Prediction, determination and understanding of structures via global exploration of their energy landscapes

Christian Schoen
Max-Planck-Institute

Predicting the structure of not-yet synthesized chemical compounds requires information about the possible stable structures, their thermodynamic weight, and their kinetic stability. Since such (meta)stable compounds correspond to locally ergodic regions on the energy landscape of the system surrounded by sufficiently high energetic and entropic barriers, we need to undertake a global study of the landscape. We first determine as many local minima as possible using various global optimization procedures. Next, we employ a threshold algorithm to determine their stability by measuring the barrier structure surrounding them, and use swarms of stochastic quench runs to find the so-called characteristic regions of the landscape, which serve as further candidates for locally ergodic regions. Finally, we perform local optimizations of these structure candidates using ab initio energy functions (Hartree-Fock and DFT), in order to identify the thermodynamically stable compounds. This general approach can also be gainfully used when dealing with compounds that have already been successfully synthesized and for which even e.g. powder diffraction data are available, but where no structure solution has been obtained for lack of a good starting model. Here, we have introduced an extended cost function consisting of the weighted sum of the potential energy and a second term R_B which measures the deviation between the observed powder diffractogram and the one calculated for a trial atom configuration. Besides presenting the general methodology and the algorithms involved, I will discuss examples ranging from the prediction of structures of not-yet-synthesized compounds over the prediction of high-pressure phases of known crystalline compounds to the structure determination from powder diffraction data.