Events

Abstract: Using numerical simulation to determine the crystal structure of a compound, based on the sole knowledge of its chemical composition, is a major challenge in materials science. The task is far from trivial: it involves identifying the lowest-energy structural arrangement from among millions of possible structures. To illustrate this challenge, the arrangement of twenty atoms in a box – a repeating lattice of variable shape and volume – can a priori generate more than 1021 possible structures that lie on the potential energy surface (PES). If it took 1 hour of computing time to numerically determine the energy associated with each optimized structure, the computing time required would exceed the age of the universe… The problem is therefore: how to access the lowest energy well (global minimum on the PES) while monopolizing a minimum of computational resources?

This talk will discuss a self-learning method for exploring the PES of a crystalline compound, an evolutionary (genetic) algorithm combined with DFT calculations. I will briefly outline the conceptual basis of this CSP algorithm, which is based on the concepts of the Darwinian evolutionary theory. I will then illustrate its use by presenting some recent results from work carried out in my Applied Quantum Chemistry group: the exploration of the Lead-Nitrogen binary phase diagram under pressure (0-100 GPa), the emergence of novel compositions in the Li-C-N ternary system at 50 and 100 GPa, the ABH3 hydrides “perovskites” under pressure,
with/without perovskite’structure…, and the recovery of selected phases to ambient conditions; the bonding and electronic properties of the in silico compounds using basic and
state-of-the-art theoretical chemistry tools (Lewis, VSEPR, ELF, COHP, COBI, …), and so on…

More info here