Data-Driven Insight into the Universal Structure–Property Relationship of Catalysts in Lithium–Sulfur Batteries

Despite
tremendous efforts in catalyzing the sulfur reduction reaction
(SRR) in high-capacity lithium–sulfur (Li–S) batteries,
understanding the universal and quantitative structure–property
relationships (UQSPRs) of SRR remains elusive. Such an unclarity results
from the limitations of first-principle calculations in analyzing
vast, high-dimensional, and heterogeneous data. Here, we present a
collaborative data-driven model for heterogeneous catalytic knowledge
fusion, detecting over 2,900 articles on SRR published between 2004
and 2024. By using sure independence screening and sparsifying operator,
we surprisingly identified a composite descriptor, D, dominated by the dispersion factor. In contrast to the classical
electronic state analysis framework, the dispersion factor directly
established UQSPRs between atom topological arrangement and catalyst-polysulfide
interaction intensity, accurately predicting the catalytic activity
of over 800 types of catalysts. Combined with a volcano plot linking
the overpotential to the interaction intensity, we determined the D value range of high catalytic activity, facilitating the
discovery of tens of novel SRR catalysts from 374,833 candidates,
many of which escaped previous human chemical intuition. As a representative,
CrB2 demonstrated superior catalytic activity under high
sulfur loadings of 12.0 mg cm–2 and low temperatures
of −25 °C. Pouch cells with CrB2 achieved a
gravimetric specific energy of 436 Wh kg–1 under
a high sulfur content of 76.1% and lean-electrolyte conditions of
2.8 μL mg–1. Our data-driven method enables
new opportunities to fundamentally identify UQSPRs using vast and
heterogeneous data, suggesting the promise of revisiting under-exploited
knowledge from the historical literature for novel catalyst discovery.