- Email: [email protected]
Dilution of the electron gas in a Cs monolayer on Cu(111) by adsorption of oxygen
Surface Science 80 (1979) 620-625 0 North-Holland Publishing Company
DILUTION OF THE ELECTRON GAS IN A Cs MONOLAYER ON Cu( 111) BY ADSORPTION OF OXYGEN S.A. LINDGREN and L. WALLDEN Physics Department, Chalmers Universityof Technology, Ciiteborg, Sweden
The Cu(ll1) oxygen sion for
characteristic loss energy of plasmons in a close packed Cs p(2 X 2) monolayer on is found to decrease gradually upon increasing the amount of oxygen adsorbed, the appearing to act as a thinner of the electron gas in the overlayer. We observe no disperthe monolayer plasmon.
1. Introduction Electron energy loss measurements have shown that plasmons in monolayers of close packed Cs atoms adsorbed on Cu( 111) and W(IOO), so far the only substrates used, in both cases have an energy, Rap, of around 1.5 eV [ 1,2] a value considerably lower than the bulk and surface plasmon energies of habp = 2.9 eV and fioJ,* = 2.1 eV respectively for Cs metal [3]. The interpretation of the excitations found for the monolayers in terms of plasmons is based on the observed smooth change of the loss structure into that of the alkali metal upon increasing the thickness of the adsorbed layer. Assuming the density of valence electrons, n, to be the same in a close packed p(2 X 2) layer of almost neutral Cs atoms on Cu(ll1) [2] as in Cs metal, existing models [4,.5] predict the overlayer plasma frequency to decrease from ‘_rbpto o,n as the wave-vector increases [6,7]. Since wbp2 a n, a factor of approximately two difference between theory and experiment could only be accounted for by making the rather unrealistic assumption that n is a factor of four lower in the Cs overlayer than in Cs metal. The low plasmon energies measured for monolayers of Cs on Cu and W as well as for Na and K on Ni [8] and K and Rb on W [9] therefore indicate that the models, which are successful for thicker films bounded by vacuum or insulators [lo], are not directly applicable to metallic monolayers on metal substrates. The energy loss experiments have also shown that the overlayer plasmon energy may be further reduced by decreasing the amount of alkali metal on the surface. The shift obtained by reducing the coverage of Cs on Cu(ll1) from that of a full monolayer to 60% of a monolayer is roughly explained be a linear relationship between n and the Cs coverage if one assumes that wP is proportional to &r [2]. Upon a further reduction of the coverage the adsorbate 620
SA. Lindgren, L. Walldh /Dilution of electron gas in Cs monolayer
621
becomes ionic due to a transfer of the Cs valence electron to the substrate and the plasmon loss peak rapidly looses strength. Here we report that considerably lower plasmon energies may be obtained by exposing a full monolayer of Cs on Cu(l11) to oxygen than by varying the amount of Cs adsorbed on Cu(ll1). The lowest plasmon energies measured are around 0.5 eV, which is the approximate limit set by our measuring system rather than a low energy limit for the overlayer plasmons. We argue that the unusually low plasmon energies obtained are due to a continuous decrease of the electron density in the overlayer as more oxygen is adsorbed. After a certain amount of oxygen has been adsorbed there are oxygen and cesium ions as well as cesium atoms in the overlayer with the valence electrons of the Cs atoms relatively free to move in the entire overlayer volume. The density of the electron gas in the overlayer is thus, we believe, controlled by the amount of adsorbed oxygen.
2. Experimental We have used a standard LAZEDapparatus to check the preparation of the Cs p(2 X 2) layer on the Cu(ll1) crystal and to measure work function changes upon oxygen adsorption via the retarding field method. To record energy loss spectra we have built a cylindrical mirror type analyzer with a circular entrance aperture, the acceptance cone having an opening angle of around 2”. The analyzer and e-gun have fured positions while the sample can be rotated. The energy loss spectra were measured with the specularly reflected beam (68” incidence angle, 32 eV primary electron energy) falling on the entrance aperture. A schematic drawing of the apparatus is shown in fig. 1. The loss spectra were obtained after exposing the sample to stepwise increased amounts of oxygen. Each exposure increase was made by raising the oxygen pressure to approximately 2 X lo-’ Torr, this pressure value represent-
E-GUN
LEE0
SCREEN
LEAK VALVE’
ARGON GUN SAMPLE
EVAPORATION
SOURCE
‘WINDOW
Fig. 1. Schematic drawing of the experimental arrangement.
622
S.J!. Lindgren, L. Walldkn/ Dihtion of electron gas in Cs monolayer
ing a compromise between the wish to keep the oxygen pressure well above the 1 X IO-” Torr background pressure and yet be able to measure the exposure times and thus exposures reasonably well.
3. Results and discussion For the range of loss energies shown in fig. 2 the substrate gives no important contribution to the loss spectra, which are dominated by the loss peak due to the overlayer plasmon. The characteristic loss energy, Mr,, and the intensity of the loss peak decreases as the oxygen exposure, Q, increases whereas the width of the loss peak remains almost constant. Plotting (aErJ2 versus exposure (fig. 3a) we find a linear dependence and an extrapolated cut off value, Qo, which due to uncertainties in the exposure values ranges between 0.3 X 10T6 and 0.5 X 10m6Torr set for different experimental runs. According to the specifications of the pressure gauge the absolute pressure and hence exposure values could be off by a factor of two. Drawn against oxygen exposure the work function, which was measured in separate runs, passes a minimum value of 1.2 eV close to Q. (fig. 3b), the absolute work function value being obtained by assuming the clean Cu(ll1) crystal to have a work function of 4.9 eV [ll]. As discussed in a previous study of the work function variations obtained when oxygen is adsorbed on a monolayer of Cs on W [12] the reduction of the work function for exposures below Q. indicates that the oxygen ions are located beneath the Cs ions. At Q. this oxygen layer is complete; the increase of the work function for exposure above Q. being associated with the growth of a second oxygen layer, likely on top of the Cs layer. To get some additional information about the atomic arrangement and the electronic structure in the overlayer we have observed the changes of the LEED pattern upon oxygen exposure and measured the strength versus oxygen exposure of an Auger peak located approximately 11 eV above the Fermi energy, in the latter case using the suppressor grids and display screen of the LEED optics as a retarding field energy analyzer. The peak is due to valence electrons emitted from the sample as a result of the Auger process started by exciting Cs 5p electrons. The peak height, which we assume is a measure of the amount of Cs valence charge in the overlayer approaches zero as the exposure approaches Q, (fig. 3~). The changes of the LEED patterns show that there is one or maybe two ordered phases between zero and Q. exposure. Close to QOhowever the same pattern is obtained as for the clean Cs p(2 X 2) layer, indicating equal numbers of oxygen and cesium ions at this exposure. As expected from adsorption measurements on similar systems [ 12,131 oxygen has a high sticking coefficient. Assuming unit sticking probability equal numbers are obtained at 0.6 X 10V6Torr set exposure, which means that a correction of at least 50% should be added to our measured exposure values. The appearance of ordered phases shows that the oxygen is spread rather evenly over the surface. The distribution of the oxygen is of course important for the properties of the overlayer. If, for example, clusters of oxide had
S.A. Lindgren, L. Wallddn /Dilution
of electron gas in Cs monolayer
623
, cu~lll)-p(2x21 \
cs
+ 02
0 \
O/O \ \ \
I
cu (Ill )-p(2x2)
a
\ , /
/
cs
0 0.2 0.4 1 0.6 O2 Exposure (L) Loss energy (eV) Fig. 2. Energy loss spectra obtained from a close packed Cs monolayer on Cu(ll1) after the oxygen exposures given in the diagram (1 L = 10T6 Torr set) showing the exposure dependence of the plasmon loss energy, A.+ The spectra were obtained for a beam of 32 eV primary electrons reflected specularly from the sample at 68” incidence angle.
2
Fig. 3. The oxygen exposure dependence of the square of the characteristic loss energy tip (a) of the work function (b) and of the height of an Auger peak due to electrons emitted as Cs 5p states are filled (c). A spectrum showing the Auger peak at zero oxygen exposure is inserted in (c). Along the horizontal axes of the insert is plotted the energy above the Fermi energy, EF.
formed on the surface one would have expected the plasmon energy to remain almost constant for the regions between the clusters. The results presented in fig. 3 suggest that at QO exposure, when the first oxygen layer is complete, all Cs on the surface has become ionic, the removal of Cs valence charge upon oxygen adsorption
624
SA. Lindgren, L. Walldtn/Dilution of electron gas in Cs monolayer
being accompanied by a gradual decrease of electron density and plasrnon loss energy, At&. No significant shift of A/??,,was found upon changing the primary electron energy between 5 and 50 eV or by turning the sample and thereby measuring the loss spectrum for electrons scattered out of the specular beam. These measurements were made without oxygen exposure using also the LEED grids and screen to record the angular integrated spectrum when studying the dependence of A,!?, upon primary energy. For metal films, which are thicker than ours though thin enough for the coupling between the waves along the two boundaries to be important, the dispersion of the plasmons makes the loss energy dependent upon primary energy and scattering angle [6,10]. For electrons inelastically backscattered from a single crystal, a case discussed by Duke et al. [ 141, a further complication arises from the angular and energy dependence of this scattering. In the present case, however, when the observed loss energy is independent of primary energy and scattering angle, we find that all our observations are consistent with a simple (A,!?r,)2= (h~+,)~ = Cn relationship where n = no (1 - 0) is the density of electrons in the overlayer and 6 the oxygen coverage (defining 19= 1 for the first full oxygen layer obtain at Q. exposure). The constant C should be determined by the dielectric function of the substrate and by the so far uncertain character of the plasma oscillations, although the dependence on the former should be rather weak since the same plasmon energy is obtained for close packed Cs on Cu as on W. The observation that A&, is independent of emission angle either means that the dispersion is small or that our loss spectra are dominated by electrons which have suffered both elastic and inelastic scattering. The effect is well known from energy loss studies of the dispersion of bulk plasmons. Due to the high probability of creating plasmons having small wavevectors, q, the loss spectrum representative of small q is superposed via elastic electron scattering on the loss spectrum characteristic of a certain scattering angle and momentum transfer [ 151. Our lowest measured loss energy, 0.5 eV, corresponds to a situation where approximately 90% of the Cs on the surface is ionic. The presence of the plasmon peak at this stage of oxidation shows, obviously, that the remaining Cs atoms are able to keep their valece electrons within the overlayer. If instead 10% of a Cu( 111) surface is covered with Cs the adsorbate is close to ionized [2], almost all of the Cs valence charge being transferred to the substrate. Two factors may contribute to this difference between the Cu( 111) t Cs and the Cu( 11 l)-p(2 X 2)Cs + 02 systems. Firstly the cesium and oxygen ions present in the latter case create a dipole field, which attracts the electrons to the overlayer. Secondly the nearest neighbour configuration is different for Cs atoms on the two surfaces. For Cu(ll1) f Cs and similar systems the Cs adsorbate is uniformly distributed over the surface at submonolayer coverages [2,16,17], which means that the nearest neighbour distance changes continuously ‘with coverage. For the Cu(l1 l)-p(2 X 2)Cs + 02 system those Cs atoms, which have not reacted with oxygen, will have nearest neighbours at almost close packed distance although their number decreases upon increasing
S_,& Lindgren, L. WalldPn/Dilution of electron gas in Cs monolayer
625
the amount of adsorbed oxygen. Even in this case there may be some change of the interatomic distance as a result of lateral contraction accompanying the oxygen adsorption [ 121. To summarize we have associated the linear relationship between the square of the characteristic loss energy and the oxygen exposure with a linear dependence of wp2 on the density, II, of valence electrons in the overlayer. This suggests that the alkali monolayers on metal substrates are systems for which the electron density can be varied continuously and substantially in a rather simple way and, as in the present case, down to quite low values without introducing much additional damping of the collective oscillations. The dispersion of the overlayer plasmons remains uncertain. The constant characteristic loss energy obtained upon changing primary energy and scattering angle suggests that the dispersion is small but for the reasons given above no firm conclusion can be drawn.
Acknowledgement We want to thank H.P. Myers for his support and encouragement, S. Anderson and U. Jostell for many helpful discussions and the Swedish Natural Science Research Council for financial assistance.
References [ 1] A.U. MacRae, K. Miiller, J.J. Lander, J. Morrison and J.C. Phillips, Phys. Rev. Letters 22 (1968) 1048. [2] S.A. Lindgren and L. Wallden, Solid State Commun. 25 (1978) 25. [3] C. Kunz, Z. Physik 196 (1966) 311. [4] R.H. Ritchie, Phys. Rev. 106 (1957) 874. [5] E.A. Stern and R.A. Ferrel, Phys. Rev. 120 (1960) 130. [6] A. Otto, Z. Physik 185 (1965) 232. [7] J.M. Gadzuk, Phys. Rev. Bl (1970) 1267. [ 81 S. Anderson and U. Jostell, Faraday Discussions Chem. Sot. 60 (1975) 255. [9] S. Thomas and T.W. Haas, Solid State Commun. 11 (1972) 193. [ 10) For a review see: H. Raether, in: Thin Solid Films, Vol. 9 (Academic Press, 1977) p. 145. [ 111 P.O. Gartland, S. Berge and B.J. Slagsvold, Phys. Norvegica 7 (1973) 39. [12] C.A. Papageorgopoulus and J.M. Chen, Surface Sci. 39 (1973) 313. [ 131 R.A. Marbrow and R.W. Lambert, Surface Sci. 61 (1976) 329. [ 141 C.B. Duke, L. Pietronero, J.0 Porteus and J.M. Wendelken, Phys. Rev. B12 (1975) 4059. [15] G. Meyer, Z. Physik 148 (1957) 61. [ 161 R.L Gerlach and T.N. Rhodin, Surface Sci. 17 (1969) 32. [ 171 S. Andersson and U. Jostell, Solid State Commun. 13 (1973) 829.
DILUTION OF THE ELECTRON GAS IN A Cs MONOLAYER ON Cu( 111) BY ADSORPTION OF OXYGEN S.A. LINDGREN and L. WALLDEN Physics Department, Chalmers Universityof Technology, Ciiteborg, Sweden
The Cu(ll1) oxygen sion for
characteristic loss energy of plasmons in a close packed Cs p(2 X 2) monolayer on is found to decrease gradually upon increasing the amount of oxygen adsorbed, the appearing to act as a thinner of the electron gas in the overlayer. We observe no disperthe monolayer plasmon.
1. Introduction Electron energy loss measurements have shown that plasmons in monolayers of close packed Cs atoms adsorbed on Cu( 111) and W(IOO), so far the only substrates used, in both cases have an energy, Rap, of around 1.5 eV [ 1,2] a value considerably lower than the bulk and surface plasmon energies of habp = 2.9 eV and fioJ,* = 2.1 eV respectively for Cs metal [3]. The interpretation of the excitations found for the monolayers in terms of plasmons is based on the observed smooth change of the loss structure into that of the alkali metal upon increasing the thickness of the adsorbed layer. Assuming the density of valence electrons, n, to be the same in a close packed p(2 X 2) layer of almost neutral Cs atoms on Cu(ll1) [2] as in Cs metal, existing models [4,.5] predict the overlayer plasma frequency to decrease from ‘_rbpto o,n as the wave-vector increases [6,7]. Since wbp2 a n, a factor of approximately two difference between theory and experiment could only be accounted for by making the rather unrealistic assumption that n is a factor of four lower in the Cs overlayer than in Cs metal. The low plasmon energies measured for monolayers of Cs on Cu and W as well as for Na and K on Ni [8] and K and Rb on W [9] therefore indicate that the models, which are successful for thicker films bounded by vacuum or insulators [lo], are not directly applicable to metallic monolayers on metal substrates. The energy loss experiments have also shown that the overlayer plasmon energy may be further reduced by decreasing the amount of alkali metal on the surface. The shift obtained by reducing the coverage of Cs on Cu(ll1) from that of a full monolayer to 60% of a monolayer is roughly explained be a linear relationship between n and the Cs coverage if one assumes that wP is proportional to &r [2]. Upon a further reduction of the coverage the adsorbate 620
SA. Lindgren, L. Walldh /Dilution of electron gas in Cs monolayer
621
becomes ionic due to a transfer of the Cs valence electron to the substrate and the plasmon loss peak rapidly looses strength. Here we report that considerably lower plasmon energies may be obtained by exposing a full monolayer of Cs on Cu(l11) to oxygen than by varying the amount of Cs adsorbed on Cu(ll1). The lowest plasmon energies measured are around 0.5 eV, which is the approximate limit set by our measuring system rather than a low energy limit for the overlayer plasmons. We argue that the unusually low plasmon energies obtained are due to a continuous decrease of the electron density in the overlayer as more oxygen is adsorbed. After a certain amount of oxygen has been adsorbed there are oxygen and cesium ions as well as cesium atoms in the overlayer with the valence electrons of the Cs atoms relatively free to move in the entire overlayer volume. The density of the electron gas in the overlayer is thus, we believe, controlled by the amount of adsorbed oxygen.
2. Experimental We have used a standard LAZEDapparatus to check the preparation of the Cs p(2 X 2) layer on the Cu(ll1) crystal and to measure work function changes upon oxygen adsorption via the retarding field method. To record energy loss spectra we have built a cylindrical mirror type analyzer with a circular entrance aperture, the acceptance cone having an opening angle of around 2”. The analyzer and e-gun have fured positions while the sample can be rotated. The energy loss spectra were measured with the specularly reflected beam (68” incidence angle, 32 eV primary electron energy) falling on the entrance aperture. A schematic drawing of the apparatus is shown in fig. 1. The loss spectra were obtained after exposing the sample to stepwise increased amounts of oxygen. Each exposure increase was made by raising the oxygen pressure to approximately 2 X lo-’ Torr, this pressure value represent-
E-GUN
LEE0
SCREEN
LEAK VALVE’
ARGON GUN SAMPLE
EVAPORATION
SOURCE
‘WINDOW
Fig. 1. Schematic drawing of the experimental arrangement.
622
S.J!. Lindgren, L. Walldkn/ Dihtion of electron gas in Cs monolayer
ing a compromise between the wish to keep the oxygen pressure well above the 1 X IO-” Torr background pressure and yet be able to measure the exposure times and thus exposures reasonably well.
3. Results and discussion For the range of loss energies shown in fig. 2 the substrate gives no important contribution to the loss spectra, which are dominated by the loss peak due to the overlayer plasmon. The characteristic loss energy, Mr,, and the intensity of the loss peak decreases as the oxygen exposure, Q, increases whereas the width of the loss peak remains almost constant. Plotting (aErJ2 versus exposure (fig. 3a) we find a linear dependence and an extrapolated cut off value, Qo, which due to uncertainties in the exposure values ranges between 0.3 X 10T6 and 0.5 X 10m6Torr set for different experimental runs. According to the specifications of the pressure gauge the absolute pressure and hence exposure values could be off by a factor of two. Drawn against oxygen exposure the work function, which was measured in separate runs, passes a minimum value of 1.2 eV close to Q. (fig. 3b), the absolute work function value being obtained by assuming the clean Cu(ll1) crystal to have a work function of 4.9 eV [ll]. As discussed in a previous study of the work function variations obtained when oxygen is adsorbed on a monolayer of Cs on W [12] the reduction of the work function for exposures below Q. indicates that the oxygen ions are located beneath the Cs ions. At Q. this oxygen layer is complete; the increase of the work function for exposure above Q. being associated with the growth of a second oxygen layer, likely on top of the Cs layer. To get some additional information about the atomic arrangement and the electronic structure in the overlayer we have observed the changes of the LEED pattern upon oxygen exposure and measured the strength versus oxygen exposure of an Auger peak located approximately 11 eV above the Fermi energy, in the latter case using the suppressor grids and display screen of the LEED optics as a retarding field energy analyzer. The peak is due to valence electrons emitted from the sample as a result of the Auger process started by exciting Cs 5p electrons. The peak height, which we assume is a measure of the amount of Cs valence charge in the overlayer approaches zero as the exposure approaches Q, (fig. 3~). The changes of the LEED patterns show that there is one or maybe two ordered phases between zero and Q. exposure. Close to QOhowever the same pattern is obtained as for the clean Cs p(2 X 2) layer, indicating equal numbers of oxygen and cesium ions at this exposure. As expected from adsorption measurements on similar systems [ 12,131 oxygen has a high sticking coefficient. Assuming unit sticking probability equal numbers are obtained at 0.6 X 10V6Torr set exposure, which means that a correction of at least 50% should be added to our measured exposure values. The appearance of ordered phases shows that the oxygen is spread rather evenly over the surface. The distribution of the oxygen is of course important for the properties of the overlayer. If, for example, clusters of oxide had
S.A. Lindgren, L. Wallddn /Dilution
of electron gas in Cs monolayer
623
, cu~lll)-p(2x21 \
cs
+ 02
0 \
O/O \ \ \
I
cu (Ill )-p(2x2)
a
\ , /
/
cs
0 0.2 0.4 1 0.6 O2 Exposure (L) Loss energy (eV) Fig. 2. Energy loss spectra obtained from a close packed Cs monolayer on Cu(ll1) after the oxygen exposures given in the diagram (1 L = 10T6 Torr set) showing the exposure dependence of the plasmon loss energy, A.+ The spectra were obtained for a beam of 32 eV primary electrons reflected specularly from the sample at 68” incidence angle.
2
Fig. 3. The oxygen exposure dependence of the square of the characteristic loss energy tip (a) of the work function (b) and of the height of an Auger peak due to electrons emitted as Cs 5p states are filled (c). A spectrum showing the Auger peak at zero oxygen exposure is inserted in (c). Along the horizontal axes of the insert is plotted the energy above the Fermi energy, EF.
formed on the surface one would have expected the plasmon energy to remain almost constant for the regions between the clusters. The results presented in fig. 3 suggest that at QO exposure, when the first oxygen layer is complete, all Cs on the surface has become ionic, the removal of Cs valence charge upon oxygen adsorption
624
SA. Lindgren, L. Walldtn/Dilution of electron gas in Cs monolayer
being accompanied by a gradual decrease of electron density and plasrnon loss energy, At&. No significant shift of A/??,,was found upon changing the primary electron energy between 5 and 50 eV or by turning the sample and thereby measuring the loss spectrum for electrons scattered out of the specular beam. These measurements were made without oxygen exposure using also the LEED grids and screen to record the angular integrated spectrum when studying the dependence of A,!?, upon primary energy. For metal films, which are thicker than ours though thin enough for the coupling between the waves along the two boundaries to be important, the dispersion of the plasmons makes the loss energy dependent upon primary energy and scattering angle [6,10]. For electrons inelastically backscattered from a single crystal, a case discussed by Duke et al. [ 141, a further complication arises from the angular and energy dependence of this scattering. In the present case, however, when the observed loss energy is independent of primary energy and scattering angle, we find that all our observations are consistent with a simple (A,!?r,)2= (h~+,)~ = Cn relationship where n = no (1 - 0) is the density of electrons in the overlayer and 6 the oxygen coverage (defining 19= 1 for the first full oxygen layer obtain at Q. exposure). The constant C should be determined by the dielectric function of the substrate and by the so far uncertain character of the plasma oscillations, although the dependence on the former should be rather weak since the same plasmon energy is obtained for close packed Cs on Cu as on W. The observation that A&, is independent of emission angle either means that the dispersion is small or that our loss spectra are dominated by electrons which have suffered both elastic and inelastic scattering. The effect is well known from energy loss studies of the dispersion of bulk plasmons. Due to the high probability of creating plasmons having small wavevectors, q, the loss spectrum representative of small q is superposed via elastic electron scattering on the loss spectrum characteristic of a certain scattering angle and momentum transfer [ 151. Our lowest measured loss energy, 0.5 eV, corresponds to a situation where approximately 90% of the Cs on the surface is ionic. The presence of the plasmon peak at this stage of oxidation shows, obviously, that the remaining Cs atoms are able to keep their valece electrons within the overlayer. If instead 10% of a Cu( 111) surface is covered with Cs the adsorbate is close to ionized [2], almost all of the Cs valence charge being transferred to the substrate. Two factors may contribute to this difference between the Cu( 111) t Cs and the Cu( 11 l)-p(2 X 2)Cs + 02 systems. Firstly the cesium and oxygen ions present in the latter case create a dipole field, which attracts the electrons to the overlayer. Secondly the nearest neighbour configuration is different for Cs atoms on the two surfaces. For Cu(ll1) f Cs and similar systems the Cs adsorbate is uniformly distributed over the surface at submonolayer coverages [2,16,17], which means that the nearest neighbour distance changes continuously ‘with coverage. For the Cu(l1 l)-p(2 X 2)Cs + 02 system those Cs atoms, which have not reacted with oxygen, will have nearest neighbours at almost close packed distance although their number decreases upon increasing
S_,& Lindgren, L. WalldPn/Dilution of electron gas in Cs monolayer
625
the amount of adsorbed oxygen. Even in this case there may be some change of the interatomic distance as a result of lateral contraction accompanying the oxygen adsorption [ 121. To summarize we have associated the linear relationship between the square of the characteristic loss energy and the oxygen exposure with a linear dependence of wp2 on the density, II, of valence electrons in the overlayer. This suggests that the alkali monolayers on metal substrates are systems for which the electron density can be varied continuously and substantially in a rather simple way and, as in the present case, down to quite low values without introducing much additional damping of the collective oscillations. The dispersion of the overlayer plasmons remains uncertain. The constant characteristic loss energy obtained upon changing primary energy and scattering angle suggests that the dispersion is small but for the reasons given above no firm conclusion can be drawn.
Acknowledgement We want to thank H.P. Myers for his support and encouragement, S. Anderson and U. Jostell for many helpful discussions and the Swedish Natural Science Research Council for financial assistance.
References [ 1] A.U. MacRae, K. Miiller, J.J. Lander, J. Morrison and J.C. Phillips, Phys. Rev. Letters 22 (1968) 1048. [2] S.A. Lindgren and L. Wallden, Solid State Commun. 25 (1978) 25. [3] C. Kunz, Z. Physik 196 (1966) 311. [4] R.H. Ritchie, Phys. Rev. 106 (1957) 874. [5] E.A. Stern and R.A. Ferrel, Phys. Rev. 120 (1960) 130. [6] A. Otto, Z. Physik 185 (1965) 232. [7] J.M. Gadzuk, Phys. Rev. Bl (1970) 1267. [ 81 S. Anderson and U. Jostell, Faraday Discussions Chem. Sot. 60 (1975) 255. [9] S. Thomas and T.W. Haas, Solid State Commun. 11 (1972) 193. [ 10) For a review see: H. Raether, in: Thin Solid Films, Vol. 9 (Academic Press, 1977) p. 145. [ 111 P.O. Gartland, S. Berge and B.J. Slagsvold, Phys. Norvegica 7 (1973) 39. [12] C.A. Papageorgopoulus and J.M. Chen, Surface Sci. 39 (1973) 313. [ 131 R.A. Marbrow and R.W. Lambert, Surface Sci. 61 (1976) 329. [ 141 C.B. Duke, L. Pietronero, J.0 Porteus and J.M. Wendelken, Phys. Rev. B12 (1975) 4059. [15] G. Meyer, Z. Physik 148 (1957) 61. [ 161 R.L Gerlach and T.N. Rhodin, Surface Sci. 17 (1969) 32. [ 171 S. Andersson and U. Jostell, Solid State Commun. 13 (1973) 829.