English

MSTAR -- a fast parallelised algorithmically regularised integrator with minimum spanning tree coordinates

Instrumentation and Methods for Astrophysics 2020-01-30 v1 Astrophysics of Galaxies

Abstract

We present the novel algorithmically regularised integration method MSTAR for high accuracy (ΔE/E1014|\Delta E/E| \gtrsim 10^{-14}) integrations of N-body systems using minimum spanning tree coordinates. The two-fold parallelisation of the O(Npart2)\mathcal{O}(N_\mathrm{part}^2) force loops and the substep divisions of the extrapolation method allows for a parallel scaling up to NCPU=0.2×NpartN_\mathrm{CPU} = 0.2 \times N_\mathrm{part}. The efficient parallel scaling of MSTAR makes the accurate integration of much larger particle numbers possible compared to the traditional algorithmic regularisation chain (AR-CHAIN) methods, e.g. Npart=5000N_\mathrm{part} = 5000 particles on 400400 CPUs for 11 Gyr in a few weeks of wall-clock time. We present applications of MSTAR on few particle systems, studying the Kozai mechanism and N-body systems like star clusters with up to Npart=104N_\mathrm{part} =10^4 particles. Combined with a tree or a fast multipole based integrator the high performance of MSTAR removes a major computational bottleneck in simulations with regularised subsystems. It will enable the next generation galactic-scale simulations with up to 10910^9 stellar particles (e.g. m=100Mm_\star = 100 M_\odot for a M=1011MM_\star = 10^{11} M_\odot galaxy) including accurate collisional dynamics in the vicinity of nuclear supermassive black holes.

Keywords

Cite

@article{arxiv.2001.03180,
  title  = {MSTAR -- a fast parallelised algorithmically regularised integrator with minimum spanning tree coordinates},
  author = {Antti Rantala and Pauli Pihajoki and Matias Mannerkoski and Peter H. Johansson and Thorsten Naab},
  journal= {arXiv preprint arXiv:2001.03180},
  year   = {2020}
}

Comments

20 pages, 16 figures, accepted for publication in MNRAS