Symposium for Data-Driven Approaches in X-ray Absorption Spectroscopy (DataXAS)

5 Jan 2026, 08:00 → 6 Jan 2026, 17:30 Europe/Zurich

Siemens Auditorium (ETH Zurich)

Adam Hugh Clark (PSI - Paul Scherrer Institut), Aram Bugaev (Paul Scherrer Institute), Maarten Nachtegaal (PSI - Paul Scherrer Institut), Tomas Aidukas (Paul scherrer institute)

The Symposium on Data-Driven Approaches in X-ray Absorption Spectroscopy (DataXAS) will bring together leading experts in spectroscopy and data science to showcase the latest advances in machine learning and computational techniques applied to XAS. The two-day in-person event at ETH Zurich will feature four main sessions covering:

• Data-Driven Decomposition and Spectral Extraction

• Machine Learning for XAS Prediction and Analysis

• Advanced Approaches to EXAFS Analysis

• Bridging Simulations and Experiments with Machine Learning

Participants will hear invited talks from internationally recognized speakers (see the list in Invited Speakers) and engage in discussions on emerging challenges and opportunities at the interface of spectroscopy and data science. The full symposium programme is available in Symposium Schedule.

---

Abstract submission

In addition to invited presentations, we have open slots in each session for 30-minute contributed talks.

A dedicated poster session with apero will take place on the evening of Day 1, providing opportunities for early-career researchers and participants to present their work and exchange ideas with peers and senior scientists.

Researchers wishing to be considered should submit an abstract as described in Abstract Submission. 

Abstract submission for talks is now closed.

Flash presentations and Posters will remain open until 1 December 2025.

 

---

Attending this event

The symposium will be held at the ETH Zurich (see Venue for more details). The registration fee includes admission to all sessions, coffee breaks, the poster reception, and the symposium dinner.

Registration will open on 22 September 2025

Registration details:

• Student registration: CHF 150 (until 01 December 2025)
  • Late registration (until 10 December 2025): 200 CHF
• Regular registration (until 01 December 2025): CHF 250
  • Late registration (until 10 December 2025): CHF 300

Places are limited and registration will close once capacity is reached.

---

 

Our sponsors

Organizing Institution

---

 

DataXAS_abstract_template.docx

dataxas_flyer.pdf

Monday 5 January

08:00 → 09:00 Registration and Coffee

09:00 → 10:00 Welcome and Plenary Talk

09:15 Multi-modal foundation models for spectroscopy and lab operations

Speaker: Dr Teodoro Laino (IBM)

10:00 → 12:30 Data-Driven Decomposition and Spectral Extraction

10:00 From Spectra to Species: Methods and Applications of XAS Spectral Decomposition

Speaker: Dr Andrea Martini (Fritz Haber Institute of the Max Planck Society)

10:40 Coffee Break

11:20 MCR-ALS, the Swiss Army Knife for Quick-EXAFS data analysis

Multivariate Curve Resolution with Alternating Least Squares (MCR-ALS) is today extensively used to solve the mixture analysis problem from Quick-EXAFS datasets recorded during the monitoring of chemical reactions, phase transitions and others [1-2]. Its objective is to decompose the XAS dataset, D, according to the bilinear relationship characteristic of Beer-Lambert spectroscopies:
D = C . ST + E (1)
where the columns of matrix C contain the relative proportions of the n pure species in the mixture, the rows of the matrix ST are the spectra of those n pure species (ST meaning the transpose of matrix S) and E is the matrix expressing the error or variance not explained by the C.ST product. The use of chemically relevant constraints (non-negativity of concentrations and/or spectra, closure relation for the concentrations …) during minimization reduces the ambiguities inherent in MCR solutions, making it possible to obtain spectra that can be satisfactorily identified by comparison with known references, by fitting or by ab initio spectra simulations.
This lecture will illustrate the power of the MCR-ALS method for isolating information of interest for the users from time-resolved datasets. The limitations of its “universal” application, particularly in the presence of co-evolving species or when spectra are too similar, will be discussed, along with strategies for overcoming them. These include data augmentation methods, such as the use of additional XAS datasets obtained by modifying experimental conditions or multimodal datasets [2-4].
The MCR-ALS use to spatially and temporally resolved datasets acquired by recently implemented full-field hyperspectral XAS imaging on the ROCK beamline [5] will conclude the presentation, highlighting the power of the chemometric method to extract high quality noise-filtered pure spectra from noisy hyperspectral imaging datasets expressed at the pixel level.
References
[1] Cassinelli, W. H., Martins, L., Passos, A. R., Pulcinelli, S. H., Santilli, C. V., Rochet, A. & Briois V. (2014). Multivariate Curve Resolution Analysis Applied to Time-Resolved Synchrotron X-ray Absorption Spectroscopy Monitoring of the Activation of Copper Alumina Catalyst. Catalysis Today 229, 114.
[2] Passos, A.R., La Fontaine, C., Rochet, A. & Briois, V. (2023). Case Studies: Time-Resolved X-Ray Absorption Spectroscopy (XAS), Springer Handbook of Advanced Catalyst Characterization (Springer), pp. 625.
[3] Rabeah, J., Briois, V., Adomeit, S., La Fontaine, C., Bentrup, U. & Brückner, A. (2020). Multivariate analysis of coupled operando EPR/XANES/ EXAFS/UV–vis/ATR–IR spectroscopy: A new dimension for mechanistic studies of catalytic gas‐liquid phase reactions. Chem. – A Eur. J. chem.202000436
[4] Plais, L., Lancelot, C., Lamonier, C., Payen, E. & Briois, V. (2021). First in-situ Temperature Quantification of CoMoS species upon Gas Sulfidation enabled by New Insight on Cobalt Sulfide Formation, Catalysis Today 377, 114.
[5] Briois, V., Itié, J.P., Polian, A., King, A., Traore, A.S., Marceau, E., Ersen, O., La Fontaine, C., Barthe, L., Beauvois, A., Roudenko, O., Belin, S. (2024). Hyperspectral Full Field Quick-EXAFS Imaging at the ROCK beamline for monitoring micrometer sized heterogeneity of functional materials under process conditions. J Synchrotron Radiation 31, 1084.

Speaker: Dr Valerie Briois (SOLEIL Synchrotron)

12:00 Spectroscopic data demixing using regularized MCR

Speaker: Dr Ilnura Usmanova (Swiss Data Science Center)

12:30 → 13:30 Lunch Break

13:30 → 15:50 Machine Learning for XAS Prediction and Analysis

13:30 Machine Learning for X-ray Spectroscopy: Hero or Zero?

X-ray spectroscopy (XS) is undergoing such a transformation, powered by next-generation, high-brilliance light sources. As experimental capability expands, a new challenge emerges: How can we efficiently and accurately analyse the resulting data so that the rich quantitative information encoded in each spectrum is fully exploited? Extracting such insight increasingly requires sophisticated theoretical modelling to connect spectral signatures to underlying structure and dynamics, yet these calculations remain computationally demanding and technically complex.
In this talk, I will present our recent progress in addressing this challenge using supervised machine-learning and deep-learning approaches to predict X-ray absorption near-edge structure (XANES) spectra directly from local geometric information around the absorbing atom. We demonstrate that these models achieve sub-eV accuracy in peak positions and predict peak intensities with errors more than an order of magnitude smaller than the intrinsic spectral variations they are trained to capture. I will discuss the model architecture, its physical underpinnings, and its application across multiple absorption edges, highlighting how data-driven approaches can accelerate spectral analysis and broaden the impact of modern X-ray spectroscopy.

Speaker: Prof. Thomas Penfold (University of Newcastle)

14:10 Fine-tuning ab-initio XANES spectra calculations using the Bayesian Optimization algorithm.

First-principle calculations of near-edge absorption spectra are widely used for experimental data analysis. Finding simulation parameters for the best match with experimental data may often be a challenge due to insufficient computational resources, ambiguous interpretation of spectral similarity criteria, or lack of expertise. When considering the simulation codes, such as FEFF or FDMNES as a black-box software, in which the output is defined as a similarity measure between the theoretical spectrum and experimental one, a Bayesian optimization technique can be applied to optimize the input configuration yielding the output most resembling the target spectra. We test this method on several experimental K-edge transmission spectra of Ni, Fe, and Pd metals, trying to find the best fits between the calculated and experimental XANES spectra using different similarity metrics such as L2 Normalized distance, cosine similarity, and correlation coefficients. The fitting process successfully converges to qualitatively and quantitatively match the resulting spectrum, with the parameter sets being reproducible within a certain margin. The test results show that this algorithm consumes, on average, three times fewer resources than the random search algorithm.

Speaker: Andrey Sapronov (IKFT KIT)

14:40 Coffee Break

15:20 Predicting noise-free XAS spectra from noisy measurements

Speaker: Dr Tomas Aidukas (PSI - Paul Scherrer Institut)

15:50 → 17:00 Flash Presentations

15:50 INSIGHTS INTO MOF SYNTHESIS WITH COMBINED IN SITU APPROACH: RAMAN SCATTERING, X-RAY ABSORPTION, X-RAY DIFFRACTION

Metal–organic frameworks with a Zr-oxo cluster [(Zr6O4(OH)4)]12− are exceptionally stable and offer vast potential for a wide range of applications. Synthesis parameters strongly affect the quality, stability, morphology, etc., of the MOFs calling for elucidating of the various reaction steps.

Figure 1 top: Reaction kinetics of Zr-oxo-cluster formation in DMF based on LCF of the in-situ XAS data: effect of temperature, modulator, and water concentration [1]; middle: in-situ synchrotron PXRD data of water-based synthesis of Zr-fumarate; bottom: in-situ Raman spectra of Zr-fumarate synthesis at HT and growth of 1666 cm –1 peak [3].
In this study, we present a combined approach of three in situ techniques: X-ray diffraction, X-ray absorption, and Raman spectroscopy, that allows for integrated monitoring of MOF synthesis in different solvents and at variable temperatures. In the early pre-crystalline synthesis phase, the local (∼2–5 Å) environment around Zr4+ ions and the Zr-oxo cluster formation is addressed by element-specific X-ray absorption at Zr k-edge. Raman scattering provides evidence of coordination between inorganic and organic blocks in the synthesis reaction. Owing to week scattering from water, Raman scattering is much more suitable for the in-situ studies of the water-based synthesis than infrared spectroscopy. Crystallization onset of the MOFs was evidenced by the appearance of the scattering peaks in the in situ patterns.

The kinetics of the various steps of the synthesis reaction can be studied using linear combination fits (LCS, X-ray absorption data, Figure 1a [1]), growth of PXRD reflections (PXRD data [2], Figure 1b), and the intensity growth of the cluster–linker coordination vibrational mode (Raman scattering, Figure 1c [3]). The dependence of the reaction kinetics on water, modulator concentrations, solvent, and metal precursor were studied as well. The results of the in-situ experiments were correlated with post-synthetic characterization of the resultant MOF products.

Speaker: Olena Zavorotynska (University of Stavanger)

15:53 Quantitative EXAFS Thermometry with Universal Machine Learning Interatomic Potentials

Extended X-ray Absorption Fine Structure (EXAFS) spectroscopy is a powerful technique for probing local atomic structure and thermal disorder. Its high sensitivity to atomic vibrations results in pronounced temperature-dependent damping of the EXAFS signal, making the technique a promising tool for non-contact thermometry, with applications ranging from nanoparticles [1] to materials under extreme conditions [2]. Absolute temperature determination from measured EXAFS spectra, however, typically requires either laborious calibration [1] or the use of analytical models, such as the Debye model, which is limited to monoatomic solids [3]. Computational approaches based on Molecular Dynamics (MD) reference standards [2] are, to date, limited in precision owing to the prohibitive cost of ab initio methods for modelling thermal disorder.

Here, we present a quantitative and computationally efficient framework to determine sample temperature based on MD simulations with Universal Machine Learning Interatomic Potentials (uMLIPs) [4] in combination with an ab initio multiple-scattering EXAFS formalism [5]. Pre-trained uMLIPs are fine-tuned on compound-specific Density Functional Theory (DFT) data, enabling accurate calculation of forces and temperature-dependent atomic vibrations at minimal computational cost [6]. Configuration-averaged EXAFS spectra are then computed at various points across a wide temperature range based on the results of MD simulations. Finally, experimental EXAFS spectra are fitted to these computational references to determine the corresponding sample temperature on an absolute temperature scale.

We validate this approach on NiO and ZnO, achieving accurate temperature determination from room temperature up to 1000 K. This framework provides a robust and computationally efficient route for quantitative EXAFS thermometry, applicable to a broad range of materials.

References
[1] Van de Broek B. et al. (2011). Temperature determination of resonantly excited plasmonic branched gold nanoparticles by X-ray absorption spectroscopy. Small, 7, 2498-2506.
[2] Sio H. et al. (2023). Extended X-ray absorption fine structure of dynamically-compressed copper up to 1 terapascal. Nature Communications, 14, 7046.
[3] Kuzmin A. et al. (2024). The Use of the correlated Debye model for extended x-ray absorption fine structure-based thermometry in body-centered cubic and face-centered cubic metals. Physica Status Solidi A, 222, 2400623.
[4] Deng B. et al. (2023). CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling. Nature Machine Intelligence, 5, 1031-1041.
[5] Rehr J.J. & Albers R.C. (2000). Theoretical Approaches to X-Ray Absorption Fine Structure. Rev. Mod. Phys. 72, 621-654.
[6] Žguns P. et al. (2025). Benchmarking CHGNet universal machine learning interatomic potential against DFT and EXAFS: The case of layered WS2 and MoS2. Journal of Chemical Theory and Computation, 21, 8142-8150.

Speaker: Dr Pjotrs Žguns (Institute of Solid State Physics, University of Latvia, Riga, Latvia)

15:56 A Data-Driven Workflow for Preprocessing, Anomaly Detection, and Simulation-Guided EXAFS Analysis of Cu in Steel

We present an integrated workflow for X-ray Absorption Spectroscopy (XAS) data processing, designed for extensible, automated analysis using Python. First, we developed a two-pass rolling median filter (with windows of 21 and 7 points and median absolute deviation thresholds of 6 and 5, respectively) to detect glitches from the incident intensity signal (I₀). The identified glitches are then removed from both I₀ and the fluorescence signal (If), yielding clean spectra for subsequent analysis. Second, we applied Z-score normalization to the absorption spectra, followed by interpolation onto a common energy grid, and then performed Principal Component Analysis (PCA) for unsupervised exploration of spectral variation across measurements. PCA reduces dimensionality, highlights systematic changes, clusters related scans, and flags outliers for inspection. Next, we applied this workflow to Cu K-edge (8.979 keV) XAS in fluorescence mode for 22MnB5 steel sheets with 0.1–0.3 wt.% Cu. Larch-based extended X-ray absorption fine structure (EXAFS) simulations were used to assess the impact of k-range on data quality. Comparing simulated EXAFS data with kmax ≈ 11 Å⁻¹ and extended k-ranges shows that higher-k measurements improve amplitude and resolution, enabling better discrimination between structural models. This modular pipeline provides a reproducible framework for XAS analysis, expandable with additional preprocessing and machine learning techniques.

Speaker: Duc-Chau Nguyen (ALBA Synchrotron - CELLS)

15:59 Similarity Metric for Automated FDMNES Parameter Optimization for XANES Simulations

Abstract
X-ray Absorption Near-Edge Spectroscopy (XANES) is a key technique for studying the local structure of an atomic specie in matter. The technique provides information about the oxidation state, coordination number, and electronic configuration. Traditionally, XANES spectra are compared with those coming from model compounds, a process known as fingerprint analysis. However, for a deeper understanding of the electronic structure and when the experimental reference spectra are not available, ab initio simulations are mandatory. A well-known real-space density functional theory code for simulating XAS spectra is FDMNES1. To get the best simulation, the typical workflow requires manually running many simulations and comparing them with experimental spectra as a function of the input atomistic model and the code parameters, until a good match is found. To automatize such workflow, a comparison metric between the experimental and simulated spectra is required. Traditional numerical comparison methods, like normalized mean square error (NMSE), often do not capture important spectral differences needed for precise matching. To address this problem, we use a feature-based similarity approach inspired by machine learning. This method identifies key points in the spectrum such as local peaks, and inflection points. These features are combined into a single similarity score that is more sensitive to meaningful spectral details than traditional error metrics by comparing differences in energy and amplitude. By using such metric, an automated pipeline has been built in order to optimize the FDMNES input parameters. We will present this method applied to a series of gold-based compounds with diverse chemical characteristics, enabling us to unravel and interpret the structural details underlying the experimentally obtained spectra.
References
[1] Bunau, O. & Joly, Y. (2009). Self-consistent aspects of x-ray absorption calculations. J. Phys.: Condens. Matter, 21, 345501 (20).

Speakers: Dr Hester Blommaert (Institut Néel, CNRS, Grenoble, France), Dr Mohamed Redhouane BOUDJEHEM (Institut Néel, CNRS, Grenoble, France)

16:02 Intelligent Analysis Pipeline for X-ray Absorption Spectroscopy: Near Real-Time Processing and Adaptive Experimentation

Automated analysis during synchrotron experiments has become crucial for modern research facilities. Rapid feedback from precise analytical results allows scientists to adjust measurement conditions, choose appropriate samples, and deploy adaptive strategies effectively, leading to optimized experiments and faster scientific breakthroughs.

The near real-time analysis pipeline presented here forms part of the ROCK-IT [1] project, which develops infrastructure for automated and remotely accessible in-situ and operando measurements. A key aspect of this project involves incorporating intelligent approaches to deliver data-driven experimental recommendations and notifications when experimental conditions are not met. The framework is designed to accommodate various potential functionalities such as energy drift forecasting in the absence of reference spectra, spectral noise reduction and signal reconstruction, anomaly detection, and intelligent determination of measurement completion. Access to analytical outcomes with minimum delay is essential to generate timely, evidence-based recommendations that improve both accuracy and operational speed throughout experiments, with some Machine Learning (ML) approaches currently under development.

We deployed an analysis framework at the Applied X-ray Absorption Spectroscopy beamline P65 located at PETRA III. The foundation of this system is Asapo [2], a data streaming framework developed at DESY to manage substantial data volumes with high efficiency. Building upon Asapo, we constructed a Python library that simplifies the creation of processing modules termed workers for individual pipeline operations. The implemented workers utilize the Larch [3] library to execute Calibration for the energy drift, Normalization and Linear Combination Analysis (LCA). This modular infrastructure scales to various experimental configurations with higher data rates and can integrate ML components that communicate with the control system or with the scientists to suggest dynamic modifications during ongoing measurements.

References
[1] https://www.rock-it-project.de/
[2] https://asapo.pages.desy.de/asapo/
[3] https://github.com/xraypy/xraylarch

Speaker: Diana Rueda (Deutsches Elektronen-Synchrotron DESY)

16:05 EXAFS study of thermoelectric Bi$_{2-x}$Sb$_x$Te$_3$ using the reverse Monte Carlo method

Bismuth telluride (Bi$_2$Te$_3$), antimony telluride (Sb$_2$Te$_3$), and their solid solutions (Bi$_{2-x}$Sb$_x$Te$_3$) are among the most widely used thermoelectric (TE) materials owing to their excellent thermoelectric performance near room temperature.

Bi$_2$Te$_3$ and Sb$_2$Te$_3$ are isostructural compounds with complex two-dimensional layered structures consisting of Te2-Bi(Sb)-Te1-Bi(Sb)-Te2 quintuple layers, which are weakly bonded by van der Waals interactions. This layered architecture plays a crucial role in enabling their high thermoelectric efficiency. The pronounced structural anisotropy facilitates efficient in-plane charge carrier transport while effectively suppressing cross-plane lattice thermal conductivity. Owing to their structural compatibility, Bi and Sb readily substitute for one another on the cation sublattice, giving rise to continuous solid solutions (Bi$_{2-x}$Sb$_x$Te$_3$) with tunable electronic and thermal transport properties.

In this study, we employed synchrotron radiation X-ray absorption spectroscopy in combination with advanced reverse Monte Carlo simulations with evolutionary algorithm approach (RMC/EA) [1] to probe the composition- and temperature-dependent (10 – 300 K) evolution of the local atomic structure in Bi$_{2-x}$Sb$_x$Te$_3$ (x=0-2). Local lattice distortions originating from the ionic radius mismatch between Bi$^{3+}$ (1.03 Å) and Sb$^{3+}$ (0.76 Å) cations were clearly resolved. By extending the structural analysis to distant coordination shells, encompassing both intralayer and interlayer interactions, we were able to capture subtle disorder effects and correlate them with TE properties [2]. The extracted mean-square relative displacement (MSRD) factors and effective interatomic force constants highlight the strong anisotropy of thermal conductivity inherent to these layered architectures.

References

[1] Timoshenko J., Kuzmin A., Purans J. (2014), EXAFS study of hydrogen intercalation into ReO$_3$ using the evolutionary algorithm, J. Phys.: Condens. Matter 26, 055401.

[2] Hamawandi, B., Parsa, P., Pudza, I., Pudzs, K., Kuzmin, A., Ballikaya, S., Welter, E., Szukiewicz, R., Kuchowicz, M., \& Toprak, M. S. (2025). Scalable solution chemical synthesis and comprehensive analysis of Bi$_2$Te$_3$ and Sb$_2$Te$_3$. Energy Materials, 5, 500065.

Speaker: Inga Pudza (Institute of Solid State Physics, University of Latvia)

16:08 Combined MCR-ALS and MEXAS-PSD to unravel the electrochemical behavior of a Co3Mn-based LDH

Layered double hydroxides (LDH), with the typical formula [M1-x2+Mx3+(OH)2]x+[Ax/n]n-.mH2O, also known as anionic clays, are hydrotalcite-type positively charged lamellar material, in which a part of the divalent M2+ cations is substituted by trivalent M3+ cations. The electroneutrality of the material is ensured by the presence of anion located in the interlayer domain. The composition of the LDH can be easily tuned by varying the nature of the divalent and/or trivalent cations within the layers and/or by changing the interlamellar counter-anions, thus allowing modulation of their physicochemical properties. This versality makes the LDH promising materials for a wide range of applications, particularly in electrochemistry [1].
The present work [2] focuses on the use of Co3Mn-based LDH for the development of electrochemical sensors for the detection of H2O2. The electrochemical activity of the Co3Mn LDH was monitored by quick-EXAFS at the Co and Mn K-edges on the ROCK beamline at the SOLEIL synchrotron [3]. The LDH was deposited on a carbon paper, used as working electrode, and placed in contact with 0.1 M NaOH in a dedicated cell designed for XAS measurements. The electrochemical treatment consisted of cyclic voltammetry (CV) between 0 and 0.6V. Between the first and the last CV at 0.66 mV s-1, modulation excitation XAS (MEXAS) measurements were performed by acquiring 20 CV cycles at 10 mV s-1. The obtained datasets were first analysed via multivariate curve resolution with alternating least squares (MCR-ALS). The results revealed that Mn(III) is irreversibly transformed to Mn(IV) during the 1st cycle, while the Co(II) oxidised in two distinct species, Co(III)1-εCo(III)ε and Co(III), which proportions perfectly matched the CV curves. The findings were further supported by applying phase sensitive detection (PSD) on the MEXAS [4] data obtained from the 20 intermediate CV cycles. While only noise was observed in the phase-resolved spectra at Mn K-edge, meaning that Mn species do not evolve during CVs, the phase-resolved spectra obtained at the Co K-edge can be superimposed to the scaled difference of the two Co species extracted from MCR-ALS.
This work highlights the excellent complementarity of MCR-ALS and MEXAS-PSD analyses in unravelling small changes occurring during electrochemical treatments of materials.

[1] Mousty C., & Farhat H. (2023). Electroanalysis, 35.
[2] Farhat H., et al (2024). J. Phys. Chem. C, 128, 21023-21037.
[3] Briois V., et al (2016). J. Phys.: Conf. Ser., 712, 012149.
[4] Urakawa A., et al (2023). in Springer Handbook of Advanced Catalyst Characterization

Speaker: Dr Anthony Beauvois (SOLEIL synchrotron)

16:11 Interpreting Non-Redox Responses in Modulation-Excitation XAS

Introduction

While operando X-ray absorption spectroscopy (XAS) offers valuable insights into the dynamics of catalysts, it remains a bulk technique. This poses a challenge given that catalysis is a surface phenomenon. To address this limitation, a modulation excitation (ME) experimental design combined with phase-sensitive detection (PSD) analysis can be used. This approach enables the separation of the active fraction of the material from the inert background.1 Traditionally, PSD results are interpreted by comparing the in-phase component to experimentally measured difference spectra. However, with the advent of faster and higher-resolution measurements we are beginning to observe dynamic, non-redox events that cannot be adequately described by conventional difference spectra. This highlights the need for more advanced analysis strategies that go beyond oxidation states, aiming to capture subtle electronic and structural changes. In this work, we explore such an approach by linking PSD results to changes in the electronic structure of the reactive Ni species.

Results & Discussion

We applied ME-XAS to investigate a Ni/SiO₂ catalyst with 16 nm Ni particles at the ROCK beamline of SOLEIL.2 To probe the dynamic response of the catalyst, two modulation periods were applied with CO₂ vs H2 in one case and O₂ vs H2 in the other.
As shown in Figure 2a, the O₂-modulation induces a strong response. Comparison with the a Ni-NiO difference spectrum (Figure 2c) confirms that the observed spectral changes are a transition from Ni⁰ to Ni²⁺. During 1 period, the NiO fraction varies from 21% to 35%. Likely only the outer shell of the Ni nanoparticles is involved.
On Figure 2b, the response to the CO2-modulation is shown. Here, the Ni-NiO difference spectrum does not match the in-phase response. It suggests non-redox changes. These could be caused by adsorbate formation, change in charge transfer at the metal–support interface or slight coordination changes that do not correspond to bulk oxidation. A new approach for interpretation is to link the observed spectral variations with changes in the electronic structure of Ni. An understanding of these phenomena allows for the incorporation of such effects into simulated XANES spectra, allowing the generation of theoretical difference spectra for direct comparison with experimental data.

References

1 D. Ferri et al., Top. Catal., 2011, 54, 1070
2 Briois, V et al., J. Phys. Conf. Ser. 2016, 712, 012149

Speaker: Mr Servaas Lips (Ghent University)

16:14 Advanced Operando XAS Methodologies for Active-Site Identification and Quantification in Materials with Complex Speciation

Quantitative structural identification of active sites in heterogeneous catalysts under reaction conditions using bulk spectroscopic techniques remains fundamentally limited. This is particularly true for Fe-exchanged zeolites, which exhibit isolated Fe ions co-existing with oligomers, and larger aggregates. This poses several challenges: (i) distinguishing active sites from spectators, (ii) quantifying the active sites, and (iii) disentangling changes in oxidation state and coordination environment under reaction conditions.
To address this challenge, we currently implement a suite of experimental protocols and data analysis methods coupled with operando X-ray absorption spectroscopy (XAS). Typical experiments consist of heating the sample to induce dehydration while acquiring spectra, followed by cooling to experiment/reaction temperature. We perform transient step-changes (addition/cut-off) of individual reactants or reactant mixtures to probe kinetics and redox properties. For transient measurements, a chemometric approach combining principal component analysis (PCA) and multivariate curve resolution (MCR) analysis is typically employed. PCA first identifies the minimum number of independent spectral components that reproduce the original data, while also obtaining an evaluation of the residuals. Then MCR decomposes overlapping spectral contributions, allowing us to track the evolution of Fe species during dehydration and reaction, and, when feasible, to extract kinetic parameters that can be correlated to online mass spectrometry and plug-flow reactor tests.
Upon reaching steady-state conditions, we can apply modulated excitation coupled with phase-sensitive detection (ME-PSD). Applying a gas pulse as a stimulus that selectively perturbs the catalytically active sites improves our ability to discriminate between them and spectator species. Stochastic noise is diminished by averaging the response over multiple modulation cycles, which enhances spectral sensitivity. Subsequently, the implementation of a lock-in amplification algorithm enhances signals that are phase-correlated with the applied perturbation, simultaneously suppresses static background contributions, and thereby enables the isolation of periodic, kinetically relevant spectral features.
When time resolution is limited, we combine modulation excitation with the step-scan technique by fixing a single energy point for each gas cycle and repeating this across an entire energy range. This generates an energy–intensity matrix from which spectra can be reconstructed, enabling, in some cases, unprecedented time resolution.
This combined advanced operando methodology and chemometric approach provides a robust basis for future integration with machine learning techniques. Neural networks trained on such decomposed spectral datasets could enable prediction of Fe K-edge XAS spectra from relevant reaction conditions (e.g., temperature, gas composition, redox environment) much faster than quantum chemical calculations. This could facilitate the screening of Fe coordination environments for experimental design and potentially optimize the usage of synchrotron beamtime.

Speakers: Marie-Gabrielle Ameres (PSI LEP ACS), Gabriela-Teodora Dutca (Paul Scherrer Institute)

16:17 Operando XRD–PDF–XAS Study on a model PdIn-based catalyst

CO2 hydrogenation to methanol offers a promising route for valorizing captured CO2 and contributes to the decarbonization of the chemical industry. A key challenge lies in designing active, selective, and stable catalysts. In2O3-based catalysts are highly active and selective for the hydrogenation of CO2 to methanol. However, unsupported In2O3 suffers from deactivation via overreduction to molten metallic In (In0) and subsequent amorphization. [1] Strategies such as the use of adequate supports, as well as the addition of a second metal, notably Pd, provide a means to tune the performance of In2O3-based catalysts (enhancing methanol yield and stability). However, the nature of active sites in the resulting complex Pd-In2O3-based catalysts is highly debated: some study claims that Pd and In2O3 interfaces are the active sites, while other study argues that the alloy formed under the reaction also plays a role. [2-3]
Here, we aim to gain insight into structure-activity relationships in Pd-In catalyst under CO2 hydrogenation conditions. To this end, we developed a model InPd nanoparticle catalyst via a colloidal approach and subsequent deposition on a high surface area amorphous SiO2 (InPd@SiO2). To ensure the formation of a single-phase InPd intermetallic structure, the catalyst was pretreated under a reductive treatment (10% H2/N2, 600°C). However, carbonaceous residues originating from the synthesis may persist on the catalyst surface. To remove these residues and assess their influence on catalytic performance, an oxidative (5% O2/N2, 600°C) pretreatment was applied. The oxidative pretreatment proved more effective at eliminating surface contaminants, resulting in a catalyst with superior activity compared In2O3@SiO2 and Pd@SiO2.
To probe the active phase, we conducted operando XRD-PDF and XAS studies under CO2 hydrogenation (260°C, 20bar, H2/CO2/N2=3:1:1) and after the reductive and oxidative pretreatments. These studies revealed that InPd intermetallic crystal structure is formed under reductive pretreatment and remains stable under CO2 hydrogenation conditions. Under oxidative treatment, the intermetallic structure transforms into In2O3, PdO and Pd. Remarkably, under the CO2 hydrogenation reaction conditions, the intermetallic InPd nanoparticles reform readily, preserving the average crystallite sizes. Operando XAS confirms that both Pd and In are in their reduced states under the reaction condition and reveals a charge transfer between the metals. While CO2 can oxidize In in InPd@SiO2, no detectable In oxidation was observed under CO2 hydrogenation.
These findings demonstrate that the intermetallic PdIn phase modifies the electronic structures of both In and Pd, distinct from those of otherwise inactive metallic In0 or Pd0 species, resulting in a highly active and stable catalyst for CO2 hydrogenation to methanol. This study also showcases that operando XRD–PDF–XAS on model catalyst can help decipher structure-activity relationships in complex catalytic systems.

[1] Tsoukalou et al., J. Am. Chem. Soc. 2019, 141, 13497−13505
[2] Potter et al., Angew. Chem. Int. Ed 2023, e202312645.
[3] Araújo et al., Angew. Chem. Int. Ed 2023, e202306563.

Speaker: Qin ZHANG (ETH Zurich)

16:20 First-principle simulated Fe4 L-edge XAS reveals redox-sensitive spectral shifts upon Li doping.

We present fully ab initio simulations of Fe L-edge X-ray absorption spectroscopy (XAS) for the archetype single-molecule magnet tetrairon Fe₄ using a linear-response time-dependent density functional theory with a spin-orbital coupling scheme. In particular, electronic and structural modifications in the Fe4 core, as induced by Li doping and by the change of R (-H and -C₅S$\cdot$), were studied systematically benchmarking hybrid functionals and basis sets. A parameter-free computational protocol is, therefore, established, which reproduces experimental spectra with excellent agreement. The simulations capture key spectroscopic signatures, including L₃/L₂ splitting, redox-induced shifts upon Li doping, and the robustness of spectral shapes against magnetic coupling schemes and structural distortions. This study establishes a practical and accurate framework for simulating 2p XAS in complex magnetic molecules, providing valuable insight into their electronic behavior and enabling a rigorous connection between experiment and theory.

Speaker: Nanchen Dongfang (Universität Zürich)

16:23 Speciation of Ru Molecular Complexes in a Homogeneous Catalytic System: Fingerprint XANES Analysis Guided by Machine Learning

Identifying the true active species in homogeneous catalytic systems remains one of the most demanding challenges in modern spectroscopy. Low metal concentrations and dynamic reaction environments often make conventional EXAFS analysis impossible, leaving XANES as the only accessible technique for probing local structure.
In this work, we combine machine learning with fingerprint XANES analysis to reveal the speciation of Ru molecular complexes that can be formed in situ in homogeneous systems. Ruthenium-based catalytic systems can be used to hydrogenate sugar alcohols to alkenes, offering an efficient pathway toward environmentally friendly and sustainable chemical processes. The system based on RuX$_3$ salts (X = Cl, Br) dissolved in the ionic liquid (Bu$_4$PBr) has been proposed as a catalyst for this type of reaction.
A comprehensive database of theoretical Ru K-edge XANES spectra was generated using ab initio simulations (FDMNES), systematically varying Ru-Br/Ru-Cl and Ru-CO distances, as well as CO coordination numbers. The resulting data were processed using a descriptor-based machine learning approach trained to predict both ligand type and interatomic distances from selected spectral features such as edge position, curvature of the white line, and PCA-derived components. The work was carried out using the original program codes written in Python, using the PyFitIt library.
The trained models achieved $R^2 > 0.98$ in cross-validation and accurately reproduced both Ru-Br/Ru-Cl and Ru-CO bond lengths and CO coordination numbers for experimental spectra measured at ESRF BM23. Remarkably, even mixed or previously uncharacterized species were correctly identified.
This study demonstrates how combining spectral descriptors with physical modeling can overcome sensitivity limitations of XAS in homogeneous catalysis. The methodology provides a general framework for extending fingerprint analysis beyond qualitative interpretation, enabling the direct identification of molecular catalyst species from XANES spectra.

Speaker: Ms Elizaveta Kozyr (Department of Chemistry and NIS Centre, University of Torino, Italy; The Smart Materials Research Institute, Southern Federal University)

16:26 Machine learning application for automatic structural information extraction from experimental operando data for Ni and Mn based cathodes materials

Over the years, X-ray absorption spectroscopy is playing an increasingly important role in the development and innovations in battery science, due to its unique ability to provide accurate information on the electronic structure of redox active elements and local structural information, also in operando conditions. Despite its potential innovative outcomes, the generation of significant developments seems hindered by the resources needed to fully and quickly exploit these capabilities and the limited cooperation between academic and industrial parties. Several challenges can be identified for the development of more performing and sustainable batteries. In this work we utilize a set of supervised and unsupervised ML algorithms including convolutional neural networks (CNNs), random forests, extra trees paired with principal component analysis (PCA) to disentangle structural and electronic parameters and to unambiguously identify transition metal oxidation state and spin variations in layered transition metal cathode materials. Complementary automated fitting approach of big in operando XAS data sets allows us to extract spectral and structural descriptors such as absorber-ligand distances, the energy positions of the rising edge, the pre-edge features, the most influenced spectral points (maxima and minima). To establish structure–property relationships, the extracted descriptors are further correlated with electrochemical parameters. In particular, we show how data driven ML analysis of in operando quasi-simultaneous Mn and Ni K-edges XANES spectra can offer a scalable pathway for real-time monitoring and interpretation of TM-based (TM = Ni, Mn) cathode materials under in operando conditions, paving the way toward data-driven design and optimization of energy-related systems.

Speaker: Oleg Usoltsev (ALBA synchrotron)

16:29 A TDDFT-based Method for the Calculation of Resonant Inelastic X-Ray Scattering in Condensed Phases

Resonant Inelastic X-ray Scattering (RIXS) is a powerful photon-in/photon-out spectroscopic technique that provides unique, orbital-specific insights into electronic structure by probing excitations from a localized core orbital. Its accurate simulation for condensed-phase systems requires a method that simultaneously describes the local electronic structure of the absorbing atom and the extended environment while balancing computational cost with accuracy. We address this by presenting a new implementation of a RIXS module within the CP2K software package. Built upon CP2K's highly efficient linear-response TDDFT and XAS frameworks, this implementation presents the first generally available method for calculating RIXS spectra in both molecular and extended systems. We demonstrate the accuracy of our implementation on molecular systems and solutions, providing a robust approach for simulating RIXS spectra in complex environments.

Speaker: Beliz Gökmen (University of Zurich)

16:32 Cation-Site Disordered Cu3PdN Nanoparticles for Hydrogen Evolution Electrocatalysis

Transition metal nitrides (TMNs) are emerging as a promising class of materials for applications in optoelectronics, as well as energy conversion and storage; however, they remain relatively unexplored, primarily due to a lack of mechanistic understanding of their synthetic pathways. Here, a one-pot synthesis is demonstrated, yielding 3 nm phase-pure Cu3PdN nanoparticles after reacting Cu methoxide and Pd acetylacetonate in benzylamine for 5 min at 140 °C. The structure of the initial complexes and their conversion to Cu3PdN are revealed by in situ X-ray absorption spectroscopy measurements, and elucidate nucleation and growth of the nitride nanocrystals by in situ total X-ray scattering measurements. Interestingly, extended X-ray absorption fine structure double-edge refinement reveals the presence of short-range cation-site disorder in the anti-perovskite structure of Cu3PdN, which has not been observed before in the Cu3PdN system. Additionally, the synthesized Cu3PdN nanoparticles are tested for the electrocatalytic hydrogen evolution reaction, revealing an overpotential as low as η10 = 212 ± 11 mV measured at 10 mA cm−2.

Speaker: Jagadesh Kopula Kesavan (University of Hamburg)

16:35 Spectroscopic insights into the electronic structure of non-critical rare earth containing permanent magnets

The extensive use of critical rare-earth elements like Nd and Sm in magnet production raises concerns about their limited availability [1]. Ongoing research explores the feasibility of cost-effective hard magnetic materials by substituting Nd or Sm with more abundant rare-earth elements such as Ce or La [2]. Here, it is crucial to deepen our understanding of the electronic and magnetic structures of compounds consisting of these low-cost rare-earth elements.

Here, we study the role of Ce as a candidate to replace heavy RE elements by investigating its 4f/5d valence state in different Ce-substituted permanent magnet systems. X-ray absorption spectroscopy was performed on Ce-Co, Ce-Co-Zn, and Ce-Co-Cu systems with varying compositions, examining the respective K, L$_{2,3}$, and M$_{4,5}$ edges of their components. We correlate the spectroscopically determined Ce valence with the composition and magnetic properties to understand the influence of Ce content on their magnetic properties.

We acknowledge the financial support of the German Research Foundation (CRC/TRR 270), the BMBF (05K2019 and 05K2022), Toyota Motor Corporation and the French National Research Agency (ANR-22-CE91-0008). Furthermore, we thank the ESRF for the allocation of beamtime at beamlines ID12 & ID32 within projects MA-5882 & MA-6819.

[1] O. Gutfleisch et al. Adv. Mater. 23, 821-842 (2011)
[2] K. P. Skokov et al. Scripta Materialia, 154, 289-294 (2018)

Speaker: Benedikt Eggert (University of Duisburg-Essen)

16:38 A Bayesian Framework for Feature Extraction in Noisy Operando X-ray Absorption Spectroscopy

We present a Bayesian framework for robustly extracting spectroscopic features from X-ray Absorption Spectroscopy measurements and synthetic reference datasets. This approach is particularly designed for complex XAS experiments with noisy spectra and irregular backgrounds, especially under operando conditions in the soft X-ray regime, where absorption by membranes in the optical path, sample environments and/or biased sample often compromises signal quality. Accurate identification of overlapping signal peaks and separation from experimental background and noise is crucial for interpreting fine structural and electronic information. Our approach combines physics-informed forward models for background, absorption edge, and composite peak shapes with Markov Chain Monte Carlo sampling to perform full posterior inference over model parameters. Using both laboratory-collected XAS spectra and controlled synthetic datasets, we show that the method can recover peak positions, widths, and absorption parameters for a broad range of spectra. We validate the pipeline by (i) recovering fiducial parameters from synthetic spectra, (ii) estimating dependence on experimental noise and parameter degeneracies using synthetic datasets, and (iii) producing posterior predictive checks showing agreement between data and reconstructed models. The implementation is computationally efficient and supports parallel MCMC sampling; results include best-fit parameter estimates, marginalized posterior distributions, and diagnostic outputs for residuals and chi-squared statistics. This probabilistic workflow facilitates rigorous interpretation of XAS features and provides a generalizable tool for spectroscopic analysis where signal-background disentanglement and uncertainty quantification are required.

Speakers: Mr Tommaso Rodani (Area Science Park), Matteo Biagetti (Area Science Park)

16:41 Real-Time TDDFT Simulations of Core-Level Spectroscopies in the Condensed Phase

Density functional theory (DFT) based methods have become standard tools for accurately describing core-level spectroscopies in systems ranging from small gas-phase molecules to periodic condensed-phase materials. Within the Kohn-Sham DFT (KS-DFT) framework, core-excited states are commonly treated using linear-response time-dependent DFT (LR-TDDFT). More recently, real-time propagation approaches have emerged which enable the simulation of absorption spectra by Fourier transforming the time-dependent dipole moment generated in response to an external perturbation [1,2].

In this work, core-level spectroscopies are simulated using the real-time TDDFT (RT-TDDFT) implementation in the CP2K set of programs [3]. A protocol is first established for calculating static X-ray absorption spectra (XAS) of gas- and liquid-phase water under periodic boundary conditions. The same computational framework is then extended to capture coupled electron-nuclear dynamics on femtosecond timescale. Finally, time-resolved XAS is obtained by propagating the electronic density from well-defined initial states within CP2K, enabling direct simulation of ultrafast core-level spectroscopic responses.

[1] Pemmaraju, C. D. et al., Velocity-gauge real-time TDDFT within a numerical atomic orbital basis set. Comput. Phys. Commun. 2018, 226, 30-38

[2] Tussupbayev, S. et al., Comparison of Real-Time and Linear-Response Time-Dependent Density Functional Theories for Molecular Chromophores Ranging from Sparse to High Densities of States. J. Chem. Theory Comput. 2015, 11, 1102-1109

[3] Kühne, T. D. et al., CP2K: An electronic structure and molecular dynamics software package - Quickstep: Efficient and accurate electronic structure calculations. J. Chem. Phys. 2020, 152, 194103

Speaker: Michael Coates (University of Zurich)

16:44 Data-Driven Analysis of Nuclear Resonance Vibrational Spectra with Machine-Learning Potentials

Nuclear Resonance Vibrational Spectroscopy (NRVS) is a synchrotron-based inelastic X-ray scattering technique that probes the vibrational density of states projected onto a Mössbauer isotope. NRVS spectra can be transformed into an element-projected phonon density of states (PDOS) integrated over the Brillouin zone. While the forward problem - calculating the PDOS from a known structure and interatomic forces - is straightforward, the inverse problem of extracting structural and kinetic information from experimental PDOS is considerably more challenging and requires advanced modelling and data-driven analysis.
In this work, we report the first operando NRVS measurements of a LiFePO₄ (LFP) electrode in a Li-ion cell. During cycling, lithium is extracted from LFP, forming FePO₄ via an intermediate metastable phase that is only rarely reported in the literature. Vibrational spectroscopy probes the local curvature of the Born–Oppenheimer surface, which is crucial for understanding phase-transformation kinetics and ionic transport. Using machine-learning-based techniques such as principal component analysis, we detect the signature of a third, metastable phase formed during charge–discharge. Non-negative matrix factorization enables us to decompose the raw spectra into contributions from individual phases, including the metastable intermediate.
To interpret these components structurally and kinetically, we perform ab initio phonon calculations for candidate structures and match the computed PDOS to experiment. For the metastable phase, we move beyond conventional DFT to a neural-network-based universal interatomic potential, that we fine-tuned on our DFT dataset, which allows us to simulate substantially larger supercells with diverse defect arrangements. We first tune the DFT-NRVS agreement for the stable phases and then identify configurations that best reproduce the experimental spectra of the intermediate. This workflow yields insight into the structure, thermodynamic stability, and transformation kinetics of the metastable phase. Finally, we apply the same data-driven NRVS–simulation framework to vibrational spectra of ceramic proton conductors, establishing quantitative links between phonons and transport of light ions such as Li⁺ and H⁺.

Speaker: Alexey Rulev (Empa)

16:47 Fast elemental composition analysis of X-ray fluorescence spectroscopy with Neural Network

The data obtained from the synchrotron generally requires a considerable amount of time to be analyzed (months or years). As part of this project (funded by the PEPR DIADEM and started in February 2025), we aim to apply neural network algorithms in order to obtain a first estimate of the results directly from the raw data, in real time during the experiment.
In the first phase of the project, we decided to select a simple technique that is common to most beamlines. Moreover, since neural networks require thousands of labeled data samples, it was necessary to choose a technique that can be easily simulated.
The choice fell on X-ray fluorescence (XRF); this technique can be simulated with a high degree of accuracy using software called McXtrace. The idea is to determine the stoichiometry of the measured material directly from the raw spectrum, thereby avoiding manual analyses.
In this presentation, we will describe the algorithm developed during this initial phase of the project. The algorithm can already predict the stoichiometry of 51 elements from XRF spectra for synthetic data, with a variable photon energy and detector energy (between 3 and 60 keV). The results on simulated data are promising, with a coefficient of determination of R² = 94%.
The next step is to apply the model to real XRF data. To achieve this, we plan to fine-tune the model parameters using the experimental spectra. Furthermore, the approach will be extended to other techniques such as XRD or XAS.

Speaker: Francesco La Porta (Soleil Synchrotron)

16:50 High-Entropy Oxides as Versatile Catalysts for Thermocatalytic CO₂ Conversion

High-entropy oxides (HEOs) are multicomponent materials containing five or more cations distributed within a single crystalline lattice and are gaining attention as promising catalysts due to their unique structural stability, redox properties, and synergistic multi-cation effects [1]. Although HEOs are promising for a wide range of applications, their rational design remains challenging due to the large compositional space and complex characterization. X-ray absorption spectroscopy (XAS) has proven to be a valuable tool to understanding high entropy materials: for instance, EXAFS can determine the random and uniform distribution of cations, a standard criterion for defining HEOs [2]. Additionally, this technique allows to observe changes in oxidation state to identify mechanisms for charge compensation during the structures formation [3]. Since HEOs are only now emerging as heterogeneous catalysts [4], there is a decided lack of fundamental knowledge regarding how the high-entropy form plays a role in catalysis and which cations are participating in the reaction.
In this work, HEOs are tailored for thermocatalytic CO₂ hydrogenation towards value-added products, such as carbon monoxide, methane [5], methanol [6], and larger hydrocarbons. In order to address the possible permutations for chemical composition, in collaboration with SwissCat+ Hub East, we have employed high-throughput synthesis and catalytic testing with machine learning to provide AI-assisted synthesis to optimize the performance for CO2 hydrogenation for different product selectivities. Furthermore, this work utilizes ex-situ XAS to elucidate structure-activity correlations by probing low-, medium-, and high-entropy oxides before and after reaction. Specifically, the K-edges of Cr, Fe, Cu, Ni, Co, and Zn were probed to track changes in the local environment of each cation and to elucidate their individual roles during the CO2 hydrogenation reaction.

Speaker: Fausto Aldegheri (ETH Zurich - Empa)

16:53 Linking Structure and Electrochemistry in Pt Supported on Mesoporous N-Doped Carbon (MPNC) Fuel-Cell Catalysts: A Complete X-ray Picture from XAS to XRS

Platinum on high-surface-area carbon supports remains the benchmark oxygen-reduction-reaction (ORR) catalyst in proton-exchange-membrane fuel cells (PEMFCs) and a leading cathode for the hydrogen-evolution reaction (HER). The mesoporous N-doped carbon (MPNC) is used as a tunable platform to control dispersion and metal, support interactions, not as a field-wide state-of-the-art. The working catalyst is dynamic: entity size distributions (single atoms (SA), sub-nanometer clusters, nanoparticles (NP)), oxidation state, and coordination evolve under bias and mass-transport stress. Pt L₃-edge X-ray absorption spectroscopy (XAS), interpreted via X-ray absorption near-edge structure (XANES) and extended X-ray absorption fine structure (EXAFS), including wavelet-domain views, enables quantitative speciation linking structure to durability and performance. We present a unified program following Pt/MPNC across synthesis, baseline characterization, stress evolution, and operation: (i) ex-situ baselining with synthesis control, (ii) accelerated-stress-test (AST)-resolved XAS with electrochemical readouts, and (iii) in-situ electrochemical XAS (EC-XAS) capturing CO-stripping transients and steady-state HER/ORR. In-situ EC-XAS then captures electrocatalyst dynamics under working conditions by tracking the spectroscopic response while co-recording the electrochemical CO-stripping signal, thereby directly correlating structural signatures with electrochemical performance.1,2
Modulating synthesis on Pt-MPNC electrocatalyst and tracking with XAS tunes the SA/cluster/NP ensemble, then advances to operando to interrogate interfacial chemistry during CO-stripping voltammetry, providing predictive baselines for durability. To complete the picture of the MPNC support and Pt–support interactions, we extend beyond absorption to hard X-ray Raman scattering (XRS) of the C/N (and O) K-edges delivers bulk-sensitive soft-edge information in Pt/MPNC, while valence-to-core X-ray emission spectroscopy (XES; Pt Lβ₂) reports on ligand field and covalency around Pt. These datasets, together with aberration-corrected high-resolution transmission electron microscopy (ac-HRTEM), scanning transmission electron microscopy–energy-dispersive X-ray spectroscopy (STEM-EDX), and TEM tomography, corroborate speciation assignments and spatial heterogeneity at the nanoscale. Overall, the continuous Pt-L₃ XAS program, augmented by XRS/XES and correlative EM, yields design rules linking initial dispersion and interfacial chemistry to operando dynamics and durability in Pt/MPNC catalysts for PEMFC ORR and HER.

References:
(1) Küspert, S.; Campbell, I. E.; Zeng, Z.; Balaghi, S. E.; Ortlieb, N.; Thomann, R.; Knäbbeler-Buß, M.; Allen, C. S.; Mohney, S. E.; Fischer, A. Small 2024, 20 (34), 2311260.
(2) Zeng, Z.; Küspert, S.; Balaghi, S. E.; Hussein, H. E. M.; Ortlieb, N.; Knäbbeler-Buß, M.; Hügenell, P.; Pollitt, S.; Hug, N.; Melke, J.; Fischer, A. Small 2023, 19 (29), 2205885.

Speaker: Dr S. Esmael Balaghi (University of Freiburg)

16:56 In-house X-ray absorption spectroscopy instrumentation at MPI CEC

At the MPI CEC, high energy resolution X-ray spectrometers are used for investigation molecular systems designed for storing and releasing energy in chemical bonds. X-ray absorption spectroscopy is, among the different spectroscopic techniques available at the institute, one option for interrogating the electronic and geometric structure of systems, permitting to investigate on an element-sensitive basis changes in oxidation state, coordination or interatomic distances upon catalytic cycling. In contrast to synchrotron radiation-based experiments, investigations are currently performed in transmission mode only. However, the flexible access to the instrumentation allows for short-notice experiments, while avoiding long-distance transfer of samples.
Depending on the required information, the near-edge or the extended range in the X-ray absorption fine structure can be explored, each with dedicated instruments. In the hard X-ray regime, XANES experiments can be realized using either a Johann- [1] or a von Hamos-type spectrometer, while for EXAFS a von Hamos-type spectrometer is available [2]. In the XANES instruments pure crystals with a rather large bending radius are used to make sure the required resolving power can be achieved, while for the EXAFS spectrometer the detection efficiency was prioritized in the design of the instrument. To this end, a highly annealed pyrolytic graphite crystal with a rather small radius of curvature is used. This design will prove beneficial in the realization of an in situ set up for thermal catalysis, with first experiments being designed around investigations of the reformation of ammonia.
Complementarily to the hard X-ray range, experiments in the soft X-ray range are possible as well on the laboratory level. These experiments are enabled by using a laser produced plasma source [3] instead of conventional X-ray tubes. These sources deliver high flux in the energy range from 200 eV to 1000 eV. Coupling them with reflective zoneplates in a dispersive scheme, allows for investigating the NEXAFS range around the K-ionization threshold of low-Z materials or the L-ionization threshold of 3d transition metals [4], or in other words studying the ligands or the metallic centers of molecular systems. Thus, complementary insights to the K-edge experiments on the 3d transition metals is offered. Further, due to the pulsed time pattern of the source even transient experiments are enabled, an option that is currently being commissioned for solid and liquid sample environments
The present status of the different instruments mentioned will be detailed in this contribution, along with selected examples and future development plans. The data that is collected using in-house X-ray spectrometers presents a valuable resource for delivering training data, at the premise that the instrumentation parameters are well controlled and accounted for. For this aspect curated data is essential. Reversely, a growing community of users of laboratory-based X-ray spectrometers will profit from advanced modelling tools.
References
[1] E. Jahrmann, et al. (2019). An improved laboratory-based x-ray absorption fine structure and x-ray emission spectrometer for analytical applications in materials chemistry research, Rev. Sci. Instrum., 90, 024106.
[2] C. Schlesiger, et al. (2020). Recent progress in the performance of HAPG based laboratory EXAFS and XANES spectrometers, J. Anal. At. Spectrom., 35, 2298-2304.
[3] I. Mantouvalou et al. (2015). High average power, highly brilliant laser-produced plasma source for soft X-ray spectroscopy, Rev. Sci. Instrum., 86, 035116.
[4] A. Jonaset al. (2019), Towards Poisson noise limited optical pump soft X-ray probe NEXAFS spectroscopy using a laser-produced plasma source, Opt. Express, 27, 36524-36537.

Speaker: Yves Kayser

17:00 → 18:30 Poster Session

19:00 → 17:00 Symposium Dinner - Restaurant Bellavista

Tuesday 6 January

09:00 → 12:30 Advanced Approaches to EXAFS Analysis

09:00 Unlocking Hidden Functionality Descriptors in Nanomaterials with Machine Learning – Driven XAFS

Understanding how nanomaterials work requires identifying their active units, pinpointing active sites, and providing data that inform theoretical models. Two major challenges are the intrinsic heterogeneity of active species and their dynamic restructuring in reactive environments. In the first part of my talk, I will introduce a descriptor-based approach that bridges structure and function in nanomaterials. X-ray absorption spectroscopy (XAS) is ideally suited to probe such descriptors in operando, yet their identities and numbers are often “hidden” within the spectra. I will describe the machine learning – based methodology our group has developed over the past decade to extract structural, compositional, electronic and dynamic descriptors from X-ray spectra and link them to material’s function. I will also discuss extensions of this approach to other structural techniques.

Speaker: Prof. Anatoly Frenkel (Stony Brook University)

09:40 Convolutional AutoEncoder for Anomaly Detection and Chemical-State Identification of XAS Spectra in Operando Catalysis

Machine Learning (ML) techniques offer powerful tools for advancing scientific discoveries and are increasingly integrated into material science, particularly X-ray Absorption Spectroscopy (XAS) studies in situ. These methods enable applications ranging from real-time spectral analysis during catalytic reaction to the prediction of structural parameters [1,2]. As a wide range of spectroscopic data becomes accessible, data-driven and ML-based approaches gain more attention along grounded theoretical methods [3,4].
In the Helmholtz-funded ROCK-IT (Remote, Operando Controlled, Knowledge-driven, and IT-based) project [5], which aims to automate Operando Catalysis experiments at synchrotron light sources, high-quality XAS data were acquired for modeling X-ray absorption spectra, as demonstrated on CO₂ methanation using various catalysts. For the analysis of these experimental data, we utilized Unsupervised Learning methods, specifically Convolutional AutoEncoder (CAE) for XAS spectral Reconstruction and Anomaly Detection for data quality improvement, in addition to Chemical-state Identification over time through latent-space analysis [6,7]. AutoEncoders are neural network architectures designed to learn compressed representation of high-dimensional input data through Encoding/Decoding processes [8]. The Convolutional Neural Network (CNN) forms the basis for many image processing frameworks, adapted in this work for spectral analysis to learn spatial properties, extract relevant features, and identify anomalies within the spectral domain.

References

[1] Cuenya, B. R., & Banares, M. A. (2024). Introduction: Operando and in situ studies in catalysis and electrocatalysis. Chemical Reviews, 124(13), 8011–8013.
[2] Timoshenko, J., & Cuenya, B. R. (2021). In situ/operando electrocatalyst characterization by X-ray absorption spectroscopy. Chemical Reviews, 121(2), 882–962.
[3] Newville, M. (2014). Fundamentals of XAFS. Reviews in Mineralogy and Geochemistry, 78(1), 33–74.
[4] Kharel, S. R., et al. (2025). OmniXAS: A universal deep-learning framework for materials X-ray absorption spectra. Physical Review Materials, 9(4), 043803.
[5] Rock-it project, https://www.rock-it-project.de/project/about/
[6] Liang, Z., Carbone, M. R., et al. (2023). Decoding structure–spectrum relationships with physically organized latent spaces. ArXiv:2301.04724.
[7] Tetef, S., Govind, N., & Seidler, G. T. (2021). Unsupervised machine learning for unbiased chemical classification in X-ray absorption spectroscopy and X-ray emission spectroscopy. Physical Chemistry Chemical Physics, 23, 23586–23601.
[8] Hinton, G. E., & Salakhutdinov, R. R. (2006). Reducing the dimensionality of data with neural networks. Science, 313(5786), 504–507.

Speaker: Athar Khodabakhsh (Helmholtz-Zentrum Berlin)

10:05 Noise-Aware Multimodal Deep Learning for Real-Time Prediction of EXAFS Fitting Quality

Hyperspectral X-ray absorption spectra (XAS) imaging implemented at the ROCK-SOLEIL beamline [1] offers a unique opportunity to combine second time-resolution and micrometer spatial resolution for monitoring electronic and local order transformations occurring during a chemical reaction. Individual spectra collected under operando conditions at each raw pixel of hyperspectral cubes have low signal-to-noise ratio (SNR) due to the speed at which they are collected (5s) and the pixel size (1.625 µm). To improve the SNR of the spectra, cube merging and pixel binning (time and spatial averaging) are used as post-processing of dozens of collected hyperspectral cubes. In this work, an intelligent data collection using noise-aware deep learning was developed to predict the optimal number of cubes to collect to accurately extract structural parameters (e.g. coordination numbers) from EXAFS spectra targeting a spatial or time resolution.
Quantifying the quality of XAS fits remains a bottleneck for hyperspectral imaging, where millions of low-SNR spectra make pixelwise FEFF fitting computationally expensive. We address this by developing a data-driven method that predicts the EXAFS amplitude fitting uncertainty directly from noisy spectra—combining classical FEFF modelling with deep learning.
We synthetically generate large training datasets by perturbing clean μ(E) XAS spectra collected at ROCK for references under controlled noise models and acquisition conditions. Three noise models are compared: Gaussian-only, Poisson-only, and a mixed model combining counting statistics with residual Gaussian noise [2], [3]. Each synthetic dataset spans variations in edge shift (±10 eV), and k-weight (2, 3), producing >600 k labelled spectra.
A multimodal convolutional neural network (CNN) architecture [4], [5], pretrained on synthetic data and fine-tuned [6] with real beamline measurements, predicts the FEFF-derived fitting uncertainty σ_amp directly from spectra. The spectral branch processes χ(k) concatenated with |χ(R)|, while a compact embedding branch encodes the absorbing element, chemical family, edge type, and k-weight, allowing the network to generalize across diverse experimental conditions and absorption edges. At the pretraining stage, the Poisson-only model provides the most realistic calibration and transferability to real data, with the mixed model performing comparably. After fine-tuning, all models converge toward similar high accuracy (R² ≈ 0.9, MAE ≈ 2%), demonstrating that real-data adaptation effectively bridges the synthetic–experimental domain gap.
Throughput. On a CPU workstation we process around 4,000 spectra/s; on a single RTX-class GPU we reach 50,000 spectra/s, yielding ~37 s for a full 600×350 cube at 70% effective pixels (147k spectra) on CPU versus 3 s on GPU.
Sweeps over temporal averaging and spatial binning reveal clear quality–cost trade-offs, showing how acquisition parameters directly influence the predicted fitting uncertainty. The resulting σ_amp maps provide near-real-time feedback during experiments, allowing users to adjust time resolution or binning to reach a desired precision while optimizing acquisition time. Overall, this method delivers a data-driven approach for quantifying spectral reliability across full hyperspectral cubes, enabling adaptive and efficient acquisition strategies at modern synchrotron beamlines.
References
[1] V. Briois et al., « Hyperspectral full-field quick-EXAFS imaging at the ROCK beamline for monitoring micrometre-sized heterogeneity of functional materials under process conditions », J. Synchrotron Radiat., vol. 31, no 5, Art. no 5, sept. 2024, doi: 10.1107/S1600577524006581.
[2] N. Acito, M. Diani, et G. Corsini, « Signal-Dependent Noise Modeling and Model Parameter Estimation in Hyperspectral Images », IEEE Trans. Geosci. Remote Sens., vol. 49, no 8, p. 2957‑2971, août 2011, doi: 10.1109/TGRS.2011.2110657.
[3] A. Foi, M. Trimeche, V. Katkovnik, et K. Egiazarian, « Practical Poissonian-Gaussian Noise Modeling and Fitting for Single-Image Raw-Data », IEEE Trans. Image Process., vol. 17, no 10, p. 1737‑1754, oct. 2008, doi: 10.1109/TIP.2008.2001399.
[4] S. G. Elnaggar, I. E. Elsemman, et T. H. A. Soliman, « Embedding-Based Deep Neural Network and Convolutional Neural Network Graph Classifiers », Electronics, vol. 12, no 12, p. 2715, juin 2023, doi: 10.3390/electronics12122715.
[5] X.-K. Ma et al., « X-ray spectra correction based on deep learning CNN-LSTM model », Measurement, vol. 199, p. 111510, août 2022, doi: 10.1016/j.measurement.2022.111510.
[6] « Neural Network-Based On-Chip Spectroscopy Using a Scalable Plasmonic Encoder | ACS Nano ». Consulté le: 13 octobre 2025. [En ligne]. Disponible sur: https://pubs-acs-org.ressources-electroniques.univ-lille.fr/doi/full/10.1021/acsnano.1c00079

Speaker: Lei JIANG (Synchrotron SOLEIL)

10:30 Coffee Break

11:10 EXAFS analysis with Graph Neural Networks

Speaker: Dr Javier Heras (University of Barcelona)

11:50 Machine learning algorithms for EXAFS analysis

Speaker: Dr Janis Timoshenko (Fritz Haber Institute of the Max Planck Society)

12:30 → 13:30 Lunch Break

13:30 → 16:50 Bridging Simulations and Experiments with Machine Learning

13:30 Machine learning potentials applied to MD-EXAFS

Speaker: Dr Alexei Kuzmin (University of Latvia)

14:10 Combining Reverse Monte Carlo Analysis of X-ray Scattering and Extended X-ray absorption Spectroscopy

Extended X-ray absorption fine structure (EXAFS) spectra contain information about the local, molecular type structure, whereas (X-ray) diffraction (XRD) data reveal the periodic structure or long-range order (crystal structure) of materials. Variations in local and periodic structure greatly influence materials properties and related applications. However, data analysis often is performed independently for EXAFS spectra and diffraction data even if measured simultaneously.
RMC simulations enable the analysis of X-ray scattering (XS) data as well as EXAFS spectra data via partial pair distribution (pPDF) functions obtained from a physical, structural model. In case of nanoparticles and scattering data this approach suffers from the termination of the pPDF’s due to the finite size of the particles. This produces artifacts in the computed scattering intensity due to the long-range probing distance of scattering which are eliminated using the Debye scattering equation (DSE) [1, 2]. Simultaneous refinement of XS data and EXAFS spectra of small nanoparticles are thus enabled using a mutual structural model. This method allows the self-consistent extraction of complementary information on local structure contained in EXAFS and long-range order in XS data. However, refinement of raw diffraction data for crystals of larger domain size (larger than about 10 nm or 50.000 atoms) are difficult. A Rietveld code embedded into the RMC code and feedback of the essential structural information between both refinement paths enables the coupled refinement of diffraction and EXAFS data are in this case [3].

[1] M. Winterer and J. Geiß, Combining reverse Monte Carlo analysis of X-ray scattering and extended X-ray absorption fine structure spectra of very small nanoparticles, J. Appl. Cryst. 56 (2023) pp. 7; doi.org/10.1107/S1600576722010858
[2] V. Mackert, T. Winter, S. Jackson, R. Kalia, A. Levish, S. Lukic, J. Geiss, and M. Winterer, Very Small Nanocrystalline Tin Dioxide Particles: Local-, Crystal-, and Micro-Structure, J. Phys. Chem. C 127 (2023) 17389–17405, 16p.; doi.org/10.1021/acs.jpcc.3c02110
[2] M. Winterer, Coupling Rietveld refinement of X‐ray diffraction data and reverse Monte Carlo analysis of extended X‐ray absorption fine structure spectra, J. Mat. Res. 40 (2025) 649-661; doi.org/10.1557/s43578-025-01545-3

Speaker: Prof. Markus Winterer (University of Duisburg-Essen)

14:50 Coffee Break

15:30 X-ray absorption spectral shapes

I will discuss some aspects of the theory and simulation of x-ray absorption spectral shapes. Concerning machine learning approaches, a complication is that the most accurate methods take too much time, so they ideally should be approximated in a reliable fashion. We can distinguish 3 different starting points for the interpretation of XAS spectral shapes:

(1) Closed shell systems can be described with Bethe-Saltpeter (BSE) or time-dependent DFT models that can be approximated with DFT based approaches
(2) 3d and 4d XAS of f-systems and 2p XAS of 3d systems can be described with local models dominated by the intra-atomic electron-electron interactions in the atomic multiplet and crystal field multiplet models
(3) Covalent 3d systems need the inclusion of charge fluctuations (charge transfer Δ and Hubbard U) in impurity multiplet or Dynamical Mean-Field (DMFT) methods.

There are a number of additional issues:
(a) In the DFT approximation to BSE/TDDFT, one has to choose an exchange-correlation potential, determine its U value and choose a core-hole treatment.
This leads to a variety of methods/choices and it is likely that the best choice is material and XAS edge dependent.
(b) The interpretation of transition metal K edges adds some additional challenges. The main dipole edge can be interpreted from DFT. However, the 1s XPS spectra of (for example) transition metal oxides show multiple peaks, implying that one photon energy gives rise to electrons with multiple kinetic energies. This means that the 1s XAS spectral shape must be described as the convolution of the empty states as calculated with DFT with the 1s XPS spectral shape [Ghiasi et al., Phys. Rev. B. 100, 075146 (2019)]. In addition, the quadrupole peaks are excitonic and have an extra core hole shift of ~3 eV and need to be described with multiplets.
(c) The complication for machine learning based multiplet methods (atomic, crystal field, impurity, DMFT) is that these are real space localised model Hamitonians. There exist no uniform first-principle method to derive the parameters in these model Hamiltonians from geometric structures. A nice example is the Hubbard U that is large in impurity multiplet calculations, but approaches zero in DFT with accurate XC potentials such as r2SCAN.

These issues together imply that we are still far from a general first-principle route of accurate XAS spectral shape simulations.

Speaker: Prof. Frank de Groot (Utrecht University)

16:10 DFT-assisted XANES simulations

Speaker: Dr Grigory Smolentsev (PSI - Paul Scherrer Institut)

16:50 → 17:00 Closing