Statistical Tools for Classifying Galaxy Group Dynamics

TL;DR

This study evaluates statistical tests to classify galaxy groups' dynamical states, finding the Anderson-Darling test most effective for identifying non-relaxed groups in observational data.

Contribution

It compares the effectiveness of goodness-of-fit tests and demonstrates the superiority of the Anderson-Darling test for galaxy group classification.

Findings

Anderson-Darling test outperforms others in detecting non-Gaussian velocity distributions.

Approximately 32% of analyzed galaxy groups are non-Gaussian and likely non-relaxed.

Velocity dispersion profiles differ significantly between Gaussian and non-Gaussian groups.

Abstract

The dynamical state of galaxy groups at intermediate redshifts can provide information about the growth of structure in the universe. We examine three goodness-of-fit tests, the Anderson--Darling (A-D), Kolmogorov and chi-squared tests, in order to determine which statistical tool is best able to distinguish between groups that are relaxed and those that are dynamically complex. We perform Monte Carlo simulations of these three tests and show that the chi-squared test is profoundly unreliable for groups with fewer than 30 members. Power studies of the Kolmogorov and A-D tests are conducted to test their robustness for various sample sizes. We then apply these tests to a sample of the second Canadian Network for Observational Cosmology Redshift Survey (CNOC2) galaxy groups and find that the A-D test is far more reliable and powerful at detecting real departures from an underlying Gaussian distribution than the more commonly used chi-squared and Kolmogorov tests. We use this statistic to classify a sample of the CNOC2 groups and find that 34 of 106 groups are inconsistent with an underlying Gaussian velocity distribution, and thus do not appear relaxed. In addition, we compute velocity dispersion profiles (VDPs) for all groups with more than 20 members and compare the overall features of the Gaussian and non-Gaussian groups, finding that the VDPs of the non-Gaussian groups are distinct from those classified as Gaussian.