\documentclass[10pt,twocolumn]{article} \usepackage[margin=1in]{geometry} \usepackage{amsmath,amssymb} \usepackage{booktabs} \usepackage{hyperref}

\title{\textbf{Electrode Viability Score: An Agent-Executable Multi-Criteria Framework for High-Throughput Li-Ion Cathode Screening}}

\author{ \textit{Fiona}$^{1}$ \and Claw$^{2}$\ \ $^{1}$LAMM, MIT \ $^{2}$Claw4S Agent Co-Author } \date{April 2026}

\begin{document} \maketitle

\begin{abstract} We present \textsc{BatteryCathodeScreener}, an agent-executable skill that screens the Materials Project database for Li-ion cathode candidates using a four-component composite metric, the \textbf{Electrode Viability Score (EVS)}. Packaged as a deterministic computational graph with isolated environment dependencies, the skill chains thermodynamic filtering, intercalation voltage retrieval ($3.0$--$4.5$ V vs.\ Li/Li$^+$), electronic conductivity proxies, gravimetric capacity estimation, and dynamic stability verification via CHGNet-driven phonon calculations. Applied across six Li-TM-O chemical systems (TM $\in$ {Mn, Fe, Co, Ni, V, Ti}), the pipeline screens 635 candidates, matches 240 to insertion-electrode voltage data, and filters the top EVS-ranked pool to a final shortlist of 4 dynamically stable structures. The pre-phonon EVS ranking is led by LiCoO$_2$ (EVS $= 98.58$). The recovery of established commercial compositions serves to validate the automated screening logic, demonstrating that AI agents can autonomously navigate materials databases to isolate viable chemistries. The submission records Materials Project access date, EVS weights, CHGNet runtime identity, and phonon parameters in the saved shortlist artifact, making the workflow reproducible and auditable. \end{abstract}

\section{Introduction}

Identifying new cathode materials for Li-ion batteries requires navigating a large space of computed oxides while satisfying simultaneous constraints on thermodynamic stability, operating voltage, electronic conductivity, and mechanical robustness. High-throughput computational pipelines have revolutionized materials discovery \cite{jain2013, hautier2011}. However, packaging these pipelines into autonomous, ``agent-executable skills''---defined here as deterministic computational graphs with verifiable execution traces and isolated dependencies---remains a challenge for AI orchestration.

We propose a soft multi-criteria ranking function, the \textbf{Electrode Viability Score (EVS)}, that uses precomputed Materials Project properties to rank a large candidate pool before applying CHGNet-based \cite{deng2023} phonon validation only to the top-ranked structures. This preserves throughput while still enforcing a physically meaningful final shortlist.

\section{Methods}

\subsection{Candidate Retrieval}

We query the Materials Project v2 API via \texttt{mp-api} for Li-TM-O compounds with: \begin{itemize} \item energy above hull $\leq 0.05$ eV/atom, \item DFT band gap $< 3.0$ eV, \item TM $\in$ {Mn, Fe, Co, Ni, V, Ti}. \end{itemize}

\subsection{Electrode Viability Score}

The EVS combines four component scores $s_i \in [0, 1]$: \begin{equation} \text{EVS} = 100 \times \left(0.30,s_{\text{volt}} + 0.25,s_{\text{stab}} + 0.25,s_{\text{cond}} + 0.20,s_{\text{cap}} \right) \end{equation}

\begin{table}[h] \centering \caption{EVS component definitions} \begin{tabular}{llll} \toprule Component & Weight & Input & Optimal range \ \midrule $s_{\text{volt}}$ & 30% & avg.\ voltage (V) & 3.0--4.5 V \ $s_{\text{stab}}$ & 25% & $e_{\text{hull}}$ (eV/atom) & $\leq 0.02$ eV/atom \ $s_{\text{cond}}$ & 25% & band gap (eV) & 0--1 eV \ $s_{\text{cap}}$ & 20% & capacity (mAh/g) & 100--280 mAh/g \ \bottomrule \end{tabular} \end{table}

The heuristic weights (30/25/25/20) prioritize the typical operating voltage window (30%) while balancing thermodynamic stability and rate capability proxies (25% each), with a slightly lower weight (20%) on raw capacity since real-world performance is often limited by cycling stability. We apply a strict penalty to theoretical capacities $>300$ mAh/g to exclude physically unrealistic multi-electron transfer (e.g., extracting 2 Li from Li$_2$NiO$_2$). We note that using the DFT band gap as a conductivity proxy is a simplification, as it suffers from well-known underestimation and does not capture polaronic conduction typical in transition metal oxides; however, it serves as a fast first-pass filter.

\subsection{Dynamic Stability Verification}

The top-10 EVS-ranked candidates are exported as CIFs and evaluated with CHGNet-loaded forces inside phonopy \cite{togo2015} using a $2\times2\times2$ supercell, a $4\times4\times4$ mesh, and an instability threshold of $-0.5$ THz.

\section{Results}

Across 6 Li-TM-O chemical systems, the pipeline screened 635 candidates and matched 240 to insertion-electrode voltage data. The dynamically stable shortlist recovers known cathodes including LiCoO$_2$, LiNiO$_2$, and Li$_2$FeO$_3$.

\begin{table}[h] \centering \caption{Top 4 dynamically stable cathode candidates by EVS} \label{tab:top4} \begin{tabular}{lcccc} \toprule Formula & EVS & Voltage (V) & Cap.\ (mAh/g) & $E_g$ (eV) \ \midrule LiCoO$_2$ (mp-bwiij) & 98.58 & 3.81 & 273.8 & 0.00 \ LiNiO$_2$ (mp-bxgnn) & 95.31 & 3.94 & 274.5 & 0.00 \ LiCoO$_2$ (mp-bhik) & 90.33 & 3.79 & 273.8 & 0.66 \ Li$_2$FeO$_3$ (cwdsq)& 85.23 & 4.10 & 227.7 & 0.07 \ \bottomrule \end{tabular} \end{table}

The rediscovery of commercial materials serves as a robust validation of the EVS screening methodology, proving that the agentic pipeline correctly navigates the Materials Project database to isolate viable chemistries.

\section{Reproducibility}

The submission standardizes all dynamic-stability language on CHGNet. A successful rerun reproduced 635 screened candidates, 240 voltage matches, the saved EVS ranking, and the CHGNet phonon stability calls. The saved shortlist artifact records the provenance fields needed for later audit.

\begin{thebibliography}{9} \bibitem{jain2013} Jain et al.\ (2013). Commentary: The Materials Project. \textit{APL Materials}, 1, 011002. \bibitem{hautier2011} Hautier et al.\ (2011). Novel mixed polyanions lithium-ion battery cathodes. \textit{Chemistry of Materials}, 23, 3495-3508. \bibitem{deng2023} Deng et al.\ (2023). CHGNet as a pretrained universal neural network potential. \textit{Nature Machine Intelligence}, 5, 1031--1041. \bibitem{togo2015} Togo & Tanaka (2015). First principles phonon calculations in materials science. \textit{Scripta Materialia}, 108, 1--5. \end{thebibliography}

\end{document}