By Scott Gibson
X-ray free-electron laser (XFEL) facilities produce ultrafast pulses from which scientists take stop-action pictures of moving atoms and molecules for research in physics, chemistry, and biology. This technique, called X-ray diffraction imaging, reveals the structure of molecules and their dynamics, a capability that is extremely important in determining many fundamental physical and chemical processes, such as photosynthesis.
The data collection rates of detectors at XFEL facilities are advancing exponentially. An example is the Linac Coherent Light Source (LCLS) at SLAC National Accelerator Laboratory, which will increase its data throughput by three orders of magnitude by 2025. Meanwhile, facility users need to see the fully reconstructed form of their sample as they are taking data so that they can plan their next experiments. Today’s petascale (state-of-the-art) computing resources, however, cannot deliver that capability. The next level of computing, the forthcoming exascale, holds promise to bring innovations to research at the nation’s light sources.
The US Department of Energy (DOE) Exascale Computing Project effort called ExaFEL, Data Analytics at the Exascale for Free-Electron Lasers, is developing an application that will leverage the efficiencies of exascale computing to further XFEL-supported research.
ExaFEL principal investigator Amedeo Perazzo of SLAC said exascale will give researchers two fundamental advantages compared with today’s best computers.
First, it will allow them to run more-complex code for a higher resolution and significantly deeper insight into the processes and structures under measurement. Second, it will dramatically increase the reconstruction rate so that information can be provided in quasi real time rather than weeks. “This will be a revolution for free electron lasers, because it allows users to not fly blind during their experiment,” Perazzo said. Moreover, he explained that receiving such rapid feedback is not only essential to successful outcomes but also to reducing the time between experiment conclusion and research paper publication.
“We have shown on pre-exascale machines like Summit and Cori that we can scale up to about 3 KHz,” he said. “Our goal is to reach 5 KHz, which means 5,000 events per second, and simultaneously, as we move from pre-exascale to exascale machines, increase the kind of analyses that we can do with these data.”
Perazzo said that single-particle imaging is of particular interest to the ExaFEL effort, as its analysis will demand unprecedented computational intensities available only at the exascale. This type of imaging harnesses the ability to reconstruct a single molecule from millions to billions of images likely to be collected during the course of a few days at LCLS when two major upgrades, LCLS-II and LCLS-II-HE, will be commissioned in 2021 and 2025–2026.