ABSTRACT
It has long been known that the local density approximation and the generalized gradient approximation do not furnish reliable band gaps, and one needs to go beyond these approximations to reliably describe these properties. Among alternatives are the use of hybrid functionals (HSE03 and HSE06 being popular), the GW approximation or the recently proposed LDA-1/2 method. In this work, we compare rigorously the performance of these four methods in describing the band gaps of alloys, employing the generalized quasi-chemical approach to treat the disorder of the alloy and to obtain judiciously the band gap for the entire compositional range. Zincblende InGaAs and InGaN were chosen as prototypes due to their importance in optoelectronic applications. The comparison between these four approaches was guided both by the agreement between the predicted band gap and the experimental one, and by the demanded computational effort (time and memory). We observed that the HSE06 method provided the most accurate results (in comparison with experiments), whereas, surprisingly, the LDA-1/2 method gave the best compromise between accuracy and computational resources. Due to its low computational cost and good accuracy, we decided to double the supercell used to describe the alloys, and employing LDA-1/2 we observed that the bowing parameter changed remarkably, only agreeing with the measured one for the larger supercell, where LDA-1/2 plays an important role.
