Improving Sensitivity to Rare Processes at the LHC with Geodesic-Aware Graph Neural Networks
This paper proposes a geodesic-aware graph neural network that maps hadronic constituents onto a unit sphere to exploit color flow information from soft QCD radiation, significantly improving sensitivity to rare top quark flavor-changing neutral current processes at the LHC compared to standard analyses.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the author. For technical accuracy, refer to the original paper. Read full disclaimer
The Large Hadron Collider is a machine built to smash protons together at incredible speeds, creating a shower of new particles that physicists study to understand the fundamental rules of the universe. Among these particles, the top quark is the heaviest known, and the Higgs boson is the particle that gives mass to other matter. Sometimes, these two particles interact in ways that are extremely rare and forbidden by the standard rules of physics. Finding these rare interactions is like looking for a single, specific grain of sand on a beach, because the background noise of ordinary particle collisions is so loud and overwhelming. The challenge is that the rare events the scientists want to find often look exactly the same as the common events they want to ignore. Both produce jets of particles that look identical when measured by the detector, making it nearly impossible to tell them apart using traditional methods of measuring speed and direction.
To solve this, a team of researchers has developed a new way of looking at the data that focuses on something invisible to standard measurements: the flow of color charge. In the world of particle physics, quarks and gluons carry a property called color charge, which dictates how they stick together. When a heavy particle like the Higgs boson decays into two quarks, those quarks are linked by a single, tight string of color charge, creating a specific pattern of soft, low-energy radiation around them. In contrast, when similar-looking jets are produced by ordinary background processes, their color connections are messy and spread out, linking them to other parts of the collision rather than just to each other. This difference in the "color flow" leaves a subtle fingerprint in the way particles are scattered, but this fingerprint is usually lost in the noise of the data.
The researchers in this study decided to treat this invisible color flow as a geometric shape. They took the particles from each collision and mapped them onto the surface of a sphere, much like placing points on a globe. They chose a specific center point for this map based on the two jets that might have come from a Higgs boson. By doing this, they could separate the particles into two distinct groups: those that fell inside the region between the two jets and those that fell outside. On their spherical map, the particles inside this region landed on one half of the globe, while the particles outside landed on the opposite half. This created a clear visual distinction between the rare signal events, which tended to fill the space between the jets, and the common background events, which were more evenly spread out or concentrated elsewhere.
To make sense of this new map, the team used a type of artificial intelligence called a graph neural network. Instead of just looking at the particles as a list of numbers, this computer program treated them as a connected web, where the distance between any two particles was measured along the curved surface of the sphere, just as one would measure the distance between two cities along the Earth's surface. This allowed the computer to learn the specific patterns of radiation that indicated a rare event. When they tested this new method against a standard computer program that looked at the same data but without this spherical mapping, the difference was striking. The new system was able to spot the rare events much more effectively, filtering out the background noise with far greater precision.
The team applied this technique to a specific search for the top quark changing into a Higgs boson and a lighter quark, a process that is predicted to happen so rarely in the standard model that it has never been observed. They simulated collisions at the energy levels expected at the Large Hadron Collider and ran their analysis on these simulated events. The results showed that their new geometric approach could improve the sensitivity of the search by about ten times compared to the standard method. Specifically, they found that with the data expected from a certain amount of running time, their method could set limits on how often this rare decay happens that are roughly ten times lower than what was previously possible. This means that if this rare event were happening at a level just slightly above the current limits, their new method would be much more likely to find it.
It is important to note that these results come from computer simulations of the data, not from a new discovery in the actual detector yet. The researchers used a fast, simplified model of the detector to generate these numbers, and they acknowledged that real-world conditions, such as the complex ways particles interact with the detector material, could change the outcome. They also noted that their method relies on assumptions about how the particles behave, which might need to be tested more rigorously with larger datasets. However, the study demonstrates that by changing the way data is represented—turning a flat map of particle positions into a curved, geometric sphere—it is possible to unlock information that was previously hidden. This suggests that in the future, looking at the shape of particle collisions, rather than just their speed and energy, could be a powerful tool for finding the most elusive secrets of the universe.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.