Beyond heuristics: structure-aware screening reveals the limits of composition-based cathode design
- 1Department of Chemical and Biomolecular Engineering
- 2University of Pennsylvania
Abstract
Composition-level heuristics based on transition-metal identity are widely used to anticipate lithium-ion cathode durability, but their physical basis is rarely tested against structure-resolved data. Screening 2,774 Li insertion electrodes from the Materials Project, we compare a metal-identity heuristic with a transparent Degradation Screening Index (DSI) built from three structure-resolved descriptors: delithiation volume strain, charged-phase energy above hull and band gap. Against 11 experimentally characterised cathodes the DSI ranks durability better (Spearman ρ = 0.76, p = 0.007) than the metal-identity heuristic (ρ = 0.55, p = 0.08) or a single volume-strain descriptor (ρ = 0.55, p = 0.08), and at database scale the heuristic correlates only weakly with the DSI (ρ = 0.37, n = 787). Resolving LiCuO2 into its polymorph-specific electrode entries shows delithiation strain from 4.7 to 47% and DSI values spanning three of four durability categories at fixed composition. For energy-density prediction, neither compositional descriptors (R2 = 0.39, 132 Magpie features) nor structural descriptors alone (R2 = 0.36; 0.27 under grouped cross-validation) are predictive, and the accuracy of an 8-feature model (R2 = 0.95) is fully accounted for by voltage, capacity and density, the arithmetic constituents of the target; its role is therefore confined to coverage and uncertainty estimation. Multi-objective Pareto analysis over energy density, DSI and supply-chain criticality shows that no single material is universally optimal. All durability estimates are reported as relative DSI categories rather than cycle counts, and structure-resolved descriptors, not composition alone, should anchor durability screening.
Related
- Comparative Study of Y2O3, ZnO, and CeO2 Additions to Magnesium Aluminosilicate Glasses: Composition–Structure–Property Relationships
- Large language model-based screening of substances and their composition from safety data sheets for high-resolution chemical exposure assessment
- Structure-based virtual screening identifies naturally derived histidine kinase VbrK inhibitors as potential countermeasures against β-lactam antimicrobial resistance
- Integrated machine learning, pharmacophore screening, and structure-based computational approaches for the identification of novel JAK3 inhibitors
- From Heuristics to Machine Learning: The Performance Ceiling for Single-Ion Magnets and Its Electronic Origin
BibTeX
@misc{kavathekar2026beyond,
title = {{Beyond heuristics: structure-aware screening reveals the limits of composition-based cathode design}},
author = {Soham Kavathekar},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010709/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010709/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Beyond heuristics: structure-aware screening reveals the limits of composition-based cathode design
AU - Soham Kavathekar
PY - 2026
DA - 2026/10/20
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010709/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010709/v1
AB - Composition-level heuristics based on transition-metal identity are widely used to anticipate lithium-ion cathode durability, but their physical basis is rarely tested against structure-resolved data. Screening 2,774 Li insertion electrodes from the Materials Project, we compare a metal-identity heuristic with a transparent Degradation Screening Index (DSI) built from three structure-resolved descriptors: delithiation volume strain, charged-phase energy above hull and band gap. Against 11 experimentally characterised cathodes the DSI ranks durability better (Spearman ρ = 0.76, p = 0.007) than the metal-identity heuristic (ρ = 0.55, p = 0.08) or a single volume-strain descriptor (ρ = 0.55, p = 0.08), and at database scale the heuristic correlates only weakly with the DSI (ρ = 0.37, n = 787). Resolving LiCuO2 into its polymorph-specific electrode entries shows delithiation strain from 4.7 to 47% and DSI values spanning three of four durability categories at fixed composition. For energy-density prediction, neither compositional descriptors (R2 = 0.39, 132 Magpie features) nor structural descriptors alone (R2 = 0.36; 0.27 under grouped cross-validation) are predictive, and the accuracy of an 8-feature model (R2 = 0.95) is fully accounted for by voltage, capacity and density, the arithmetic constituents of the target; its role is therefore confined to coverage and uncertainty estimation. Multi-objective Pareto analysis over energy density, DSI and supply-chain criticality shows that no single material is universally optimal. All durability estimates are reported as relative DSI categories rather than cycle counts, and structure-resolved descriptors, not composition alone, should anchor durability screening.
ER -