Collaborative Research: Data Driven Discovery of Singlet Fission Materials

材料研究部和数学科学部为本奖项提供资金。该项目利用基于量子力学的材料模拟、数据科学和人工智能的进展来发现新的太阳能电池材料。以高能光子形式被太阳能电池材料吸收的多余光能的一小部分可以转化为热而不是电流。这就造成了能量转换效率的损失。许多晶体是由原子组成的，这些原子以规则的空间周期模式自我排列。晶体可以在有机分子代替原子形成基本单位的地方形成。这些分子晶体可能会经历一个被称为单线态裂变的过程，这个过程可以将多余的光能转换成携带电流的电荷，而不是热量。基于单线态裂变的太阳能电池还不是一项商业技术，因为缺乏合适的材料来展示这一过程。已知的分子晶体结构超过一百万种，但不知道哪一种会发生单线态裂变。要测试许多已知或预测的新分子晶体，实验甚至先进的量子力学模拟都过于昂贵和耗时。为了克服这一障碍，计算材料科学和统计学领域的跨学科专家团队将利用数据科学和人工智能的方法。该团队将研究使用机器学习方法，通过快速识别有潜力的候选分子晶体来加速材料发现的过程。该团队将从基于量子力学的模拟中生成数据，以构建和迭代改进机器学习模型。这项研究有望导致发现单线态裂变的新材料，这将推动太阳能电池技术的发展，并可能降低太阳能电池的成本。这项研究将导致方法论的发展，从而使材料的发现能够用于其他应用。该项目还通过组织研讨会和会议，支持在百亿亿级上进行高级高性能计算培训，以及在材料科学和数据科学交叉领域建立社区。外展活动将吸引教育工作者和K-12学生提高对材料工程和数据科学的职业和研究机会的认识。技术摘要材料研究部和数学科学部为该奖项提供资金。单线态裂变是一个光产生的单线态激子转化为两个三重态激子。人们对单线态裂变非常感兴趣，因为它有可能通过从一个光子中捕获两个载流子来显着提高有机太阳能电池的效率。然而，目前很少有材料能够高效地表现出分子间单线态裂变，这阻碍了固态单线态裂变太阳能电池的实现。可能的发色团的化学合成空间是无限广阔的，很大程度上尚未被探索。为了能够通过计算发现单线态裂变的材料，跨学科团队将开发一种多保真度筛选方法，将基于量子力学的不同保真度模拟与机器学习和数据库挖掘相结合。机器学习算法将用于从模拟生成的数据中学习，并引导模拟进行进一步的数据采集。随着获取的数据越来越多，机器学习模型将动态调整。这将在一个完全自动化的迭代工作流程中实现，设计用于运行在百亿亿级高性能计算机上。这项研究将推动发现新的分子间单线态裂变发色团，这将催化“第三代”固态单线态裂变有机太阳能电池的实现。机器学习揭示的结构-性质相关性将通过提高单线态裂变效率，为发色团和晶体形式提供化学见解和设计规则，从而推进对单线态裂变的基本理解。除了单线裂变发色团外，还可能发现具有其他有机电子器件应用所需性能的材料。此外，该项目将带来新的方法，以解决寻找单线态裂变材料所带来的数据科学挑战。该项目将导致涉及许多数据源的高维问题的最佳抽样策略的统计发展，以及随着获得更多数据而动态发展的自适应模型。所开发的方法可以扩展到其他问题，其中感兴趣的属性来自复杂现象，数据采集成本高或耗时长，预测描述符未知。该项目还通过组织研讨会和会议，支持在百亿亿级上进行高级高性能计算培训，以及在材料科学和数据科学交叉领域建立社区。外展活动将吸引教育工作者和K-12学生提高对材料工程和数据科学的职业和研究机会的认识。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。

NONTECHNICAL SUMMARYThe Division of Materials Research and the Division of Mathematical Sciences contribute funds to this award. This project leverages advances in quantum-mechanics-based materials simulations, data science, and artificial intelligence to discover new solar cell materials. A fraction of excess energy of light in the form of high-energy photons absorbed by solar cell materials can be converted to heat rather than to electric current. This results in a loss of energy conversion efficiency. Many crystals are made of atoms that self-arrange in a regular spatially periodic pattern. Crystals can be made where organic molecules can form the fundamental unit instead of atoms. These molecular crystals may undergo a process known as singlet fission, which enables the conversion of that excess light energy into current carrying charges rather than heat. Singlet-fission-based solar cells are not yet a commercial technology due to dearth of suitable materials that exhibit this process. Over a million molecular crystal structures are known, but it is unknown for which ones singlet fission can occur. Experiments and even advanced quantum mechanical simulations are too costly and time consuming to test many known or predicted new molecular crystals. To overcome this barrier, the interdisciplinary team of experts in computational materials science and statistics will harness methods of data science and artificial intelligence. The team will investigate the use of machine learning methods to accelerate the process of materials discovery by rapidly identifying potentially promising candidate molecular crystals. The team will generate data from quantum-mechanics-based simulations to construct and iteratively improve machine-learned models. This research is expected to lead to the discovery of new materials that exhibit singlet fission, which would advance solar cell technology and could reduce the cost of solar cells. The research will lead to methodological developments that would enable materials discovery for other applications. This project also supports training in advanced high-performance computing at the exascale, and community building at the intersection of materials science and data science through organizing workshops and conferences. An outreach activity will engage educators and K-12 students to raise awareness of careers and research opportunities in materials engineering and data science. TECHNICAL SUMMARYThe Division of Materials Research and the Division of Mathematical Sciences contribute funds to this award. singlet fission is the conversion of one photogenerated singlet exciton into two triplet excitons. There has been much interest in singlet fission because of its potential to significantly increase the efficiency of organic solar cells by harvesting two charge carriers from one photon. However, few materials are presently known to exhibit intermolecular singlet fission with high efficiency, hindering the realization of solid-state singlet-fission-based solar cells. The chemical compound space of possible chromophores is infinitely vast and largely unexplored. To enable computational discovery of materials that exhibit singlet fission, the interdisciplinary team will develop a multi-fidelity screening approach, integrating quantum-mechanics-based simulations at different levels of fidelity with machine learning and database mining. Machine learning algorithms will be used to learn from data generated by simulations and to steer simulations for further data acquisition. The machine learning models will dynamically adapt as more data is acquired. This will be implemented in a fully automated iterative workflow, designed to run on exascale high performance computers. This research will advance the discovery of new intermolecular singlet fission chromophores, which will catalyze the realization of "third generation" solid-state singlet-fission-based organic solar cells. Structure-property correlations revealed by machine learning will advance the fundamental understanding of singlet fission by deriving chemical insights and design rules for chromophores and crystal forms with enhanced singlet-fission efficiency. In addition to singlet-fission chromophores, materials may be discovered with desirable properties for other organic electronic device applications. In addition, this project will lead to new methodology needed to tackle the data-science challenges posed by searching for singlet fission materials. This project will lead to statistical developments in optimal sampling strategies for high-dimensional problems involving many data sources and in adaptive models that dynamically evolve as more data is acquired. The developed approach may be extended to other problems, where properties of interest arise from complex phenomena, for which data acquisition is costly or time consuming, and predictive descriptors are unknown. This project also supports training in advanced high-performance computing at the exascale, and community building at the intersection of materials science and data science through organizing workshops and conferences. An outreach activity will engage educators and K-12 students to raise awareness of careers and research opportunities in materials engineering and data science.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.