Avaliação de diferentes potenciais interatômicos no cálculo do tensor de elasticidade do tungstato de zircônio

Abstract

Zirconium tungstate (ZrW2O8) is a material that exhibits negative thermal expansion (NTE), over a wide temperature range (0.3 at 1050 K). Although thoroughly studied, controversies still remain concerning the microscopic mechanisms responsible for this behavior. The cubic phase of this compound, denominated a-ZrW2O8, was already the subject of study through computer simulations using interatomic potentials (IP) and lattice dynamics in quasiharmonic approximation (LDQH). In two different IPs proposed in the literature succeeded in reproducing the a-ZrW2O8 NTE, but not the dependence with temperature of the elasticity tensor. Starting from this observation, this work intends to evaluate of existent IPs and same proposed new IPs in computer simulations aiming the calculation of the tensor of elasticity for a-ZrW2O8 between 0 and 300 K. Experimental data (such as atomic positions, lattice parameters and the tensor of elasticity of a-ZrW2O8 at 0 and 300 K) and, in another series of calculations, the ab initio energy hypersurface in the athermic limit, were used to obtain the parameters of the IPs. Different strategies were used in the search for the parameters of IP, including line minimization, Newton-Raphson/BFGS and genetic algorithm (GA). At the end of an exhaustive search we were led to conclude that it is not possible to describe the structure and elastic properties of a-ZrW2O8 as a function of temperature with simple IPs and that this incapacity is not related the any limitation of LDQH or of the parameters of the IPs, but instead to the analytical form of the tested IPs. This suggests that same results obtained with IPs already available in the literature may be unreliable. As alternatives for the solution of this problem, it can be considered the use of a neural network for the representation of the ab initio energy hypersurface, the use of more sophisticated IPs than take into account the atomic neighborhood (bond order potentials) and even (with a computational cost much higher) ab initio calculations at T > 0.