Résumé
The art of deriving classical potential consists in finding the correct balance between the complexity of the functional form whose implementation should be as simple as possible, the computational efficiency, and the reliabilty. In this talk, we will present a new methodology which uses ab initio data for parametrizing classical potentials suited for amorphous systems.We have considered a simple and widespread empirical Ansatz, i.e. Born-Mayer-Coulomb, with the goal to investigate large time/size scales. As a first condition, a classical potential should be able to describe the atomic structure of the system under investigation. In order to optimize the parameters of the classical potential we have then used a cost function that involves the partial pair correlation functions extracted from ab initio simulations. The system considered is liquid silica, a liquid which represents a prototype of strong glass formers exibiting a network structure. From a methodological point of view, we will show, through the different comparisons between ab initio and classical MD, that this structural fitting procedure is a very promising approach to obtain new effective potentials suited for describing not only amorphous material but also crystalline properties.We will present simulations carried out using this new potential and showing that CHIK reproduces with more accuracy a number of properties of molten and glass silica, than the well known BKS potential. Some tests performed on a-quartz also showed that the new potential can also be used to simulate crystalline systems with a good accuracy regarding properties such as asymmetric coordinates and elastic constants. Finally, we will discuss molecular dynamics simulations carried out to represent the structures of amorphous dehydroxylated silica surface in liquid and glass states, and show that the new potential gives a also fair description of the surface.