Retrieval, the initial stage of a recommendation system, is tasked with down-selecting items from a pool of tens of millions of candidates to a few thousands. Embedding Based Retrieval (EBR) has been a typical choice for this problem, addressing the computational demands of deep neural networks across vast item corpora. EBR utilizes Two Tower or Siamese Networks to learn representations for users and items, and employ Approximate Nearest Neighbor (ANN) search to efficiently retrieve relevant items. Despite its popularity in industry, EBR faces limitations. The Two Tower architecture, relying on a single dot product interaction, struggles to capture complex data distributions due to limited capability in learning expressive interactions between users and items. Additionally, ANN index building and representation learning for user and item are often separate, leading to inconsistencies exacerbated by representation (e.g. continuous online training) and item drift (e.g. items expired and new items added). In this paper, we introduce the Hierarchical Structured Neural Network (HSNN), an efficient deep neural network model to learn intricate user and item interactions beyond the commonly used dot product in retrieval tasks, achieving sublinear computational costs relative to corpus size. A Modular Neural Network (MoNN) is designed to maintain high expressiveness for interaction learning while ensuring efficiency. A mixture of MoNNs operate on a hierarchical item index to achieve extensive computation sharing, enabling it to scale up to large corpus size. MoNN and the hierarchical index are jointly learnt to continuously adapt to distribution shifts in both user interests and item distributions. HSNN achieves substantial improvement in offline evaluation compared to prevailing methods.

Machine learning interatomic potentials (MLIPs) are a promising technique for atomic modeling. While small errors are widely reported for MLIPs, an open concern is whether MLIPs can accurately reproduce atomistic dynamics and related physical properties in molecular dynamics (MD) simulations. In this study, we examine the state-of-the-art MLIPs and uncover several discrepancies related to atom dynamics, defects, and rare events (REs), compared to ab initio methods. We find that low averaged errors by current MLIP testing are insufficient, and develop quantitative metrics that better indicate the accurate prediction of atomic dynamics by MLIPs. The MLIPs optimized by the RE-based evaluation metrics are demonstrated to have improved prediction in multiple properties. The identified errors, the evaluation metrics, and the proposed process of developing such metrics are general to MLIPs, thus providing valuable guidance for future testing and improvements of accurate and reliable MLIPs for atomistic modeling.

All-solid-state Li-ion batteries are a next generation electrochemical energy storage technology and an emerging alternative to conventional Li-ion batteries owing to higher energy density, wider operation temperature range and better safety properties1. Solid state ionic conductor as the solid electrolyte (SEs) is the key component in all-solid-state Li-ion batteries. Among existing SEs, sulfides SEs attract extensive attentions due to high room temperature ionic conductivity and favorable mechanical properties. However, currently known sulfide SEs only present a few limited types of crystal structures, which limits the materials design and hinders our understanding of ionic conduction mechanisms. Exploring new compound and new crystal structures beyond known groups is much desired in order to expand the ionic conductor database and advances the knowledge about ionic conduction. Recently, our group successfully synthesized and identified a new lithium chloro-thiophosphate in Li2S-P2S5-LiCl system, namely Li15P4S16Cl3 2. This compound shows a low ionic conductivity of ~10-7 S/cm at room temperature. However, this crystal structure provides abundant choices for composition and unit cell size tuning, which may trigger high ionic conductivity. We successfully substituted P in Li15P4S16Cl3 with aliovalent cations to tune the Li content and the interstitial site occupancy. As a result, high room temperature conductivity of ~10-5 S/cm was achieved. The structure of the doped compounds was examined with using synchrotron X-ray and neutron diffractions. The Li+ diffusion pathway was also theoretically investigated with using ab initio molecular dynamics (AIMD) method and experimentally examined with using 7Li solid state nuclear magnetic resonance (NMR) techniques. Auvergniot, J.; Cassel, A.; Ledeuil, J.-B.; Viallet, V.; Seznec, V.; Dedryvère, R., Interface Stability of Argyrodite Li6PS5Cl toward LiCoO2, LiNi1/3Co1/3Mn1/3O2, and LiMn2O4 in Bulk All-Solid-State Batteries. Chemistry of Materials 2017, 29 (9), 3883-3890. Liu, ZT.; Zinkevich, T.; He, XF.; Indris, S.; Xiong, S.; Liu, J.; Xu, W.; Bai, J.; Mo, Yifei; Chen, HL., Li15P4S16Cl3, a new lithium chloro-thiophosphate as a solid state ionic conductor. Submitted.

The development of all-solid-state Li-ion batteries requires solid electrolyte materials with many desired properties, such as ionic conductivity, chemical and electrochemical stability, and mechanical durability. Computation-guided materials design techniques are advantageous in designing and identifying new solid electrolytes that can simultaneously meet these requirements. In this joint computational and experimental study, a new family of fast lithium ion conductors, namely, LiTaSiO5 with sphene structure, are successfully identified, synthesized, and demonstrated using a novel computational design strategy. First-principles computation predicts that Zr-doped LiTaSiO5 sphene materials have fast Li diffusion, good phase stability, and poor electronic conductivity, which are ideal for solid electrolytes. Experiments confirm that Zr-doped LiTaSiO5 sphene structure indeed exhibits encouraging ionic conductivity. The lithium diffusion mechanisms in this material are also investigated, indicating the sphene materials are 3D conductors with facile 1D diffusion along the [101] direction and additional cross-channel migration. This study demonstrates a novel design strategy of activating fast Li ionic diffusion in lithium sphenes, a new materials family of superionic conductors.

All-solid-state batteries as an emerging electrochemical energy storage technology are attracting extensive attentions, owing to the good safety properties and potentially very high energy density with using Li-metal anode. Solid electrolyte (SE) is the key component in all-solid-state Li-ion batteries and solid state Li ion conductors with high room temperature conductivity, ideal electrochemical stability and good mechanical compatibility are very much desired. Oxides and sulfides based SEs have their own advantages and disadvantages. Oxides can be handled and stored in ambient environment, while their room temperature ionic conductivity is relatively low and the grain boundary is high. Sulfides are commonly not very stable with air and moisture, but they have much higher ionic conductivity, lower grain boundary resistance and favorable mechanical properties for cold processing of solid cells. Here we report recent progresses in our research group in both oxide and sulfide-based lithium ion conductors. A group of zirconia doped lithium tantalum oxosilicate previously identified through a computation assisted search and screening were successfully synthesized and showed encouraging ionic conductivity of >10 -5 S/cm at room temperature. Analysis of electrochemical impedance spectroscopy data and computation predictions indicate that higher conductivity from 10 -4 to 10 -3 S/cm may be achieved by further optimizations. With using another experiment-based design strategy, a group of lithium chloro-thiophosphate compounds were also experimentally identified assisted by in situ X-ray diffraction investigations and showed high conductivity of >10 -4 S/cm. The structure of both sulfide and oxide new compounds were characterized with using synchrotron X-ray diffraction and neutron diffraction. The relationship between their ionic conductivity and structural features, such as unit cell size, diffusion pathway, site occupancy, interstitial sites, etc., were revealed by the characterization and will be discussed in the presentation. Caption of Figure 1. a) Crystal structure and ionic conductivity of Zr doped LiTaSiO 5 with sphene structure. b) Crystal structure of new lithium chloro-thiophosphate Figure 1

Although machine learning has gained great interest in the discovery of functional materials, the advancement of reliable models is impeded by the scarcity of available materials property data. Here we propose and demonstrate a distinctive approach for materials discovery using unsupervised learning, which does not require labeled data and thus alleviates the data scarcity challenge. Using solid-state Li-ion conductors as a model problem, unsupervised materials discovery utilizes a limited quantity of conductivity data to prioritize a candidate list from a wide range of Li-containing materials for further accurate screening. Our unsupervised learning scheme discovers 16 new fast Li-conductors with conductivities of 10 −4 –10 −1 S cm −1 predicted in ab initio molecular dynamics simulations. These compounds have structures and chemistries distinct to known systems, demonstrating the capability of unsupervised learning for discovering materials over a wide materials space with limited property data.

As technologically important materials for solid‐state batteries, Li super‐ionic conductors are a class of materials exhibiting exceptionally high ionic conductivity at room temperature. These materials have unique crystal structural frameworks hosting a highly conductive Li sublattice. However, it is not understood why certain crystal structures of the super‐ionic conductors lead to high conductivity in the Li sublattice. In this study, using topological analysis and ab initio molecular dynamics simulations, the crystal structures of all Li‐conducting oxides and sulfides are studied systematically and the key features pertaining to fast‐ion conduction are quantified. In particular, a unique feature of enlarged Li sites caused by large local spaces in the crystal structural framework is identified, promoting fast conduction in the Li‐ion sublattice. Based on these quantified features, the high‐throughput screening identifies many new structures as fast Li‐ion conductors, which are further confirmed by ab initio molecular dynamics simulations. This study provides new insights and a systematic quantitative understanding of the crystal structural frameworks of fast ion‐conductor materials and motivates future experimental and computational studies on new fast‐ion conductors.

Tremendous efforts have been devoted to the design of solid Li+ electrolytes and the development of all-solid-state batteries. Compared with conventional Li-ion batteries, which use flammable liquid organic electrolytes, all-solid-state batteries show significant advantages in safety. In this work, a novel lithium chlorothiophosphate compound, Li15P4S16Cl3, is discovered. The crystal structure and electrochemical properties are investigated. Li15P4S16Cl3 can be synthesized as a pure phase via a facile solid-state reaction by heating a ball-milled mixture of Li2S, P2S5, and LiCl at 360 °C. The crystal structure of Li15P4S16Cl3 was refined against neutron and synchrotron powder X-ray diffraction data, revealing that it crystallizes in the space group I4̅3d. The Li+ transport in Li15P4S16Cl3 was also investigated by multiple solid-state NMR methods, including variable-temperature NMR line-shape analysis, NMR relaxometry, and pulsed-field-gradient NMR. Li15P4S16Cl3 shows good thermodynamic stability and can be synthesized at relatively low temperature. Although it exhibits a low ionic conductivity at room temperature, it can serve as a new motif crystal structure for the design and development of new solid-state electrolytes.

Ab initio molecular dynamics (AIMD) simulation is widely employed in studying diffusion mechanisms and in quantifying diffusional properties of materials. However, AIMD simulations are often limited to a few hundred atoms and a short, sub-nanosecond physical timescale, which leads to models that include only a limited number of diffusion events. As a result, the diffusional properties obtained from AIMD simulations are often plagued by poor statistics. In this paper, we re-examine the process to estimate diffusivity and ionic conductivity from the AIMD simulations and establish the procedure to minimize the fitting errors. In addition, we propose methods for quantifying the statistical variance of the diffusivity and ionic conductivity from the number of diffusion events observed during the AIMD simulation. Since an adequate number of diffusion events must be sampled, AIMD simulations should be sufficiently long and can only be performed on materials with reasonably fast diffusion. We chart the ranges of materials and physical conditions that can be accessible by AIMD simulations in studying diffusional properties. Our work provides the foundation for quantifying the statistical confidence levels of diffusion results from AIMD simulations and for correctly employing this powerful technique.

H- ion conductor materials have the great potential to enable high-energy density electrochemical storage based on hydrogen. Fast H- conduction has been recently demonstrated in the La2-x-ySrx+yLiH1-x+yO3-y oxyhydride materials. However, little is known about the H- diffusion mechanism in this new material and its unique structure. The origin of such exceptional H- conduction in the oxide-based materials is of great interest. Using first-principles calculations, we studied the energetics and diffusion mechanisms of H- ions as a function of structures and compositions in this oxyhydride system. Our study identified that fast H- diffusion is mediated by H- vacancies and that the fast two-dimensional or three-dimensional H- diffusion is activated by different anion sublattices in different compositions. In addition, novel doping was predicted from ab initio computation to increase H- conductivity in these materials. The unique two-anion-site feature in this structural framework enables highly tunable lattice and minimizes the blocking of anion diffusion by oxygen sublattice, allowing high mobile-carrier concentration and good diffusion network. This conclusion offers general guidance for future design and discovery of novel oxide-based anion conductors.

The all-solid-state lithium-ion battery is a promising next-generation battery technology. However, the realization of all-solid-state batteries is impeded by limited understanding of solid electrolyte materials and solid electrolyte-electrode interfaces. In this review, we present an overview of recently developed computation techniques and their applications in understanding and advancing materials and interfaces in all-solid-state batteries. We review the role of ab initio molecular dynamics simulations in studying fast ion conductors and discuss the capabilities of thermodynamic calculations powered by materials databases for identifying the chemical and electrochemical stability of solid electrolyte materials and solid electrolyte-electrode interfaces. We highlight the computational studies in the design and discovery of new solid electrolyte materials and outline design guidelines for solid electrolytes and their interfaces. We conclude with discussion of future directions in computation techniques, materials development, and interface engineering for all-solid-state lithium-ion batteries.

Lithium metal battery is a promising candidate for high‐energy‐density energy storage. Unfortunately, the strongly reducing nature of lithium metal has been an outstanding challenge causing poor stability and low coulombic efficiency in lithium batteries. For decades, there are significant research efforts to stabilize lithium metal anode. However, such efforts are greatly impeded by the lack of knowledge about lithium‐stable materials chemistry. So far, only a few materials are known to be stable against Li metal. To resolve this outstanding challenge, lithium‐stable materials have been uncovered out of chemistry across the periodic table using first‐principles calculations based on large materials database. It is found that most oxides, sulfides, and halides, commonly studied as protection materials, are reduced by lithium metal due to the reduction of metal cations. It is discovered that nitride anion chemistry exhibits unique stability against Li metal, which is either thermodynamically intrinsic or a result of stable passivation. The results here establish essential guidelines for selecting, designing, and discovering materials for lithium metal protection, and propose multiple novel strategies of using nitride materials and high nitrogen doping to form stable solid‐electrolyte‐interphase for lithium metal anode, paving the way for high‐energy rechargeable lithium batteries.

SrCeO3 perovskites are promising materials for hydrogen separation membranes. High hydrogen flux in SrCeO3 is achieved by various elemental doping to increase protonic and electronic conductivity. While the effect of B-site dopants on protonic conductivity is established, the polaronic mechanism induced by B-site cations, which is essential for electronic transport, has been less understood. Using first principles hybrid functional calculations, we investigated the polaron formation and migration in SrCeO3 perovskites doped with different elements. Our computation results revealed distinctive behaviors of different dopant elements in localizing polarons and explained previous literature results of doping SrCeO3 for increasing electronic conductivity and hydrogen flux. In addition, new promising dopants are predicted to increase electronic conductivity. The computation approach demonstrated in this study provides a general scheme to design materials with tailored polaron formation and enhanced functional properties.

Super-ionic conductor materials have great potential to enable novel technologies in energy storage and conversion. However, it is not yet understood why only a few materials can deliver exceptionally higher ionic conductivity than typical solids or how one can design fast ion conductors following simple principles. Using ab initio modelling, here we show that fast diffusion in super-ionic conductors does not occur through isolated ion hopping as is typical in solids, but instead proceeds through concerted migrations of multiple ions with low energy barriers. Furthermore, we elucidate that the low energy barriers of the concerted ionic diffusion are a result of unique mobile ion configurations and strong mobile ion interactions in super-ionic conductors. Our results provide a general framework and universal strategy to design solid materials with fast ionic diffusion.

The technology of H-2 separation is essential for the utilization of hydrogen fuel, which is strongly motivated by the grand challenge of climate change. In particular, SrCeO3 perovskites with mixed protonic and electronic conductivities are promising materials for H-2 separation due to negligible oxygen permeability. However, low electronic conductivity of SrCeO3 often limits the overall H-2 flux. Yb-doped SrCeO3 has been demonstrated to exhibit excellent performance as H-2 separation membranes with high electronic conductivity. Our study reveals the enhancement on electronic conductivity from Yb dopant originates from the polaron mechanism, based on first principles hybrid calculations. Electron polarons on Yb and Ce are mobile, contributing to electronic conductivity in Yb-doped SrCeO3. The insight of the polaron enhancement mechanism will shed light on predicting promising dopants to increase electronic conductivity. Our study demonstrates first principles techniques in accelerating deign of mixed conductors.

The electrochemical stability window of solid electrolyte is overestimated by the conventional experimental method using a Li/electrolyte/inert metal semiblocking electrode because of the limited contact area between solid electrolyte and inert metal. Since the battery is cycled in the overestimated stability window, the decomposition of the solid electrolyte at the interfaces occurs but has been ignored as a cause for high interfacial resistances in previous studies, limiting the performance improvement of the bulk‐type solid‐state battery despite the decades of research efforts. Thus, there is an urgent need to identify the intrinsic stability window of the solid electrolyte. The thermodynamic electrochemical stability window of solid electrolytes is calculated using first principles computation methods, and an experimental method is developed to measure the intrinsic electrochemical stability window of solid electrolytes using a Li/electrolyte/electrolyte‐carbon cell. The most promising solid electrolytes, Li 10 GeP 2 S 12 and cubic Li‐garnet Li 7 La 3 Zr 2 O 12 , are chosen as the model materials for sulfide and oxide solid electrolytes, respectively. The results provide valuable insights to address the most challenging problems of the interfacial stability and resistance in high‐performance solid‐state batteries.

Garnet-type solid-state electrolytes have attracted extensive attention due to their high ionic conductivity, approaching 1 mS cm−1, excellent environmental stability, and wide electrochemical stability window, from lithium metal to ∼6 V. However, to date, there has been little success in the development of high-performance solid-state batteries using these exceptional materials, the major challenge being the high solid–solid interfacial impedance between the garnet electrolyte and electrode materials. In this work, we effectively address the large interfacial impedance between a lithium metal anode and the garnet electrolyte using ultrathin aluminium oxide (Al2O3) by atomic layer deposition. Li7La2.75Ca0.25Zr1.75Nb0.25O12 (LLCZN) is the garnet composition of choice in this work due to its reduced sintering temperature and increased lithium ion conductivity. A significant decrease of interfacial impedance, from 1,710 Ω cm2 to 1 Ω cm2, was observed at room temperature, effectively negating the lithium metal/garnet interfacial impedance. Experimental and computational results reveal that the oxide coating enables wetting of metallic lithium in contact with the garnet electrolyte surface and the lithiated-alumina interface allows effective lithium ion transport between the lithium metal anode and garnet electrolyte. We also demonstrate a working cell with a lithium metal anode, garnet electrolyte and a high-voltage cathode by applying the newly developed interface chemistry. Garnet-type electrolytes are attractive for lithium metal batteries due to their high ionic conductivity. A strategy to decrease interfacial impedance between a lithium metal anode and garnet electrolyte is found promising for all-solid-state batteries.

We perform a first principles computational study of designing the Na0.5Bi0.5TiO3 (NBT) perovskite material to increase its oxygen ionic conductivity. In agreement with the previous experiments, our computation results confirm fast oxygen ionic diffusion and good stability of the NBT material. The oxygen diffusion mechanisms in this new material were systematically investigated, and the effects of local atomistic configurations and dopants on oxygen diffusion were revealed. Novel doping strategies focusing on the Na/Bi sublattice were predicted and demonstrated by the first principles calculations. In particular, the K doped NBT compound achieved good phase stability and an order of magnitude increase in oxygen ionic conductivity of up to 0.1 S cm(-1) at 900 K compared to the previous Mg doped compositions. This study demonstrated the advantages of first principles calculations in understanding the fundamental structure-property relationship and in accelerating the materials design of the ionic conductor materials.