Rebound Alternatives

rebound reviews and mentions

• General Relativity and Solar System Stability
  3 projects | news.ycombinator.com | 14 Mar 2023

  Thank you! It's exciting to be here. The Guidelines seemed brief, but informative. Thank you as well for pointing out the FAQ.

  We used REBOUND <https://rebound.readthedocs.io/> to run our N-body simulations and a modified version of REBOUNDx <https://reboundx.readthedocs.io/> to add the first-order post-Newtonian correction from GR and adjust it over time <https://github.com/zyrxvo/reboundx/tree/GR_Sweep>.

  We used the exact same setup and initial conditions for each of the 1280 simulations. Thus, the only variation between them was our choice of how slowly to turn off GR. "Turning off GR" is similar too, but not the same as increasing the speed of light from ~3e8 m/s to infinity. We started with GR at the currently accepted value (equivalent to c=3e8) and then ramped it down linearly so that c=∞ at some time t=τ (again, not exactly the same as changing the speed of light, but related to it).

  We would expect that without any differences, we could run two N-body simulations with REBOUND and reproduce them exactly, bit for bit even though the system is chaotic. Therefore, with even the slightest changes, such as slowly altering the perihelion precession rate of Mercury, chaos would cause any two similar or nearby trajectories to diverge to different results.

  For the most part running the simulations themselves didn't pose any challenges. The most difficult aspect of running them was the walltime, or the real-world time it takes for a simulation to run. With a timestep of ~3 days, it took about 82 days to run one simulation to 12.5 billion years. Based on the ending times of all 1280 simulations, it required about 132 years of walltime to run. This is why we were very grateful for Compute Canada and the Niagara supercomputer, because we were able to run all of these simulations in less than 3-4 months.