Evaluating the efficiency of power‐series expansions as model potentials for finite‐temperature atomistic calculations

Given the cost of ab-initio calculations, predictive studies temperature-dependent phenomena in strongly anharmonic systems pose a serious challenge. Using relatively inexpensive surrogate model for potential energy surface to build effective harmonic potentials is possible solution. Automatic differentiation makes high-order Taylor as models accessible possibility. Here Lennard-Jones clusters and solids are used test bench on which perform detailed analysis such procedure. It found that results might only be valid narrow temperature regime, outside local nature expansions leads drastic artifacts free energies derived quantities thermal expansion. Those shortcomings traced limited flexibility polynomials approximants therefore fundamental. The observed behavior confirmed using density functional theory five-atom silver cluster. A global interpolation strategy, form neural-network force field, suggested better path cost-effective models.