As computers get faster, researchers are on the lookout for more reliable and practicable methods to model materials on the atomic scale. The increasing range of such computational techniques—broadly divided into a class that treats electrons explicitly and another that does not⎯permit a better trade-off between computational resources and accuracy in results.
Analytic interatomic potentials are difficult to calculate accurately; those that work for bulk phases may not accurately predict observable properties, which often depend on what happens at the surface. Writing in Physical Review Letters, Albert Bartók, Mike Payne, and Gábor Csányi from the University of Cambridge, UK, and Risi Kondor from the California Institute of Technology, US, introduce a technique to model the potential energy surface of a set of atoms that allows them to work around having to simulate electrons explicitly; in effect, they autogenerate interatomic potentials from existing calculations of atomic forces and energies. Though independent of the specifics of the functional form, these potentials appear to be remarkably accurate in reproducing complex energy landscapes. The hope is that this admittedly generalized approach will work well in modeling specific metals and semiconductors. – Sami Mitra