The interfacial incompatibility between sodium anodes and sulfide/selenide solid-state electrolytes (SSEs) remains a central bottleneck for all-solid-state sodium batteries (ASSBs). Representative Na3PS4 (NPS), Na3SbS4 (NSS), Na3PSe4 (NPSe), and Na3SbSe4 (NSSe) electrolytes were coupled with Na, Na-Sb, Na-Sn, and Na-K anodes to establish baseline interface energetics, followed by W/Mo/Cr and B/O/F/Cl/Br/I co-doping to generate 304 interfaces. Candidate interfaces were prioritized using tree-based ensemble machine learning with compositional, structural, and low-cost density functional theory (DFT) descriptors. Controlled ablation was used to determine whether the single-point descriptors added information beyond composition and structure. Under grouped validation, modest ranking ability was retained by the formation energy model within the chemical space represented in the training data, whereas the migration model was used primarily for candidate prioritization. Cr/halogen-doped NSS/Na interfaces were prioritized through the integrated ranking for explicit first-principles assessment. The prioritized Cr/halogen-doped NSS/Na interfaces exhibited more negative relaxed formation energies than undoped NSS/Na, while subsequent Dimer searches along predefined directions yielded lower local Na migration barriers. Within the enumerated phase set, complete reduction of both undoped NSS and the CrX candidates was predicted by the Na-rich grand-potential analysis, showing that favorable contact-formation energetics do not imply equilibrium electrochemical stability. Frozen-ion single-point calculations required only approximately 1/248 of the computational cost of full interface relaxation. The cost of candidate prioritization is reduced by the workflow, while the range in which the models can be applied is clearly defined. Final assessments were based on relaxed DFT, Dimer, and grand-potential calculations.
原文链接
分享
引用
