Are diffusion models ready for materials discovery in unexplored chemical space?
Korea Institute of Science and Technology (KIST)（韩国科学技术院）；Seoul National University（首尔大学）；Korea Institute of Ceramic Engineering and Technology (KICET)（韩国陶瓷工程技术研究院）；Korea Institute of Advanced Study（韩国高等研究院）；University of Science and Technology (UST)（韩国科学技术院）；Kumho Petrochemical（锦湖石化）
Sanghyun Kim, Gihyeon Jeon, Seungwoo Hwang, Jiho Lee, Jisu Jung, Seungwu Han, Sungwoo Kang
While diffusion models are attracting increasing attention for the design of novel materials, their ability to generate low-energy structures in unexplored chemical spaces has not been systematically assessed. Here, we evaluate the performance of two diffusion models, MatterGen and DiffCSP, against three databases: a ternary oxide set (constructed by a genetic algorithm), a ternary nitride set (constructed by template informatics), and the GNoME database (constructed by a combination of both). We find that diffusion models generally perform stably in well-sampled chemical spaces (oxides and nitrides), but are less effective in uncommon ones (GNoME), which contains many compositions involving rare-earth elements and unconventional stoichiometry. Finally, we assess their size-extrapolation capability and observe a significant drop in performance when the number of atoms exceeds the trained range. This is attributed to the limitations imposed by periodic boundary conditions, which we refer to as the curse of periodicity. This study paves the way for future developments in materials design by highlighting both the strength and the limitations of diffusion models.
EC-MOF/Phase-I: A computationally ready database of electrically conductive metal-organic frameworks with high-throughput structural and electronic properties
An LLM agent for end-to-end computational materials discovery
Northwestern Polytechnical University（西北工业大学）；National University of Singapore（新加坡国立大学）；University of Toronto（多伦多大学）；Chinese Academy of Sciences（中国科学院）；Huazhong University of Science and Technology（华中科技大学）；The University of Tokyo（东京大学）
Catlas: an automated framework for catalyst discovery demonstrated for direct syngas conversion
在后化石燃料时代，催化剂的发现对于支持能源和关键化学原料的获取至关重要。使用从头算方法如密度泛函理论（DFT）对大的材料设计空间进行穷举计算搜索是不可行的。我们试图通过利用大型的、广义的、基于图的机器学习（ML）模型，以相对较低的计算成本探索大型设计空间，这些模型是预先训练的，因此不需要前期数据收集或训练。我们提出了catlas，一个框架，分布和自动生成的吸附物表面配置和ML推断的DFT能量，以实现这一目标。Catlas是开源的，使ML辅助催化剂筛选变得容易并可供所有人使用。为了证明其有效性，我们使用catlas来探索用于将合成气直接转化为多碳含氧化合物的催化剂候选物。对于这种情况下的研究，我们探索947稳定/亚稳二元，过渡金属间化合物作为可能的催化剂候选人。在这个子集的材料，我们能够预测的吸附能的关键描述符，*CO和 *OH，与近DFT精度（0.16，0.14 eV MAE，分别）。使用现有微动力学模型对C2+含氧化合物的预测选择性，我们确定了144种候选材料。对于10个有希望的候选人，DFT计算显示出与我们使用ML的评估有很好的相关性。最常见的元素组合是Pt-Ti、Pd-V、Ni-Nb和Ti-Zn，所有这些都是未经实验研究的。
Reaction-Network-Level Discovery of Ammonia Synthesis Catalysts via Ten-Million-Scale Generative Exploration
Shanghai Institute of Applied Physics, Chinese Academy of Sciences（中国科学院上海应用物理研究所）；Shanghai Advanced Research Institute, Chinese Academy of Sciences（中国科学院上海高等研究院）；Titan Holdings（泰坦控股）；University of Chinese Academy of Sciences（中国科学院大学）
Framework for an AI-Ready ARPES Community Database
Stanford University（斯坦福大学）；SLAC National Accelerator Laboratory（斯坦福直线加速器中心）；University of Geneva（日内瓦大学）；Leibniz Institute for Solid State and Materials Research, IFW Dresden（德累斯顿固体与材料研究所）；Lawrence Berkeley National Laboratory（劳伦斯伯克利国家实验室）；University of Washington（华盛顿大学）；University of Oxford（牛津大学）；University of British Columbia（不列颠哥伦比亚大学）；University of Colorado Boulder（科罗拉多大学博尔德分校）；Carnegie Mellon University（卡内基梅隆大学）；The University of Tokyo（东京大学）；National Synchrotron Radiation Research Center（国家同步辐射研究中心）；National Tsing Hua University（国立清华大学）；Rice University（莱斯大学）；Princeton University（普林斯顿大学）；Yale University（耶鲁大学）；Aarhus University（奥胡斯大学）；Massachusetts Institute of Technology（麻省理工学院）；University of Florida（佛罗里达大学）；University of St Andrews（圣安德鲁斯大学）；Tohoku University（东北大学）；Cornell University（康奈尔大学）；University of California Davis（加州大学戴维斯分校）；University of Chicago（芝加哥大学）；Fritz Haber Institute of the Max Planck Society（马克斯·普朗克学会弗里茨·哈伯研究所）
