arXiv subjects

Romain Meriot

Publications and source records attributed to Romain Meriot.

9 recordsLinked to original sources

arXiv2026

Cross-simulator transfer with foundation model summaries: Towards robust SKA-era reionization inference

Simulation-based inference (SBI) for parameter estimation is vulnerable to model misspecification: neural summaries and density estimators trained on a specific forward model typically fail when applied to data drawn from another model, or from real observations, and no training simulator can capture the full observational pipeline of a real measurement exactly. We show that a self-supervised Vision Transformer (ViT), pretrained label-free on a fast approximate simulator, produces transferable data summaries that generalize across simulators. Without retraining, it can be reused as a frozen encoder to infer astrophysical parameters from a completely different simulator that resolves the radiative transfer explicitly, on which it has never seen either data or parameters. As a concrete use case in 21cm cosmology, SKATR, a ViT pretrained with a Joint Embedding Predictive Architecture (JEPA), serves as a foundation model for reionization inference from upcoming SKA measurements: SKATR is pretrained once on 67k low-cost, noiseless semi-numerical 21cmFAST lightcones, then frozen and applied to hydrodynamical Loreli II lightcones, where a lightweight conditional flow matching head infers five astrophysical parameters; the encoder is never shown Loreli data, its parameters, or any noise. In our comparison, SKATR yields the most precise and best-calibrated posteriors across all five parameters, matching the accuracy of the fully-supervised in-domain baseline while requiring 2.6x fewer radiative-transfer simulations. Under realistic SKA AA* noise, only SKATR remains simultaneously accurate, informative, and calibrated, outperforming even a supervised baseline retrained from scratch on noisy data. Self-supervised pretraining on computationally efficient semi-numerical simulations is therefore a viable route to calibrated, simulator- and noise-agnostic reionization inference for the SKA-era.

astro-ph.CO↗

arXiv2026

How X-rays heat the IGM in different 21-cm simulation codes: a comparison between Licorice and Beorn

Any interpretation of the 21-cm signal of neutral hydrogen using Bayesian inference methods can only be as accurate as the underlying simulation code used to model the state of the intergalactic medium (IGM). 3D radiative transfer (RT) simulation codes may capture complex physics, but are computationally expensive and, therefore, faster, more approximate codes have been developed. To improve our understanding of the convergence of simulation codes in the 21-cm science community, we present a comparison of the X-ray heating of the IGM modelled in Licorice, a 3D RT simulation code, and Beorn, a 1D RT code. We use Beorn to process sources extracted from Licorice simulations, using the same physics of the sources, in order to obtain two versions of the temperature of the IGM heated by X-rays. We observe a good agreement between the luminosity fields, mean temperatures, and global 21-cm signal of the two setups, but discrepancies in the distribution of temperature and 21-cm signal, which result in a $\sim 30\%$ difference in the 21-cm power spectrum. We attempt to isolate the approximations that lead to these differences and find that common approximations used in 1D RT codes produce effects of that magnitude. Using an emulator of the Licorice power spectra in an MCMC pipeline, we translate these differences between power spectra into differences between posterior distributions over the astrophysical parameters. We observe a typical bias between 1D posteriors of $\gtrsim 1 \sigma$ (with a noise level corresponding to 100h of SKA observations).

astro-ph.CO↗

arXiv2026

Overview of 21cm Experiments at high redshift with SKAO

We provide an overview of the eight SKAO Science Book chapters that motivate the Epoch of Reionisation and Cosmic Dawn experiments with SKA-Low. We describe the individual SKA-Low experiments and expected sensitivity - power spectrum, tomography, 21-cm forest, cross-correlations, building on the broad observational plan laid out in the 2015 SKA Science Book. Finally, we outline features of the telescope that will be critical for the success of EoR/CD science, e.g., beam apodization, substations, and multi-beaming.

astro-ph.CO↗

arXiv2025

Astrophysical constraints from future measurements of the kinetic Sunyaev-Zel'dovich power spectrum

High-precision measurements of the Cosmic Microwave Background (CMB) will soon allow for the unprecedented detection of small-scale secondary anisotropies, such as the kinetic Sunyaev-Zel'dovich (kSZ) effect. Linking the kSZ power spectrum to the properties of ionising sources would provide an opportunity to use such observations to access astrophysical and cosmological information from the Epoch of Reionisation, including the morphology of ionised regions, while simultaneously improving CMB analyses. The aim of this work is to assess this potential of the kSZ power spectrum to measure reionisation-era galaxy properties. We repurpose the publicly available LoReLi II simulations, which track the evolution of neutral hydrogen during reionisation, to generate a training set of patchy kSZ angular power spectra. We then train an emulator using neural network regression in order to allow for efficient Bayesian inference, and conduct forecasts assuming mock observations from current and future CMB experiments. We find that measurements of the kSZ power spectrum from such surveys can provide meaningful constraints on several of the astrophysical model parameters of the LoReLi II suite, including the ionising escape fraction for which we expect a 14% relative error, on average. They also provide an independent measurement of the CMB optical depth, marginalised over the astrophysics and with error bars competitive with the cosmic variance limit from large scale surveys. The kSZ power spectrum offers a promising avenue for probing the properties of reionisation-era galaxies and providing an independent measurement of the CMB optical depth with upcoming CMB experiments. Since the error budget of our mock observations is dominated by emulator reconstruction errors, we expect our results could be further improved with a more extended simulation training set.

astro-ph.CO↗

arXiv2025

Simulation based inference of the ionization history from the 2D 21 cm power spectrum

The 21 cm signal contains a wealth of information about the formation of the first stars and the reionization of the intergalactic medium during the Cosmic Dawn (CD) and Epoch of Reionization (EoR). The timing of these important milestones has only roughly been constrained through indirect measurements, such as from the cosmic microwave background (CMB) optical depth, and Lyman-$\alpha$ forest. Therefore, inferring the neutral fraction over cosmic time is a goal of upcoming 21 cm experiments, such as the Square Kilometer Array (SKA). We contrast two approaches to infer astrophysical parameters and ionization history from 21 cm 2D power spectra (2DPS). We develop an emulator of the 21 cm 2DPS, trained on 21cmFAST simulations, taking into account the expected instrumental noise from the SKA and sample variance. We then perform simulation based inference (SBI) using neural posterior estimation (NPE). We compare training on datasets of noisy 2DPS obtained from 21cmFAST simulations and an emulator, to infer astrophysical parameters of interest. Using an emulator of the ionization history, which has been trained on simulations from the same astrophysical parameters, we then obtain posterior distributions of the ionization history over the redshift range z $\sim$ 5-12. We demonstrate that both methods are capable of accurately recovering the ionization history and astrophysical parameters. However, coverage tests indicate that adding emulated samples does not improve predictions. This work suggests that due to the stochastic nature of the 2DPS, using an emulator of this summary statistic may result in poorer inference.

astro-ph.CO↗

arXiv2025

Generative Models of 21cm EoR Lightcones with 3D Scattering Transforms

The 21cm signal from the Epoch of Reionization (EoR) is observed as a three-dimensional data set known as a lightcone, consisting of a redshift (frequency) axis and two spatial sky plane axes. When observed by radio interferometers, this EoR signal is strongly obscured by foregrounds that are several orders of magnitude stronger. Due to its inherently non-Gaussian nature, the EoR signal requires robust statistical tools to accurately separate it from these foreground contaminants, but current foreground separation techniques focus primarily on recovering the EoR power spectrum, often neglecting valuable non-Gaussian information. Recent developments in astrophysics, particularly in the context of the Galactic interstellar medium, have demonstrated the efficacy of scattering transforms - novel summary statistics for highly non-Gaussian processes - for component separation tasks. Motivated by these advances, we extend the scattering transform formalism from two-dimensional data sets to three-dimensional EoR lightcones. To this end, we introduce a 3D wavelet set from the tensor product of 2D isotropic wavelets in the sky plane domain and 1D wavelets in the redshift domain. As generative models form the basis of component separation, our focus here is on building and validating generative models that can be used for component separation in future projects. To achieve this, we construct maximum entropy generative models to synthesise EoR lightcones, and statistically validate the generative model by quantitatively comparing the synthesised EoR lightcones with the single target lightcone used to construct them, using independent statistics such as the power spectrum and Minkowski Functionals. The synthesised lightcones agree well with the target lightcone both statistically and visually, opening up the possibility of developing for component separation methods using 3D scattering transforms.

astro-ph.CO↗

arXiv2024

Comparison of Bayesian inference methods using the Loreli II database of hydro-radiative simulations of the 21-cm signal

While the observation of the 21 cm signal from the Cosmic Dawn and Epoch of Reionization is an instrumental challenge, the interpretation of a prospective detection is still open to questions regarding the modelling of the signal and the Bayesian inference techniques that bridge the gap between theory and observations. To address some of these questions, we present Loreli II, a database of nearly 10 000 simulations of the 21 cm signal run with the Licorice 3D radiative transfer code. With Loreli II, we explore a 5-dimensional astrophysical parameter space where star formation, X-ray emissions, and UV emissions are varied. We then use this database to train neural networks and perform Bayesian inference on 21 cm power spectra affected by thermal noise at the level of 100 hours of observation with the Square Kilometer Array. We study and compare three inference techniques : an emulator of the power spectrum, a Neural Density Estimator that fits the implicit likelihood of the model, and a Bayesian Neural Network that directly fits the posterior distribution. We measure the performances of each method by comparing them on a statistically representative set of inferences, notably using the principles of Simulation-Based Calibration. We report errors on the 1-D marginalized posteriors (biases and over/under confidence) below $15 \%$ of the standard deviation for the emulator and below $25 \%$ for the other methods. We conclude that at our noise level and our sampling density of the parameter space, an explicit Gaussian likelihood is sufficient. This may not be the case at lower noise level or if a denser sampling is used to reach higher accuracy. We then apply the emulator method to recent HERA upper limits and report weak constraints on the X-ray emissivity parameter of our model.

astro-ph.CO↗

arXiv2023

The LoReLi database: 21 cm signal inference with 3D radiative hydrodynamics simulations

The Square Kilometer array is expected to measure the 21cm signal from the Epoch of Reionization (EoR) in the coming decade, and its pathfinders may provide a statistical detection even earlier. The currently reported upper limits provide tentative constraints on the astrophysical parameters of the models of the EoR. In order to interpret such data with 3D radiative hydrodynamics simulations using Bayesian inference, we present the latest developments of the \textsc{Licorice} code. Relying on an implementation of the halo conditional mass function to account for unresolved star formation, this code now allows accurate simulations of the EoR at $256^3$ resolution. We use this version of \textsc{Licorice} to produce the first iteration of \textsc{LoReLi}, a public dataset now containing hundreds of 21cm signals computed from radiative hydrodynamics simulations. We train a neural network on \textsc{LoReLi} to provide a fast emulator of the \textsc{Licorice} power spectra, \textsc{LorEMU}, which has $\sim 5\%$ rms error relative to the simulated signals. \textsc{LorEMU} is used in a Markov Chain Monte Carlo framework to perform Bayesian inference, first on a mock observation composed of a simulated signal and thermal noise corresponding to 100h observations with the SKA. We then apply our inference pipeline to the latest measurements from the HERA interferometer. We report constraints on the X-ray emissivity, and confirm that cold reionization scenarios are unlikely to accurately represent our Universe.

astro-ph.CO↗

arXiv2022

The Cosmic Mach Number as an Environment Measure for the Underlying Dark Matter Density Field

Using cosmological dark matter only simulations of a $(1.6$ Gpc$/h)^3$ volume from the Legacy simulation project, we calculate Cosmic Mach Numbers (CMN) and perform a theoretical investigation of their relation with halo properties and features of the density field to gauge their use as an measure of the environment. CMNs calculated on individual spheres show correlations with both the overdensity in a region and the density gradient in the direction of the bulk flow around that region. To reduce the scatter around the median of these correlations, we introduce a new measure, the rank ordered Cosmic Mach number ($\hat{\mathcal{M}}_g$), which shows a tight correlations with the overdensity $\delta=\frac{\rho-\bar{\rho}}{\bar{\rho}}$. Measures of the large scale density gradient as well as other average properties of the halo population in a region show tight correlations with $\hat{\mathcal{M}}_g$ as well. Our results in this first empirical study suggest that $\hat{\mathcal{M}}_g$ is an excellent proxy for the underlying density field and hence environment that can circumvent reliance on number density counts in a region. For scales between $10$ and $100 Mpc$/h, Mach numbers calculated using dark matter halos $(> 10^{12}$ M$_{\odot})$ that would typically host massive galaxies are consistent with theoretical predictions of the linear matter power spectrum at a level of $10\%$ due to non-linear effects of gravity. At redshifts $z\geq 3$, these deviations disappear. We also quantify errors due to missing large scale modes in simulations. Simulations of box size $\leq 1 $ Gpc/$h$ typically predict CMNs 10-30\% too small on scales of$\sim 100$ Mpc$/h$.

astro-ph.CO↗