WorksPhysics and Astronomy
On-the-Fly Gaussian Process Reconstruction of the Galaxy Redshift Distribution for 32pt Cosmology
Abstract
The joint analysis of weak gravitational lensing, photometric galaxy clustering, and their cross-correlation, the so-called 32pt analysis, is one of the primary probes Stage-IV surveys will use to constrain the dark-energy equation of state. Its accuracy depends critically on the calibration of the underlying photometric redshift distributions, , whose residual uncertainties can bias cosmological inference. However, conventional parametric nuisance models may not adequately capture the full range of plausible uncertainties. We therefore present a cosmological inference framework that instead reconstructs on the fly using Gaussian Processes (GPs), replacing per-bin shift-and-width parameterisations by a non-parametric, data-driven model. We validate the method on synthetic Euclid-like 32pt data vectors representative of Euclid Data Release 1, sampling GP hyperparameters jointly with cosmological and nuisance parameters to reliably propagate uncertainties into the posterior distributions. We find that the dependence of the inferred parameters on the adopted model is markedly stronger in CDM than in CDM. Opening the dark-energy sector both amplifies the biases induced by a given treatment and changes which reconstruction the data prefer, so that a model validated in CDM does not automatically remain adequate once and are free. Our results thus indicate that Stage-IV surveys should define uncertainty marginalisation requirements separately in cosmologies beyond CDM to ensure unbiased dark-energy constraints.