Neural Network Approach for Characterizing Structural Transformations by X-Ray Absorption Fine Structure Spectroscopy

A Combined X-ray Absorption and UV-Vis Spectroscopic Study of the Iron-Catalyzed Belousov-Zhabotinsky Reaction

The iron-catalyzed Belousov-Zhabotinsky (BZ) oscillating reaction was investigated in an unstirred reactor by combining Br K-edge X-ray absorption and UV-vis spectroscopies. The experimental data were analyzed through an integrated approach based on principal component analysis, multivariate curve resolution, and ab initio theoretical X-ray absorption spectroscopy (XAS), providing quantitative insights into the properties of the key reaction bromine species while contextually tracking the Fe2+ to Fe3+ oscillatory transformation. The high-quality XAS experimental data supported by the multivariate and theoretical analyses provide clear-cut evidence of the conversion of bromate, initially predominant in the reaction mixture, to the brominated derivative of the employed allylmalonic acid substrate. The described interdisciplinary method was proven to be valuable to monitor the fate of the main BZ reaction brominated species, which are silent to conventional spectroscopic methods of detection, and the developed approach may support future mechanistic investigations of other oscillatory systems.

A Graph Neural Network-Based Approach to XANES Data Analysis

The determination of three-dimensional structures (3D structures) is crucial for understanding the correlation between the structural attributes of materials and their functional performance. X-ray absorption near edge structure (XANES) is an indispensable tool to characterize the atomic-scale local 3D structure of the system. Here, we present an approach to simulate XANES based on a customized 3D graph neural network (3DGNN) model, XAS3Dabs, which takes directly the 3D structure of the system as input, and the inherent relation between the fine structure of spectrum and local geometry is considered during the model construction. It turns out to be faster than the traditional XANES fitting method when the simulation approach and XANES optimization algorithm are combined to fit the 3D structure of the given system. The geometric features of the system are included in the weighted message passing block of XAS3Dabs and their importance is investigated. XAS3Dabs model demonstrates superior accuracy in XANES prediction compared to most machine learning models. By extracting graphs constituted by edges related to the absorbing atom, our model reduces redundant information, thereby not only enhancing the model's performance but also improving its robustness across different hyperparameters. XAS3Dabs model can be generalized to simulate the spectra for the systems with the absorber having the designed absorption edge so as to meet the expectations of online data processing. The method is expected to be the key part of the online 3D structure analysis framework for the XAS-related beamlines of high-energy photon source (HEPS) now under construction.

Progress and challenges in structural, in situ and operando characterization of single-atom catalysts by X-ray based synchrotron radiation techniques

Single-atom catalysts (SACs) represent the ultimate size limit of nanoscale catalysts, combining the advantages of homogeneous and heterogeneous catalysts. SACs have isolated single-atom active sites that exhibit high atomic utilization efficiency, unique catalytic activity, and selectivity. Over the past few decades, synchrotron radiation techniques have played a crucial role in studying single-atom catalysis by identifying catalyst structures and enabling the understanding of reaction mechanisms. The profound comprehension of spectroscopic techniques and characteristics pertaining to SACs is important for exploring their catalytic activity origins and devising high-performance and stable SACs for industrial applications. In this review, we provide a comprehensive overview of the recent advances in X-ray based synchrotron radiation techniques for structural characterization and in situ/operando observation of SACs under reaction conditions. We emphasize the correlation between spectral fine features and structural characteristics of SACs, along with their analytical limitations. The development of IMST with spatial and temporal resolution is also discussed along with their significance in revealing the structural characteristics and reaction mechanisms of SACs. Additionally, this review explores the study of active center states using spectral fine characteristics combined with theoretical simulations, as well as spectroscopic analysis strategies utilizing machine learning methods to address challenges posed by atomic distribution inhomogeneity in SACs while envisaging potential applications integrating artificial intelligence seamlessly with experiments for real-time monitoring of single-atom catalytic processes.

Rational partitioning of spectral feature space for effective clustering of massive spectral image data

We have successfully proposed and demonstrated a clustering method that overcomes the "needle-in-a-haystack problem" (finding minuscule important regions from massive spectral image data sets). The needle-in-a-haystack problem is of central importance in the characterization of materials since in bulk materials, the properties of a very tiny region often dominate the entire function. To solve this problem, we propose that rational partitioning of the spectral feature space in which spectra are distributed, or defining of the decision boundaries for clustering, can be performed by focusing on the discrimination limit defined by the measurement noise and partitioning the space at intervals of this limit. We verified the proposed method, applied it to actual measurement data, and succeeded in detecting tiny (~ 0.5%) important regions that were difficult for human researchers and other machine learning methods to detect in discovering unknown phases. The ability to detect these crucial regions helps in understanding materials and designing more functional materials.

Coupled X-ray Absorption/UV-vis Monitoring of a Prototypical Oscillating Reaction

Oscillating reactions are among the most intriguing phenomena in chemistry, but many questions on their mechanisms still remain unanswered, due to their intrinsic complexity and to the low sensitivity of the most common spectroscopic techniques toward the reaction brominated species. In this work, we investigate the cerium ion-catalyzed Belousov-Zhabotinsky (BZ) oscillating reaction by means of time-resolved X-ray absorption spectroscopy (XAS), in combination with UV-vis spectroscopy and unsupervised machine learning, multivariate curve resolution, and kinetic analyses. Altogether, we provide new insights into the collective oscillatory behavior of the key brominated species involved in the classical BZ reaction and measure previously unreported oscillations in their concentrations through Br K-edge XAS, while simultaneously tracking the oscillatory Ce4+-to-Ce3+ transformation by coupling XAS with UV-vis spectroscopy. Our work evidences the potential of the XAS technique to investigate the mechanisms of oscillatory chemical systems whose species are often not detectable with conventional experimental methods.

Artificial neural network for deciphering the structural transformation of condensed ZnO by extended x-ray absorption fine structure spectroscopy

Pressure-induced structural phase transitions play a pivotal role in unlocking novel material functionalities and facilitating innovations in materials science. Nonetheless, unveiling the mechanisms of densification, which relies heavily on precise and comprehensive structural analysis, remains a challenge. Herein, we investigated the archetypalB4 →B1 phase transition pathway in ZnO by combining x-ray absorption fine structure (XAFS) spectroscopy with machine learning. Specifically, we developed an artificial neural network (NN) to decipher the extended-XAFS spectra by reconstructing the partial radial distribution functions of Zn-O/Zn pairs. This provided us with access to the evolution of the structural statistics for all the coordination shells in condensed ZnO, enabling us to accurately track the changes in the internal structural parameteruand the anharmonic effect. We observed a clear decrease inuand an increased anharmonicity near the onset of theB4 →B1 phase transition, indicating a preference for the iT phase as the intermediate state to initiate the phase transition that can arise from the softening of shear phonon modes. This study suggests that NN-based approach can facilitate a more comprehensive and efficient interpretation of XAFS under complexin-situconditions, which paves the way for highly automated data processing pipelines for high-throughput and real-time characterizations in next-generation synchrotron photon sources.

Machine learning for analysis of experimental scattering and spectroscopy data in materials chemistry

The rapid growth of materials chemistry data, driven by advancements in large-scale radiation facilities as well as laboratory instruments, has outpaced conventional data analysis and modelling methods, which can require enormous manual effort. To address this bottleneck, we investigate the application of supervised and unsupervised machine learning (ML) techniques for scattering and spectroscopy data analysis in materials chemistry research. Our perspective focuses on ML applications in powder diffraction (PD), pair distribution function (PDF), small-angle scattering (SAS), inelastic neutron scattering (INS), and X-ray absorption spectroscopy (XAS) data, but the lessons that we learn are generally applicable across materials chemistry. We review the ability of ML to identify physical and structural models and extract information efficiently and accurately from experimental data. Furthermore, we discuss the challenges associated with supervised ML and highlight how unsupervised ML can mitigate these limitations, thus enhancing experimental materials chemistry data analysis. Our perspective emphasises the transformative potential of ML in materials chemistry characterisation and identifies promising directions for future applications. The perspective aims to guide newcomers to ML-based experimental data analysis.

Localization of quantum walks with classical randomness: Comparison between manual methods and supervised machine learning

A transition of quantum walk induced by classical randomness changes the probability distribution of the walker from a two-peak structure to a single-peak one when the random parameter exceeds a critical value. We first establish the generality of the localization by showing its emergence in the presence of random rotation or translation. The transition point can be located manually by examining the probability distribution, momentum of inertia, and inverse participation ratio. As a comparison, we implement three supervised machine learning methods, the support vector machine (SVM), multilayer perceptron neural network, and convolutional neural network with the same data and show they are able to identify the transition. While the SVM sometimes underestimates the exponents compared to the manual methods, the two neural-network methods show more deviations for the case with random translation due to the fluctuating probability distributions. Our work illustrates potentials and challenges facing machine learning of physical systems with mixed quantum and classical probabilities.

Tracking the Evolution of Single-Atom Catalysts for the CO2 Electrocatalytic Reduction Using Operando X-ray Absorption Spectroscopy and Machine Learning

Transition metal-nitrogen-doped carbons (TMNCs) are a promising class of catalysts for the CO2 electrochemical reduction reaction. In particular, high CO2-to-CO conversion activities and selectivities were demonstrated for Ni-based TMNCs. Nonetheless, open questions remain about the nature, stability, and evolution of the Ni active sites during the reaction. In this work, we address this issue by combining operando X-ray absorption spectroscopy with advanced data analysis. In particular, we show that the combination of unsupervised and supervised machine learning approaches is able to decipher the X-ray absorption near edge structure (XANES) of the TMNCs, disentangling the contributions of different metal sites coexisting in the working TMNC catalyst. Moreover, quantitative structural information about the local environment of active species, including their interaction with adsorbates, has been obtained, shedding light on the complex dynamic mechanism of the CO2 electroreduction.

Fast extraction of three-dimensional nanofiber orientation from WAXD patterns using machine learning

Structural disclosure of biological materials can help our understanding of design disciplines in nature and inspire research for artificial materials. Synchrotron microfocus X-ray diffraction is one of the main techniques for characterizing hierarchically structured biological materials, especially the 3D orientation distribution of their interpenetrating nanofiber networks. However, extraction of 3D fiber orientation from X-ray patterns is still carried out by iterative parametric fitting, with disadvantages of time consumption and demand for expertise and initial parameter estimates. When faced with high-throughput experiments, existing analysis methods cannot meet the real time analysis challenges. In this work, using the assumption that the X-ray illuminated volume is dominated by two groups of nanofibers in a gradient biological composite, a machine-learning based method is proposed for fast and automatic fiber orientation metrics prediction from synchrotron X-ray micro-focused diffraction data. The simulated data were corrupted in the training procedure to guarantee the prediction ability of the trained machine-learning algorithm in real-world experimental data predictions. Label transformation was used to resolve the jump discontinuity problem when predicting angle parameters. The proposed method shows promise for application in the automatic data-processing pipeline for fast analysis of the vast data generated from multiscale diffraction-based tomography characterization of textured biomaterials.

Towards structural reconstruction from X-ray spectra

We report a statistical analysis of Ge K-edge X-ray emission spectra simulated for amorphous GeO₂ at elevated pressures. We find that employing machine learning approaches we can reliably predict the statistical moments of the Kβ'' and Kβ₂ peaks in the spectrum from the Coulomb matrix descriptor with a training set of ∼ 10⁴ samples. Spectral-significance-guided dimensionality reduction techniques allow us to construct an approximate inverse mapping from spectral moments to pseudo-Coulomb matrices. When applying this to the moments of the ensemble-mean spectrum, we obtain distances from the active site that match closely to those of the ensemble mean and which moreover reproduce the pressure-induced coordination change in amorphous GeO₂. With this approach utilizing emulator-based component analysis, we are able to filter out the artificially complete structural information available from simulated snapshots, and quantitatively analyse structural changes that can be inferred from the changes in the Kβ emission spectrum alone.

Deciphering the Structural and Chemical Transformations of Oxide Catalysts during Oxygen Evolution Reaction Using Quick X-ray Absorption Spectroscopy and Machine Learning

Bimetallic transition-metal oxides, such as spinel-like CoxFe3-xO4 materials, are known as attractive catalysts for the oxygen evolution reaction (OER) in alkaline electrolytes. Nonetheless, unveiling the real active species and active states in these catalysts remains a challenge. The coexistence of metal ions in different chemical states and in different chemical environments, including disordered X-ray amorphous phases that all evolve under reaction conditions, hinders the application of common operando techniques. Here, we address this issue by relying on operando quick X-ray absorption fine structure spectroscopy, coupled with unsupervised and supervised machine learning methods. We use principal component analysis to understand the subtle changes in the X-ray absorption near-edge structure spectra and develop an artificial neural network to decipher the extended X-ray absorption fine structure spectra. This allows us to separately track the evolution of tetrahedrally and octahedrally coordinated species and to disentangle the chemical changes and several phase transitions taking place in CoxFe3-xO4 catalysts and on their active surface, related to the conversion of disordered oxides into spinel-like structures, transformation of spinels into active oxyhydroxides, and changes in the degree of spinel inversion in the course of the activation treatment and under OER conditions. By correlating the revealed structural changes with the distinct catalytic activity for a series of CoxFe3-xO4 samples, we elucidate the active species and OER mechanism.

Strategy to Probe the Local Atomic Structure of Luminescent Rare Earth Complexes by X-ray Absorption Near-Edge Spectroscopy Simulation Using a Machine Learning-Based PyFitIt Approach

Rare earth(III) β-diketonates are highly remarkable luminophores in the visible spectral region among the rare earth compounds, owing to the efficient contribution from the 4f-4f intraconfigurational transitions. To get detailed structural insight into the RE³⁺ sites (RE = Eu, Gd, and Sm), X-ray absorption near-edge spectroscopy (XANES) can be very potent in probing the local chemical environment around the RE³⁺ ion. In this work, a PyFitIt machine learning approach was employed as a new strategy to simulate the Eu, Gd, and Sm L₃-edge XANES and thereby determine the local atomic structure of the luminescence RE³⁺ β-diketonate complexes, [Eu(tta)₃(H₂O)₂], [C₄mim][Eu(dbm)₄], [Gd(tta)₃(H₂O)₂], and [Sm(dbm)₃(phen)] (tta, 3-thenoyltrifluoroacetonate; dbm, dibenzoylmethane; phen, phenanthroline; and C₄mim, 1-butyl-3-methylimidazolium bromide). Continuous Cauchy wavelet transform validated the PyFitIt calculated XANES by visualizing very efficiently the coordination geometries, composed of O and O/N backscatterers around the RE³⁺ (RE = Eu and Gd) and Sm³⁺ ions, respectively, as a pinkish-red color map in the two-dimensional images of the corresponding complexes. Extended X-ray absorption fine structure fit in Artemis also corroborated the three-dimensional structures generated by PyFitIt XANES simulation for all the compounds. Though, relatively slightly higher bond distance values for the Sm³⁺ complex are due to the higher atomic radius of the Sm³⁺ ion when compared to the Eu³⁺ and Gd³⁺ complexes. Meanwhile, higher Debye-Waller factor (σ²) values for the [C₄mim][Eu(dbm)₄] when compared to the [Eu(tta)₃(H₂O)₂] indicated the structure disorder, owing to the distortion in the local geometry. It is noteworthy that the optical properties, described mainly by the Ω_λ (λ = 2 and 4) 4f-4f intensity parameters, are very sensitive to the local coordination environment around the Eu³⁺ ion. Thus, a close agreement between the experimental and theoretically calculated Ω_λ parameter values confirmed that the PyFitIt calculated square antiprismatic structures are precisely similar to the real structures of the Eu³⁺ complexes.

Design of Single-Atom Catalysts and Tracking Their Fate Using Operando and Advanced X-ray Spectroscopic Tools

The potential of operando X-ray techniques for following the structure, fate, and active site of single-atom catalysts (SACs) is highlighted with emphasis on a synergetic approach of both topics. X-ray absorption spectroscopy (XAS) and related X-ray techniques have become fascinating tools to characterize solids and they can be applied to almost all the transition metals deriving information about the symmetry, oxidation state, local coordination, and many more structural and electronic properties. SACs, a newly coined concept, recently gained much attention in the field of heterogeneous catalysis. In this way, one can achieve a minimum use of the metal, theoretically highest efficiency, and the design of only one active site-so-called single site catalysts. While single sites are not easy to characterize especially under operating conditions, XAS as local probe together with complementary methods (infrared spectroscopy, electron microscopy) is ideal in this research area to prove the structure of these sites and the dynamic changes during reaction. In this review, starting from their fundamentals, various techniques related to conventional XAS and X-ray photon in/out techniques applied to single sites are discussed with detailed mechanistic and in situ/operando studies. We systematically summarize the design strategies of SACs and outline their exploration with XAS supported by density functional theory (DFT) calculations and recent machine learning tools.

Transferable Deep Learning Potential Reveals Intermediate-Range Ordering Effects in LiF-NaF-ZrF4 Molten Salt

LiF-NaF-ZrF₄ multicomponent molten salts are promising candidate coolants for advanced clean energy systems owing to their desirable thermophysical and transport properties. However, the complex structures enabling these properties, and their dependence on composition, is scarcely quantified due to limitations in simulating and interpreting experimental spectra of highly disordered, intermediate-ranged structures. Specifically, size-limited ab initio simulations and accuracy-limited classical models used in the past are unable to capture a wide range of fluctuating motifs found in the extended heterogeneous structures of liquid salt. This greatly inhibits our ability to design tailored compositions and materials. Here, accurate, efficient, and transferable machine learning potentials are used to predict structures far beyond the first coordination shell in LiF-NaF-ZrF₄. Neural networks trained at only eutectic compositions with 29% and 37% ZrF₄ are shown to accurately simulate a wide range of compositions (11-40% ZrF₄) with dramatically different coordination chemistries, while showing a remarkable agreement with theoretical and experimental Raman spectra. The theoretical Raman calculations further uncovered the previously unseen shift and flattening of bending band at ∼250 cm^-1 which validated the simulated extended-range structures as observed in compositions with higher than 29% ZrF₄ content. In such cases, machine learning-based simulations capable of accessing larger time and length scales (beyond 17 Å) were critical for accurately predicting both structure and ionic diffusivities.

Machine learning for a sustainable energy future

Transitioning from fossil fuels to renewable energy sources is a critical global challenge; it demands advances - at the materials, devices and systems levels - for the efficient harvesting, storage, conversion and management of renewable energy. Energy researchers have begun to incorporate machine learning (ML) techniques to accelerate these advances. In this Perspective, we highlight recent advances in ML-driven energy research, outline current and future challenges, and describe what is required to make the best use of ML techniques. We introduce a set of key performance indicators with which to compare the benefits of different ML-accelerated workflows for energy research. We discuss and evaluate the latest advances in applying ML to the development of energy harvesting (photovoltaics), storage (batteries), conversion (electrocatalysis) and management (smart grids). Finally, we offer an overview of potential research areas in the energy field that stand to benefit further from the application of ML.

Informed Chemical Classification of Organophosphorus Compounds via Unsupervised Machine Learning of X-ray Absorption Spectroscopy and X-ray Emission Spectroscopy

We analyze an ensemble of organophosphorus compounds to form an unbiased characterization of the information encoded in their X-ray absorption near-edge structure (XANES) and valence-to-core X-ray emission spectra (VtC-XES). Data-driven emergence of chemical classes via unsupervised machine learning, specifically cluster analysis in the Uniform Manifold Approximation and Projection (UMAP) embedding, finds spectral sensitivity to coordination, oxidation, aromaticity, intramolecular hydrogen bonding, and ligand identity. Subsequently, we implement supervised machine learning via Gaussian process classifiers to identify confidence in predictions that match our initial qualitative assessments of clustering. The results further support the benefit of utilizing unsupervised machine learning as a precursor to supervised machine learning, which we term Unsupervised Validation of Classes (UVC), a result that goes beyond the present case of X-ray spectroscopies.

Emulator-based decomposition for structural sensitivity of core-level spectra

We explore the sensitivity of several core-level spectroscopic methods to the underlying atomistic structure by using the water molecule as our test system. We first define a metric that measures the magnitude of spectral change as a function of the structure, which allows for identifying structural regions with high spectral sensitivity. We then apply machine-learning-emulator-based decomposition of the structural parameter space for maximal explained spectral variance, first on overall spectral profile and then on chosen integrated regions of interest therein. The presented method recovers more spectral variance than partial least-squares fitting and the observed behaviour is well in line with the aforementioned metric for spectral sensitivity. The analysis method is able to independently identify spectroscopically dominant degrees of freedom, and to quantify their effect and significance.

Theoretical Perspective on Operando Spectroscopy of Fluxional Nanocatalysts

Improvements in operando spectroscopy have enabled the catalysis community to investigate the dynamic nature of catalysts under operating conditions with increasing detail. Still, the highly dynamic nature of some catalysts, such as fluxional supported subnano clusters, presents a formidable challenge even for the most state-of-the-art techniques. The reason is that such fluxional catalytic interfaces contain a variety of thermally accessible states. Operando spectroscopies used in catalysis generally fall into two categories: ensemble-based techniques, which provide spectra containing the signals of the entire ensemble of states of the catalyst and are not necessarily dominated by the most active species, and localized techniques, which provide atomistic-level information about the dynamics of active sites in a very small area, which might not include the most active species. Combining many different kinds of techniques can provide detailed insight; however, we propose that effective utilization of specific computational techniques and approaches within the fluxionality paradigm can fill the gap and enable atomistic characterization of the most relevant catalytic sites.

Tracking heterogeneous structural motifs and the redox behaviour of copper-zinc nanocatalysts for the electrocatalytic CO2 reduction using operando time resolved spectroscopy and machine learning

Copper-based catalysts are established catalytic systems for the electrocatalytic CO₂ reduction reaction (CO₂RR), where the greenhouse gas CO₂ is converted into valuable industrial chemicals, such as energy-dense C₂₊ products, using energy from renewable sources. However, better control over the catalyst selectivity, especially at industrially relevant high current density conditions, is needed to expedite the economic viability of the CO₂RR. For this purpose, bimetallic materials, where copper is combined with a secondary metal, comprise a promising and a highly tunable catalyst for the CO₂RR. Nevertheless, the synergy between copper and the selected secondary metal species, the evolution of the bimetallic structural motifs under working conditions and the effect of the secondary metal on the kinetics of the Cu redox behavior require careful investigation. Here, we employ operando quick X-ray absorption fine structure (QXAFS) spectroscopy coupled with machine-learning based data analysis and surface-enhanced Raman spectroscopy (SERS) to investigate the time-dependent chemical and structural changes in catalysts derived from shape-selected ZnO/Cu₂O nanocubes under CO₂RR conditions at current densities up to −500 mA cm⁻². We furthermore relate the catalyst transformations observed under working conditions to the catalytic activity and selectivity and correlate potential-dependent surface and subsurface processes. We report that the addition of Zn to a Cu-based catalyst has a crucial impact on the kinetics of subsurface processes, while redox processes of the Cu surface layer remain largely unaffected. Interestingly, the presence of Zn was found to contribute to the stabilization of cationic Cu(i) species, which is of catalytic relevance since Cu(0)/Cu(i) interfaces have been reported to be beneficial for efficient electrocatalytic CO₂ conversion to complex multicarbon products. At the same time, we attribute the increase of the C₂₊ product selectivity to the formation of Cu-rich CuZn alloys in samples with low Zn content, while Zn-rich alloy phases result in an increased formation of CO paralleled by an increase of the parasitic hydrogen evolution reaction.

Elucidating the role of Zn in bimetallic CuZn nanocatalysts for the electrocatalytic CO₂ reduction reaction (CO₂RR), where the greenhouse gas CO₂ is converted into valuable industrial chemicals using energy from renewable sources.

Automatic determination of the spectrum-structure relationship by tree structure-based unsupervised and supervised learning

Spectroscopy is widely used for the analysis of chemical, vibrational, and bonding information. Interpretations of the spectral features have been performed by comparing the objective spectra with reference spectra from experiments or simulations. However, the interpretation process by humans is not always straightforward, especially for spectra obtained from unknown or new materials. In the present study, we developed a method using machine learning techniques to obtain human-like interpretation automatically. We combined unsupervised and supervised learning methods; then applied it to the spectrum database which includes more than 400 spectra of water and organic molecules containing various ligands and chemical bonds. The proposed method has successfully found the correlations between the spectral features and descriptors of the atoms, bonds, and ligands. We demonstrated that the proposed method enabled the automatic determination of reasonable spectrum-structure relationships such as between π* resonance in C-K edges and multiple bonds. The proposed method enables the automatic determination of physically and chemically reasonable spectrum-structure relationships without arbitrariness in data-driven manner, which is considerably difficult only with simulation or conventional machine leaning techniques. Such relationships are useful for understanding what structural parameters cause changes in the spectrum, providing a way for the better interpretation of spatial distributed or time evolutionary data. Furthermore, although the present work focused on the ELNES/XANES spectrum from small organic molecules, the proposed method can be readily extended to other spectral data. It is expected to contribute to a better understanding of the spectrum-structure relationship in various spectroscopy applications.

Adsorption Sites on Pd Nanoparticles Unraveled by Machine-Learning Potential with Adaptive Sampling

Catalytic properties of noble-metal nanoparticles (NPs) are largely determined by their surface morphology. The latter is probed by surface-sensitive spectroscopic techniques in different spectra regions. A fast and precise computational approach enabling the prediction of surface-adsorbate interaction would help the reliable description and interpretation of experimental data. In this work, we applied Machine Learning (ML) algorithms for the task of adsorption-energy approximation for CO on Pd nanoclusters. Due to a high dependency of binding energy from the nature of the adsorbing site and its local coordination, we tested several structural descriptors for the ML algorithm, including mean Pd-C distances, coordination numbers (CN) and generalized coordination numbers (GCN), radial distribution functions (RDF), and angular distribution functions (ADF). To avoid overtraining and to probe the most relevant positions above the metal surface, we utilized the adaptive sampling methodology for guiding the ab initio Density Functional Theory (DFT) calculations. The support vector machines (SVM) and Extra Trees algorithms provided the best approximation quality and mean absolute error in energy prediction up to 0.12 eV. Based on the developed potential, we constructed an energy-surface 3D map for the whole Pd₅₅ nanocluster and extended it to new geometries, Pd₇₉, and Pd₈₅, not implemented in the training sample. The methodology can be easily extended to adsorption energies onto mono- and bimetallic NPs at an affordable computational cost and accuracy.

Unsupervised machine learning for unbiased chemical classification in X-ray absorption spectroscopy and X-ray emission spectroscopy

We report a comprehensive computational study of unsupervised machine learning for extraction of chemically relevant information in X-ray absorption near edge structure (XANES) and in valence-to-core X-ray emission spectra (VtC-XES) for classification of a broad ensemble of sulphorganic molecules. By progressively decreasing the constraining assumptions of the unsupervised machine learning algorithm, moving from principal component analysis (PCA) to a variational autoencoder (VAE) to t-distributed stochastic neighbour embedding (t-SNE), we find improved sensitivity to steadily more refined chemical information. Surprisingly, when embedding the ensemble of spectra in merely two dimensions, t-SNE distinguishes not just oxidation state and general sulphur bonding environment but also the aromaticity of the bonding radical group with 87% accuracy as well as identifying even finer details in electronic structure within aromatic or aliphatic sub-classes. We find that the chemical information in XANES and VtC-XES is very similar in character and content, although they unexpectedly have different sensitivity within a given molecular class. We also discuss likely benefits from further effort with unsupervised machine learning and from the interplay between supervised and unsupervised machine learning for X-ray spectroscopies. Our overall results, i.e., the ability to reliably classify without user bias and to discover unexpected chemical signatures for XANES and VtC-XES, likely generalize to other systems as well as to other one-dimensional chemical spectroscopies.

A Holistic Approach for Elucidating Local Structure, Dynamics, and Speciation in Molten Salts with High Structural Disorder

To examine ion solvation, exchange, and speciation for minority components in molten salts (MS) typically found as corrosion products, we propose a multimodal approach combining extended X-ray absorption fine structure (EXAFS) spectroscopy, optical spectroscopy, ab initio molecular dynamics (AIMD) simulations, and rate theory of ion exchange. Going beyond conventional EXAFS analysis, our method can accurately quantify populations of different coordination states of ions with highly disordered coordination environments via linear combination fitting of the EXAFS spectra of these coordination states computed from AIMD to the experimental EXAFS spectrum. In a case study of dilute Ni(II) dissolved in the ZnCl₂+KCl melts, our method reveals heterogeneous distributions of coordination states of Ni(II) that are sensitive to variations in temperature and melt composition. These results are fully explained by the difference in the chloride exchange dynamics at varied temperatures and melt compositions. This insight will enable a better understanding and control of ion solubility and transport in MS.

Physically inspired deep learning of molecular excitations and photoemission spectra

Modern functional materials consist of large molecular building blocks with significant chemical complexity which limits spectroscopic property prediction with accurate first-principles methods. Consequently, a targeted design of materials with tailored optoelectronic properties by high-throughput screening is bound to fail without efficient methods to predict molecular excited-state properties across chemical space. In this work, we present a deep neural network that predicts charged quasiparticle excitations for large and complex organic molecules with a rich elemental diversity and a size well out of reach of accurate many body perturbation theory calculations. The model exploits the fundamental underlying physics of molecular resonances as eigenvalues of a latent Hamiltonian matrix and is thus able to accurately describe multiple resonances simultaneously. The performance of this model is demonstrated for a range of organic molecules across chemical composition space and configuration space. We further showcase the model capabilities by predicting photoemission spectra at the level of the GW approximation for previously unseen conjugated molecules.

Machine learning powered by principal component descriptors as the key for sorted structural fit of XANES

Modern synchrotron radiation sources and free electron laser made X-ray absorption spectroscopy (XAS) an analytical tool for the structural analysis of materials under in situ or operando conditions. Fourier approach applied to the extended region of the XAS spectrum (EXAFS) allows the estimation of the number of structural and non-structural parameters which can be refined through a fitting procedure. The near edge region of the XAS spectrum (XANES) is also sensitive to the coordinates of all the atoms in the local cluster around the absorbing atom. However, in contrast to EXAFS, the existing approaches of quantitative analysis provide no estimation for the number of structural parameters that can be evaluated for a given XANES spectrum. This problem exists both for the classical gradient descent approaches and for modern machine learning methods based on neural networks. We developed a new approach for rational fit based on principal component descriptors of the spectrum. In this work the principal component analysis (PCA) is applied to a dataset of theoretical spectra calculated a priori on a grid of variable structural parameters of a molecule or cluster. Each principal component of the dataset is related then to a combined variation of several structural parameters, similar to the vibrational normal mode. Orthogonal principal components determine orthogonal deformations that can be extracted independently upon the analysis of the XANES spectrum. Applying statistical criteria, the PCA-based fit of the XANES determines the accessible structural information in the spectrum for a given system.

Revisiting the Extended X-ray Absorption Fine Structure Fitting Procedure through a Machine Learning-Based Approach

A novel approach for the analysis of extended X-ray absorption fine structure (EXAFS) spectra is developed exploiting an inverse machine learning-based algorithm. Through this approach, it is possible to explore and account for, in a precise way, the nonlinear geometry dependence of the photoelectron backscattering phases and amplitudes of single and multiple scattering paths. In addition, the determined parameters are directly related to the 3D atomic structure, without the need to use complex parametrization as in the classical fitting approach. The applicability of the approach, its potential and the advantages over the classical fit were demonstrated by fitting the EXAFS data of two molecular systems, namely, the KAu (CN)₂ and the [RuCl₂(CO)₃]₂ complexes.

Extracting Local Symmetry of Mono-Atomic Systems from Extended X-ray Absorption Fine Structure Using Deep Neural Networks

In recent years, neural networks have become a new method for the analysis of extended X-ray absorption fine structure data. Due to its sensitivity to local structure, X-ray absorption spectroscopy is often used to study disordered systems and one of its more interesting property is the sensitivity not only to pair distribution function, but also to three-body distribution, which contains information on the local symmetry. In this study, by considering the case of Ni, we show that by using neural networks, it is possible to obtain not only the radial distribution function, but also the bond angle distribution between the first nearest-neighbors. Additionally, by adding appropriate configurations in the dataset used for training, we show that the neural network is able to analyze also data from disordered phases (liquid and undercooled state), detecting small changes in the local ordering compatible with results obtained through other methods.

Technetium immobilization by chukanovite and its oxidative transformation products: Neural network analysis of EXAFS spectra

The uptake of the fission product technetium (Tc) by chukanovite, an Fe^II hydroxy carbonate mineral formed as a carbon steel corrosion product in anoxic and carbonate-rich environments, was studied under anoxic, alkaline to hyperalkaline conditions representative for nuclear waste repositories in deep geological formations with cement-based inner linings. The retention potential of chukanovite towards Tc^VII is high in the pH range 7.8 to 12.6, evidenced by high solid-water distribution coefficients, log R_d ~ 6, and independent of ionic strength (0.1 or 1 M NaCl). Using Tc K-edge X-ray absorption spectroscopy (XAS) two series of samples were investigated, Tc chukanovite sorption samples and coprecipitates, prepared with varying Tc loadings, pH values and contact times. From the resulting 37 XAS spectra, spectral endmembers and their dependence on chemical parameters were derived by self-organizing (Kohonen) maps (SOM), a neural network-based approach of machine learning. X-ray absorption near-edge structure (XANES) data confirmed the complete reduction of Tc^VII to Tc^IV by chukanovite under all experimental conditions. Consistent with mineralogical phases identified by X-ray diffraction (XRD), SOM analysis of the extended X-ray absorption fine-structure (EXAFS) spectra revealed the presence of three species in the sorption samples, the speciation predominately controlled by pH: Between pH 7.8 and 11.8, TcO₂-dimers form inner-sphere sorption complexes at the surface of the initial chukanovite as well as on the surface of secondary magnetite formed due to redox reaction. At pH ≥ 11.9, Tc^IV is incorporated in a mixed, chukanovite-like, Fe/Tc hydroxy carbonate precipitate. The same species formed when using the coprecipitation approach. Reoxidation of sorption samples resulted in a small remobilization of Tc, demonstrating that both the original chukanovite mineral and its oxidative transformation products, magnetite and goethite, contribute to the immobilization of Tc in the long term, thus strongly attenuating its environmental transport.

Local electronic structure rearrangements and strong anharmonicity in YH3 under pressures up to 180 GPa

The discovery of superconductivity above 250 K at high pressure in LaH₁₀ and the prediction of overcoming the room temperature threshold for superconductivity in YH₁₀ urge for a better understanding of hydrogen interaction mechanisms with the heavy atom sublattice in metal hydrides under high pressure at the atomic scale. Here we use locally sensitive X-ray absorption fine structure spectroscopy (XAFS) to get insight into the nature of phase transitions and the rearrangements of local electronic and crystal structure in archetypal metal hydride YH₃ under pressure up to 180 GPa. The combination of the experimental methods allowed us to implement a multiscale length study of YH₃: XAFS (short-range), Raman scattering (medium-range) and XRD (long-range). XANES data evidence a strong effect of hydrogen on the density of 4d yttrium states that increases with pressure and EXAFS data evidence a strong anharmonicity, manifested as yttrium atom vibrations in a double-well potential.

Latent Representation Learning for Structural Characterization of Catalysts

Supervised machine learning-enabled mapping of the X-ray absorption near edge structure (XANES) spectra to local structural descriptors offers new methods for understanding the structure and function of working nanocatalysts. We briefly summarize a status of XANES analysis approaches by supervised machine learning methods. We present an example of an autoencoder-based, unsupervised machine learning approach for latent representation learning of XANES spectra. This new approach produces a lower-dimensional latent representation, which retains a spectrum-structure relationship that can be eventually mapped to physicochemical properties. The latent space of the autoencoder also provides a pathway to interpret the information content "hidden" in the X-ray absorption coefficient. Our approach (that we named latent space analysis of spectra, or LSAS) is demonstrated for the supported Pd nanoparticle catalyst studied during the formation of Pd hydride. By employing the low-dimensional representation of Pd K-edge XANES, the LSAS method was able to isolate the key factors responsible for the observed spectral changes.

In Situ/Operando Electrocatalyst Characterization by X-ray Absorption Spectroscopy

During the last decades, X-ray absorption spectroscopy (XAS) has become an indispensable method for probing the structure and composition of heterogeneous catalysts, revealing the nature of the active sites and establishing links between structural motifs in a catalyst, local electronic structure, and catalytic properties. Here we discuss the fundamental principles of the XAS method and describe the progress in the instrumentation and data analysis approaches undertaken for deciphering X-ray absorption near edge structure (XANES) and extended X-ray absorption fine structure (EXAFS) spectra. Recent usages of XAS in the field of heterogeneous catalysis, with emphasis on examples concerning electrocatalysis, will be presented. The latter is a rapidly developing field with immense industrial applications but also unique challenges in terms of the experimental characterization restrictions and advanced modeling approaches required. This review will highlight the new insight that can be gained with XAS on complex real-world electrocatalysts including their working mechanisms and the dynamic processes taking place in the course of a chemical reaction. More specifically, we will discuss applications of in situ and operando XAS to probe the catalyst's interactions with the environment (support, electrolyte, ligands, adsorbates, reaction products, and intermediates) and its structural, chemical, and electronic transformations as it adapts to the reaction conditions.

Neural Network Model for Predicting Student Failure in the Academic Leveling Course of Escuela Politécnica Nacional

The purpose of this study is to train an artificial neural network model for predicting student failure in the academic leveling course of the Escuela Politécnica Nacional of Ecuador, based on academic and socioeconomic information. For this, 1308 higher education students participated, 69.0% of whom failed the academic leveling course; besides, 93.7% of the students self-identified as mestizo, 83.9% came from the province of Pichincha, and 92.4% belonged to general population. As a first approximation, a neural network model was trained with twelve variables containing students’ academic and socioeconomic information. Then, a dimensionality reduction process was performed from which a new neural network was modeled. This dimension reduced model was trained with the variables application score, vulnerability index, regime, gender, and population segment, which were the five variables that explained more than 80% of the first model. The classification accuracy of the dimension reduced model was 0.745, while precision and recall were 0.883 and 0.778, respectively. The area under ROC curve was 0.791. This model could be used as a guide to lead intervention policies so that the failure rate in the academic leveling course would decrease.

Melting properties by X-ray absorption spectroscopy: common signatures in binary Fe-C, Fe-O, Fe-S and Fe-Si systems

X-ray absorption spectroscopy (XAS) is a widely used technique to probe the local environment around specific atomic species. Applied to samples under extreme pressure and temperature conditions, XAS is sensitive to phase transitions, including melting, and allows gathering insights on compositional variations and electronic changes occurring during such transitions. These characteristics can be exploited for studies of prime interest in geophysics and fundamental high-pressure physics. Here, we investigated the melting curve and the eutectic composition of four geophysically relevant iron binary systems: Fe-C, Fe-O, Fe-S and Fe-Si. Our results show that all these systems present the same spectroscopic signatures upon melting, common to those observed for other pure late 3d transition metals. The presented melting criterion seems to be general for late 3d metals bearing systems. Additionally, we demonstrate the suitability of XAS to extract melt compositional information in situ, such as the evolution of the concentration of light elements with increasing temperature. Diagnostics presented in this work can be applied to studies over an even larger pressure range exploiting the upgraded synchrotron machines, and directly transferred to time-resolved extreme condition studies using dynamic compression (ns) or fast laser heating (ms).

Deep learning networks for the recognition and quantitation of surface-enhanced Raman spectroscopy

Surface-enhanced Raman spectroscopy (SERS) based on machine learning methods has been applied in material analysis, biological detection, food safety, and intelligent analysis. However, machine learning methods generally require extra preprocessing or feature engineering, and handling large-scale data using these methods is challenging. In this study, deep learning networks were used as fully connected networks, convolutional neural networks (CNN), fully convolutional networks (FCN), and principal component analysis networks (PCANet) to determine their abilities to recognise drugs in human urine and measure pirimiphos-methyl in wheat extract in the two input forms of a one-dimensional vector or a two-dimensional matrix. The best recognition result for drugs in urine with an accuracy of 98.05% in the prediction set was obtained using CNN with spectra as input in the matrix form. The optimal quantitation for pirimiphos-methyl was obtained using FCN with spectra in the matrix form, and the analysis was accomplished with a determination coefficient of 0.9997 and a root mean square error of 0.1574 in the prediction set. These networks performed better than the common machine learning methods. Overall, the deep learning networks provide feasible alternatives for the recognition and quantitation of SERS.

Machine-Learning X-Ray Absorption Spectra to Quantitative Accuracy

Simulations of excited state properties, such as spectral functions, are often computationally expensive and therefore not suitable for high-throughput modeling. As a proof of principle, we demonstrate that graph-based neural networks can be used to predict the x-ray absorption near-edge structure spectra of molecules to quantitative accuracy. Specifically, the predicted spectra reproduce nearly all prominent peaks, with 90% of the predicted peak locations within 1 eV of the ground truth. Besides its own utility in spectral analysis and structure inference, our method can be combined with structure search algorithms to enable high-throughput spectrum sampling of the vast material configuration space, which opens up new pathways to material design and discovery.

Infrared spectroscopy data- and physics-driven machine learning for characterizing surface microstructure of complex materials

There is a need to characterize complex materials and their dynamics under reaction conditions to accelerate materials design. Adsorbate vibrational excitations are selective to adsorbate/surface interactions and infrared (IR) spectra associated with activating adsorbate vibrational modes are accurate, capture details of most modes, and can be obtained operando. Current interpretation depends on heuristic peak assignments for simple spectra, precluding the possibility of obtaining detailed structural information. Here, we combine data-based approaches with chemistry-dependent problem formulation to develop physics-driven surrogate models that generate synthetic IR spectra from first-principles calculations. Using synthetic IR spectra of carbon monoxide on platinum, we implement multinomial regression via neural network ensembles to learn probability distributions functions (pdfs) that describe adsorption sites and quantify uncertainty. We use these pdfs to infer detailed surface microstructure from experimental spectra and extend this methodology to other systems as a first step towards characterizing complex interfaces and closing the materials gap.

Linking the evolution of catalytic properties and structural changes in copper-zinc nanocatalysts using operando EXAFS and neural-networks

Understanding the evolution of unique structural motifs in bimetallic catalysts under reaction conditions, and linking them to the observed catalytic properties is necessary for the rational design of the next generation of catalytic materials. Extended X-ray absorption fine structure (EXAFS) spectroscopy is a premier experimental method to address this issue, providing the possibility to track the changes in the structure of working catalysts. Unfortunately, the intrinsic heterogeneity and enhanced disorder characteristic of catalytic materials experiencing structural transformations under reaction conditions, as well as the low signal-to-noise ratio that is common for in situ EXAFS spectra hinder the application of conventional data analysis approaches. Here we address this problem by employing machine learning methods (artificial neural networks) to establish the relationship between EXAFS features and structural motifs in metals as well as oxide materials. We apply this approach to time-dependent EXAFS spectra acquired from copper–zinc nanoparticles during the electrochemical reduction of CO₂ to reveal the details of the composition-dependent structural evolution and brass alloy formation, and their correlation with the catalytic selectivity of these materials.

A neural network is used to reveal composition-dependent structural evolution under operando conditions in CuZn nanocatalysts for CO₂ electroreduction.

Machine learning and big scientific data

This paper reviews some of the challenges posed by the huge growth of experimental data generated by the new generation of large-scale experiments at UK national facilities at the Rutherford Appleton Laboratory (RAL) site at Harwell near Oxford. Such 'Big Scientific Data' comes from the Diamond Light Source and Electron Microscopy Facilities, the ISIS Neutron and Muon Facility and the UK's Central Laser Facility. Increasingly, scientists are now required to use advanced machine learning and other AI technologies both to automate parts of the data pipeline and to help find new scientific discoveries in the analysis of their data. For commercially important applications, such as object recognition, natural language processing and automatic translation, deep learning has made dramatic breakthroughs. Google's DeepMind has now used the deep learning technology to develop their AlphaFold tool to make predictions for protein folding. Remarkably, it has been able to achieve some spectacular results for this specific scientific problem. Can deep learning be similarly transformative for other scientific problems? After a brief review of some initial applications of machine learning at the RAL, we focus on challenges and opportunities for AI in advancing materials science. Finally, we discuss the importance of developing some realistic machine learning benchmarks using Big Scientific Data coming from several different scientific domains. We conclude with some initial examples of our 'scientific machine learning' benchmark suite and of the research challenges these benchmarks will enable. This article is part of a discussion meeting issue 'Numerical algorithms for high-performance computational science'.

Machine learning approaches for ELNES/XANES

Materials characterization is indispensable for materials development. In particular, spectroscopy provides atomic configuration, chemical bonding and vibrational information, which are crucial for understanding the mechanism underlying the functions of a material. Despite its importance, the interpretation of spectra using human-driven methods, such as manual comparison of experimental spectra with reference/simulated spectra, is becoming difficult owing to the rapid increase in experimental spectral data. To overcome the limitations of such methods, we develop new data-driven approaches based on machine learning. Specifically, we use hierarchical clustering, a decision tree and a feedforward neural network to investigate the electron energy loss near edge structures (ELNES) spectrum, which is identical to the X-ray absorption near edge structure (XANES) spectrum. Hierarchical clustering and the decision tree are used to interpret and predict ELNES/XANES, while the feedforward neural network is used to obtain hidden information about the material structure and properties from the spectra. Further, we construct a prediction model that is robust against noise by data augmentation. Finally, we apply our method to noisy spectra and predict six properties accurately. In summary, the proposed approaches can pave the way for fast and accurate spectrum interpretation/prediction as well as local measurement of material functions.

Neural network assisted analysis of bimetallic nanocatalysts using X-ray absorption near edge structure spectroscopy

Trained neural networks are used to extract the first partial coordination numbers from XANES spectra. In bimetallic nanoparticles, the four local structure descriptors provide rich information on structural motifs.

Data-Driven Materials Science: Status, Challenges, and Perspectives

Data-driven science is heralded as a new paradigm in materials science. In this field, data is the new resource, and knowledge is extracted from materials datasets that are too big or complex for traditional human reasoning-typically with the intent to discover new or improved materials or materials phenomena. Multiple factors, including the open science movement, national funding, and progress in information technology, have fueled its development. Such related tools as materials databases, machine learning, and high-throughput methods are now established as parts of the materials research toolset. However, there are a variety of challenges that impede progress in data-driven materials science: data veracity, integration of experimental and computational data, data longevity, standardization, and the gap between industrial interests and academic efforts. In this perspective article, the historical development and current state of data-driven materials science, building from the early evolution of open science to the rapid expansion of materials data infrastructures are discussed. Key successes and challenges so far are also reviewed, providing a perspective on the future development of the field.

Solving the Structure and Dynamics of Metal Nanoparticles by Combining X-Ray Absorption Fine Structure Spectroscopy and Atomistic Structure Simulations

Extended X-ray absorption fine structure (EXAFS) spectroscopy is a premiere method for analysis of the structure and structural transformation of nanoparticles. Extraction of analytical information about the three-dimensional structure and dynamics of metal-metal bonds from EXAFS spectra requires special care due to their markedly non-bulk-like character. In recent decades, significant progress has been made in the first-principles modeling of structure and properties of nanoparticles. In this review, we summarize new approaches for EXAFS data analysis that incorporate particle structure modeling into the process of structural refinement.

Deep Learning Spectroscopy: Neural Networks for Molecular Excitation Spectra

Deep learning methods for the prediction of molecular excitation spectra are presented. For the example of the electronic density of states of 132k organic molecules, three different neural network architectures: multilayer perceptron (MLP), convolutional neural network (CNN), and deep tensor neural network (DTNN) are trained and assessed. The inputs for the neural networks are the coordinates and charges of the constituent atoms of each molecule. Already, the MLP is able to learn spectra, but the root mean square error (RMSE) is still as high as 0.3 eV. The learning quality improves significantly for the CNN (RMSE = 0.23 eV) and reaches its best performance for the DTNN (RMSE = 0.19 eV). Both CNN and DTNN capture even small nuances in the spectral shape. In a showcase application of this method, the structures of 10k previously unseen organic molecules are scanned and instant spectra predictions are obtained to identify molecules for potential applications.