$\texttt{unimpeded}$: A Public Nested Sampling Database for Bayesian Cosmology

TL;DR

Bayesian inference in cosmology hinges on evidences and normalised posteriors, which are computationally expensive to obtain and hamper model comparison and tension analyses. The paper introduces unimpeded, a pip-installable library and data repository offering a public grid of pre-computed nested sampling and MCMC chains for 8 cosmological models (including $Λ$CDM and seven extensions) across 39 datasets, with normalised posteriors and built-in six tension metrics: $R$, $I$, $S$, $d_G$, $\sigma$, and $p$-value. Evidence and Kullback–Leibler divergence can be calculated with anesthetic, enabling rapid model comparison and quantification of dataset constraining power. The workflow includes Zenodo archiving with permanent DOIs via DatabaseCreator, public data access via DatabaseExplorer, and an end-to-end pipeline using YAML for HPC nested sampling, Cobaya, PolyChord, CAMB, and anesthetic for analysis and visualization. This resource lowers computational barriers, promotes reproducibility, and accelerates systematic tension studies across cosmological probes.

Abstract

Bayesian inference is central to modern cosmology. While parameter estimation is achievable with unnormalised posteriors traditionally obtained via MCMC methods, comprehensive model comparison and tension quantification require Bayesian evidences and normalised posteriors, which remain computationally prohibitive for many researchers. To address this, we present $\texttt{unimpeded}$, a publicly available Python library and data repository providing DiRAC-funded (DP192 and 264) pre-computed nested sampling and MCMC chains with their normalised posterior samples, computed using $\texttt{Cobaya}$ and the Boltzmann solver $\texttt{CAMB}$. $\texttt{unimpeded}$ delivers systematic analysis across a grid of eight cosmological models (including $Λ$CDM and seven extensions) and 39 modern cosmological datasets (comprising individual probes and their pairwise combinations). The built-in tension statistics calculator enables rapid computation of six tension quantification metrics. All chains are hosted on Zenodo with permanent access via the unimpeded API, analogous to the renowned Planck Legacy Archive but utilising nested sampling in addition to traditional MCMC methods.

Figures (2)

• Figure 1: The unimpeded ecosystem and workflow. At the centre, unimpeded manages data archival and retrieval through Zenodo, providing permanent DOIs and public access to pre-computed chains. For data generation, unimpeded configures YAML files for resource-intensive HPC nested sampling using Cobaya, PolyChord, and CAMB. For analysis, users download chains via DatabaseExplorer and leverage anesthetic for visualisation (corner plots, posterior distributions, constraint contours) and tension quantification (six metrics: $R$ statistic, information ratio $I$, suspiciousness $S$, Bayesian model dimensionality $d_G$, significance $\sigma$, and $p$-value).
• Figure 2: Tension analysis heatmap produced by unimpeded and anesthetic displaying p-value derived tension significance ($\sigma$ values) for 31 pairwise dataset combinations across 8 cosmological models. Rows are sorted by significance, with the most problematic dataset pairs (highest tension) at the top. This demonstrates unimpeded's capability to systematically quantify tensions and their model dependence.