Exploring the universe at larger distances requires the use of sometimes very complex tracers of the mass. A particularly interesting tracer is provided by a very tight sequence of absorption features in the light emitted by distant by quasars, which are very bright distant astronomical objects. These features, called Lyman-Alpha forest, has been related to hydrogen clouds present on each quasar lines of sights at various cosmological distances. Some crude model is generally used to model these troughs in observations. At the same time the large cosmological simulations including the physics of the gas at small scales are reaching a good degree of maturity. This comes however at a CPU cost, which makes cosmological inference with automated, Bayesian, machine learning impossible. At the same time the simulations have shown that the simplistic model is largely insufficient to interpret the Lyman-α forests that we have already observed.
The aim of the present PhD subject is to develop new algorithm based on deep learning able to predict the Lyman-Alpha forest from cheap N-body simulations. This will be built upon the expertise acquired from the Lymas project by Sébastien Peirani, deep learning models developped at the Institut d’Astrophysique de Paris by Guilhem Lavaux, and modern hydrodynamical simulations available at the same institute. The model will then be incorporated into a larger inference framework, which can automatically explore cosmological datasets to make inferences and predict cosmic density fields in unobserved regions.
PhD student: Nai Boonkongkird
PhD supervisor: Guilhem Lavaux
Research laboratory: IAP - Institut d’Astrophysique de Paris