Periodic Reporting for period 1 - FORWARD (Forward-Models of Cosmic Dawn: connecting 21cm simulations to the real world)
Periodo di rendicontazione: 2023-09-01 al 2025-08-31
The first billion years of our Universe's history are almost completely unexplored, and hold the key to answering fundamental questions about how the first massive structures in the Universe formed: dark matter halos, the first stars and galaxies, and the supermassive black holes that power quasars. More subtly, these early structures shaped the evolution of the intergalactic medium (IGM)---a diffuse gas of neutral hydrogen that fills the space between galaxies---by emitting radiation and heating the surrounding gas, leading to a complex interplay between the first galaxies and their environment. The 21cm line of neutral hydrogen is considered the most powerful probe of this epoch, with the potential to map the distribution and state of the IGM during these first billion years, and therefore indirectly inform us about the first cosmic structures. Large radio interferometers such as HERA and the upcoming SKA are designed to detect this faint signal from the early Universe. However, detecting and interpreting the 21cm signal is extremely challenging due to the presence of bright foregrounds, instrumental effects, and the complexity of the underlying astrophysical processes. The FORWARD project aimed to develop advanced forward-modeling techniques to simulate realistic 21cm observations, incorporating the effects of foregrounds, instrumental response, and noise. In particular FORWARD aimed to develop an end-to-end simulation and inference pipeline that consistently accounts for the known limiting systematics of the current world-leading 21cm experiments--HERA and EDGES--and jointly infers their effects alongside the astrophysical parameters of interest. This is the only principled way of generating robust scientific insight from this unexplored epoch, and thus represents an opportunity for huge scientific impact.
To achieve the objectives of the action, we pursued a number of profitable methodological directions, all aligned with the ultimate goal of building a robust end-to-end inference pipeline for 21cm cosmology. Forward-models of high-redshift 21cm observations comprise a number of distinct steps, from modeling and simulating the underlying cosmological signal, through modeling the foregrounds, computing the ideal instrumental response, adding instrumental systematics and noise, and finally performing data reduction and statistical inference. This project advanced the state-of-the-art in each of these steps. We developed a novel modeling technique for simulating the 21cm signal that includes the stochastic contributions of discrete populations of galaxies, implemented within the popular 21cmFAST simulation code. This allows for more accurate predictions of the 21cm signal, but also crucially enables joint analysis with other observations from Cosmic Dawn such as high-redshift galaxy observations from the James Webb Space Telescope. Computing the instrumental response of radio interferometers is computationally expensive, so in this project we developed two new algorithms to accelerate this step: a GPU-accelerated visibility simulator (matvis), and a novel method that exploits Non-Uniform Fast Fourier Transforms (NUFFTs) to compute visibilities more efficiently (fftvis). Both methods reduce computational cost of visibility simulation by up to a factor of 1000 and make complementary trade-offs. We also developed new models for the dominant instrumental systematic affecting HERA: mutual coupling between antennas. We implemented these models within the open-source simulation pipeline hera_sim. We developed a new open-source noise estimator for 21cm power spectrum measurements, 21cmSense v2, which enables fast and flexible forecasting of experimental sensitivities. Notably, the action supported the the PI's work as corresponding author on the very first multi-redshift power spectrum upper limits from HERA Phase II---a significant achievement that has placed the strongest limits on the signal during Cosmic Dawn from any experiment, and establishes HERA's end-to-end analysis pipeline as providing unbiased estimates of the signal over a wide range of scales. This result has been submitted to be published in a peer-reviewed journal. Supporting this significant result, we produced the largest-ever end-to-end instrumental simulations using the tools developed in this action: fftvis and hera_sim. To interpret the result in terms of astrophysics, we used a Bayesian inference tool also developed within this action: 21cmEMU. This is a novel machine-learning based emulator for 21cm power spectra, which enables fast parameter inference over different subsets of data.
Gains were also made in the analysis of EDGES data to validate the first claimed detection of the 21cm global signal. We developed a new data analysis pipeline that incorporates components that can be used in a forward-modeling framework, and this work is close to completion. Simultaneously, we have been applying this pipeline towards two important goals: exploring the impact of various analysis choices on the seminal results of the EDGES-2 experiment (which claimed the first detection of the global 21cm signal from Cosmic Dawn in 2018), and the first processing of the next-generation EDGES-3 experiment. These results are all in the final stages of preparation.
The action and the collaborations it fostered have supported a substantial driving contribution to 9 publications in peer-reviewed journals so far, with three more already submitted, and five more in the latter stages of preparation.
Gains were also made in the analysis of EDGES data to validate the first claimed detection of the 21cm global signal. We developed a new data analysis pipeline that incorporates components that can be used in a forward-modeling framework, and this work is close to completion. Simultaneously, we have been applying this pipeline towards two important goals: exploring the impact of various analysis choices on the seminal results of the EDGES-2 experiment (which claimed the first detection of the global 21cm signal from Cosmic Dawn in 2018), and the first processing of the next-generation EDGES-3 experiment. These results are all in the final stages of preparation.
The action and the collaborations it fostered have supported a substantial driving contribution to 9 publications in peer-reviewed journals so far, with three more already submitted, and five more in the latter stages of preparation.
The work performed within FORWARD has significantly advanced the state-of-the-art along multiple fronts in 21cm cosmology. Our discrete-galaxy modeling of the 21cm signal, implemented in 21cmFAST v4, is the first of its kind, and opens up new avenues for joint analysis of 21cm data with other high-redshift observations such as those from JWST. Our two new visibility simulation methods increase the efficiency of this crucial step by up to three orders of magnitude, and have already been used to perform the largest end-to-end simulations of HERA to date, supporting the first upper limits from HERA Phase II. As they are open-source, they can be adopted by the wider community, and will be especially impactful for upcoming large interferometers, such as the SKA. Combining 21cmFAST v4 with these visibility simulators, we are preparing the largest 21cm simulation ever performed, covering HERA's full field-of-view and updating the state-of-the-art theoretical reference for the community to include recent insights from JWST. Our new open-source noise estimator, 21cmSense v2, is already being widely used for forecasting the sensitivity of current and future 21cm experiments, including HERA and the SKA. Our new emulator, 21cmEMU, is the first to simultaneously emulate multiple observables, making it the most powerful tool of its kind for 21cm parameter inference. It has already been used to perform the first Bayesian inference of 21cm astrophysical parameters from HERA Phase II data, and its broad impact is evident from the 33 citations it has already received since publication. This action supported the first upper limits from HERA Phase II, which are the deepest ever placed during Cosmic Dawn, and are consistent with thermal noise at all redshifts, indicating that the pipeline is in good shape and that adding more data will continue to decrease the limits over time. This will have monumental impact across the field. As the community continues to develop the full framework required for end-to-end 21cm parameter inference, the tools and methods developed within FORWARD will remain crucial components. While support from the action has ended, the seeds of work started in this action will continue to be developed towards a seamless end-to-end pipeline that not only enables deep data validation, but also comprehensive scientific inference.