NCSA Researchers Explore Stars In The Making

By Herbert Morgan, NCSA

Paul Ricker, NCSA research scientist and assistant professor of astronomy at UIUC, Ramesh Balakrishnan, research programmer in Performance Engineering and Computational Methods (PECM), and Zarija Lukic, UIUC astronomy graduate student, are developing a physics code module for FLASH that incorporates the effects of star formation and supernova feedback in cosmological simulations.

Stars are grouped together into galaxies; some fraction of galaxies is grouped together into galaxy clusters. Consider the range of length scales that are important in these different situations: for stars, a range of length scale is much less than a light year; for galaxies, perhaps a hundred thousand light years across; for clusters of galaxies, maybe a million light years across.

Suppose you want simulations that include all of these scales. According to Ricker, it is not possible to do a simulation that simultaneously captures, from first principles, stars and star-forming clouds and galaxies and clusters of galaxies and large-scale structure.

He further explains that the problem is usually approached by trying to solve the equations that describe how matter behaves from first principles on the largest scales, but also by introducing models for the small-scale behavior of star-forming clouds and stars too small to capture on those large scales.

So researchers develop models to describe how stars of different masses form, how rapidly they form in different environments, and, when they die, how much energy, matter, and enriched material they re-inject into the interstellar medium. Researchers incorporate these models into the cosmological simulations and try to understand the large-scale effects of star formation processes.

Star formation history is also of interest in understanding how the universe emerged from the Dark Ages—the period of time following the decoupling of matter from the cosmic background radiation but before new sources of light, such as stars and quasars, had yet emerged. Being able to pick where stars should form and then to model their effects appropriately on the intergalactic medium is important to study the physics of that era.

The FLASH code was developed at the University of Chicago for creating simulations of Type-1A supernovae. It only has gas dynamics, expansion of the universe, collision-less dynamics of dark matter particles, gravity, and other pieces of physics. But it doesn't include a model for the subgrid and the small-scale star formation processes. So, last year Ricker and Balakrishnan submitted a Strategic Applications Program (SAP) proposal to develop a model and an extension to the FLASH framework that would allow models of star formation to be included in the code and let them begin testing those against observations of real galaxies.

"We have reached the point," says Ricker, "where we now have a simple model for star formation in the code that is based on observed statistics of star formation … so we have an estimate of where stars will form and we have some information about the statistical properties of stars where they form."

In the next six months they plan to elaborate on this model, to tune its parameters using observations, and to achieve full-production status in order to do simulations.

Ricker has been running his code on NCSA's Mercury and Tungsten. The code uses HDF 5 for I/O, which can be used in two different ways: using the parallel I/O capabilities of HDF 5 and using serial I/O to send all the processors and information to one processor, which writes to disk.

Ricker and Balakrishnan did encounter some subtle problems having to do with memory alignment. They attributed those to structures that included both integers and reals. Ricker suggests that one should test on multiple types of platforms. "For NCSA users, who have a variety of different machines that they could run on, it would be well-advised to ask for at least some amount of time on more than one architecture."

Ricker plans to submit a new SAP proposal whose result, he hopes, will culminate in cosmological simulations that include star formation. "It has been extremely helpful," says Ricker, "to have Ramesh working with me on this project. [SAP] is a very good program that more groups ought to take advantage of."