Big Computers, Little Galaxies: 1st March 2019

Dr David Williamson, Southampton University 

The Sloan Digital Sky Survey has revealed huge numbers of galaxies, which typically have masses between 108 and 1012 solar masses. They are not evenly distributed: there are clusters and voids. The Local Group is not a cluster, because clusters are regarded as having structure, and typically up to four big galaxies. The Local Group has the just the Milky Way (MW), M31 and M33, plus lots of dwarf galaxies orbiting MW and M31, of which the Magellanic Clouds (MC) are the best known.

The MC are quite large for dwarf galaxies, and naked-eye objects in the southern sky, but are blob-shaped rather than looking like galaxies. The Large MC has a bar, arm-like features and HII regions, and both MCs have tails of neutral hydrogen. Other dwarf galaxies have similar features, and most appear to be interacting with the large galaxies. The gas and stars in galaxies are not independent: hot thin gas collapses into stars, and stellar winds blow gas out of the galaxy, after which it may either fall back or become part of the intergalactic medium. The question is how dwarf galaxies differ from large galaxies. One difference is chemistry.

Astronomers classify the elements into hydrogen, helium and “metals”, i.e. all the remaining elements. All pass through cycles of star formation followed by ejection back into the interstellar medium, and the proportions and distribution of metals vary with galaxy size and type. Massive galaxies have a higher proportion of metals, but the reason is not clear. One possibility is that most dwarf galaxies are in groups, and from time to time collide with the massive ones. The collision of the gas clouds results in an effect called “ram-pressure stripping” in which the metal-rich gas is lost from the dwarf galaxy.

This process can be modelled by solving the Navier-Stokes equations in a supercomputer, but this is an expensive operation. The space is subdivided into cells, the equations are solved in each cell taking into account conditions in neighbouring cells, and this is repeated over a million time steps. With 1 million cells and 10 equations each requiring 5 seconds to solve, a single simulation would take 1.5 million years using a normal computer. Supercomputers are computer clusters: the one used by Southampton is IRIDIS, which has 20,000 processing cores. The problem is split between the cores; the hard part of the process lies in ensuring they all communicate properly.

There are two options for the simulation. One is to simulate a large part of the universe, which allows for galactic interactions and gives statistics, but has limited resolution. The second is to simulate one galaxy in detail, but then interactions and statistics are not available. This is the approach taken by the Southampton group, because it works best for dwarf galaxies and gives good resolution. The simulation looks at the interaction between the Milky Way and the Large Magellanic Cloud, although as a simplification, the effect of the LMC on the Milky Way is ignored.

The simulations have shown that the outflows of gas resulting from the interaction do not contain enough metals to affect the concentration in the LMC, in fact it appears that the gas in the Milky Way is constraining the outflow of metal-rich gas from the LMC. This is an effect called “ram-pressure confinement”. The conclusion is that massive galaxies are not, after all, very good at removing gas from dwarf galaxies, so the interactions do not explain the lack of metals in them.

Another possible explanation is that the interactions remove gas from around the dwarf galaxies, and this slows the rate of star formation so fewer metals are formed. Unfortunately, this scenario is too complex to simulate with current computers. Alternatively, it may be that dwarf galaxies are simply less good at forming stars, for reasons that are unknown at present. Research is continuing.

Notes and summary by Chris Hooker.