Quantifying resolution in cosmological N-body simulations using self-similarity

Michael Joyce, Lehman H. Garrison, Daniel J. EisensteinMonthly Notices of the Royal Astronomical Society 2020 · DOI

TL;DR

This paper introduces a novel method using self-similarity in scale-free cosmological simulations to precisely quantify the resolution limits of N-body simulations. By analyzing deviations from self-similarity in the two-point correlation function, they can determine the range of scales and times for which simulation results are physically converged, achieving accuracies below 1%.

What Problem Does This Paper Address?

Cosmological N-body simulations are crucial for predicting the evolution of large-scale structure in the Universe. However, these simulations inherently have limitations in their spatial and mass resolution due to the finite number of particles and the use of force softening. Quantifying precisely how converged a simulation is to the true physical result, especially at small scales and high redshifts where non-linear gravitational clustering is significant, has been a persistent challenge. Traditional "brute force" studies, while informative, are computationally expensive and can be misleading if numerical convergence is mistaken for physical convergence. This paper aims to provide a more reliable and quantitative method for assessing simulation resolution.

Methods

The core of the paper's methodology lies in leveraging the property of self-similarity in scale-free cosmological models. These models, characterized by a power-law initial power spectrum and an Einstein-de-Sitter expansion, exhibit a unique property: their statistical properties should evolve in a self-similar manner, meaning temporal evolution is equivalent to a simple rescaling of spatial coordinates.

The authors use the ABACUS N-body code to run these simulations. They analyze the two-point correlation function (2PCF), denoted by $\xi(r, a)$, where $r$ is the separation and $a$ is the scale factor. The key idea is that in a perfectly self-similar simulation, the 2PCF, when plotted against a rescaled distance $r/R_{NL}$ (where $R_{NL}$ is the non-linearity scale), should be independent of time.

Deviations from this self-similar behavior in the plots of $\xi(r/R_{NL}, a)$ reveal the scales and times at which the simulation is not converged to the physical continuum result. By carefully analyzing these deviations and their fractional variations ($\Delta\xi/\xi$) between simulation snapshots, they can:

  1. Quantify numerical convergence: Assess the accuracy of the simulation's integration with respect to time-stepping parameters.
  2. Determine physical resolution: Identify the range of comoving scales that are reliably resolved at different times, based on the degree of self-similarity.
  3. Infer precision: Estimate the accuracy of the resolved 2PCF, defined as the maximum difference between the measured 2PCF in resolved regions and the converged value.

Datasets

The paper does not introduce new observational datasets. Instead, it utilizes newly generated N-body simulation data from the ABACUS code. These simulations are specifically designed to be scale-free with an initial power spectrum slope $n=-2$, chosen to be representative of the non-linear regime of LCDM models. The simulations are run with specific parameters for particle number ($N=1024^3$), force softening ($\epsilon/\Lambda = 1/30$), and initial conditions.

Key Insights and Ideas

  • Self-similarity as a resolution probe: The most significant insight is the demonstration that self-similarity in scale-free models provides a powerful, quantitative tool to assess simulation resolution. Deviations from self-similarity are direct indicators of where the simulation is failing to capture the physical continuum.
  • Spatial resolution is mass-limited: For small force softening lengths, the spatial resolution of N-body simulations is primarily determined by the mass resolution (related to the particle number density). This challenges the intuition that force softening alone dictates spatial resolution.
  • Two-body collisionality sets late-time resolution: The paper provides strong evidence that the evolution of the lower cut-off scale for resolution at late times is governed by two-body collisionality within halos, leading to a characteristic scaling of $a^{-1/2}$ for the resolved comoving scale.
  • "Apparent convergence" vs. "physical convergence": A crucial corollary is that simulations can appear numerically converged (with respect to time-stepping and force softening) over a significant range of scales, yet still not be physically converged to the true continuum result. This method helps distinguish between these two.

Key Results

  • The method can quantify the resolution of the 2PCF to below 1% accuracy in well-converged scales.
  • For small force softening, the spatial resolution is found to be essentially determined by the mass resolution.
  • The lower cut-off scale for convergence evolves in comoving units as $a^{-1/2}$ at later times, consistent with two-body collisionality.
  • The paper provides detailed plots (Figures 5 and 6) showing the resolved comoving scales as a function of time for different precision levels (e.g., 1% and 5%), offering a direct way to assess simulation fidelity.
  • The authors demonstrate that their method can be extended to infer resolution limits for non-scale-free cosmologies like LCDM.

Limitations and Future Work

  • Model Dependence: The current study focuses on a single scale-free model ($n=-2$). While the method is general, the specific quantitative results for resolution limits might depend on the initial power spectrum. Future work will explore scale-free models with different exponents and apply the method to LCDM simulations.
  • Force Softening: The paper uses a small force softening ($\epsilon/\Lambda = 1/30$). The authors acknowledge that a larger $\epsilon/\Lambda$ would set an absolute lower limit on resolution earlier in time. Further studies will investigate the impact of varying force softening.
  • Other Statistics: The current analysis focuses on the 2PCF. The authors state that the method can be applied to other clustering statistics (e.g., halo mass function, power spectrum) and plan to do so in future work.
  • Other Processes: While two-body collisionality is identified as a key driver of late-time resolution evolution, the authors acknowledge that other processes like unphysical merging could also play a role and warrant further investigation.
  • Extrapolation to LCDM: The application of the scale-free results to LCDM involves approximations, particularly regarding the non-power-law nature of the LCDM power spectrum and its deviation from the Einstein-de-Sitter expansion. These will be explored further.

Summary updated

Abstract

ABSTRACT We demonstrate that testing for self-similarity in scale-free simulations provides an excellent tool to quantify the resolution at small scales of cosmological N-body simulations. Analysing two-point correlation functions measured in simulations using abacus , we show how observed deviations from self-similarity reveal the range of time and distance scales in which convergence is obtained. While the well-converged scales show accuracy below 1 per cent, our results show that, with a small force softening length, the spatial resolution is essentially determined by the mass resolution. At later times, the lower cut-off scale on convergence evolves in comoving units as a−1/2 (a being the scale factor), consistent with a hypothesis that it is set by two-body collisionality. A corollary of our results is that N-body simulations, particularly at high red-shift, contain a significant spatial range in which clustering appears converged with respect to the time-stepping and force softening but has not actually converged to the physical continuum result. The method developed can be applied to determine the resolution of any clustering statistic and extended to infer resolution limits for non-scale-free simulations.
BibTeX
@inproceedings{joyce2020quantifying,
  author = {Michael Joyce and Lehman H. Garrison and Daniel J. Eisenstein},
  title = {{Quantifying resolution in cosmological <i>N-</i>body simulations using self-similarity}},
  booktitle = {Monthly Notices of the Royal Astronomical Society},
  year = {2020},
  doi = {10.1093/mnras/staa3434},
}
Figure