A gravitational lens is a cosmic funhouse mirror. A galaxy sits almost exactly in front of a more distant one and bends its light into rings and arcs. Fewer than 0.1% of galaxies line up that way, and many rings are too small to resolve from the ground, so catalogues of good lenses stayed small for decades.
The clever bit
Euclid's team built a funnel instead of asking people to look at everything. Five neural networks, trained on lenses already found in earlier Euclid data, scored 68 million objects. The top scorers went to Space Warps, where about 7,000 volunteers made more than 4.5 million classifications of 300,000 objects. Then 67 lens researchers graded the survivors, inspecting more than 30,000 systems. Each stage shrank the pile for the next one.
The catalogue lists 15,072 high-confidence candidates: 6,073 graded as sure lenses and 8,999 as probable. The team says that is more than all earlier catalogues combined, and Euclid's small 2025 preview, covering 63 square degrees, had produced about 600. It includes more than 1,000 candidate disc-galaxy lenses and 84 candidate double-source lenses, two exotic kinds with few earlier examples. Checked against 549 lenses that astronomers had reported earlier in the same patch of sky, the search recovered 428 of them (78%), including 290 of 303 (96%) that a fresh inspection graded as sure lenses.
Citizen scientists have played a central role in delivering what is the largest set of strong lenses known to date.
— Aprajita Verma, associate professor of physics, University of Oxford, and a Space Warps project lead, via ox.ac.uk
Hooray
The paper credits the volunteers by name. The release also comes with automated lens models for the sure and probable systems, plus distances and star properties for the foreground galaxies, so astronomers can start studying the lenses now.
What comes next
This release covers about 14% of Euclid's planned survey area. Extrapolating by area, the team expects more than 100,000 high-confidence lens candidates in the full survey, and it expects the neural networks to keep improving.
