SciPost Submission Page
Estimation of Temporal Muon Signals in Water-Cherenkov Detectors of the Surface Detector of the Pierre Auger Observatory
by Margita Kubátová
This is not the latest submitted version.
Submission summary
| Authors (as registered SciPost users): | Margita Kubátová |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2509.18333v1 (pdf) |
| Date submitted: | Sept. 24, 2025, 9:29 a.m. |
| Submitted by: | Margita Kubátová |
| Submitted to: | SciPost Physics Proceedings |
| Proceedings issue: | The 2nd European AI for Fundamental Physics Conference (EuCAIFCon2025) |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
The author(s) disclose that the following generative AI tools have been used in the preparation of this submission:
ChatGPT (OpenAI, GPT-5) was used to correct grammar and improve clarity of text.
Abstract
The Surface Detector (SD) of the Pierre Auger Observatory is a 3000 km$^2$ array of stations, whose main components are Water-Cherenkov Detectors (WCDs) recording ground-level signals from extensive air showers (EASs) initiated by Ultra-High-Energy Cosmic Rays (UHECRs). Understanding the physics of UHECRs requires knowledge of their mass composition, for which the number of ground muons is a key probe. Isolating the muon component is difficult, as different types of particles contribute to the SD signal. We apply a recurrent neural network to estimate the muon content of the SD signals, showing small bias in simulations and weak dependence on selected hadronic interaction model.
Current status:
Reports on this Submission
Report
I find the presentation of the material succinct and descriptive enough to understand the approach of the authors. The findings demonstrate the viability of the applied machine-learning method in the context of Pierre Auger data. In this regard, I support the publication of this proceedings article. Yet, I have very minor comments that the authors may address.
Requested changes
- In Section 2, I read the acronym "LSTM", which seems to be undefined.
- In the left panel of Fig. 4, the points for all tested models seem to be either systematically above or below the 0% line without crossing between negative and positive deviations. Is this some systematic bias that one should expect or an artefact of the neural network?
- In the conclusions, it might be interesting for the reader to find a few other sources of uncertainty that one might need to address once real data is being analysed with this method.
Recommendation
Publish (meets expectations and criteria for this Journal)
Author: Margita Kubátová on 2025-12-15 [id 6147]
(in reply to Report 1 on 2025-12-01)Thank you for the positive review, the responses to the individual points are provided below: 1. The missing LSTM definition has been added. 2. The plot shows the bias for proton and iron induced showers for two hadronic interaction models. Since proton showers contain fewer muons than iron showers at the same energy, predictions from the NN tend to fall between the two cases. As a result, some overprediction for protons and underprediction for irons is expected. Thus, this behaviour reflects the underlying shower physics rather than an artefact of the network architecture. 3. Work on systematics is currently in the progress. Due to the limited length of the proceedings, we only provided a relevant source for readers interested in potential sources of corrections.