Enabling Scientific Discovery at Next-Generation Light Sources with Advanced AI and HPC

Abstract

The synchrotron and free electron laser light sources, large scientific user facilities, are in the position to help solve some of the most challenging and novel scientific questions facing the world, ranging from the design of new materials to manipulate classical and quantum information with high fidelity and ultra low power consumption, to enabling systems for efficient energy storage, transportation, and conversion that will drive the emerging economy based on renewable energy, to understanding the structure and motion of protein molecules to enable individualized medicine. These scientific opportunities will be addressed by new measurement techniques, technological advances in detectors, multi-modal data utilization, and advances in data analysis algorithms, all of which are being driven to a new level of sophistication. Over the next decade, it is estimated that the US light sources will generate in the exabyte (EB) range of data, require tens to 1,000 PFLOPS of peak on-demand computing resources, and utilize billions of core hours per year. Scientific discovery on this scale will be enabled by data management and workflow tools that integrate user facility instruments with sufficient computing, networking, and storage resources, on-demand utilization of super-computing environments to enable real-time data processing, real-time data analysis capabilities to significantly reduce data volumes and provide feedback during experiments to improve data quality and to drive the direction of ongoing measurements, the application of advanced machine learning algorithms to make crucial experiment decisions, and the integration of simulations and model-based approaches to facilitate automated experiment design and steering of data collection.