Machine-learning insights into mechanical and ionic transport processes in thiophosphate solid electrolytes

Résumé fourni par la source

Understanding the coupling between mechanical integrity and ionic conductivity in solid electrolytes is central to the development of high-performance solid-state lithium batteries. In this work, a machine-learning interatomic potential (MLIP) was developed and iteratively refined to investigate the structural, mechanical, and transport properties of lithium thiophosphate (LPS) electrolytes. The model was trained on over 15,000 DFT-calculated structures using the aenet-PyTorch framework and subsequently employed in molecular dynamics (MD) simulations to probe pressure-dependent lithium-ion conductivity and mechanical response. Stress-strain analysis established a direct connection between local structural motifs and mechanical properties: Li2PS3 exhibits high stiffness (Young's modulus 52.9 GPa) due to dense P-S-P bridging of PS3(2-) chains and P2S7(2-) dimers, whereas Li3PS4 and Li7P3S11 are more compliant (16–17 GPa) because of a higher concentration of PS4(3-) tetrahedra, consistent with radial distribution function analysis. Ionic conductivity, computed via the Nernst-Einstein relation from 2 ns MD simulations across strain and temperature, shows that Li3PS4 exhibits pronounced strain-dependent transport, peaking at 0.55 S/cm at 600 K and 12% strain with a minimum activation energy of 0.11 eV. In contrast, Li7P3S11 displays strain-independent conductivity due to its robust three-dimensional diffusion network, in contrast to the quasi-two-dimensional network of Li3PS4. These results demonstrate the power of MLIPs for resolving the interplay between mechanical behavior and lithium-ion transport and highlight how polyanion chemistry and strain can be leveraged to tune solid-state electrolyte performance.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.