Quantitative Operando IR Spectroscopy: Multivariate Analysis, Reproducible Workflows, and AI – Assisted Perspectives

Arnaud Travert (LCS, Université de Caen Normandie, ENSICAEN, Caen, France)

Through selected examples, this lecture will present the state of the art and current challenges in the quantitative spectroscopic investigation of diffusion, adsorption, and surface reactions on complex heterogeneous surfaces such as working catalysts, with a particular emphasis on the user’s perspective.
In situ and operando IR spectroscopy are now well established, enabling the acquisition of high-quality spectra with excellent spectral and temporal resolution under realistic conditions. When spectroscopically distinct surface species are present and experimental parameters are well controlled, classical analysis of spectral evolution—monitoring band intensity or area as a function of time, pressure, or contact time—can provide direct access to internal diffusion coefficients [1], adsorption thermodynamics, or intrinsic surface kinetics [2].
However, in most catalytic systems, straightforward analysis is hindered by the coexistence of multiple adsorbed species with strongly correlated spectra and concentration profiles. In many cases, particularly for diffusion and adsorption, concentration profiles can be approximated by two-parameter models such as the Langmuir isotherm or fractional uptake curves. The inversion method (2D-IRIS) then allows determination of diffusion or adsorption constant distributions together with the corresponding spectral signatures [3,4].
When concentration profiles are more complex, a general framework based on exploratory multivariate analysis and multivariate curve resolution can be applied. This approach retrieves both spectra and concentration profiles while accommodating constraints ranging from soft constraints [5,6] to hard constraints based on explicit kinetic or thermodynamic models [7,8].
Despite these advances, practical implementation of multivariate and model-based methods remains demanding, requiring rigorous preprocessing, appropriate constraint selection, model validation, and full traceability. Computational frameworks such as SpectroChemPy [9], built around a structured multidimensional data model with integrated metadata management and chemometric tools in a reproducible Python environment, contribute to robust quantitative workflows. Nevertheless, the learning barrier remains significant for most users. As a perspective, developing a large language model (LLM)-based interface on top of such an environment could broaden their adoption by translating high-level scientific instructions into transparent and executable workflows, while preserving rigor and reproducibility [10].

[1] P. Peng, D. Stosic, A. Aitblal, A. Vimont, P. Bazin, X._M. Liu, et al., ACS Catal. 10, 6822 (2020),
[2] S. Kadam, H. Li, R. Wormsbecher, A. Travert, Chem. Eur. J. 24, 4 (2018),
[3] A. Ait Blal, D. Stosic, P. Bazin, A. Vimont, A. Travert, Phys.Chem.Chem.Phys. 25, 27170 (2023)
[4] L. Oliviero, A. Travert, E. Dominguez Garcia, J. Chen, F. Maugé, J. Catal. 403, 87 (2021)
[5] S. van Daele, et al. Applied Catal. B 284, 119699 (2021)
[6] S.A. Kadam, S. Sandoval, Z. Bastl, K. Simkovičová, L. Kvítek, J. Jašík, et al. ACS Catal. 13 13484 (2023)
[7] F. Vilmin, P. Bazin, F. Thibault-Starzyk, A. Travert, Analytica Chimica Acta 891, 79 (2015)
[8] R Aboulayt, E. Vottero, A. Vimont, P. Bazin, E. Bloch, S. Bourrelly,et al. ChemRxiv 10001501/vs (2025)
[9] A. Travert, C. Fernandez, SpectroChemPy, 2026, https://zenodo.org/doi/10.5281/zenodo.3823841
[10] Y. Li, H.N. Moussa, Z. Chen, S. Chen, B. Yu, M. Xue, et al. arXiv:2506.08140 (2025)