🔭 astrophysics

Interpretation of 21 cm Auto Power Spectrum Measurement at $z\sim 1$ by the Canadian Hydrogen Intensity Mapping Experiment

The Canadian Hydrogen Intensity Mapping Experiment (CHIME) measured the 21 cm auto power spectrum at $z\sim 1$ and interpreted it through both parametric modeling to constrain HI bias and density parameters, and comparison with IllustrisTNG simulations, which revealed a significant tension suggesting discrepancies in the simulated nonlinear redshift-space clustering of neutral hydrogen.

Original authors: CHIME Collaboration, Mandana Amiri, Kevin Bandura, Arnab Chakraborty, Zhuo Yu Brian Chu, Matt Dobbs, Simon Foreman, Liam Gray, Mark Halpern, Gary Hinshaw, Albin Joseph, Nolan Kruger, Joshua MacEachern

Published 2026-03-27

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: CHIME Collaboration, Mandana Amiri, Kevin Bandura, Arnab Chakraborty, Zhuo Yu Brian Chu, Matt Dobbs, Simon Foreman, Liam Gray, Mark Halpern, Gary Hinshaw, Albin Joseph, Nolan Kruger, Joshua MacEachern, Kiyoshi W. Masui, Juan Mena-Parra, Laura Newburgh, Tristan Pinsonneault-Marotte, Alex Reda, Shabbir Shaikh, Seth R. Siegel, Yukari Uchibori, Keith Vanderlinde, Haochen Wang, Dallas Wulf

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Mapping the Invisible Forest

Imagine the universe as a giant, dark forest. Most of the trees are made of "normal" matter (stars, planets, us), but the forest is also filled with a vast, invisible mist made of Hydrogen gas. This gas is the raw material for making new stars.

For a long time, astronomers could only see the "trees" (galaxies) because they glow. But the "mist" (the hydrogen gas) is invisible to our eyes. The CHIME telescope in Canada is a giant radio dish designed to listen to a specific "whisper" that this hydrogen gas makes—a radio signal called the 21 cm line. By listening to this whisper, CHIME can map where the invisible gas is hiding, even when there are no stars around to light it up.

The Experiment: Listening to the Echo

The scientists used CHIME to listen to this hydrogen whisper from a time in the universe's history when it was about 10 billion years old (a redshift of $z \sim 1$ ). They didn't just listen for a single sound; they measured the clumpiness of the gas.

Think of the hydrogen gas like popcorn kernels floating in a giant bowl.

Smooth: If the kernels are spread out perfectly evenly, the bowl is smooth.
Clumpy: If the kernels are bunched together in clusters, the bowl is lumpy.

The scientists measured how "lumpy" the hydrogen was at different sizes. They found that the gas was clumping together much more strongly than the universe's most famous computer simulations predicted.

The Two Ways They Tried to Understand It

The paper describes two different ways the team tried to make sense of their measurements.

Approach 1: The "Volume Knob" and the "Shape Shifter"

The team tried to fit a mathematical model to their data. They realized the signal they heard depends on two main things:

How much gas is there? (The total volume of the popcorn).
How clumpy is it? (How tightly the popcorn is bunched).

The Problem: It's like trying to guess the volume of a room just by listening to how loud a speaker is. If the speaker is loud, is it because the room is huge, or because the volume knob is turned up? You can't tell the difference just by listening. In physics terms, the "amount of gas" and the "clumpiness" are degenerate—they are stuck together in a way that makes them hard to separate.

The Solution:

They created a "Master Volume Knob" called $A^2_{HI}$ . This combines the amount of gas and its clumpiness into one number. They measured this number very precisely.
To separate the two, they borrowed a "rule of thumb" from other scientists. If they assume the "clumpiness" is a certain amount (based on computer models), they can calculate exactly how much gas is there.
The Result: They found the gas is clumping together quite a bit. When they checked their numbers against other studies, they found a match, giving them confidence that their measurement is real and not a mistake.

Approach 2: The "Video Game" Comparison

The second approach was like comparing a real-life photo to a video game screenshot.

The Real Photo: The actual data CHIME collected from the sky.
The Video Game: The IllustrisTNG simulations. These are super-computer models that try to recreate the history of the universe, including how gas behaves, how stars form, and how galaxies collide.

The Surprise:
When they compared the real photo to the video game, they found a huge mismatch.

The Real Universe (the photo) showed the gas clumping together very tightly.
The Video Game (the simulation) showed the gas being much more spread out and lazy.

The difference was so big that it was statistically impossible to be a coincidence.

For the TNG100 simulation, the real data was 3.1 times more "clumpy" than the game predicted.
For the TNG300 simulation, the real data was 4.0 times more clumpy.

Why Does This Matter?

This is a big deal for two reasons:

The "Game" Needs a Patch: The computer simulations are our best tools for understanding the universe. If the game says the gas should be spread out, but reality shows it's clumpy, then the "physics engine" inside the game is missing something.
- Analogy: Imagine a video game where gravity is too weak, so characters float away. You know the game is broken because you see people floating in real life.
- The scientists suspect the simulations aren't simulating the "baryonic effects" (the complex physics of gas, supernovae, and black holes) correctly. The gas in the real universe seems to be getting pulled into clumps more aggressively than the computer models think it should.
We Can "See" the Invisible: This paper proves that we can use radio telescopes to measure the "nonlinear" scales of the universe. These are the small, messy, clumpy scales where the simple rules of physics break down and things get complicated. CHIME successfully navigated this messy terrain and found a signal that tells us something new about how the universe works.

The Conclusion

The CHIME team has successfully mapped the invisible hydrogen gas in the universe. They found that this gas is clumping together much more tightly than our best computer models predict.

The takeaway: The universe is more "messy" and "clumpy" than our simulations think. This doesn't mean the simulations are useless, but it means they need an update. The "physics" inside the computer code needs to be tweaked to explain why the real hydrogen gas is so eager to bunch together. This discovery opens a new door for understanding how galaxies form and how gas moves around in the cosmic web.

1. Problem Statement

The paper addresses the challenge of interpreting the first measurement of the 21 cm auto power spectrum by the Canadian Hydrogen Intensity Mapping Experiment (CHIME) at redshift $z \sim 1$ . The measurement covers the frequency range 608.2–707.8 MHz ( $0.4 \lesssim k \lesssim 1.5 \, h \, \text{Mpc}^{-1}$ ), a regime where structure formation is nonlinear.

The core difficulties are:

Parameter Degeneracy: The amplitude of the 21 cm power spectrum depends on the product of the neutral hydrogen density ( $\Omega_{\text{HI}}$ ) and the linear bias ( $b_{\text{HI}}$ ), making them difficult to constrain individually without external priors.
Nonlinear Modeling: Standard perturbative theories fail at these small scales ( $k > 0.4 \, h \, \text{Mpc}^{-1}$ ). Modeling requires phenomenological parameters to account for nonlinear clustering and redshift-space distortions (Fingers of God).
Simulation Discrepancies: It is unclear whether current cosmological hydrodynamical simulations (like IllustrisTNG) accurately predict the clustering of HI on these nonlinear scales.

2. Methodology

The authors employ two distinct approaches to interpret the CHIME data, utilizing a custom pipeline that forward-models the telescope response and analysis steps.

A. Data Processing

Dataset: 94 sidereal days of CHIME data (Jan–Nov 2019).
Pipeline: Includes RFI masking, beamforming, foreground removal (DAYENU filter), bandpass correction (HyFoReS), and cross-correlation of even/odd days to eliminate noise bias.
Output: An auto power spectrum measured in 8 bandpowers within the nonlinear regime.

B. Approach 1: Parametric Bayesian Modeling

Model: A 4-parameter phenomenological model for the 21 cm power spectrum $P_{21}(k, \mu)$ $P_{21} (k, μ)$ :
1. $\Omega_{\text{HI}}$ : Mean HI density.
2. $b_{\text{HI}}$ : Linear bias.
3. $\alpha_{\text{NL}}$ : Interpolates between linear and nonlinear matter power spectra.
4. $\alpha_{\text{FoG}}$ : Scales the Finger-of-God damping kernel (Lorentzian).
Template-Based Forward Modeling: Instead of a general transfer function, the authors generate 120 simulated datasets by varying the input parameters. They use these to pre-compute "templates" of the observed power spectrum. The final model is a linear combination of these templates, allowing for fast Markov Chain Monte Carlo (MCMC) sampling.
Effective Amplitude: Due to strong degeneracy between $\Omega_{\text{HI}}$ and $b_{\text{HI}}$ , they define a single effective amplitude parameter:
$A^2_{\text{HI}} \equiv 10^6 \Omega^2_{\text{HI}} (b^2_{\text{HI}} + \langle f\mu^2 \rangle)^2$
where $\langle f\mu^2 \rangle$ accounts for redshift-space distortions.
Validation: The model is validated against 20 mock simulations with Latin hypercube sampling and against scaled IllustrisTNG power spectra to ensure unbiased recovery of parameters.

C. Approach 2: Comparison with IllustrisTNG

Simulation Propagation: The authors take HI power spectra from the TNG100 and TNG300 runs of the IllustrisTNG suite (at $z=1$ ).
Forward Modeling: These spectra are processed through the full CHIME simulation pipeline (including beam effects, masking, and foreground filtering) to generate "observed" predictions.
Scaling: They fit a free amplitude parameter to scale the simulation predictions to the CHIME data to quantify the discrepancy.

3. Key Contributions

First Nonlinear Auto-Power Spectrum: Provides a robust measurement of the 21 cm auto power spectrum in the nonlinear regime ( $k \sim 1.5 \, h \, \text{Mpc}^{-1}$ ) with a signal-to-noise ratio of 12.4.
Robust Parameter Constraints: Successfully constrains the effective amplitude $A^2_{\text{HI}}$ despite strong parameter degeneracies, using a template-based approach that marginalizes over uncertain nonlinear shape parameters.
Systematic Uncertainty Budget: Quantifies systematic errors arising from template accuracy (1.3–2.2%), sample variance (1.4–1.6%), beam modeling (15%), and flux scale (8%).
Simulation Discrepancy Detection: Identifies a significant tension between observational data and state-of-the-art hydrodynamical simulations (IllustrisTNG).

4. Results

Parametric Constraints (Approach 1)

Effective Amplitude: The marginalized constraint on the amplitude parameter is:
$A^2_{\text{HI}} = 3.55^{+0.96}_{-1.32}(\text{stat}) \pm 0.61(\text{sys})$
at the effective redshift $z_{\text{eff}} = 1.16$ .
Individual Parameters (with Priors):
- Assuming a prior on $b_{\text{HI}}$ (based on simulations): $\Omega_{\text{HI}}(z=1.16) \times 10^3 = 0.79^{+0.20}_{-0.16}(\text{stat}) \pm 0.07(\text{sys})$ .
- Assuming a prior on $\Omega_{\text{HI}}$ (based on literature): $b_{\text{HI}}(z=1.16) = 1.99^{+1.22}_{-0.74}(\text{stat}) \pm 0.23(\text{sys})$ .
Consistency: The results are consistent with previous CHIME cross-correlation stacking analyses and other HI density estimates from the literature.

Simulation Comparison (Approach 2)

Significant Discrepancy: The measured power spectrum amplitude is significantly higher than predictions from IllustrisTNG.
- TNG100: Discrepant at 3.1 $\sigma$ (Best-fit scaling factor $2.38 \pm 0.18$ ).
- TNG300: Discrepant at 4.0 $\sigma$ (Best-fit scaling factor $3.97 \pm 0.30$ ).
Nature of Discrepancy: The mismatch is primarily in the overall amplitude of the clustering, not the shape (scale dependence). When the simulation curves are rescaled by the best-fit amplitude, they fit the data well.
Cause: The discrepancy is unlikely due to the total abundance of HI ( $\Omega_{\text{HI}}$ ), which is consistent between simulations and data. Instead, it suggests that the nonlinear redshift-space clustering of HI in the simulations is too weak compared to observations. This points to potential issues in how the simulations model baryonic physics (e.g., feedback from supernovae or AGN, gas stripping in satellite galaxies).

5. Significance

Astrophysical Insight at Nonlinear Scales: This work demonstrates that 21 cm intensity mapping can extract valuable astrophysical information even in the nonlinear regime, where traditional perturbative methods fail.
Constraint on Galaxy Formation Physics: The tension with IllustrisTNG suggests that current subgrid models for gas physics (specifically the coupling between dark matter halos and HI gas on small scales) may be incomplete. The simulations may underestimate the clustering strength of HI or overestimate gas stripping in satellites.
Methodological Advancement: The template-based forward-modeling technique provides a robust framework for analyzing intensity mapping data without requiring a fully analytic transfer function, which is difficult to derive for complex pipelines.
Future Directions: The results highlight the need for improved simulations and further CHIME observations (including cross-correlations with other tracers and bispectrum analysis) to disentangle the specific physical mechanisms driving the observed clustering.