A fast method of reionization parameter space exploration using GPR trained SCRIPT

Abstract

ABSTRACT Efficient exploration of parameter spaces is crucial to extract physical information about the Epoch of Reionization from various observational probes. To this end, we propose a fast technique based on Gaussian process regression training applied to a semi-numerical photon-conserving reionization model, SCRIPT. Our approach takes advantage of the numerical convergence properties of SCRIPT and constructs a training set based on low-cost, coarse-resolution simulations. A likelihood emulator is then trained using this set to produce results in approximately two orders of magnitude less computational time than a full Markov Chain Monte Carlo (MCMC) run, while still generating reasonable 68 per cent and 95 per cent confidence contours. Furthermore, we conduct a forecasting study using simulated data to demonstrate the applicability of this technique. This method is particularly useful when full MCMC analysis is not feasible due to expensive likelihood computations.