Cosmological constraints from BOSS with analytic covariance matrices

TL;DR

This paper validates the use of analytic covariance matrices for galaxy power spectrum analysis, showing they match mock-based results, reduce computational costs, and improve precision for future surveys like DESI and Euclid.

Contribution

It demonstrates the effectiveness of analytic covariance matrices in cosmological parameter estimation, offering advantages over sample covariances in accuracy and computational efficiency.

Findings

Analytic covariances agree with mock-based results.

Analytic approach reduces systematic errors and computational costs.

Non-Gaussian contributions have a marginal impact on parameter errors.

Abstract

We use analytic covariance matrices to carry out a full-shape analysis of the galaxy power spectrum multipoles from the Baryon Oscillation Spectroscopic Survey (BOSS). We obtain parameter estimates that agree well with those based on the sample covariance from two thousand galaxy mock catalogs, thus validating the analytic approach and providing substantial reduction in computational cost. We also highlight a number of additional advantages of analytic covariances. First, the analysis does not suffer from sampling noise, which biases the constraints and typically requires inflating parameter error bars. Second, it allows us to study convergence of the cosmological constraints when recomputing the analytic covariances to match the best-fit power spectrum, which can be done at a negligible computational cost, unlike when using mock catalogs. These effects reduce the systematic error budget of cosmological constraints, which suggests that the analytic approach may be an important tool for upcoming high-precision galaxy redshift surveys such as DESI and Euclid. Finally, we study the impact of various ingredients in the power spectrum covariance matrix and show that the non-Gaussian part, which includes the regular trispectrum and super-sample covariance, has a marginal effect ($\lesssim 10 \%$) on the cosmological parameter error bars. We also suggest improvements to analytic covariances that are commonly used in Fisher forecasts.