PHOTOEMISSION IN YBCU2SI2 - PROBLEMS WITH THE KONDO IMPURITY MODEL

Abstract We report valence band photoemission results for YbCu 2 Si 2 , The 4 f 13 (J = 7 2 ) final state peak, centered 60 meV below the Fermi level ϵ F , lacks the temperature dependence and is broader than predicted for a Kondo resonance. Together with recent photoemission results for cerium compounds, these results raise serious doubts about the Kondo impurity explanation of heavy fermion photoemission.

The prevailing theory [1,2] of photoemission in heavy fcrmion compounds is the Kondo impurity theory in the I/N approximation. This approach predicts that a resonance should appear in the spectrum, located (for cerium) a distance T K above the Fermi level ,sv (where T K is the Kondo temperature) with a spectral weight proportional TK: the resonance should vanish as the temperature increase above T K. We have recently shown [3] for a series of cerium compounds that there is no correlation between the spectral weights observed in the 4f peaks near E v and the known Kondo temperatures, furthermore the temperature dependence is negligible. This suggests that the spectra are not related to a Kondo resonance.
A problem of cerium compounds is that the resonance is expected to lic abot'e e v; hence, only tlat. w il below e v can bc observed in photocmission. For ytterbium compounds the resonance should bc below e v, and Iwncc i: directly observable [2]. In this paper wc report results for YbCu2Si 2, which has a Kondo temperature of order 50 K as deduced from specific heat [4] or quasielastic neutron linewidth [5]. While a peak which might be identified as a Kondo resonance is observed, our main results are that we find no temperature dependence over a temperature interval [10,300 K], and that the peak is substantially broader than predicted by the Kondo theory.
The spectrum at hu = 182 eV is shown in fig. 1. The large peak near 4.5 cV is due to Cu emission. The peaks in the interval 5-12 eV arise from bulk 4f ~3 Permanent address: Physics department, University of California, lrvine, CA 92717, USA.
4f 12 transitions while the two peaks within 3 eV of the Fermi energy arise from 4f 14 ~4f 13 transitions, with the 4f7/2 final state closest to Ev and the 4f5/-, transition at approximately 1.5 eV binding energy [6]. Also included in fig. 1 is a high resolution spectrum taken at hu = 50 eV. This shows that the 4f 14 spectrum consisits of bulk and surface doublets, with the latter at higher binding energy. From the ratio of the 4f 14 and 4f 13 bulk transitions we estimate [6] that the Yb valence is 2.8-2.9 (the estimate varies somewhat with photon energy).
In the language of Kondo theory [2] the bulk 4f ~4 ---, 4f ~3 transitions are the Kondo resonance (4f7/2, n,'arcst ~t.) and a spin-orbit sideband (4f .. normalized to the Cu emission peak, which should be temperature independent; the surface doublet was fitted to a pair of Gaussians and subtracted off.) It is clear that the temperature dependence is negligible. A complication [2] arises from the known existence [7] of crystal fields in YbCu.,Si,. The level scheme as deduced from neutron scattering [7] is shown in fig. 3. The ground state doublet has a width 4 meV. Excited state doublets at 12, 30 and 80 meV have neutron linewidths of order 8 meV; this width is belicved to arise from Kondo scattering, and it givcs the scalc on which thc crystal field Kondo sidcbands should rcnormalizc to zcro [2]. Tire obsc~,cd 4f7/2 pcak near ¢1should bc thc sum of taesc sidcbands, convoluted with inslrumcntal rcsolutior, as shown in fig. 3. Direct comparison to our data for figs. 2 and 3 dcmonstratcs that th,zrc is no tcmpcratu,c dependence, neither on the scale of the ground state Kondo temperature, nor on the scale of the sideband Kondo temperature (100 K); secondly the observed emission peak is substantially broader than the peak predicted by convoluting the sideband peaks with instrumental resolution.
The prediction of Kondo impurity theory that the Kondo resonance should rcnormalize to zero implies that the valence should approach the value 3 at high tempcrature [2]. Our results seem to imply that thc valence remains constant. Perhaps in the Kondo lattice the valence is stabilized. Alternatively, photoemission may be incorrectly measuring valence, perhaps due to   Fig. 3. The 4f t3 (J= 7/2) f'-tate peak. measured at hv = 60 eV with 6(I meV reso..tlon and at T = 20 K. The dashed curve in both the bottom and top plots is a convolution of the reported crystal field sidebands with instrumental resolution. The data are substantially broader than this fit.