Large Language Model-Guided Design of Anti-Swelling Hybrid Dual Network Membranes for Long-Duration Alkaline Zinc Iron Flow Batteries.

Membranes applied in alkaline flow batteries suffer swelling from spontaneous polymer chain motion, which degrades ion selectivity and mechanical strength. Here, we employ a large language model (LLM)-driven screening workflow to efficiently shortlist crosslinkers from a curated knowledge base, guiding the rational design of a hybrid dual network (H-DN) membrane. The workflow identifies N,N,N',N'-tetramethylethylenediamine as the optimal candidate to build a robust, chemically crosslinked polysulfone network interpenetrated within a widely used sulfonated poly(ether ether ketone) (SPEEK) matrix. Mechanistically, the in situ formed quaternary ammonium motifs strongly ion-pair with sulfonic acid groups, effectively suppressing chain motion without blocking ion transport channels. Consequently, the H-DN membrane exhibits a 68% reduction in swelling and an ultra-high wet-state hardness of 216 MPa (a sixfold increase over SPEEK), while retaining high ionic conductivity (10.7 mS cm-1). Enabled by this mechanically robust and highly selective architecture, the alkaline zinc-iron flow battery achieves an energy efficiency of 88.87% at a challenging areal capacity of 240 mAh cm-2 and operates stably for over 820 h without dendrite-induced failure, which surpassed the commercial Nafion212 and SPEEK benchmarks by threefold and sevenfold, respectively. This work establishes a paradigm for LLM-accelerated material discovery in addressing the stability-selectivity dilemma of ion-exchange membranes.