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.

Ultralow work function of the electride Sr3CrN3.

Electrides have valence electrons that occupy free space in the crystal structure, making them easier to extract. This feature can be used in catalysis for important reactions that usually require a high-temperature and high-pressure environments, such as ammonia synthesis. In this paper, we use density functional theory to investigate the behaviour of interstitial electrons of the 1-dimensional electride Sr3CrN3. We find that the bulk excess electron density persists on introduction of surface terminations, that the crystal termination perpendicular to the 1D free-electron channel is highly stable and we confirm an extremely low work function with hybrid functional methods. Our results indicate that Sr3CrN3 is a potentially important novel catalyst, with accessible, directional and extractable free electron density.

Origin of Ferroelectricity in Two Prototypical Hybrid Organic-Inorganic Perovskites.

Hybrid organic-inorganic perovskite (HOIP) ferroelectrics are attracting considerable interest because of their high performance, ease of synthesis, and lightweight. However, the intrinsic thermodynamic origins of their ferroelectric transitions remain insufficiently understood. Here, we identify the nature of the ferroelectric phase transitions in displacive [(CH3)2NH2][Mn(N3)3] and order-disorder type [(CH3)2NH2][Mn(HCOO)3] via spatially resolved structural analysis and ab initio lattice dynamics calculations. Our results demonstrate that the vibrational entropy change of the extended perovskite lattice drives the ferroelectric transition in the former and also contributes importantly to that of the latter along with the rotational entropy change of the A-site. This finding not only reveals the delicate atomic dynamics in ferroelectric HOIPs but also highlights that both the local and extended fluctuation of the hybrid perovskite lattice can be manipulated for creating ferroelectricity by taking advantages of their abundant atomic, electronic, and phononic degrees of freedom.

Tilt and shift polymorphism in molecular perovskites.

Molecular perovskites, i.e. ABX3 coordination polymers with a perovskite structure, are a chemically diverse material platform for studying fundamental and applied materials properties such as barocalorics and improper ferroelectrics. Compared to inorganic perovskites, the use of molecular ions on the A- and X-site of molecular perovskites leads to new geometric and structural degrees of freedom. In this work we introduce the concept of tilt and shift polymorphism, categorising irreversible perovskite-to-perovskite phase transitions in molecular perovskites. As a model example we study the new molecular perovskite series [(nPr)3(CH3)N]M(C2N3)3 with M = Mn2+, Co2+, Ni2+, and nPr = n-propyl, where different polymorphs crystallise in the perovskite structure but with different tilt systems depending on the synthetic conditions. Tilt and shift polymorphism is a direct ramification of the use of molecular building units in molecular perovskites and as such is unknown for inorganic perovskites. Given the role of polymorphism in materials science, medicine and mineralogy, and more generally the relation between physicochemical properties and structure, the concept introduced herein represents an important step in classifying the crystal chemistry of molecular perovskites and in maturing the field.

Cycling Rate-Induced Spatially-Resolved Heterogeneities in Commercial Cylindrical Li-Ion Batteries.

Synchrotron high-energy X-ray diffraction computed tomography has been employed to investigate, for the first time, commercial cylindrical Li-ion batteries electrochemically cycled over the two cycling rates of C/2 and C/20. This technique yields maps of the crystalline components and chemical species as a cross-section of the cell with high spatiotemporal resolution (550 × 550 images with 20 × 20 × 3 µm3 voxel size in ca. 1 h). The recently developed Direct Least-Squares Reconstruction algorithm is used to overcome the well-known parallax problem and led to accurate lattice parameter maps for the device cathode. Chemical heterogeneities are revealed at both electrodes and are attributed to uneven Li and current distributions in the cells. It is shown that this technique has the potential to become an invaluable diagnostic tool for real-world commercial batteries and for their characterization under operating conditions, leading to unique insights into "real" battery degradation mechanisms as they occur.

Entropy-based active learning of graph neural network surrogate models for materials properties.

Graph neural networks trained on experimental or calculated data are becoming an increasingly important tool in computational materials science. Networks once trained are able to make highly accurate predictions at a fraction of the cost of experiments or first-principles calculations of comparable accuracy. However, these networks typically rely on large databases of labeled experiments to train the model. In scenarios where data are scarce or expensive to obtain, this can be prohibitive. By building a neural network that provides confidence on the predicted properties, we are able to develop an active learning scheme that can reduce the amount of labeled data required by identifying the areas of chemical space where the model is most uncertain. We present a scheme for coupling a graph neural network with a Gaussian process to featurize solid-state materials and predict properties including a measure of confidence in the prediction. We then demonstrate that this scheme can be used in an active learning context to speed up the training of the model by selecting the optimal next experiment for obtaining a data label. Our active learning scheme can double the rate at which the performance of the model on a test dataset improves with additional data compared to choosing the next sample at random. This type of uncertainty quantification and active learning has the potential to open up new areas of materials science, where data are scarce and expensive to obtain, to the transformative power of graph neural networks.

Determining the maximum information gain and optimizing experimental design in neutron reflectometry using the Fisher information.

An approach based on the Fisher information (FI) is developed to quantify the maximum information gain and optimal experimental design in neutron reflectometry experiments. In these experiments, the FI can be calculated analytically and used to provide sub-second predictions of parameter uncertainties. This approach can be used to influence real-time decisions about measurement angle, measurement time, contrast choice and other experimental conditions based on parameters of interest. The FI provides a lower bound on parameter estimation uncertainties, and these are shown to decrease with the square root of the measurement time, providing useful information for the planning and scheduling of experimental work. As the FI is computationally inexpensive to calculate, it can be computed repeatedly during the course of an experiment, saving costly beam time by signalling that sufficient data have been obtained or saving experimental data sets by signalling that an experiment needs to continue. The approach's predictions are validated through the introduction of an experiment simulation framework that incorporates instrument-specific incident flux profiles, and through the investigation of measuring the structural properties of a phospholipid bilayer.

Bandgap Engineering in the Configurational Space of Solid Solutions via Machine Learning: (Mg,Zn)O Case Study.

Computer simulations of alloys' properties often require calculations in a large space of configurations in a supercell of the crystal structure. A common approach is to map density functional theory results into a simplified interaction model using so-called cluster expansions, which are linear on the cluster correlation functions. Alternative descriptors have not been sufficiently explored so far. We show here that a simple descriptor based on the Coulomb matrix eigenspectrum clearly outperforms the cluster expansion for both total energy and bandgap energy predictions in the configurational space of a MgO-ZnO solid solution, a prototypical oxide alloy for bandgap engineering. Bandgap predictions can be further improved by introducing non-linearity via gradient-boosted decision trees or neural networks based on the Coulomb matrix descriptor.

Interpretable, calibrated neural networks for analysis and understanding of inelastic neutron scattering data.

Deep neural networks (NNs) provide flexible frameworks for learning data representations and functions relating data to other properties and are often claimed to achieve 'super-human' performance in inferring relationships between input data and desired property. In the context of inelastic neutron scattering experiments, however, as in many other scientific scenarios, a number of issues arise: (i) scarcity of labelled experimental data, (ii) lack of uncertainty quantification on results, and (iii) lack of interpretability of the deep NNs. In this work we examine approaches to all three issues. We use simulated data to train a deep NN to distinguish between two possible magnetic exchange models of a half-doped manganite. We apply the recently developed deterministic uncertainty quantification method to provide error estimates for the classification, demonstrating in the process how important realistic representations of instrument resolution in the training data are for reliable estimates on experimental data. Finally we use class activation maps to determine which regions of the spectra are most important for the final classification result reached by the network.

Linking in situ charge accumulation to electronic structure in doped SrTiO3 reveals design principles for hydrogen-evolving photocatalysts.

Recently, high solar-to-hydrogen efficiencies were demonstrated using La and Rh co-doped SrTiO3 (La,Rh:SrTiO3) incorporated into a low-cost and scalable Z-scheme device, known as a photocatalyst sheet. However, the unique properties that enable La,Rh:SrTiO3 to support this impressive performance are not fully understood. Combining in situ spectroelectrochemical measurements with density functional theory and photoelectron spectroscopy produces a depletion model of Rh:SrTiO3 and La,Rh:SrTiO3 photocatalyst sheets. This reveals remarkable properties, such as deep flatband potentials (+2 V versus the reversible hydrogen electrode) and a Rh oxidation state dependent reorganization of the electronic structure, involving the loss of a vacant Rh 4d mid-gap state. This reorganization enables Rh:SrTiO3 to be reduced by co-doping without compromising the p-type character. In situ time-resolved spectroscopies show that the electronic structure reorganization induced by Rh reduction controls the electron lifetime in photocatalyst sheets. In Rh:SrTiO3, enhanced lifetimes can only be obtained at negative applied potentials, where the complete Z-scheme operates inefficiently. La co-doping fixes Rh in the 3+ state, which results in long-lived photogenerated electrons even at very positive potentials (+1 V versus the reversible hydrogen electrode), in which both components of the complete device operate effectively. This understanding of the role of co-dopants provides a new insight into the design principles for water-splitting devices based on bandgap-engineered metal oxides.

Modeling the dielectric constants of crystals using machine learning.

The relative permittivity of a crystal is a fundamental property that links microscopic chemical bonding to macroscopic electromagnetic response. Multiple models, including analytical, numerical, and statistical descriptions, have been made to understand and predict dielectric behavior. Analytical models are often limited to a particular type of compound, whereas machine learning (ML) models often lack interpretability. Here, we combine supervised ML, density functional perturbation theory, and analysis based on game theory to predict and explain the physical trends in optical dielectric constants of crystals. Two ML models, support vector regression and deep neural networks, were trained on a dataset of 1364 dielectric constants. Analysis of Shapley additive explanations of the ML models reveals that they recover correlations described by textbook Clausius-Mossotti and Penn models, which gives confidence in their ability to describe physical behavior, while providing superior predictive power.

Differentiating the role of organic additives to assemble open framework aluminosilicates using INS spectroscopy.

Presently, there is little clarity concerning how organic additives control structure formation in the synthesis of zeolite catalysts. Such ambiguity is a major obstacle towards synthesis design of new bespoke zeolites with intended applications. Herein, we have applied inelastic neutron scattering (INS) spectroscopy to experimentally probe the nature of organic-framework interactions, which are crucial in understanding structure direction. With this technique we have studied the dynamics of 18-crown-6 ether, which can be used as an additive to direct the formation of four zeolites: Na-X, EMC-2, RHO and ZK-5. We observed significant softening of the 18-crown-6 ether molecule's dynamics upon occlusion within a zeolite host, with a strong influence on both the circular and radial vibrational modes. Furthermore, there is a strong correlation between the size/geometry of the zeolite framework cages and perturbations in the dynamics of the 18C6 oxyethylene chain. We propose that the approach used herein can be used to study other zeolites, and hence gain a more comprehensive view of organic-framework interactions.

Inflammation in Mental Disorders: Is the Microbiota the Missing Link?

Research suggests that inflammation is important in the pathophysiology of mental disorders. In addition, a growing body of evidence has led to the concept of the microbiota-gut-brain axis. To understand the potential interactions, we begin by exploring the liaison between the immune system and mental disorders, then we describe the evidence that the microbiota impact the immune response in the developing brain. Next, we review the literature that has documented microbiome alterations in major mental disorders. We end with a summary of therapeutic applications, ranging from psycho-biotics to immunomodulatory drugs that could affect the microbiota-gut-brain axis, and potential treatments to alleviate the adverse effects of antipsychotics. We conclude that there is promising evidence to support the position that the microbiota plays an important role in the immunological pathophysiology of mental disorders with an emphasis on psychotic disorders and mood disorders. However, more research is needed to elucidate the mechanisms.

Understanding the Balance of Entropy and Enthalpy in Hydrogen-Halide Noncovalent Bonding.

Hydrogen bonds are of great scientific interest, determining the free energy landscape and hence chemical and physical properties of many materials systems, for example, the hybrid organic-inorganic perovskites. Although these interactions are critical, understanding them is difficult in complex, multicomponent systems; hydrogen halides are ideal as simple binary model compounds for understanding the role of hydrogen bonding in physical properties like phase transitions. Here we investigate the orthorhombic low-temperature phase and the cubic high-temperature phase in HX (X = F, Cl, Br, or I) systems to understand how different hydrogen-halide bonds influence free energy profiles. We show that hydrogen fluoride has a qualitatively different behavior due to strong hydrogen bonding and hence a very different vibrational entropy. Heavier halides are in contrast rather similar in their physical properties; however, dispersion interactions play a more crucial role in these. These results have implications for the rational design of materials with hydrogen-halide bonds and tuning material properties in systems like mixed anion CH3NH3PbX3 perovskites.

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'.

Heterostructure Engineering of a Reverse Water Gas Shift Photocatalyst.

To achieve substantial reductions in CO2 emissions, catalysts for the photoreduction of CO2 into value-added chemicals and fuels will most likely be at the heart of key renewable-energy technologies. Despite tremendous efforts, developing highly active and selective CO2 reduction photocatalysts remains a great challenge. Herein, a metal oxide heterostructure engineering strategy that enables the gas-phase, photocatalytic, heterogeneous hydrogenation of CO2 to CO with high performance metrics (i.e., the conversion rate of CO2 to CO reached as high as 1400 µmol g cat-1 h-1) is reported. The catalyst is comprised of indium oxide nanocrystals, In2O3- x (OH) y , nucleated and grown on the surface of niobium pentoxide (Nb2O5) nanorods. The heterostructure between In2O3- x (OH) y nanocrystals and the Nb2O5 nanorod support increases the concentration of oxygen vacancies and prolongs excited state (electron and hole) lifetimes. Together, these effects result in a dramatically improved photocatalytic performance compared to the isolated In2O3- x (OH) y material. The defect optimized heterostructure exhibits a 44-fold higher conversion rate than pristine In2O3- x (OH) y . It also exhibits selective conversion of CO2 to CO as well as long-term operational stability.

Mutual Insight on Ferroelectrics and Hybrid Halide Perovskites: A Platform for Future Multifunctional Energy Conversion.

An insight into the analogies, state-of-the-art technologies, concepts, and prospects under the umbrella of perovskite materials (both inorganic-organic hybrid halide perovskites and ferroelectric perovskites) for future multifunctional energy conversion and storage devices is provided. Often, these are considered entirely different branches of research; however, considering them simultaneously and holistically can provide several new opportunities. Recent advancements have highlighted the potential of hybrid perovskites for high-efficiency solar cells. The intrinsic polar properties of these materials, including the potential for ferroelectricity, provide additional possibilities for simultaneously exploiting several energy conversion mechanisms such as the piezoelectric, pyroelectric, and thermoelectric effect and electrical energy storage. The presence of these phenomena can support the performance of perovskite solar cells. The energy conversion using these effects (piezo-, pyro-, and thermoelectric effect) can also be enhanced by a change in the light intensity. Thus, there lies a range of possibilities for tuning the structural, electronic, optical, and magnetic properties of perovskites to simultaneously harvest energy using more than one mechanism to realize an improved efficiency. This requires a basic understanding of concepts, mechanisms, corresponding material properties, and the underlying physics involved with these effects.

Tuning the Negative Thermal Expansion Behavior of the Metal-Organic Framework Cu3BTC2 by Retrofitting.

The modular building principle of metal-organic frameworks (MOFs) presents an excellent platform to explore and establish structure-property relations that tie microscopic to macroscopic properties. Negative thermal expansion (NTE) is a common phenomenon in MOFs and is often ascribed to collective motions that can move through the structure at sufficiently low energies. Here, we show that the introduction of additional linkages in a parent framework, retrofitting, is an effective approach to access lattice dynamics experimentally, in turn providing researchers with a tool to alter the NTE behavior in MOFs. By introducing TCNQ (7,7,8,8-tetracyanoquinodimethane) into the prototypical MOF Cu3BTC2 (BTC = 1,3,5-benzenetricarboxylate; HKUST-1), NTE can be tuned between αV = -15.3 × 10-6 K-1 (Cu3BTC2) and αV = -8.4 × 10-6 K-1 (1.0TCNQ@Cu3BTC2). We ascribe this phenomenon to a general stiffening of the framework as a function of TCNQ loading due to additional network connectivity, which is confirmed by computational modeling and far-infrared spectroscopy. Our findings imply that retrofitting is generally applicable to MOFs with open metal sites, opening yet another way to fine-tune properties in this versatile class of materials.

Subwavelength Spatially Resolved Coordination Chemistry of Metal-Organic Framework Glass Blends.

Microstructured metal-organic framework (MOF) glasses have been produced by combining two amorphous MOFs. However, the electronic structure of these materials has not been interrogated at the length scales of the chemical domains formed in these glasses. Here, we report a subwavelength spatially resolved physicochemical analysis of the electronic states at visible and UV energies in a blend of two zeolitic imidazolate frameworks (ZIFs), ZIF-4-Co and ZIF-62-Zn. By combining spectroscopy at visible and UV energies as well as at core ionization energies in electron energy loss spectroscopy in the scanning transmission electron microscope with density functional theory calculations, we show that domains less than 200 nm in size retain the electronic structure of the precursor crystalline ZIF phases. Prototypical signatures of coordination chemistry including d- d transitions in ZIF-4-Co are assigned and mapped with nanoscale precision.