Polar coordinate transformations for machine-learning-based dark matter subhalo detection in strong gravitational lenses
- 1School of Physics and Astronomy, University of Nottingham , University Park, Nottingham NG7 2RD ,
Abstract
Strong gravitational lensing provides a powerful probe of dark matter, particularly on small scales where the gravitational effects of dark matter subhaloes within galaxies can manifest as perturbations within the extended arcs of gravitationally lensed sources. We investigate whether transforming lensed images into polar coordinates improves the ability of convolutional neural networks to infer subhalo mass. We introduce a machine learning architecture that outputs a prediction uncertainty alongside a mass prediction to enable assessment of network reliability. Using simulated Hubble Space Telescope observations, we compare our models trained on Cartesian and polar representations under different initialization schemes, noise levels, and subhalo concentrations (c=60, c=30). We find that polar-transformed inputs consistently yield higher subhalo detection fractions than standard Cartesian images across all tested masses. For subhaloes with mass 109 M⊙ ≤ M ≤ 109.5 M⊙, the fraction of subhaloes the network is able to detect increases by ∼ 15 per cent. Pretrained networks outperform randomly initialized networks, and the polar transform consistently improves network performance in both low signal-to-noise data and for lower-concentration subhaloes. These results demonstrate that presenting strong lensing images in a polar representation provides a computationally inexpensive way of improving convolutional neural network (CNN)-based subhalo detection.
Related
- Deep Learning Detection of Beyond-General-Relativity Deviations in Gravitational-Wave Signals: A Detection-Threshold Study with Real LIGO Noise
- Gravitational-wave parameter estimation with machine-learning generated surrogate waveforms
- Architecture Acceleration of Machine Learning Gravitational Waveform Surrogate Models
- BRAHMa: Bar Recognition And Hatching using Machine learning -- a framework for oriented bounding box detection of galactic bars
- Toward claiming a detection of gravitational memory
BibTeX
@article{campbell2026polar,
title = {{Polar coordinate transformations for machine-learning-based dark matter subhalo detection in strong gravitational lenses}},
author = {J M Campbell and S Dye and E Chapman and A Moss},
journal = {Monthly Notices of the Royal Astronomical Society},
year = {2026},
month = oct,
doi = {10.1093/mnras/stag1955},
url = {https://academic.oup.com/mnras/article/doi/10.1093/mnras/stag1955/8923674},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Polar coordinate transformations for machine-learning-based dark matter subhalo detection in strong gravitational lenses
AU - J M Campbell
AU - S Dye
AU - E Chapman
AU - A Moss
PY - 2026
DA - 2026/10/15
JO - Monthly Notices of the Royal Astronomical Society
DO - 10.1093/mnras/stag1955
UR - https://academic.oup.com/mnras/article/doi/10.1093/mnras/stag1955/8923674
AB - Strong gravitational lensing provides a powerful probe of dark matter, particularly on small scales where the gravitational effects of dark matter subhaloes within galaxies can manifest as perturbations within the extended arcs of gravitationally lensed sources. We investigate whether transforming lensed images into polar coordinates improves the ability of convolutional neural networks to infer subhalo mass. We introduce a machine learning architecture that outputs a prediction uncertainty alongside a mass prediction to enable assessment of network reliability. Using simulated Hubble Space Telescope observations, we compare our models trained on Cartesian and polar representations under different initialization schemes, noise levels, and subhalo concentrations (c=60, c=30). We find that polar-transformed inputs consistently yield higher subhalo detection fractions than standard Cartesian images across all tested masses. For subhaloes with mass 10^9 M_⊙ ≤ M ≤ 10^9.5 M_⊙, the fraction of subhaloes the network is able to detect increases by ∼ 15 per cent. Pretrained networks outperform randomly initialized networks, and the polar transform consistently improves network performance in both low signal-to-noise data and for lower-concentration subhaloes. These results demonstrate that presenting strong lensing images in a polar representation provides a computationally inexpensive way of improving convolutional neural network (CNN)-based subhalo detection.
SN - 0035-8711
SN - 1365-2966
ER -