Recent advances in artificial intelligence(AI)have led to the development of sophisticated algorithms that significantly improve image analysis capabilities.This combination of AI and microscopic imaging is transformi...Recent advances in artificial intelligence(AI)have led to the development of sophisticated algorithms that significantly improve image analysis capabilities.This combination of AI and microscopic imaging is transforming the way we interpret and analyze imaging data,simplifying complex tasks and enabling innovative experimental methods previously thought impossible.In smart manufacturing,these improvements are especially impactful,increasing precision and efficiency in production processes.This review examines the convergence of AI with particle image analysis,an area we refer to as“particle vision analysis(PVA).”We offer a detailed overview of how this technology integrates into and impacts various fields within the physical sciences and materials sectors,where it plays a crucial role in both innovation and operational improvements.We explore four key areas of advancement-namely,particle classification,detection,segmentation,and object tracking-along with a look into the emerging field of augmented microscopy.This paper also underscores the vital role of the existing datasets and implementations that support these applications,which provide essential insights and resources that drive continuous research and development in this fast-evolving field.Our thorough analysis aims to outline the transformative potential of AI-driven PVA in improving precision in future manufacturing at the microscopic scale and thereby preparing the ground for significant technological progress and broad industrial applications in nanomanufacturing,biomanufacturing,and pharmaceutical manufacturing.This exploration not only highlights the advantages of integrating AI into conventional manufacturing processes but also anticipates the rise of next-generation smart manufacturing,which is set to revolutionize industry standards and operational practices.展开更多
The rapid advancement of machine learning based tight-binding Hamiltonian(MLTB)methods has opened new avenues for efficient and accurate electronic structure simulations,particularly in large-scale systems and long-ti...The rapid advancement of machine learning based tight-binding Hamiltonian(MLTB)methods has opened new avenues for efficient and accurate electronic structure simulations,particularly in large-scale systems and long-time scenarios.This review begins with a concise overview of traditional tight-binding(TB)models,including both(semi-)empirical and first-principles approaches,establishing the foundation for understanding MLTB developments.We then present a systematic classification of existing MLTB methodologies,grouped into two major categories:direct prediction of TB Hamiltonian elements and inference of empirical parameters.A comparative analysis with other ML-based electronic structure models is also provided,highlighting the advancement of MLTB approaches.Finally,we explore the emerging MLTB application ecosystem,highlighting how the integration of MLTB models with a diverse suite of post-processing tools from linear-scaling solvers to quantum transport frameworks and molecular dynamics interfaces is essential for tackling complex scientific problems across different domains.The continued advancement of this integrated paradigm promises to accelerate materials discovery and open new frontiers in the predictive simulation of complex quantum phenomena.展开更多
By adopting stochastic density functional theory(SDFT)and mixed stochastic-deterministic density functional theory(MDFT)methods,we perform first-principles calculations to predict the shock Hugoniot curves of boron(pr...By adopting stochastic density functional theory(SDFT)and mixed stochastic-deterministic density functional theory(MDFT)methods,we perform first-principles calculations to predict the shock Hugoniot curves of boron(pressure P=7.9×10^(3)-1.6×10^(6) GPa and temperature T=25-2800 eV),silicon(P=2.6×10^(3)-7.9×10^(5) GPa and T=21.5-1393 eV),and aluminum(P=5.2×10^(3)-9.0×10^(5) GPa and T=25-1393 eV)over wide ranges of pressure and temperature.In particular,we systematically investigate the impact of different cutoff radii in norm-conserving pseudopotentials on the calculated properties at elevated temperatures,such as pressure,ionization energy,and equation of state.By comparing the SDFT and MDFT results with those of other first-principles methods,such as extended first-principles molecular dynamics and path integral Monte Carlo methods,we find that the SDFT and MDFT methods show satisfactory precision,which advances our understanding of first-principles methods when applied to studies of matter at extremely high pressures and temperatures.展开更多
Solid electrolytes with fast ion transport are crucial for solid state lithium metal batteries.Chemical doping has been the most effective strategy for improving ion condictiviy,and atomistic simulation with machine-l...Solid electrolytes with fast ion transport are crucial for solid state lithium metal batteries.Chemical doping has been the most effective strategy for improving ion condictiviy,and atomistic simulation with machine-learning potentials helps optimize doping by predicting ion conductivity for various composition.Yet most existing machine-learning models are trained on narrow chemistry,requiring retraining for each new system,which wastes transferable knowledge and incurs significant cost.Here,we propose a pre-trained deep potential model purpose-built for sulfide solid electrolytes with attention mechanism,known as DPA-SSE.The training set includes 15 elements and consists of both equilibrium and extensive out-of-equilibrium configurations.DPA-SSE achieves a high energy resolution of less than 2 meV/atom for dynamical trajectories up to 1150 K,and reproduces experimental ion conductivity with remarkable accuracy.DPA-SSE generalizes well to complex electrolytes with mixes of cation and anion atoms,and enables highly efficient dynamical simulation via model distillation.DPA-SSE also serves as a platform for continuous learning and can be fine-tuned with minimal downstream data.These results demonstrate the possibility of a new pathway for the AIdriven development of solid electrolytes with exceptional performance.展开更多
Solid-state batteries have received increasing attention in scientific and industrial communities,which benefits from the intrinsically safe solid electrolytes(SEs).Although much effort has been devoted to designing S...Solid-state batteries have received increasing attention in scientific and industrial communities,which benefits from the intrinsically safe solid electrolytes(SEs).Although much effort has been devoted to designing SEs with high ionic conductivities,it is extremely difficult to fully understand the ionic diffusion mechanisms in SEs through conventional experimental and theoretical methods.Herein,the temperature-dependent concerted diffusion mechanism of ions in SEs is explored through machinelearning molecular dynamics,taking Li_(10)GeP_(2)S_(12) as a prototype.Weaker diffusion anisotropy,more disordered Li distributions,and shorter residence time are observed at a higher temperature.Arrhenius-type temperature dependence is maintained within a wide temperature range,which is attributed to the linear temperature dependence of jump frequencies of various concerted diffusion modes.These results provide a theoretical framework to understand the ionic diffusion mechanisms in SEs and deepen the understanding of the chemical origin of temperature-dependent concerted diffusions in SEs.展开更多
Multi-component alloys have demonstrated excellent performance in various applications,but the vast range of possible compositions and microstructures makes it challenging to identify optimized alloys for specific pur...Multi-component alloys have demonstrated excellent performance in various applications,but the vast range of possible compositions and microstructures makes it challenging to identify optimized alloys for specific purposes.To overcome this challenge,large-scale atomic simulation techniques have been widely used for the design and optimization of multi-component alloys.The capability and reliability of large-scale atomic simulations essentially rely on the quality of interatomic potentials that describe the interactions between atoms.This work provides a comprehensive summary of the latest advances in atomic simulation techniques for multi-component alloys.The focus is on interatomic potentials,including both conventional empirical potentials and newly developed machine learning potentials(MLPs).The fitting processes for different types of interatomic potentials applied to multi-component alloys are also discussed.Finally,the challenges and future perspectives in developing MLPs are thoroughly addressed.Overall,this review provides a valuable resource for researchers interested in developing optimized multicomponent alloys using atomic simulation techniques.展开更多
To fill the gap between accurate(and expensive)ab initio calculations and efficient atomistic simulations based on empirical interatomic potentials,a new class of descriptions of atomic interactions has emerged and be...To fill the gap between accurate(and expensive)ab initio calculations and efficient atomistic simulations based on empirical interatomic potentials,a new class of descriptions of atomic interactions has emerged and been widely applied;i.e.machine learning potentials(MLPs).One recently developed type of MLP is the deep potential(DP)method.In this review,we provide an introduction to DP methods in computational materials science.The theory underlying the DP method is presented along with a step-by-step introduction to their development and use.We also review materials applications of DPs in a wide range of materials systems.The DP Library provides a platform for the development of DPs and a database of extant DPs.We discuss the accuracy and efficiency of DPs compared with ab initio methods and empirical potentials.展开更多
Machine learning-assisted modeling of the inter-atomic potential energy surface(PES)is revolutionizing the field ofmolecular simulation.With the accumulation of high-quality electronic structure data,a model that can ...Machine learning-assisted modeling of the inter-atomic potential energy surface(PES)is revolutionizing the field ofmolecular simulation.With the accumulation of high-quality electronic structure data,a model that can be pretrained on all available data and finetuned on downstream tasks with a small additional effort would bring the field to a new stage.Here we propose DPA-1,a Deep Potentialmodel with a gated attentionmechanism,which is highly effective for representing the conformation and chemical spaces of atomic systems and learning the PES.We tested DPA-1 on a number of systems and observed superior performance compared with existing benchmarks.When pretrained on large-scale datasets containing 56 elements,DPA-1 can be successfully applied to various downstream tasks with a great improvement of sample efficiency.Surprisingly,for different elements,the learned type embedding parameters form a spiral in the latent space and have a natural correspondence with their positions on the periodic table,showing interesting interpretability of the pretrained DPA-1 model.展开更多
Machine learning has been widely used for solving partial differential equations(PDEs)in recent years,among which the random feature method(RFM)exhibits spectral accuracy and can compete with traditional solvers in te...Machine learning has been widely used for solving partial differential equations(PDEs)in recent years,among which the random feature method(RFM)exhibits spectral accuracy and can compete with traditional solvers in terms of both accuracy and efficiency.Potentially,the optimization problem in the RFM is more difficult to solve than those that arise in traditional methods.Unlike the broader machine-learning research,which frequently targets tasks within the low-precision regime,our study focuses on the high-precision regime crucial for solving PDEs.In this work,we study this problem from the following aspects:(i)we analyze the coeffcient matrix that arises in the RFM by studying the distribution of singular values;(ii)we investigate whether the continuous training causes the overfitting issue;(ii)we test direct and iterative methods as well as randomized methods for solving the optimization problem.Based on these results,we find that direct methods are superior to other methods if memory is not an issue,while iterative methods typically have low accuracy and can be improved by preconditioning to some extent.展开更多
Minimizing disorder and defects is crucial for realizing the full potential of two-dimensional transition metal dichalcogenides(TMDs) materials and improving device performance to desired properties. However, the meth...Minimizing disorder and defects is crucial for realizing the full potential of two-dimensional transition metal dichalcogenides(TMDs) materials and improving device performance to desired properties. However, the methods in defect controlcurrently face challenges with overly large operational areas and a lack of precision in targeting specific defects. Therefore,we propose a new method for the precise and universal defect healing of TMD materials, integrating real-time imaging withscanning transmission electron microscopy (STEM). This method employs electron beam irradiation to stimulate the diffusionmigration of surface-adsorbed adatoms on TMD materials grown by low-temperature molecular beam epitaxy (MBE),and heal defects within the diffusion range. This approach covers defect repairs ranging from zero-dimensional vacancydefects to two-dimensional grain orientation alignment, demonstrating its universality in terms of the types of samples anddefects. These findings offer insights into the use of atomic-level focused electron beams at appropriate voltages in STEMfor defect healing, providing valuable experience for achieving atomic-level precise fabrication of TMD materials.展开更多
The escalating energy crisis has spurred extensive research into organic compounds for energyefficient applications,taking advantage of their environmental friendliness,cost-effective synthesis,and adaptable molecular...The escalating energy crisis has spurred extensive research into organic compounds for energyefficient applications,taking advantage of their environmental friendliness,cost-effective synthesis,and adaptable molecular structures.Traditional trial-and-error methods for discovering highly functional organic compounds are expensive and time-consuming.We employed a 3D transformerbased molecular representation learning algorithm to create the Org-Mol pre-trained model,using 60 million semi-empirically optimized small organic molecule structures.After fine-tuning with public experimental data,the model can accurately predict various physical properties of pure organics,with test set R2 values exceeding 0.92.These fine-tuned models are used in high-throughput screening among millions of ester molecules to identify novel immersion coolants,resulting in the experimental validation of two promising candidates.This work not only demonstrates the potential of Org-Mol in predicting bulk properties for pure organic compounds but also paves the way for the rational and efficient development of ideal candidates for energy-saving materials.展开更多
Platinum complexes are one of the most promising emitters for organic light-emitting diodes(OLEDs)and they have been studied for decades.However,it is still challenging to realize high-performance solution-processed O...Platinum complexes are one of the most promising emitters for organic light-emitting diodes(OLEDs)and they have been studied for decades.However,it is still challenging to realize high-performance solution-processed OLEDs by using the platinum complex as an emissive dopant.Herein,a nonplanar terdentate[Pt(N^C^N)Cl]emitter(Pt1)with good solubility and film-forming ability,ultrahigh photoluminescence quantum yield(98%in PMMA film),and short excited-state lifetime(1.5μs in solution)is presented to achieve this goal.This complex is designed and synthesized by using a nonplanar ligand strategy.Solution-processed OLEDs of Pt1 are successfully fabricated to show a _(max)imum external quantum efficiency(EQE_(max))of 16.14%,a _(max)imum current efficiency(CE_(max))of 46.36 cd A^(-1),a _(max)imum luminance(L_(max))of 13990 cd m^(-2),and operating half-lifetime(LT_(50))of 288 min,respectively,representing the highest performance recorded so far based on the terdentate platinum complexes.More importantly,the large-area(144 mm^(2))OLEDs of Pt1 with high EQE_(max) of 10.28%and uniform luminance of 8076 cd m^(-2)have been achieved,indicating a high potential of Pt1 for commercial applications.展开更多
In this paper,we propose a machine learning approach via model-operatordata network(MOD-Net)for solving PDEs.A MOD-Net is driven by a model to solve PDEs based on operator representationwith regularization fromdata.Fo...In this paper,we propose a machine learning approach via model-operatordata network(MOD-Net)for solving PDEs.A MOD-Net is driven by a model to solve PDEs based on operator representationwith regularization fromdata.For linear PDEs,we use a DNN to parameterize the Green’s function and obtain the neural operator to approximate the solution according to the Green’s method.To train the DNN,the empirical risk consists of the mean squared loss with the least square formulation or the variational formulation of the governing equation and boundary conditions.For complicated problems,the empirical risk also includes a fewlabels,which are computed on coarse grid points with cheap computation cost and significantly improves the model accuracy.Intuitively,the labeled dataset works as a regularization in addition to the model constraints.The MOD-Net solves a family of PDEs rather than a specific one and is much more efficient than original neural operator because few expensive labels are required.We numerically show MOD-Net is very efficient in solving Poisson equation and one-dimensional radiative transfer equation.For nonlinear PDEs,the nonlinear MOD-Net can be similarly used as an ansatz for solving nonlinear PDEs,exemplified by solving several nonlinear PDE problems,such as the Burgers equation.展开更多
Accurate prediction of protein-ligand complex structures is a crucial step in structure-based drug design.Traditional molecular docking methods exhibit limitations in terms of accuracy and sampling space,while relying...Accurate prediction of protein-ligand complex structures is a crucial step in structure-based drug design.Traditional molecular docking methods exhibit limitations in terms of accuracy and sampling space,while relying on machine-learning approaches may lead to invalid conformations.In this study,we propose a novel strategy that combines molecular docking and machine learning methods.Firstly,the protein-ligand binding poses are predicted using a deep learning model.Subsequently,position-restricted docking on predicted binding poses is performed using Uni-Dock,generating physically constrained and valid binding poses.Finally,the binding poses are re-scored and ranked using machine learning scoring functions.This strategy harnesses the predictive power of machine learning and the physical constraints advantage of molecular docking.Evaluation experiments on multiple datasets demonstrate that,compared to using molecular docking or machine learning methods alone,our proposed strategy can significantly improve the success rate and accuracy of protein-ligand complex structure predictions.展开更多
The ability to rapidly evaluate materials properties through atomistic simulation approaches is the foundation of many new artificial intelligence-based approaches to materials identification and design.This depends o...The ability to rapidly evaluate materials properties through atomistic simulation approaches is the foundation of many new artificial intelligence-based approaches to materials identification and design.This depends on the availability of accurate descriptions of atomic bonding and an efficient means for determining materials properties.We present an efficient,robust platform for calculating materials properties from a wide-range of atomic bonding descriptions,i.e.,APEX,the Alloy Property Explorer.APEX enables the rapid evolution of interatomic potential development and optimization,which is of particular importance in fine-tuning new classes of general AI-based foundation models for applications in materials science and engineering.APEX is an open-source,extendable,cloud-native platform for material property calculations using a range of atomistic simulation methodologies that effectively manages diverse computational resources and is built upon user-friendly features including automatic results visualization,a web-based platform and a NoSQL database client.It is designed for expert and non-specialist users,lowering the barrier to entry for interdisciplinary research within an“AI for Materials”framework.We describe the foundation and use of APEX,as well as provide two examples of its application to properties of titanium and 179 metals and alloys for a wide-range of bonding descriptions.展开更多
Quantum transport simulations are essential for understanding and designing nanoelectronic devices,yet the long-standing trade-off between accuracy and computational efficiency has limited their practical applications...Quantum transport simulations are essential for understanding and designing nanoelectronic devices,yet the long-standing trade-off between accuracy and computational efficiency has limited their practical applications.We present DeePTB-NEGF,an integrated framework combining deep learning tight-binding Hamiltonian prediction with non-equilibrium Green’s function methodology to enable accurate quantum transport simulations in open boundary conditions with 2–3 orders of magnitude acceleration.We demonstrate DeePTB-NEGF through two challenging applications:comprehensive break junction simulations with over 104 snapshots,showing excellent agreement with experimental conductance histograms;and carbon nanotube field-effect transistors(CNT-FETs)at experimental dimensions,reproducing measured transfer characteristics for a 41 nm channel CNT-FET(~8000 atoms,3×10^(4)orbitals)and predicting zero-bias transmission spectra for a 180 nm CNT(~3×10^(4)atoms,10^(5)orbitals),showcasing the framework’s capability for large-scale device simulations.Our systematic studies across varying geometries confirm the necessity of simulating realistic experimental structures for precise predictions.DeePTB-NEGF bridges the longstanding gap between first-principles accuracy and computational efficiency,providing a scalable tool for highthroughput and large-scale quantum transport simulations that enable previously inaccessible nanoscale device investigations.展开更多
The rational design of organic functional devices relies on understanding structure-propertyperformance relationships through multi-scale characterization.However,traditional characterizations are costly and require m...The rational design of organic functional devices relies on understanding structure-propertyperformance relationships through multi-scale characterization.However,traditional characterizations are costly and require multidisciplinary expertise.Here we present OCNet,a domain-knowledge-enhanced representation learning framework that,for the first time,enables unified virtual characterization from molecules to devices.Pre-trained on over ten million selfgenerated conjugated molecules and dimers,OCNet learns generalizable microscopic representations comparable to expert-crafted features.As a result,it surpasses state-of-the-art models by over 20%in predicting key computed and experimental molecular optoelectronic properties.OCNet further provides the first transferable model for predicting transfer integrals in thin films,enabling accurate mesoscale carrier mobility estimation via multiscale simulations.By integrating tight-binding-level electronic descriptors,OCNet achieves near real-time,accurate prediction of device power conversion efficiency.Together,OCNet offers a unified and scalable foundation for virtual characterization of organic materials across multiple scales,with broad applicability in photovoltaics,displays,and sensing.展开更多
We computationally investigated the molecular aggregation effects on the excited state deactivation processes by considering both the direct vibrational relaxation and the S0/S1 surface crossing,that is,the minimum en...We computationally investigated the molecular aggregation effects on the excited state deactivation processes by considering both the direct vibrational relaxation and the S0/S1 surface crossing,that is,the minimum energy conical intersection(MECI).Taking classical AIEgens bis(piperidyl)anthracenes(BPAs)isomers and the substituted silole derivatives as examples,we show that the deformation ofMECI always occurs at the atom with greater hole/electron overlap.Besides,the energetic and structural changes of MECI caused by substituent has been investigated.We find that effective substituent such as the addition of the electron-donating groups,which can polarize the distribution of hole/electron density of molecules,will lead to the pyramidalization deformation of MECI occurring at the substituent position and simultaneously reduce the required energy to reach MECI.And MECI is sterically restricted by the surrounding molecules in solid phase,which remarkably hinders the non-radiative decay through surface crossing.Through quantitative computational assessments of the fluorescence quantum efficiency for both solution and solid phases,we elucidate the role of MECI and its dependence on the substitutions through pyramidalization deformation,which give rise to the aggregation-induced emission(AIE)phenomenon for 9,10-BPA,to aggregation-enhance emission(AEE)behavior for 1,4-BPA,and to conventional aggregation-caused quenching(ACQ)for 1,5-BPA.We further verify such mechanism for siloles,for which we found that the substitutions do not change the AIE behavior.Our findings render a general molecular design approach to manipulating the aggregation effect for optical emission.展开更多
Molecular dynamics(MD)is an indispensable atomistic-scale computational tool widely-used in various disciplines.In the past decades,nearly all ab initio MD and machine-learning MD have been based on the general-purpos...Molecular dynamics(MD)is an indispensable atomistic-scale computational tool widely-used in various disciplines.In the past decades,nearly all ab initio MD and machine-learning MD have been based on the general-purpose central/graphics processing units(CPU/GPU),which are well-known to suffer from their intrinsic“memory wall”and“power wall”bottlenecks.Consequently,nowadays MD calculations with ab initio accuracy are extremely time-consuming and power-consuming,imposing serious restrictions on the MD simulation size and duration.To solve this problem,here we propose a special-purpose MD processing unit(MDPU),which could reduce MD time and power consumption by about 103 times(109 times)compared to state-of-the-art machine-learningMD(ab initio MD)based on CPU/GPU,while keeping ab initio accuracy.With significantly-enhanced performance,the proposed MDPU may pave a way for the accurate atomistic-scale analysis of large-size and/or longduration problems which were impossible/impractical to compute before.展开更多
The rapid advancements in artificial intelligence(AI)are catalyzing transformative changes in atomic modeling,simulation,and design.AI-driven potential energy models havedemonstrated the capability to conduct large-sc...The rapid advancements in artificial intelligence(AI)are catalyzing transformative changes in atomic modeling,simulation,and design.AI-driven potential energy models havedemonstrated the capability to conduct large-scale,long-duration simulations with the accuracy of ab initio electronic structure methods.However,the model generation process remains a bottleneck for large-scale applications.We propose a shift towards a model-centric ecosystem,wherein a large atomic model(LAM),pretrained across multiple disciplines,can be efficiently fine-tuned and distilled for various downstream tasks,thereby establishing a new framework for molecular modeling.In this study,we introduce the DPA-2 architecture as a prototype for LAMs.Pre-trained on a diverse array of chemical and materials systemsusing a multi-task approach,DPA-2demonstrates superior generalization capabilities across multiple downstream tasks compared to the traditional single-task pre-training and fine-tuning methodologies.Our approach sets the stage for the development and broad application of LAMs in molecular and materials simulation research.展开更多
基金funding support from the US National Science Foundation(2229092)supported by the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship,a program of Schmidt Sciences,LLC.
文摘Recent advances in artificial intelligence(AI)have led to the development of sophisticated algorithms that significantly improve image analysis capabilities.This combination of AI and microscopic imaging is transforming the way we interpret and analyze imaging data,simplifying complex tasks and enabling innovative experimental methods previously thought impossible.In smart manufacturing,these improvements are especially impactful,increasing precision and efficiency in production processes.This review examines the convergence of AI with particle image analysis,an area we refer to as“particle vision analysis(PVA).”We offer a detailed overview of how this technology integrates into and impacts various fields within the physical sciences and materials sectors,where it plays a crucial role in both innovation and operational improvements.We explore four key areas of advancement-namely,particle classification,detection,segmentation,and object tracking-along with a look into the emerging field of augmented microscopy.This paper also underscores the vital role of the existing datasets and implementations that support these applications,which provide essential insights and resources that drive continuous research and development in this fast-evolving field.Our thorough analysis aims to outline the transformative potential of AI-driven PVA in improving precision in future manufacturing at the microscopic scale and thereby preparing the ground for significant technological progress and broad industrial applications in nanomanufacturing,biomanufacturing,and pharmaceutical manufacturing.This exploration not only highlights the advantages of integrating AI into conventional manufacturing processes but also anticipates the rise of next-generation smart manufacturing,which is set to revolutionize industry standards and operational practices.
基金supported by the Advanced Materials-National Science and Technology Major Project(Grant No.2025ZD0618401)the National Natural Science Foundation of China(Grant No.12504285)+1 种基金the Natural Science Foundation of Jiangsu Province(Grant No.BK20250472)NFSG grant from BITS-Pilani,Dubai campus。
文摘The rapid advancement of machine learning based tight-binding Hamiltonian(MLTB)methods has opened new avenues for efficient and accurate electronic structure simulations,particularly in large-scale systems and long-time scenarios.This review begins with a concise overview of traditional tight-binding(TB)models,including both(semi-)empirical and first-principles approaches,establishing the foundation for understanding MLTB developments.We then present a systematic classification of existing MLTB methodologies,grouped into two major categories:direct prediction of TB Hamiltonian elements and inference of empirical parameters.A comparative analysis with other ML-based electronic structure models is also provided,highlighting the advancement of MLTB approaches.Finally,we explore the emerging MLTB application ecosystem,highlighting how the integration of MLTB models with a diverse suite of post-processing tools from linear-scaling solvers to quantum transport frameworks and molecular dynamics interfaces is essential for tackling complex scientific problems across different domains.The continued advancement of this integrated paradigm promises to accelerate materials discovery and open new frontiers in the predictive simulation of complex quantum phenomena.
基金supported by the National Key R&D Program of China under Grant No.2025YFB3003603the National Natural Science Foundation of China under Grant Nos.12135002 and 12105209.
文摘By adopting stochastic density functional theory(SDFT)and mixed stochastic-deterministic density functional theory(MDFT)methods,we perform first-principles calculations to predict the shock Hugoniot curves of boron(pressure P=7.9×10^(3)-1.6×10^(6) GPa and temperature T=25-2800 eV),silicon(P=2.6×10^(3)-7.9×10^(5) GPa and T=21.5-1393 eV),and aluminum(P=5.2×10^(3)-9.0×10^(5) GPa and T=25-1393 eV)over wide ranges of pressure and temperature.In particular,we systematically investigate the impact of different cutoff radii in norm-conserving pseudopotentials on the calculated properties at elevated temperatures,such as pressure,ionization energy,and equation of state.By comparing the SDFT and MDFT results with those of other first-principles methods,such as extended first-principles molecular dynamics and path integral Monte Carlo methods,we find that the SDFT and MDFT methods show satisfactory precision,which advances our understanding of first-principles methods when applied to studies of matter at extremely high pressures and temperatures.
基金supported in part by the National Science and Technology Major Project(Grants No.2023ZD0120702)Key Research Program of Frontier Sciences of CAS(Grant No.ZDBS-LY-SLH008)National Nature Science Foundation of China(Grants No.12304049).We thank Bowen Deng and Dr.Peichen Zhong,the authors of CHGNet,for inspiring discussions.We also appreciate valuable advice by Dr.Qisheng Wu.The computational resource was supported by the Bohrium Cloud Platform at DP technology.
文摘Solid electrolytes with fast ion transport are crucial for solid state lithium metal batteries.Chemical doping has been the most effective strategy for improving ion condictiviy,and atomistic simulation with machine-learning potentials helps optimize doping by predicting ion conductivity for various composition.Yet most existing machine-learning models are trained on narrow chemistry,requiring retraining for each new system,which wastes transferable knowledge and incurs significant cost.Here,we propose a pre-trained deep potential model purpose-built for sulfide solid electrolytes with attention mechanism,known as DPA-SSE.The training set includes 15 elements and consists of both equilibrium and extensive out-of-equilibrium configurations.DPA-SSE achieves a high energy resolution of less than 2 meV/atom for dynamical trajectories up to 1150 K,and reproduces experimental ion conductivity with remarkable accuracy.DPA-SSE generalizes well to complex electrolytes with mixes of cation and anion atoms,and enables highly efficient dynamical simulation via model distillation.DPA-SSE also serves as a platform for continuous learning and can be fine-tuned with minimal downstream data.These results demonstrate the possibility of a new pathway for the AIdriven development of solid electrolytes with exceptional performance.
基金supported by the National Key Research and Development Program(2021YFB2500210)the Beijing Municipal Natural Science Foundation(Z20J00043)+4 种基金the National Natural Science Foundation of China(22109086 and 21825501)the China Postdoctoral Science Foundation(2021TQ0161 and 2021 M691709)the Guoqiang Institute at Tsinghua University(2020GQG1006)the support from the Shuimu Tsinghua Scholar Program of Tsinghua Universitythe support from the Tsinghua National Laboratory for Information Science and Technology for theoretical simulations。
文摘Solid-state batteries have received increasing attention in scientific and industrial communities,which benefits from the intrinsically safe solid electrolytes(SEs).Although much effort has been devoted to designing SEs with high ionic conductivities,it is extremely difficult to fully understand the ionic diffusion mechanisms in SEs through conventional experimental and theoretical methods.Herein,the temperature-dependent concerted diffusion mechanism of ions in SEs is explored through machinelearning molecular dynamics,taking Li_(10)GeP_(2)S_(12) as a prototype.Weaker diffusion anisotropy,more disordered Li distributions,and shorter residence time are observed at a higher temperature.Arrhenius-type temperature dependence is maintained within a wide temperature range,which is attributed to the linear temperature dependence of jump frequencies of various concerted diffusion modes.These results provide a theoretical framework to understand the ionic diffusion mechanisms in SEs and deepen the understanding of the chemical origin of temperature-dependent concerted diffusions in SEs.
基金the National Key Research and Development Program of China(No.2022YFB3709000)the National Natural Science Foundation of China(Nos.52122408,52071023,52101019,and 51901013)the Fundamental Research Funds for the Central Universities(University of Science and Technology Beijing,Nos.06500135 and FRF-TP-2021-04C1).
文摘Multi-component alloys have demonstrated excellent performance in various applications,but the vast range of possible compositions and microstructures makes it challenging to identify optimized alloys for specific purposes.To overcome this challenge,large-scale atomic simulation techniques have been widely used for the design and optimization of multi-component alloys.The capability and reliability of large-scale atomic simulations essentially rely on the quality of interatomic potentials that describe the interactions between atoms.This work provides a comprehensive summary of the latest advances in atomic simulation techniques for multi-component alloys.The focus is on interatomic potentials,including both conventional empirical potentials and newly developed machine learning potentials(MLPs).The fitting processes for different types of interatomic potentials applied to multi-component alloys are also discussed.Finally,the challenges and future perspectives in developing MLPs are thoroughly addressed.Overall,this review provides a valuable resource for researchers interested in developing optimized multicomponent alloys using atomic simulation techniques.
基金T W and D J S gratefully acknowledge the support of the Research Grants Council,Hong Kong SAR,through the Collaborative Research Fund Project No.8730054The work of H W is supported by the National Science Foundation of China under Grant Nos.11871110 and 12122103The work of W E is supported in part by a gift from iFlytek to Princeton University。
文摘To fill the gap between accurate(and expensive)ab initio calculations and efficient atomistic simulations based on empirical interatomic potentials,a new class of descriptions of atomic interactions has emerged and been widely applied;i.e.machine learning potentials(MLPs).One recently developed type of MLP is the deep potential(DP)method.In this review,we provide an introduction to DP methods in computational materials science.The theory underlying the DP method is presented along with a step-by-step introduction to their development and use.We also review materials applications of DPs in a wide range of materials systems.The DP Library provides a platform for the development of DPs and a database of extant DPs.We discuss the accuracy and efficiency of DPs compared with ab initio methods and empirical potentials.
基金supported by the National Key R&D Program of China under Grant No.2022YFA1004300the National Natural Science Foundation of China under Grant No.12122103supported by the Bohrium Cloud Platform at DP technology.
文摘Machine learning-assisted modeling of the inter-atomic potential energy surface(PES)is revolutionizing the field ofmolecular simulation.With the accumulation of high-quality electronic structure data,a model that can be pretrained on all available data and finetuned on downstream tasks with a small additional effort would bring the field to a new stage.Here we propose DPA-1,a Deep Potentialmodel with a gated attentionmechanism,which is highly effective for representing the conformation and chemical spaces of atomic systems and learning the PES.We tested DPA-1 on a number of systems and observed superior performance compared with existing benchmarks.When pretrained on large-scale datasets containing 56 elements,DPA-1 can be successfully applied to various downstream tasks with a great improvement of sample efficiency.Surprisingly,for different elements,the learned type embedding parameters form a spiral in the latent space and have a natural correspondence with their positions on the periodic table,showing interesting interpretability of the pretrained DPA-1 model.
基金supported by the NSFC Major Research Plan--Interpretable and Generalpurpose Next-generation Artificial Intelligence(No.92370205).
文摘Machine learning has been widely used for solving partial differential equations(PDEs)in recent years,among which the random feature method(RFM)exhibits spectral accuracy and can compete with traditional solvers in terms of both accuracy and efficiency.Potentially,the optimization problem in the RFM is more difficult to solve than those that arise in traditional methods.Unlike the broader machine-learning research,which frequently targets tasks within the low-precision regime,our study focuses on the high-precision regime crucial for solving PDEs.In this work,we study this problem from the following aspects:(i)we analyze the coeffcient matrix that arises in the RFM by studying the distribution of singular values;(ii)we investigate whether the continuous training causes the overfitting issue;(ii)we test direct and iterative methods as well as randomized methods for solving the optimization problem.Based on these results,we find that direct methods are superior to other methods if memory is not an issue,while iterative methods typically have low accuracy and can be improved by preconditioning to some extent.
基金the Beijing Natural Science Foundation(Grant Nos.JQ24010 and Z220020)the Fundamental Research Funds for the Central Universities,and the National Natural Science Foundation of China(Grant No.52273279)Project supported by the Electron Microscopy Laboratory of Peking University,China for the use of Nion U-HERMES200 scanning transmission electron microscopy.We thank Materials Processing and Analysis Center,Peking University,for assistance with TEM characterization.The electron microscopy work was through a user project at Center of Oak Ridge National Laboratory(ORNL)for Nanophase Materials Sciences(CNMS),which is a DOE Office of Science User Facility.
文摘Minimizing disorder and defects is crucial for realizing the full potential of two-dimensional transition metal dichalcogenides(TMDs) materials and improving device performance to desired properties. However, the methods in defect controlcurrently face challenges with overly large operational areas and a lack of precision in targeting specific defects. Therefore,we propose a new method for the precise and universal defect healing of TMD materials, integrating real-time imaging withscanning transmission electron microscopy (STEM). This method employs electron beam irradiation to stimulate the diffusionmigration of surface-adsorbed adatoms on TMD materials grown by low-temperature molecular beam epitaxy (MBE),and heal defects within the diffusion range. This approach covers defect repairs ranging from zero-dimensional vacancydefects to two-dimensional grain orientation alignment, demonstrating its universality in terms of the types of samples anddefects. These findings offer insights into the use of atomic-level focused electron beams at appropriate voltages in STEMfor defect healing, providing valuable experience for achieving atomic-level precise fabrication of TMD materials.
基金supported by research grants from China Petroleum&Chemical Corp(funding number 124014)the financial support from the National Key R&D Program of China(Grant No.2024YFA1510200).
文摘The escalating energy crisis has spurred extensive research into organic compounds for energyefficient applications,taking advantage of their environmental friendliness,cost-effective synthesis,and adaptable molecular structures.Traditional trial-and-error methods for discovering highly functional organic compounds are expensive and time-consuming.We employed a 3D transformerbased molecular representation learning algorithm to create the Org-Mol pre-trained model,using 60 million semi-empirically optimized small organic molecule structures.After fine-tuning with public experimental data,the model can accurately predict various physical properties of pure organics,with test set R2 values exceeding 0.92.These fine-tuned models are used in high-throughput screening among millions of ester molecules to identify novel immersion coolants,resulting in the experimental validation of two promising candidates.This work not only demonstrates the potential of Org-Mol in predicting bulk properties for pure organic compounds but also paves the way for the rational and efficient development of ideal candidates for energy-saving materials.
基金supported by the National Natural Science Foundation of China(22004041,21925112,22090021,and 22305251)the National Key R&D Program of China(2023YFE0125200).
文摘Platinum complexes are one of the most promising emitters for organic light-emitting diodes(OLEDs)and they have been studied for decades.However,it is still challenging to realize high-performance solution-processed OLEDs by using the platinum complex as an emissive dopant.Herein,a nonplanar terdentate[Pt(N^C^N)Cl]emitter(Pt1)with good solubility and film-forming ability,ultrahigh photoluminescence quantum yield(98%in PMMA film),and short excited-state lifetime(1.5μs in solution)is presented to achieve this goal.This complex is designed and synthesized by using a nonplanar ligand strategy.Solution-processed OLEDs of Pt1 are successfully fabricated to show a _(max)imum external quantum efficiency(EQE_(max))of 16.14%,a _(max)imum current efficiency(CE_(max))of 46.36 cd A^(-1),a _(max)imum luminance(L_(max))of 13990 cd m^(-2),and operating half-lifetime(LT_(50))of 288 min,respectively,representing the highest performance recorded so far based on the terdentate platinum complexes.More importantly,the large-area(144 mm^(2))OLEDs of Pt1 with high EQE_(max) of 10.28%and uniform luminance of 8076 cd m^(-2)have been achieved,indicating a high potential of Pt1 for commercial applications.
基金sponsored by the National Key R&D Program of China Grant No.2019YFA0709503(Z.X.)and No.2020YFA0712000(Z.M.)the Shanghai Sailing Program(Z.X.)+9 种基金the Natural Science Foundation of Shanghai Grant No.20ZR1429000(Z.X.)the National Natural Science Foundation of China Grant No.62002221(Z.X.)the National Natural Science Foundation of China Grant No.12101401(T.L.)the National Natural Science Foundation of China Grant No.12101402(Y.Z.)Shanghai Municipal of Science and Technology Project Grant No.20JC1419500(Y.Z.)the Lingang Laboratory Grant No.LG-QS-202202-08(Y.Z.)the National Natural Science Foundation of China Grant No.12031013(Z.M.)Shanghai Municipal of Science and Technology Major Project No.2021SHZDZX0102the HPC of School of Mathematical Sciencesthe Student Innovation Center at Shanghai Jiao Tong University.
文摘In this paper,we propose a machine learning approach via model-operatordata network(MOD-Net)for solving PDEs.A MOD-Net is driven by a model to solve PDEs based on operator representationwith regularization fromdata.For linear PDEs,we use a DNN to parameterize the Green’s function and obtain the neural operator to approximate the solution according to the Green’s method.To train the DNN,the empirical risk consists of the mean squared loss with the least square formulation or the variational formulation of the governing equation and boundary conditions.For complicated problems,the empirical risk also includes a fewlabels,which are computed on coarse grid points with cheap computation cost and significantly improves the model accuracy.Intuitively,the labeled dataset works as a regularization in addition to the model constraints.The MOD-Net solves a family of PDEs rather than a specific one and is much more efficient than original neural operator because few expensive labels are required.We numerically show MOD-Net is very efficient in solving Poisson equation and one-dimensional radiative transfer equation.For nonlinear PDEs,the nonlinear MOD-Net can be similarly used as an ansatz for solving nonlinear PDEs,exemplified by solving several nonlinear PDE problems,such as the Burgers equation.
基金supported by the National Key Research and Development Program of China(2022YFA1004302)
文摘Accurate prediction of protein-ligand complex structures is a crucial step in structure-based drug design.Traditional molecular docking methods exhibit limitations in terms of accuracy and sampling space,while relying on machine-learning approaches may lead to invalid conformations.In this study,we propose a novel strategy that combines molecular docking and machine learning methods.Firstly,the protein-ligand binding poses are predicted using a deep learning model.Subsequently,position-restricted docking on predicted binding poses is performed using Uni-Dock,generating physically constrained and valid binding poses.Finally,the binding poses are re-scored and ranked using machine learning scoring functions.This strategy harnesses the predictive power of machine learning and the physical constraints advantage of molecular docking.Evaluation experiments on multiple datasets demonstrate that,compared to using molecular docking or machine learning methods alone,our proposed strategy can significantly improve the success rate and accuracy of protein-ligand complex structure predictions.
基金supported by the Research Grants Council,Hong Kong SAR through the General Research Fund(17210723,17200424)the support of The University of Hong Kong via seed fund(2201100392)+2 种基金supported by the National Key R&D Program of China(Grant No.2022YFA1004300)the National Natural Science Foundation of China(Grant No.12122103)startup funding from Materials Innovation Institute for Life Sciences and Energy(MILES),HKU-SIRI in Shenzhen for support of this manuscript.
文摘The ability to rapidly evaluate materials properties through atomistic simulation approaches is the foundation of many new artificial intelligence-based approaches to materials identification and design.This depends on the availability of accurate descriptions of atomic bonding and an efficient means for determining materials properties.We present an efficient,robust platform for calculating materials properties from a wide-range of atomic bonding descriptions,i.e.,APEX,the Alloy Property Explorer.APEX enables the rapid evolution of interatomic potential development and optimization,which is of particular importance in fine-tuning new classes of general AI-based foundation models for applications in materials science and engineering.APEX is an open-source,extendable,cloud-native platform for material property calculations using a range of atomistic simulation methodologies that effectively manages diverse computational resources and is built upon user-friendly features including automatic results visualization,a web-based platform and a NoSQL database client.It is designed for expert and non-specialist users,lowering the barrier to entry for interdisciplinary research within an“AI for Materials”framework.We describe the foundation and use of APEX,as well as provide two examples of its application to properties of titanium and 179 metals and alloys for a wide-range of bonding descriptions.
基金supported by the Natural Science Foundation of Jiangsu Province(BK20250472)the Advanced Materials-National Science and Technology Major Project(2025ZD0618401)+2 种基金National Natural Science Foundation of China(12504285)National Key R&D Program of China(2024YFA1208203)Q.G.also acknowledges the funding support from the AI for Science Institute,Beijing(AISI).
文摘Quantum transport simulations are essential for understanding and designing nanoelectronic devices,yet the long-standing trade-off between accuracy and computational efficiency has limited their practical applications.We present DeePTB-NEGF,an integrated framework combining deep learning tight-binding Hamiltonian prediction with non-equilibrium Green’s function methodology to enable accurate quantum transport simulations in open boundary conditions with 2–3 orders of magnitude acceleration.We demonstrate DeePTB-NEGF through two challenging applications:comprehensive break junction simulations with over 104 snapshots,showing excellent agreement with experimental conductance histograms;and carbon nanotube field-effect transistors(CNT-FETs)at experimental dimensions,reproducing measured transfer characteristics for a 41 nm channel CNT-FET(~8000 atoms,3×10^(4)orbitals)and predicting zero-bias transmission spectra for a 180 nm CNT(~3×10^(4)atoms,10^(5)orbitals),showcasing the framework’s capability for large-scale device simulations.Our systematic studies across varying geometries confirm the necessity of simulating realistic experimental structures for precise predictions.DeePTB-NEGF bridges the longstanding gap between first-principles accuracy and computational efficiency,providing a scalable tool for highthroughput and large-scale quantum transport simulations that enable previously inaccessible nanoscale device investigations.
基金supported in part by NSFC’s Major Research Project 92270001Z.Z.’s work is supported in part by the Beijing Nova Program(20250484934).
文摘The rational design of organic functional devices relies on understanding structure-propertyperformance relationships through multi-scale characterization.However,traditional characterizations are costly and require multidisciplinary expertise.Here we present OCNet,a domain-knowledge-enhanced representation learning framework that,for the first time,enables unified virtual characterization from molecules to devices.Pre-trained on over ten million selfgenerated conjugated molecules and dimers,OCNet learns generalizable microscopic representations comparable to expert-crafted features.As a result,it surpasses state-of-the-art models by over 20%in predicting key computed and experimental molecular optoelectronic properties.OCNet further provides the first transferable model for predicting transfer integrals in thin films,enabling accurate mesoscale carrier mobility estimation via multiscale simulations.By integrating tight-binding-level electronic descriptors,OCNet achieves near real-time,accurate prediction of device power conversion efficiency.Together,OCNet offers a unified and scalable foundation for virtual characterization of organic materials across multiple scales,with broad applicability in photovoltaics,displays,and sensing.
基金National Natural Science Foundation of China,Grant/Award Numbers:21788102,2017YFA0204501。
文摘We computationally investigated the molecular aggregation effects on the excited state deactivation processes by considering both the direct vibrational relaxation and the S0/S1 surface crossing,that is,the minimum energy conical intersection(MECI).Taking classical AIEgens bis(piperidyl)anthracenes(BPAs)isomers and the substituted silole derivatives as examples,we show that the deformation ofMECI always occurs at the atom with greater hole/electron overlap.Besides,the energetic and structural changes of MECI caused by substituent has been investigated.We find that effective substituent such as the addition of the electron-donating groups,which can polarize the distribution of hole/electron density of molecules,will lead to the pyramidalization deformation of MECI occurring at the substituent position and simultaneously reduce the required energy to reach MECI.And MECI is sterically restricted by the surrounding molecules in solid phase,which remarkably hinders the non-radiative decay through surface crossing.Through quantitative computational assessments of the fluorescence quantum efficiency for both solution and solid phases,we elucidate the role of MECI and its dependence on the substitutions through pyramidalization deformation,which give rise to the aggregation-induced emission(AIE)phenomenon for 9,10-BPA,to aggregation-enhance emission(AEE)behavior for 1,4-BPA,and to conventional aggregation-caused quenching(ACQ)for 1,5-BPA.We further verify such mechanism for siloles,for which we found that the substitutions do not change the AIE behavior.Our findings render a general molecular design approach to manipulating the aggregation effect for optical emission.
基金supported by the National Natural Science Foundation of China(62474058 and 61804049)the Yuelushan Center for Industrial Innovation(2023YCII0104)+2 种基金the Huxiang High Level Talent Gathering Project(2019RS1023)the Technology Innovation and Entrepreneurship Funds of Hunan Province,P.R.China(2019GK5029)the Fund for Distinguished Young Scholars of Changsha(kq1905012).
文摘Molecular dynamics(MD)is an indispensable atomistic-scale computational tool widely-used in various disciplines.In the past decades,nearly all ab initio MD and machine-learning MD have been based on the general-purpose central/graphics processing units(CPU/GPU),which are well-known to suffer from their intrinsic“memory wall”and“power wall”bottlenecks.Consequently,nowadays MD calculations with ab initio accuracy are extremely time-consuming and power-consuming,imposing serious restrictions on the MD simulation size and duration.To solve this problem,here we propose a special-purpose MD processing unit(MDPU),which could reduce MD time and power consumption by about 103 times(109 times)compared to state-of-the-art machine-learningMD(ab initio MD)based on CPU/GPU,while keeping ab initio accuracy.With significantly-enhanced performance,the proposed MDPU may pave a way for the accurate atomistic-scale analysis of large-size and/or longduration problems which were impossible/impractical to compute before.
基金supported by the National Key R&D Program of China(grantno.2022YFA1004300)the National Natural Science Foundation of China(grant no.12122103)+11 种基金supported by the National Key Research and Development Project of China(grant no.2022YFA1004302)the National Natural Science Foundation of China(grants nos.92270001 and 12288101)supported by the National Institutes of Health(grant no.GM107485 to D.M.Y.)the National Science Foundation(grant no.2209718 to D.M.Y.)supported by the Natural Science Foundation of Zhejiang Province(grant no.2022XHSJJ006)supported by the National Natural Science Foundation of China(grants nos.22222303 and 22173032)supported by the National Key R&D Program of China(grants nos.2021YFA0718900 and 2022YFA1403000)supported by the National Natural Science Foundation of China(grants nos.12034009 and 91961204)supported by the National Science Fund for Distinguished Young Scholars(grant no.22225302)Laboratory of AI for Electrochemistry(AI4EC),and IKKEM(grants nos.RD2023100101 and RD2022070501)supported by the National Natural Science Foundation of China(grants nos.12122401,12074007,and 12135002)Lastly,the computational resource was supported by the Bohrium Cloud Platform at DP Technology and Tan Kah Kee Supercomputing Center(IKKEM).
文摘The rapid advancements in artificial intelligence(AI)are catalyzing transformative changes in atomic modeling,simulation,and design.AI-driven potential energy models havedemonstrated the capability to conduct large-scale,long-duration simulations with the accuracy of ab initio electronic structure methods.However,the model generation process remains a bottleneck for large-scale applications.We propose a shift towards a model-centric ecosystem,wherein a large atomic model(LAM),pretrained across multiple disciplines,can be efficiently fine-tuned and distilled for various downstream tasks,thereby establishing a new framework for molecular modeling.In this study,we introduce the DPA-2 architecture as a prototype for LAMs.Pre-trained on a diverse array of chemical and materials systemsusing a multi-task approach,DPA-2demonstrates superior generalization capabilities across multiple downstream tasks compared to the traditional single-task pre-training and fine-tuning methodologies.Our approach sets the stage for the development and broad application of LAMs in molecular and materials simulation research.
