Constraining the Hubble Constant with a Simulated Full Covariance Matrix Using Neural Networks

Abstract: The Hubble parameter, $H(z)$, plays a crucial role in understanding the expansion history of the universe and constraining the Hubble constant, $\mathrm{H}_0$. The Cosmic Chronometers (CC) method provides an independent approach to measuring $H(z)$, but existing studies either neglect off-diagonal elements in the covariance matrix or use an incomplete covariance matrix, limiting the accuracy of $\mathrm{H}_0$ constraints. To address this, we use a Fully Connected Neural Network (FCNN) to simulate the full $33 \times 33$ covariance matrix based on a previously proposed $15 \times 15$ covariance matrix. We find that two key hyperparameters, epochs and batch size, significantly affect the simulation and introduce two criteria for selecting optimal values. Using the simulated covariance matrix, we constrain $\mathrm{H}_0$ via two independent methods: EMCEE and Gaussian Process. Our results show that different hyperparameter selection criteria lead to variations in the chosen combinations but have little impact on the final constrained $\mathrm{H}_0$. However, different epochs and batch size settings do affect the results. Incorporating the simulated covariance matrix increases the uncertainty in $\mathrm{H}_0$ compared to using no covariance matrix or only the proposed $15 \times 15$ covariance matrix. The comparison between EMCEE and GP suggests that the constraint method itself also influences the final $\mathrm{H}_0$. These findings highlight the importance of properly modeling covariance in CC-based $\mathrm{H}_0$ constraints.