We have performed a comparative study of six classical many-body potentials for silicon (Pearson, Takai, Halicioglu, and Tiller; Biswas and Hamann; Stillinger and Weber; Dodson, Tersoff 2, and Tersoff 3). Extensive static calculations have been performed using these potentials on ${\mathrm{Si}}_{\mathit{n}}$ clusters (n=2–6), bulk point defects, elastic constants, polytypes, pressure-induced phase transformations, and surfaces [(111), (100), and (110)]. Similarities and differences between the six potentials have been identified, and their transferability as well as their accuracy with respect to experiment and first-principles methods have been assessed. In general, all of these potentials do a relatively poor job of modeling the energetics of small clusters as well as the various reconstructions of the Si(111) surface. They provide a fair to good description of the properties of bulk diamond cubic silicon, its intrinsic defects, and the Si(100) surface. Besides the fact that none of them models π bonding, their inability to be more transferable lies in their inadequate description of the angular forces. Each potential has its strengths and limitations, but none of them appears to be clearly superior to the others, and none is totally transferrable. However, despite their shortcomings we feel that some of these potentials will be useful in large-scale simulations of materials-related problems. They can give valuable insights into phenomena that are otherwise intractable to investigate either experimentally or via first-principles methods.

• Received 13 January 1992

DOI:https://doi.org/10.1103/PhysRevB.46.2250