A Noise-Robust Data Assimilation Method for Crystal Structure Prediction Using Powder Diffraction Intensity

TL;DR

This paper introduces a noise-robust data assimilation method using a correlation-coefficient penalty function to improve crystal structure prediction from powder X-ray diffraction data, especially under noisy conditions.

Contribution

The paper presents a novel correlation-coefficient-type penalty function that enhances noise robustness in powder diffraction data assimilation for crystal structure prediction.

Findings

Effective in predicting structures of SiO₂ coesite and ε-Zn(OH)₂.

Improves success rate under noisy experimental conditions.

Demonstrates robustness compared to previous methods.

Abstract

Crystal structure prediction for a given chemical composition has long been a challenge in condensed-matter science. We have recently shown that experimental powder X-ray diffraction (XRD) data are helpful in a crystal structure search using simulated annealing, even when they are insufficient for structure determination by themselves (N. Tsujimoto et al., Phys. Rev. Materials 2, 053801 (2018)). In the method, the XRD data are assimilated into the simulation by adding a penalty function to the physical potential energy, where we used a crystallinity-type penalty function defined by the difference between experimental and simulated diffraction angles. To improve the success rate and noise robustness, we introduce a correlation-coefficient-type penalty function adaptable to XRD data with significant experimental noise. We apply the new penalty function to SiO$_2$ coesite and $\epsilon$-Zn(OH)$_2$ to determine its effectiveness in the data assimilation method.