Fast Baryonic Field Painting for Sunyaev-Zel'dovich Analyses: Transfer Function vs. Hybrid Effective Field Theory

This paper introduces and compares two rapid methods for reconstructing Sunyaev-Zel'dovich effect fields on $N$-body simulations, demonstrating that a Gaussian Process transfer function approach excels for highly correlated fields while Hybrid Effective Field Theory is superior for enhancing weaker correlations.

The Gist

Imagine the universe as a giant, invisible ocean. Most of this ocean is made of "dark matter," which we can't see but know is there because of its gravity. However, there's also a smaller amount of "normal" matter (baryons)—the gas, stars, and planets we can actually detect. Understanding how this normal matter is distributed is crucial for testing our theories about how the universe works.

The problem? Simulating the universe is incredibly expensive and slow. To get a perfect picture of the gas, scientists need to run "full-physics" simulations that track every particle of gas and dark matter. These take millions of hours on supercomputers. On the other hand, simulating just the dark matter is fast and easy, but it misses the gas entirely.

This paper is about a clever shortcut: "Painting" the gas onto the fast, dark-matter-only simulations. The authors want to take a fast, cheap simulation of dark matter and quickly "paint" the properties of the gas onto it, so they can study the gas without waiting for the slow, expensive simulations to finish.

They tested two different "painting" techniques using a massive, high-quality simulation called MillenniumTNG as their benchmark (the "gold standard" they are trying to match).

The Two Painting Techniques

1. The "Transfer Function" Method (The Copy-Paste Artist)

Think of this method as a sophisticated copy-paste tool.

• How it works: The authors looked at the relationship between the dark matter and the gas in their training data. They created a "transfer function"—a mathematical rule that says, "If the dark matter looks like this, the gas should look like that."
• The Analogy: Imagine you have a black-and-white photo (dark matter) and you want to turn it into a color photo (gas). This method uses a pre-made filter (the transfer function) to instantly add the right colors based on the shades of gray.
• The Result: This method is amazing at preserving the exact details if the black-and-white photo and the color photo are already very similar. For example, when looking at Optical Depth (a measure of how much gas blocks light), the dark matter and gas are almost identical twins. The copy-paste method worked perfectly here, keeping the high-quality details intact.
• The Limitation: If the black-and-white photo and the color photo are very different (low correlation), this method can't invent new details. It can only rearrange what's already there. It can't make the gas look more like the gas if the starting dark matter doesn't match well.

2. The "Hybrid Effective Field Theory" (HEFT) Method (The Sculptor)

Think of this method as a sculptor working from a rough block of clay.

• How it works: Instead of just copying, this method starts with the initial conditions of the universe (the "clay") and uses physics rules to "sculpt" the gas field forward in time. It then adjusts a few knobs (called "bias parameters") to make the sculpture match the real gas as closely as possible.
• The Analogy: Imagine you have a rough lump of clay (dark matter) and you want to carve a specific statue (gas). The sculptor doesn't just copy a photo; they use their knowledge of how clay behaves to carve the shape, then tweak the final touches to match the reference.
• The Result: This method shines when the starting materials are very different. For the Compton-y field (a measure of gas pressure and heat), the dark matter and the gas don't look very much alike initially. The "copy-paste" method struggled here, but the "sculptor" (HEFT) was able to carve out a much better match, improving the correlation significantly.
• The Limitation: For fields that were already very similar (like the Optical Depth), the sculptor sometimes over-complicated things and didn't match the fine details as well as the simple copy-paste method.

The "Super-Method" (Extended HEFT)

The authors also tried a hybrid approach they call Extended HEFT. This is like giving the sculptor a magic chisel that can also directly copy parts of the original photo if needed.

• The Result: This new method combined the best of both worlds. It could match the high-quality details of the similar fields (like the copy-paste method) and still improve the difficult, dissimilar fields (like the sculptor method).

What Did They Find?

• If the gas and dark matter are already very similar: Use the Transfer Function (the copy-paste method). It's fast and keeps the details sharp.
• If the gas and dark matter are quite different: Use HEFT (the sculptor method). It can "invent" the missing correlations and build a better match.
• The Best of Both Worlds: The Extended HEFT method seems to be the most versatile, handling both easy and difficult cases better than either method alone.

Why Does This Matter?

By proving these "painting" methods work, the authors show that we can use fast, cheap computer simulations to study the complex behavior of gas in the universe. This allows scientists to run thousands of simulations to test different theories about the universe's history, something that would be impossible if they had to run the slow, expensive full-physics simulations every time.

In short, they found two different ways to quickly and accurately "fake" the gas in the universe, depending on how much the gas resembles the dark matter to begin with.

Technical Summary

Technical Summary: Fast Baryonic Field Painting for Sunyaev-Zel'dovich Analyses

Problem Statement
Understanding the distribution of baryonic matter is critical for validating cosmological models, yet observations at late times account for only a fraction of the predicted baryon abundance. The Sunyaev-Zel'dovich (SZ) effects—specifically the thermal (tSZ) and kinematic (kSZ) effects—serve as primary probes for mapping this baryonic content. While full-physics hydrodynamic simulations (e.g., IllustrisTNG) can accurately model these effects, they are computationally prohibitive, often requiring millions of CPU hours and lacking the statistical volume necessary for robust cosmological inference. Conversely, dark-matter-only N-body simulations are fast and accurate but lack baryonic physics. The challenge lies in developing computationally efficient methods to "paint" baryonic properties, specifically the optical depth ($\tau$) and Compton-$y$ parameters, onto dark-matter-only simulations to facilitate future SZ analyses.

Methodology
The authors evaluate two distinct field-level reconstruction methods using the MillenniumTNG simulation suite as a benchmark, trained on the CAMELS (Cosmology and Astrophysics with MachinE Learning Simulations) suite.

1. Gaussian Process (GP) Transfer Function Emulator:
   This approach models the relationship between the dark matter overdensity field ($\delta_m$) and the target baryonic field ($\delta_b$) via a transfer function $T(k) = \sqrt{P_b(k)/P_m(k)}$. Since the true transfer function is unknown for new simulations, the authors employ a Gaussian Process emulator trained on the CAMELS IllustrisTNG suite. The emulator predicts the transfer function based on cosmological parameters ($\Omega_m, \sigma_8$) and astrophysical feedback parameters. For the MillenniumTNG application, the astrophysical parameters are treated as nuisance parameters and optimized via differential evolution to match the target transfer function.

2. Hybrid Effective Field Theory (HEFT):
   This method combines Lagrangian Perturbation Theory (LPT) with numerical simulations. It treats baryons as biased tracers of the initial Lagrangian density field. The reconstruction involves advecting initial conditions forward in time using simulation-derived displacement fields and applying a bias expansion. The reconstructed field is a linear combination of the dark matter field and higher-order operators (e.g., $\delta^2$, shear squared, $\nabla^2\delta$) with fitted bias parameters. The authors test both scale-independent and scale-dependent bias fits.

3. Extended HEFT (E-HEFT):
   As a synthesis of the two approaches, the authors propose an extension where the dark matter field ($\delta_m$) itself is treated as a bias term in the expansion, rather than just the baseline. This allows for a field-level fit to the $\delta_m$ component, effectively synthesizing Lagrangian perturbation theory with Eulerian bias fitting.

Key Contributions and Results
The study compares these methods against the true $\tau$ and Compton-$y$ fields in MillenniumTNG using cross-correlation coefficients, power spectra, and power spectrum error metrics.

• Optical Depth ($\tau$) Reconstruction:
  The $\tau$ field exhibits a high initial correlation with the dark matter field ($r \approx 1$). In this regime, the Transfer Function method yields the most accurate reconstructions, successfully replicating the power spectrum up to the Nyquist frequency ($k \approx 6.78 \, h/\text{Mpc}$) with an order-of-magnitude lower error than HEFT methods. The HEFT methods, while capable of matching power spectra, degrade the cross-correlation coefficient at small scales compared to the transfer function approach. The E-HEFT method slightly improves upon standard HEFT but does not surpass the transfer function for this highly correlated field.

• Compton-$y$ Reconstruction:
  The Compton-$y$ field has a weaker initial correlation with dark matter ($r \sim 0.5$). Here, the Transfer Function method performs sub-optimally because it cannot enhance the cross-correlation between uncorrelated fields; it merely scales the existing correlation. In contrast, the HEFT approach proves superior, increasing the cross-correlation coefficient by 10–15% while maintaining power spectrum accuracy. The scale-independent HEFT variant achieves this with only four free bias parameters. The E-HEFT method provides the best overall performance for this field, significantly improving cross-correlation up to the box Nyquist frequency.

• Halo Cross-Correlations:
  Both methods generally reproduce the cross-correlation between the reconstructed fields and dark matter haloes well at large scales ($k < 1 \, h/\text{Mpc}$). However, discrepancies increase at smaller scales. The Compton-$y$ field shows larger deviations from truth in halo cross-correlations compared to $\tau$, suggesting that dark matter alone is insufficient to fully capture the Compton-$y$ distribution, particularly for high-mass haloes.

Significance and Conclusions
The paper demonstrates that fast, field-level reconstruction of SZ observables is feasible with high computational efficiency. The authors conclude that the choice of method depends on the intrinsic correlation between the dark matter and target baryonic fields:

• For highly correlated fields (like optical depth), the Transfer Function emulator is the preferred method, preserving high cross-correlations and minimizing power spectrum errors.
• For weakly correlated fields (like Compton-$y$), the HEFT framework is more effective, as it can actively enhance correlations through bias fitting.
• The Extended HEFT (E-HEFT) serves as a robust synthesis, offering improved performance for both regimes by incorporating the dark matter field as a bias term.

The authors note that while these methods are promising, future improvements could involve integrating generative machine learning models (such as diffusion models) to address small-scale correlations that current perturbative or transfer-function approaches may miss. The work establishes a foundation for using fast N-body simulations to model baryonic feedback effects for upcoming SZ surveys.