Astronomy Colloquium: 2017 Neugebauer Lecture

The discoveries of all extra-solar planets, our percent-level knowledge of the cosmological parameters, and the measurements of proper motions for millions of stars all depend on exceedingly good calibration of astronomical instruments. Gerry Neugebauer cared deeply about accurate calibration, and he was the US lead of IRAS, which was one of the first large sky surveys from space. These large sky surveys have transformed how we think about making accurate measurements; the most precise calibration of telescopes no longer comes from calibration programs or standard stars; it comes from the science data themselves. Indeed, GXN might have been upset to learn that we now often advise new projects not to take any calibration data whatsoever! Self-calibration capitalizes on causal structure in our generative model of the data set, and therefore there are important design choices to be made for an astronomical project that maximize the evidence for this causal structure in the data. The CMB and astrometry communities have understood this for decades, and it is now becoming true in a much wider range of areas. I will give unsolicited advice for the designers of any new instruments, telescopes, spacecraft, or surveys. I will comment a bit on the differences between accuracy and precision, on linear algebra (!), and on what self-calibration can and cannot do. I will bring examples from SDSS, Kepler, HST, Gaia, radio interferometry, coronography, stellar spectroscopy, and CMB polarimetry.