Collaborative Research: CDS&E: Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) to maximize the science return of next generation cosmological experiments

The Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) project leverages recent major advances in computational galaxy formation to produce the largest suite of cosmological simulations with full baryonic physics designed to train machine learning algorithms for a broad range of applications, including thousands of cosmological and feedback parameter variations. This project will use this unique dataset to study how to maximize the science return from next generation cosmological surveys. Although the surveys will constrain the value of the cosmological parameters with unprecedented accuracy, achieving this goal requires overcoming two major obstacles: (1) the optimal summary statistic is unknown, and (2) a lot of the information is on scales significantly affected by baryonic processes that are still poorly understood. CAMELS will (1) develop neural networks to help extract the most cosmological information, and (2) perform thousands of simulations over a wide range of parameters to quantify uncertainties in baryonic effects. All CAMELS data products will be publicly available, to enable research and engagement by the broader community. The team will work to increase the participation and success of women and underrepresented minorities by providing dedicated mentoring and early access to research, through three programs for undergraduate students: (1) a summer research program co-organized by the National Society of Black Physicists and the Simons Observatory; (2) the AstroCom NYC program, joining other mentors from the City University of New York, the American Museum of Natural History, and the Flatiron Institute; and (3) the new Colors of Astrophysics program at the University of Connecticut.

Upcoming experiments such as DES, DESI, LSST, WFIRST, SKA, and Euclid will improve our understanding of fundamental physics and the origin and fate of the Universe. CAMELS will help to determine the optimal summary statistic to apply to the non-Gaussian density fields observed in most cosmological surveys, and to quantify uncertainties in subgrid models for key astrophysical processes such as feedback from stars and massive black holes, which limit the use of hydrodynamic simulations. The neural networks and thousands of simulations that will be used by CAMELS will produce a distinct qualitative improvement over previous work.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.