Access: A first glimpse of first stars

Adaptive mesh refinement (AMR) is the key to the incredible resolution of the simulation. Cosmology researchers, however, only recently began using the technique.

"AMR has been around since the 1980s, but this is one of the first times it has been used in cosmology," says Greg Bryan, a Hubble fellow at the Massachusetts Institute of Technology who developed the three-dimensional AMR cosmology code while Norman's Ph.D. student.

In the simulation, a cubic grid with 64 cells on each side is laid over the 18,000 light year cube that represents the hypothetical section of the universe. The AMR code both runs the simulation and monitors its progress. With each time step, the data changes - gravity draws the gases into more and more dense clouds. Then, the code checks the cells to see if gravity has pooled four times as much gas into a given cell since the previous phase of the simulation. If the cell's density has not quadrupled, the code advances to the next time step.


	Simulation of a forming protostellar cloud. The images are a visualization of gases pooling into the clouds that formed the first stars. Each subsequent image shown is a power of 10 magnification of the previous, meaning that final image in the series is magnified to a power of one million compared to the first. The density of the clouds are illustrated by color. The least dense areas are violet; the most dense areas are red and orange.

If a cell's density has quadrupled, the code flags that cell. It then builds new regions from adjacent flagged cells. The code can keep track of any number of regions, and they can all be processed simultaneously.

The code draws a new three-dimensional grid on top of each of these new regions. The information necessary to continue the simulation is taken from the previous grid. Then the code zooms in on the action. After each time step, the process repeats - the data changes, and the data is checked. The grids are redrawn as necessary, constantly focusing in on the forming gas clouds with better and better resolution.

"When you start out at such a vast volume, the adaptive mesh refinement is very important," Bryan says. "Without it, the focus of your data almost immediately collapses into a tiny portion of your total simulation, and there's nothing you can do to improve the resolution."

For the work with the first star formations, the grids were redrawn and the attention of the code was refocused 19 times to achieve such high resolution. "This is something that no one has ever done before. Something totally unprecedented," Norman says.