CE

Computing

Anvil

  • Baowei has gotten the code working on Anvil (Purdue University)
    • No queue (as of now)
    • Maximum 16 nodes/2048 cores, 96 hours
    • More than twice as efficient as Frontera with 64 nodes/3584 cores, in terms of node-hours per frame
    • Implies that the code does much better with more cores/node and less nodes, for the same number of cores.
    • Anvil has AMD nodes, so nice that our code runs with AMD nodes (before we always used Intel)
    • Allocation ends June 30, so need to ask for extension.

Computer time

  • Extension for Anvil
  • Renew XSEDE (now under a different scheme)?
  • Renew Frontera?

New Analysis

Energy terms (star gas only, excludes ambient), evolution with time

https://www.pas.rochester.edu/~lchamandy/EoS/Figures/Env_E_terms_277_282_spancols_individ_totals_upto115days.png

  • The black and grey lines show that the envelope gas is being energized by the ambient after 40-50 days — problematic
    • Is the ambient energizing bound or unbound gas? If unbound, not a big problem. If bound, then implies that our estimates for the unbound mass are higher than they should be.
      • Ans: mainly unbound gas, but some bound gas too. The proof is that total energy of particles+bound gas rises, even though it must be transferring energy to unbound gas so should be losing energy rather than gaining:

https://www.pas.rochester.edu/~lchamandy/EoS/Figures/Env_E_terms_bnd_277_282_spancols_individ_totals.png Energy terms, bound envelope gas
https://www.pas.rochester.edu/~lchamandy/EoS/Figures/EnvMass_282_283_277.png Unbound mass with factors of two in particle-gas PE
https://www.pas.rochester.edu/~lchamandy/EoS/Figures/EnvMass_282_283_277_half.png Unbound mass without factors of two in particle-gas PE

So what to do?

  1. Admit the limitation and write up the paper as "part I". Focus on the bound gas.
  2. Redo the simulations with smaller ambient density and pressure
    • Requires some experimentation to get it working
    • Assuming we can get it to work, requires a few months of running and analysis, but computer time is available and pipeline is ready

Comments

No comments.