ADS: Hezaveh et al. 2013b
arXiv: Hezaveh et al. 2013b
This paper, another part of the trio we published at the same time, describes the initial results of our ALMA Cycle 0 program to take one-minute snapshot images of a large number of SPT galaxies. While we were pretty sure the galaxies were lensed, we didn’t have much high-resolution imaging that proved it conclusively. In this paper, we presented images of four galaxies which were clearly lensed even in the low-resolution data we had at the time, and used those data to determine what the galaxies would look like if they weren’t lensed.
As part of this paper, we presented a code that models the effects of gravitational lensing, and it’s the best in the business. It models each system directly in the Fourier plane, since every pixel in the images is correlated with every other pixel. It also accounts for various calibration errors present in the data – for example, not knowing the position of each ALMA antenna perfectly well. The code uses a Markov Chain Monte Carlo technique to fully sample the complex parameter space that results from varying the properties of both the lens and the background source.
The models indicate that the four sources in this paper are magnified by factors of 5 to 20. This lets us observe them at higher effective resolutions much more cheaply than would otherwise be possible. The unique modeling procedure also lets you discern features not visible in the images by eye – we later observed these galaxies at higher-resolution than what was presented here, and those images agreed perfectly with what was predicted from the lower-resolution models.
My favorite example of this is shown below. If you look at the top of the page, it looks like SPT0346-52 in the top left is doubly imaged, but the best model based on those data predict that the southern image should split up into multiple images. After we made that model, we got higher resolution data which confirmed the previous prediction almost perfectly. This goes to show that doing the lens modeling directly with the visibilities instead of on the images is vital to getting the right model.