main concepts scientists sections
SimplePopularAdvanced

Rings of Light: How to Find Invisible Matter by Straightening the Picture ✛ ✎ ✦

Original: "Polar coordinate transformations for machine learning based dark matter subhalo detection in strong gravitational lenses"
· Jacob M Campbell, Simon Dye, Emma Chapman, Adam Moss
arXiv:2607.02663 · 2026-07-02 · arXiv non-exclusive · independent analysis · 1 min
Galaxies Cosmology
By turning a ring into a straight line, scientists taught a computer to better detect invisible clumps of dark matter.

The gravity of galaxies bends light from distant stars — this is gravitational lensing. Sometimes the light gathers into rings, described by Einstein. Hidden within these rings is dark matter: it's five times more abundant than ordinary matter, and without it, galaxies would fall apart (as shown by Vera Rubin and Fritz Zwicky). Tiny clumps of dark matter, like bubbles under a sticker, betray its composition by their number.

Clumps are invisible and only slightly jitter the bright arcs. Images from Hubble are analyzed by machine learning programs, but they divide the frame into squares, while the ring is round. The fix is simple: mathematically unroll the ring into a straight band — like peeling a label off a jar. After this straightening, the program finds 15% more clumps.

Strikingly, the accuracy gain reached 50% when the program had no prior training. It's like solving a puzzle in your native language rather than in a mirror reflection.

Upcoming telescopes (Euclid) will deliver thousands of such rings. By comparing the number of clumps with computer simulations (accounting for light stretched by the expansion of space), we will learn whether dark matter is cold (many clumps) or warm (few).

Whole graph →

To get a sharp image of a lens, Hubble sometimes collects light over 50 orbits around Earth.

\theta_E = \sqrt{\frac{4GM D_{ls}}{c^2 D_l D_s}}
The Einstein angle is the radius of the gravitational ring. The more massive the lens and the closer it is to the line of sight, the larger it is. Tiny subhalos, like invisible satellites, slightly deform the perfect circle, creating microscopic kinks—by these we detect them.
Scientists
Fritz ZwickyBernhard RiemannJoseph WeberKarl SchwarzschildKip ThorneRainer Weiss
Concepts
neural network polarimetry Convolutional Neural Networks weak lensing gravitational lensing analysis generalization dark matter subhalo Large Language Models Machine Learning deep learning in medical imaging dark matter gravitational waves microlensing Lorentz transformations ai astrophysics training data polarimetric remote sensing image classification computer vision molecular property prediction polarimetry Natural Language Processing