🔭 astrophysics

Balancing bias, baryons, and scale cuts in LSST 3x2pt analysis

This paper presents a new open-source pipeline (MGL) for LSST 3x2pt analyses that utilizes the BACCO emulator and hybrid-effective field theory to demonstrate that while minimal bias models allow for unbiased cosmological constraints at small scales, they can introduce significant biases in neutrino mass inference, highlighting the critical trade-off between model complexity, scale cuts, and systematic degeneracies.

Original authors: Ottavia Truttero, Maria Tsedrik, Joe Zuntz, Alkistis Pourtsidou, Nikolina Šarčević, Christos Georgiou, the LSST Dark Energy Science Collaboration

Published 2026-06-10

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ottavia Truttero, Maria Tsedrik, Joe Zuntz, Alkistis Pourtsidou, Nikolina Šarčević, Christos Georgiou, the LSST Dark Energy Science Collaboration

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

Imagine the universe as a giant, invisible ocean of dark matter. We can't see this ocean directly, but we can see the "boats" floating on it: galaxies. By studying how these boats cluster together and how their shapes get stretched by the ocean's currents (a phenomenon called gravitational lensing), astronomers try to map the hidden ocean and understand the rules governing the universe.

This paper is about a new, highly detailed map-making project for the LSST (a massive telescope survey), specifically looking at data from its first year (Y1) and its tenth year (Y10). The authors are trying to solve a tricky puzzle: How do we get the most accurate map without getting confused by the "noise" in the water?

Here is a breakdown of their findings using simple analogies:

1. The Two Big Sources of "Noise"

When trying to measure the universe, there are two main things that mess up the picture if you look too closely (at small scales):

The "Boat" Problem (Galaxy Bias): Galaxies aren't perfect tracers of the dark matter ocean. They are like boats that might cluster in specific patterns that don't perfectly match the water underneath. If you assume the boats are just floating randomly, your map will be wrong once you zoom in.
The "Swirl" Problem (Baryonic Feedback): The ocean isn't just dark matter; it also has gas, stars, and black holes (baryons). These act like underwater engines and propellers, churning the water and creating swirls that push the dark matter around. If you ignore these swirls, your map of the ocean's density will be distorted.

2. The "Zoom" Dilemma

The authors wanted to know: How close can we zoom in before the noise ruins our measurements?

The Old Way (Linear Bias): In the past, scientists used a simple rule: "Boats follow the water perfectly." This works great if you only look at the big picture (zooming out). But as soon as you zoom in a little bit, this simple rule breaks, and your map becomes biased (wrong).
The New Way (HEFT & Minimal Bias): The authors tested more complex rules. One is a "Hybrid" rule (HEFT) that uses supercomputer simulations to understand exactly how boats move. The other is a "Minimal" rule that simplifies the complex math by ignoring a few tiny, hard-to-measure details.

The Finding: They discovered that the simple "Linear" rule only works if you stay zoomed out. However, the "Minimal" rule is surprisingly powerful. It allows them to zoom in much closer (up to a specific limit) without getting a biased map, almost as well as the super-complex "Hybrid" rule. This is a huge win because it saves time and computing power while keeping the map accurate.

3. The Great Mimicry

One of the most interesting discoveries in the paper is a case of cosmic impersonation.

The authors found that the "swirls" caused by the gas and stars (baryonic feedback) look very similar to the complex patterns caused by the "boat" behavior (higher-order bias).

The Analogy: Imagine you are looking at a shadow on a wall. You can't tell if the shadow is being cast by a person waving their arms (bias) or by a fan blowing dust around (baryons). The paper shows that the "boat" patterns can mimic the "swirl" patterns so well that, in some cases, you can't tell the difference just by looking at the map.
The Catch: While they look similar, they aren't identical. The "swirls" (baryons) can't perfectly copy every possible "boat" pattern. But because they are so similar, if you use a simplified model, you might accidentally blame the "boats" for the "swirls" or vice versa.

4. The "Ghost" in the Data (Neutrinos)

The team also asked: Can we detect the total mass of neutrinos (tiny, ghost-like particles that pass through everything) using this survey?

The Result: Yes, but only if they zoom in far enough. If they stay zoomed out, the signal is too weak.
The Risk: Here is where the "Mimicry" comes back to bite them. Because the "boat" patterns can look like the "ghost" particles, the specific value they calculate for the neutrino mass depends heavily on which "boat" rule they use.
- If they use the simple "Minimal" rule, they might get a number for the neutrino mass that is slightly off (biased) in certain scenarios, even though they will still detect that the mass is not zero.
- It's like trying to weigh a ghost on a scale that is also being shaken by a fan. You know the ghost is there, but the exact weight you read might change depending on how you account for the fan.

5. The New Tool (MGL)

To do all this, the authors built a new, open-source software tool called MGL. Think of this as a new, high-precision camera lens and image-processing software that allows them to take these complex measurements and test different "rules" for how the boats and swirls interact.

Summary

The paper concludes that for the upcoming LSST survey:

Don't stay zoomed out: To get the best data, we need to look at smaller scales, but we must use better math (like the "Minimal" or "Hybrid" bias models) to handle the complexity.
Simplification works: We don't always need the most complex supercomputer models; a slightly simpler "Minimal" model works just as well for the main cosmological measurements.
Beware of look-alikes: The effects of gas (baryons) and galaxy clustering (bias) can trick each other. This makes it hard to pin down the exact mass of neutrinos, though we will definitely know they have mass.
The Goal: The ultimate goal is to balance the complexity of the math with the amount of data we use, ensuring we get the most accurate map of the universe without getting lost in the noise.

Technical Summary: Balancing Bias, Baryons, and Scale Cuts in LSST 3 × 2pt Analysis

Problem Statement
Stage IV cosmological surveys, such as the Rubin Observatory Legacy Survey of Space and Time (LSST), will probe the nonlinear regime of structure formation where systematic effects from galaxy bias and baryonic feedback become dominant. In a $3 \times 2$ pt analysis (combining galaxy clustering, cosmic shear, and their cross-correlation), these effects introduce nuisance parameters that are often weakly constrained and can exhibit degeneracies with cosmological parameters. Specifically, the interplay between nonlinear galaxy bias and baryonic suppression in the matter power spectrum poses a challenge: purely linear bias models are insufficient at small scales ( $k \gtrsim 0.1 \, h/\text{Mpc}$ ), yet more complex perturbative models introduce additional degrees of freedom that may be degenerate with baryonic physics. The central problem addressed is how to balance model complexity against scale cuts to maximize constraining power on cosmological parameters ( $\Omega_m, \sigma_8$ ) and extensions (neutrino mass $M_\nu$ ) without introducing significant biases.

Methodology
The authors present a comprehensive analysis using a new, open-source Bayesian inference pipeline called MGL (Modified Gravity Lensing). The study generates mock data vectors resembling LSST Year 1 (Y1) and Year 10 (Y10) specifications, including cosmic shear and galaxy clustering spectra across 20 log-spaced bins in $\ell \in [20, 5000]$ .

Key methodological components include:

Modeling Framework: The analysis employs the BACCO emulator to model both the nonlinear matter power spectrum and baryonic feedback via a "baryonification" mechanism. For galaxy bias, three models are compared:
1. Linear Bias: Valid only at large scales.
2. Hybrid Effective Field Theory (HEFT): A perturbative approach combining Lagrangian bias expansion with N-body simulation evolution, valid up to $k_{\text{max}} \approx 0.7 \, h/\text{Mpc}$ .
3. Minimal Bias: A simplified HEFT variant where higher-order bias terms ( $b_{s^2}$ and $b_{\nabla^2}$ ) are fixed to zero.
Baryonic Treatment: Baryonic feedback is modeled using a suppression factor $S_{mm}$ applied to the matter power spectrum. For galaxy-galaxy lensing ( $P_{gm}$ ), the authors adopt the approximation from Zennaro et al. (2024), where $P_{gm}$ is corrected by $\sqrt{S_{mm}}$ , while $P_{gg}$ relies on the flexibility of the bias expansion.
Analysis Strategy: The authors perform likelihood analyses on mock data generated with the full HEFT model and fiducial parameters. They test various analysis assumptions, including mismatched models (e.g., analyzing HEFT-generated data with a linear or minimal bias model) and varying the inclusion of baryonic parameters.
Metrics: Performance is quantified using the Figure of Merit (FoM) for constraining power and the Figure of Bias (FoB) to measure deviations from fiducial values.

Key Contributions

Development of MGL Pipeline: The paper introduces MGL, an independent, open-source pipeline designed for Stage IV surveys that utilizes emulators for efficient computation of power spectra and supports MCMC analysis with the nautilus sampler.
Systematic Comparison of Bias Models: The work provides a direct comparison of linear, HEFT, and minimal bias models in the context of LSST Y1 and Y10, specifically evaluating their robustness against baryonic feedback and scale cuts.
Degeneracy Analysis: The study explicitly investigates the degeneracy between higher-order bias parameters and baryonic feedback, as well as the interplay between bias modeling and the detection of massive neutrinos.

Results

Scale Cuts and Bias Models: A linear bias model yields unbiased constraints on $\Omega_m$ and $\sigma_8$ (FoB $< 0.5\sigma$ ) only up to $k_{\text{max}} \approx 0.1 \, h/\text{Mpc}$ . Extending to smaller scales requires a perturbative approach.
Minimal Bias Efficacy: The "minimal bias" model (fixing $b_{s^2}$ and $b_{\nabla^2}$ to zero) performs comparably to the full HEFT model. It remains unbiased for $\Omega_m$ and $\sigma_8$ even at $k_{\text{max}} = 0.7 \, h/\text{Mpc}$ , provided the fiducial bias parameters are within standard ranges. This suggests that the full complexity of HEFT may not be necessary for standard cosmological constraints, offering a path to reduce the parameter space.
Baryonic-Bias Interplay: Higher-order bias parameters can mimic the suppression effects of baryonic feedback in the galaxy-galaxy power spectrum. However, baryonic feedback cannot fully reproduce the range of behaviors generated by higher-order bias terms. Neglecting baryonic corrections in the galaxy-galaxy lensing component does not significantly bias $\Omega_m$ or $\sigma_8$ , though it may shift the inferred baryonic parameter ( $\log_{10} M_c$ ).
Neutrino Mass Detection: A detection of the total neutrino mass ( $M_\nu$ ) is possible for both Y1 and Y10 when extending to nonlinear scales ( $k \geq 0.3 \, h/\text{Mpc}$ ). However, the specific inferred value of $M_\nu$ is not robust across all plausible scenarios. When mock data with elevated higher-order bias parameters are analyzed using the minimal bias model, significant biases in the inferred $M_\nu$ (up to $\sim 2\sigma$ for Y10) occur. This indicates a strong degeneracy between neutrino mass, higher-order bias terms, and projection effects.
Y1 vs. Y10: Moving from Y1 to Y10 improves the FoM by a factor of $\sim 2.3$ at $k_{\text{max}} = 0.7 \, h/\text{Mpc}$ . While $\Omega_m$ and $\sigma_8$ remain robust, the fragility of $M_\nu$ constraints under simplified bias models persists.

Significance and Claims
The paper claims that for standard $\Lambda$ CDM parameters ( $\Omega_m, \sigma_8$ ), a minimal bias parameterization is sufficient to extract unbiased constraints from LSST data up to the validity limit of HEFT ( $k_{\text{max}} = 0.7 \, h/\text{Mpc}$ ), even in the presence of baryonic feedback. This allows for the inclusion of quasi-nonlinear scales without the computational and statistical penalty of a full 15-term bias expansion.

However, the authors modestly caution that this conclusion is sensitive to the specific fiducial values of the bias parameters. While LSST should be capable of an unambiguous detection of non-zero neutrino mass, the precise significance and value of this detection are highly dependent on the accuracy of the bias modeling. The study highlights that simplified modeling approaches (like minimal bias) may yield robust results for standard cosmology but can introduce significant biases for extension parameters like neutrino mass if the underlying galaxy bias physics deviates from the assumed fiducial. Consequently, the authors advocate for careful testing of simplified models against specific science cases before adoption.