Invert Everything Everywhere All at Once
What if we could give any combination of observations of a particular area or event on the Sun along with equations describing the physics to a universal code that creates a physically self-consistent model that explains all the observations? What if we do not need to tell the code how to solve the physics equations? What if the resulting model lives in a space-time continuum without any grid? What if the model could even predict how the Sun evolves after the observations have stopped? What if this inversion science fiction could become reality? In this presentation, I will explain how this could be done using physics-informed neural networks and illustrate it with an extreme inversion problem: deriving velocities, densities, and temperatures as a function of three spatial dimensions and time using only a sequence of continuum images.
Christoph Keller’s research expertise spans exoplanets, solar and stellar magnetic fields, planetary atmospheres, and innovative optical instruments for astronomy, remote sensing, and biomedical imaging. He authored or co-authored over 450 publications and received the Friedrich Wilhelm Bessel award for outstanding astrophysics research. Christoph’s career includes a decade at the National Solar Observatory in Tucson, where he led the iconic McMath-Pierce Solar Telescope. He then joined Utrecht University, becoming its Astronomical Institute director. He served on oversight and visiting committees and chaired the board of the Isaac Newton Group of telescopes on the Canary Islands. After 10 years at Leiden University, he joined Lowell Observatory as Director of Science in 2022 and returned to NSO as Director in May 2024, focusing on fostering a conducive environment for scientific research.