Key points are not available for this paper at this time.
In this study, we evaluate the functionals of different embedded-atom methods (EAM) by fitting their free parameters to ab-initio results for copper. Our emphasis lies on testing the transferability of the potentials between systems which vary in their spatial dimension and geometry. The model structures encompass zero-dimensional clusters, one-dimensional chains, two-dimensional tilings, and three-dimensional bulk systems. To avoid having to mimic charge transfer, which is outside the scope of conventional EAM potentials, we focus on structures, in which all atoms are symmetrically equivalent. We find that the simple, four-parameter Gupta EAM potential is overall satisfactory. Adding complexity to it decreases the errors on our set of structures only by marginal amounts, unless EAM is modified to depend also on density gradients, higher-order derivatives, or related terms. All tested conventional EAM functions reveal similar problems: the binding energy of closed-packed systems is overestimated in comparison to open or planar geometries, and structures with small coordination tend to be too rigid. These deficiencies can be fixed in terms of a systematically modified embedded-atom method (SMEAM), which reproduces DFT results on bond lengths, binding energies, and stiffnesses or bulk moduli by typically O(1%), O(5%), and O(15%) accuracy, respectively. SMEAM also predicts the radial distribution function of liquid copper quite accurately. Yet, it does not overcome the difficulty to reproduce the elastic tensor elements of a hypothetical diamond lattice.
Jalkanen et al. (Fri,) studied this question.