Evan Walter Clark Spotte-Smith

Applications of deep learning (DL) to design nanomaterials are hampered by a lack of suitable data representations and training data. Here we report efforts to overcome these limitations and leverage DL to optimize the nonlinear optical properties of core-shell upconverting nanoparticles (UCNPs). UCNPs, which have applications in fields such as biosensing, super-resolution microscopy and three-dimensional printing, can emit visible and ultraviolet light from near-infrared excitations. We report a large-scale dataset of UCNP emission spectra based on accurate but expensive kinetic Monte Carlo simulations (N > 6,000) and use these data to train a heterogeneous graph neural network using a physically motivated representation of UCNP nanostructure. Applying gradient-based optimization on the trained graph neural network, we identify structures with 6.5× higher predicted emission under 800-nm illumination than any UCNP in our training set. Our work reveals design principles for UCNP heterostructures and presents a roadmap for DL-based inverse design of nanomaterials.

Foundation models, including large language models, vision-language models, and similar large-scale machine learning tools, are quickly becoming ubiquitous in society and in the professional world. Chemical practitioners are not immune to the appeal of foundation models, nor are they immune to the many risks and harms that these models introduce. In this work, I present the first analysis of foundation models using the combined lens of scientific ethics and chemical professional ethics. I find that general-purpose generative foundation models are in many ways incompatible with the ethical practice of chemistry, though there are fewer ethical problems with chemistry-specific foundation models. My discussion concludes with an examination of how the harm associated with foundation models can be minimized and further poses a set of lingering questions for chemical practitioners and scientific ethicists.

The Materials Project was launched formally in 2011 to drive materials discovery forwards through high-throughput computation and open data. More than a decade later, the Materials Project has become an indispensable tool used by more than 600,000 materials researchers around the world. This Perspective describes how the Materials Project, as a data platform and a software ecosystem, has helped to shape research in data-driven materials science. We cover how sustainable software and computational methods have accelerated materials design while becoming more open source and collaborative in nature. Next, we present cases where the Materials Project was used to understand and discover functional materials. We then describe our efforts to meet the needs of an expanding user base, through technical infrastructure updates ranging from data architecture and cloud resources to interactive web applications. Finally, we discuss opportunities to better aid the research community, with the vision that more accessible and easy-to-understand materials data will result in democratized materials knowledge and an increasingly collaborative community.

Foundation models, including large language models, vision-language models, and similar large-scale machine learning tools, are quickly becoming ubiquitous in society and in the professional world. Chemical practitioners are not immune to the appeal of foundation models, nor are they immune to the many risks and harms that these models introduce. In this work, I present the first analysis of foundation models using the combined lens of scientific ethics and chemical professional ethics. I find that general-purpose generative foundation models are in many ways incompatible with the ethical practice of chemistry, though there are fewer ethical problems with chemistry-specific foundation models. My discussion concludes with an examination of how the harm associated with foundation models can be minimized and further poses a set of lingering questions for chemical practitioners and scientific ethicists.

Applications of deep learning (DL) to design nanomaterials are hampered by a lack of suitable data representations and training data. We report efforts to overcome these limitations and leverage DL to optimize the nonlinear optical properties of core-shell upconverting nanoparticles (UCNPs). UCNPs, which have applications in e.g., biosensing, super-resolution microscopy, and 3D printing, can emit visible and ultraviolet light from near-infrared excitations. We report the first large-scale dataset of UCNP emission spectra based on accurate but expensive kinetic Monte Carlo simulations (N > 6,000) and use this data to train a heterogeneous graph neural network (GNN) using a novel representation of UCNP nanostructure. Applying gradient-based optimization on the trained GNN, we identify structures with 6.5x higher predicted emission under 800nm illumination than any UCNP in our training set. Our work reveals new design principles for UCNPs and presents a roadmap for DL-based inverse design of nanomaterials.

Hydrolysis is a fundamental family of chemical reactions where water facilitates the cleavage of bonds. The process is ubiquitous in biological and chemical systems, owing to water's remarkable versatility as a solvent. However, accurately predicting the feasibility of hydrolysis through computational techniques is a difficult task, as subtle changes in reactant structure like heteroatom substitutions or neighboring functional groups can influence the reaction outcome. Furthermore, hydrolysis is sensitive to the pH of the aqueous medium, and the same reaction can have different reaction properties at different pH conditions. In this work, we have combined reaction templates and high-throughput ab initio calculations to construct a diverse data set of hydrolysis free energies. The developed framework automatically identifies reaction centers, generates hydrolysis products, and utilizes a trained graph neural network (GNN) model to predict ΔG values for all potential hydrolysis reactions in a given molecule. The long-term goal of the work is to develop a data-driven, computational tool for high-throughput screening of pH-specific hydrolytic stability and the rapid prediction of reaction products, which can then be applied in a wide array of applications including chemical recycling of polymers and ion-conducting membranes for clean energy generation and storage.

Macroscopic chemical and physical phenomena are driven by microscopic interactions at the atomic and molecular scales. In order to capture complex processes with high fidelity, simulation methods that bridge disparate time and length scales are needed. While techniques like molecular dynamics and ab initio simulations capture dynamics and reactivity at high resolution, they cannot be used beyond relatively small length (hundreds to thousands of atoms) and time scales (picoseconds to microseconds). Kinetic Monte Carlo (kMC) approaches overcome these limitations to bridge length and time scales across several orders of magnitude while retaining relevant microscopic resolution, making it a powerful and flexible tool.

Hydrolysis is a fundamental family of chemical reactions where water facilitates the cleavage of bonds. The process is ubiquitous in biological and chemical systems, owing to water's remarkable versatility as a solvent. However, accurately predicting the feasibility of hydrolysis through computational techniques is a difficult task, as subtle changes in reactant structure like heteroatom substitutions or neighboring functional groups can influence the reaction outcome. Furthermore, hydrolysis is sensitive to the pH of the aqueous medium, and the same reaction can have different reaction properties at different pH conditions. In this work, we have combined reaction templates and high-throughput ab-initio calculations to construct a diverse dataset of hydrolysis free energies. The developed framework automatically identifies reaction centers, generates hydrolysis products, and utilizes a trained Graph Neural Network(GNN) model to predict values for all potential hydrolysis reactions in a given molecule. The long-term goal of the work is to develop a data-driven, computational tool for high-throughput screening of pH-specific hydrolytic stability and the rapid prediction of reaction products, which can then be applied in a wide array of applications including chemical recycling of polymers and ion-conducting membranes for clean energy generation and storage.

Understanding the intricate pathways of ethylene carbonate (EC) decomposition is pivotal for enhancing the stability and efficiency of electrolytes in various electrochemical applications. In this study, we delve into the decomposition mechanisms of EC in the presence of superoxide and peroxide species. The decomposition of EC-based electrolytes causes gas (carbon dioxide, carbon monoxide and hydrogen) evolution. Since this gas evolution is observed at potentials much lower than the oxidation potential of EC, it has been suggested that the mechanisms leading to decomposition are primarily chemical in nature (i.e., not related to the electrochemical oxidation of EC) and related to the reaction of reactive oxygen [singlet, superoxide, peroxide] with EC. [1,2,3]. Utilizing high-throughput Density Functional Theory (DFT) calculations, we meticulously explored the reaction networks of EC with superoxide and peroxide species across different cathode voltages. Our findings unveil compelling insights into the primary pathways dictating EC decomposition, which notably occur at potentials significantly lower than the oxidation potential of EC. Monte Carlo simulations were performed to understand the major products of the reaction network. Carbon dioxide is one of the major products in the reaction networks. Other major products include mono-carbonate, formaldehyde and organic products containing O-O in the fragments. We find that the products with O-O bond in the organic chain were also previously observed through mass spectroscopy experiments in [4]. Thus, using high-throughput data-driven approach, we are able to identify novel decomposition pathways of ethylene carbonate near the cathode. Reference [1] Freiberg, Anna TS, et al. “Singlet oxygen reactivity with carbonate solvents used for Li-ion battery electrolytes.” The Journal of Physical Chemistry A 122.45 (2018): 8828-8839. [2] Wandt, Johannes, et al. “Singlet oxygen evolution from layered transition metal oxide cathode materials and its implications for lithium-ion batteries.” Materials Today 21.8 (2018): 825-833. [3] Rinkel, Bernardine LD, et al. “Electrolyte oxidation pathways in lithium-ion batteries.” Journal of the American Chemical Society 142.35 (2020): 15058-15074. [4] Kaufman, et al. “Surface Lithium Carbonate Influences Electrolyte Degradation via Reactive Oxygen Attack in Lithium-Excess Cathode Materials” Journal of the American Chemical Society (2021): 4170–4176

Preventing the decomposition reactions of electrolyte solutions is essential for extending the lifetime of lithium-ion batteries. However, the high voltage (>4.5 V) reactivity of commonly used electrolyte solutions is still poorly understood; electrolyte decomposition at the positive electrode is generally believed to proceed via electrochemical oxidation of the organic carbonate solvent,¹ while more recent work has revealed an alternative pathway involving reactive lattice oxygen species originating from the positive electrode (i.e., singlet oxygen).² Unfortunately, we still have a limited understanding of these reactions, the products that form, and which chemical structural units in the electrolyte solution make them reactive - all of which make it challenging to assess and mitigate the impact of electrolyte decomposition on the battery’s lifetime. In this work, operando gas measurements and solution NMR were combined to study the electrochemical and chemical decomposition of widely used electrolyte salts, solvents, and additives. First, the electrochemical stability of the electrolyte components was reviewed against C65 electrodes in a standard H-cell setup: all solvents reveal little decomposition up to 5.5 V (vs Li/Li⁺), consistent with previous electrochemical measurements.^3,4 Based on the experimentally detected soluble and gaseous products, the electrochemical decomposition mechanism is inferred. Subsequently, the chemical stability against reactive oxygen species (i.e., singlet oxygen; ¹O₂, peroxide anion; O₂ ^{2 −}, and superoxide radical anion, O₂ ⁻) was explored. All carbonate solvents demonstrated instability against reactive oxygen species. As before, the experimentally detected gaseous and soluble products were used to infer the decomposition mechanisms. The proposed mechanisms were verified using ¹⁷O and ¹⁸O isotopic labeling for NMR spectroscopy and mass spectrometry measurements, respectively. The findings on the reactivity between the electrolyte components and reactive oxygen species provide a deeper understanding of electrolyte decomposition processes occurring at high voltage positive electrodes (e.g., Li-rich layered and disordered rocksalt oxides, high-voltage spinels, and polyanionic materials). Moreover, these insights will support the development of more stable electrolyte formulations, further extending the lifetime of lithium-ion batteries. (1) Imhof, R.; Novak, P. Oxidative Electrolyte Solvent Degradation in Lithium-Ion Batteries: An In Situ Differential Electrochemical Mass Spectrometry Investigation. J Electrochem Soc 1999, 146 (5), 1702–1706. (2) Jung, R.; Metzger, M.; Maglia, F.; Stinner, C.; Gasteiger, H. A. Oxygen Release and Its Effect on the Cycling Stability of LiNi x Mn y Co z O 2 (NMC) Cathode Materials for Li-Ion Batteries. J Electrochem Soc 2017, 164 (7), A1361–A1377. https://doi.org/10.1149/2.0021707jes. (3) Zhang, X.; Soc, J. E.; Zhang, X.; Kostecki, R.; Richardson, T. J.; Pugh, J. K.; Ross, P. N. Electrochemical and Infrared Studies of the Reduction of Organic Carbonates Electrochemical and Infrared Studies of the Reduction of Organic Carbonates. Journal of The Electrochemical Society 2001, 148 (12), A1341–A1345. https://doi.org/10.1149/1.1415547. (4) Moshkovich, M.; Cojocaru, M.; Gottlieb, H. E.; Aurbach, D. The Study of the Anodic Stability of Alkyl Carbonate Solutions by in Situ FTIR Spectroscopy, EQCM, NMR and MS. Journal of Electroanalytical Chemistry 2001, 497, 84–96. https://doi.org/http://dx.doi.org/10.1016/S0022-0728(00)00457-5.