Stacked Hybrid Recurrent Convolutional Neural Network Reconstruction of X-Ray Influence on 21 cm Brightness Temperature

Abstract X-ray photons substantially affect the thermal and ionization states of the intergalactic medium during the epoch of reionization (EoR), thereby significantly influencing the 21 cm line observables, such as its sky-averaged (global) brightness temperature. Nevertheless, the complicated dependency of astrophysical processes on a broad spectrum of parameters, including X-ray efficiency, spectral characteristics, and gas dynamics, makes precisely simulating the effect of X-ray flux challenging. Traditional approaches, including N -body and hydrodynamical simulations, are computationally intensive and struggle to explore high-dimensional parameter spaces efficiently. We present a stacked hybrid model trained on a specific simulation intended to reconstruct the effect of X-ray flux on the global 21 cm brightness temperature during the EoR. Along with convolutional neural networks (CNNs), this architecture combines two powerful variants of recurrent neural networks, long short-term memory (LSTM) and gated recurrent unit (GRU), thereby enabling fast adaptation to a range of X-ray flux levels. Without demanding repeated simulations, this emulator preserves temporal and spatial dependencies and generalizes to unseen parameter combinations. This reduces computation time by a factor of 1 million while preserving an excellent prediction accuracy of 99.93%, facilitating studies on high-dimensional parameter inference and sensitivity with an error margin of less than 0.266 mK. Our LSTM-GRU-CNN emulator combines recurrent and convolutional architectures to enable a robust and scalable analysis of X-ray heating effects on the global 21 cm signal during the EoR.