Data CI Pilot: CI-Based Collaborative Development of Data-Driven Interatomic Potentials for Predictive Molecular Simulations

该项目满足了分子模拟社区的迫切需求，为材料研究人员提供了一种强大的新能力，可以有效地合成与其特定研究问题相关的所有可用数据和知识。它召集了一个材料研究人员联盟，围绕提高由NSF和其他机构支持的现有和新兴材料网络基础设施的互操作性这一共同目标。PIS将开发一个新的计算框架，使研究人员能够通过连接现有的第一原理计算和实验数据的网络基础设施资源，快速开发和部署复杂材料系统的数据驱动的原子间相互作用势。这将提供对材料行为的定性见解，以及设计新材料和纳米结构所需的预测能力。该项目旨在通过消除材料研究人员面临的现有障碍，加快采用数据驱动的原子间相互作用势(DDIP)技术。学术界、政府和工业界很少有团体有能力开发DDIP，因为这需要同时具备被建模材料的物理、“第一原理”(FP)计算和机器学习技术方面的专业知识，以及获得构建适应高质量DDIP所需的极其庞大和多样化的训练集所需的广泛计算资源。为了解决这一问题，正在开发一种名为“ColabFit”的计算基础设施，它将使研究人员能够通过汇集他们的知识和数据来合作开发DDIP。研究人员将能够使用受支持的匹配选择代码来培训最先进的DDIP，无缝访问来自FP数据的现有网络基础设施(CI)资源的培训数据，并通过原子间模型开放知识库(OpenKIM)项目以标准格式交换DDIP，以便他们能够在彼此工作的基础上再接再厉。这一努力满足了材料模拟社区的迫切需求，由领先的DDIP开发人员、CI项目和材料标准组织组成的大型联盟就证明了这一点，这些组织已经聚集在一起支持它。为了使ColabFit框架的开发专注于真实世界的材料研究需求，它将围绕用于2D过渡金属二卤化物相变的DDIP开发的目标应用程序进行组织。该项目由计算机和信息科学局的高级数字基础设施办公室和数学和物理科学局的材料研究司联合支持。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。

The project addresses a pressing need of the molecular simulation community by providing materials researchers with a powerful new ability to efficiently synthesize all available data and knowledge related to their particular problem of study. It convenes a consortium of materials researchers around the common goal of increasing interoperability of existing and emerging materials cyberinfrastructures supported by NSF and others. The PIs will develop a new computational framework that enables researchers to rapidly develop and deploy data-driven interatomic potentials for complex material systems, by connecting existing cyberinfrastructure resources of first-principles calculations and experimental data. This will provide qualitative insights into material behavior, as well as predictive capability necessary to design new materials and nanostructures.This project aims to accelerate the adoption of data-driven interatomic potential (DDIP) technology by removing existing barriers faced by materials researchers. Few groups in academia, government and industry have the capacity to develop DDIPs as this requires simultaneous expertise in the physics of the material being modeled, in “first principles” (FP) calculations, and in machine learning techniques, as well as access to extensive computational resources needed to construct the extremely large and diverse training sets required to fit high-quality DDIPs. To address this, a computational infrastructure called “ColabFit” is being developed that will enable researchers to collaborate on DDIP development by pooling their knowledge and data. Researchers will be able to train state-of-the-art DDIPs using a supported fitting code of choice, seamlessly access training data from existing cyberinfrastructure (CI) resources of FP data, and exchange DDIPs in a standard format through the Open Knowledgebase of Interatomic Models (OpenKIM) project so that they can build on each other’s work. This effort fills a pressing need in the materials simulation community, as evidenced by the large consortium of leading DDIP developers, CI projects, and materials standards organization that has been assembled to support it. To keep development of the ColabFit framework focused on real-world materials research needs, it will be organized around a target application of DDIP development for phase transformations in 2D transition metal dichalcogenides.This project is jointly supported by the Office of Advanced Cyberinfrastructure in the Computer and Information Sciences Directorate, and the Division of Materials Research in the Mathematical and Physical Sciences Directorate.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.