Milky Way
A sub-parsec resolution simulation of the Milky Way
Check out the full version, with more info, guided tour etc...
About the simulation:
Details in
- Adaptive Mesh Refinement (AMR) with the RAMSES code (Teyssier 2002)
- 21 levels of adaptive refinement on a 100x100x100 kpc3 box
- Resolution: 0.05 pc
- 6x107 particles (initially)
- Dark matter halo + stars (spheroid + bulge + thick disk + thin disk) + gas (disk) + super massive black hole
- Based on the Besancon model (Robin et al. 2003)
- Initial conditions generated by pyMGE (Emsellem et al. 1994, Emsellem & Renaud, in prep)
- Star formation + stellar feedback: HII regions, radiative pressure, supernovae
The simulation has been running for 12 million CPU hours on 6080 cores of the Curie supercomputer hosted at the TGCC (France).
The simulation in numbers
- 240 million gas cells (on average)
- 60 million particles (initially)
- 13 million files generated
- 40 Tb generated
- 12 million CPU hours consumed
- 2000 yr: equivalent time on one CPU
- 6080 CPUs used
- 830 Myr: time simulated
- 0.05 pc: smallest cell
- 14 K: coolest cell
- 18 yr: shortest timestep
- 1264 swear words pronounced
Yes, you are probably older than the shortest timestep in the simulation!
Adaptive Mesh Refinment
In the simulation, the gas is rendered on a grid. But there is no need to have high resolution (= small cells) where there are no structures, like outside the galactic disk, or between the spiral arms. However, maximum resolution is required in the densest structures, like the star forming clouds. The Adaptive Mesh Refinement (AMR) technique does the job of refining the grid only when and where it is needed. One level of refinement corresponds to cells smaller by a factor 2. Therefore, at level 6, cells are 26=64 times smaller than the cell at level 0 (which is the entire simulated volume).
Without AMR, the entire simulation would require 9 223 372 036 854 775 808 cells.
Using this simulation
All our data is publicly available, but no interface is provided. The amount of data, the complex structure of the AMR and the fact that I have something else to do that coding an interface make the distribution of data quite inefficient. However, I am more than happy to share the simulation with you. I can extract the data that you need upon request. Don't hesitate to ask!