Galaxy Simulations

Machine learning in cosmological models

+ See all authors and affiliations

Science  01 Apr 2016:
Vol. 352, Issue 6281, pp. 49
DOI: 10.1126/science.352.6281.49-a

Machine learning can help model the formation and evolution of galaxies

PHOTO: NASA/ESA/AND THE HUBBLE SM4 ERO TEAM

A cosmological simulation containing only dark matter is relatively easy to run, but adding gas, stars, and galaxies to the model requires sophisticated hydrodynamics and subgrid physics. These are very computationally expensive, but Kamdar et al. have used a machine learning algorithm to massively speed up the process. By training the algorithm on part of a large hydrodynamic simulation, they are able to reproduce many galactic properties in the rest of the simulation just from the dark matter information, and in much less time. The technique could be used to quickly scale up detailed hydrodynamic simulations to larger dark matter–only ones, aiding in the interpretation of observational surveys.

Mon. Not. R. Astron. Soc. 457, 1162 (2016).

Navigate This Article