Apr 8, 2025
5:00pm - 7:00pm
Summit, Level 2, Flex Hall C
Seungwoo Hwang1,Jiho Lee1,Youngho Kang2,Seungwu Han1,Sungwoo Kang3
Seoul National University1,Incheon National University2,Korea Institute of Science and Technology3
Seungwoo Hwang1,Jiho Lee1,Youngho Kang2,Seungwu Han1,Sungwoo Kang3
Seoul National University1,Incheon National University2,Korea Institute of Science and Technology3
The discovery of novel oxide solid electrolytes with high ionic conductivity is crucial for advancing all-solid-state batteries. In previous work, high-throughput theoretical screening of the experimentally existing materials identified promising candidates by harnessing the corner-sharing characteristics of oxide electrolytes.[1] However, large portions of the chemical space remain unexplored, leaving the possibility of undiscovered, synthesizable materials. Crystal structure prediction (CSP), such as evolutionary algorithms, can generate ground-state and metastable structures without predefined prototypes by exploring the potential energy surface (PES). Machine learning potentials (MLPs) further accelerate the evaluation of the PES and enhance CSP throughput. In our previous work, we demonstrated the effectiveness of this approach in exploring materials with up to 50 atoms per unit cell.[2,3] However, most oxide solid electrolyte structures are even more complex, often containing over 100 atoms per unit cell and comprising more than four elements, leading to a high complexity that cannot be addressed via conventional CSP methods.
Due to the large ion-diffusion pathways, many oxide solid electrolytes exhibit a corner-sharing topology in multi-element systems. By leveraging this structural property, we aim to simplify the CSP process by narrowing the configurational space to oxide solid electrolytes with corner-sharing topologies. In this work, we present a CSP method for crystal structures with corner-sharing topology, named TOPIC (TOpology-based crystal structure Prediction of Inorganic solid electrolytes with Corner-sharing frameworks), to explore uncharted quaternary chemical spaces. To address the complexity of quaternary PES of oxide electrolytes, we first predict the host crystal structure without incorporating Li atoms. First, we train MLPs using a training set that consists of melt-quench trajectories [4]. Then, we develop a framework prediction method using MLPs that enforces corner-sharing topology without Li atoms. This approach significantly reduces the positional degrees of freedom compared to conventional methods. Within the predicted host-structure framework, Li atoms are placed using Voronoi tessellation and Monte Carlo simulation. Finally, we evaluate the diffusivity of low-energy candidate materials using MLPs.
For validation, we test whether our method can accurately predict structures of known oxide solid electrolytes (e.g. NASICON and LiTa
2PO
8). Our method successfully predicts several solid electrolytes with corner-sharing topology, each containing approximately 100 atoms per conventional unit cell, at a reasonable computational cost. Furthermore, we explore novel quaternary oxide solid electrolytes with corner-sharing topology in unexplored compositions, evaluating their stability, bandgap, and ionic conductivity to assess their potential for solid electrolyte applications.
[1] K. Jun,
et. al., Nat. Mater. 21, 924, (2022)
[2] S. Kang,
et. al., npj Comput. Mater. 8, 108 (2022)
[3] S. Hwang,
et. al., J. Am. Chem. Soc. 145, 19378 (2023)
[4] C. Hong,
et. al., Phys. Rev. B 102, 224104 (2020)