Abstract
The performance demands of future particle-physics experiments investigating the high-energy frontier pose a number of new challenges, forcing us to find new solutions for the detection, identification, and measurement of final-state particles in subnuclear collisions. One such challenge is the precise measurement of muon momenta at very high energy, where the curvature provided by conceivable magnetic fields in realistic detectors proves insufficient to achieve the desired resolution. In this work we show the feasibility of an entirely new avenue for the measurement of the energy of muons based on their radiative losses in a dense, finely segmented calorimeter. This is made possible by the use of the spatial information of the clusters of deposited photon energy in the regression task. Using a homogeneous lead-tungstate calorimeter as a benchmark, we show how energy losses may provide significant complementary information for the estimate of muon energies above 1 TeV.
The Challenge of High-Energy Muon Measurement
In current particle physics experiments like those at the Large Hadron Collider (LHC), muon momentum is traditionally measured by tracking the particle's curvature in a magnetic field. However, as we look toward future colliders operating at the multi-TeV (tera-electronvolt) scale, this method faces a fundamental limitation: the faster a muon travels, the straighter its path becomes. Consequently, the resolution of magnetic tracking scales linearly with energy, making precise measurements increasingly difficult at the highest energies. For a 1 TeV muon, current resolutions range between 6% and 20%, a figure that would degrade further at future facilities like a Muon Collider or the FCC.
While muons are typically characterized as "minimum ionizing particles" (MIPs) that pass through matter with minimal interaction, this behavior changes at extremely high energies. Above approximately 100 GeV, muons enter a regime where they lose significant energy through radiative processes, including bremsstrahlung, pair production, and photonuclear interactions. As shown in Figure 1, the mass stopping power begins to rise significantly in this "radiative" region. Historically, these stochastic energy bursts were considered background noise for tracking; this research proposes using them as a primary signal for energy estimation.
Calorimetric Reconstruction and Machine Learning
To test the feasibility of this approach, researchers utilized Geant4 to simulate a homogeneous lead-tungstate ($PbWO_4$) calorimeter. This benchmark detector was designed to be exceptionally deep (2 meters) and highly granular, consisting of 51,200 individual cells. This high granularity allows for the capture of detailed spatial patterns of energy deposition.
The core of the methodology lies in extracting information from the messy patterns of energy deposition (visualized in Figure 2). The team developed a regression pipeline that moved beyond simple energy summation:
- Feature Extraction: 16 high-level features were extracted per event, including total energy deposited, spatial moments (describing the spread of energy), and cluster-based statistics.
- The "Hyperball" Algorithm: A customized k-Nearest Neighbors (kNN) regressor was employed. This model used an ensemble of weak learners and gradient boosting to handle the high-dimensional feature space.
- Bias Correction: To counteract the "regression to the mean" typical of kNN models, a specific bias-correction function was implemented to ensure accurate predictions across the 100 GeV to 2 TeV range.
The study utilized a dataset of 411,000 simulated muon interactions, with 396,000 used for training and 15,000 for validation. The energy distribution was kept flat to ensure the model learned effectively across all scales.
Performance and Resolution
The most significant finding of this study is that the spatial pattern of energy deposition—the "fingerprint" of the muon's path—contains more information than the total energy sum alone. The customized kNN regressor significantly outperformed a baseline "energy-sum" model in both linearity and resolution.
For muons in the 1.5 to 2 TeV range, the kNN model achieved a 68.3% central interval of [904, 2063] GeV. While seemingly broad, this represents a massive improvement over the energy-sum model, which produced a nearly uninformative interval of [70, 2303] GeV. At the 2 TeV mark, this calorimetric approach begins to show competitive performance with the extrapolated resolutions of state-of-the-art magnetic trackers like those in the ATLAS or CMS experiments. Crucially, calorimetric resolution typically improves with energy (scaling with $1/\sqrt{E}$), making it perfectly complementary to tracking, which degrades as energy increases.
Outlook and Future Directions
This feasibility study demonstrates that the spatial information of radiative losses can serve as a viable proxy for muon energy at the multi-TeV scale. However, several challenges remain for practical implementation. The current simulation uses an idealized cuboid geometry and does not account for electronic noise or "pileup" (background radiation from simultaneous collisions).
Future work may involve applying more advanced deep learning architectures, such as Convolutional Neural Networks (CNNs) or Graph Neural Networks (GNNs), which are naturally suited to extracting features from the raw 3D energy images provided by granular calorimeters. By proving that radiative "bursts" are a signal rather than noise, this work opens a new avenue for detector design in the next generation of high-energy physics experiments.
Summary updated