A sophisticated AI algorithm analyzing the latest DESI (Dark Energy Spectroscopic Instrument) data has identified 1,210 new gravitational lenses, effectively doubling the catalog of these rare cosmic phenomena.
Einstein's general theory of relativity predicts that massive objects like galaxies or galaxy clusters warp spacetime. When Earth aligns perfectly behind such a massive foreground object with a distant background source, the light from that background is distorted and magnified, often appearing as dramatic arcs encircling the lens.
These "gravitational lenses" allow astronomers to observe otherwise invisible distant objects.
Yet, gravitational lensing is rare: astronomers estimate it occurs in only about one in 10,000 massive galaxies. Detecting them requires precise alignment. "A massive galaxy distorts spacetime around it, but you usually don't notice unless a background object is perfectly aligned behind it," explains Xiaosheng Huang from the University of San Francisco.
Just two years ago, only around 300 such lenses were known; recent efforts had doubled that to about 600—still insufficient for robust studies. To accelerate discoveries, Huang's team applied machine learning to the vast DESI Legacy Imaging Surveys dataset, gathered at the NSF-operated Cerro Tololo Inter-American Observatory (CTIO) and Kitt Peak National Observatory (KPNO).
Leveraging the supercomputing power of Berkeley Lab's National Energy Research Scientific Computing Center (NERSC), they deployed a deep residual neural network trained specifically for lens detection. The results were transformative: the AI pinpointed 1,210 new candidates.
These discoveries will refine measurements of cosmological parameters. By spotting distant supernovae lensed by massive galaxies and analyzing their light curves, researchers can gauge true distances—key to better constraining the Hubble constant and understanding universe expansion.
