Zadoks Austin, Marrazzo Antimo, Marzari Nicola
Theory and Simulation of Materials (THEOS), École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland.
Dipartimento di Fisica, Università di Trieste, I-34151 Trieste, Italy.
NPJ Comput Mater. 2024;10(1):278. doi: 10.1038/s41524-024-01446-9. Epub 2024 Dec 2.
Machine learning in atomistic materials science has grown to become a powerful tool, with most approaches focusing on atomic geometry, typically decomposed into local atomic environments. This approach, while well-suited for machine-learned interatomic potentials, is conceptually at odds with learning complex intrinsic properties of materials, often driven by spectral properties commonly represented in reciprocal space (e.g., band gaps or mobilities) which cannot be readily partitioned in real space. For such applications, methods that represent the electronic rather than the atomic structure could be more promising. In this work, we present a general framework focused on electronic-structure descriptors that take advantage of the natural symmetries and inherent interpretability of physical models. We apply this framework first to material similarity and then to accelerated screening, where a model trained on 217 materials correctly labels 75% of entries in the Materials Cloud 3D database, which meet common screening criteria for promising transparent-conducting materials.
机器学习在原子材料科学中已发展成为一种强大的工具，大多数方法聚焦于原子几何结构，通常将其分解为局部原子环境。这种方法虽然非常适合机器学习的原子间势，但在概念上与学习材料复杂的固有属性不一致，这些属性通常由倒易空间中常见的光谱特性驱动（例如带隙或迁移率），而这些特性在实空间中不易划分。对于此类应用，表征电子结构而非原子结构的方法可能更具前景。在这项工作中，我们提出了一个通用框架，该框架聚焦于利用物理模型的自然对称性和内在可解释性的电子结构描述符。我们首先将此框架应用于材料相似性，然后应用于加速筛选，在加速筛选中，一个在217种材料上训练的模型正确标记了材料云3D数据库中75%符合有前景的透明导电材料常见筛选标准的条目。
