From Static Spectra to Operando Infrared Dynamics: Physics Informed Flow Modeling and a Benchmark

Abstract

The Solid Electrolyte Interphase (SEI) is critical to the performance of lithium-ion batteries, yet its analysis via Operando Infrared (IR) spectroscopy remains experimentally complex and expensive, which limits its accessibility for standard research facilities. To overcome this bottleneck, we formulate a novel task, Operando IR Prediction, which aims to forecast the time-resolved evolution of spectral ``fingerprints'' from a single static spectrum. To facilitate this, we introduce OpIRSpec-7K, the first large-scale operando dataset comprising 7,118 high-quality samples across 10 distinct battery systems, alongside OpIRBench, a comprehensive evaluation benchmark with carefully designed protocols. Addressing the limitations of standard spectrum, video, and sequence models in capturing voltage-driven chemical dynamics and complex composition, we propose Aligned Bi-stream Chemical Constraint (ABCC), an end-to-end physics-aware framework. It reformulates MeanFlow and introduces a novel Chemical Flow to explicitly model reaction trajectories, employs a two-stream disentanglement mechanism for solvent-SEI separation, and enforces physics and spectrum constraints such as mass conservation and peak shifts. ABCC significantly outperforms state-of-the-art static, sequential, and generative baselines. ABCC even generalizes to unseen systems and enables interpretable downstream recovery of SEI formation pathways, supporting AI-driven electrochemical discovery.

Figures (5)

• Figure 1: We formulate a novel task: Operando IR Prediction. Starting from only a single, easily obtainable static spectrum with specific voltage profiles and electrolyte chemistries, the goal is to predict the evolution of IR spectra. This output absorbance spectra is from Cu-PVDF//LP30//Li metal cell, where LP30 is 1 M LiPF$_6$ in EC:DMC (1:1 v/v), shown at three representative voltages ($V_{\text{begin}}$, $V_{\text{middle}}$, $V_{\text{end}}$). Arrows and trajectories provide a visual guide to spectral dynamics with different voltages, $\uparrow$ for formation and $\downarrow$ for decomposition. The highlighted windows isolate the $\nu$C=O region, e.g., semi-carbonates (LEDC, LMC, etc.) (green rectangle) and Li$_2$CO$_3$ absorb (orange rectangles).
• Figure 2: OpIRSpec-7K shows A) high measurement fidelity with strong SNR, B) broad voltage coverage concentrated near operating potentials, and C) RMS spectral changes that combine subtle solvent-driven physical variation with sparse SEI chemical reaction-induced transitions, supporting high-quality operando data with chemically meaningful dynamics.
• Figure 3: Training pipeline of the proposed ABCC framework. The input IR spectrum $\mathbf{s}_{v_0}$ is converted to an initial waveform spectrum $I_{v_0}$ and fed into ABCC together with the voltage condition $v_m$. The electrolyte mixture, represented by 3D molecular structures $X\in\mathbb{R}^{k\times 3}$ and their ratios $R$, is encoded by a latent encoder to produce a mixture embedding used as an additional condition $\mathcal{C}$. ABCC models the conditional probability river of MeanFlow and is optimized by matching the target waveform spectrum $I_{v_m}$ with the objective $\mathcal{L}(\theta)$. Two-stream design separates reaction flow and solvent flow to promote solvent and SEI disentanglement. A peak constraint $\mathcal{L}_{\mathrm{peak}}$ and a mass conservation loss $\mathcal{L}_{\mathrm{mass}}$ incorporate physics-informed constraints. At inference, ABCC forecasts full operando IR spectral dynamics from a single static spectrum, conditioned on voltage profiles and electrolyte composition, generalizing to unseen systems.
• Figure 3: System Split quantitative evaluation. Our ABCC still dominates metrics, and shows strong generalization to unseen systems.
• Figure 4: Ablation under system split. MAE is in $10^{-2}$, MSE in $10^{-3}$, $R^2$ in $10^{-1}$. Chemical Flow, two-stream disentanglement, 3D mixture, and physics constraints drive generalization.