Collaborative Research: DMREF: Data-Driven Prediction of Hybrid Organic-Inorganic Structures

杂化有机无机结构，特别是金属卤化物钙钛矿结构，由于太阳能电池、发光二极管性能的前所未有的进步，以及在晶体管、传感器、自旋电子学和催化剂方面的新应用，最近引起了人们的极大关注。有机-无机混合体系产生了极其广阔的化学和结构空间，这既为性质可调提供了令人兴奋的机会，也带来了与探索这一广阔空间以适合特定应用的结构的艰苦过程相关的巨大挑战。该项目旨在通过开发最近精选的X射线结构数据库、分子动力学模拟、机器学习(ML)、迭代反馈循环中的合成和结构研究来大力加速hOIS空间中的结构预测。这项研究将对组成-结构关系提供关键的见解，包括首选的结构维度，无机晶格中的扭曲，不同类钙钛矿型结构的相对稳定性，以及潜在的分子特征。其结果将是从原始材料快速预测有机-无机钙钛矿型混合结构，这对于优化广泛应用的光学、电子和自旋性质至关重要。将探索大约1000个新的hOI，使已知结构的范围增加一倍以上。空军研究实验室和国家可再生能源实验室与联邦合作伙伴的外部合作将测试新合成的结构和理论模型的应用。该团队包括来自三所大学的四名首席调查人员，其中包括为拉美裔服务的机构新墨西哥高地大学。该项目将培训本科生、研究生和博士水平的研究人员，包括代表性不足的少数民族和女性。私营部门还计划在国家会议上组织专题讨论会，以传播成果，并让更多的专家参与这项活动。技术描述：这项研究将利用多个hOIS数据库中约1000个已报道的晶体结构和带有界面力场的分子动力学模拟来提供描述符，并训练ML算法来预测晶体结构的相对稳定性和维度，结构特征，如相邻八面体之间的扭曲，以及晶格参数。然后，这些工具将被应用于预测~1000个未知的钙钛矿成分的结构，在合成和表征的迭代反馈循环中，预计相对于系列实验发现，至少会加速10倍。在合成、表征、建模和数据库开发的迭代过程中，将显著增加已知的hO I的数量，并阐明关键的分子间相互作用的作用，如多极电荷分布、原子半径、π堆积、异常氢键以及构筑块的手性对hO I多晶型的晶体结构和相对稳定性的影响。这项活动将解决材料科学中的一个重大挑战，这包括获得加权描述符，以精确控制高氧化合物的结构和与晶体生长的关系。这项努力将把材料科学、化学、计算和数据科学领域的专家和共同建议的学生聚集在一起，通过利用数据革命和融合的多学科研究来加速知识的创造。结构预测的描述符、ML算法和训练数据将公开共享，将多个结构数据库、网络基础设施工具和计算资源提升到一个新的水平。新的数据库条目、ML算法、迭代改进的力场参数和实验技术可以用于超出本项目范围的HO。该奖项反映了NSF的法定使命，并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，被认为值得支持。

Non-technical Description: Hybrid Organic Inorganic Structures (HOIS), specifically in the form of metal-halide perovskites, have recently attracted much attention due to unprecedented performance advancements in solar cells, light emitting diodes, as well as emerging applications in transistors, sensors, spintronics and catalysts. The extremely wide chemical and structural space engendered by hybrid organic-inorganic systems presents both exciting opportunities for property tunability, but also substantial challenges associated with the laborious process of exploring this wide space for suitable structures for a given application. This project aims to strongly accelerate structure prediction within the HOIS space through exploitation of recently curated X-ray structure databases, molecular dynamics simulation, machine learning (ML), synthetic and structural studies in an iterative feedback loop. The research will provide critical insights into composition-structure relationships, including the preferred structural dimensionality, distortions in the inorganic lattice, relative stabilities of different perovskite-like structures, and the underlying molecular features. The outcome will be the rapid prediction of hybrid organic-inorganic perovskite-type structures from the starting materials, which is essential to optimize optical, electronic and spin properties for a wide range of applications. Approximately one thousand new HOIS will be explored, more than doubling the range of known structures. External collaborations with federal partners at the Air Force Research Laboratory and at the National Renewable Energy Laboratory will test applications of newly synthesized structures and theoretical models. The team includes four Principal Investigators at three universities, including New Mexico Highlands University, a Hispanic-serving institution. The project will train undergraduate, graduate, and PhD-level researchers, including under-represented minorities and females. The PIs also plan to organize symposia at national meetings to disseminate the results and engage further experts in this activity. Technical Description: This research will utilize approximately 1000 reported crystal structures in multiple HOIS databases and molecular dynamics simulations with the INTERFACE force field to inform descriptors and train ML algorithms to predict the relative stability and dimensionality of crystal structures, structural features such as distortions between adjoining octahedra, and lattice parameters. The tools will then be applied to predict the structure of ~1000 yet unknown perovskite compositions in an iterative feedback loop with synthesis and characterization, expecting at least 10 times acceleration relative to serial experimental discovery. Iterations in synthesis, characterization, modeling, and database development will significantly increase the number of known HOIS and elucidate the role of critical intermolecular interactions such as multipolar charge distributions, atomic radii, π-stacking, unusual hydrogen bonds, and chirality of building blocks for the crystal structure and relative stability of HOIS polymorphs. The activity will address a grand challenge in materials science, which consists in obtaining weighted descriptors for precise structural control of HOIS and relationships to crystal growth. The effort will bring together experts and co-advised students across the fields of materials science, chemistry, computation, and data science for accelerated creation of knowledge by Harnessing the Data Revolution and convergent multidisciplinary research. The descriptors, ML algorithms, and training data for structure prediction will be openly shared, taking multiple structure databases, cyberinfrastructure tools, and computing resources to the next level. New database entries, ML algorithms, iteratively improved force field parameters, and experimental techniques can be used for HOIS beyond the scope of this project.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.