Predictability in Semi-Analytic Models of Galaxy Formation

TL;DR

This paper introduces a framework to evaluate the predictability of semi-analytic galaxy formation models by analyzing their response to input perturbations, revealing mass-dependent variability in galaxy properties.

Contribution

It presents a novel approach using perturbations and a predictability metric to benchmark and understand semi-analytic models of galaxy formation.

Findings

Small halos show highly predictable galaxy responses.

Massive halos exhibit large, seemingly random fluctuations.

Predictability metric enables quantitative benchmarking.

Abstract

We propose a general framework to scrutinize the performance of semi-analytic codes of galaxy formation. The approach is based on the analysis of the outputs from the model after a series of perturbations in the input parameters controlling the baryonic physics. The perturbations are chosen in a way that they do not change the results in the luminosity function or mass function of the galaxy population. We apply this approach on a particular semi-analytic model called GalICS. We chose to perturb the parameters controlling the efficiency of star formation and the efficiency of supernova feedback. We keep track of the baryonic and observable properties of the central galaxies in a sample of dark matter halos with masses ranging from 10^{10} M_sol to 10^{13} M_sol. We find very different responses depending on the halo mass. For small dark matter halos its central galaxy responds in a highly predictable way to small perturbation in the star formation and feedback efficiency. For massive dark matter halos, minor perturbations in the input parameters can induce large fluctuations on the properties of its central galaxy, at least $\sim 0.1$ in (B-V) color or $\sim 0.5$ mag in U or r filter, in a seemingly random fashion. We quantify this behavior through an objective scalar function we call predictability. We argue that finding the origin of this behavior needs additional information from other approximations and different semi-analytic codes. Furthermore, the implementation of an scalar objective function, such as the predictability, opens the door to quantitative benchmarking of semi-analytic codes based on its numerical performance.