SIMulation-Based Inference of Galaxies¶
SimBIG is a forward modeling framework for extracting cosmological information
from the 3D spatial distribution of galaxies.
It uses simulation-based inference (SBI) to perform highly efficient cosmological parameter
inference using neural density estimation from machine learning.
SimBIG enables us to leverage high-fidelity simulations that model the full details
of the observed galaxy distribution and robustly analyze higher-order clustering
on small, non-linear, scales, beyond current standard analyses.
A schematic illustration of the SimBIG framework
In Hahn et al. (2023) we analyzed the galaxy power spectrum from the Sloan Digital Sky Survey-III
Baryon Oscillation Spectroscopic Survey (BOSS) and demonstrated that we can rigorously analyze galaxy clustering down to smaller
scales than ever before and extract more cosmological information than current
standard anlayses.
In upcoming papers (Hahn et al. in prep, Lemos et al. in prep, Régaldo-Saint Blancard
et al. in prep), we will present cosmological constraints from the SimBIG analyses
of higher-order clustering.
Forward Model¶
The SimBIG forward model is based on the high-resolution Quijote
N-body simulations that accurately model non-linear structure formation.
It uses a flexible state-of-the-art halo occupation model to model the
galaxy-halo connection.
It includes survey realism and observational systematics
(e.g. survey geometry, masking, fiber collisions).
For details of the forward model, see 3D comparison and Hahn et al. (2023b).
