Abstract/Summary

Understanding the atomic-scale local properties of solid solutions is crucial for deciphering their structure–property relationships. In this work, we present a computational approach that combines solid-state nuclear magnetic resonance (NMR) spectroscopy with density functional theory (DFT) calculations to investigate local chemical environments in solid solutions. Previous canonical ensemble models, which only sample configurations at a fixed composition of the simulation cell, fail to capture local compositional fluctuations that can significantly influence the NMR spectra. To address this limitation, we employ a grand-canonical ensemble approach enabling a more comprehensive representation of the contributions of all possible local chemical environments to the NMR spectrum, using a La2(Zr1−xSnx)2O7 pyrochlore solid solution as a case study. To mitigate the high computational cost of such simulations, we also explore ensemble truncation strategies and the use of machine learning (ML) to aid predictions of NMR chemical shifts, achieving a significant reduction in computational cost while maintaining most of the predictive power. Our results show that combining the grand-canonical approach with machine learning and ensemble truncation offers an efficient framework for modelling and interpreting NMR spectra in disordered crystalline materials.