Gaussian-Process Emulation of the Redshift-Space Halo Power Spectrum Monopole in Cosmologies with Massive Neutrinos

TL;DR

This paper introduces a Gaussian-process emulator for the redshift-space halo power spectrum in cosmologies with massive neutrinos, enabling fast and accurate likelihood evaluations for neutrino mass inference.

Contribution

The authors develop and validate a GP-based emulator trained on simulations, achieving better than 2% accuracy across relevant scales and redshifts for cosmological parameter estimation.

Findings

Emulator reproduces simulated spectra with better than 2% accuracy.

Provides a fast, reliable tool for neutrino mass inference from clustering data.

Enables efficient likelihood evaluations in MCMC analyses.

Abstract

We present a Gaussian-process (GP) emulator for the monopole of the redshift-space halo power spectrum in $\Lambda$CDM cosmologies with massive neutrinos. The emulator is trained on 1000 COLA simulations distributed in a Latin-hypercube design over the six-dimensional cosmological parameter space $\{\Omega_m h^2,\Omega_b h^2,\Omega_\nu h^2,\sigma_8,h,n_s\}$, with outputs at 11 snapshots spanning $0.5 \le z \le 2.0$. From redshift-space halo catalogues we measure shot-noise-subtracted monopole spectra over $0.01 \le k \le 0.50\,h\,\mathrm{Mpc}^{-1}$. We also generate 1000 fixed-cosmology realizations to estimate the covariance matrix and to construct synthetic data vectors for likelihood tests. On held-out cosmologies, the emulator reproduces the simulated spectra to typically better than $2\%$ across the scales and redshifts considered. Combined with its GP-based estimate of interpolation uncertainty, this speed and accuracy make the emulator well suited to repeated likelihood evaluations in Markov Chain Monte Carlo analyses. The resulting framework provides an efficient route toward neutrino-mass inference from DESI-motivated redshift-space clustering measurements.