Gradient corrections have been shown to improve the accuracy of density-functional theory (DFT) when applied to homonuclear dimers, small molecules, and bulk properties of transition metals (such as Fe). A more thorough evaluation is needed before adopting these corrections for large-scale computations in solids. We investigate a broad range of different systems by using the most recently proposed gradient correction to DFT by Perdew and Wang, which is expected to give a marked improvement over earlier attempts. We find that when this correction is used to calculate the energy of atoms it gives better agreement with experiment than the local-density approximation (LDA). When applied to bulk properties, this correction gives results which are not consistently better and tend to overcorrect LDA results.

• Received 25 February 1993

DOI:https://doi.org/10.1103/PhysRevB.48.14944