Peiyuan Gao

Molecular property prediction is essential in a variety of contemporary scientific fields, such as drug development and the design of energy storage materials. Although there are many machine learning models available for this purpose, those that achieve high accuracy while also offering the interpretability of predictions are uncommon. We present a graph neural network that not only matches the prediction accuracies of leading models but also provides insights into four levels of molecular substructures. This model helps identify which atoms, bonds, molecular fragments, and connections among fragments are significant for predicting a specific molecular property. Understanding the importance of connections between fragments is particularly valuable for molecules with substructures that do not connect through standard bonds. The model additionally can quantify the impact of specific fragments on the prediction, allowing for the identification of fragments that may improve or degrade a property value. These interpretable features are essential for deriving scientific insights from the model's learned relationships between molecular structures and properties.

Thick electrodes are critical for long-duration energy storage, but their practical development is hindered by sluggish ion transport and elevated viscosity. Here, we present a solvation-based design strategy—Localized Diluted Electrolytes (LDEs)—that decouples ionic mobility from bulk viscosity through spatially engineered solvation heterogeneity. By promoting the nanoscale preferential distribution of salt-rich and water-rich domains, the LDE approach forms interconnected water-rich regions that sustain fluid-like ionic transport within a mechanically constrained gel matrix. Implemented in a zinc-iodine (Zn-I2) polymeric framework gel battery, this design enables a 2-mm-thick monolithic gel electrode with an areal capacity of ~ 28 mAh/cm² and an energy density of ~ 240 Wh/L based on cathode materials (~ 153 Wh/L for a full cell), achieving stable long-duration cycling. Nuclear magnetic resonance (NMR) spectroscopy, molecular dynamics simulations, and electrochemical analyses confirm rapid, sustained water and ion mobility throughout the polymer network, despite ~ 1200-fold increase in viscosity compared to the liquid-phase electrolyte. This work redefines how solvation environments can be spatially structured to decouple ionic transport from viscosity-driven limitations, establishing a new framework for electrolyte design.

New electrolytes are critical for high‐energy lithium (Li)–sulfur (S) batteries (LSBs) to ensure their stability against Li metal anode and polysulfides (PSs) shuttling which hinder the large‐scale application of LSBs. In this study, the design principle of moderately solvating electrolytes (MSEs) for LSBs is demonstrated by using a multiple‐solvent system comprising of a highly solvating solvent, a weakly solvating solvent, and a non‐solvating solvent to create a well‐balanced electrolyte system. This resulting electrolyte significantly improves the cycle life of LSBs, achieving 300 cycles, which is twice as long as that of similar cells with the conventional electrolyte and it also ensures stable calendar life for at least seven months. The optimal MSE forms robust passivation layers enhancing the structural integrity of both S and Li metal electrodes after cycling. These virtues effectively hinder parasitic side reactions and self‐discharge behavior of LSBs. This electrolyte design principle is versatile and can be applied to other battery chemistries, providing a potential path toward the development of a more efficient and stable battery system. By addressing key challenges such as the instability of electrodes and shuttling of polysulfides, this electrolyte approach offers promising solutions for advancing LSB technology.

With the fast-paced advancement of artificial intelligence (AI), emerging applications of large language models (LLMs) have demonstrated useful applications in material science and self-driving labs (SDLs). Lacking understanding of the relationship between the probability distribution of an LLM and token output, especially in a scientific setting, some concerns are maintaining scientific rigor and autonomy when integrating AI tools. In this correspondence, we propose to use LLM with function calling as an automation node. Such automation can be virtual automation or experimental automation. The proposed agentic automation would project a balance between an elevated level of autonomy in the automation process while maintaining scientific rigor by acquiring data through traditional computing software or experimental tools.

Molecular property prediction is a crucial step in many modern-day scientific applications including drug discovery and energy storage material design. Despite the availability of numerous machine learning models for this task, we are lacking in models that provide both high accuracies and interpretability of the predictions. We introduce the FragNet architecture, a graph neural network not only capable of achieving prediction accuracies comparable to the current state-of-the-art models, but also able to provide insight on four levels of molecular substructures. This model enables understanding of which atoms, bonds, molecular fragments, and molecular fragment connections are critical in the prediction of a given molecular property. The ability to interpret the importance of connections between fragments is of particular interest for molecules which have substructures that are not connected with regular covalent bonds. The interpretable capabilities of FragNet are key to gaining scientific insights from the model’s learned patterns between molecular structure and molecular properties.

Li2O is rarely used for cathode material synthesis due to its high melting point of 1438 °C. However, this work has discovered an unusual Li2O sublimation that occurs at high temperatures of 800-1000 °C, even under 1 atm pressure. Utilizing Li2O vapor sublimed directly from solids, we propose and validate a fundamentally new reaction mechanism that may find broad applications. This mechanism enables the synthesis of single crystals such as LiNi0.8Mn0.1Co0.1O2 (NMC811) or LiNi0.9Mn0.05Co0.05O2 (NMC90) without direct contact with Li2O salts. Moreover, Li2O vapor successfully converts spent polycrystalline NMC811 into well-segregated single crystals without any milling or post-treatment, further confirming the proposed Li2O sublimation mechanism. Li2O vapor, derived from Li2O solids, diffuses rapidly and reacts with other precursors, mimicking a molten salt environment without actual melting, which facilitates single crystal growth. The chemical lithiation process continuously drives Li2O to sublime, sintering crystals, as confirmed experimentally and theoretically. Single crystals derived from Li2O salts, using either fresh precursors or spent polycrystals, demonstrate outstanding cycling stability after 1000 cycles in full cells. The discovery of the Li2O sublimation phenomenon and its universal role in promoting single crystal growth provides a novel direction for broad single crystal synthesis, scale-up, and recycling.

The properties of constrained fluids have increasingly gained relevance for applications ranging from materials to biology. In this work, we propose a multiscale model using twin neural networks to investigate the properties of a fluid constrained between solid surfaces with complex shapes. The atomic scale model and the mesoscale model are connected by the coarse-grained potential which is represented by the first neural network. Then we train the second neural network model as a surrogate to predict the velocity profile of the constrained fluid with complex boundary conditions at the mesoscale. The effect of complex boundary conditions on the fluid dynamics properties and the accuracy of the neural network model prediction are systematically investigated. We demonstrate that the neural network-enhanced multiscale framework can connect simulations at atomic scale and mesoscale and reproduce the properties of a constrained fluid at mesoscale. This work provides insight into multiscale model development with the aid of machine learning techniques and the developed model can be used for modern nanotechnology applications such as enhanced oil recovery and porous materials design.

The lifespan of lithium (Li) metal batteries (LMBs) can be greatly improved by the formation of inorganic‐rich electrode‐electrolyte interphases (EEIs) (including solid‐electrolyte interphase on anode and cathode‐electrolyte interphase on cathode). In this work, a localized high‐concentration electrolyte containing lithium bis(fluorosulfonyl)imide (LiFSI) salt, 1,2‐dimethoxyethane (DME) solvent and 1,2‐bis(1,1,2,2‐tetrafluoroethoxy)ethane (BTFEE) diluent is optimized. BTFEE is a fluorinated ether with weakly‐solvating ability for LiFSI so it also acts as a co‐solvent in this electrolyte. It can facilitate anion decomposition at electrode surfaces and promote the formation of more inorganic‐rich EEI layers. With an optimized molar ratio of LiFSI:DME:BTFEE = 1:1.15:3, LMBs with a high loading (4 mAh cm ⁻² ) lithium nickel manganese cobalt oxide (LiNi _0.8 Mn _0.1 Co _0.1 ) cathode can retain 80% capacity in 470 cycles when cycled in a voltage range of 2.8–4.4 V. The fundamental understanding on the functionality of BTFEE revealed in this work provides new perspectives on the design of practical high‐energy density battery systems.

Sodium (Na)‐metal batteries (SMBs) are considered one of the most promising candidates for the large‐scale energy storage market owing to their high theoretical capacity (1,166 mAh g ⁻¹ ) and the abundance of Na raw material. However, the limited stability of electrolytes still hindered the application of SMBs. Herein, sulfolane (Sul) and vinylene carbonate (VC) are identified as effective dual additives that can largely stabilize propylene carbonate (PC)‐based electrolytes, prevent dendrite growth, and extend the cycle life of SMBs. The cycling stability of the Na/NaNi _0.68 Mn _0.22 Co _0.1 O ₂ (NaNMC) cell with this dual‐additive electrolyte is remarkably enhanced, with a capacity retention of 94% and a Coulombic efficiency (CE) of 99.9% over 600 cycles at a 5 C (750 mA g ⁻¹ ) rate. The superior cycling performance of the cells can be attributed to the homogenous, dense, and thin hybrid solid electrolyte interphase consisting of F‐ and S‐containing species on the surface of both the Na metal anode and the NaNMC cathode by adding dual additives. Such unique interphases can effectively facilitate Na‐ion transport kinetics and avoid electrolyte depletion during repeated cycling at a very high rate of 5 C. This electrolyte design is believed to result in further improvements in the performance of SMBs.