The embedded atom method ansatz: validation and violation

Abstract

The addition of the embedding energy term to pair interaction contribution has made the embedded atom method (EAM) potentials a simple and vastly superior alternative to popular classical pair potentials. EAM relies on the ansatz that the embedding energy is a function of a linear superposition of spherically averaged atomic electron densities. This ansatz is taken to be self-evident and inviolate. Using density functional theory (DFT) calculations of a model face-centered cubic (fcc) Cu system, we systematically investigate the validity of this foundational ansatz of EAM. We conclude that it (1) agrees well with DFT calculations along a path with changing coordination and symmetry, (2) captures the exponential decrease of the background electron density with respect to distance, (3) demonstrates transferability as seen by agreement of electron densities for other non-fcc structures with first nearest neighbor (NN) coordination ranging from 4 to 12 and (4) fails to explain the behavior of background electron density with respect to second NN distance and arrangements. This failure may be remedied by including a fraction of the second NN atomic electron density in the background electron density, including angular contributions to the density, or including electron density rearrangement. These insights likely make EAM approaches more broadly applicable, more predictive and perhaps unique, and in the process broadly impact atomistic modeling. A new EAM potential is presented that for the first time reproduces electron densities from DFT calculations as well as experimental properties of Cu in the potential fitting.