Euclid: The linear-construction covariance and cosmology

This paper validates the linear-construction method for estimating galaxy cluster covariance matrices as a computationally efficient alternative to standard sample covariance, demonstrating that it yields cosmological parameter constraints for $\Omega_{\rm m}$ and $\sigma_8$ that are statistically indistinguishable from those derived using the more expensive sample-covariance approach.

The Gist

Imagine you are an astronomer trying to understand the architecture of the entire universe. You have a massive map of millions of galaxy clusters (huge groups of galaxies held together by gravity). To figure out how the universe is built and what it's made of, you need to measure how these clusters are spaced out relative to one another. This measurement is called the 2-point correlation function.

But here's the catch: to trust your measurements, you need to know how much "noise" or random error is in your data. In statistics, this is called the covariance matrix. It's like knowing the margin of error on a ruler. If you don't know the margin of error, you can't trust your measurement of the universe's shape.

The Problem: The "Brute Force" Bottleneck

Traditionally, to figure out this margin of error, scientists use a method called Sample Covariance.

• The Analogy: Imagine you want to know how much a specific type of dice roll varies. You roll the dice 1,000 times, write down the results, and calculate the spread.
• The Reality: In astronomy, "rolling the dice" means running a supercomputer simulation of the entire universe. To get a reliable error margin, you need to run this simulation 1,000 times.
• The Cost: Running one simulation takes about a day on a powerful supercomputer. Running 1,000 of them would take 1,000 days (nearly 3 years!). This is too slow for the upcoming Euclid space telescope, which will generate data faster than we can simulate it.

The Solution: The "Linear Construction" (LC) Shortcut

The authors of this paper tested a new, super-fast method called Linear Construction (LC).

• The Analogy: Instead of rolling the dice 1,000 times from scratch, the LC method is like a clever magician. It rolls the dice a few times, then uses a mathematical trick to construct the rest of the results instantly. It doesn't simulate the whole universe 1,000 times; it simulates it a few times and mathematically fills in the gaps.
• The Speed: This method is 20 times faster. Instead of waiting 3 years, you get your answer in about 5 days.

The Big Question: Is the Shortcut Accurate?

The authors asked: "If we use this fast shortcut, will we get the wrong answer about the universe?"

They set up a massive test:

1. They created 1,000 fake universes (simulations) with known properties (they knew the "true" answer).
2. They calculated the error margins using the slow, traditional method (Sample Covariance).
3. They calculated the error margins using the fast LC method.
4. They used both sets of error margins to try and guess the properties of the universe (specifically, how much matter is in the universe and how clumpy it is).

The Results: The Shortcut Works!

The results were fantastic.

• The "True" Universe: The simulations were built with specific values for matter density ($\Omega_m$) and clumpiness ($\sigma_8$).
• The Slow Method: Estimated the values almost perfectly.
• The Fast LC Method: Estimated the values almost identically to the slow method.

The difference between the two methods was so tiny (less than 0.16% of the uncertainty) that it was practically invisible. It was like two different maps of the same city; one was drawn by walking every street (slow), and the other by using a satellite algorithm (fast). Both maps showed the same streets in the same places.

A Small Hiccup: The "Inverted" Problem

There was one technical snag. In statistics, to use these error margins, you have to "invert" the matrix (a bit like flipping a fraction upside down).

• The fast LC method sometimes produces a matrix that is "wobbly" or unstable when you try to flip it, leading to small mathematical errors.
• The authors developed a new mathematical "stabilizer" (a correction factor) to fix this wobble. Even with this fix, the results remained incredibly accurate.

The Bottom Line

This paper proves that the Linear Construction method is a game-changer.

• Old Way: Wait 3 years to get a reliable answer.
• New Way: Wait 5 days to get the same reliable answer.

This means that when the Euclid telescope starts sending back terabytes of data, scientists won't be stuck waiting for supercomputers to catch up. They can use this fast method to analyze the data in real-time, helping us understand the dark energy and dark matter that make up our universe much sooner than we thought possible.

In short: They found a way to cheat the math to save time, and they proved that the cheat code doesn't break the game.

Technical Summary

Here is a detailed technical summary of the paper "Euclid: The linear-construction covariance and cosmology" by Lindholm et al. (2026).

1. Problem Statement

The paper addresses the computational bottleneck associated with estimating the covariance matrix of the galaxy cluster 2-point correlation function (2PCF) for upcoming large-scale structure surveys like Euclid.

• The Challenge: Standard cosmological parameter estimation requires a precise covariance matrix, typically estimated via the "sample covariance" method using thousands of mock catalogues (e.g., 1,000–10,000). For a survey like Euclid with $3 \times 10^7$ galaxies, computing the 2PCF for a single mock takes ~1 day on a 48-CPU node. Generating the full covariance matrix from 1,000 mocks would take ~1,000 days.
• The Goal: To validate the Linear-Construction (LC) method (Keihänen et al. 2022), which claims to be ~20 times faster than the standard sample-covariance method, for use in constraining cosmological parameters ($\Omega_m$ and $\sigma_8$). The authors aim to determine if the LC method introduces biases that compromise cosmological inference compared to the standard sample covariance.

2. Methodology

A. The Linear-Construction (LC) Method

The LC method exploits the mathematical structure of the Landy–Szalay (LS) estimator when paired with the "split" random catalogue option.

• Principle: The covariance of the 2PCF, $\text{cov}[\xi(r_1), \xi(r_2)]$, can be expressed as $A + M^{-1}B$, where $M$ is the ratio of random to data points.
• Implementation: Instead of computing the covariance with a large $M$ (e.g., $M=50$), the LC method computes 2PCFs with small $M$ values (specifically $M=1$ and $M=2$) and solves for the coefficients $A$ and $B$. This allows the reconstruction of the covariance for any $M$ with significantly fewer random pairs, reducing computation time by a factor of ~20.

B. Inverting the Covariance (Precision Matrix)

Cosmological likelihoods require the inverse of the covariance matrix (precision matrix).

• Bias Issue: The direct inverse of the LC covariance is biased. Unlike the sample covariance, which follows an inverse Wishart distribution (correctable via the Hartlap factor), the LC covariance is a linear combination of sample covariances and does not follow this distribution.
• Solution: The authors derived an approximate bias correction for the LC precision matrix using a Neumann series expansion. They implemented an iterative procedure to solve for the corrected precision matrix, achieving convergence within 14–36 steps.

C. Covariance Modeling

To avoid the noise inherent in numerical covariance matrices, the authors fitted a theoretical 4-parameter covariance model to the numerical estimates.

• Model: Based on the work of EC24 (Fumagalli et al. 2024), the model includes Gaussian and non-Gaussian terms.
• Adaptation: The original EC24 fitting procedure (using parameters $\beta, \alpha, \gamma$) was incompatible with the LC method's linear structure. The authors re-parametrized the model as a linear combination of four component matrices ($C = \sum p_k C_k$) to allow fitting against the LC covariance.

D. Simulation Setup

• Data: 1,000 mock dark matter halo catalogues generated using the PINOCCHIO algorithm.
• Cosmology: Flat $\Lambda$CDM with $\Omega_m = 0.30711$, $\sigma_8 = 0.8288$.
• Selection: Halos with virial mass $> 10^{14} h^{-1} M_\odot$ in four redshift shells ($z=0.0\text{--}1.6$).
• Analysis: The authors compared cosmological constraints derived from:
  1. Sample covariance (standard).
  2. LC covariance (raw and bias-corrected).
  3. Covariance models fitted to both types.

3. Key Contributions

1. Bias Correction for LC Precision Matrix: The paper provides a novel derivation and iterative algorithm to correct the bias in the inverse LC covariance matrix, a necessary step for likelihood analysis that was previously missing.
2. Adapted Covariance Modeling: The authors successfully adapted the EC24 covariance modeling framework to work with the LC method by re-parametrizing the model coefficients, enabling a direct comparison between LC-fitted and sample-fitted models.
3. Validation of LC for Cosmology: This is the first study to rigorously test the LC method's impact on final cosmological parameter constraints (posteriors) rather than just covariance matrix properties.

4. Results

• Covariance Fitting: The 4-parameter model fitted to the LC covariance reproduced the numerical sample covariance with residuals of a few percent to ~10%, comparable to the performance of the model fitted to the sample covariance.
• Precision Matrix Bias: While the iterative bias correction improved the LC precision matrix, the Kolmogorov-Smirnov test on $\chi^2$ distributions showed that the corrected LC precision matrix still deviated from the theoretical $\chi^2$ distribution (p-values $< 10^{-7}$). However, this bias was found to propagate weakly to the final parameter constraints.
• Cosmological Parameter Constraints:
  • Sample Covariance: $\Omega_m = 0.307 \pm 0.003$, $\sigma_8 = 0.826 \pm 0.009$.
  • LC Covariance: $\Omega_m = 0.308 \pm 0.003$, $\sigma_8 = 0.825 \pm 0.009$.
  • Comparison: The posterior widths are identical. The difference in median values is < 0.16$\sigma$ for both parameters.
  • Significance: The shift introduced by the LC method is negligible compared to the variance between different mock realisations (cosmic variance).

5. Significance

• Computational Efficiency: The LC method reduces the computational cost of constructing the covariance matrix by a factor of ~20 (e.g., reducing a 200-day calculation to <1 day for a realistic Euclid setup) without sacrificing the accuracy of cosmological parameter estimation.
• Feasibility for Future Surveys: The results validate the use of the LC method for the Euclid mission and similar large-scale structure surveys, making high-precision covariance estimation computationally tractable.
• Applicability: While tested on galaxy clusters (where evolution is mostly linear), the authors note that the method is theoretically applicable to galaxy clustering as well, though modeling the covariance for galaxies (affected by non-linear growth and baryonic physics) would be more complex.

Conclusion: The Linear-Construction method is a robust, highly efficient alternative to the standard sample-covariance method for estimating 2PCF covariances. When combined with a fitted covariance model, it yields cosmological constraints indistinguishable from the standard method, enabling faster analysis of next-generation cosmological data.