Machine Learning Finds More Gravitational Lenses Than All Astronomers Combined

One of Einstein’s most revolutionary predictions is that mass bends light.

Starlight bent around the eclipsed Sun in 1919 confirmed this.

The 1930s first developed a prediction for gravitational lenses.

Large foreground masses would bend and magnify fortuitously aligned background sources.

Multiple images or even “Einstein rings” could occur.

MORE FOR YOU

For decades, they were solely theoretical.

Finally, in 1979, the “Twin QSO” was found: two lensed images of the same quasar.

Since that time, many more gravitational lenses have been found.

Features include:

  • quadruple images,
  • magnified arcs,
  • hidden background objects,
  • and nearly-perfect rings.

Hubble’s deep imaging uncovered many more strong lenses.

Lensing affects only 1 in ~10,000 galaxies.

Hubble, unfortunately, only offers narrow-field capabilities.

After 30 years, it’s imaged less than 1% of the sky.

However, DESI (Dark Energy Spectroscopic Instrument) Legacy Imaging Surveys are both deep and wide.

Spanning ~20,000 square degrees, its full map requires over 10 trillion pixels.

Machine Learning is required to handle that much data.

That process discovered 1,210 new gravitational lenses.

That’s more than previously discovered by all astronomers, combined.

Occasionally, Hubble followed up, revealing additional details.

With Euclid, Vera Rubin, and Nancy Roman telescopes coming soon, we’ll certainly find even more.


Mostly Mute Monday tells an astronomical story in images, visuals, and no more than 200 words. Talk less; smile more.