Collaborative Research: Elements: Building an open source DFT+eDMFT database for quantum materials

新量子材料的发现在技术进步中起着至关重要的作用。数据科学工具的使用，加上人工智能（AI），有可能大大加速材料的发现。然而，这些工具的有效性依赖于拥有大型、高质量的材料数据库。不幸的是，过去十年建立的许多现有数据库在准确预测量子材料的性质方面是不够的。量子材料在原子尺度上表现出高度相关的量子力学现象，在宏观尺度上表现出对社会有用的性质，如磁性和超导性。该项目旨在通过使用现代量子多体方法实现高吞吐量算法来开发一个开源的高保真材料数据库，这些算法尚未用于大规模数据库创建。此外，该项目使用人工智能工具增强了现有的大型材料数据库。为了确保可访问性，该项目将向科学界免费提供这些数据库的基础设施和教程。重要的材料数据库只包括由密度泛函理论（DFT）引擎创建的物理性质数据，DFT引擎根据平均电位存在下的独立电子来描述简单材料的物理性质。对于不能以这种平均方式处理的强相关量子材料，DFT通常无法预测正确的物理性质，而动态平均场论（DMFT）方法允许以更高但仍然实用的成本对相同性质进行更精确的计算。为了克服现有材料数据库基于DFT的缺点，并将数据科学驱动的材料发现转变为一个新时代，该团队旨在通过开发新的高通量DFT+DMFT工作流程，建立一个开源的高保真量子材料特性数据库，其中DFT引擎被基于DFT和DMFT结合的更精确的多体方法（DFT+DMFT）所取代。这种高保真但较小的数据库将用于使用人工智能工具（如迁移学习）纠正现有的大规模DFT数据库，一旦知道较小域的高保真数据，就可以自动修复不太准确的DFT数据中的系统错误。这项研究还旨在扩大学生在量子理论、人工智能职业以及连接材料和数据科学的跨学科领域的知识和技能。该项目由美国国家科学基金会高级网络基础设施办公室、刺激竞争性研究的既定计划（EPSCoR）和材料研究部联合资助。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The discovery of new quantum materials plays a crucial role in technological advancements. The use of data science tools, coupled with artificial intelligence (AI), has the potential to greatly accelerate materials discovery. However, the effectiveness of these tools relies on having large, high-quality material databases. Unfortunately, many existing databases built over the past decade have been inadequate in accurately predicting the properties of quantum materials. Quantum materials exhibit highly correlated quantum mechanical phenomena at the atomic scale, which manifest socially useful properties like magnetism and superconductivity at the macroscopic scale. This project aims to develop an open-source, high-fidelity materials database by implementing high-throughput algorithms using modern quantum many-body methods, which have not yet been employed for large-scale database creation. Additionally, the project enhances existing large-scale materials databases using AI tools. To ensure accessibility, this project will make the infrastructures and tutorials for these databases freely available to the scientific community. Important materials databases include exclusively physical property data created from Density Functional Theory (DFT) engines which describe the physical properties of simple materials in terms of independent electrons in the presence of an average potential. For strongly correlated quantum materials, which cannot be treated in this averaged manner, DFT often fails to predict correct physical properties, while the Dynamical Mean Field Theory (DMFT) approach allows far more accurate calculations of the same properties at a higher but still practical cost. To overcome the DFT-based shortcomings of existing materials databases and to transform data-science-driven materials discovery into a new era, the team aims to build an open-source high-fidelity database of quantum materials properties in which the DFT engine is replaced by the more precise many-body method based on a combination of DFT and DMFT (DFT+DMFT), through the development of a new high-throughput DFT+DMFT workflow. This high-fidelity but the smaller database will be used to correct the existing large-scale DFT databases using artificial intelligence tools, such as transfer learning, allowing the automatic repair of the systematic errors in the less accurate DFT data once the high-fidelity data in a smaller domain is known. This research also aims to expand the knowledge and skills of students in quantum theory, AI careers, and interdisciplinary fields that bridge materials and data science.This project is jointly funded by the NSF Office of Advanced Cyberinfrastructure, the Established Program to Stimulate Competitive Research (EPSCoR), and the Division of Materials Research.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.