Yi Mao

We introduce a native tensor-based framework for foreground mitigation in 21 cm intensity mapping (IM), utilizing the Oriented Singular Value Decomposition (O-SVD) algorithm. While 21 cm IM is a powerful probe of the large-scale structure of the Universe, its efficacy is severely limited by astrophysical foregrounds that are orders of magnitude brighter than the cosmological signal. Traditional mitigation strategies often necessitate flattening multidimensional data cubes into 2D matrices, a process that potentially compromises the intrinsic spatial–spectral correlations by treating distinct spatial pixels as independent samples. By treating multifrequency sky maps and angular power spectra as third-order tensors, the O-SVD method performs decomposition directly on the multilinear manifold, preserving the underlying physical topology and leveraging the distinct coherence properties of astrophysical foregrounds across different dimensions. We demonstrate the performance and versatility of the O-SVD framework through its application to high-fidelity simulations from the Square Kilometre Array Science Data Challenge 3a and real-world observational data from the Tianlai Cylinder Pathfinder Array. Our results indicate that O-SVD provides a robust and universal approach for foreground subtraction, achieving high-fidelity signal recovery while offering superior performance compared to conventional matrix-based SVD methods.

The advent of increasingly advanced surveys and cosmic tracers has motivated the development of new inference techniques and novel approaches to extracting information from cosmic fields. A central challenge in this endeavor is to quantify the information content carried by these summary statistics in cosmic fields. In particular, how should we assess which statistics are more informative than others and assess the exact degree of complementarity of the information from each statistic? Here, we introduce mutual information (MI), which provides, from an information-theoretic perspective, a natural framework for assessing the sufficiency and complementarity of summary statistics in cosmological data. We demonstrate how MI can be applied to typical inference tasks to make information-theoretic evaluations, using two representative examples: the cosmic microwave background map, from which the power spectrum (PS) extracts almost all information as is expected for a Gaussian random field, and the 21 cm brightness temperature map, from which the scattering transform extracts the most non-Gaussian information but which is complementary to the PS and bispectrum. Our results suggest that MI offers a robust theoretical foundation for evaluating and improving summaries, thereby enabling a deeper understanding of cosmic fields from an information-theoretic perspective.

The advent of increasingly advanced surveys and cosmic tracers has motivated the development of new inference techniques and novel approaches to extracting information from cosmic fields. A central challenge in this endeavor is to quantify the information content carried by these summary statistics in cosmic fields. In particular, how should we assess which statistics are more informative than others and assess the exact degree of complementarity of the information from each statistic? Here, we introduce mutual information (MI) that provides, from an information-theoretic perspective, a natural framework for assessing the sufficiency and complementarity of summary statistics in cosmological data. We demonstrate how MI can be applied to typical inference tasks to make information-theoretic evaluations, using two representative examples: the cosmic microwave background map, from which the power spectrum extracts almost all information as is expected for a Gaussian random field, and the 21~cm brightness temperature map, from which the scattering transform extracts the most non-Gaussian information but is complementary to power spectrum and bispectrum. Our results suggest that MI offers a robust theoretical foundation for evaluating and improving summaries, thereby enabling a deeper understanding of cosmic fields from an information-theoretic perspective.

Emulators using machine learning techniques have emerged to efficiently generate mock data matching the large survey volume for upcoming experiments, as an alternative approach to large-scale numerical simulations. However, high-fidelity emulators have become computationally expensive as the simulation volume grows to hundreds of megaparsecs. Here, we present a multifidelity emulation of large-scale 21 cm lightcone images from the epoch of reionization, which is realized by applying the few-shot transfer learning to training generative adversarial networks (GAN) from small-scale to large-scale simulations. Specifically, a GAN emulator is first trained with a huge number of small-scale simulations, and then transfer-learned with only a limited number of large-scale simulations, to emulate large-scale 21 cm lightcone images. We test the precision of our transfer-learned GAN emulator in terms of representative statistics including global 21 cm brightness temperature history, 2D power spectrum, and scattering transform coefficients. We demonstrate that the lightcone images generated by the transfer-learned GAN emulator can reach the percentage level precision in most cases on small scales, and the error on large scales only increases mildly to the level of a few tens of percent. Nevertheless, our multifidelity emulation technique saves a significant portion of computational resources that are mostly consumed for generating training samples for GAN. On estimate, the computational resource by training GAN completely with large-scale simulations would be 1 to 2 orders of magnitude larger than using our multifidelity technique. This implies that our technique allows for emulating high-fidelity, traditionally computationally prohibitive, images in an economic manner.

The 21 cm line and the patchy kinetic Sunyaev–Zel’dovich (kSZ) effect are promising and complementary probes of the Epoch of Reionization. A challenge for cross correlating these two signals is that foreground avoidance or removal algorithms applied to the 21 cm data inevitably sacrifice Fourier modes with long wavelengths along the line of sight (i.e., low-k _∥ modes), yet only these same modes contribute to the kSZ signal. Here we show that a suitable kSZ² × 21 cm² cross-correlation statistic nevertheless remains nonvanishing, even after filtering out the corrupted low-k _∥Fourier modes from the 21 cm data. We simulate the kSZ² × 21 cm² cross-correlation signal across reionization-era redshifts and find distinctive redshift evolution. This signal peaks early in the reionization history, when the volume-averaged fraction is around 0.1 ≲ x _HII ≲ 0.2, after which it changes sign and reaches a minimum near reionization’s midpoint (x _HII ∼ 0.5), while the signal gradually vanishes as reionization completes. These trends appear generic across three simulated models which differ in their reionization histories. We forecast the detectability of the kSZ² × 21 cm² cross-power spectrum for the HERA and SKA1-Low 21 cm experiments in combination with current and next-generation cosmic microwave background (CMB) surveys including the Simons Observatory, CMB-S4, and CMB-HD. We find that a high-significance detection (signal-to-noise ratio ≳ 5σ) is possible with SKA1-Low and CMB-S4.

Atomic hydrogen (H i) gas, mostly residing in dark matter halos after cosmic reionization, is the fuel for star formation. Its relation to the properties of the host halo is the key to understand the cosmic H i distribution and to mock H i large scale structure accurately and conveniently for the large area H i intensity mapping surveys in the future. In this work, we propose a flexible empirical model of H i–halo relation. In this model, while the H i mass depends primarily on the mass of host halo, there is also secondary dependence on other halo properties. We apply our model to the observation data of the Arecibo Fast Legacy ALFA (ALFALFA) survey, and find it can successfully fit the cosmic H i abundance (Ω_{H I} ), average H i–halo mass relation 〈M _{H I} ∣M _h〉, and the H i clustering. The best fit of the ALFALFA data rejects the model with no secondary halo dependence of H i and the model with secondary dependence on the halo spin parameter (λ), and shows a strong dependence on halo formation time (a _1/2) and halo concentration (c _vir). In an attempt to explain these findings from the perspective of hydrodynamical simulations, the IllustrisTNG simulation confirms the dependence of H i mass on secondary halo parameters. However, the IllustrisTNG results show a strong dependence on λ and a weak dependence on c _vir and a _1/2, and also predict a much larger value of H i clustering on large scales than observations. This interesting difference between simulation and observation calls for improvements in understanding the H i–halo relation from both theoretical and observational sides.

The 21 cm line from neutral hydrogen captures rich cosmological information. However, the 21 cm signals are highly contaminated by foreground emission, making accurate component separation vital for cosmological analyses. Gaussian processes (GPs) have been utilized for component separation in 21 cm data analyses, exploiting the distinct spectral behavior of the cosmological and foreground signals, which are modeled through the GP covariance kernel. Previous approaches have employed a global GP kernel along all lines of sight (LoS). In this work, we study Bayesian approaches that allow for spatial variations in foreground kernel parameters, testing them against simulated H i intensity mapping observations, which measures the total H i intensity in large voxels. We consider a no-pooling (NP) model, treating each LoS independently by fitting for separate covariance kernels, and a hierarchical Gaussian process (HGP) model, allowing for variation in kernel parameters between different LoS, regularized through a global hyperprior. We find that including spatial variations in the GP kernel parameters results in a significant improvement in cosmological signal recovery, achieving up to a 30% reduction in the standard deviation of the residual distribution. Allowing for spatial variations also improves the recovery of the H i power spectra and wavelet scattering transform coefficients. While the NP model achieves superior recovery as measured by the residual distribution, it is computationally demanding, faces convergence challenges, and is prone to overfitting. The HGP model strikes a balance between accuracy and robustness. Further improvements to the HGP model will require more physically-motivated modeling of foreground spatial variations. Our code can be found in the following repository: https://github.com/dkn16/H21F.

We introduce synax (https://github.com/dkn16/Synax), a novel library for automatically differentiable simulation of Galactic synchrotron emission. Built on the JAX framework, synax leverages JAX’s capabilities, including batch acceleration, just-in-time compilation, and hardware-specific optimizations (CPU, GPU, TPU). Crucially, synax uses JAX’s automatic differentiation (AD) mechanism, enabling precise computation of analytical derivatives with respect to any model parameters. This facilitates powerful inference algorithms, such as Hamiltonian Monte Carlo (HMC) and gradient-based optimization, which enables inference over models that would otherwise be computationally prohibitive. In its initial release, synax supports synchrotron intensity and polarization calculations down to GHz frequencies, alongside several models of the Galactic magnetic field (GMF), cosmic-ray spectra, and thermal electron density fields. When running synax on the CPU we obtain identical performance to hammurabi, a state-of-the-art synchrotron simulation package, while on the GPU synax brings a 20-fold enhancement in efficiency. We further demonstrate the potential of AD in enabling full posterior inference using gradient-based inference algorithms. Using synax with HMC to perform inference over a four-parameter test model, we attain a twofold improvement compared to standard random walk Metropolis–Hastings (RWMH). When applied to a more complex 16-parameter model, HMC is still able to obtain accurate posterior expectations, while RWMH fails to converge. We also showcase the application of synax to optimizing the GMF based on the Haslam 408 MHz map, achieving residuals with a standard deviation below 1 K.