DeepModeling

1. Introduction

Recently, a collaborative team from Peking University, University of Science and Technology of China, Institute of Physics, and other institutions has implemented a unified heterogeneous computing framework for real-time time-dependent density functional theory (RT-TDDFT) based on numerical atomic orbitals (NAO) within the domestically developed open-source density functional theory software ABACUS. By introducing a hardware-agnostic abstraction layer into the ABACUS code, the team successfully achieved efficient cross-platform acceleration from single-node multi-core CPUs to large-scale multi-GPU architectures while keeping the physics algorithm code clean and maintainable.

Recently, the team of Likun Pan and Chenglong Wang from East China Normal University, together with the team of Jinliang Li from Jinan University, and researchers from Shanghai Maritime University and the National University of Singapore, published a research paper titled "A Quantitative Electrostatic Potential Descriptor Enables Deep Learning-Accelerated Discovery of High-Performance Lithium-Ion Battery Electrolytes" in Angewandte Chemie International Edition. The study proposes a quantitative electrostatic potential ratio descriptor, ESPratio, and combines it with a fine-tuned Uni-Mol self-supervised pre-trained model to conduct a three-tier screening of approximately 1 million PubChem molecules, yielding 353 candidate electrolyte molecules. Experimental validation further confirmed the application potential of molecules such as TBDN and PIV in high-voltage lithium-ion batteries, providing a new strategy for the rational design of high-performance electrolytes.

Accuracy Rivals Mainstream Models, Speed Surpasses Empirical Potentials: DPA4 OMat24 and OMol25 Pre-trained Models Unveiled

Whether a new material can withstand high temperatures, whether a battery will swell, or where an alloy part will crack—to answer these questions, the most thorough approach is to lay out the atoms in the material one by one, let them move according to physical laws, and play an "atomic-level movie" in a computer. This is cheaper and faster than actually smelting a batch, burning it, or breaking it.

Yet for decades, this movie has been stuck in the same place: you have to choose between fidelity and scale.

Molten salt reactors (MSRs), as important candidates for Generation IV advanced nuclear energy systems, feature fluoride molten salts (such as the LiF-NaF-KF ternary eutectic system FLiNaK) as core functional materials. Their high-temperature stability, excellent thermal conductivity, and ability to dissolve actinides make them ideal carriers for liquid nuclear fuel and coolants. However, the microstructural coordination environment and thermophysical property evolution of the key actinide ion Th⁴⁺ in thorium-based MSRs under different alkali metal cation ratios have lacked systematic study. How to efficiently and accurately screen fuel salt formulations that balance thermal storage, transport properties, and structural stability across a wide compositional space has become a critical bottleneck limiting MSR optimization.

In silicon-on-insulator (SOI) technology, the core challenge of the separation by implanted oxygen (SIMOX) process lies in precisely controlling the thickness and oxygen distribution of the buried oxide (BOX) layer and understanding cross-interface thermal transport behavior. Traditional methods often suffer from limited accuracy or efficiency in predicting oxygen diffusion dynamics, layer thickness evolution, and interfacial thermal resistance during post-implantation annealing. Recently, a team led by Prof. Guangping Zheng and Dr. Jiashu Chen from the Department of Mechanical Engineering at The Hong Kong Polytechnic University, in collaboration with Prof. Zhuo Tang’s group from the College of Information Science and Engineering, Hunan University, published a research achievement in Communications Materials (IF=9.6) titled “Deep learning modeling of oxygen redistribution and thermal transport in silicon on insulator and buried oxide layers”. They proposed a computational framework integrating deep neural network potential with molecular dynamics (DFT-MD + DP), systematically addressing the multiscale modeling challenges from oxygen implantation and diffusion to interfacial heat transport.

Overview of APEX 1.3.0 Core Upgrades

As a key component of the DeepModeling open-source ecosystem, APEX (Alloy Property Explorer) [1] has undergone continuous iterative optimization since the release of V1.2.0, focusing on high-throughput materials property calculation scenarios. These efforts have steadily enhanced workflow automation, computational efficiency, and user experience, driving the AI for Materials (AI4M) infrastructure toward greater intelligence and standardization.

The newly released APEX V1.3.0 introduces comprehensive upgrades across multiple fronts, including automated structure construction, novel property calculation workflows, a graphical user interface (GUI), and task fault-tolerance with diagnostic mechanisms. This release further reduces manual intervention in complex computational pipelines while improving execution stability and traceability.

More importantly, this upgrade marks a critical step for APEX in evolving from a “computational tool” toward a “scientific agent infrastructure”: automated structure generation, standardized task encapsulation, and workflow composability lay the foundation for building materials computation services that can be directly invoked by AI agents. Whether it is composition-aware structure generation, sublattice-aware random solid-solution construction, or batch task management and multi-workflow tracking within the graphical interface, all these features reflect APEX’s significant progression toward an agent-ready research infrastructure.

Phase diagrams and thermodynamic properties of deep Earth materials are fundamental to geophysics, geodynamics and geological research. As the core chemical system of mantle mineralogy, the Mg-Al-Si-O system governs the stability and physical properties of dominant mantle minerals and melts, further modulating mantle dynamics and plate tectonics. Recently, Xin Zhong, Timm John from Free University of Berlin and Yifan Li from Princeton University published their research A general purposed machine learning interatomic potential for Mg-Al-Si-O system suitable for Earth materials at high pressure and temperature conditions in npj Computational Materials. The team developed a universal machine learning interatomic potential for the Mg-Al-Si-O system. By combining the r2SCAN functional with pairwise Gaussian energy correction, the average enthalpy error of over 20 mineral phases was reduced from 5.2 kJ/mol to 1.2 kJ/mol. The potential reproduces phase diagrams of systems including SiO₂ Al₂SiO₅ and Mg₂SiO₄ with excellent consistency against experimental measurements. The study quantitatively characterizes the anisotropy of solid–melt interfacial free energy for periclase and forsterite at the atomic scale, and quantifies how non-hydrostatic stress modulates the α-β quartz phase transition.

Next-generation nuclear fission and fusion reactors impose extremely stringent requirements on structural materials, which must simultaneously withstand high temperatures, high-dose irradiation, and strongly corrosive coolants. Multi-principal element alloys (MPEAs) are regarded as highly promising candidate materials owing to their unique high-entropy effect, lattice distortion, sluggish diffusion, and cocktail effect. Nevertheless, understanding irradiation damage and mechanical behaviors at the atomic scale in these complex alloys demands interatomic potentials (IAPs) with high precision and universal transferability. Although conventional machine learning interatomic potentials (MLIAPs) achieve decent accuracy, the volume of training datasets rises exponentially for quinary or higher-order complex systems, leading to prohibitive computational costs for generating DFT reference labels.

Recently, the OpenLAM Team of the AI fof Science Institute,Beijing(AISI), Peking University, DP Technology, and the Institute of Applied Physics and Computational Mathematics have jointly launched DPA4, a new-generation model architecture tailored for the era of Large Atomic Models (LAMs). DPA4 claimed the top spot worldwide with its comprehensive performance score (CPS) on Matbench Discovery, an authoritative global benchmark for materials discovery, emerging as the latest State-of-the-Art (SOTA) model.

DPA4’s highlight lies in its ultra-low training threshold: the prior leading eSEN needed over 300 GPU days for training, yet DPA4 reaches matching accuracy with merely one consumer RTX 5090 running for roughly one day, and its parameter volume is less than one-tenth of eSEN’s.

In short, the SOTA-level accuracy once reliant on costly supercomputing is now accessible via a single consumer graphics card. DPA4 reshapes the accuracy-efficiency Pareto frontier of large atomic models.

Official Screenshot of Matbench Discovery (Data as of May 22, 2026)

A research team led by Dr. Xue Hongtao and Dr. Chang Zhen from State Key Laboratory of Advanced Processing and Recycling of Non-ferrous Metals & Gansu Center for Materials Genome and Fundamental Structural Research, Lanzhou University of Technology, has published new findings in Journal of Materials Research and Technology. The paper titled Deep potential driven molecular dynamics simulations on tensile properties of nano-grained Al–Ce alloys: Mechanisms of Ce segregation and strengthening develops a high-precision deep potential (DP) for Al-Ce alloys and clarifies the microscale strengthening mechanism induced by grain boundary segregation of trace rare earth cerium (Ce) in nanocrystalline aluminum alloys.