Improving Performance of Zoom-In Cosmological Simulations using Initial Conditions with Customized Grids

TL;DR

This paper introduces a method to customize initial conditions for zoom-in cosmological simulations by creating tailored root grids, enabling efficient simulations with maintained accuracy and nearly doubled speed.

Contribution

The authors develop a novel approach to generate customized root grids for zoom-in simulations, improving computational efficiency while preserving key physical properties.

Findings

Dark matter halo masses match within 15%

Major merger timings and masses are well reproduced

Achieved nearly a twofold speedup in simulations

Abstract

We present a method for customizing the root grid of zoom-in initial conditions used for simulations of galaxy formation. Starting from the white noise used to seed the structures of an existing initial condition, we cut out a smaller region of interest and use this trimmed white noise cube to create a new root grid. This new root grid contains similar structures as the original, but allows for a smaller box volume and different grid resolution that can be tuned to best suit a given simulation code. To minimally disturb the zoom region, the dark matter particles and gas cells from the original zoom region are placed within the new root grid, with no modification other than a bulk velocity offset to match the systemic velocity of the corresponding region in the new root grid. We validate this method using a zoom-in initial condition containing a Local Group analog. We run collisionless simulations using the original and modified initial conditions, finding good agreement. The dark matter halo masses of the two most massive galaxies at $z=0$ match the original to within 15%. The times and masses of major mergers are reproduced well, as are the full dark matter accretion histories. While we do not reproduce specific satellite galaxies found in the original simulation, we obtain qualitative agreement in the distributions of the maximum circular velocity and the distance from the central galaxy. We also examine the runtime speedup provided by this method for full hydrodynamic simulations with the ART code. We find that reducing the root grid cell size improves performance, but the increased particle and cell numbers can negate some of the gain. We test several realizations, with our best runs achieving a speedup of nearly a factor of two.