Stability and Phase Transition of Cobalt Oxide Phases by Machine Learning Global Potential Energy Surface

Fan-Chen Kong,Ye-Fei Li,Cheng Shang,Zhi-Pan Liu

DOI: https://doi.org/10.1021/acs.jpcc.9b02842

2019-01-01

The Journal of Physical Chemistry C

Abstract:Cobalt oxides (CoOx), despite its wide applications, are notoriously complex on structure, in particular under experimental reduction/oxidation conditions. Here, using the newly developed machine learning method, stochastic surface walking global optimization in combination with global neural network potential, we are able to, for the first time, explore the global potential energy surface (PES) of CoOx at different Co/O ratios. Rich information on the thermodynamics and kinetics of CoOx is thus gleaned from more than 10(7) PES data, which helps to resolve the long-standing puzzles on CoOx chemistry. We show that (i) only CoO and Co3O4 are thermodynamically stable compositions in CoOx, whereas Co3O4 is the most stable phase. The trivalent Co2O3, although having a well-defined global minimum, tends to decompose to Co3O4 and O-2 at finite temperatures. (ii) The solid phase transition between wurtzite CoO (h-CoO) and rock salt CoO (c-CoO) follows the reconstructive phase transition mechanism with a high barrier. Because a high temperature is required for transition, the strong preference of structural defects inside c-CoO instead of h-CoO contributes to the one-way solid-phase transition from h-CoO to c-CoO. (iii) It is c-CoO that can achieve a coherent interface with Co3O4 in forming a biphasic junction, which implies the reversibility of Co3O4 and c-CoO transition under reduction/oxidation conditions. Our results demonstrate the power of global neural network potential in material discovery for fast exploration of polymorphism and transition kinetics and lay the structural foundation for understanding CoOx applications.