CDS&E: Systematic Exploration of the High Entropy Alloy Space through High-Dimensional Thermodynamic Modeling from High-Throughput Computations and Experimental Data

非技术概述材料科学领域被一类被称为高熵合金的合金的发现所吸引，这种合金的特点是通过大量不同元素的组合以相当的量意外地稳定简单的晶体结构。这类合金的发现和设计需要详细了解控制高维组成空间稳定性的热力学因素，该空间可能有许多相互竞争的晶体结构。这种复杂性测试了当前热力学建模能力的极限，这是由于所需的输入数据和进入模型的参数的绝对数量。该项目通过使用正式统计和主动机器学习技术，将(I)实验得出的热力学模型的大型元数据库与(Ii)来自高通量量子力学计算的热力学数据相结合来解决这一问题。这项工作的最终成果是一个开放分布的大范围包罗万象的热力学模型，该模型可以在高维组成空间中进行查询，并返回作为交互三维横截面的结构稳定性信息，或者作为依赖于组成和温度的热力学性质。由于与已知高熵合金的数量相比，这一可能的合金空间是如此之大，通过该工具发现新合金的潜力是巨大的，这可能会广泛地影响许多能力受到材料性质限制的工程应用。该项目利用并集成了PI小组的两个最新发展：(I)搜索引擎(热力学数据库或TDBDB)，其索引所有可用的、以标准格式在科学文献中电子获得的实验得出的热力学数据；(Ii)一套软件工具(合金理论自动化工具包或ATAT)，其简化了从从头算数据生成热力学数据库的过程。这种混合方法旨在结合最先进的实验和计算方法的独特优势，即前者的较高精度和后者的高通量特性。主动机器学习和统计技术被用于(I)针对可能形成具有简单晶体结构的固体溶液的有前景的组成区域的探索，以及(Ii)开发需要很少从头计算输入的非化学计量固体的高效统计力学模型，从而实现高通量操作。虽然现有的计算高通量工作主要集中在绝对零度下无缺陷的化学计量化合物的性质上，但该项目的目标是在所有温度下更广泛的材料范围，包括可能具有短程有序的无序合金和可能具有点缺陷的有序合金。这就需要(I)短程有序，(Ii)强非谐相和(Iii)磁有序的有效统计力学模型，通过利用已知的和数据挖掘的趋势，每个模型都需要比现有的蛮力方法更少的从头算输入。在这个项目中设计的数据和工具预计将产生更广泛的影响：几乎所有的工程材料都是非化学计量合金，其性能可以通过控制添加大量组分来调整(高熵合金仅代表一个极端例子)。因此，这种资源可以通过增加现有的计算高通量努力，极大地促进材料发现和优化，目前这些努力只针对绝对零度的有序化合物。该奖项反映了NSF的法定使命，通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为是值得支持的。

Nontechnical summaryThe field of materials science has been captivated by the discovery of a class of alloys known as "high entropy" alloys, which are characterized by the unexpected stabilization of simple crystal structures through the combination of a large number of different elements in comparable amounts. The discovery and design of such alloys demand a detailed knowledge of the thermodynamic factors governing stability in a very high dimensional composition space with potentially many competing crystal structures. This complexity tests the limits of current thermodynamic modeling capabilities due to the sheer number of input data needed and of parameters entering the model. The project addresses this by combining (i) large-scale meta-databases of experimentally-derived thermodynamic models with (ii) thermodynamic data from high-throughput quantum-mechanical calculations, using formal statistical and active machine learning techniques. The end product of this effort is an openly distributed large-scale encompassing thermodynamic model that can be queried in a high-dimensional compositional space and that returns structural stability information as interactive tridimensional cross-sections, or as composition- and temperature-dependent thermodynamic properties. As this space of possible alloys is so vast compared to the number of known high-entropy alloys, the potential for discoveries of novel alloys through this tool is significant and this could broadly impact numerous engineering applications where capabilities are limited by materials properties.Technical summaryThe project leverages and integrates two recent developments from the PI's group: (i) a search engine (the Thermodynamic DataBase DataBase or TDBDB), that indexes all available experimentally-derived thermodynamic data electronically available in standardized format in the scientific literature; (ii) a suite of software tools (the Alloy Theoretic Automated Toolkit or ATAT) that streamlines the generation of thermodynamic databases from ab initio data. This hybrid approach aims to combine the distinct advantage of state-of-the-art experimental and computational methods, namely, the higher accuracy of the former and the high-throughput nature of the latter.Active machine learning and statistical techniques are used to (i) target the exploration of promising composition regions likely to form solid solutions with simple crystal structures and (ii) develop efficient statistical mechanics models of non-stoichiometric solids that require few ab initio inputs, thus enabling "high-throughput" operation. Whereas existing computational high-throughput efforts primarily focus on the properties of defect-free stoichiometric compounds at absolute zero, this project targets, at all temperatures, the broader range of materials including disordered alloys with possible short-range order and ordered alloys with possible point defects. This demands efficient statistical mechanical models of (i) short-range order, (ii) strongly anharmonic phases and (iii) magnetic ordering, each of which requiring fewer ab initio input than existing brute force methods, by exploiting both known and data-mined trends.The data and tools devised during this project are expected to have broader impacts: virtually all engineering materials are nonstoichiometric alloys whose properties are tuned by controlled additions of numerous components (with high entropy alloys only representing an extreme example). Hence, this resource could greatly facilitate materials discovery and optimization by augmenting existing computational high-throughput efforts that currently only target ordered compounds at absolute zero. The broad compositional gamut covered also enables applications that could range from identifying glass-forming metallic alloys to determining possible exoplanet core compositions and structures.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.