Interatomic potentials (IAPs), which describe the potential energy surface of atoms, are a fundamental input for atomistic simulations. However, existing IAPs are either fitted to narrow chemistries or too inaccurate for general applications. Here, we report a universal IAP for materials based on graph neural networks with three-body interactions (M3GNet). The M3GNet IAP was trained on the massive database of structural relaxations performed by the Materials Project over the past 10 years and has broad applications in structural relaxation, dynamic simulations and property prediction of materials across diverse chemical spaces. About 1.8 million materials were identified from a screening of 31 million hypothetical crystal structures to be potentially stable against existing Materials Project crystals based on M3GNet energies. Of the top 2000 materials with the lowest energies above hull, 1578 were verified to be stable using DFT calculations. These results demonstrate a machine learning-accelerated pathway to the discovery of synthesizable materials with exceptional properties.
Expanding Universal Machine Learning Interatomic Potentials to 97 Elements Towards Nuclear Applications
Center for Quantum Information and Quantum Biology, The University of Osaka（大阪大学量子信息与量子生物学中心）；Graduate School of Science, The University of Osaka（大阪大学理学研究科）
Equivariant Machine Learning Interatomic Potentials with Global Charge Redistribution
Department of Mechanical Engineering, Texas Tech University, Lubbock, Texas 79409, USA（德克萨斯理工大学机械工程系）；Samsung Advanced Institute of Technology, Samsung Electronics, Suwon 16678, Republic of Korea（三星电子首尔先进技术研究所）
National Institute of Information and Communications Technology（国立情报学研究所）
We verified that the deep learning method named reading periodic table introduced by ref. Deep Learning Model for Finding New Superconductors, which utilizes deep learning to read the periodic table and the laws of the elements, is applicable not only for superconductors, for which the method was originally applied but also for other problems of materials by demonstrating band gap estimations. We then extended the method to learn the laws better by directly learning the cylindrical periodicity between the right- and left-most columns in the periodic table at the learning representation level, …
We obtain parameters for non-orthogonal and orthogonal TB models from two-atomic molecules for all combinations of elements of period 1 to 6 and group 3 to 18 of the periodic table. The TB bond parameters for 1711 homoatomic and heteroatomic dimers show clear chemical trends. In particular, using our parameters we compare to the rectangular d-band model, the reduced sp TB model as well as canonical TB models for sp- and d-valent systems which have long been used to gain qualitative insight into the interatomic bond. The transferability of our dimer-based TB bond parameters to bulk systems is d…
We propose and demonstrate that thermal laser evaporation can be applied to all solid, non-radioactive elements in the periodic table. By depositing thin films, we achieve growth rates exceeding 1 angstrom/s with output laser powers less than 500 W, using identical beam parameters for many different elements. The source temperature is found to vary linearly with laser power within the examined power range. High growth rates are possible using free-standing sources for most of the elements tested, eliminating the need for crucibles.
Classical force fields (FF) based on machine learning (ML) methods show great potential for large scale simulations of materials. MLFFs have hitherto largely been designed and fitted for specific systems and are not usually transferable to chemistries beyond the specific training set. We develop a unified atomisitic line graph neural network-based FF (ALIGNN-FF) that can model both structurally and chemically diverse materials with any combination of 89 elements from the periodic table. To train the ALIGNN-FF model, we use the JARVIS-DFT dataset which contains around 75000 materials and 4 mill…
The University of Tokyo（东京大学）；Massachusetts Institute of Technology（麻省理工学院）
A universal interatomic potential for an arbitrary set of chemical elements is urgently needed in computational materials science. Graph convolution neural network (GCN) has rich expressive power, but previously was mainly employed to transport scalars and vectors, not rank $\ge 2$ tensors. As classic interatomic potentials were inspired by tight-binding electronic relaxation framework, we want to represent this iterative propagation of rank $\ge 2$ tensor information by GCN. Here we propose an architecture called the tensor embedded atom network (TeaNet) where angular interaction is translate…
Single-Atom Adsorption on h-BN along the Periodic Table of Elements: From Pristine Surface to Vacancy-Engineered Sites
University of Belgrade − Faculty of Physical Chemistry（贝尔格莱德大学物理化学学院）；Luleå University of Technology（吕勒奥理工大学）；Serbian Academy of Sciences and Arts（塞尔维亚科学与艺术院）
We study the dynamical stability of elemental two-dimensional (2D) metals from Li to Pb by calculating the phonon band structure from first principles, where 2D structures are assumed to be planer hexagonal, buckled honeycomb, and buckled square lattice structures. We show the relationship between the stability of 2D structures and that of three-dimensional structures. This provides a material design concept for alloys, where the similarity with regard to the stable 2D structures, rather than the energetic stability of alloy, is important to yield dynamically stable alloys.
The understanding of atomic adsorption on graphene is of high importance for many advanced technologies. Here we present a complete database of the atomic adsorption energies for the elements of the Periodic Table up to the atomic number 86 (excluding lanthanides) on pristine graphene. The energies have been calculated using the projector augmented wave (PAW) method with PBE, long-range dispersion interaction corrected PBE (PBE+D2, PBE+D3) as well as non-local vdW-DF2 approach. The inclusion of dispersion interactions leads to an exothermic adsorption for all the investigated elements. Dispers…
A well-known drawback of state-of-the-art machine-learning interatomic potentials is their poor ability to extrapolate beyond the training domain. For small-scale problems with tens to hundreds of atoms this can be solved by using active learning which is able to select atomic configurations on which a potential attempts extrapolation and add them to the ab initio-computed training set. In this sense an active learning algorithm can be viewed as an on-the-fly interpolation of an ab initio model. For large-scale problems, possibly involving tens of thousands of atoms, this is not feasible becau…
Geometric Algebra Meets Cartesian Tensors: Higher-Order Equivariance for Interatomic Potentials
Dept. of Electrical & Computer Eng.（电气与计算机工程系）；Texas A&M Univ.（德克萨斯A&M大学）；Dept. of Prosthetics & Orthotics（假肢与矫形器系）；College of Science & Eng.（科学与工程学院）；Ankara Univ.（安卡拉大学）；Hamad Bin Khalifa Univ. and Texas A&M Univ. at Qatar（哈马德·本·卡西姆大学和德克萨斯A&M大学（多哈））
Vacancies in graphene present sites of altered chemical reactivity and open possibilities to tune graphene properties by defect engineering. The understanding of chemical reactivity of such defects is essential for successful implementation of carbon materials in advanced technologies. We report the results of a systematic DFT study of atomic adsorption on graphene with a single vacancy for the elements of rows 1 to 6 of the Periodic Table of Elements (PTE), excluding lanthanides. The calculations have been performed using PBE, long-range dispersion interaction-corrected PBE (PBE+D2 and PBE+D3…
Two-dimensional transition metal dichalcogenides (TMDs) exist in two polymorphs, referred to as $1T$ and $1H$, depending on the coordination sphere of the transition metal atom. The broken octahedral and trigonal prismatic symmetries lead to different crystal and ligand field splittings of the $d$ electron states, resulting in distinct electronic properties. In this work, we quantify the crystal and ligand field parameters of two-dimensional TMDs using a Wannier-function approach. We adopt the methodology proposed by Scaramucci et al. [A. Scaramucci et al., J. Phys.: Condens. Matter 27, 175503…
In this comment we report on two misprints of a classical and still widely used tight-binding table contained in a seminal, 65-years-old paper by Slater and Koster, and suggest the corresponding corrections.
Projector augmented-wave and all-electron calculations across the periodic table: a comparison of structural and energetic properties
我们构建了一个参考数据库的材料性能计算使用密度泛函理论在本地或广义梯度近似，和一个全电子或投影仪增强波（PAW）制定，验证和验证的第一性原理模拟。全电子计算使用全电势线性增强平面波方法，如在\texttt{Elk}开源代码中实现的，而PAW计算使用我们中的一些人在开源\texttt{PSlibrary}存储库和\texttt{Quantum ESPRESSO}分布中开发的数据集。我们首先计算晶格参数，体积模量，和能量差的碱金属，碱土金属，和$3d$和$4d$过渡金属在三个理想的，参考相（简单立方，面心立方，体心立方），代表一个标准化的晶体单原子固态测试。然后，如K. Lejaeghere { \it等}，[Critical Reviews in Solid State and Material Sciences 39，p 1（2014）]，我们比较了所有元素的状态方程，除了镧系元素和锕系元素，在他们的实验阶段（或者偶尔更简单，密切相关的一个）。PAW和全电子能量差和结构参数同意在大多数情况下，在几个毫电子伏/原子和百分之一的分数，分别。这种水平的协议，与以前的研究相比，还包括其他PAW和全电子数据的电子结构代码\texttt{VASP}和\texttt{WIEN2K}，并强调了目前的整体可靠性，国家的最先进的电子结构计算。与此同时，即使在简单的基本结构特性的同一公式中也会出现差异，…
