Galactic Cannabilism in the Local Universe

Tim Connors

My project, which I will be involved with for about the next 3 years, is the modelling of galaxy evolution in the local universe. In particular, concentrating on the Milky Way + Large and Small Magellanic Cloud system. The two clouds seem to have been bound either for a very long time, or perhaps forever, and are currently executing a polar orbit around our galaxy - going straight over the top of it, to within a few degrees. I have a talk I gave - in PDF format here. The movies that accompanied the talk are too big to be put on the net, except for one of them - the movie of the (7 MB) particles from the SMC in the stream. On linux, you can see the movie by running

 
xanim +o +Zp0 +Zpe +Zr +j150 clouds.fli 

I started off modelling galaxy formation, by semi-analytic means. Most people modelling the formation of galaxies use "N-body" codes. These track individual virtual "particles", which might be a few hundreds or thousands of solar mass objects. These particles interact gravitationally, and possibly by other means as well, and by a suitable choice of initial conditions, you can model how sattelite galaxies collide with the Milky Way, or how you form galaxy clusters, or how significant portions of the universe evolve. These codes are slow. The problem is naively an O(N²) problem (although one or two people have gotten O(N) performance out of theirs - with certains tricks and their accompanying loss of accuracy) - double the number of particles you look at, and quadruple the time you take. Current big simulations at Swinburne take of the order of a month to execute on our ~120 node super-cluster.

I was tackling the problem from a different direction. I had taken code from a couple of other people (Paul Nulsen and Kelvin Wu), which is a "block-model" implementation of a semi-analytic merger tree. I had run quite a few simulations, taking the code as was supplied to me. Instead of waiting for a month, I wait for about 30 minutes, maximum, using only one of the nodes in the cluster (or I could run a bunch of simulations on several nodes at the same time, using different inputs, and they all finish within the half hour).

This block model uses an approximation that two objects of equal mass in space collide, into one object of double the mass. It doesn't worry about colliding two objects of different mass, but this does not affect too many of the results. The method is a monte-carlo method - we run a random simulation hundreds to thousands of times, and average the results. Thus we can only get global, or average properties of galaxies or clusters. Randomness is created by not allowing the masses to change, but allowing the densities to change. Whether an object "collapses" after colliding is determined by its density. If it collapses, then stars form, abundances (or metalicities) change, gas properties change, temperatures and X-ray properties change. Some of these can be compared to observational data, and this way we can verify aspects of the model. Our model currently has the best treatment of gas physics out there - Paul and Kelvin are great X-ray people - but is lacking in other areas. The block model was be upgraded, chemical evolution was be more sophisticated, star formation was to be more sophisticated.....

Star formation and anything that evolves with time, such as metalicities, are really hard to get a handle on, observationally, and this is very current research. So no matter what model I was to come up with after 3 years, no-one will be able to prove me wrong ;) Except that I encountered rather large difficulties that are peculiar to semi-analytic codes. I changed areas slightly, and now use an N-body code, as I talked about above.

As my thesis is very much evolving, this page will be updated. Plans for the near future will to put up some movies I have been making, and well as some perhaps perty pictures. We'll see....

- Tim Connors