Research Interests
Supernova Feedback
Early simulations of galaxy formation suffered from extreme over-collapse.
Governato et al (2004) explains how part of the problem is lack of resolution.
The other problem is that no one was using effective feedback mechanisms in
their large scale, cosmological simulations. Thus, when stars formed, they
could keep on forming because the energy they release in stellar winds and
supernovae was immediately lost because of details of how smoothed particle
hydrodynamics (SPH) works. When I arrived at grad school,
a n-body SPH solver, Gasoline, had been written, but it did not include any
effective feedback effect from supernovae.
James Wadsley related to us how well adiabatic feedback worked for Thacker and
Couchmann (2000, 2001), so we put it into Gasoline. The initial recipe they
used and the original recipe from Gerritsen (1997) were great, but they were
slightly inflexible. I decided as a first shot to use a parameter to decide
how many gas particles should have their cooling shut off. The details are in
Stinson et al. (2006). That was fine, but why parameterize when
McKee & Ostriker (1977) gives an explicit formula for the effect of supernovae
blastwaves on the ISM? So, I put that physics in and eliminated a parameter.
How Well It Works
 |
I am currently running simpler virialized halo model galaxies of different
masses. So far, the results are encouraging and they should come out in a paper
soon. The most interesting galaxy was the dwarf at left. The link plays a
movie of how supernovae act as a significant feedback on star formation. Stars
form, supernovae explode and stars cannot form any more until the gas cools
back down and collapses.
|
- I have also begun running a billion particle cosmological simulation.
New Development
- In addition to all the IDL required to do the analysis of these simulations,
I've worked on a new more powerful analysis tool, NChilada
that is parallel. It uses a new output format that takes advantage of
directories. It puts different attributes into their own files and files them
in directories according to time step and particle type. The viewer,
Salsa has a java interface that allows you to view and
manipulate the large datasets in real time (we wouldn't be able to view the
billion particles any other way). Salsa also hooks into python so that you
can manipulate all your data in python.
- I made sure that Gasoline writes output in the new format. Breaking data
up in this way makes it possible to analyze the data more easily on desktop
machines as you only have to read in one or two attribute files to get the
data that you desire.
- Another exciting part of the whole NChilada is a brand new n-body gravity
solver, Changa. It has been written by our collaborators in the CS department
at UIUC. Using their parallel framework,
Charm, it scales great on a commodity
cluster. Thanks to the efforts of Fillippo Gioachin, Changa
allows for domain decomposition in several different ways, uses a very
efficient cache management scheme for communcation, and implements several
varieties of tree. Once we implement SPH into the code, I expect to use it
for my runs that include gas. For now, Gasoline is a fine option.
- I wrote documentation (password required, hopefully open sourced soon?) for
pkdgrav and gasoline.
- The bread and butter of the simulation game is particle tracing. How did
particles get to where they are today? (Or, in the real world, how did the
elements that make up the earth get where they are today?) Quite often, it
is interesting to figure out which halos got swallowed by which and which
particles got blown out of which halos. You might call them "merger trees".
I have written IDL and C functions to determine which end halos came from which
halos in an earlier era.
|