A detailed reexamination of the (110) surface structure of rutile TiO 2 has been carried out using first-principles total-energy methods. This investigation is in response to a recent high-precision LEED-IV measurement revealing certain significant quantitative discrepancies between experiment and previous theoretical calculations. We have been able to resolve these discrepancies and achieve excellent quantitative agreement with experiment by judicious attention to reducing computational approximation errors. Our analysis also reveals that bond lengths converge much faster with slab thickness than do relaxed absolute atomic positions, which are the structural parameters typically reported in the literature for this and related systems. The effect this observation has on both the choice of slab models and the way in which surfaces structures should be reported for covalently bonded solids is discussed. Finally, the efficacy of freezing the lowest several atomic layers of TiO 2 slab models in their bulk-like positions is examined.
A recent LEED-IV experiment1 has produced what is arguably the highest precision measurement to date of the structure of the (110) surface of rutile TiO 2 .Comparison of these data to various published firstprinciples theoretical investigations of this system [2][3][4][5][6][7][8] reveals that, while some predicted atomic positions are in good agreement with the new experiment, othersmost noticeably those of the surface oxygen atomsare not. Indeed, the vertical relaxation of the bridging oxygen atoms was consistently predicted to be negative (i.e., inward relaxation relative to bulk termination), but was measured to be positive in Ref.[1]. An earlier surface x-ray diffraction study 9 had reported a significant inward relaxation for the bridging oxygen. Thus, the relaxed structure of the surface oxygen atoms has been a source of confusion both experimentally and theoretically.Given the prominence of TiO 2 (110) in applications and basic research, 10 as well as the extensive number of first-principles investigations on systems that build on TiO 2 (110) as a starting point, we felt this confusion was significant and merited further inquiry. As a result, we have undertaken an extensive first-principles analysis of the structure of TiO 2 (110), focusing on convergence and characterization issues. Our calculations resolve the discrepancy between theory and experiment on the relaxations of the surface oxygen atoms, while maintaining agreement in the other atomic relaxations. Furthermore, we find that much thicker slabs than have been typically used are necessary to achieve convergence not only in surface energy, as found previously, 11 but also in absolute atomic positions. However, if bond lengths are chosen as the "figures of merit" for assessing convergence, then our results show that the thinner slabs used to date in theoretical studies of this system are acceptable. We argue below in favor of switching to reporting bond lengths (and perhaps bond angles) as a more physically relevant descri...