Using Machine Learning to Speed Up the Analysis of Cosmological Models

Many cosmological models remain untested or under-tested due to computational limitations. In order to compare a theoretical model to observed data, codes which calculate these models have to be called upwards of 10⁵ to 10⁶ times in a Monte Carlo Markov Chain (MCMC) algorithm. Furthermore, a complete analysis consists of not just one of these MCMC algorithms but multiple, utilizing different datasets. One of these computationally intensive cosmological models is a model which introduces two scalar fields which interact through their kinetic terms. It has been argued that such a model may address both the Hubble and S₈ tensions. However, as of yet, this model has not been fully analyzed due to the fact that it is computationally intensive. In order to perform the first analysis of this model we compute an emulator using CONNECT, a code for generating cosmological emulators. Our preliminary work finds that, contrary to claims in the literature, this model is unable to provide a good fit to the Planck satellite’s measurements of the cosmic microwave background power spectra.