Frameworks: Data-Driven Software Infrastructure for Next-Generation Molecular Simulations

This project focuses on the development and optimization of the MB-Fit/MBX software infrastructure, a tool designed to advance the field of molecular simulations. By providing a machine-learned representation of molecular interactions, the software provides researchers worldwide with the possibility to model and predict the behavior of complex systems at the molecular level with unprecedented accuracy. As an open-source tool, MB-Fit/MBX not only promotes the progress of science by encouraging contributions from researchers worldwide but also democratizes access to cutting-edge computational tools. The project is committed to fostering education and diversity through the organization of workshops and training programs, thereby creating a vibrant community of users and developers. These initiatives include undergraduate summer research programs and collaborations with Historically Black Colleges and Universities (HBCUs), which aim to attract students to computational molecular sciences and promote STEM disciplines in underrepresented and underprivileged communities. The significance of this project is underscored by its potential to advance scientific knowledge and contribute to national progress by providing a tool with wide-ranging applications. From drug design to materials science, the MB-Fit/MBX software infrastructure is poised to catalyze breakthroughs across various scientific domains. The project also includes a blog that highlights all scientific accomplishments and new discoveries enabled by the MB-Fit/MBX software infrastructure. This platform not only disseminates the project's achievements but also helps to raise the visibility of contributors in the community, thereby fostering a culture of recognition and collaboration. The primary goals of this project include the development of the MBX-Fit/MBX software infrastructure for data-driven many-body molecular simulations, acceleration of computationally intensive terms to GPU accelerators, and integration of the MBX-Fit/MBX software with LAMMPS, i-PI, and RASPA, which are widely used open-source software for molecular simulations. The project also aims to enhance the scalability of the iterative electrostatic solver in MBX and develop mini-apps for efficient evaluation of PIPs on CPUs and GPU accelerators. The project employs a variety of methods and approaches, including machine learning, high-performance computing, and open-source software development. Specifically, the project will explore multi-partition algorithms where a subset of the CPUs will be tasked with computing the 3D FFTs, providing additional opportunities for performance optimization. The development of mini-apps will be a key element of the software design strategy, enabling rapid algorithm development and performance measurement on target architectures. The project will also focus on performance tuning and optimization guided by profiling tools, with a particular focus on achieving better performance with code refactoring. As a result, this project will enable more accurate and efficient molecular simulations, foster a community of users and developers, and provide a platform for training and education in the field. The progress and achievements of the project will be disseminated through a blog that highlights scientific accomplishments and new discoveries enabled by the MB-Fit/MBX software infrastructure. The project also includes a manual on the MB-Fit/MBX website, providing comprehensive guidance for users and developers. The project will benefit from periodic interactions with LAMMPS, i-PI, and RASPA developers to ensure the best performance of the MBX/LAMMPS, MBX/i-PI, and MBX/RASPA interfaces. The project's broader impacts include the organization of workshops and training programs, undergraduate summer research programs, and collaborations with Historically Black Colleges and Universities (HBCUs), all aimed at promoting STEM disciplines in underrepresented and underprivileged communities. This award by the Office of Advanced Cyberinfrastructure is jointly supported by the Division of Chemistry within the Directorate for Mathematical and Physical Sciences. 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.