The team is harnessing the power of machine learning to analyze data from a particle physics experiment testing our fundamental understanding of the universe
Study: Improved analysis of non-resonant Higgs boson pair production in the bb̄τ+τ− final state with 196 fb−1 of data collected at √s = 13 TeV and 13.6 TeV with the ATLAS detector (DOI: 10.48550/arXiv.2607.26879)
Is the universe as stable as we think it is?
That's one of the big questions that particle physicists worldwide are preparing to answer with the Large Hadron Collider, or LHC-the world's most powerful particle accelerator-when its upgrade is completed in about four years. In the meantime, researchers, including a cohort at the University of Michigan, are working to sharpen their analytical tools and techniques to make the most of LHC's current and future data.
A team including U-M physicists has now reported record-setting sensitivity in spotting a specific interaction within LHC's data that may help answer fundamental questions about our universe. In particular, the analysis uses an advanced AI algorithm to spot signatures researchers are looking for to understand how the Higgs boson, the famous fundamental particle that helps explain how subatomic particles have mass, interacts with itself.
"This analysis is the most sensitive in the world to this specific physics," said Greg Myers, a research fellow in the U-M Department of Physics.
Myers and U-M doctoral student Tamas Baer played key roles in developing the new method, with contributions from former postdoc Kevin Nelson and recent doctoral graduate Dustyn Hofer. The U-M team was led by Tom Schwarz, a professor of physics who has been working on this particular analysis for a decade. U-M has also been deeply involved with this branch of research at LHC, starting with Bing Zhou and Jianming Qian in 2010.
The new analysis is about 60 times more sensitive than those early results, with an improvement of about 65% compared with its immediate predecessor, Schwarz said.
This work was presented at the International Conference on High-Energy Physics in Natal, Brazil, in August and was recently published as a preprint scientific article on arXiv. The research was supported by agencies in more than three dozen countries, including the U.S. Department of Energy and the U.S. National Science Foundation.
Massive questions and subatomic particles
When the LHC fired up in 2008, the biggest looming question at the time was whether the new accelerator could find the elusive Higgs boson. The particle had been predicted to exist decades earlier as theorists refined the Standard Model of Particle Physics, the best and most comprehensive accounting of the universe's fundamental particles and forces.
Yet experiments could not provide direct physical evidence of the Higgs boson's existence before the LHC, which is operated by CERN, the European Organization for Nuclear Research. In 2012, two independent LHC experiments-one called ATLAS and the other CMS-would provide that evidence. A year later, two European scientists behind the theory predicting the particle, François Englert and Peter Higgs, the particle's namesake, earned the Nobel Prize in Physics.
With the discovery, physicists could then start asking and answering questions about how the Higgs worked. But one of those questions, how it interacts with itself, is a critical process within the Standard Model that remains an important mystery.
"It's kind of the big thing left on the table," Schwarz said. "Understanding that will really tell us a lot about not just how the subatomic world works, but also how the universe works."
The LHC upgrade will help power the experiments to probe that question. Depending on what physicists find, it could hint at undiscovered physics beyond the Standard Model or even suggest the universe is more or less stable than we currently think.
As the upgrade gets underway, though, Baer, Myers, Schwarz and their colleagues on the ATLAS experiment are also evolving how they analyze data with modern computational techniques. This not only helps prepare for future data, but lets them squeeze as much knowledge as possible from the data the LHC has already collected.
U-M has had a large team of scientists working on all aspects of the ATLAS experiment, including Schwarz, Qian and Zhou, as well as Junjie Zhu and Christian Herwig.
Trillions of collisions
The tell-tale signatures that the team is looking for are found in the aftermath of the LHC sending two beams of protons-subatomic particles themselves-careening into each other at nearly the speed of light.
What ensues is a quantum kerfuffle that produces a shower of new particles. Many of these new particles are unstable, like the Higgs boson, meaning they decay or transform into other particles before they can be observed directly. But some of those resulting particles-including electrons, their heavier cousins called muons and massless particles of light known as photons-generate a signal in the ATLAS experiment.
Knowing which particles are detected along with their properties, such as their energy and trajectory, allows researchers to work backward to reconstruct and study what came before. Sometimes, a single proton-proton collision should produce two Higgs bosons, which would generate a signal scientists can look for to find an opportunity to probe the particle's self-interaction.
"It's really the only way to access it at the moment," Myers said.
The rub is that those specific double-Higgs events would be exceedingly rare.
"It happens maybe once every trillion collisions," Baer said. "If something is that rare, the best way to find it is to just go for it a trillion times and then it'll be in there once."
The team's project included data from all of LHC's Run 2, which ran from 2015 to 2018, and the first two years of Run 3, the full extent of which went from 2022 until this June. During those times, the experiment was running day and night, creating roughly 30 million collisions per second. Baer and Myers were instrumental in developing a new approach that spots potential double-Higgs events with the help of AI in the oceans of resulting data.
"What we worked on was to create a graph neural network, a type of machine learning AI algorithm, to pick out as best we can the events we're looking for from the trillions of other background events that we're not interested in," Baer said.
Beyond raising the bar in the sensitivity of spotting those events, the team's work also found that the rate of double-Higgs production could be more than twice what the Standard Model predicts. While that figure might sound tantalizing, Schwarz stressed that there's enough uncertainty that it could easily be explained by reasons that will prove to be mundane as the work continues.
So, for now, there is no new evidence suggesting the universe is more unstable than we thought. On the flipside, though, the research also doesn't slam the door on that possibility.
