The material BaCoS\({}_{2}\) is a layered antiferromagnetic Mott insulator which transitions into a metal under pressure concommitant with the suppression of the magnetic ordering temperature. Many novel superconductors are found in the region of phase diagram where magnetism is suppressed. However, it’s not always the case that if (a) we have a magnetic layered magnetic compound, and (b) we suppress ordering, then (c) we necessarily get superconductivity. There are other, less-understood factors at play, and BaCoS\({}_{2}\) is an example.

Motivated by these ideas and the observation of a pressure-tuned metal-insulator transition, there was an experimental effort to apply chemical pressure via substitution of the large sulfur ion by the smaller oxygen ion in BaCoS\({}_{2}\) to form BaCoSO. The fully-substituted end member BaCoO\({}_{2}\) was already known. In this work, we attempted to predict the structure of BaCoSO without input from experiment.

Electronic structure – We have performed DFT calculations for BaCoSO using the experimentally-confirmed structure, as described below. The orbitally-resolved densities of state are plotted in Fig. \ref{fig:bacoso-dos} [Ran, I think it would be a good idea to have at least a DFT DOS plot for BaCoSO]. From DFT, we find a \(d\)-metal where the density at the Fermi level is composed mainly of the Co \(3d\) states. As expected, these states hybridize moderately with the oxygen and sulfur \(p\) states, which are located XXXeV below the Fermi level.

Structure prediction – It has been commonly assumed that LDA/GGA is sufficient for structure prediction. Whereas comparisons between compounds with differing compositions certainly require the corrections detailed above, it has been assumed that the systematic errors in LDA/GGA energies should cancel for differing structures of a *single* given composition. We argue this is not the case.

We propose the following two-step algorithm: use a structure prediction algorithm running on top of LDA to produce a list of candidate structures, then reorder these structures according to energies computed via LDA+U. This method has the advantage of being relatively fast, as structure prediction using LDA+U or LDA+DMFT/GW would be prohibitively expensive, yet capturing the essential effect of correlations via the second re-ordering step.

We apply this strategy to the Ba-Co-S-O system by first using USPEX, a structure prediction package based on genetic algorithms, to sample the local minima in the energy landscape. We choose a 1:1:1:1 ratio for the elements and allowed for two formula units in a unit cell. We use VASP as our DFT engine. We use spin-polarized LDA/GGA and do not include U corrections. Around 300 k-points are used, with convergence criteria EDIFF=1e-6, NELM=40.

To capture the relevant local minima, we retain all candidate structures produced in any of the USPEX generations that lie within 0.5eV/(unit cell) of the final lowest energy structure. We group together similar structure using the criterion that their symmetry groups are identical and their computed energies are less than 3meV apart. As a result, of the original set of 152 structures we keep 58 distinct ones. The energies of this set of structures are then examined as a function of \(U\), which we plot in Fig. \ref{fig:reordering}. We kept the structures fixed as tuned \(U\), i.e. no structural relaxation.