So, how do you get those images from telescopes anyway?
The other day I came across this post on X
I’m no astronomer and I wondered why it took this person two hours to generate an image. I mean, doesn’t the telescope do that already? Aren’t modern telescopes able to capture images of things far, far away?
Turns out, not! I asked Claude (not Argons) for help. What’s captured in the Event Horizon Telescope (mentioned in the post) are measurements of 2D positions in Fourier space. Not a picture. For that, you’d need an optical telescope which isn’t powerful enough to image something like a black hole.
This is what the raw data looks like for M87*, the supermassive black hole that I decided to try reconstructing.
So it would just be a question of performing an inverse Fourier transform, right? Maybe. But based on the data measured you’d get a dirty image. Like this. No good. We don’t have enough data.
Instead, what Claude did was a complex fitting process. Over 700 rounds, it iteratively guessed an image, computed the expected (u,v) pairs, compared them against the real measurements, and updated its guess. And finally arrived at this.
Fascinating stuff. It took Claude under 30s to do this and 10 minutes for installing the tooling and generating artifacts. Took me a good chunk of my Sunday to understand all of this.
#astronomy#signal-processing#imaging#ai#claude