Neutrinos stand as one of the most enigmatic components within the Standard Model, our fundamental framework for understanding natural forces and particles. Despite being among the most plentiful particles throughout the cosmos, they rarely engage with matter, presenting significant challenges for detection. A persistent puzzle in neutrino research emerged from the Mini Booster Neutrino Experiment (MiniBooNE), conducted at Illinois' Fermi National Accelerator Laboratory (Fermilab) between 2002 and 2017. This experiment detected substantially more electron-producing neutrino interactions than our current Standard Model understanding would predict—leaving physicists searching for explanations.
By 2007, researchers conceived MicroBooNE as a successor experiment, which has recently completed its data collection phase at Fermilab. MicroBooNE represents an ideal investigation into the MiniBooNE excess through its implementation of innovative liquid argon time projection chamber (LArTPC) technology, generating high-resolution visualizations of particles created during neutrino interactions. This advancement in deep learning neutrino detection technology has opened new frontiers in particle physics research.
Physics graduate students Nicholas Kamp and Lauren Yates, working alongside Professor Janet Conrad from MIT's Laboratory for Nuclear Science, have spearheaded MicroBooNE's artificial intelligence-driven search for neutrino excesses in the Fermilab Booster Neutrino Beam. Their work exemplifies how AI applications in particle physics research are revolutionizing our understanding of fundamental particles. In this conversation, Kamp explores the future of the MiniBooNE anomaly in light of MicroBooNE's latest discoveries.
Q: What makes the MiniBooNE anomaly so significant in modern physics?
A: One of the most compelling questions in neutrino physics centers on the potential existence of a theoretical particle known as the "sterile neutrino." The discovery of any new particle would represent a monumental breakthrough, offering insights into the broader theoretical framework that explains the multitude of particles we observe. The predominant explanation for the MiniBooNE excess involves incorporating such a sterile neutrino into the Standard Model. Through neutrino oscillation effects, this sterile neutrino would appear as an increased presence of electron neutrinos in MiniBooNE's observations.
Numerous additional anomalies in neutrino physics suggest this particle might exist. However, reconciling these anomalies with MiniBooNE through a single sterile neutrino model proves challenging—the complete picture doesn't quite align. Our MIT research group focuses on developing new physics models that might potentially explain this comprehensive puzzle, leveraging machine learning physics experiments Fermilab to push boundaries.
Q: How has our understanding of the MiniBooNE excess evolved recently?
A: Our comprehension has advanced considerably due to progress in both experimental and theoretical domains.
Our team has collaborated with physicists from Harvard, Columbia, and Cambridge universities to investigate new photon sources that could appear in a theoretical model also featuring a 20 percent electron signature. We developed a "mixed model" incorporating two types of exotic neutrinos—one transforming into electron flavor and another decaying into photons. This research will soon appear in Physical Review D.
On the experimental front, more recent MicroBooNE findings—including a deep-learning analysis where our MIT group made substantial contributions—detected no excess of neutrinos producing electrons in the MicroBooNE detector. Considering MicroBooNE's measurement precision, this indicates the MiniBooNE excess cannot be entirely attributed to additional neutrino interactions. If not electrons, then photons must be responsible, as they're the only particles capable of producing similar signatures in MiniBooNE. However, we're confident these aren't photons from known interactions, as those occur at limited rates. Therefore, they must originate from something novel, such as the exotic neutrino decay proposed in our mixed model. Currently, MicroBooNE is developing a search method to isolate and identify these additional photons. The scientific community eagerly awaits these results!
Q: You mentioned your group's involvement in deep-learning-based MicroBooNE analysis. Why implement deep learning in neutrino physics research?
A: When humans examine images of cats, they can typically distinguish between species with relative ease. Similarly, when physicists analyze images from a LArTPC, they can generally differentiate between particles produced in neutrino interactions without significant difficulty. However, due to the subtle nature of these differences, both tasks present considerable challenges for conventional algorithms.
MIT serves as a hub for deep-learning innovation. Recently, for instance, it became home to the National Science Foundation AI Institute for Artificial Intelligence and Fundamental Interactions. It was logical for our group to leverage the extensive local expertise in this field. We've also had the privilege of collaborating with outstanding teams at SLAC, Tufts University, Columbia University, and IIT, each bringing substantial knowledge about the connections between deep learning and neutrino physics.
A fundamental concept in deep learning involves "neural networks"—algorithms that make decisions (such as identifying particles in a LArTPC) based on previous exposure to training data. Our group published the first paper on particle identification using deep learning in neutrino physics, demonstrating its power as a technique. This represents a primary reason why MicroBooNE's recently released deep learning analysis results place strong constraints on interpreting the MiniBooNE excess as electron neutrinos.
Overall, it's incredibly fortunate that much of the groundwork for this analysis occurred in MIT's AI-rich environment, where artificial intelligence solving neutrino mysteries has become a reality rather than science fiction.
