When MAGIC met IceCube

Gamma-Ray Astronomy with Neutrino Event Reconstruction Techniques

Jarred Green

Copenhagen, HAMLET-PHYSICS 2025

20 August 2025

Max Planck Institute for Physics

jgreen@mpp.mpg.de

Outline

  1. The Gamma-Ray Sky
  2. Current Event Reconstruction
  3. Graphs
  4. ML Playground
  5. Conclusions

Jarred Green - jgreen@mpp.mpg.de

Crab Nebula

NASA (JWST)

Centaurus A

ESO

1. The Gamma-Ray Sky

Gamma-ray Energy [log]

Flux

[log]

 [ number / area / time ]

a few gamma rays every second in 1 m²

a few gamma rays every year in 1 m²

MeV

GeV

TeV

HIGH ENERGY

Our galaxy (Fermi)

1. The Gamma-Ray Sky

γ ray

e+

e-

e-

γ ray

Flux

[log]

a few gamma rays every second in 1 m²

a few gamma rays every year in 1 m²

"Imaging Atmospheric Cherenkov Telescopes"

MeV

GeV

TeV

VERY HIGH ENERGY

HIGH ENERGY

2. Event Reconstruction

0. Raw data

1. Calibration

2. Cleaning

3. Parameterization

4. ML

5. Science

}

Dimensionality Reduction

Jarred Green - jgreen@mpp.mpg.de

2. Event Reconstruction

0. Raw data

1. Calibration

2. Cleaning

3. Parameterization

4. ML

5. Science

One Pixel, 30ns

Time [ns]
Counts

Jarred Green - jgreen@mpp.mpg.de

2. Event Reconstruction

0. Raw data

1. Calibration

2. Cleaning

3. Parameterization

4. ML

5. Science

Jarred Green - jgreen@mpp.mpg.de

2. Event Reconstruction

0. Raw data

1. Calibration

2. Cleaning

3. Parameterization

4. ML

5. Science

Jarred Green - jgreen@mpp.mpg.de

2. Event Reconstruction

0. Raw data

1. Calibration

2. Cleaning

3. Parameterization

4. ML

5. Science

Jarred Green - jgreen@mpp.mpg.de

2. Event Reconstruction

0. Raw data

1. Calibration

2. Cleaning

3. Parameterization

4. ML

5. Science

{
	'size': 546.88637,
	'xc': -3.78983,
	'yc': 7.29427,
	'length': 7.76769,
	'width': 1.76480,
	'delta': -0.56875,
	 ...
}

"Hillas Parameters" (1977!)

Jarred Green - jgreen@mpp.mpg.de

2. Event Reconstruction

0. Raw data

1. Calibration

2. Cleaning

3. Parameterization

4. ML

5. Science

{
	'size': 546.88637,
	'xc': -3.78983,
	'yc': 7.29427,
	'length': 7.76769,
	'width': 1.76480,
	'delta': -0.56875,
	 ...
}

Random Forests

Jarred Green - jgreen@mpp.mpg.de

2. Event Reconstruction

0. Raw data

1. Calibration

2. Cleaning

3. Parameterization

4. ML

5. Science

  • Particle Type
  • Energy
  • Arrival Direction

Detections

Skymaps

Spectra

Jarred Green - jgreen@mpp.mpg.de

2. Event Reconstruction

ML

1039 \ \text{pixels} \\ \times 50 \ \text{time slices} \\ \times 2 \ \text{telescopes} \\ = 103900 \ \text{params}
{
	'size': 546.88637,
	'xc': -3.78983,
	'yc': 7.29427,
	'length': 7.76769,
	'width': 1.76480,
	'delta': -0.56875,
	 ...
}
  • Particle Type
  • Energy
  • Arrival Direction

Dimensionality

reduction

tl;dr

Jarred Green - jgreen@mpp.mpg.de

3. Graphs for MAGIC?

Proton

Gamma

Muon

103900 \ \text{params per event (raw)}
\sim 300 \frac{\text{triggers}}{\text{s}}

Jarred Green - jgreen@mpp.mpg.de

\text{S/N ratio} \sim \frac{1}{10000}

3. Graphs for MAGIC

an IceCube case study

  • A GNN framework for Neutrino observatories
  • Easy to adapt to other detectors

DynEdge

DeepIce

Proven models

Jarred Green - jgreen@mpp.mpg.de

3. Graphs for MAGIC

We effectively deal with:

  • irregular camera geometry
  • sparse data
  • non gridlike pixels
  • arrays with different camera types

Why Graphs?

Jarred Green - jgreen@mpp.mpg.de

3. Graphs for MAGIC

time

Jarred Green - jgreen@mpp.mpg.de

4. ML Playground

Particle Classification with

Model:

  • Binary classification
  • DynEdge with 12 NN
  • Separate each telescope
  • Cross-attention 
  • Only need 5-10% of training data
  • z = time

DynEdge

Jarred Green - jgreen@mpp.mpg.de

Confidence of gamma classification

Gammas

Protons

4. ML Playground

Direction Reconstruction with

Model:

  • 2nd place winner of IceCube Kaggle Competition
  • Transformer + DynEdge
  • Very robust model for neutrino direction reconstruction
  • z = telescope number

DeepIce

Radial Error [deg]
Cumulative Fraction

Jarred Green - jgreen@mpp.mpg.de

MC 68% containment <0.08 deg

5. Conclusions

MAGIC Raw Data

  • Contains full time evolution of events

  • Saves on data processing

  • Are naturally represented as graphs

Reconstruction with GraphNET

  • Apply domain knowledge from IceCube
  •  'Built-in' neutrino models work on gamma-ray data
  • Robust performance with minimal modifications

Further evidence for the benefits
of cross-domain collaboration

Jarred Green - jgreen@mpp.mpg.de

5. Future Motivation

70+ Telescopes!!

Cherenkov Telescope Array

Jarred Green - jgreen@mpp.mpg.de

Thank you!

Jarred Green - jgreen@mpp.mpg.de

Credit: G. Ceribella