Theory-Guided Discovery of Tin-Based Materials

该奖项支持计算和理论研究，通过快速搜索大空间的化学成分来推进理论指导下的新材料发现。该项目的中心目标是对锡合金进行系统筛选，并为一系列化学元素开发基于神经网络的原子间电位库。神经网络模型将用于加速在原子水平上对可能结构的搜索。由于锡合金具有展示新物理特性和下一代功能特征的潜力，因此对锡合金的研究引起了人们新的兴趣。精细调谐的锡基拓扑绝缘体可以在自旋电子学、量子计算和热电材料中找到未来的用途，这些材料可以从热量中产生电力。具有更高耐久性的高容量锡基电极可以使电池更便宜、更安全。精心优化的锡基焊料可以减少有毒含铅材料的使用。该项目包括教育活动，以吸引年轻学生和代表性不足的群体成员参与科学研究。该项目将为宾厄姆顿大学面向中学生的物理拓展项目贡献一个新主题，为本科生开发一套关于神经网络的实践演示，并招募宾厄姆顿大学进化研究项目不同专业的学生进行跨学科的本科生研究。该奖项支持计算和理论研究，通过快速搜索大空间的化学成分来推进理论指导下的新材料发现。项目将集中于锡基材料的系统研究。这项工作的动机是锡合金的结构和电子特性的显著丰富性，使其能够用作拓扑绝缘体、电池阳极、焊料等。锡合金的研究和开发面临的主要挑战在于其结构的复杂性和键合机制的多样性。这些因素限制了基于从头算的研究范围和经典的基于势的建模的准确性。该项目旨在证明材料预测的效率和可靠性可以通过以下方式大大提高：(i)使用一套不同的搜索策略筛选大型材料类别，以及（ii）使用新兴的神经网络方法来模拟原子间相互作用以加速搜索。为了对结构和组成的构型空间进行全面采样，研究团队将依靠高通量、进化和理性设计搜索的结合。新型锡基宽间隙拓扑绝缘体、耐用电池阳极和稳定无铅焊料的鉴定将在基础和应用驱动研究的几个领域推进知识。为了系统地构建可重用的神经网络模型，研究团队将使用最近开发的分层训练程序，使库能够自然扩展到更大的化学系统集。神经网络模型将作为该组织开源MAISE包的一部分免费提供。这一努力将促进新兴机器学习方法在材料研究中的发展和应用。科学工作将与教育和推广活动相结合，这将培养下一代对科学、技术、工程和数学学科的兴趣。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。

NON-TECHNICAL SUMMARYThis award supports computational and theoretical research to advance theory-guided discovery of new materials through rapid search of the large space of chemical compositions. The central aims of this project are to perform a systematic screening of tin alloys and develop a library of neural network-based interatomic potentials for an extended set of chemical elements. The neural network models will be used to accelerate the search over possible structures at the level of atoms. Research into tin alloys has attracted renewed interest due to their potential to display novel physics and next-generation functional features. Finely tuned tin-based topological insulators could find future use in spintronics, quantum computing, and thermoelectric materials which can generate electricity from heat. High-capacity tin-based electrodes with improved durability could make batteries cheaper and safer. Carefully optimized tin-based solders may reduce the use of toxic lead-containing materials. This project includes educational activities to attract young students and members of underrepresented groups to scientific research. The PI will contribute a new theme to Binghamton University's Physics Outreach Program for middle school students, develop a set of hands-on presentations on neural networks for undergraduate students, and recruit students from different majors enrolled in Binghamton University's Evolutionary Studies program to carry out interdisciplinary undergraduate research.TECHNICAL SUMMARYThis award supports computational and theoretical research to advance theory-guided discovery of new materials through rapid search of the large space of chemical compositions. The PI will focus on the systematic study of tin-based materials. The work is motivated by the remarkable richness of the tin alloys' structural and electronic properties enabling their use as topological insulators, battery anodes, solders, and more. The main challenges associated with the study and development of tin alloys lie in the complexity of their structures and the diversity of their bonding mechanisms. These factors limit the scope of ab initio-based study and the accuracy of classical potential-based modeling. The PI aims to demonstrate that the efficiency and reliability of materials prediction can be improved considerably by: (i) screening a large materials class with a suite of diverse search strategies, and (ii) using emerging neural network methodology for modeling interatomic interactions to accelerate the search. For the comprehensive sampling of the configuration space of structures and compositions, the research team will rely on a combination of high-throughput, evolutionary, and rational design searches. Identification of new tin-based wide-gap topological insulators, durable battery anodes, and stable lead-free solders will advance knowledge in several areas of basic and application-driven research. For the systematic construction of reusable neural network models, the research team will use a recently developed stratified training procedure enabling a natural extension of libraries to larger sets of chemical systems. The neural network models will be freely available as a part of the group's open-source MAISE package. This effort will promote the development and application of emerging machine learning methods in materials research. The scientific work will be integrated with educational and outreach activities which will foster the interest of the next generation in science, technology, engineering, and mathematics disciplines.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.