Linda Zhenyu Jin

We present a neural network emulator to constrain the thermal parameters of the intergalactic medium (IGM) at \5.4 \\le z \\le 6.0 using the Lyman-\\α(Ly \\α\) forest flux autocorrelation function. Our autodifferentiable JAX-based framework accelerates the surrogate model generation process using approximately 100 sparsely sampled Nyx hydrodynamical simulations with varying combinations of thermal parameters, i.e. the temperature at mean density \T_0 the slope of the temperature–density relation \\γ and the mean transmission flux \\⟨F \⟩\. We show that this emulator has a typical accuracy of 1.0 per cent across the specified redshift range. Bayesian inference of the IGM thermal parameters, incorporating emulator uncertainty propagation, is further expedited using NumPyro Hamiltonian Monte Carlo. We compare both the inference results and computational cost of our framework with the traditional nearest-neighbour interpolation approach applied to the same set of mock Ly \\αflux. By examining the credibility contours of the marginalized posteriors for \T_0, \γ, \\text{and} \⟨F \⟩ obtained using the emulator, the statistical reliability of measurements is established through inference on 100 realistic mock data sets of the autocorrelation function.

Using Neural Emulators and Hamiltonian Monte Carlo to constrain the Epoch of Reionization’s History with the Lyman-alpha Forest Power Spectrum

Adding baryonic feedback correction to the dark matter field is crucial to extracting the non-Gaussian statistics that answer fundamental questions about our Universe. Accurately modeling this process requires small-scale reconstruction of baryonic physics in high-resolution hydrodynamical simulations, which directly impacts cosmological and feedback parameter inference. However, running such simulations is significantly more computationally expensive than gravity-only N-body simulations—a key limitation in modern theoretical cosmology. In this work, we present a simple and easy-to-train AI-assisted model based on a convolutional neural network architecture that enables fast mapping from N-body to hydrodynamical simulations—specifically the total matter density fields—for parameter inference from future weak lensing observations. Our model exhibits scale-attentive training and improves the quality and accuracy of predicted fields across hydrodynamical simulation suites, outperforming previous field-level methods.