Solar Collisons at the limit of HPC

  • Visualisation of results, not during.
  • Spatial and temporal discretisation of complex equations -> errors
  • Chemistry of dust: hard to do (q.v. Sheffield)
  • 300,000 Blue Waters (NCSA)
  • Not all simulations scale (gravity)
  • Simulations on different scales.
  • Simulations of binary stars and their ultimate end.
  • Nebula formation. Takes a long time (6 months wall clock) due to poor scalability (200 cores) due to gravity even with common techniques such as approximating distant groups to blocks to avoid the number of point-to-point calculations required. Huge numbers of point-to-point communications means poor MPI performance.
  • C++ used for new codes. New codes scale better. Computer Scientists (RSEs) needed to make things better.


  • Can trinary stars be simulated? Yes but very complex. Easier with multiple collisions of planets.
  • Not all binary pairs ultimately destroy each other, e.g. a large giant star and a white dwarf. Can even survive all the way to being sucked into the giant’s core.
  • Collisions can be relatively fast: approach on centuries, about a year for the final part of a collision, which takes about a year to simulate.
  • Don’t really understand what causes supernova. It might be due to collisions.