HIGH-ENERGY SPECTROSCOPIC STUDY OF THE ELECTRONIC-STRUCTURE OF UBE13

X-ray photoemission and bremsstrahlung isochromat spectroscopies have been used to probe the occupied and unoccupied states of UBe». Between two and three electrons are found to populate the tail of a surprisingly broad 5f band ( — 5 eV) of extended states. With our resolution ( & 0. 5 eV) it is not possible to observe directly at the Fermi energy any peculiarity of the density of states explaining the extraordinary properties of this compound. However, the drastic differences between the core-level spectra of U and Be indicate that the Sf states remain essentially confined around the U atoms and are only weakly hybridized with the sp-band states originating from the Be atoms.

The recent discovery of bulk superconductivity in UBe» below 1 K (Ref. 1) has given clear evidence for the existence of strongly interacting electrons in some metallic materials, giving rise to anomalous properties at low temperatures.
Other examples for this type of materials are CeA13 (Ref. 2) and CeCu2Si2 (Ref. 3), again two compounds containing f electrons, obviously a prerequisite for the observation of the behavior to be discussed here. All these compounds show an anomalously large specific heat c~( T) at low temperatures.
In CeA13, c~( T) decreases linearly with decreasing temperature below 1 K but, in comparison with normal metals, with a very large coefficient y of the order of 1.5 J/mole K, indicating a considerable renormalization of the electronic subsystem. In UBet3 (Ref. 1) and CeCu2Si2, this linear decrease with a slope of about 1 J/mole K is intercepted by a discontinuity Ac due to a superconducting transition. In both cases, the magnitude of Ac is compatible with the large values of c~/T just above the transitions and the experimental verification that the entropy difference between the superconducting and the normal state is zero below T, demonstrates that it is indeed the strongly interacting electrons that are involved in the superconducting state. These high y values are necessarily also based on large densities of electronic states at the Fermi energy E+, implying very narrow features in the energy dependence of the electronic structure at E+. The possible occurrence of such features is well known from theoretical work concerning the electronic structure of simple metals containing transitionmetal impurities in the dilute limit. " Recent work considering concentrated systems claims that similar narrow resonances at E~also appear in this case. Information on the energy dependence of the electronic structure of a metal is provided by using photoemission techniques. A first attempt concerning UBe» involved resonant photoemission, favoring the emission of electrons with f symmetry, by scanning the energy range below EF to about 12-eV binding energy.
Since also core-electron spectra and particularly the energy distribution of empty electron states provide valuable information, we chose to map parts of the electronic spectrum of UBet3 by using x-ray photoemission (XPS) and bremsstrahlung-isochromat spectroscopy (BIS).
The sample investigated in the present study was a platelet cut from the same polycrystalline batch of material that was used for previous specific-heat measurements. ' The XPS and BIS spectra were obtained in a combined instrument described elsewhere.~The contamination was removed from the sample surface by scraping it in situ with an A1203 file until the 0 1s and C 1s XPS signals could no longer be detected in a 5-min scan. The base pressure of 1x10 " Torr in the instrument allowed us to accumulate the different spectra during periods of many hours without any sizable degradation of the surface cleanliness. separation which is more than S0% larger in UBe13 than in a-U.
In 8 limited energy range above E~the unoccupied states have been probed by BIS and thc corresponding spectrum is shown on the right side of Fig. 1. Since the matrix elements for the BIS transitions are practically the same as those accounting for XPS transitions, also the BIS spectrum represents nearly exclusively the empty 5 f states superimposed on a background increasing toward higher energies and attributable to electrons which have been scattered inelastically prio~to the BIS process. Whereas two marked peaks are observed in o. -U, ' the BIS spectrum of UBe13 shows a single and featureless peak with its maximum at 1.3 CV. Qn the low-energy side, the clear change of slope at 0.4 cV allows us to observe the position of the Fermi edge which is found to coincide with the position determined by the usual energy-scale calibration performed with Au. The Koopmans's approximation is very accurate at the Fermi energy of metals, '3 so that it is well justified to join the XPS and HIS spectra at EF in order to depict the whole 5f band.
The main uncertainty in this attempt is the calibration of the relative intensities observed with the two different techniques. In thc pl'cscnt sltuatlon, thc comparison of thc FerII-cdge intensities is not suitable: the two techniques have different line shapes and linewidths89 and, furthermore, E~is located in a very steep DOS. A more reliable approach ln determining thc rclatlvc lntcnsltlcs consists ln joining thc two spectra above and belo~the range affected by the Fermi cutoff by a smooth curve. As shown in Fig. I, this criterion allo~s us to represent rather precisely a continuous DOS crossing E~. After subtraction in both spectra of a background proportional to the integrated signal intensity from E~to the respective energy and joining smoothly the spectra away from E~, the area of the two curves represent the total band intensity.
As mentioned previously, the contribution of the I=2 projected DOS to the spectra is small but not entirely negligible. For this reason the experimental intensity ratio Ists/Ixps=3. 7 does not directly represent the ratio of the empty to occupied 5 f states but it must be corrected for the presence of the ten 6d states located in the same energy range. It is straightforward to express Ims/Ixps as a function of the numbers of the occupied states (n/, nd) and of the In a metallic bond, only a very small charge transfer is expected between the Be and U atoms, as confirmed belo~in the discussion of the core level binding energies. The approximate neutrality around the U atoms yields the additional condition for the number of occupied f and d states nz+ n»--6 which allows Us to extract thc value n»--2.8.
The uncertainties affecting the numbers used in this estimation are not influencing markedly the value found for n». In order to demonstrate this fact, we have calculated the range of each parameter for which n» varies from 2 to 3: 3 5~I sis/Ixps 1.75~a (5f')/a. (6d')~19 5.4~ng+ n»~10.4 The variation of (nz+ nf) describes implicitly the neglected presence in U of s and p states which have different cross sections than the d states. The acceptable errors in these nuITlbcls 8rc smaller than thcsc limits and w'c can safely conclude that the metallic band of UBe13 is occupied by a number of 5 f electrons not far from 3, but in any case between 2 and 3. This 1'csUlt ls ln dlsagrccmcnt with thc valUc of n» -dcrlvcd from plcvloUs photocmlsslon spectra excited by low-energy photons.
Less surprising is the discrepancy with the value of n» between 1 and 2 predicted by a free-ion interpretation of the effective paramagnetic moment of 3.08@~. 7 Figure 2 shows the XPS spectrum of the 5f core level of U. The maximum of the 4f7/2 line located at 377.35 eV can be considered to yield accurately the binding energy of this level. It lies in the narrow energy range from 377.1 to 377.4 CV where the 4f7/2 binding energies of n-U, US, UAs, USb are found. ' It is interesting to note that the peculiar behavior of UTe manifests itself by a binding energy of 378.1 cV (Rcf. 14) and thc tctl'avalcncc of thc U lolls con- eV, respectively. These observations confirm our intensity analysis of the outer level spectra leading to the conclusion that the Sf population is undoubtedly larger than 2. The very pronounced asymmetrical shape of the lines shown in Fig. 2 reflects a high metallic DOS around EF allowing the excitation of a great number of low-energy electron-hole pairs in the screening mechanism of the core hole around the U ions. An intense and broad satellite is also observed at about a 7 eV larger binding energy than each main 4f signal. The satellites most likely account for an atomiclike reaction of the outer electrons to the deep hole creation and are due to a local breakdown of the band behavior around the ionized atom. Such effects are particularly intense in narrow bands where the correlation is so important that the large relaxation energy requires the existence of such highly excited final states. The exact nature of this puzzling final state is not established but it may correspond to an integral Sf occupation (Sf or Sf3). It has been pointed out'7 that the intensity of this satellite observed in many different U compounds is to some extent correlated with the value of the effective paramagnetic moment, a fact which is also verified in UBet3. Probably the hybridization strength of the Sf states of U with the sp orbitals of the numerous surrounding Be atoms is the key parameter of this problem and not the U-U spacing. ' ' In UBe~3 this hybridization seems to be just strong enough to involve the Sf states in the band but can still be locally broken by the atomic potential increase resulting from the core-level ionization. This situation could provide an explanation for the correlation observed between high values of the effective paramagnetic moment and intense satellites. ' Finally, the absence of shake-down satellite is a further confirmation of the extend-