Packs of around a billion electrons each move in close proximity to the speed of light via metal tubes whenever the linear accelerator at the SLAC National Accelerator Laboratory is operating. The particle beam produced by these electron bunches is used to examine a variety of topics, including the atomic behaviour of molecules and new materials.
However, it is challenging to predict the exact appearance of a particle beam as it passes through an accelerator; as a result, scientists frequently have only a rough idea of how a beam will behave in an experiment.
Now, scientists at the University of Chicago, the Department of Energy’s Argonne National Laboratory, and the SLAC National Accelerator Laboratory have created an algorithm that more accurately forecasts a beam’s distribution of particle positions and velocities as it flies through an accelerator.
As accelerator facilities operate at higher and higher energies and produce more complex beam profiles, this in-depth beam information will help scientists conduct their experiments more reliably. In Physical Review Letters, the researchers described their methodology and algorithm in full.
“We have a lot of different ways to manipulate particle beams inside of accelerators, but we don’t have a really precise way to describe a beam’s shape and momentum,” SLAC accelerator scientist and lead author Ryan Roussel said. “Our algorithm takes into account information about a beam that is normally discarded and uses that information to paint a more detailed picture of the beam.”
A lot of potentially important information is lost when researchers characterise the positions and speeds of particles in a beam in terms of a few summary statistics that give a general idea of the beam. As an alternative, beam scientists can make several measurements of the beam and attempt to reconstruct, occasionally using machine learning, what the beam would appear to be under various experimental conditions. However, those techniques demand a significant amount of data and processing capacity.
In order to conduct this work, the team adopted a novel strategy: they developed a machine learning model that makes use of our knowledge of beam dynamics to forecast the distribution of particle locations and speeds within the beam, also referred to as the beam’s phase space distribution.
The team applied their model to analyse experimental data from the Argonne Wakefield Accelerator at the DOE’s Argonne National Laboratory in order to put their theories to the test. The researchers were able to precisely reconstruct the fine characteristics of the beam using only 10 data points by incorporating the physics of particle beam dynamics with the experimental data. This task might have required up to 10,000 data points for some machine learning algorithms that do not contain a model of beam physics.
“Most machine learning models don’t directly include any notion of particle beam dynamics to speed up learning and reduce the amount of data required,” SLAC accelerator scientist and co-author Auralee Edelen said. “We’ve shown that we can infer very complicated high-dimensional beam shapes from astonishingly small amounts of data.”
Currently, the algorithm can recreate a model of a beam along its left-right and up-down axes, simulating a pancake-like motion of the particle bunch down the accelerator path. 4D beam phase space reconstruction is this form of reconstruction. The algorithm will then be experimentally tested on the reconstruction of complete 6D phase space distributions, which contain particle locations and speeds along the beam’s direction of travel.
According to Roussel, the programme represents a significant paradigm shift in the way experimental accelerator data is now examined at sites.
“We can now use particle beam data in a more comprehensive, powerful way to improve our scientific goals at accelerators everywhere,” he said.
