.e., asp = 2qdo16 DIFFERENT ADSORPTION STATESFIG. 3.-Work functions of individualplanes against average coverage 3 foran equilibrated K layer on tungsten.FIG.2.-Ratios of actual to averageK atom densities as - function ofaverage coverage B (defined as7 3 . 9 x 1014) for a K layer equili-brated at 400°K. I I tI 1 1 1 I.I0 20 .30 -40 .50 *cOe' (monolayers)1 3 0.2 0.4 06 0.8 1.0 1.2 -8 (monolayersR. GOMER 17Fig. 4 shows the dipole moments so computed as function of coverage, and alsothe fictitious average dipole moment fi, obtained from $. The latter is much lowerthan any of the real moments at low coverage because the adsorbate is then con-centrated on the high work function regions, which do not yet contribute appreci-ably to emission. Consequently the observed A+ values, which correspond toemission from the intrinsically low work function regions are in fact caused by con-siderably smaller adsorbate concentrations than the average Ei used for computing $.4 a 2 ' A 4 I -6 ' I -8 I ' I.0 ' ' 1-2coverage 8actual (not average) K coverage for several planes of a tungsten emitter.FIG.4.-Dipols moments p &do (q adsorbate charge, do surface-adsorbate distance) againstThe n/Z values may also be used to compute anisotropies in the heat H of ad-sorption, entropy effects presumably being small at 400°K where equilibrium is as-sumed to be frozen in. Since adsorption predominates on 110 at low 8, it is reason-able that the average heat of adsorption be equated with H110 so that absolute Hvalues for the other planes j can be obtained from njjrzllo at low 6. Since the largen anisotropies turn out to correspond to fairly small H anisotropies, it is permissibleto neglect the latter altogether as 0 increases and the n anisotropies become small.Thus, N against 8 curves over the entire coverage interval 0 1 can be constructedfrom the l7 against 0 curve, after converting 8 to 0 by means of the data of fig. 2and by taking H anisotropies into account at low 8.The results, together withthe previously 1 determined II against B curve are shown in fig. 5.Although the adsorbate charges cannot be determined in the absence of dovalues, certain bounds can be placed on them from the fact that the heat of ionicadsorption is given by(3) Hioni, == 4e -1,e + e2j4d, - Rep,where Rep is a repulsive ion core interaction and 10 the ionization potential of theadsorbate atom. Since the heat computed from eqn.(3) must be less than or equalto the actual H value, it turns out for the (I 10) plane at 8 = 0 that do = 2.85 A ifRep = 0, leading to a value of q = 0.58 electron charges, or that Rep = 0-9 eV ifq = 1 and do = 1.6A. The limits on the do values obtained in this way can thenbe used for the other planes to get estimates of the adsorbate charge and its variation18 DIFFERENT ADSORPTION STATESwith coverage. The results, shown in fig. 6, are that y, whatever its assignmentat 8 = 0, decreases rapidly with increasing coverage as is obvious from the dipolemoment curves.1 --T"0 -2 *4 -6 -8 to0FIG. 5.-Heats of K adsorption against actual (not average) coverage for several planes of tungsten.The average H against average coverage curve is also shown.I0FIG.6.-K atom charges as function of coverage for the limiting cases,do = 2.85 A, and Rep = 0.9 eV.To summarize the experimental findings, adsorption appears to be polar (possiblywholly ionic on (110) at low coverage), maxinium adsorbate charge occurring onthe highest work function region. As coverage increases, adsorbate charges de-crease markedly but smoothly on all regions. The heats of adsorption appear tR. GOMER 19be relatively insensitive to substrate structure or work function, but decrease fromthe initial values of 243-2.5 eV by over 1 eV as O+l.It is fairly clear from these findings that K adsorption on W must be essentiallymetallic, the large decrease in adsorbate charge with increasing 8 resulting from theincreased filling of the very broadened and split adsorbate level 1, 3, 4 as the latteris lowered by the decrease in work function.It is possible to set up a quantummechanical mode1,2 which utilizes system wave functions constructed from linearcombinations of one-electron metal wave functions and the appropriate K atomicorbitals (corrected for actual adsorbate charge), and, at least in principle, to deter-mine the coefficients and calculate the energy. If the nature of K adsorption hasbeen correctly diagnosed, this model would then indicate that system states far inenergy from that of the (corrected) adsorbate atom would be just metal states,with negligible atomic coefficients, while states lying fairly near the atomic levelwould have appreciable atomic wave function coefficients.Thus, the electroniccharge at the adsorbate consists of small contributions from a large number ofelectrons, the exact charge being fixed by the fact that all system levels below theFermi energy must be filled. For our present purposes it suffices, and is in someways more illuminating, to accept the general validity of the statements just madeand to consider the sharp adsorbate level so split by interaction with the substrate,that the adatom can be regarded as a tiny piece of metal, which we shall call a“ metallet ”. The heat of adsorption relative to substrate and isolated atom canthen be found by considering the energy changes in the following steps.(i) The atom far from the surface is ionized, the electron transferred to the metaland the ion allowed to approach to a distance do with the restriction that no electrontransfer back into the atom is as yet allowed.The energy decrease is that givenby eqn. (3), #e - loe + e2/4do - Rep.(ii) Since the sharp atomic level at -10 has now been broadened into a bandof metallet states we fill the latter to the equilibrium Fermi level I(q) by transferringamounts of electronic charge dq- from the metal into the metallet levels. Theenergy decrease is - #dq-+ I(q-)dq-- qdy-/2d, where the last term arises from 1: S6 c the fact that work must be done against the (to the electron, repulsive) image potentialq/2do in moving a charge dq- to the adsorbate. Since the net positive adsorbatecharge q = e-q- we obtain as the sum of steps (i) and (ii)H = +q + 1 eI(q)dq + q2/4d, - Ioe - Rep.4(4)Eqn. (4), which reduces to eqn.(3) for q = e casts the heats of adsorption largelyin electrostatic terms except for the integral which represents the energy of electronsactually in the metallet, so to speak. Eqn. (4) makes the variation of q and # withcoverage explicitly responsible for a large part of the decrease in H, the compensatingfactor being the increase in the integral as q decreases. # must be taken as thezero coverage work function plus the portion of the contact potential effective atdo, which can be obtained from the measured A# and the appropriate dipole sums,l’ 4since the absolute coverage n is known. Since q varies with coverage, it is moreconvenient to recast eqn. (4) asH = 4q + q2/4d0 - 1 ‘l(q)dq +A - Rep,0whereA = I(qjdq-f,e.J20 DIFFERENT ADSORPTION STATESSince I(q)dq is the average energy of the metallet levels (or better its absolute value,measured from the zero of field free space) A is simply the amount by which thisenergy has been shifted downward from the value -1oe for the sharp level by inter-action with the metal.The numerical value of 1(q) can be obtained from the equality of Fermi levelsat equilibrium, taking due account of electrostatic potentials,We are thus in a position to determine I(q) as a function of q. If the adsorbate-substrate geometry did not change with increasing coverage (which is the way inwhich q is varied) it would be legitimate to integrate the resultant curve from 0 toq and thus to obtain A from the actual H values and eqn.(5). This procedurehas been used rather than the more legitimate one of obtaining I(q)dq from Hand eqn. (4) to make the comparison between different planes easier, since A in-volves the same range of integration in all cases. Changes in adsorbate substrateinteraction then show up in A, although the numerical values will be incorrect ifbased on the distorted Z(q) against q curve obtainable from experiment. The resultof the Z(q) determination is shown in fig. 7 and 8 for the two extreme assumptionscI(q) = 4 +@do. (7)c-Ii -14+9/2do, e vFIG. 7.-K atom charge against Fermi level $+q/2do for several planes of tungsten, calculated onthe assumption of do = 2~85A.I , 110; 0,211 ; 0, 100; A, 111.d = 2.85 A, Rep = 0 ; and Rep = 0.9 eV, do = 1-6-24 A. The resultant +-widths ofthe metallet band are -3 eV in each case. The values of A obtained for the twoassumptions are shown in fig. 9. While the numerical values differ for the twoassumptions, the trends are very similar and do show some dependence on struc-ture and also on coverage. The behaviour can be rationalized by postulating thatA is greatest where the adsorbate can embed itself most deeply into the substrate,i.e., greatest on the (111) plane, less on the (211) plane and least on the (110) plane.The same qualitative results are obtained when the net delocalization energy isdetermined by means of the detailed model referred to earlier, by subtracting fromthe observed H values the explicitly electrostatic interactions given by that model.The reason why the curves seem to show maxima can be rationalized on the basiR.GOMER 21of fit with the substrate and lateral, attractive adsorbate-adsorbate interactions.At low coverage, optimal fit to the substrate is possible. As 8 increases, but notenough to cause departure from best fit, metallic adsorbate interactions (all repulsiveinteractions have already been taken into account by using the proper value of#) can slightly increase A. However, since the binding energy of pure K metal ismuch less than the heat of adsorption on W, this effect is soon overcome by thefact that further increases in coverage force departure of all adatoms from optimalsites, in order to accommodate more adsorbate in the first layer, leading to anoverall decrease in A.4+9/2do, evFIG.&-Data as in fig. 7, with the assumption Rep = 0.9 eV.I 1FIG. 9.-A against coverage for the two cases do = 2.85 A, and Rep = 0.9 eV.Thus, the present model rationalizes the fact that differences in substrate workfunction and geometry lead to only small heat anisotropies by casting the resultsinto a torm where the greater electrostatic contribution from the high work function,close-packed regions, is largely compensated by the increased resonance or exchangeeffects on regions where intrinsically lower work function is balanced by a morehospitable geometry22 DIFFERENT ADSORPTION STATESAlthough there is no obvious reason why adsorption on 100 should be less favour-able than on 211, the intrinsic work functions being almost equal, it is possible toinvoke differences in geometry and to point out that the nearest neighbour contacton 211 is 5 as compared with 4 on 100, while 111 can accommodate a close-packedlayer of K.However, it would be stretching the argument to go much beyondthe general trends just discussed. For present purposes it suffices to note that evenfor metallic adsorption substrate structure plays a role apart from explicitly electro-static effects, in determining adsorption properties.CARBON MONOXIDE ADSORPTIONThe adsorption of carbon monoxide on tungsten has been studied intensivelyby a number of authors in recent years and there is now evidence from flash fila-ment desorption,S'-*o field emission 11, 12 and electron impact desorption 12 for theexistence of at least three types of chemisorption states, which differ not only in theirheats of adsorption and electron desorption cross-sections but even in the sign oftheir dipole moments. Briefly some of the evidence for this Fehaviour is thefollowing.When CO is adsorbed at low temperature on a clean field emitter aweakly bound electronegative (4 increased) virgin state results, characterized byfairly uniform dipole moment and surface concentration. This is shown by theappearance of the pattern which indicates little change in emission anisotropyfrom the clean emitter, and also by the fact that the average work function increasesfairly linearly with amount adsorbed.The result of heating such a layer to >400"Kor of exposing it to low energy electron impact 12 is partial desorption, as will beshown presently,l3 and creation of a more tightly bound electronegative p layercharacterized by a high-temperature flash desorption spectrum showing considerablesfructure,~-7~ 10 and by a very low electron desorption cross-section.12 When COis re-adsorbed on a p layer the virgin layer is not restored, but a third type of ad-sorption is observed, characterized by a very high electron desorption 12 cross-sectionand by a positive dipole moment 8 , 9, 11- 12 (i.e., the work function is reduced).This layer can be desorbed by reheating 11, 12 to 300-400°K or by electron impact 12without affecting the underlying p layer.Menzel and Gomer12 were able to show, by means of electron desorption,that this picture is somewhat oversimplified.Since the electron desorption cross-sections for the different adsorption types differ by orders of magnitude, it is possibleto dissect the composition of a given layer by noting the observed rates and extra-polating the corresponding A& to zero desorption. In this way it was found thatthere is a small amount of a-CO on the virgin layer even before heating, and byimplication a corresponding amount of P-CO, since the formation of a-COseems to hinge on the presence of p states. Further, heating a virgin layer to 270°Kleads not only to p formation, but to some a formation as well, indicating that virgindesorption may proceed through an intermediate a-state.Also, in addition tocc-CO a small amount of virgin GO can be adsorbed on a full p-layer.Although all of these results refer to averages over the entire emitter there isless reason to suspect their basic validity than would be the case with electropositiveadsorbates, in large part because adsorption is immobile below 600°K and becausethere is less reason to suspect severe coverage anisotropies as shown by the factthat the intrinsic high work function regions stay dark, although the overall workfunction (i.e., that of the intrinsically low 4 regions) is raised to 54-56 eV at 8 = 1.On the other hand, this means that no information is obtained on the 110, 21 1 and100 regions since they do not contribute to emission at any coverage.MeasurementR. GOMER 23on single planes may force some modification of these assumptions as well asproviding information on the high cf, regions.While a quantitative corrclation of work function changes with amounts adsorbedand a determination of the relative abundances of the various adsorption statesrequires measurements on single crystal faces, it is already of considerable interestto obtain these quantities averaged over the surface. To this end, Bell and Gomer 13measured average work function changes as function of the relative amount ofCO impinged and then converted the amounts impinged into amounts adsorbed bydetermining sticking coeffcients in a separate experiment. The method used forthe latter consisted of “illuminating” a tungsten ribbon with a CO sublimationsource.A moveable field emitter serving as detector could be raised in front ofthe ribbon in such a way as to receive a CO deposit only by reflection (or desorption)from the ribbon. In this way the amount reflected per dose impinged on the ribboncould be determined quite accurately acd step desorption spectra could be obtainedas well. The resultant sticking coefficients are shown in fig. 10. 4 as functions ofSticking Coefficientson Initially Clean W2 4 6 8 10 12 14 16number of doses adsorbedFIG. 10.-Sticking coefficients of CO on initially clean tungsten against relative coverage; the 500and 780°K curves coincide.doses impinged and doses adsorbed (converted to the latter by means of the stickingcoefficient results) are shown for virgin, /3 and a adsorption in fig.11-13. The cicurve is linear, while the virgin and /3 curves show two essentially linear segments.For the /3 layer formed at 400°K this may mean that the initial portion of the layercorresponding to a high sticking coefficient has an intrinsically lower dipole moment.For the virgin curve where the sticking coefficient is unity throughout, the higherdipole moment is observed at lower coverage, and it might therefore be supposedthat the terminal section of the 20°K layer consisted of the initial portion of thep layer, i.e., that the low temperature layer is a composite of virgin and p states.This is improbable, however, for a number of reasons : (i) only a small amount ofM-CO is detectable on a low temperature layer by electron desorption; since theratio of a to fl is close to unity (see later), this implies the presence of only a smallamount of P-CO.(ii) The total amount of the low dipole 20°K deposit is roughl24 DIFFERENT ADSORPTION STATEStwice the amount of the low dipole B deposit obtainable on a saturated layer.(iii) It is improbable that under conditions of unit sticking coefficient virgin sitesshould be occupied first, and that only after the completion of the virgin layer a---- 3 --6I 2 3 4 5 6 7 8 9 10 II 12 13 14 15 16 17 18 19 2021 22 232425number of doses impingedBFIG. 11.-Work function against number of 30 sec CO doses adsorbed at 20°K. The curve markedB represents the change in the logarithm of the Fowler-Nordheim preexponential term.7 - 1 I.I I5 2 0 400'K Adsorption-460 -450 I 2 3 4 5 6 7 8 9 1 0 1432 I'I0number of doses impingedFIG. 12.-Work function against number of 60sec CO doses impinged (open circles) and dosesadsorbed (full circles) for adsorption at 400°K. The B curve refers to the change in the logarithmof the Fowler-Nordheim pre-exponential term.fl deposit should be formed. It therefore seems more likely that the decrease indipole moment at higher coverages for the low temperature layer is simply theresult of less optimal adsorption under conditions of crowding, as well as of the forma-tion of some a-CO at high coverage.Fig. 14 shows the step desorption spectrum of a complete low-temperaturelayer, and superimposed on it an a desorption spectrum.The low-temperaturelayer shows a main desorption peak at 320°K with desorption of this part completR. GOMER 25- I4 I I 5.4 0Alpha Adsorptionon 600°K Beta 15Layer5.20 16-5.10 '\5.004 4.90%'.0 I 2 3 4 5 6 7 8 9 10number of dosesFIG. 13.-Work function against number of 30 sec CO doses impinged and adsorbed at 100°K ona B layer prepared by heating a virgin layer to 600°K. Where the impinged and adsorbed dosesdiffer the latter are shown by open circles. The In A curve represents the logarithm of the Fowler-Nordheim pre-exponential term.I 11 Virgin, Alpha and Beta SpectraI -I -!i <,/fp&q$ i0 100 200 300 40C 700 800 900 1000 1100 1200 1300 1400T"KX4s s aLFIG.14.4uperimposed step-desorption spectra for virgin and alpha layers. Virgin ordinatesare shown by vertical bars, alpha ordinates by circles. The virgin spectrum was obtained byheating a full 20°K layer in steps to 450°K. The emitter was then redosed at 20°K and the alphaspectrum obtained. Continuation of heating then gives the beta spectrum shown. Two pointson the latter (x ) have been normalized to the average T interval. Scale for beta spectrum givenon right side of figure. A V/Vo represents relative simal size of detector26 DIFFERENT ADSORPTION STATESat 450"K, corresponding to formation of the p layer. The latter shows desorptioncommencing at 700°K. The a spectrum is remarkably similar to the low-temper-ature portion of the virgin spectrum, despite the fact that the y-states have a dipolemoment of opposite sign.The desorption results leave the possibility that virgin and j? states are adsorbedsimultaneously at low temperature, with virgin desorption simply leaving behinda pure p layer.This is ruled out, however, by the adsorption hysteresis, i.e., thefact that re-adsorption leads to a rather than to virgin adsorption, and also by thedata shown in fig. 15 where the ratio of CO desorbed below 500°K to the total amount*I '2 ' 3 - 4 -5 -6 -7 -8 *9 to' 38 (fraction of full 20°K layer)FIG. 15.-Ratio of virgin CO desorbed (i.e., CO desorbed below 500°K) to total amount depositedon the surface at 20°K against coverage (in terms of the full low temperature layer).adsorbed at 20°K is plotted against the initial coverage, expressed as a fraction ofthe 20°K layer maximum coverage.If the ratio of virgin to p-CO adsorbed at20°K were fixed, this ratio should be constant for immobile adsorption. The S-shape actually observed is consistent with the assumption that most of the CO isadsorbed as virgin, and that heating leads to desorption and conversion to p, de-sorption being rate-controlling at high coverage.If this interpretation is accepted the work function data can be used to determinerelative dipole moments. The results relative to the dipole moment of the initialportion of the virgin deposit (p,) are as follows: ,u, (at high coverage)/p, = 0.7;pb (low coverage)/p, = 0-68 ;Relative abundances can be obtained from the desorption data, if the presentinterpretation of the layer composition is accepted.As already pointed out, electronimpact measurements indicate that there is a small amount of a-CO on the lowtemperature layer, and by implication a similar amount of p-CO. If the corres-ponding work function increments obtained from electron desorption are convertedinto relative coverages by means of the dipole moment ratios just obtained, thecorrections turn out to be -12 % each of the total low-temperature coverage.It further turns out from electron desorption that redosing a P-layer leads to a smallamount of virgin adsorption, corresponding to -7 % of the total low temperaturedeposit. With these corrections it is possible to calculate the ratio of total p tototal virgin as well as the convertible virgin to converted p. It is also possible to(high coverage)/p, = 1.0 ; pa/pv = -0.68R.GOMER 27correct the amount of o! obtained in desorption for the small amount of re-adsorbedvirgin CO. The results of these manipulations are shown in table 1.TABLE 1 .a unwrr. as fraction of: 01 con. function of: virgin and fi (corr.)0.61 0.69 1.4 0.78 1.2 0.68 0-60 1.8 2.1Summary of abundances of virgin, /3 and cc CO on a polycrystalline tungsten surface. cc un-corrected refers to the total amount desorbed after redosing a b layer. 0: corrected is this amountminus the amount of re-adsorbed virgin CO, as determined from electron impact desorption and thedipole ratios. The virgin and /3 data have been corrected for residual ct and j? on a low-temperatureI ayer .While it is apparent that these results are only averages over a polycrystallinesurface, it is nevertheless probable that valid qualitative conclusions can be drawnfrom them, because there is no evidence for special anisotropies from field emissionor electron desorption.These conclusions are the following. The dipole momentsof virgin and p-CO are negative and of approximately equal magnitude. Thedipole moment of a-CO is positive androughly 70 % of that of virgin CO. If amean value of 6 x 1014 molecules is taken as the density of the monolayer, the dipolemoments are -0-44 D for the high dipole forms of virgin and p-CO, -0-3 D for thelow dipole forms of virgin and p, and + 0.3 D for a-CO.For every molecule desorbedfrom a virgin layer, roughly one p state can be formed and one molecule re-adsorbedas a, i.e., the ratio of virgin to p sites is roughly 2 and the ratio of /? to a sites isroughly 1.Even if these numbers have to be modified severely on certain regions of thesurface not yet accessible to measurement, the conclusion seems very strong thatthe different adsorption states correspond to different geometric substrate-adsorbateconfigurations. Thus, adsorption at low temperature may occur via a weak, un-activated mode utilizing sp2 carbon orbitals ; since these are half-filled it is reasonablethat adsorption should be electronegative. Heating or electron impact may leadto desorption and to activated rearrangement of the remainder to tightly boundstates utilizing partly filled molecular orbitals and thus having negative dipolemoments.If this is a lying-down mode, the occupancy of two virgin sites by each/J molecule would be explained. On a full /? layer, single substrate atom sites wouldstill be available for M adsorption, presumably through a doubly occupied sp carbonorbital, with some electron transfer from CO, thus accounting for the positivedipole moment. It seems reasonable, but is by no means proved, that this generalscheme applies in varying degrees to all regions of the surface and that the structurein the p spectrum, for instance, is due to variations of configuration, either fromregion to region or possibly within a given region as well.A UNIFIED PICTURE OF CHEMISORPTION?Even if the particular scheme advanced for CO is incorrect, it is apparent thatstrong chemisorption of molecules with high ionization potentials is markedlydifferent from metallic adsorption.In its outward manifestations covalent ad-sorption shows a much greater sensitivity to substrate structure but less sensitivityto work function, a greater variety of discrete adsorption states, but much less leewayfor gradual change within a given adsorption type. It is not difficult to understandthe reasons for this in a qualitative way. In metallic adsorption the relevant energ28 DIFFERENT ADSORPTION STATESlevels fall within the conduction band of the total system, and a very large numberof levels are involved in binding, i.e., the basis wave functions contain in additionto the relevant adsorbate orbitals a very large number of substrate orbitals, andmany electrons participate in adsorption.In electronegative adsorption of thetype just considered it may be concluded that the system states involved lie belowthe conduction band (even when adsorption is weak). Consequently they can beconsidered as composed of the appropriate adsorbate orbitals and interband surfacestates of the substrate, which are not utilized in metallic adsorption. Since thesurface states are highly localized it may be concluded that (a) the number of elec-trons involved in adsorption is small, so that the adsorption complex is much closerto a conventional chemical compound than for metallic adsorption, and (b) thatthe types of surface compounds so formed will be much more sharply defined, withdiscontinuous changes, as the configurations and hence the orbitals involved change.Since the number of electrons involved in bonding is small, the polarizability ofthe ad-complex, and its ability to vary its charge significantly will be very smallcompared to metallic adsorption. Further, the net electron transfer involved inbonding is also likely to be much less, since binding must be covalent. The dis-parity in 10 and 4 precludes electropositive polar binding, and the relatively lowelectron affinity of the adsorbate precludes very polar electronegative binding.So far the emphasis has been on differences between metallic and covalent ad-sorption, but significant similarities exist. In both cases, exchange integrals whosenumerical values depend sensitively on configuration play a role, even though thelatter is less pronounced in metallic adsorption. To a reasonable degree the con-cept of geometric fit, i.e., of hard-sphere models, would appear to have the significancewith which intuition has long credited it.1 Schmidt and Gomer, J. Chem. Physics, 1965, 42, 3573.2 Schmidt and Gomer, J. Chem. Physics, in press.3 Gurney, Physic. Rev., 1935, 47, 479.4 Gomer and Swanson, J. Chem. Physics, 1963,38, 1613.5 Ehrlich, J. Chem. Physics, 1962, 36, 1171.6 Redhead, Trans. Faraduy Soc., 1961, 57, 641.7 Rigby, Can. J. Physics, 1964, 42, 1256.8 Eisinger, J. Chem. Physics, 1957, 27, 1206.9 Gavrilyuk and Medvedev, Fiz. Tver. Tela, 1962, 4, 2372. [English trans., Soviet Physics-Solid10 Madey, Yates and Stern, J. Chem. Physics, 1965, 42, 1372.11 Swanson and Gomer, J. Chem. Physics, 1963, 39,2813.12 Menzel and Gomer, J. Chem. Physics, 1964, 41, 3329.13 Bell and Gomer, J. Chem. Physics, 1966, 44, 1065.State, 1963, 4, 1737.