Abstract
Understanding the origin of potential shifts in concentrated electrolytes is crucial for the design of stable and high-energy-density lithium-metal batteries. In our previous work, we identified a strong correlation between experimental electrode potential and Li(+)-anion interactions, suggesting the importance of intermolecular structure beyond conventional molecular properties. In this study, we revisit this issue from the perspective of causal discovery. We applied the Linear Non-Gaussian Acyclic Model (LiNGAM) to a data set of 75 electrolyte solutions containing LiFSI salt and various solvent molecules, using a comprehensive set of 132 descriptors including molecular properties (e.g., HOMO, LUMO, binding energy) and intermolecular descriptors derived from radial distribution functions (RDF) and their integrals (NDF) from molecular dynamics simulations. To mitigate overfitting due to high descriptor dimensionality, we employed a genetic algorithm to select reduced sets of descriptors. LiNGAM analysis revealed that descriptors associated with Li(+)-anion spatial distribution, especially long-range NDF descriptors, exhibit a direct causal effect with experimentally measured potential values. In contrast, no such causal relationships were identified for intrinsic molecular properties. These findings further support the recently proposed advanced theoretical framework, which incorporates a Debye-Hückel-based model and identifies the liquid-phase Madelung interaction as a dominant factor governing the potential shift in highly concentrated electrolytes. Moreover, we demonstrate that causal discovery methods can reveal fundamental physical mechanisms underlying electrochemical behavior.