Matching X-ray and optical stars without using their positions

Two telescopes spot the same point of light in the sky. One sees X-rays, the other sees visible light. Are they looking at the same object, or two different things that happen to sit close together? The usual answer is brute geometry: if the two dots are close enough, call them a match. But the sky is crowded, and "close enough" often catches the wrong neighbor.

This team tried a different trick. Instead of asking where a source is, they asked what it looks like — its brightness, its color, its distance. They trained a pattern-spotting program (LightGBM, a tool that stacks many simple yes/no rules into one strong guess) on a set of matches everyone already trusts. The program learned which combinations of properties signal a genuine pairing versus a coincidence.

The result: of about 254,000 X-ray sources, it confidently paired roughly 113,000 with an optical twin. For about 7,000 it flagged more than one believable candidate. And it rejected about 20,000 matches that the old distance-only method had accepted — calling half of those mere accidents of crowding.

The proof it works: on a dense, well-studied patch around the Orion nebula, it reproduced 95% of the trusted matches using zero position data. That also marks the limit — it leans on those trusted matches as its teacher, so it inherits whatever blind spots they carry.

Why you should care: This released catalog hands astronomers 113,000 cleaner star pairings, with the coincidental fakes already filtered out — a sturdier foundation for studying objects that glow in both X-ray and visible light.