Imagine being able to observe the inner workings of a chemical reaction or material as it changes and reacts to its environment—that’s the kind of thing researchers can do with a high-speed “electronic camera” called a Ultrafast electron diffraction in megaelectronvolts Instrument (MeV-UED) at Linac coherent light source (LCLS) to the United States Department of Energy SLAC National Accelerator Laboratory.
Now, thanks to two new studies, researchers at SLAC, Stanford and other institutions have discovered how to capture these tiny, ultra-fast details more accurately and efficiently. first studyRecently published in Structural Dynamics, a team invented a technique to improve the temporal resolution of the electronic camera. In a second, published In Nature Communications, researchers trained and used artificial intelligence (AI) to tune the MeV-UED electron beam and adapt it to a variety of experimental needs.
“These effects are profound for the advancement of instrumentation and beam diagnostics for SLAC’s electron accelerators and will enable a new frontier in exploring novel effects with unprecedented precision,” said Mohamed Othman, an associate scientist at SLAC and co-author of both papers.
Timing is everything
Chemical reactions happen quickly: sometimes key events occur in millionths of a billionth of a second, or femtoseconds. Capturing these femtosecond events is the terrain of a field known as ultra-fast science This requires some of the most advanced scientific instruments in the world – instruments like MeV-UED.
The MeV-UED takes snapshots by hitting samples with a beam of electrons and record what happens to matter as electrons pass through it. The result is a molecular movie that allows scientists to observe the behavior of molecules and atoms at ultrafast speeds and gain insights into processes that are critical to energy solutions and innovative new materials and medicines, among other things.
The problem is that the MeV-UED beam is made up of packets of electrons, or electron pulses, which can be unruly. When the electron pulses arrive at the material sample, there is a slight gap in the arrival time between the first and last electrons in the pulse. This time gap, along with the time variations between pulses, called jitter, make it difficult to accurately determine when events occur in each frame of the electron camera.
The SLAC Team Previously reported The use of terahertz radiation, which lies between microwaves and infrared light on the electromagnetic spectrum, and the addition of a compressor in the MeV-UED have improved the instrument’s time resolution. The compressor uses terahertz radiation to shorten the time propagation of an electron pulse using a method aptly called packet compression.
In their quest to tame electron bunches further, the team combined bunch compression with another method called time stamping: After the pulse interacts with the sample and hits the detector, timing information is encoded into the electron camera image. With simple time sorting, users can more precisely determine the timing of each frame or movie.
The combination of packet compression and timestamping increased time precision and reduced jitter. “Researchers could use this technique to observe extremely fast time scales, especially for atomic motion in materials,” Othman said. “This atomic microscope can be used in fundamental sciences: materials science, chemistry, green energy, quantum information and much more. Reaching femtosecond time scales is essential to study these scientific fields.”
The success of this prototype led to the next step of building an instrument with these combined capabilities. “We’re trying to push the boundaries of what the MeV-UED can do, for example in terms of time. The MeV-UED is a facility used by the DOE, so we want to build this instrument that can be an option for users,” Othman said.
The power of AI
Researchers from all over the world come to SLAC’s MeV-UED to conduct their experiments, and their needs vary widely. For each experiment, beam operators must optimize 20 to 30 parameters, such as the beam spot size, and consider tradeoffs among all the parameters. Fuhao Ji, a SLAC scientist and lead author of the paper, likened the tuning process to changing the ingredients in a recipe when baking bread to suit a customer’s taste: There are many factors to consider, and everyone’s taste is a little different.
Currently, experienced operators make all of these choices themselves with the help of an automated process, but it’s not as efficient as it could be. To make this work more smoothly, SLAC researchers on the lab’s accelerator and instrumentation side teamed up with the lab’s AI experts to implement a special AI model, called multi-objective Bayesian optimization (MOBO), to tune the MeV-UED’s electron beam directly, online. This approach could tune as well as an experienced operator and at least ten times faster than the automated process. Because users have a fixed beam duration, this means less hands-on time and more time to run their experiments and collect data.
Before launching the AI model, the SLAC team had to train it to know not only what to look for, but also how to evaluate tradeoffs between beam parameters. The model learned by doing: The researchers ran experiments and collected data as they normally would, then fed that data into the model, which learned how different parameters interacted to shape the beam.
Like other AI models, MOBO can predict new outcomes from new parameters, which is particularly useful when a user needs a beam setting that has never been used before. The model also provides a more complete picture of the experimental system.
“This is the result of close collaboration between MeV-UED and the Machine Learning Group of the Accelerator Directorate and paves the way for the ultimate goal of establishing an end-to-end automated intelligent scientific user facility at MeV-UED,” Ji said, where AI algorithms would co-optimize all components of the entire system, from the electron source to the accelerator, light source, sampling parameters and detector.
Ji and his colleagues are looking to expand the capabilities of the MOBO tool. Their next step is to adopt another AI tool, running Bayesian algorithms, to further speed up the optimization process and achieve better performance.
“We hope this will have a broad impact on research in different disciplines, such as physics, chemistry, biology and quantum materials, in complex and large-scale scientific facilities,” Ji said.
The research was supported by the DOE Office of Science and SLAC’s Laboratory Directed Research and Development program. LCLS is an Office of Science user facility.
Quotes:
M. Othman et al., Structural Dynamics, April 22, 2024 (https://doi.org/10.1063/4.0000230)
F. Ji et al., Nature Communications, June 3, 2024 (https://doi.org/10.1038/s41467-024-48923-9)