DFT-Vorhersage der Kristallstruktur und Eigenschaften intermetallischer Actinoidverbindungen: Die Beispiele UCo und UIr

In this work, we tested the quality of density functional theory (DFT) for predicting the crystal structures and physical properties of the two intermetallic actinoid compounds UCo (I2_13, cI16) and UIr (P2_1/c, mP16). Scalar-relativistic calculations with the evolutionary algorithm USPEX using the LDA-PW, the GGA-PBE, and the DFT+U methodology, predict for both UCo and UIr not the observed structure but the NaTl structure type (Fd3-m, cF16) as the energy minimum. The relative energetics of UCo can be corrected by including spin-orbit interactions. For UIr, the energy difference between the predicted and experimentally determined structural model is too large. It would not be considered in further higher accuracy analyses without prior knowledge of the expected outcome. We explain the difference in prediction quality by the extent of electron correlation in UCo and UIr. UCo, similar to α-U, is a moderately correlated metal and still well described by the chosen methodology. The properties of UIr are similar to those of δ-Pu, which is considered a strongly correlated metal. This limits the usability of DFT at the LDA or GGA level. A boundary of predictability thus runs between UCo and UIr. Therefore, reliable predictions can be expected only for compounds with thorium or protactinium, since they should exhibit even lower correlation effects. For other actinide systems, methods of higher accuracy in calculating the internal energy are needed. For this purpose, we investigated the influence of spin-orbit interactions and the DFT+U correction. Spin-orbit effects improve the relative energetics and we recommend including them due to their physical motivation. This is not so for the DFT+U method and we do not recommend it for the systems in this work. In addition to technical problems, it does not provide a systematically controllable improvement in relative energetics. We expect an improved prediction quality for intermetallic actinoid compounds from many-body models, such as DFT+DMFT, which take dynamic effects into account. We investigated qualitatively causes for the limited prediction quality in terms of the fundamental errors of the DFT approximation. The self-interaction or delocalization error manifests itself primarily in an overestimation of the bonding interactions of the 5f electrons. This favors structure types with short distances between actinoid atoms, here the UCo and NaTl type, which form broad 5f bands. Experimentally, attractive interactions of the 5f orbitals are induced by pressure. Thus, both structure types are potential high-pressure modifications of the group of intermetallic actinoids. The DFT self-interaction error possibly leads to an underestimation of the phase transformation pressures of these compounds. The static correlation error results primarily in an overestimation of spin polarization of the 5f electrons, which in some sense mimics the localization and correlation of these electrons. Since both, localized and spin-polarized electrons do not contribute to bonding, good structural predictions are still possible at the expense of a possibly spin-polarized ground state. This is demonstrated by DFT structure optimizations on UCo and UIr. In the first case, the proposed structure of the UCo type (I2_13, cI16) can be confirmed. In the case of UIr, the DFT calculations provide the motivation for the experimental redetermination and correction of the crystal structure from the acentric space group P2_1 to the centrosymmetric space group P2_1/c. Accordingly, UIr is not an acentric superconductor, as has been widely discussed. The overestimation of spin polarization can lead to problems in the prediction of physical properties. This is illustrated by the probably untenable prediction of half-metallic ferromagnetism in NaTl-type UCo or the magnetic stabilization of CrB-type UIr. Helpful DFT predictions are possible especially around the local minimum of the experimentally observed structures. Here, error compensation increases the accuracy of the calculations of relative energies. An example of this is the predicted magnetic easy-axis of UIr, which agrees with experimental observations. A comparison with valence band spectra shows that the electronic structures of the compounds studied here are reproduced qualitatively. Their analysis provides an approach to rationalize the occurrence of the structure types found: Both the UCo and NaTl type allow the formation of short, homoatomic contacts of the actinoid atoms, which most effectively enhance the attractive interactions of the 5f orbitals. This reduces the energy needed for compressing these potential high-pressure compounds.