Research Article

Orbital Reconstruction and Covalent Bonding at an Oxide Interface

See allHide authors and affiliations

Science  16 Nov 2007:
Vol. 318, Issue 5853, pp. 1114-1117
DOI: 10.1126/science.1149338


Orbital reconstructions and covalent bonding must be considered as important factors in the rational design of oxide heterostructures with engineered physical properties. We have investigated the interface between high-temperature superconducting (Y,Ca)Ba2Cu3O7 and metallic La0.67Ca0.33MnO3 by resonant x-ray spectroscopy. A charge of about –0.2 electron is transferred from Mn to Cu ions across the interface and induces a major reconstruction of the orbital occupation and orbital symmetry in the interfacial CuO2 layers. In particular, the Cu d3z2r2 orbital, which is fully occupied and electronically inactive in the bulk, is partially occupied at the interface. Supported by exact-diagonalization calculations, these data indicate the formation of a strong chemical bond between Cu and Mn atoms across the interface. Orbital reconstructions and associated covalent bonding are thus important factors in determining the physical properties of oxide heterostructures.

In semiconductor heterostructures, high-mobility electron systems with tunable density have led to prominent advances in science and technology over the past decades. Such systems have recently been replicated in heterostructures of complex transition metal oxides (1), leading to the observation of transistor effects (2) and the quantum Hall effect (3). Because transition metal oxides exhibit a notably rich phase behavior in the bulk (4), these developments have raised expectations that quantum states with properties and functionalities qualitatively beyond those attainable in semiconductors can be generated at oxide interfaces.

The large variety of phases (often with radically different physical properties) in transition metal oxides is due to the delicate sensitivity of the charge transfer and magnetic interaction between metal ions to the occupation of d orbitals (5). Which linear combination of the five possible d orbitals is occupied on a given transition metal site depends, in turn, on parameters such as electron density, ligand positions, magnetic order, and chemical bonding, which are generally different at the interface than in the bulk. Despite its pivotal role in determining the phase behavior and physical properties of oxides, almost no experimental information is available about the occupation of orbitals at oxide interfaces, and theoretical work (611) has thus far hardly addressed this issue. We report the results of soft x-ray absorption spectroscopy (XAS) and soft x-ray linear dichroism (XLD) experiments on heterostructures of copper and manganese oxides tailored to probe the electronic structure and orbital occupation at the interface. The cuprate-manganate interface is well suited as a model system for this purpose, because nearly strain-free, atomically sharp heterostructures can be synthesized (1214) and because the electronic properties of both materials have been studied extensively in the bulk.

Probing heterostructure interfaces. The experiments were performed at the 4-ID-C beam-line at the Advanced Photon Source on epitaxial trilayers and superlattices of the high-temperature superconductor (Y,Ca)Ba2Cu3O7 (YBCO) in c-axis orientation, combined with ferromagnetic metallic La1–xCaxMnO3 (LCMO) at a doping level x = ⅓. The quality of these multilayer structures was checked by a variety of characterization methods (15). In order to discriminate the electronic structure at the interface from surface and bulk contributions, we have performed a systematic series of experiments on heterostructures with different capping layers, taking advantage of the element specificity and shallow probing depth of resonant XAS and XLD in the total electron yield (TEY) mode (Fig. 1A). For instance, the occupation of Cu d orbitals on the YBCO side of the interface was studied on heterostructures with LCMO capping layers, so that no surface Cu is present. If the photon energy is tuned to the Cu L absorption edge, the capping layer does not influence the detected signal apart from an overall attenuation factor. As a result of the low electron escape depth (a few nanometers), the TEY signal is dominated by the CuO2 layers immediately adjacent to the first interface; contributions from deeper layers are exponentially reduced. The interface sensitivity is further enhanced with the use of a low angle of incidence for the x-ray beam (11.2°). The converse procedure was used to probe the electronic structure of MnO2 layers on the LCMO side of the interface. Control experiments in the bulk-sensitive fluorescence-yield (FY) mode were simultaneously carried out in both cases.

Fig. 1.

(A) Schematic of the experimental setup used to obtain the XAS and XLD data in TEY and FY modes. Data sensitive to interfacial Cu (Mn) atoms were taken in TEY mode with photon energies near the Cu (Mn) L absorption edge, on samples with LCMO (YBCO) capping layers. To obtain a sizable dichroism, we tilted the film plane with respect to the photon beam propagation direction. C indicates the c-axis of the film; H is the applied magnetic field; h and v denote the linear polarization state of the incident x-ray. (B) Atomic positions near the LCMO-YBCO interface (14, 29). The MnCuO10 cluster used for the exact-diagonalization calculations is highlighted.

Spectroscopy at the interface.Figure 2 shows normalized absorption spectra near the Cu L3 edge in bulk- and interface-sensitive modes. The bulk-sensitive FY data are in excellent agreement with previous XAS data at the Cu L edge of nearly optimally doped YBCO (16). The main narrow absorption peak around 931 eV corresponds to the intra-ionic transition 2p63d9 → 2p53d10. The shoulder on the right-hand side of the peak is attributed to the intersite transition 2p63d9L → 2p53d10L, where L denotes a hole on the oxygen ligand. The line shape of the main absorption peak is a signature of the “Zhang-Rice singlet” (17), a bound state of charge carriers on oxygen and copper sites that keeps the Cu plane site in the nominal valence state 2+ as the hole density in the CuO2 sheets is tuned by doping. The polarization dependence of the FY signal also contains important information about the electronic structure near the Fermi level of YBCO. In particular, the absorption for photon polarization parallel to the CuO2 sheets greatly exceeds that for polarization along the c axis. This implies that holes in the conduction band of YBCO predominantly occupy the planar Cu dx2y2 orbital, which hybridizes strongly with oxygen p orbitals in the CuO2 layers. Similar observations have been made in all other high-temperature superconductors investigated thus far, and together they have become one of the basic tenets of our current understanding of this class of materials (16, 18).

Fig. 2.

Normalized x-ray absorption spectra at the Cu L3 absorption edge, taken in bulk-sensitive (FY, top panel) and interface-sensitive (TEY, bottom panel) detection modes with varying photon polarization as indicated in the legend. a.u., arbitrary units.

Evidence for orbital reconstruction and charge transfer. The interface-sensitive data shown in Fig. 2 are very different. One first notices that the interfacial absorption peak is shifted to lower energy with respect to the bulk by ∼0.4 eV and that the high-energy shoulder is no longer present. The shift of the peak is evidence of a change in valence state of Cu ions near the interface. This indicates that charge is transferred across the interface and that a charged double layer is formed, as generally expected for heterostructures of materials with different work functions. In agreement with specific predictions for the system at hand (11), the charge-transfer direction is such that the hole density in YBCO is reduced at the interface. Because of the strong influence of the core hole created by absorbing the photon, the relationship between the x-ray absorption edge and the Cu valence is not straightforward, but a comparison to XAS spectra of reference materials containing Cu1+ and Cu2+ ions [for a review, see (19)] yields a rough estimate of 0.2e (where e in the charge on the electron) per copper ion for the charge-transfer amplitude. At first sight, this seems to correspond with the line shape of the interfacial absorption peak, which bears a strong resemblance to XAS data in undoped YBCO (16). Notably, however, numerous XAS experiments on YBCO and other bulk hole-doped high-temperature superconductors have shown that the position of the Cu L-absorption peak is independent of doping. This has been attributed to the Zhang-Rice singlet state and, consequently, the doped holes have predominantly oxygen character (17). The observed shift of the L3 absorption peak in our interface-sensitive experiment thus cannot be attributed to a readjustment of the hole density alone and indicates an extreme modification of the electronic structure of the CuO2 layer adjacent to the interface.

In order to uncover the origin of the unexpected shift of the absorption peak and to obtain further information about the electronic states at the interface, we have varied the photon polarization in the interface-sensitive detection mode (Fig. 2). In marked contrast to the bulk-sensitive data, the strengths of the absorption signals for polarization perpendicular and parallel to the layers are almost equal. This is a manifestation of an “orbital reconstruction.” Whereas the holes are constrained to the Cu dx2y2 orbital in the bulk, at least some of them occupy the d3z2r2 orbitals at the interface. The distribution of holes over the two Cu orbitals cannot be precisely determined, because the XLD experiment probes not only the CuO2 layer directly at the interface but also the deeper layers (albeit with exponentially reduced sensitivity). However, the nearly isotropic cross section shown in Fig. 2 implies that the hole content of the Cu d3z2r2 orbital is at least equal to that of the dx2y2 orbital. We repeated the measurement at several temperatures (from 300 to 30 K) and confirmed that the peak position and polarization dependence do not depend on temperature. Similar observations were also made on heterostructures in which the doping level of YBCO was raised into the over-doped regime by Ca substitution. The orbital reconstruction and the charge transfer are hence general, robust characteristics of the YBCO-LCMO interface.

Before discussing possible mechanisms and potential implications of the orbital reconstruction, we briefly discuss XAS spectra near the Mn L2 and L3 absorption edges taken in bulk- and interface-sensitive modes (Fig. 3). The bulk-sensitive data are again in good agreement with corresponding data in the literature. The spectra are much broader than those taken near the Cu L edge, because all of the five Mn d orbitals are partially occupied, giving rise to a complicated multiplet splitting of the absorption peak. The peak intensity is independent of photon polarization within the experimental error. This finding has been taken as evidence of an orbitally disordered state with equal occupation of Mn dx2y2 and d3z2r2 orbitals in bulk metallic LCMO. In the interface-sensitive detection mode, neither the peak position nor its polarization dependence are noticeably different from the bulk data. This does not imply, however, that the Mn ions maintain their bulk charge density and electronic structure at the interface. Indeed, as a result of charge conservation, one generally expects a shift in Mn valence matching that of the interfacial Cu ions (Fig. 2), but because of the strong multiplet broadening of the Mn peak, such a shift is much harder to recognize than in the case of Cu (20). Based on the data of Fig. 3, one can set an upper bound of 0.4 eV on the difference between the positions of Mn L absorption edges in bulk- and interface-sensitive detection modes. Because a valence change from Mn3+ to Mn4+ results in a shift of the L edge of ∼1.5 eV, this translates into an upper bound of ∼0.3e per Mn atom on the amplitude of the charge transfer across the interface, which is consistent with the estimated amplitude of ∼0.2e based on the Cu XAS spectra discussed above. Likewise, because the polarization dependence of the intensity at the Mn L edge is influenced to a large extent by the completely unoccupied minority t2g and eg orbitals, it is difficult to see a rearrangement of the majority dx2y2 and d3z2r2 orbitals comparable to that observed on Cu.

Fig. 3.

Normalized x-ray absorption spectra at the Mn L2 and L3 absorption edges, taken in bulk-sensitive (FY, top panel) and interface-sensitive (TEY, bottom panel) detection modes with varying photon polarization as indicated in the legend. The line shape of the FY spectra is distorted by self-absorption effects.

Mechanism of orbital reconstruction. Our data therefore imply that the interfacial Cu d3z2y2 orbitals, which are fully occupied in bulk YBCO, are partially populated by holes at the interface. In principle, two distinct physical mechanisms could lead to such an orbital reconstruction. First, it is possible that the different crystal-field environment of Cu ions at the interface could raise the energy of the d3z2y2 orbital above that of the dx2y2 orbital. Because the ligand positions at the interface are not precisely known, this scenario cannot be firmly ruled out, but it is highly unlikely because of the large energy difference between Cu d3z2r2– and dx2y2–derived bands in bulk YBCO. A reversal of this hierarchy would require a substantially shorter distance between the copper and apical oxygen O(2) ions as compared with the in-plane Cu-O bond length, which is unrealistic. Furthermore, the major difference between the bulk and interface crystal-field environments is the substitution of Cu-chain ions (with valence close to 2+ in bulk YBCO) by Mn ions (with valence ∼3.3+ in bulk LCMO). The higher ligand charge should lower the energy of the d3z2r2 orbital and further increase the energy difference with the dx2y2 level. A major rearrangement of the orbital level scheme due to readjustments of ligand positions is therefore implausible. The second scenario is based on the observation that the Cu d3z2r2 orbital points directly toward the interface and can hybridize effectively with the Mn d3z2r2 orbital via the apical oxygen ion [O(2) in Fig. 1B], generating a covalent chemical bond bridging the interface. In this scenario, covalency results in the formation of extended “molecular orbitals” consisting of atomic Cu and Mn d3z2y2 orbitals with an admixture of the pz orbitals on the apical oxygen (insets in Fig. 4).

Fig. 4.

Occupancy of Cu d orbitals at the LCMO-YBCO interface as a function of Mn hole on-site energy, as predicted by the exact-diagonalization calculations described in the text. The occupancy is given by the total number of holes, measured from the full-shell (3d10) electron configuration. The corresponding formal Cu valence states are indicated for clarity. The insets show the orbital level scheme at the interface, including extended bonding (B) and antibonding (AB) “molecular orbitals” formed by hybridized Cu and Mn d3z2r2 orbitals. The hole is indicated as the green circle.

Covalent bonding at the interface. To check whether the covalent-bonding scenario is viable, we performed exact-diagonalization calculations of the MnCuO10 cluster shown in Fig. 1B, including a single hole and the full set of Cu d orbitals and interaction parameters described in the literature (15, 21). Because the Mn dx2y2 orbital does not hybridize with Cu, the Mn ion is represented by a single d3z2r2 orbital with a classical Hund's rule coupling to the core t2g spins. In order to simulate the difference in YBCO and LCMO work functions, we tuned the on-site energy of the hole on Mn. Figure 4 shows the hole density in the Cu d shell (measured from the full-shell configuration 3d10) as a function of this parameter. For large values, the hole resides completely on the Cu ion, and the ionic valencies are close to their bulk values (i.e., Cu is in the formal valence state 2+, and the Mn valence is 3+). The formal Cu2+ valence state, realized at high Mn hole on-site energy, corresponds to ∼ 0.76 holes in the Cu d shell, which reside in the Cu dx2y2 orbital. The remaining hole density is in the in-plane oxygen p orbitals [O(3) in Fig. 1B], which hybridize strongly with Cu dx2y2. The plot also shows that the charge transfer across the interface leads to a major rearrangement of the electron distribution in the Cu eg orbital manifold. Indeed, when the transfer is complete, the Cu dx2y2 orbital is completely full, and holes partially occupy the d3z2r2 orbital. This indicates that the Mn and Cu d3z2y2 orbitals are indeed strongly hybridized and that molecular orbitals are formed. The charge-transfer transition occurs when the antibonding molecular orbital crosses the Cu dx2y2 level, as shown in the insets. The change in hole density on Cu induced by the charge-transfer transition is another signature of the formation of a covalent bond: Figure 4 shows that only about half of the hole charge ends up on Cu; the remaining hole charge is distributed over the Mn- and O(2)-derived components of the molecular orbital.

The cluster calculation demonstrates that the formation of a strong covalent bond between Cu and Mn ions at the interface is indeed realistic. Partial occupancy of the corresponding antibonding molecular orbital qualitatively explains the modification of the XLD spectra at the interface. The filling factor of this orbital is expected to match that of the Mn d3z2y2 orbital, which is about one-third in bulk LCMO. According to the calculation, a substantial fraction of the charge density in the hybrid orbital resides on Cu, which is in good agreement with the experimentally observed shift of the Cu valence. Because holes in this orbital are not subject to Zhang-Rice singlet formation, the shift of the Cu XAS peak at the interface is also explained. Also, the Cu-Mn orbital hybridization naturally results in a strong antiferromagnetic exchange coupling (15), as recently observed at LCMO-YBCO interfaces (22, 23).

Needless to say, the cluster calculations have some limitations. In particular, the energy levels in the cluster are sharp, and the Cu-Mn molecular orbitals are nearly orthogonal to Cu dx2y2, because mixing is only due to weak spin-orbit coupling effects. The charge-transfer transition shown in Fig. 4 is therefore abrupt. In an extended system, the energy levels are broadened into bands, and the two bands near the Fermi level can be partially mixed. The charge carriers on interfacial Cu ions are thus generally expected to have mixed dx2y2 and d3z2y2 character, as experimentally observed. More elaborate ab initio calculations are required to obtain a quantitative description of the band dispersions at the interface. However, given the complexity of the unit cells and the broken translational symmetry at the interface, this presents a formidable challenge to current computational capabilities.

Concluding remarks. Our experiments show that the electronic structure of the CuO2 layer is modified by covalent bonds across the interface. These results suggest that the orbital rearrangement and strong hybridization are at least partially responsible for the unusual magnetic behavior previously observed at cuprate-manganate interfaces (22, 23) and contribute to the suppression of superconductivity near the interface (13). Further, the valence electrons of a large variety of transition metal oxides, whose properties in heterojunctions have been extensively investigated [including manganates (24), titanates (25), vanadates (26), ruthenates (27), and ferrites (28)], reside in nearly degenerate d orbitals and are hence subject to hybridization at interfaces.

Supporting Online Material

Materials and Methods

Figs. S1 to S3


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article