🔭 astrophysics

Self-calibration of weak lensing cosmic shear biases

This paper introduces a novel, simulation-free methodology that jointly infers and marginalizes over multiplicative and additive cosmic shear biases using parameterized ellipticity distributions, offering a cosmology-independent calibration solution for current and future weak lensing surveys.

Original authors: Giuseppe Congedo, Andy Taylor

Published 2026-08-21

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Giuseppe Congedo, Andy Taylor

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

Deep in the night sky, billions of distant galaxies are being stretched and distorted by the invisible scaffolding of the universe. This phenomenon, known as weak gravitational lensing, occurs when the gravity of massive structures, such as dark matter, bends the path of light traveling from those distant galaxies to Earth. By measuring these tiny distortions, astronomers can map the distribution of dark matter and understand how the universe has evolved over time. However, to turn these faint shapes into precise measurements of the cosmos, scientists must first ensure their tools are perfect. Even the slightest error in how a telescope records an image, or how a computer calculates a galaxy's shape, can introduce a systematic bias. If these errors are not corrected, they can lead to false conclusions about the nature of dark energy and the fate of the universe.

For years, the standard way to fix these errors has been to rely on massive computer simulations. Scientists create fake universes with known properties, run them through their measurement software, and see how the results differ from the truth. They then use these differences to calibrate their real-world data. But as telescopes become more powerful and the required precision becomes stricter, relying solely on simulations is becoming a bottleneck. Simulations are expensive, complex, and raise questions about whether they truly reflect the messy reality of the sky. A new approach is needed—one that can find and correct these errors using the real data itself, without needing a separate simulation to tell it what is wrong.

In a recent study, researchers Giuseppe Congedo and Andy Taylor have proposed a method to do exactly that. They developed a way to self-calibrate the measurements of cosmic shear by looking directly at the statistical distribution of galaxy shapes in the sky. Instead of treating every galaxy as an individual puzzle piece that needs to be fixed one by one, they looked at the entire collection of shapes as a single, shifting pattern. Their core idea is that while individual galaxy measurements are noisy and unpredictable, the overall shape of the distribution of millions of galaxies follows strict mathematical rules. When measurement errors, known as biases, creep in, they distort this entire pattern in a specific, predictable way.

The researchers focused on two types of errors that commonly plague these measurements. The first is an additive error, which acts like a constant shift, pushing the average shape of the galaxies in a specific direction. The second is a multiplicative error, which acts like a zoom or a stretch, making the entire distribution of shapes appear wider or narrower than it really is. In the past, scientists have been able to correct the additive shifts by looking at the data, but the multiplicative stretching has been much harder to pin down without external simulations. Congedo and Taylor showed that by analyzing the full distribution of galaxy shapes, including its sharp peak at the center and its long, thin tails, they could mathematically separate the true cosmic signal from these two types of errors simultaneously.

To test their method, the team created a realistic simulation of a galaxy survey. They started with a known, perfect distribution of galaxy shapes and then deliberately introduced specific amounts of additive and multiplicative errors, along with the random noise expected from a real telescope. They then applied their new statistical technique to this noisy, biased data. The method worked by building a flexible mathematical model of what the galaxy shapes should look like and then asking: "What combination of stretching, shifting, and noise would turn this perfect model into the messy data we see?" By running this process millions of times, they were able to find the most likely values for the errors.

The results were striking. Even with the presence of noise and the complex interplay between the different parameters, the method successfully recovered the exact amount of error that had been injected into the simulation. The researchers found that the technique could determine the multiplicative error to within a fraction of a percent, a level of precision required for the next generation of space telescopes. Crucially, this was achieved without knowing the true answer beforehand and without relying on any external simulation to provide a baseline. The method effectively "solved" for the errors by using the internal consistency of the data itself.

One of the most significant aspects of this work is that it does not depend on assumptions about the specific shapes of galaxies or the details of cosmology. It treats the galaxy shapes as a statistical population, meaning it works regardless of whether the galaxies are spiral or elliptical, or whether the universe is expanding fast or slow. The researchers also demonstrated that the method is robust against small imperfections in the model. Even if the mathematical description of the galaxy shapes was not a perfect match for reality, the errors in the final calibration remained very small. This suggests that the technique is practical for real-world application, where the true distribution of galaxy shapes is never perfectly known.

The study also addressed a common concern: that the errors might be so tangled with the natural variations in galaxy shapes that they cannot be separated. The researchers showed that while some of the underlying parameters are indeed linked, the statistical process they used effectively untangles them. By looking at the entire distribution rather than just the average, they could distinguish between a natural spread in galaxy shapes and an artificial stretch caused by measurement errors. This ability to distinguish between the two is what allows the method to work without needing to know the exact properties of the galaxies in advance.

This new approach offers a powerful tool for future surveys, such as those planned by the Euclid and Rubin telescopes, which aim to map the universe with unprecedented accuracy. By providing a way to calibrate measurements directly from the data, it reduces the reliance on massive, resource-intensive simulations. While the method was tested on simulated data, the researchers argue that the same principles apply to real observations. The flexibility of the model, which can adapt to different types of data, means it can be applied to various subsets of galaxies, such as those at different distances or in different parts of the sky.

Ultimately, this work represents a shift in how astronomers think about calibration. Instead of viewing errors as a nuisance that must be corrected by an external reference, the method treats the errors as a feature of the data that can be decoded. It turns the challenge of measurement uncertainty into a solvable statistical problem. By proving that multiplicative and additive biases can be inferred jointly from the observed distribution of galaxy shapes, the researchers have opened a path toward more independent and reliable cosmological measurements. As the field moves toward even higher precision, this self-calibrating capability could become essential for unlocking the secrets of the dark universe.

Technical Summary: Self-calibration of weak lensing cosmic shear biases

Problem Statement
Current and future Stage-III and Stage-IV weak lensing surveys (e.g., Euclid, Rubin, Roman) require percent-level precision in cosmological measurements. A primary obstacle to this accuracy is the presence of residual systematic biases in cosmic shear measurements, specifically multiplicative ( $m$ ) and additive ( $c$ ) biases. These arise from imperfect modeling of the point-spread function (PSF), detector effects, and measurement methodologies. Historically, correcting these residuals has relied heavily on external simulations to calibrate the biases. However, this approach raises concerns regarding the sensitivity of calibration to simulation setups and the computational volume required. While additive biases can be empirically corrected on data, multiplicative biases are notoriously difficult to correct without simulations, as current methods (e.g., Metacalibration) may still suffer from residual biases when applied to the stringent requirements of Stage-IV surveys.

Methodology
The authors propose a novel, unified methodology to infer both multiplicative and additive biases directly from the observed distribution of galaxy ellipticities. While the approach aims to reduce reliance on external simulations for the final bias values, it requires a model of the unbiased lensed distribution. The paper notes that while deep fields or simulations can be used to constrain this model, the method is designed to allow the unbiased distribution to be constrained from the data itself, provided the functional form is sufficiently flexible and accurate.

The core of the method rests on the statistical transformation of the probability distribution function (PDF) of ellipticities. The paper establishes that the observed ellipticity distribution, $\hat{p}_{\hat{\epsilon}}$ , is related to the unbiased lensed distribution, $p_{\epsilon}$ , via an affine transformation involving the biases and a convolution with noise:
$\hat{p}_{\hat{\epsilon}}(\hat{\epsilon}|m, c) = \frac{1}{|1+m|} p_{\epsilon}\left(\frac{\hat{\epsilon}-c}{1+m}\right) * p_{\epsilon_n}(\epsilon_n)$
where $m$ acts as a rescaling factor and $c$ as a shift.

Key technical components include:

Moment Analysis: The authors analyze how central and raw moments of the ellipticity distribution transform under bias. They demonstrate that while central moments are insensitive to additive bias $c$ , raw moments and higher-order statistics (skewness, kurtosis) contain information about both $m$ and $c$ . Crucially, the non-Gaussian nature of the ellipticity distribution (characterized by a sharp peak at zero and truncation at $|\epsilon| < 1$ ) provides sufficient information to break degeneracies that would exist in a Gaussian distribution.
Bayesian Inference: The problem is framed as a Bayesian forward-modeling exercise. The authors construct a likelihood function based on the Poisson statistics of binned ellipticity counts. They model the underlying unbiased distribution using a flexible parametric form (Eq. 16) characterized by hyper-parameters $\theta = \{\sigma, \beta, \gamma, \kappa\}$ .
Marginalization: To address the degeneracy between the multiplicative bias $m$ and the dispersion of the underlying distribution, the method marginalizes over the unknown hyper-parameters $\theta$ . The joint posterior distribution is computed, and the final constraints on $m$ and $c$ are derived by integrating out the nuisance parameters.

Key Contributions and Results

Unified Estimation: The paper demonstrates that multiplicative and additive biases can be estimated simultaneously from a single observed ellipticity distribution, removing the need for separate calibration pipelines.
Simulation-Reduced Calibration: The methodology aims to infer biases directly from data without relying on external simulations for the final bias values, provided a sufficiently accurate functional form for the unbiased distribution can be identified. The paper acknowledges that the accurate modeling of this unbiased distribution is the main limitation; while deep fields or simulations can aid in constraining this model, the authors argue that the functional form can often be constrained by the data itself (e.g., via symmetry and boundary conditions) or aided by deep fields.
Performance Validation: The authors tested the estimator on representative data generated from pre-launch Euclid simulations and Q1 data.
- In noise-free simulations with injected biases of $m_{true} = 0.05$ and $c_{true} = -0.01$ , the method recovered the values within one standard deviation ( $m - m_{true} = (2.7 \pm 8.4) \times 10^{-3}$ ).
- The analysis showed that while degeneracies exist between $m$ and the distribution's dispersion parameter $\sigma$ , marginalization effectively mitigates their impact on the final bias estimates.
- Convergence tests indicated that for sample sizes $N > 10^6$ , the residual error on $m$ drops below 1%.
Robustness to Mismodeling: The study tested the impact of ignoring the measurement noise term ( $\sigma_{\epsilon_n}$ ) in the likelihood. Results showed that even if this term is ignored, the residual bias on $m$ remains below 1% for noise levels comparable to those found in current data ( $\sigma_{\epsilon_n} \approx 0.05$ ).

Significance and Claims
The paper claims that this self-calibration technique offers a high potential for applicability to current and future weak lensing surveys. By inferring biases directly from the data's statistical properties, the method:

Reduces reliance on large-scale, realistic simulations for the final bias calibration, though it may still require them (or deep fields) to constrain the underlying unbiased distribution model.
Provides a cosmology-agnostic and morphology-agnostic approach, making it applicable to broad samples, including tomographic and morphological bins.
Offers a way to provide empirical priors for bias calibration, potentially aiding simulation-based calibration rather than replacing it entirely.

The authors conclude that while the accurate modeling of the unbiased lensed distribution is a prerequisite, the functional forms required are broadly known and can be constrained by the data itself (potentially aided by deep fields). The method represents a step toward achieving the rigorous calibration standards required for Stage-IV cosmological constraints.

More like this