Accelerating Simulated Annealing of Glassy Materials with Data Assimilation

The ultra-long relaxation time of glass transition makes it difficult to construct atomic models of amorphous materials by conventional methods. We propose a novel method for building such atomic models using data assimilation method by simulated annealing with an accurately computed interatomic potential augmented by penalty from experimental data. The advantage of this method is that not only can it reproduce experimental data as the structure refinement methods like reverse Monte Carlo but also obtain the reasonable structure in terms of interatomic potential energy. In addition, thanks to the interatomic potential, we do not need high $Q$ range diffraction data, which is necessary to take into account the short-range order. Persistent homology analysis shows that the amorphous ice obtained by the new method is indeed more ordered at intermediate range.