UC Santa Barbara

One of the great successes of modern cosmology is the percent-level concordance between theory and observations of the intergalactic medium (IGM) at z ≳ 1.7. Yet, the Lyα forest at z < 1.7, which can only be studied via HST UV spectra, has pointed out a puzzling discrepancy, i.e., the Doppler b-parameters of these absorption lines are, on average, ∼ 10 km/s wider than those in any existing hydrodynamic simulation. This discrepancy implies that the low-z IGM might be substantially hotter than expected, contradicting one of the fundamental predictions in current cosmology that the IGMshould cool down owing to the Hubble expansion after He ii reionization ( z < 2.5). Moreover, the IGM thermal state degenerates with its ionization state characterized by the UV background (UVB) photoionization rate, ΓHI , which dictates the abundance of the Lyα absorbers, dN /dz. Such a degeneracy requires any reliable measurement to adopt a careful statistical inference procedure. To overcome these difficulties, in this thesis, a novel machine-learning-based inference framework is employed to jointly measure the thermal and ionization state of the IGM, using the 2D distribution of b-parameter and H i column density and dN /dz. This method effectively resolves the degeneracies between the thermal and ionization state of the IGM and achieves high precision, even with limited-sized data. I apply this method to 94 archival HST COS and STIS quasar spectra distributed across the seven redshift bins, yielding a comprehensive evolutionary history of the IGM thermal and ionization state at z < 1.5. The results suggest that the IGM may be significantly hotter than previously expected at low-z and is potentially isothermal, with IGM temperature at mean density, T0 ∼ 30, 000K and power-law index of the temperature-density, γ ∼ 1.0 at z = 0.1. The inferred thermal history suggests that this unexpected IGM temperature possibly emerges around z ∼ 1. Additionally, while the ΓHI measurements align with the theoretical model at z ∼ 1, the values measured at z < 0.5 are substantially lower than predicted, posing challenges to low-z UV background synthesis models.