The Galaxy Factory

Team: Justin Read (PI); Oscar Agertz; Maxime Delorme

Simulations of galaxy formation over cosmic time have continued to improve over the past two decades due to ever increasing hardware speed and improved software algorithms. In the past two years, we have passed a key threshold where it has become possible to resolve the sites of star formation within galaxies. The key goal of this project is to run extreme resolution “ab-initio” simulations of galaxy formation for a range of different cosmological models, for the first time. This will allow us to understand galaxy formation more deeply; to explain why galaxy formation appears to cease below a stellar mass of ~100,000 times the mass of the Sun; to probe our cosmological model on smaller scales than ever before; and to place new constraints on the nature of the elusive “dark matter” in the Universe. The project has been funded by a prize from the Merac Foundation; a Royal Society equipment grant; and the University of Surrey.


The Galaxy Factory Machine

One of the key challenges in simulating galaxy formation is capturing the enormous dynamic range in both space and time. Small star forming regions just a few hundred light years across give birth to massive stars that through stellar winds, radiation and supernovae explosions return energy, momentum, mass and metals to the surrounding interstellar medium. Such “feedback” is the key to regulating star formation in galactic discs. This in turn is likely responsible for the observed properties of disc galaxies in the Universe today. However, to properly capture the physics of such stellar feedback we must resolve scales of ~10 light years in a simulation volume some ~150 million light years across with a time resolution of ~1000 years over a simulation time of some ~14 billion years. This means a spatial and temporal dynamic range of ~1:10,000,000! Such highly “multi-scale” simulations are extremely challenging to efficiently parallelise beyond ~100 CPU cores. Our solution to this problem was to build a special machine - the Galaxy Factory - designed to run such multiscale simulations on just 20 super-fast cores each over long periods of time. A machine of this type is unusual in typical supercomputer centres as it requires significant fast RAM; state-of-the-art CPUs that are “over-clocked”; and very long uninterrupted run-times. The exact design of the current Surrey Galaxy Factory machine was determined from a test node funded by a Royal Society equipment grant. The current machine, funded by the University of Surrey, comprises 13 such nodes allowing us to run up to 13 galaxies at a time. The nodes are infiniband connected so that “grand challenge” problems can also be run using larger subsets of the machine. Special software tuned to run on this machine will be developed by a new “Astrophysics Scientific Programmer” position within our group, funded by the MERAC Foundation and the University of Surrey. The Galaxy Factory saw “first light” in August 2014 and is already enabling exciting new science.


Published results

So far, we have focussed on understanding the tiniest star forming galaxies in the Universe - isolated 'dwarf irregular' galaxies. In a series of three papers, we have shown that bursty star formation in these dwarfs 'heats up' their dark matter, transforming their internal dark matter distribution. This solves a long-standing cosmological puzzle known as the 'cusp-core' problem. It also opens the door to using these dwarfs to directly probe the nature of dark matter. We have now used data for 18 such nearby systems to place a new constraint on the “temperature” of dark matter, showing that dark matter most be quite “cold” (i.e. non-relativistic). This supports the idea that dark matter is a new particle that lies beyond the standard model of particle physics.

Papers: